This study was carried out at Hospital General Universitario Gregorio Marañón in Madrid (Spain), a tertiary university hospital with 1,350 beds. Toxigenic C. difficile is routinely investigated in all diarrhoeic stool samples from patients older than 2 years. The microbiology laboratory receives samples both from the hospital itself and from 13 outpatient centres in the same area.

We conducted a prospective study from January 2019 to April 2021. We enrolled patients over 2 years old with a positive test result for toxigenic C. difficile and met the clinical criteria for CDI. We excluded patients younger than 2 years old because they have a high colonisation rate, which in many cases makes it difficult to determine the true cause of diarrhoea, in infants less than 7 days are 15%, in infants between 6 and 12 months 41% and 22% in 2 year old (Tougas et al., 2021). It appears that the infants’ gut is immune to the effects of toxins A and B although the mechanism is unknown, very few cases of CDI have been described in infants under 2 years of age (Jangi and Lamont, 2010). We collected clinical data and faecal samples from all patients. Patients were classified into 2 groups: patients with a primary CDI episode who did not subsequently develop recurrent CDI (non-recurrent) and patients with a primary CDI episode who subsequently developed recurrent CDI (recurrent).

An episode of CDI was defined as the presence of a positive toxigenic CDI test result, together with diarrhoea (≥3 unformed stools in 24 h) or findings of pseudomembranous colitis by colonoscopy, following the definitions set out in the guidelines of the Society for Healthcare Epidemiology of America (SHEA) and the Infectious Diseases Society of America (IDSA) (McDonald et al., 2018).

CDI is considered recurrent when it recurs within 8 weeks after a previous episode, provided the symptoms from the previous episode resolved after completion of the initial treatment (van Prehn et al., 2021). Having new symptoms and a positive sample after 60 days was considered a new episode.

Defined daily dose (DDD) is defined in the “WHO Collaborating Centre for Drug Statistics Methodology” as the assumed average maintenance dose per day in the case of a drug used for its main indication in adults (WHO, 2021).

We classified antimicrobial drugs as anaerobic, as previously described by Mulder et al, “We also classified the antimicrobial drugs in antimicrobial drugs with anaerobic activity (consisting of combinations of penicillins, including beta-lactamase inhibitors (J01CR), lincosamides (J01 FF) and imidazole derivatives (metronidazole) (J01XD): anaerobic+)…” (Mulder et al., 2020).

The severity of the CDI episode was defined according to the SHEA and IDSA guidelines (McDonald et al., 2018).

Death was considered CDI-related when there were no other attributable causes and/or it occurred within 10 days after the diagnosis of CDI and/or was due to known complications of CDI.

As for microbiome-related definitions, we considered richness as the number of different species found in a sample (Whittaker et al., 2001). Evenness (Pielou index) was defined as the degree to which different species are similar or uniform in abundance. Diversity indicated the degree of species richness and abundance, where alpha diversity is referred to the diversity within an individual -Inverse Simpson index: an indication of the richness in a community with the same evenness; and Shannon index: takes into account the number of species living in a habitat (richness) and their relative abundance (evenness)- and beta diversity referred to the difference in diversity between individuals (Whittaker et al., 2001).

We considered as microbiota-related diseases the following: cholelithiasis, colorectal cancer, hepatic encephalopathy, idiopathic constipation, inflammatory bowel disease, irritable bowel syndrome, familial Mediterranean fever, gastric lymphoma or carcinoma, arthritis, asthma, atopy, dermatitis, psoriasis, autoimmune disease, fatigue syndrome, diabetes mellitus, hypercholesterolaemia, idiopathic thrombocytopenic purpura, myocardial ischaemia, metabolic syndrome, behavioural disorders, multiple sclerosis, myoclonus dystonia, non-alcoholic fatty liver disease, oxalate kidney stones and Parkinson’s disease.

Samples were processed using a rapid detection kit for toxigenic C. difficile. This rapid test consists of glutamate dehydrogenase antigen and toxins A and B detection by immunochromatography (C Diff Quik-Chek Complete assay, TechLab, Blacksburg, VA, USA) (Sensitivity 90.5%, specificity 93.1%, predictive positive value 76.4%, predictive negative value 97.6% in comparison with bacterial culture) and a real-time polymerase chain reaction (PCR) assay of the C. difficile toxin B gene (Xpert.C. difficile Assay, GeneXpert, Cepheid, Sunnyvale, CA, USA) (Sensitivity 98.79%, specificity 90.82%, predictive positive value 56.58%, predictive negative value 99.83% in comparison with bacterial culture).

