Microbiota-based markers predictive of development of Clostridioides difficile infection

ARTICLE Microbiota-based markers predictive of development of Clostridioides difficile infection Matilda Berkell 1,46, Mohamed Mysara 1,2,46, Basil Britto Xavier 1,46, Cornelis H. van Werkhoven 3, Pieter Monsieurs2,45, Christine Lammens1, Annie Ducher4, Maria J. G. T. Vehreschild 5,6,7, Herman Goossens1, Jean de Gunzburg 4, Marc J. M. Bonten3,8, Surbhi Malhotra-Kumar 1✉& the ANTICIPATE study group* Antibiotic-induced modulation of the intestinal microbiota can lead to Clostridioides difficile infection (CDI), which is associated with considerable morbidity, mortality, and healthcarecosts globally. Therefore, identification of markers predictive of CDI could substantially contribute to guiding therapy and decreasing the infection burden. Here, we analyze the intestinal microbiota of hospitalized patients at increased CDI risk in a prospective, 90-day cohort-study before and after antibiotic treatment and at diarrhea onset. We show that patients developing CDI already exhibit significantly lower diversity before antibiotic treatment and a distinct microbiota enriched in Enterococcus and depleted of Ruminococcus,Blautia, Prevotella and Bifidobacterium compared to non-CDI patients. We find that antibiotic treatment-induced dysbiosis is class-specific with beta-lactams further increasing enterococcal abundance. Our findings, validated in an independent prospective patient cohort developing CDI, can be exploited to enrich for high-risk patients in prospective clinical trials, and to develop predictive microbiota-based diagnostics for management of patients at risk for CDI. https://doi.org/10.1038/s41467-021-22302-0 OPEN 1Laboratory of Medical Microbiology, Vaccine & Infectious Disease Institute, University of Antwerp, Wilrijk, Belgium. 2Microbiology Unit, Interdisciplinary Biosciences, Belgian Nuclear Research Centre, SCK–CEN, Mol, Belgium. 3Julius Center for Health Sciences and Primary Care, University Medical Center Utrecht, Utrecht University, Utrecht, the Netherlands. 4Da Volterra, Paris, France. 5Department of Internal Medicine, Center for Integrated Oncology Aachen Bonn Cologne Duesseldorf, University of Cologne, Cologne, Germany. 6German Centre for Infection Research (DZIF), partner site Bonn-Cologne, Cologne, Germany. 7Department of Internal Medicine, Infectious Diseases, University Hospital Frankfurt, Goethe University Frankfurt, Frankfurt am Main, Germany. 8Department of Medical Microbiology, University Medical Center Utrecht, Utrecht University, Utrecht, the Netherlands. 45 Present address: Department of Biomedical Sciences, Institute of Tropical Medicine, Antwerp, Belgium. 46 These authors contributed equally: Matilda Berkell, Mohamed Mysara, Basil Britto Xavier. *A list of authors and their affiliations appears at the end of the paper. ✉email: [email protected] NATURE COMMUNICATIONS | (2021) 12:2241 | https://doi.org/10.1038/s41467-021-22302-0 | www.nature.com/naturecommunications 1 1234567890():,; C lostridioides difficile is the most common cause of infectious antibiotic-associated diarrhea (AAD) and is the pathogen responsible for the largest number of healthcare-associated infections world-wide1–3.C. difficile infection (CDI) is characterized by watery stool accompanied by toxin-mediated inflammation of the bowel where primary risk factors include hospitalization, age, colonization by toxigenic C. difficile, and most importantly, antibiotic exposure where use of fluoroquinolones (FQNs), clindamycin, carbapenems, cephalosporins, and penicillins combined with beta-lactamase inhibitors (PBLs) are associated with increased CDI risk4–9. However, markers predictive of CDI or AAD development are as yet lacking. Such markers could be utilized to stratify patients into different risk categories and to enrich patient populations for clinical trials assessing preventive measures against or therapeutics for CDI. Patients suffering from CDI, as well as from other forms of AAD harbor a disrupted intestinal microbiota characterized by reduced diversity and elevated levels of Enterococcus alongside reduced levels of members of the Bacteroidetes phylum, the Lachnospiraceae and Ruminococcaceae families, and Prevotella spp. during disease manifestation10–20. 