Gastroenterology

Inhibiting Growth of Clostridioides difficile by Restoring Valerate, Produced by the Intestinal Microbiota --Manuscript Draft--

Manuscript Number: GASTRO-D-18-00449R1 Full Title: Inhibiting Growth of Clostridioides difficile by Restoring Valerate, Produced by the Intestinal Microbiota Article Type: Basic - Alimentary Tract Corresponding Author: Julian Marchesi Cardiff University Cardiff, UNITED KINGDOM Corresponding Author Secondary Information: Corresponding Author's Institution: Cardiff University Corresponding Author's Secondary Institution: First Author: Julie A. K. McDonald, PhD First Author Secondary Information: Order of Authors: Julie A. K. McDonald, PhD Benjamin H. Mullish, MB BChir Alexandros Pechlivanis, PhD Zhigang Liu, PhD Jerusa Brignardello, MSc Dina Kao, MD Elaine Holmes, PhD Jia V. Li, PhD Thomas B Clarke, PhD Mark R. Thursz, MD Julian R. Marchesi, PhD Order of Authors Secondary Information: Abstract: Background & Aims: Faecal microbiota transplantation (FMT) is effective for treating recurrent Clostridioides difficile infection (CDI), but there are concerns regarding its long-term safety. Understanding the mechanisms of the effects of FMT could help us design safer, targeted therapies. We aimed to identify microbial metabolites that are important for C difficile growth.

Methods: We used a CDI chemostat model as a tool to study the effects of FMT in vitro. The following analyses were performed: C difficile plate counts, 16S rRNA gene sequencing, 1H-NMR spectroscopy, and UPLC mass spectrometry bile acid profiling. FMT mixtures were prepared using fresh faecal samples provided by donors enrolled in the FMT Programme. Results from chemostat experiments were validated using human stool samples, C difficile batch cultures, and C57BL/6 mice with CDI. Human stool samples were collected from 16 patients with recurrent CDI and 5 healthy donors participating in an FMT trial.

Results: In the CDI chemostat model, clindamycin decreased valerate and deoxycholic acid concentrations and increased C difficile total viable counts (TVC) and valerate precursors, taurocholic acid, and succinate concentrations. After we stopped adding

Powered by Editorial Manager® and ProduXion Manager® from Aries Systems Corporation clindamycin, levels of bile acids and succinate recovered, whereas levels of valerate and valerate precursors did not. In the CDI chemostat model FMT increased valerate concentrations and decreased C difficile TVC (94% reduction), spore counts (86% reduction), and valerate precursor concentrations—concentrations of bile acids were unchanged. In stool samples from patients with CDI, valerate was depleted before FMT, but restored after FMT. C difficile batch cultures confirmed that valerate decreased vegetative growth, and that taurocholic acid is required for germination but had no effect on vegetative growth. C difficile TVC were decreased by 95% in mice with CDI given glycerol trivalerate compared to phosphate-buffered saline.

Conclusions: We identified valerate as a metabolite that is depleted with clindamycin and only recovered with FMT. Valerate is a target for a rationally designed recurrent CDI therapy.

Key words: bacteria; stool transplant; gut microbiome; pathogen

Powered by Editorial Manager® and ProduXion Manager® from Aries Systems Corporation Cover Letter

Dr. Julian R. Marchesi Professor of Digestive Health

Centre for Digestive and Gut Health Department of Surgery and Cancer Faculty of Medicine Imperial College London