In addition, all samples were cultured on C. difficile selective agar (bioMeriéux, Marcy l’Etoile, France). Suspected toxigenic C. difficile colonies were confirmed using immunochromatography (C Diff Quik-Chek Complete assay, TechLab, Blacksburg, VA, USA).

Calprotectin levels were measured using EDITM Quantitative Faecal Calprotectin ELISA (Measurement range: 1 – 2000 µg/g stool) (Epitope Diagnostics, Inc, San Diego, CA, USA), according to the manufacturers’ instructions.

The demographic data collected included age and sex. Regarding clinical data, the underlying conditions were recorded using the McCabe and Jackson score for prognosis of underlying diseases. The McCabe score was used to categorise the risk of death by underlying comorbidities. The 3 categories were rapidly fatal (within 1 year), ultimately fatal (within 5 years) and non-fatal (>5 years) (Jackson and McCabe, 1962); comorbidity was graded according to the Charlson comorbidity index (Charlson et al., 1987). The Charlson comorbidity index is one of the most extensively used comorbidity indices (Huang et al., 2014). This index consists of 19 conditions, each of which is assigned a weighting of 1, 2, 3 or 6 based on the adjusted hazard ratio for each comorbidity. The sum of the weighted scores gives the total score (Yurkovich et al., 2015).

Other clinical data collected included antibiotic treatment, proton pump inhibitor use, nasogastric tube use, mechanical ventilation, surgery and chemotherapy or radiotherapy in the month prior to diagnosis of CDI. Regarding antibiotic consumption, the DDDs for each antibiotic were recorded. For the CDI episode, data on severity of the episode, treatment received, treatment failure, recurrence, mortality and CDI-related mortality were recorded.

Immediately upon receipt, the faecal samples were homogenized, aliquoted and stored at −80°C until the day of analysis. Total DNA was extracted from faecal samples using the Qiagen Fast QiaAmp DNA stool mini kit (QIAGEN, Valencia, CA, USA) according to the manufacturer’s protocol with the inclusion of a physical lysis step. The sample was lysed twice at 6.5m/s for 45 seconds in FastPrep-24 (MPBio, Derby, UK) with lysis matrix tubes E (MPBio, Derby, UK). The hypervariable V4 region of the 16S rRNA gene was amplified by PCR, with 515-806 primers (515:GTGCCAGCMGCCGCGGTAA, 806:GGACTACHVGGGTWTCTAAT) tailed with sequences to incorporate Illumina flow cell adapters and indexing barcodes (Illumina, San Diego, CA, USA). PCR amplification program was 1 cycle of 98°C 30 seconds; 25 cycles of 98°C 10 seconds, 60°C 20 seconds, 72°C 20 seconds; 1 cycle of 72°C 2 minutes. Q5® High-Fidelity 2X Master Mix (New England Biolabs, Ipswich, MA, USA)

Primer dimers and low-molecular-weight products were removed using Agencourt Ampure Beads (Beckman Coulter, Spain). Samples were quantified and quality was checked for amplicon size using the 4200 TapeStation (Agilent Technologies, Santa Clara, CA, USA). Amplicons were equimolar pooled and sequenced (2 × 250) on an Illumina Miseq platform (Illumina, San Diego, CA, USA) according to standard protocols.

The raw data were pre-processed, grouped by operational taxonomic units (OTUs) with 97% similarity and taxonomically classified using MOTHUR software (Patrick D. Schloss, PhD, © 2019, Michigan, USA) and SILVA and RDP databases. Analyses of species richness (OTUs observed), evenness (Pielou index), alpha diversity (Shannon index) and beta diversity (Bray-Curtys distance, unweighted unifrac distance) were performed with MOTHUR and R software (R Core Team, 2021, Vienna, Austria).

The statistical analyses were performed using R (R Core Team, 2021, Vienna, Austria). Frequencies were calculated for qualitative variables, and proportions were calculated with their 95% confidence interval (CI) following a binomial distribution. For quantitative variables, the median and interquartile range (IQR) or mean and standard deviation (SD) were calculated. Microbiota analyses were performed with R using the packages phyloseq, microbiome, microbiomeStat, vegan, DESeq2 and microeco.

