Low Gut Ruminococcaceae Levels are Associated with Occurrence of Antibiotic-associated Diarrhea

Patients receiving antibiotics often suffer from antibiotic-associated diarrhea (AAD). AAD is of clinical significance as it can result in premature antibiotic discontinuation and suboptimal treatment of infection. The drivers of AAD however, remain poorly understood. We sought to understand if differences in the gut microbiome, both at baseline and during antibiotic administration, would influence the development of AAD. We administered a 3-day course of oral amoxicillin-clavulanate to 30 healthy adult volunteers, and performed a detailed interrogation of their stool microbiome at baseline and up to 4-weeks post antibiotic administration, using 16S rRNA gene sequencing. Lower levels of Ruminococcaceae were significantly and consistently observed from baseline till Day 7 in participants who developed AAD. The probability of AAD could be predicted based on qPCR-derived levels of Faecalibacterium prausnitzii, the most dominant species within the Ruminococcaceae family. Overall, participants who developed AAD experienced a greater decrease in microbial diversity during antibiotic dosing. Our findings suggest that a lack of gut Ruminococcaceae at baseline influences development of AAD. In addition, quantification of F. prausnitzii in stool prior to antibiotic administration may help identify patients at risk of AAD, and aid clinicians in devising individualised treatment regimens to minimise such adverse effects.


INTRODUCTION 64
Ruminococcaceae levels were distinctly different between the two groups, both at baseline 138 and post-antibiotic treatment. At baseline, the AAD group had a lower proportion of 139 Ruminococcaceae (mean 8.4% vs 14.4%, median 7.9% [IQR 4.2-11.9] vs 14.2% .7], Bonferroni-corrected p=0.02, n=30) ( Figure 2B). The AAD group also experienced a 141 consistently lower proportion of Ruminococcaceae compared to the non-AAD group till day 142 7. On average, across the duration of the study, the AAD group had a lower proportion of 143 Ruminococcaceae (mean 7.5% vs 15.3%, median 5.5 [IQR 3.0-10.8] vs 15.6 [IQR 11.8-19.2], 144 Bonferroni-corrected p=2.1e-14, n=197). These differences were not observed with any of the 145 other major taxonomies. 146 We identified potential drivers of AAD at the genus level within the Ruminococcaceae 147 family across the duration of the study, with Faecalibacterium being the most abundant 148 (mean 67.1%, median 66.6 [IQR 56.8-77.7]), followed by Subdoligranulum (mean 11.7%, 149 median 10.4 [IQR 6.5-14.5]) and Ruminococcus (7.7%, median 6.5 [IQR 3.1-10.1]). We 150 observed that the genera Faecalibacterium, Subdoligranulum and Ruminococcus were 151 significantly less abundant in the AAD group across most days between day 0 to day 7. 152 Amoxicillin-clavulanate causes greater gut microbiome diversity loss and community 153 disturbance in the AAD group compared to the non-AAD group 154 We next determined the extent to which the gut microbiota was disrupted during antibiotic 155 treatment and the timescale of recovery, in terms of the abundance of microbial amplicon 156 sequence variants (ASVs) and composition. By day 3, the faecal microbiota from the AAD 157 group was distinctly different from microbiomes sampled at the other timepoints and from 158 the non-AAD group, forming a separate cluster along the first and second axes of a principal-159 coordinate analysis (PCoA) ( Figure 3A). We quantified the microbial diversity within each 160 individual at a given time point ( diversity) and the differences between each individual's 161 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 24, 2021. ; https://doi.org/10.1101/2021.09.23.21263551 doi: medRxiv preprint baseline and post-treatment gut microbiota ( diversity) ( Figure 3B). We observed a greater 162 decrease in diversity in the AAD group, compared to the non-AAD group on days 2 and/or 3 163 ( Figure 3B, P < 0.05, Bonferroni-corrected Mann-Whitney U test). Analysis of the relative 164 abundance of bacterial taxonomic groups at the phylum level supported our finding that the 165 AAD group was more severely impacted than the non-AAD group ( Figure 3C). This 166 difference in diversity between the two groups was driven by a sharp increase of 167 Proteobacteria, and a decrease of Firmicutes and Actinobacteria in the AAD group, as 168 compared to the non-AAD group on days 2 and 3 ( Figure 3C). Individuals in both groups 169 returned to their baseline taxonomy and diversity by day 7, as shown by permutational 170 multivariate analysis of variance (PERMANOVA) ( Figure 3A). 