Room 1003 Liver Unit, 10th Floor St Mary’s Hospital South Wharf Road London W2 1NY Tel: +44 02033126197 [email protected] 21 June 2018 Dear Professors Peek and Corley, Please consider our revised manuscript entitled "A novel route for inhibiting Clostridioides difficile growth by restoring gut microbiota-produced valerate" by McDonald and colleagues for consideration for publication in Gastroenterology (Category: Basic and Translational, Alimentary Tract, manuscript number GASTRO-D-18-00449R1). We appreciate the positive feedback and interest that the editors and reviewers have taken in our manuscript. We have addressed all the comments and made the requested revisions. The major outstanding concern was the limited clinical perspective without an in vivo mouse interventional study. As such, we performed an interventional study in a C. difficile infection (CDI) mouse model, where C. difficile infected mice were administered valerate (in the form of glycerol trivalerate) or PBS (as a vehicle control). We showed a significant reduction in C. difficile total viable counts in faeces from glycerol trivalerate-treated mice compared to PBS-treated mice (95% reduction after 3 doses, see Figure 6). No adverse effects were noted in glycerol trivalerate-treated mice over the course of the experiment. These results support our other findings and strengthen our manuscript. We have also reworded the text and provided additional clarification as requested by Reviewers 1 and 2. Moreover, we have added a schematic of the key metabolite interactions with C. difficile as requested by Reviewer 2 (see Figure S12). We believe our manuscript has been significantly improved by making these suggested changes. A point-by-point response to the reviewer comments is provided that describes the revisions in more detail. Due to the addition of the mouse experimental data, we have moved text from the main text discussion section to the supplementary discussion section. We have also moved Figure 2 to the supplementary figures to ensure the total number of figures and tables does not exclude 7 in total. The main conclusions of our manuscript remain unchanged. We believe we have exhaustively demonstrated the role of valerate in CDI by presenting results from chemostat experiments, human samples, batch culture experiments, and mouse experiments. We showed that valerate directly inhibits C. difficile growth, and therefore reduction of valerate-producing bacteria in the gut microbiota following antibiotic exposure decreases valerate production and permits C difficile growth. Consequently, restoration of valerate and valerate-producing bacteria in the gut microbiota following FMT decreases C. difficile growth. We believe these findings are important for the development of safer, microorganism- free treatments for CDI/recurrent CDI. Treatment with valerate may be of particular interest to immunocompromised, paediatric, or frail/elderly patients with CDI. Moreover, valerate administration may be beneficial for patients who are expected to undergo further antibiotic treatment for other infections, where antibiotic exposure may render FMT treatment less effective. Therefore, the clear clinical relevance of these findings would be of great interest to Gastroenterology readers. Thank you for your consideration. Should you require any further information, please do not hesitate to contact me. Yours sincerely,

Prof. Julian R. Marchesi Professor of Digestive Health at Imperial College London Professor of Human Microbiome Research at Cardiff University Point by Point Response

Response to Reviewer Comments

EDITORS: This is an interesting study into the pathobiology of C. diff infection in a model system allowing to study the molecular effects of fecal transplantation. The authors show the detrimental effects of Clindamycin, which increased valerate precursors, [taurocholic] acid and succinate, and decreased valerate. After stopping clindamycin the effects on valerate and its precursors persisted, while [TCA] and succinate normalized. FMT counteracted this effect on valerate, and valerate affected C. diff growth. Thus, the authors suggest that valerate is a good target for prevention and treatment of C diff infection and suggest therapeutic studies. While there is general enthusiasm for the study the authors will need to perform to interventional study in mice (a concern also raised by Reviewer 3). In absence of this experiment, this contribution lacks sufficient novelty and clinical perspective for further consideration. The authors will also need to revise the title of the manuscript, as "suppressive metabotype" is too vague. We would like to thank the reviewers and editors for the time and effort put into reviewing the previous version of this manuscript. We appreciate their constructive comments, and we believe this feedback has improved our manuscript. As the editors and Reviewer 3 suggested, we performed an interventional study with valerate (in the form of glycerol trivalerate) in a CDI mouse model. We found a significant decrease in C. difficile total viable counts (TVC) in glycerol trivalerate-treated mice compared to PBS-treated mice (see Figure 6). After only 3 doses, glycerol trivalerate-treated mice had an average of 95% less C. difficile TVC per gram of faeces compared to PBS-treated mice. These results corroborate the other findings presented in this study and support the therapeutic use of valerate to treat CDI. We have discussed these results in more detail in response to Reviewer 3’s comments below. We have also edited the title of the manuscript (as reviewed by Dr. Kristine Novak, Science Editor for Gastroenterology) to remove the term “suppressive metabotype,” and the title now reads: “Inhibiting Growth of Clostridioides difficile by Restoring Valerate, Produced by the Intestinal Microbiota.”