16S rRNA gene profiling provides a useful method for studying changes in microbial composition. More importantly, it might allow identification of microbial markers predictive of the risk of CDI development. Although the microbial composition at CDI onset has been wellstudied, the pre-CDI microbiota remains largely unexplored. To our knowledge, only one prospective study has investigated microbial composition as a potential predictor of CDI18,21. This single-center Canadian study18, part of a larger clinical study assessing CDI risk factors21, demonstrated that the absence or reduction in Clostridiales, namely members of Clostridiales Incertae Sedis XI, in the intestinal microbiota was associated with an increased risk of CDI18. Collateral damage on the microbiota induced by antibiotic treatment has been further shown to result in reduced alpha diversity, as well as an increase in the presence of antibiotic resistance genes, and long-lasting effects ranging from weeks to years depending on the antibiotic22–25. Several small studies in healthy adults have investigated specific short-term changes induced by antibiotics, often in combination with multiple compounds25–28. None of these, however, link antibiotic-induced dysbiosis to development of CDI. In this multi-center, observational, prospective study, we investigated the intestinal microbiota of hospitalized patients aged 50 years and above in 34 hospitals across six European countries prior to antibiotic therapy with the aim of identifying robust microbial markers predictive of CDI and AAD development utilizing 16S rRNA gene profiling combined with a highresolution sequence typing approach. Furthermore, longitudinal perturbations induced by different antibiotic classes were studied in relation to the CDI-specific and AAD-specific microbiota. Here we show that patients developing CDI exhibit significantly lower microbial diversity prior to antibiotic treatment and a distinct microbiota enriched in Enterococcus and depleted of Ruminococcus,Blautia, Prevotella, and Bifidobacterium spp. compared to non-CDI patients. By validating our findings in an independent prospective cohort, we show that these microbial markers are applicable in a geographically diverse patient population. Further, we show that broad-spectrum antibiotic treatment induces classspecific dysbiosis with beta-lactam antibiotics particularly increasing enterococcal abundance. Collectively, our findings can be exploited to enrich for patients at high risk of CDI development in prospective clinical trials and to develop predictive microbiota-based diagnostics for management of patients at risk for CDI. Results Study population and design. The ANTICIPATE study recruited 1007 patients of which 1002 provided rectal swabs at D1 and 848 at D6 (Fig. 1). Of the 135 patients who developed diarrhea during a period of 90 days following study inclusion, 15 were diagnosed with CDI, as described in van Werkhoven et al.29 Stool samples from 33 of the 135 patients with diarrhea, including 6 patients diagnosed with CDI, collected at the occurrence of the first diarrheal episode (S1) were available for analysis (Fig. 1, Table 1). Following 16S rRNA gene sequencing, 945 and 775 rectal swab samples collected at D1 and D6, respectively, and S1 samples from 32 diarrheic patients (including stools from 6 CDI patients) passed data pre-processing and quality filtering criteria and were analyzed further. Demographics of the 945 patients included in this study are detailed in Table 1. In the studied population, 390 patients received antibiotic treatment with drugs belonging to one of the following classes of broad-spectrum antibiotics: PBLs, other beta-lactam antibiotics (OBLs), and FQNs (Table 1). Influence of baseline characteristics—gender, country of origin, and age by decades—on the fecal microbiome of the 945 patients was assessed at D1 (Supplementary Table 1). Certain differences in both alpha and beta diversity were observed in the microbiota, namely when patients were stratified by countries of origin (Supplementary Results, Supplementary Table 1, and Supplementary Fig. 1). However, the aim of this study was to identify biomarkers predictive of CDI that were robust enough to transcend country, diet, and gender differences in the intestinal microbiota. Therefore, patients were not stratified by baseline characteristics. High-resolution 16S rRNA profiling defines C. difficile prevalence. Upon OTU clustering using a 97% cut-off, a single OTU classified as Clostridium XI was identified in 556 of the 945 (58.8%) patient samples at D1. High-resolution analysis of this OTU by oligotyping revealed two distinct types of raw reads, one corresponding to C. difficile and the other to Clostridium bartlettii (also known as Intestenibacter bartlettii30), a known anaerobic occupant of the human intestine31 (Supplementary Results). Oligotyping identified C. difficile in 51 patients (5.40%) at D1 with an average relative abundance of 0.82% (SD =2.67, Fig. 2a). Of these, four patients developed CDI within 1–3 days of the D1 sampling (i.e., study inclusion). Relative abundance of C. difficile at D1 was slightly higher in the four CDI patients (1.72%, SD =2.18) than in the 47 patients who did not develop CDI within the 90-day study period (0.74%, SD =2.70). All four CDI patients underwent treatment with metronidazole and/or vancomycin between the D1 and D6 sampling time-points and in two patients, C. difficile was not detected at D6. Of the 945 studied patients, 737 provided both D1 and D6 samples. C. difficile was detected in 50 patients at D6 (6.78%) at a 2.5-fold higher average relative abundance (2.02%, SD = 6.14) compared to D1 (Fig. 2b). Only 13 patients carried C. difficile at both sampling time-points. C. difficile was detected in all six analyzed stools originating from patients with confirmed CDI at an average relative abundance of 0.26% (SD =0.29), and in two stools from AAD patients that tested negative for CDI with a relative abundance of 0.01% and 0.02%. Distinct microbial markers predictive of CDI and AAD development. Alpha diversity at D1 was lower in patients developing CDI, compared to those developing AAD or ND (Fig. 3a, b, p≤0.049). AAD patients also had lower Chao1 indices compared to ND patients whereas a decreasing trend was observed in Shannon indices, however, not sufficient to be significant (Fig. 3a, b, p=0.017 and p=0.087, respectively). ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-22302-0 2NATURE COMMUNICATIONS | (2021) 12:2241 | https://doi.org/10.1038/s41467-021-22302-0 | www.nature.com/naturecommunications Furthermore, at D1, the microbiota composition differed between patients developing CDI and the other two groups (Fig. 3c, p≤0.025), and between AAD patients and ND patients (p=0.002), with the most pronounced differences observed between CDI and ND patients (p< 0.001). Microbial composition at D1 was compared between the three patient groups (CDI, AAD, and ND) for identification of predictive microbial markers. Patients who developed CDI harbored elevated levels of the pathobiont Enterococcus (Fig. 3c, Table 2, and Supplementary Fig. 4). AAD patients harbored elevated levels of the family Clostridiales Incertae Sedis XI, Ruminococcus (R. torques, R. faecis, R. lactaris, and Clostridium glycyrrhizinilyticum), and Blautia spp. (B. wexlerae,B. obeum, and B. faecis, Supplementary Tables 2–6), whereas ND patients primarily exhibited a higher abundance of Blautia luti,Prevotella, uncultured Clostridiales, Porphyromonas (P. bennonis,P. asaccharolytica,P. uenonis), Campylobacter, and Ezakiella, in addition to members of the Bifidobacteriales family. When comparing CDI versus non-CDI (including AAD, ND, and non-CDI tested diarrheic cases) patients at D1, several OTUs showed lower abundance among CDI patients, which included six Bifidobacterium spp. and three Blautia spp. B. wexlerae,B. obeum, and B. faecis (Supplementary Tables 2, 3, and 6). The OTU with the strongest association at D1 with the subsequent development of CDI, classified as Enterococcus, consisted of two types of raw reads distinguished by seven nucleotides (constituting a 98.4% sequence homology). One was classified as one or multiple species of the group consisting of E. hirae,E. villorum,E. ratti,E. faecium,orE. durans (oligotype 1, 63.5% of all OTU reads) and the other as E. faecalis (oligotype 2, 36.5% of all OTU reads) both with 100% sequence identity and length. In order to further elucidate the specific species constituting oligotype 1, we performed shotgun metagenomic sequencing on D1 samples from patients developing CDI. Oligotype 1 was primarily constituted by E. faecium and E. villorum corresponding to average relative abundances of 57% and 32%, respectively, within the oligotype (Supplementary Table 7). Similarly D1 samples from non-CDI patients underwent shotgun sequencing and two of the most prevalent OTUs classified as Bifidobacterium spp. and Blautia spp. were delineated to be Bifidobacterium adolescentis,B. catenulatum,B. dentium, and Blautia wexlerae and B. obeum, respectively (Supplementary Table 7). OTUs classified as Ruminococcus and Alistipes were also abundant in non-CDI patients and were further speciated as, R. torques and R. lactaris as the most prevalent of the oligotypes together with R. bromii, and A. onderdonkii and A. finegoldia, respectively. Validation of microbial markers predictive of CDI. To verify the generalizability and potential of the identified microbiota-based biomarkers predictive of CDI, we analyzed a previously published dataset18. This dataset, generated from a Canadian cohort of elderly patients who developed CDI with matched non-diarrheic patient Fig. 1 Patient and sample flow in the study. The flow chart provides an overview of participating patients in each processing step, number of samples collected at each timepoint, and reasons for sample exclusion or non-collection. D1: rectal swab sample collected upon study enrollment. D6: rectal swab collected ~6 days after initiation and at the end of antibiotic treatment. S1: stool sample collected at the first occurrence of diarrhea (variable time-point). AAD: patients with non-C. difficile antibiotic-associated diarrhea. CDI: patients with confirmed C. difficile infection. ND: non-diarrheic patients. PBL: penicillin +beta-lactamase inhibitor. OBL: other beta-lactamase antibiotics. FQN: fluoroquinolones. NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-22302-0 ARTICLE NATURE COMMUNICATIONS | (2021) 12:2241 | https://doi.org/10.1038/s41467-021-22302-0 | www.nature.com/naturecommunications 3 controls served as an independent validation of the microbial markers identified in this study to be predictive of CDI. Briefly, prospective fecal samples had been collected from patients developing