171 Predicting the risk of AAD based on the relative abundance of Ruminococcaceae at 172 baseline 173 While post-treatment changes in the gut microbiome could explain AAD, predicting which 174 patients are at higher risk of developing AAD would be clinically useful, to enable 175 personalization of antibiotic prescription to minimize the incidence of AAD. Through 176 hierarchical clustering of microbiome composition at baseline, the majority of non-AAD 177 individuals ( Figure 4A, blue labels) were grouped into a single cluster (cluster 1), while AAD 178 individuals ( Figure 4A, red labels) were separated into various clusters (non-cluster 1). This 179 clustering suggests that some features could be a potential indicator to identify individuals at 180 risk of AAD. Moreover, it also suggests the presence of common characteristics amongst the 181 non-AAD group, but not for the AAD group. Our findings suggest that individuals who went 182 on to develop AAD could be differentiated from those who did not, by the relative abundance 183 of Ruminococcaceae at baseline. This was evident from PCoA based on Bray-Curtis 184 dissimilarities ( Figure S2). Hence, we sought to determine if we could predict the risk of 185 AAD based on the relative abundance of Ruminococcaceae at the baseline (D0). Upon 186 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 24, 2021. ; https://doi.org/10.1101/2021.09.23.21263551 doi: medRxiv preprint ranking the subjects based on their relative-abundance of Ruminococcaceae, we found that 187 there was a clear separation between the AAD and non-AAD groups at the extremes of 188 relative abundance (<0.05 or >0.16) ( Figure 4B). Next we quantified absolute values of one 189 species under Ruminococcaceae (Faecalibacterium prausnitzii [F. prausnitzii], the most 190 abundant Ruminococcaceae species, using a qPCR assay (14). We found that the gene copies 191 of F. prausnitzii followed a similar trend with 16S rRNA gene relative abundance of 192 Ruminococcaceae (Spearman's ρ = 0.85, p = 3.0e-9) ( Figure 4C-D). We calculated the risk of 193 developing AAD based on the absolute abundance of F. prausnitzii derived using qPCR at 194 baseline. Lower relative abundance of F. prausnitzii at baseline were predictive of risk of 195 AAD. The probability of developing AAD was above 0.7 if F. prausnitzii levels were less 196 than 2.4⨉10 7 GC/µL, and below 0.3 if F. prausnitzii levels were above 8.0⨉10 7 GC/µL 197 ( Figure 4E). 198

DISCUSSION 199
Our study has shed important insights on differences in gut microbiome responses between 200 individuals who developed AAD and those who did not. We found that individuals who 201 developed AAD experienced greater gut microbiome community changes, accompanied by 202 lower diversity and a greater disturbance in abundance across taxonomies. Individuals in the 203 AAD group experienced a sharp increase in Proteobacteria, belonging to the genus 204 Escherichia-Shigella. Looking for taxonomic signatures that could differentiate between the 205 two groups, we found that amongst all bacterial families, gut Ruminococcaceae levels were 206 significantly and consistently different between the two groups. Ruminococcaceae levels 207 were lower both at baseline prior to antibiotic treatment, and up till day 7 post-dose. In fact, 208 ranking of gut Ruminococcaceae levels, or simply F. prausnitzii (the most dominant species 209 . CC-BY 4.0 International license It is made available under a perpetuity. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint Ruminococcaceae is a group of strictly anaerobic bacteria that is present in the colonic 212 mucosal biofilm of healthy individuals (15). Decreased abundance of Ruminococcaceae has 213 been implicated in a number of inflammatory bowel diseases, including ulcerative colitis and 214 Crohn's disease (16-18), inflammatory diseases such as hepatic encephalopathy (19) and has 215 also been associated with C. difficile infection and C. difficile-negative nosocomial diarrhea 216 (20). Ruminococcaceae plays an important role in the maintenance of gut health through its 217 ability to produce butyrate and other SCFAs. These SCFAs are essential carbon and energy 218 sources to colonic enterocytes (21), in the absence of which, functional disorders of the 219 colonic mucosa may occurwhich may manifest in the form of osmotic diarrhea (9). Indeed, 220 supplementation of butyrate and other SCFAs has been shown to reduce colonic 221 inflammation and improve diarrhea in conditions such as inflammatory bowel diseases, 222 irritable bowel syndrome, and diverticulitis (22-24). We thus posit that a lack of 223 Ruminococcaceae resulting in decreased SCFA production may be driving the development 224 of AAD in our cohort of otherwise healthy individuals who received amoxicillin-clavulanate. 