REVIEWER 1: This outstanding manuscript is an extension of work by the authors (their refs 12, 14, and 27) and especially the lead author in her investigations of an in vitro, chemostat model of the microbiome. The effects of fecal transplantation were studied on C. difficile infection. Biochemical and bacteriologic information were convincingly reported. Clindamycin augmented C. difficile growth, valerate precursors, taurocholate, and succinate concentrations, reverting to normal with clindamycin's cessation. Fecal transplantation promoted increased valerate. The statistical documentation of the changes was clear-cut. The investigations are important in developing strategies for treatment of C. difficile, but they also provide information on the microbiome, both important investigative activities. The manuscript is clearly written, with little grammatical or spelling errors. I have only to suggest that they make data plural and that they explain the terms "richness" and "evenness" in their bacteriologic cultures (for us non-bacteriologists). We would like to thank Reviewer 1 for the positive comments. We have edited the manuscript to make “data” plural. We have also defined the terms “richness” and “evenness” (see lines 261-263): “Clindamycin dosing also resulted in a decrease in bacterial diversity (p=0.002), richness (the number of species present, p=0.001) and evenness (how evenly distributed each species is, p=0.004) (Figure S2).”

REVIEWER 2: As I understand, the authors demonstrate in vitro the impact of clindamycin on C. difficile (CD) vegetative growth and spore germination, its transient inhibition of the resident microbiota and downstream changes to the metabolic environment that determine the likelihood of recurrent CD infection. It is a logical and encouraging study that goes a long way to improving this reviewer's understanding of pseudo-membranous colitis and related diseases. It will stimulate further research into this complex area with a prospect for the development of reliable, safe and acceptable therapies. We would like to thank Reviewer 2 for the helpful comments. We believe addressing these comments has improved our manuscript. Comments: 1) The use of acronyms, particularly names used infrequently and bile acids. Suggest avoid BSH and any others infrequently used acronym, and consider a prefix before bile acid acronyms to identify between primary and secondary compounds. We have removed infrequently used acronyms from the manuscript, including BSH. We have referred the reader to which bile acids are primary vs secondary bile acids in the text (see lines 311-319): “During the clindamycin dosing period there was a significant increase in the conjugated primary bile acids taurocholic acid (TCA, p=0.004), glycocholic acid (GCA, p=0.005), and glycochenodeoxycholic acid (GCDCA, p=0.005), in the unconjugated primary bile acids cholic acid (CA, p=0.004) and chenodeoxycholic acid (CDCA, p=0.037), and in the conjugated secondary bile acids taurodeoxycholic acid (TDCA, p=0.045) and glycodeoxycholic acid (GDCA, p=0.004) (Figures 3 and S9). During the clindamycin dosing period there was a significant decrease in the secondary bile acids deoxycholic acid (DCA, p=0.006), lithocholic acid (LCA, p=0.005), and ursodeoxycholic acid (UDCA, p=0.037) (Figures 3 and S9).” 2) Provide a brief summary of the general relevance or valerate (new to me). We have provided a brief summary of valerate on lines 383-384: “Valerate is a short chain fatty acid produced via amino acid fermentation by members of the gut microbiota.33” 3) Consider a schematic (editor's approval permitting?) highlighting the sequence of key interactions. We have provided a schematic to summarise the findings from our study (see Figure S12).