The MaAslin2 (Multivariate microbial Association by Linear models) library was used to study the relationship between microbiota and clinical variables (Mallick et al., 2021). Differences between groups were determined using the χ² test. Continuous variables were compared using the t test or the Mann-Whitney test (when a normal distribution could not be assumed). The normality of the distribution of continuous variables was assessed using the Kolmogorov-Smirnov test with the Lilliefors correction. A multivariate logistic regression model was used to assess predictors of R-CDI. The odds ratio (OR) and 95% CI were calculated. A p value <0.05 was considered significant. All significant variables in the bivariate study were included in the multivariate model, in the multivariate study a logistic regression was performed in which the weight of each variable was scored according to its odds ratio.

This study was approved by the Ethics Committee of Hospital General Universitario Gregorio Marañón in Madrid (number MICRO.HGUGM.2016-029).

Participants’ informed consent was obtained before enrolment. All patients included in the study were given the sufficient time to make the decision to participate and understood each of the components of the study. Patients were also informed about their rights of access, rectification, cancellation and opposition.

The data for this study have been deposited in the European Nucleotide Archive (ENA) at EMBL-EBI under accession number PRJEB57947 (https://www.ebi.ac.uk/ena/browser/view/PRJEB57947).

The median age of the patients was 67 years for those with non-recurrent disease and 78.50 years for those with recurrent disease. In both groups, there were more females than males (non-recurrent 54.8% [80/146]; recurrent 74.1% [40/54]; p = 0.013).

The most common underlying diseases were cardiovascular, metabolic, endocrine, nephrourological and gastrointestinal diseases ( Table 1 ). The median Charlson comorbidity index was 4 (IQR: 2-6) in the non-recurrent group and 3.50 (IQR: 3-6) in the recurrent group (p= 0.256). Both groups contained a high percentage of patients with underlying diseases related to alterations of the microbiota, although this difference was not statistically significant (non-recurrent 65.8% [96/146]; recurrent 75.9% [41/54]; (p=0.169). Within this group of diseases, diabetes mellitus and ischemic heart disease were the most common ( Table 1 ). Approximately 36% of all patients were immunocompromised ( Table 1 ). Surprisingly, a higher percentage of patients had inflammatory bowel disease in the non-recurrent group (13.5%) than in the recurrent group (2.4%), and a higher percentage of patients had arthritis in the recurrent group (12.2%) than in the non-recurrent group (3.1%) (all p <0.05).

No significant differences regarding risk factors for CDI were recorded in either of the groups. The main risk factor for developing CDI was antibiotic therapy in the month prior to sampling. In the patients with non-recurrent disease, the most frequent antibiotics were cephalosporins, followed by penicillins, carbapenems and quinolones, while in the patients with recurrent disease the most frequently administered antibiotics were cephalosporins, followed by quinolones, penicillins and carbapenems ( Table 2 ). As for other CDI risk factors, treatment with proton pump inhibitors and being hospitalised were widely present ( Table 3 ).

Regarding the severity of the CDI episodes, most were mild (non-recurrent 65.8%; recurrent 64.8%; p=0.901). There were 2 cases of toxic megacolon and 2 cases of pseudomembranous colitis, all in the non-recurrent group. Most CDI episodes (non-recurrent 57.5%; recurrent 38.9%; p=0.020) were hospital onset, healthcare facility–associated ( Table 4 ). The median toxin B PCR cycle was lower in patients with recurrent disease than in those with non-recurrent disease (23.25 vs 26.10; p=0.003). Those with recurrent disease had higher levels of creatinine and faecal calprotectin than those with non-recurrent disease (all p<0.05) ( Table 4 ).

In both groups, most patients received treatment for CDI (non-recurrent, 97.9% [142/146]; recurrent, 100% [54/54]), mainly with vancomycin (non-recurrent, 85.6%, recurrent, 88.7%), followed by metronidazole (non-recurrent, 34.2%, recurrent, 26.4%). Fourteen patients included in this study had been enrolled in a clinical trial with faecal microbiota transplantation as their initial treatment. Few patients received fidaxomicin, and few patients received bezlotoxumab. There were no significant differences between the groups with respect to the CDI treatment they received ( Table 4 ).

One patient with recurrent disease died (probably CDI-related), and no patients in the non-recurrent group died from a probably CDI-related condition. Both 30-day mortality and 90-day mortality (3.7% and 11.1%, respectively) were higher in the recurrent group than in the non-recurrent group (0% and 1.4% respectively) (all p<0.05) ( Table 5 ).

We analyzed 179 samples for calprotectin levels (128 from non-recurrent patients and 51 from recurrent patients). The median values obtained were 98.92µg/mg (IQR 8.33-191.05). We observed significant differences between non-recurrent and recurrent patients (92.91 µg/mg vs 150.36 µg/mg; p=0.01).We obtained a median calprotectin value of 81.32 µg/mg (IQR: 0.00-171.02) for mild CDI episodes and 148.16 µg/mg (IQR: 55.83-230.38) for severe CDI episodes (p=0.002).