CDI (n=25) as well as from age-matched and gendermatched non-diarrheic control patients (n=25), and resulted in 24 and 25 baseline samples from patients developing CDI and from non-CDI controls, respectively, after processing. In concordance with our results, baseline samples from patients who subsequently developed CDI (median time to development, 5 days) in the validation dataset also harbored elevated levels of a single enterococcal OTU (Table 3). Also, in concordance with our data, the OTU classified as Enterococcus spp. corresponded to two oligotypes, one classified as E. faecalis (73.81% overall relative abundance) and the other as E. faecium,E. hirae,E. villorum,E. ratti,orE. durans (26.19% overall relative abundance, Supplementary Table 7). It was further noted that 91.7% (n=22) of the patients who proceeded to develop CDI within this dataset (n=24) were colonized by C. difficile (OTU classified as Clostridium XI) upon study enrollment. Patients who did not develop CDI in the validation dataset harbored elevated levels of Ezakiella,Odoribacter,andRuminococcus spp. together with uncultured Clostridiales (Table 3). These also corresponded to the species identified by oligotyping in our data (Supplementary Tables 6 and 8). In addition, several taxa were also unique in the two datasets, especially those associated with non-CDI patients in our study were not identified in the validation dataset, and vice versa (Supplementary Table 9). These differences are likely due to variable sequencing technologies and depths in either study (15,000 reads in this study compared to 2000 in the validation dataset18). Finally, one OTU classified as Finegoldia spp. was associated with non-CDI patients in the validation dataset whereas our study revealed it to be linked to patients developing CDI when compared with non-CDI patients (Table 3). However, the non-CDI group in our study is constituted by (non-CDI) AAD as well as ND patients. Further assessment of taxa linked to CDI and AAD in our study (Supplementary Table 5) revealed Finegoldia spp. to be primarily associated with patients developing AAD rather than CDI, thereby correlating with the findings in the validation dataset. Antibiotic treatment induces class-specific microbial dysbiosis. Samples from 390 patients were available for longitudinal analysis of class-specific antibiotic-induced intestinal dysbiosis. Of these, 194, 133, and 63 patients had received treatment with PBL, OBL, and FQN, respectively, between the D1 and D6 sampling timepoints (Table 1). Comparison of D1 and D6 samples from the 390 patients showed that treatment with all three antibiotic classes induced a decrease in alpha diversity together with a shift in beta diversity of the intestinal microbiota (p< 0.001 for both). Further investigation into specific changes induced by the individual antibiotic classes (PBL, OBL, and FQN) revealed a decrease in both Shannon (p≤0.007, Fig. 4a) and Chao1 diversity indices (p≤0.011, Fig. 4b) at D6, together with a change in microbiota composition (p< 0.001). Comparison of beta diversity distances between D1 and D6 samples within each antibiotic class showed that patients treated with OBLs exhibited the largest alterations in microbial composition compared to those receiving PBLs and FQNs (Fig. 4c). Next, we compared the microbial compositional differences between the D1 and D6 samples within each of the three antibiotic classes. Treatment with all classes resulted in alterations within the four most dominant phyla of the intestinal microbiota; Firmicutes, Bacteroidetes, Actinobacteria, and Proteobacteria (Fig. 4d, Supplementary Table 10). The majority of these observed changes affected members of the Clostridia class. Treatment with beta-lactams (PBL and OBL) primarily affected members of the Firmicutes phylum where Clostridiales (primarily Lachnospiraceae members, especially after OBL treatment), and Lactobacillales were the major affected orders. FQN treatment resulted in alterations within the same taxa asbeta-lactamtreatment,aswellasotherssuchastheclassBacteroidia, where the Porphyromonadaceae and Prevotellaceae families were most affected. Although each antibiotic class resulted in its own unique dysbiotic microbiota after treatment, some of the observed changes were also common (Fig. 4d, Supplementary Table 10). All three classes reduced Streptococcus spp. and increased relative abundances of Sphingomonas spp. after treatment (Supplementary Results). Beta-lactam treatment (PBL and OBL) resulted in a remarkable increase in Enterococcus and Clostridium XIVa spp. Additionally, PBL and FQN treatment both resulted in reductions in uncultured Clostridiales, Anaerococcus and Peptoniphilus spp., as well as in Porphyromonas and Prevotella spp., both members of the Bacteroidales order, and in Campylobacter spp. Similarly, OBL and FQN treatment both resulted in a reduction in Escherichia/Shigella spp. However, some classes resulted in an Table 1 Patient demographics of the analyzed study population. Total number of