Figure S12: Schematic of key metabolite interactions with C. difficile during health, CDI, and recurrent CDI. Initial antibiotic treatment decreases the diversity of the gut microbiota, killing bacteria that produce valerate and bile salt hydrolase (an enzyme that degrades TCA). This results in increased levels of TCA and decreased levels of valerate, allowing for C. difficile spore germination and vegetative growth (respectively). Treatment for CDI (vancomycin/metronidazole) decreases C. difficile vegetative cells, but C. difficile spores remain and microbial community diversity remains low. Again, this antibiotic exposure results in an environment with high TCA and low valerate, allowing the remaining C. difficile spores to germinate and grow once vancomycin/metronidazole is stopped. When a patient receives FMT for recurrent CDI (usually within 1-2 days of stopping suppressive antibiotics) valerate levels (and valerate producing bacteria) and bile salt hydrolase levels (and bile salt hydrolysing bacteria) are restored, resulting in an environment that inhibits both C. difficile germination and vegetative growth. 4) Line 260-64 need to be written as it only makes sense after reading the next sentence. We have edited the following sentences as follows (see lines 282-287): “Following the end of the clindamycin-dosing period the levels of succinate, butyrate, acetate, and isobutyrate recovered to pre-antibiotic levels, and these levels were not affected by FMT treatment (Figures 2 and S3). However, after stopping clindamycin dosing the levels of valerate (p=0.009) were still significantly decreased and the levels of 5-aminovalerate (p=0.009) and ethanol (p=0.013) were still significantly increased compared to pre-antibiotic levels (Figure 2), indicating the levels of these metabolites did not recover after stopping clindamycin dosing.” 5) Sentence 281 and others might be better constructed starting with a subject - "The proposed valerate peaks ....". We have edited the following sentences as follows (see lines 303-307): “The identity of the proposed valerate peaks was confirmed in the chemostat culture supernatants by performing 1D-1H-NMR spectroscopy of a sample before and after valerate spike-in, as well as 2D-1H-NMR spectroscopy of a valerate-containing sample (Figures S5 and S6, Supplementary Results). The identities of other metabolites were also confirmed in the chemostat culture supernatants by performing 2D-1H-NMR and STOCSY (Figures S7 and S8).” 6) Sentence starting line 288 is ambiguous. We have revised this sentence to clarify (see lines 311-316): “During the clindamycin dosing period there was a significant increase in the conjugated primary bile acids taurocholic acid (TCA, p=0.004), glycocholic acid (GCA, p=0.005), and glycochenodeoxycholic acid (GCDCA, p=0.005), in the unconjugated primary bile acids cholic acid (CA, p=0.004) and chenodeoxycholic acid (CDCA, p=0.037), and in the conjugated secondary bile acids taurodeoxycholic acid (TDCA, p=0.045) and glycodeoxycholic acid (GDCA, p=0.004) (Figures 3 and S9).” 7) Sentence starting line 388 is too speculative. We have revised this sentence to be less speculative (see lines 454-456): “A more controlled method of restoring bile salt hydrolase activity in the gut microbiomes of CDI patients could be to avoid the administration of live microorganisms by administering purified bile salt hydrolase enzyme preparations.” 8) The use of "correlate" might be better replaced with "associate" in some cases given the complexity of these manifest interactions. In this manuscript we performed regularised canonical correlation analysis and Spearman's rho statistic, and when we use the term “correlate” in our manuscript we are referring to these analyses. Therefore, we believe the term “correlate” is appropriate to use in these contexts. 9) The ordination figures (S3) were not clear in the material provided me. We agree that the figures included in the merged pdf for review look less clear than our original uploaded files. The figure resolution is downgraded upon upload to the manuscript submission website. If the reviewer clicks the link at the top of the figure “Click here to download Figure FigureS4.tiff” a file will be downloaded with a clearer resolution image.