We classified faecal calprotectin levels as >185µg/mg and <185µg/mg (45.1% and 21.1% of the recurrent group, respectively; p=0.01) ( Table 6 ). Patients with faecal calprotectin >185µg/mg had a lower median toxin B PCR cycle threshold than patients with faecal calprotectin levels <185µg/mg (24.00 vs 25.60; p=0.034). Patients with faecal calprotectin higher than 185µg/mg had higher levels of leucocytes and more days of diarrhoea than patients with lower levels of faecal calprotectin. In addition, more patients had severe and severe-complicated CDI episodes with a calprotectin level >185µg/mg and more mild episodes with calprotectin <185µg/mg (all p<0.05) ( Table 6 ).

We ran a multivariate model and found the following independent risk factors for developing R-CDI: age, faecal calprotectin level >185µg/mg, toxin B PCR cycle threshold <23, immunosuppression, female sex, and creatinine levels higher than 1mg/dL (all p<0.05). We built a risk index for developing R-CDI based on the risk factors we obtained in the multivariable model ( Table 7 ). The ROC curve analysis showed that this risk index had an area under the curve (AUC) of 0.783, with a power of 0.999, significance level of 0.05 and an overall accuracy of 79.0% ( Figure 1 ).

When studying the primary episode microbiome diversity of patients who went on to develop R-CDI and those who did not, we found significant differences in the inverse Simpson index (p=0.021), with lower values in patients with non-recurrent disease.

We found significant differences between those who presented any type of inflammatory bowel disease in the Shannon index, richness and evenness, with lower values than those of patients who did not present this condition (all p<0.05). Patients with a nasogastric tube during the month prior to the episode had lower values in the Shannon index, richness and evenness (all p<0.05).We found lower evenness values for patients who had taken probiotics in the month prior to the episode and those who had undergone colectomy or ileostomy and lower richness values for smokers and patients who had received antifungal treatment in the month prior to the episode.

We also studied the effects of antibiotics on the diversity of the microbiota and found that patients who received more than 3 antibiotics had less alpha diversity, less evenness (all p<0.05) and significant differences in beta diversity with respect to patients who received fewer than 3 antibiotics. The differences in alpha diversity and evenness remained significant in patients with recurrent disease compared with those who had non-recurrent disease.

In terms of beta diversity and of the variance of the intra-group samples, we found no significant differences between the recurrent and the non-recurrent group.

Regarding the relative abundance of the different taxonomic groups, we found that Bacteroidetes was the major phylum in patients with non-recurrent disease (40.18%), followed by Firmicutes (35.69%), whereas in those with recurrent disease, the major phylum was Firmicutes (38.26%), followed by Bacteroidetes (37.63%). Differences in relative abundance between the groups were not significant.

At the taxonomic family level, we found Bacteroidaceae to be the most abundant taxon in all patients (non-recurrent, 30.57%; recurrent, 28.02%), followed by Enterobacteriaceae (non-recurrent, 16.77%; recurrent, 13.75%) ( Figure 2 ). When we analysed the taxonomic level of genus, we found that Fusobacterium increased in abundance in recurrent versus non-recurrent disease and that Collinsella, Senegalimassilia, Prevotella and Ruminococcus decreased in abundance in recurrent versus non-recurrent disease (all p<0.05) ( Figure 3 ).

Our analysis of the differences between patients with calprotectin levels >185µg/mg and those with levels <185µg/mg revealed that, as in the recurrent group, Fusobacterium increased and Senegalimassilia, Prevotella and Ruminococcus decreased. We also found other differences such as lower abundance of Lactobacillus, Lactococcus, Streptococcus, Blautia, Dorea, Bacteroides, Faecalicoccus, Clostridium XVIa and Parasutterella in patients with faecal calprotectin >185µg/mg than in those with lower values (all p<0.05) ( Figure 4 ).

We performed a multivariable association study between microbial communities and clinical metadata and found that the genera Senegalimassilia, Alistipes, Blautia and Lactococcus had a negative association with faecal calprotectin concentration and that the genera Bacteroides, Fusobacterium and Dialister had a positive association with faecal calprotectin concentration (all p<0.05) ( Table 8 ).

In terms of similarities between the OTUs of both groups, we found that they shared 1921 OTUs: 1119 unique OTUs in the recurrent group and 2241 unique OTUs in the non-recurrent group.