225 To date, many others have studied the role of probiotic bacteria such as Lactobacillus, 226 Bifidobacterium, Clostridium, Bacillus and Lactococcus in the development and prevention 227 of AAD (25). In a systematic review and a meta-analysis, administration of such probiotics 228 were associated with a reduced incidence of AAD (26, 27). However, a clinical trial 229 performed on a cohort of 3000 older patients concluded that Lactobacillus and 230 Bifidobacterium probiotic administration was not found to be effective in preventing C. 231 difficile -associated diarrhea (1). Likewise, Lactobacillus reuteri was not effective in 232 preventing AAD in a cohort of 250 children (28, 29). There remains a need to search for 233 probiotic candidates that will be effective against AAD. Our findings suggest that certain 234 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 24, 2021. ; species within the Ruminococcaceae family may be useful as probiotics to prevent AAD, but 235 this will need to be further evaluated in randomised-controlled clinical trials. 236 We have identified clear differences in baseline gut microbial composition that may enable 237 pre-identification of individuals at higher risk of developing AAD. Although our findings at 238 present are only applicable in the context of AAD caused by amoxicillin-clavulanate, our 239 study provides a framework to identify potential drivers of AAD caused by other classes of 240 antibiotics. We acknowledge that a limitation of our study is the relatively small sample size, 241 which may have reduced the statistical power given the variability of inter-individual 242 differences in the gut microbiome. Despite this, we were still able to observe clear and 243 significant differences between the AAD and non-AAD groups. 244 Our findings provide evidence for the first time that baseline differences in the individual's 245 gut microbial composition can influence the risk of developing AAD with certain antibiotics, 246 and would guide the development of point-of-care diagnostics. Being able to pre-identify 247 individuals at increased risk of AAD would aid clinicians in devising an individualised 248 antibiotic regime best suited to the patient that is least likely to result in premature antibiotic 249 discontinuation and suboptimal treatment of infection -An example of such a clinical 250 workflow is presented in Figure 5. In addition, the use of Ruminococcaceae as a prebiotic to 251 prevent AAD in patients who receive amoxicillin-clavulanate also warrants further 252 exploration. Overall, our study provides insights into how the gut microbiome influences 253 development of AAD, and opens a window of opportunity for further research in this area. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 24, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 24, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 24, 2021. ; https://doi.org/10.1101/2021.09.23.21263551 doi: medRxiv preprint (31) to merge pairs of reads when the original DNA fragments are shorter than twice of the 304 reads length. The obtained splicing sequences were called raw tags. Quality filtering were 305 then performed on the raw tags under specific filtering conditions of QIIME (V1.7.0) (32) 306 quality control process. After filtering, high-quality clean tags were obtained. The tags were 307 compared with the reference database (Gold database, 308 http://drive5.com/uchime/uchime_download.html) using UCHIME algorithm (33) to detect 309 chimeric sequences, and then the chimeric sequences were removed to obtain the Effective 310 Tags finally. Sequence analysis and processing of paired-end demultiplexed sequences were 311 performed in Quantitative Insights into Microbial Ecology pipeline (QIIME 2, v 2020.6) (34). 312 Demultiplexed sequences were imported using QIIME 2 "Fastq manifest" format by mapping 313 the sample identifiers to absolute file paths containing sequence information for each sample. 314 PairedEndFastqManifestPhred33V2 format was used. Interactive quality plot and a summary 315 distribution of sequence qualities at each base pair position in the sequence data was 316 visualized to determine input parameters for denoising. DADA2 was used to denoise, 317 dereplicate and filter chimaeras in paired-end sequences to identify all amplicon sequence 318 variants (ASVs), equivalent to 100% Operational Taxonomy Unit (OTUs) (35, 36). Forward 319 and reverse reads were truncated at 200 bases to retain high quality bases respectively. In 320 total, we characterized an average of 150,087 ± 14,526 (mean ± SD) 16S rRNA sequences for 321 197 samples. 322

Microbiome composition and diversity 323
The alpha diversity metrics (Shannon entropy) and beta diversity metrics (Jensen-Shannon 324 Distance) from ASVs were generated via q2-diversity plugin. A sampling depth of 96,935 325 sequences per sample was used and optimal alpha-rarefaction curves were achieved. 326 Taxonomy assignment to ASVs was performed using q2-feature-classifier plugin using a pre-327 trained Naive Bayes classifier against the reference 515F/806R region of sequences in Silva 328 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 24, 2021. ; 138 at 99% OTUs (37). We computed ASV pairwise distances using the Pearson correlation 329 (ASV abundances across 30 subjects at baseline). The resulting distance matrix was 330 subsequently inputted into a hierarchical clustering function ('fcluster'). The linkage 331 approach was set as 'average'. The colour threshold was set to '0·4'. Principal coordinate 332 analysis (PCoA) was performed using the 'scipy' package in python based on the ASV-level 333 Bray-Curtis dissimilarities between the composition of baseline samples. PERMANOVA 334 analysis was calculated using the 'skbio' package in python based on the ASV-level Bray-335 Curtis dissimilarities and 9999 permutations. 336