REVIEWER 3: This is an interesting study into the pathobiology of C. diff infection in a model system allowing to study the molecular effects of fecal transplantation. The authors show the detrimental effects of clindamycin, which increased valerate precursors, [taurocholic] acid and succinate, and decreased valerate. After stopping clindamycin the effects on valerate and its precursors persisted, while [TCA] and succinate normalized. FMT counteracted this effect on valerate, and valerate affected C. diff growth. Thus, the authors suggest that valerate is a good target for prevention and treatment of C. diff infection and suggest therapeutic studies. This paper could be greatly enhanced, if the authors actually performed such an interventional study in the mouse system in line with their own suggestion. Without this experiment the data are still [interesting], but lack the novelty and convincing clinical perspective to make it a top paper. After all, the effect of antibiotics on valerate has been described before (Theriot et al. 2014), and the reconstitution by FMT is expected. Furthermore, in view of the new data on trehalose on C. diff growth, for example, it is likely that we will see a competition of various C. diff nutrient-targeted interventions very soon, which will have to prove their effectiveness in [a] mouse model first before being taken into clinical trials. Furthermore, and this could also enhance this paper, in addition to the direct effects on C. diff growth, the competition in the gut of protective species with C. diff may explain nutrient intervention effects and FMT effects - the role of secondary bile acids is here probably more prominent, and they thus are likely to benefit by enhancing protective Clostridium species as a perhaps more important mechanism in [vivo] than the effect of valerate. We would like to thank Reviewer 3 for the suggestion to investigate valerate as an intervention in a mouse CDI model. We performed an interventional study testing the effects of valerate (in the form of glycerol trivalerate) in a mouse model of CDI using methods previously described by Winston and colleagues.2 The results are summarised on lines 361-373: “Next, we determined whether valerate could inhibit C. difficile growth when administered as an intervention in a CDI mouse model (Figure 6A). To avoid the rapid uptake of valerate and obtain sufficient release in the gastrointestinal tract, valerate was administered to the mice in the form of glycerol trivalerate (which is hydrolysed by lipases to release valerate32). One-day post-infection (prior to glycerol trivalerate or PBS administration) there was no significant difference in the levels of C. difficile TVC in mouse faeces in each group (p=0.584). Glycerol trivalerate or PBS was administered to C. difficile-infected mice by oral gavage 1, 2, and 3 days post-infection (n=5 per group). There was a significant decrease in C. difficile TVC in glycerol trivalerate-treated mice compared to PBS-treated mice (Figure 6B, p=0.027 on day 2, p=0.007 on day 3, and p=0.024 on day 4). After only 3 doses of glycerol trivalerate or PBS, glycerol trivalerate-treated mice had an average of 95% less C. difficile TVC per gram of faeces compared to PBS-treated mice. No adverse effects were noted in glycerol trivalerate-treated mice over the course of the experiment. These results corroborate the other findings presented in this study and support the therapeutic use of valerate to treat CDI.” As summarised above, we saw a significant reduction in C. difficile total viable counts in faeces from glycerol trivalerate-treated mice compared to PBS-treated mice (95% reduction after 3 doses of glycerol trivalerate/PBS). No adverse effects were noted in glycerol trivalerate-treated mice over the course of the experiment. These results corroborate the other findings presented in this study and support the therapeutic use of valerate to reduce C. difficile TVC in vivo. The advantage of using valerate (instead of a live microorganism or relatively undefined FMT preparation) is that valerate is a well-defined small molecule that is normally present in the healthy gut and appears to have no adverse outcomes. Valerate may be a preferred treatment for CDI in immunocompromised patients where administration of bacteria may increase the risk of an infection, or in CDI patients that require further antibiotic treatment for other purposes (where the antibiotic treatment would kill bacteria present in FMT preparations and render the treatment less effective). In this study glycerol trivalerate was orally gavaged to mice, but glycerol trivalerate has also been included as an additive in animal feed, so there is the potential to administer this treatment via a less invasive route compared to FMT. The findings from this study led us to propose a novel mechanism of FMT for treatment of CDI: (1) antibiotic exposure kills bacteria in the intestine that are responsible for valerate production; (2) decreased valerate concentrations permits C. difficile vegetative growth, and (3) FMT restores valerate and valerate-producing bacteria concentrations in the intestine, decreasing C. difficile vegetative growth. We believe we have exhaustively demonstrated the role of valerate in CDI by presenting results from chemostat experiments, human samples, batch culture experiments, and mouse experiments. Our findings are novel and of clear clinical relevance given the potential of valerate to be used as a novel approach to treat CDI with no risk for the development of antimicrobial resistance.

Graphical AbstractC. difficile spores Click here to download Graphical Abstract Gastro-graphical-abstract- 2Mar2018 JRM.pptx Valerate + clindamycin C. difficile FMT (VB), spores Saline (VA) Vessel B 14 Clindamycin (VB)

) 12

mM 10 VA VB VA VB 8 Day 25 Day 32 6 Saline FMT 4 Vessel A Metabonomics 2 (VA) Concentration ( Concentration 0 18 22 26 30 34 38 42 46 50 54 VA VB VA VB Day Day 54 Day 42 Revised Manuscript in Word or RTF (with changes marked)