patients 945 Age (median [IQR]) 70 [61–79] Male gender (%) 557 (58.9) Myocardial infarction (%) 78 (8.25) Congestive heart failure (%) 134 (14.2) Peripheral vascular disease (%) 143 (15.1) Cerebrovascular disease (%) 81 (8.57) COPD (%) 144 (15.2) Connective tissue disease (%) 53 (5.61) Peptic ulcer disease (%) 50 (5.29) Diabetes mellitus (%) 268 (28.4) Moderate to severe chronic kidney disease (%) 127 (13.4) Hemiplegia (%) 16 (1.69) Leukemia (%) 55 (5.82) Malignant lymphoma (%) 68 (7.20) Solid tumor (%) 208 (22.0) Liver disease (%) 88 (9.31) AIDS (%) 10 (1.06) Intestinal obstruction (%) 5 (0.53) Inflammatory bowel disease (%) 14 (1.48) Other non-specified comorbidities (%) 538 (56.9) Has history of CDI (%) 14 (1.48) Developed CDI within study period (%) 14 (1.48) Developed AAD within study period (%) 64 (6.77) Country of origin France (%) 210 (22.2) Germany (%) 145 (15.3) Greece (%) 85 (8.99) The Netherlands (%) 14 (1.48) Romania (%) 184 (19.5) Spain (%) 307 (32.2) Antibiotic treatment received from D1 to D5 Penicillin +beta-lactamase inhibitor (PBL, %) 194 (20.5) Other beta-lactam antibiotics (OBL, %) 133 (14.1) Fluoroquinolones (FQN, %) 63 (6.67) Combination therapy and other (%) 347 (36.7) Patients without a D6 sample (%) 208 (22.0) IQR interquartile range, COPD chronic obstructive pulmonary disease, AIDS acquired immune deficiency syndrome, AAD patients with non-C. difficile antibiotic-associated diarrhea, CDI patients with confirmed C. difficile infection, D1 rectal swab sample collected upon study enrollment, D6 rectal swab collected ~6 days after initiation and at the end of antibiotic treatment. ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-22302-0 4NATURE COMMUNICATIONS | (2021) 12:2241 | https://doi.org/10.1038/s41467-021-22302-0 | www.nature.com/naturecommunications a b Fig. 2 C. difficile and C. bartlettii carriage in the analyzed population at D1 and D6. Oligotyping revealed diversity within the OTU classified as Clostridium XI wherein reads were divided into C. difficile (green) and C. bartlettii (blue). aRectal swab samples collected at the time of study enrollment (D1), and ~6 days later at the end of treatment (D6) harbored reads classified as both C. difficile and C. bartlettii. Patients that were defined as C. difficile carriers due to the presence of C. difficile reads after oligotyping showed no clear link to clinical outcome. bSimilar carriage rates of the OTU Clostridium XI were identified in patients both at D1 (5.40%, n=51) and D6 (6.78%, n=50) with varying relative abundances of C. difficile and C. bartlettii reads as demonstrated in the Circos plot. Patients at D6 generally show higher relative abundance of C. difficile than of C. bartlettii compared to D1. D1: rectal swab sample collected upon study enrollment. D6: rectal swab collected ~6 days after initiation and at the end of antibiotic treatment. AAD: patients with non-C. difficile antibiotic-associated diarrhea. CDI: patients with confirmed C. difficile infection. ND: non-diarrheic patients. NA: patients without known CDI status and/or early study termination. OTU: operational taxonomic unit. NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-22302-0 ARTICLE NATURE COMMUNICATIONS | (2021) 12:2241 | https://doi.org/10.1038/s41467-021-22302-0 | www.nature.com/naturecommunications 5 opposing effect of the affected taxa; OBL treatment reduced Staphylococcus spp., whereas the same taxon was found to increase after FQN treatment. Similarly, PBL treatment reduced Dialister spp., whereas FQN treatment resulted in an increase. Additionally, OBL treatment resulted in a reduction of Blautia spp. among others. Comparison of the D6 samples of patients receiving any of the three antibiotic classes showed dysbiotic profiles that were classspecific(p< 0.001 overall, and ≤0.006 for the pairwise comparisons). Patients treated with OBLs harbored the highest levels of Enterococcus spp. at D6 together with elevated levels of Prevotella spp., whereas PBL-treated patients harbored elevated levels of Fig. 3 Characterization of microbial diversity in baseline (D1) samples. a CDI patients (n=14, brown) display distinctly lower alpha diversity expressed by the Shannon index compared to AAD (n=64, blue, p=0.037) and ND patients (n=669, green, p=0.005) at D1. AAD patients similarly display lower Shannon diversity compared to ND patients, however not sufficient to be statistically significant (p=0.087). bBoth CDI and AAD patients display lower diversity expressed by the Chao1 index compared to ND patients (p=0.001 and 0.017, respectively) at D1. Compared to patients who develop AAD, CDI patients display lower Chao1 indices (p=0.049). cCladogram generated by LEfSe demonstrating significantly higher abundances of Actinobacteria, Alphaproteobacteria, and Enterococcus spp. in the gut microbiota of CDI patients at baseline (D1) compared to AAD and ND patients. The cladogram shows distinctly abundant taxa of interest. For more details, see Supplementary Fig. 4. Alpha diversity indices were compared using the non-parametric two-sided Wilcoxon rank sum test followed by Bonferroni correction of p-values. Box plots indicate median (middle line), 25th, 75th percentile(box),and5thand95thpercentile(whiskers)aswellas outliers (gray single dots). AAD: patients with non-C. difficile antibiotic-associated diarrhea. CDI: patients with confirmed C. difficile infection. ND: non-diarrheic patients. LEfSe: Linear discriminant analysis effect size. LDA: linear discriminant analysis score. *p<0.05. **p≤0.01. ***p≤0.001. ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-22302-0 6NATURE COMMUNICATIONS | (2021) 12:2241 | https://doi.org/10.1038/s41467-021-22302-0 | www.nature.com/naturecommunications Escherichia/Shigella compared to patients treated with other antibiotics (Table 4). Similar to OBL-treated patients, patients receiving PBLs also harbored increased levels of Prevotella spp. at D6, but distinct members of this genus compared to those observed after OBL treatment. FQN treatment resulted in a higher relative abundance of members of the Lachnospiraceae and Ruminococcaceae families as well as Blautia spp. (Table 4). Large alterations in microbial composition in AAD and CDI patients. To better understand associations between antibioticinduced perturbations in the intestinal microbiota and the occurrence of diarrhea, we determined alterations in alpha and beta diversity between D1, D6, and at the time of first diarrheal onset in 32 patients (S1; AAD n=26, CDI n=6). The median time to S1 was 6 days (Interquartile range [IQR]: 3–24) for patients with AAD. A decreasing trend in overall diversity described by the Shannon index was observed in patients with AAD between D1, D6, and S1 samples (Fig. 5a, p≤0.018), indicating that the dysbiosis induced by antibiotic treatment worsened at the onset of diarrhea. Microbial richness described by the Chao1 index revealed a slight decrease between D1 and D6 as well as between D1 and S1 (Friedman rank sum: p=0.054, Fig. 5b). Additionally, multi-dimensional scaling (MDS) revealed a shift in microbial composition between different time-points with the largest differences observed between D1 and S1 (Fig. 5c), primarily due to decreasing abundances of the Proteobacteria phylum, and the Clostridia and Bacteroidia classes. Members of the Clostridiales Incertae Sedis XI family were significantly reduced, together with Prevotella,Escherichia/Shigella, and Finegoldia amongst others (Fig. 5d, Supplementary Fig. 5). Further, a remarkable shift in members of the Firmicutes phylum was observed between D1 and S1 samples of AAD patients: from dominance of Clostridia to an increasing proportion of Bacilli, wherein Lactobacillales increased in relative abundance together with Enterococcus,Streptococcus,Lactobacillus, and Akkermansia. Interestingly, the increase in Enterococcus in the AAD-associated microbiota (S1 samples) was due to the same OTU that was identified in this study as a predictive marker of CDI development and that also increased after PBL and OBL treatments. Due to the low number of stool samples obtained from patients with ongoing CDI, the alterations in microbial composition observed at the occurrence of CDI (S1) compared to that observed at baseline (D1) were less well-defined. However, many similarities were observed between the AAD and CDI microbiota (Supplementary Fig. 6, Supplementary Results). At the occurrence of diarrhea, both the CDI and AAD microbiota were vastly dominated by Firmicutes, whereas the abundance of Bacteroidetes and Proteobacteria was reduced. In addition, both the CDI and AAD microbiota (S1 samples) showed elevated levels of Enterococcus and a drastic decrease in Prevotella spp. Discussion In this prospective study, we performed longitudinal characterization of the intestinal microbiota of 945 hospitalized patients from six European countries included in the ANTICIPATE study based on known risk factors for development of CDI. Our primary aim was to identify specific microbial markers or patterns in the intestinal microbiota predisposing to CDI and AAD that could be detected in asymptomatic patients at the time of hospital Table 2 Identification of microbiota-based markers at D1. Group OTU LDA p-value Genus AAD Otu9 4.25 0.0423 Finegoldia Otu30 3.85 0.0280 Blautia Otu31 3.66 0.0207 Ruminococcus2 Otu106 3.32 0.0131 Uncultured Clostridiales Otu165 2.66 0.0402 Uncultured Lachnospiraceae Otu300 2.81 0.0054 Odoribacter CDI Otu1 4.29 0.0002 Enterococcus Otu143 3.75 0.0421 Phenylobacterium Otu163 3.28 0.0455 Dorea Otu359 2.87 0.0169 Oscillibacter ND Otu56 3.79 0.0048 Porphyromonas Otu69 3.59 0.0047 Porphyromonas Otu648 3.18 0.0035 Blautia Otu2540 3.37 0.0401 Ruminococcus Otu127 3.33 0.0001 Oscillibacter Otu137 3.11 0.0354 Uncultured Lachnospiraceae Otu1026 2.86 0.0277 Uncultured Lachnospiraceae Otu974 3.06 0.0003 Roseburia Otu262 2.58 0.0055 Butyricoccus Otu294 2.85 0.0309 Uncultured Lachnospiraceae Otu399 2.94 0.0006 Uncultured Ruminococcaceae The microbiota of patients developing CDI (n=14) and AAD (n=64), and ND patients (n= 