Calculating predictive probability of developing AAD from baseline abundance 337
To calculate the predictive probability of developing AAD from F. prausnitzii baseline 338 abundance, the absolute abundance of F. prausnitzii was min-max normalized to their 339 transformed value between 0 and 1. The predictive probability of developing AAD was 340 calculated using kernel density estimation with a Gaussian distribution kernel. We assumed 341 that each sample had an independent and identical distribution with a mean at its 342 concentration and a standard deviation, which is the hyperparameter of this model. We 343 aggregated the distribution of every sample from the AAD group and obtained the AAD 344 probability density function (PDF). The same approach was applied to calculating the non-345 AAD PDF. The probability of developing AAD was calculated as is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 24, 2021.  (Table S5). is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint the amplification regions of the primer-probe sets (Table S5) is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 24, 2021. ; https://doi.org/10.1101/2021.09.23.21263551 doi: medRxiv preprint 60°C for 1 min for annealing/extension, 4°C for 5 min and 90 °C for 5 min for signal 402 stabilization). During the process, the ramp rate was set as 2°C/s. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 24, 2021. is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint  is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 24, 2021. ; https://doi.org/10.1101/2021.09.23.21263551 doi: medRxiv preprint that there were no significant differences between day 0 and 7 in both AAD (P = 0.43, N = 24) 474 and non-AAD groups (P = 0.51, N = 24). (B) Within-sample species diversity ( diversity of 475 ASVs, Shannon entropy index) greatly decreased in the AAD group compared to the non-476 AAD group on day 2. The similarity of each individual's gut microbiota to their baseline 477 communities ( diversity of ASVs, Jensen-Shannon distance) greatly decreased in the AAD 478 group compared to the non-AAD group cross days 2-3. Significant difference between the 479 AAD and non-AAD groups are labelled with asterisks (Bonferroni-corrected, two-sided 480 Mann-Whitney U test, P ≤ 0·05, *; P ≤ 0·01, **). (C) A sharp increase of Proteobacteria, and 481 a decrease of Firmicutes and Actinobacteria were observed in the AAD group on days 2 and 482 3. 483 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 24, 2021. ; https://doi.org/10.1101/2021.09.23.21263551 doi: medRxiv preprint Ruminococcaceae relative abundance at baseline among the AAD and non-AAD groups. 493 Each number refers to the individual at baseline. (C) qPCR concentration for the species F. 494 prausnitzii was normalized to 16S rRNA gene copy (Data represent median±IQR range, n=3). 495 Each number refers to the individual at baseline. (D) Correlations between Ruminococcaceae 496 relative abundance and F. prausnitzii median absolute quantification (Spearman's ρ = 0·850, 497 p = 3·0e-9). Line depicts the best linear fit and blue shading the 95% confidence interval of 498 the linear fit. (E) Calculated predictive precision of developing AAD using F. prausnitzii 499 absolute abundance quantified by qPCR assay. 500 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 24, 2021. ; https://doi.org/10.1101/2021.09.23.21263551 doi: medRxiv preprint Figure 5. Risk-stratifying patients at risk of AAD using stool F. prausnitzii qPCR . We 502 propose a workflow to evaluate AAD risk at baseline prior to antibiotic administration. This 503 would enable clinicians to risk-stratify patients, and identify those in whom use of 504 amoxicillin-clavulanate should be avoided due to higher risk of AAD. Although 505 quantification of F. prausnitzii is currently via qPCR, this could potentially be translated into 506 rapid point-of-care diagnostics in furture. This framework could also serve as basis for 507 identifying potential drivers in AAD caused by other classes of antibiotics. (This figure was 508 created using BioRender.com.) 509 510 . CC-BY 4.0 International license It is made available under a perpetuity.
is the author/funder, who has granted medRxiv a license to display the preprint in (which was not certified by peer review) preprint The copyright holder for this this version posted September 24, 2021. ; https://doi.org/10.1101/2021.09.23.21263551 doi: medRxiv preprint