1 Inhibiting Growth of Clostridioides difficile by Restoring Valerate, Produced by the Intestinal 1 2 2 Microbiota 3 4 5 3 6 7 4 Short title: Valerate inhibits Clostridioides difficile 8 9 5 10 11 1 1 1 1 12 6 Julie A. K. McDonald , Benjamin H. Mullish , Alexandros Pechlivanis , Zhigang Liu , Jerusa 13 14 7 Brignardello1, Dina Kao2, Elaine Holmes1, Jia V. Li1, Thomas B. Clarke3, Mark R. Thursz1, Julian R. 15 16 8 Marchesi1,4 17 18 19 9 20 21 10 1. Division of Integrative Systems Medicine and Digestive Disease, Department of Surgery and Cancer, 22 23 24 11 Faculty of Medicine, Imperial College London, London, UK. 25 26 12 2. Division of Gastroenterology, Department of Medicine, University of Alberta, Edmonton, Alberta, 27 28 13 Canada. 29 30 31 14 3. MRC Centre for Molecular Bacteriology and Infection, Imperial College London, London, UK. 32 33 15 4. School of Biosciences, Cardiff University, Cardiff, UK. 34 35 16 36 37 38 17 Grant support: The Division of Integrative Systems Medicine and Digestive Disease at Imperial College 39 40 18 London receives financial support from the National Institute of Health Research (NIHR) Imperial 41 42 43 19 Biomedical Research Centre (BRC) based at Imperial College Healthcare NHS Trust and Imperial 44 45 20 College London. This article is independent research funded by the NIHR BRC, and the views expressed 46 47 21 in this publication are those of the authors and not necessarily those of the NHS, NIHR, or the 48 49 50 22 Department of Health. BHM is the recipient of a Medical Research Council (MRC) Clinical Research 51 52 23 Training Fellowship (grant reference: MR/R000875/1). DK received research funding from Alberta 53 54 24 Health Services and University of Alberta Hospital Foundation. TBC is a Sir Henry Dale Fellow jointly 55 56 57 25 funded by the Wellcome Trust and Royal Society (Grant Number 107660/Z/15Z). 58 59 26 60 61 62 1 63 64 65 27 Abbreviations: 1H-NMR, proton nuclear magnetic resonance; CA, cholic acid; CDCA, chenodeoxycholic 1 2 28 acid; CDI, Clostridioides difficile infection; COSY, correlation spectroscopy; DCA, deoxycholic acid; FDR, 3 4 5 29 false discovery rate; FMT, faecal microbiota transplantation; GC-MS, gas chromatography-mass 6 7 30 spectrometry; GCA, glycocholic acid; GCDCA, glycochenodeoxycholic acid; GDCA, glycodeoxycholic 8 9 31 acid; LCA, lithocholic acid; NOESY, nuclear Overhauser enhancement spectroscopy; OD600, optical 10 11 12 32 density at 600 nm; OTU, operational taxonomic unit; PBS, phosphate buffered saline; rCCA, 13 14 33 regularised Canonical Correlation Analysis; SANTA, Short AsyNchronous Time-series Analysis; STOCSY, 15 16 34 statistical total correlation spectroscopy; TCA, taurocholic acid; TDCA, taurodeoxycholic acid; TOCSY, 17 18 19 35 total correlation spectroscopy; TVC, total viable counts; UDCA, ursodeoxycholic acid; UPLC-MS, ultra- 20 21 36 performance liquid chromatography-mass spectrometry; VA, vessel A; VB, vessel B. 22 23 24 37 25 26 38 Correspondence: 27 28 39 Prof. Julian R. Marchesi, 29 30 31 40 Division of Integrative Systems Medicine and Digestive Disease, 32 33 41 Department of Surgery and Cancer, Faculty of Medicine, 34 35 42 Imperial College London, St Mary’s Hospital Campus, 36 37 38 43 South Wharf Road, London, UK W2 1NY, 39 40 44 Tel: +44 02033126197, [email protected] or [email protected] 41 42 43 45 44 45 46 Disclosures: DK has received research funding from Rebiotix. All other authors disclose no conflicts. 46 47 47 48 49 50 48 Author Contributions: JAKM and JRM designed the study. BHM sourced faecal samples and helped 51 52 49 with data analysis. JAKM conducted the chemostat experiments, 16S rRNA gene sequencing, 16S rRNA 53 54 50 gene qPCR, batch culture experiments, data integration, and performed statistical analysis. JAKM, AP, 55 56 57 51 and BHM performed bile acid UPLC-MS and data analysis. JAKM, JVL, and ZL performed 1D- and 2D- 58 59 52 NMR and data analysis. DK provided human stool samples for GC-MS analysis. BHM and JB performed 60 61 62 2 63 64 65 53 GC-MS and data analysis. TBC and JAKM performed the mouse experiments. JAKM wrote the 1 2 54 manuscript and all authors edited the manuscript. All authors read and approved the final version of 3 4 5 55 the manuscript. 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 3 63 64 65 56 ABSTRACT: 1 2 57 Background & Aims: Faecal microbiota transplantation (FMT) is effective for treating recurrent 3 4 5 58 Clostridioides difficile infection (CDI), but there are concerns regarding its long-term safety. 6 7 59 Understanding the mechanisms of the effects of FMT could help us design safer, targeted therapies. 8 9 60 We aimed to identify microbial metabolites that are important for C difficile growth. 