699) was compared to identify predictive biomarkers at D1. Distinctly abundant OTUs associated with each condition were identified using linear discriminant analysis effect size (LEfSe, LDA > 2.0). AAD: patients with non-C. difficile antibiotic-associated diarrhea. For more details, see Fig. 3and Supplementary Fig. 4. CDI: patients with confirmed C. difficile infection, ND: non-diarrheic patients, D1: rectal swab sample collected upon study enrollment, LDA: linear discriminant analysis score, OTU: Operational taxonomic unit. Table 3 Validation of identified microbiota-based markers predictive of CDI. OTU OTU classification ANTICIPATE dataset Validation dataset Patient group Average relative abundance (%) LDA p-value Patient group Average relative abundance (%) LDA p-value Otu1 Enterococcus CDI 1.86 4.27 0.0001 CDI 6.40 4.41 0.0037 Otu9* Finegoldia CDI 4.18 4.32 0.0200 Non-CDI 2.76 3.95 0.0009 Otu31 Ruminococcus2 Non-CDI 1.32 3.53 0.0059 Non-CDI 2.04 3.70 0.0302 Otu64 Ezakiella Non-CDI 0.77 3.59 0.0320 Non-CDI 1.17 3.79 0.0190 Otu106 Uncultured Clostridiales Non-CDI 0.34 3.18 0.0117 Non-CDI 1.09 4.06 0.0006 Otu300 Odoribacter Non-CDI 0.20 2.79 0.0496 Non-CDI 0.39 3.64 0.0022 Microbiota-based biomarkers identified in the main study were validated by conducting biomarker identification in an independent patient cohort18. Potential biomarkers associated with CDI (n=24) and non-CDI (n=25) patients were identified in the validation cohort. Distinctly abundant OTUs associated with each condition were identified using linear discriminant analysis effect size (LEfSe, LDA > 2.0). OTU9(*) classified as Finegoldia was associated with patients developing CDI (n=14) when comparing with non-CDI patients (n=733) in the main study. The non-CDI group in our study is constituted by (non-CDI) AAD as well as ND patients, and further investigation (Supplementary Tables 3 and 5) revealed this OTU to be strongly associated with patients developing AAD at D1. AAD: patients with non-C. difficile antibiotic-associated diarrhea, CDI: patients with confirmed C. difficile infection, ND: non-diarrheic patients, Non-CDI: patients with non-C. difficile diarrhea or non-diarrheic patients, D1: rectal swab sample collected upon study enrollment, LDA: linear discriminant analysis score, OTU: operational taxonomic unit. NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-22302-0 ARTICLE NATURE COMMUNICATIONS | (2021) 12:2241 | https://doi.org/10.1038/s41467-021-22302-0 | www.nature.com/naturecommunications 7 admission. We also characterized the changes associated with antibiotic treatment potentially impacting the subsequent risk of CDI and AAD development. This was achieved using 16S rRNA profiling combined with a high-resolution sequence typing approach (oligotyping), and when required, further delineation of species using shotgun metagenomic sequencing. At the start of antibiotic treatment, intestinal microbial diversity was lowest in patients who subsequently developed CDI (median time to development: 18 days, range 1–78) and highest in patients not developing AAD or CDI. In the low-diversity microbiota of patients subsequently developing CDI, pathobionts belonging to the Enterococcus genus were present in significantly ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-22302-0 8NATURE COMMUNICATIONS | (2021) 12:2241 | https://doi.org/10.1038/s41467-021-22302-0 | www.nature.com/naturecommunications higher abundances compared to AAD or ND patients. Higher abundance was attributed to a single OTU classified as Enterococcus, and constituted primarily by E. faecium and E. faecalis, which are also the two most frequently occurring enterococci in the human intestine32. We further identified the same OTU in higher abundance in diarrheic stool samples of patients with CDI (Supplementary Fig. 6, Supplementary Results), as has also been repeatedly observed in previous studies analyzing stools from patients with active CDI10,12–17. Baseline samples of CDI patients also showed remarkably low abundances of Blautia (mainly B. wexlerae,B. obeum, and B. luti), Ruminococcus (R. torques,R. lactaris, and R. bromii), Porphyromonas (P. bennonis and P. uenonis),Bifidobacteria (B. adolescentis,B. catenulatum, and B. dentium), Odoribacter, and Ezakiella spp. in comparison to non- CDI patients. In order to validate our results in independent patient cohorts, we screened the literature for studies in CDI patients with a prospective sampling design, and identified one study that had investigated the microbial composition of the intestinal microbiota in patients prior to CDI development18. This dataset, generated from a Canadian cohort of elderly patients who developed CDI and from non-diarrheic controls served as an independent validation to verify the generalizability and predictive potential of the microbiota-based markers identified here. Reanalysis of this decade-old Canadian dataset, which was generated using a different sequencing technology than that utilized in this study, also revealed a low-diversity microbiota