10 11 12 61 Methods: We used a CDI chemostat model as a tool to study the effects of FMT in vitro. The following 13 14 62 analyses were performed: C difficile plate counts, 16S rRNA gene sequencing, 1H-NMR spectroscopy, 15 16 63 and UPLC mass spectrometry bile acid profiling. FMT mixtures were prepared using fresh faecal 17 18 19 64 samples provided by donors enrolled in the FMT Programme. Results from chemostat experiments 20 21 65 were validated using human stool samples, C difficile batch cultures, and C57BL/6 mice with CDI. 22 23 24 66 Human stool samples were collected from 16 patients with recurrent CDI and 5 healthy donors 25 26 67 participating in an FMT trial. 27 28 68 Results: In the CDI chemostat model, clindamycin decreased valerate and deoxycholic acid 29 30 31 69 concentrations and increased C difficile total viable counts (TVC) and valerate precursors, taurocholic 32 33 70 acid, and succinate concentrations. After we stopped adding clindamycin, levels of bile acids and 34 35 71 succinate recovered, whereas levels of valerate and valerate precursors did not. In the CDI chemostat 36 37 38 72 model FMT increased valerate concentrations and decreased C difficile TVC (94% reduction), spore 39 40 73 counts (86% reduction), and valerate precursor concentrations—concentrations of bile acids were 41 42 43 74 unchanged. In stool samples from patients with CDI, valerate was depleted before FMT, but restored 44 45 75 after FMT. C difficile batch cultures confirmed that valerate decreased vegetative growth, and that 46 47 76 taurocholic acid is required for germination but had no effect on vegetative growth. C difficile TVC 48 49 50 77 were decreased by 95% in mice with CDI given glycerol trivalerate compared to phosphate-buffered 51 52 78 saline. 53 54 79 Conclusions: We identified valerate as a metabolite that is depleted with clindamycin and only 55 56 57 80 recovered with FMT. Valerate is a target for a rationally designed recurrent CDI therapy. 58 59 81 Key words: bacteria; stool transplant; gut microbiome; pathogen 60 61 62 4 63 64 65 82 INTRODUCTION: 1 2 83 Clostridioides difficile (formerly Clostridium difficile) is an anaerobic, spore-forming, Gram-positive 3 4 5 84 bacterium that causes opportunistic infections in the human colon, usually after antibiotic exposure. 6 7 85 C difficile infection (CDI) can lead to diarrhoea, pseudomembranous colitis, toxic megacolon, intestinal 8 9 86 perforation, multi-organ failure, and death.1 A recent study showed that the incidence of recurrent 10 11 12 87 CDI has disproportionately increased relative to CDI overall, therefore the demand for recurrent CDI 13 14 88 therapies may be rising.2 15 16 89 The principle behind faecal microbiota transplantation (FMT) is to use stool from a healthy donor 17 18 19 90 to replace the microorganisms and ecosystem functions that are depleted in the gut of recurrent CDI 20 21 91 patients. While FMT is highly effective at treating recurrent CDI,3 it lacks a detailed mechanism of 22 23 24 92 action and it is unclear whether all the microbes included in the preparations are required to resolve 25 26 93 disease. There are concerns regarding the long-term safety, reproducibility, composition, and stability 27 28 94 of FMT preparations,4 and potential risks include transmission of infections, invasive administration 29 30 31 95 routes, and concerns treating high-risk individuals (frail/elderly or immunosuppressed patients). In 32 33 96 addition, as more studies describe the role of the gut microbiota in disease, it is unclear whether FMT 34 35 97 could result in the transfer of a gut microbiota which later contributes to disease (e.g. colorectal 36 37 38 98 cancer, obesity, inflammatory bowel disease, etc). 39 40 99 C difficile causes disease after germination, where cells change from their dormant spore state to 41 42 5 43 100 their active vegetative state. Previous studies have suggested that exposure to antibiotics alters the 44 45 101 composition and functionality of the gut microbiota, changing the global metabolic profile to an 46 47 102 environment that supports C difficile germination and vegetative growth.6 Studies have shown that 48 49 50 103 antibiotic exposure and FMT alters many microbial metabolic pathways, including bile acid 51 52 104 metabolism7 and succinate metabolism.8 In fact, Ott and colleagues showed that sterile faecal filtrate 53 54 105 from healthy stool donors was able to cause remission from recurrent CDI in a preliminary 55 56 9 57 106 investigation of five patients with the condition. It is possible the sterile faecal filtrate contained 58 59 60 61 62 5 63 64 65 107 bacterial metabolites or enzymes that were sufficient to inhibit C difficile spore germination and 1 2 108 vegetative growth. 