with elevated levels of Enterococcus spp. (mainly E. faecium and E. faecalis) and depleted of Ruminococcus,Ezakiella, and Odoribacter spp. 5 days prior to development of CDI in this cohort. These markers were dichotomized, developed into predictive OTU ratios or relative abundance thresholds, and after assessment in competing event models, the best predictive ratio was that of Enterococcus relative to Ruminococcus predictive of a 5- fold higher risk of CDI in our dataset. The best abundance-based OTU model utilized relative abundances of Enterococcus relative to Alistipes and was similarly found to be predictive of a 5-fold higher CDI risk. Both markers predicted an increased CDI risk in the Canadian cohort of 4.6 or 6.2 times, respectively, as described in van Werkhoven et al.29. Remarkably, treatment with beta-lactams, both PBLs and OBLs, was also associated with an increased relative abundance of the enterococcal OTU. Enterococcus spp. are intrinsically resistant to most cephalosporins and clindamycin, have reduced sensitivity to staphylococcal-targeting penicillins and carbapenems, and clinical strains are frequently resistant to FQNs32. Since the 1950s, ampicillin has been the first choice of treatment of enterococcal infections33. Yet, since 2000, a specific clade of E. faecium has become dominant in healthcare settings worldwide and is characterized by, amongst others, resistance to ampicillin33,34. Fig. 4 Longitudinal analysis of the impact of antibiotic therapy on the intestinal microbiota. Microbial diversity was compared prior to broad-spectrum antibiotic treatment at D1 (green) and after treatment at D6 (purple) following treatment with PBLs (n=194), OBLs (n=133), and FQNs (n=63). aA distinct reduction was observed in Shannon diversity in patients treated with all antibiotic classes (p=8.98*10−6,p=2.06*10−5, and p=0.007, respectively). bSimilarly, distinct reductions in Chao1 diversity was observed following treatment with all antibiotic classes (p=0.011, p=0.001, and p=9.26 × 10−5, respectively). cTreatment with each antibiotic class resulted in a shift in microbial composition illustrated by the Jaccard distances between the D1 and D6 samples. dThe heatmap illustrates distinctly abundant taxa (LDA > 3.0) identified using LEfSe. For more details, see Supplementary Table 10. Alpha diversity indices were compared using the paired two-sided non-parametric Wilcoxon signed rank test. Box plots in indicate median (middle line), 25th, 75th percentile (box), and 5th and 95th percentile (whiskers) as well as outliers (gray single dots). PBL: penicillin +betalactamase inhibitor. OBL: other beta-lactam antibiotics. FQN: fluoroquinolones. LDA: Linear discriminant analysis score. D1: rectal swab sample collected upon study enrollment. D6: rectal swab collected ~6 days after initiation and at the end of antibiotic treatment. LEfSe: linear discriminant analysis effect size. LDA: linear discriminant analysis score. OTU: operational taxonomic unit. *p< 0.05, **p≤0.01, ***p≤0.001. Table 4 Characterization of class-specific differences in microbial composition after broad-spectrum antibiotic treatment. Antibiotic class OTU LDA p-value Genus PBL Otu24 3.97 0.002 Prevotella Otu2 3.85 0.000 Escherichia/Shigella Otu1392 3.48 0.004 Clostridium_XlVa Otu56 3.14 0.019 Porphyromonas OBL Otu1 4.38 0.000 Enterococcus Otu21 3.92 0.018 Uncultured Lachnospiraceae Otu18 3.83 0.004 Prevotella Otu48 3.79 0.000 Prevotella Otu68 3.63 0.045 Alistipes Otu167 3.51 0.000 Uncultured Erysipelotrichaceae FQN Otu30 4.06 0.000 Blautia Otu43 3.92 0.000 Collinsella Otu117 3.81 0.000 Roseburia Otu137 3.78 0.000 Uncultured Lachnospiraceae Otu49 3.60 0.012 Uncultured Ruminococcaceae Otu3 3.57 0.000 Staphylococcus Otu75 3.55 0.020 Anaerococcus Otu163 3.52 0.000 Dorea Otu527 3.47 0.000 Coprococcus Otu1026 3.45 0.000 Uncultured Lachnospiraceae Otu22 3.38 0.000 Lactobacillus Otu1086 3.29 0.020 Uncultured Ruminococcaceae Otu97 3.26 0.000 Anaerostipes Otu2858 3.25 0.000 Uncultured Lachnospiraceae Otu165 3.12 0.000 Uncultured Lachnospiraceae Otu262 3.10 0.000 Butyricicoccus Otu399 3.10 0.046 Uncultured Ruminococcaceae Otu315 3.07 0.000 Uncultured Lachnospiraceae Otu3495 3.05 0.001 Fusicatenibacter Otu252 3.03 0.011 Uncultured Ruminococcaceae The microbiota of patients treated with PBLs (n=194), OBLs (n=133), and FQNs (n=63) was compared at D6 to identify the class-specific impact of broad-spectrum antibiotic treatment. Distinctly abundant OTUs associated with each timepoint were identified using linear discriminant analysis effect size (LEfSe, LDA > 3.0). D6: rectal swab collected ~6 days after initiation and at the end of antibiotic treatment. PBL: penicillin +beta-lactamase inhibitor. OBL: other beta-lactamase antibiotics. FQN: fluoroquinolones. LDA: linear discriminant analysis score. OTU: operational taxonomic unit. NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-22302-0 ARTICLE NATURE COMMUNICATIONS | (2021) 12:2241 | https://doi.org/10.1038/s41467-021-22302-0 | www.nature.com/naturecommunications 9