3 4 5 109 Mechanistic studies are challenging to conduct in vivo due to the wide variety of factors which 6 7 110 influence the composition and functionality of the gut microbiota. Firstly, samples from recurrent CDI 8 9 111 patients prior to FMT are usually collected while they are still on suppressive vancomycin. Therefore, 10 11 12 112 it is difficult to determine whether changes in specific bacteria or metabolites following FMT are due 13 14 113 to the FMT administration, or whether these changes could have occurred in the absence of FMT due 15 16 114 to recovery of the gut microbiota following cessation of antibiotic treatment. Changes in diet can also 17 18 19 115 cause profound changes in the composition and functionality of the gut microbiota, especially short 20 21 116 chain fatty acid production,10 and diet is especially difficult to control in human studies. Recurrent CDI 22 23 24 117 patients may eat differently before and after receiving FMT, and may eat differently from healthy 25 26 118 controls. Studies on diet, the gut microbiota, and short chain fatty acid production have often relied 27 28 119 on fermentation data in vitro and animal data due to the challenges associated with human studies.11 29 30 31 120 Therefore, human studies could lead to “false positives” for mechanisms of C difficile pathogenesis. 32 33 121 Data collected from chemostat studies can be used to complement microbiome data collected 34 35 122 from human and animal studies to more easily determine a mechanism of action for specific disease 36 37 12 38 123 states or interventions. Chemostat models are artificial systems that mimic some of the spatial, 39 40 124 temporal, and environmental conditions found in the human gut.13 Chemostats have many advantages 41 42 12 43 125 over human and animal studies, which have been discussed in detail previously. Bacterial 44 45 126 communities cultured in these models are highly reproducible, stable, complex, and representative of 46 47 127 the bacterial communities found in vivo.14,15 This means researchers can perform longitudinal studies 48 49 50 128 in these systems that can directly link changes in the gut microbiota structure and function to an 51 52 129 experimental intervention. Chemostats have previously been used to model CDI and test the effects 53 54 130 of several treatments on C difficile growth and pathogenesis (e.g. antibiotics,16-19 bacteriophages,20 55 56 21 57 131 and lactoferrin ). 58 59 60 61 62 6 63 64 65 132 We used a twin-vessel single-stage distal gut chemostat model as a tool to study CDI and the effects 1 2 133 of FMT under tightly-controlled conditions in vitro. We hypothesised that exposure to antibiotics kills 3 4 5 134 bacteria that perform important functions in the gut microbial ecosystem, resulting in a “metabolic 6 7 135 dysbiosis” where the loss or reduction of specific microbial metabolic pathways creates an 8 9 136 environment that promotes C difficile germination and growth. We also hypothesised that FMT 10 11 12 137 administration would reverse these effects by restoring the bacteria responsible for performing these 13 14 138 key metabolic functions. Our aim was to identify these metabolites so we could propose new 15 16 139 therapeutic approaches to treat recurrent CDI that are well-defined, effective, and safe. 17 18 19 140 20 21 141 MATERIALS AND METHODS: 22 23 24 142 Chemostat model of CDI: 25 26 143 The chemostat models used in this study were two identical Electrolab FerMac 200 series 27 28 144 bioreactor systems (Electrolab, Tewkesbury, UK). Chemostat inoculum and growth medium were 29 30 31 145 prepared and vessels were inoculated and operated as previously described (see Supplementary 32 33 146 Methods).14 Stool samples were collected under approval from the UK National Research Ethics 34 35 147 Centres (13/LO/1867). We performed three separate twin-vessel chemostat experiments. In each 36 37 38 148 experiment, two identical vessels (“VA” receiving saline vehicle control and “VB” receiving FMT 39 40 149 preparation) were inoculated with a 10% (w/v) faecal slurry prepared using fresh faeces from a healthy 41 42 43 150 donor not exposed to antibiotics within the previous 2 months (Run 1= male in his 40's; Run 2= male 44 45 151 in his 60’s; Run 3= male in his 80’s). We used clindamycin and C difficile spores to induce CDI in our 46 47 152 chemostat model following a modified version of the methods previously described by Freeman and 48

49 22 50 153 colleagues (see Table 1 and Supplementary Methods). After stopping clindamycin dosing, microbial 51 52 154 communities were allowed to stabilise before administering the FMT preparation or saline vehicle 53 54 155 control. FMT mixtures were prepared using fresh faecal samples provided by donors enrolled within 55 56 23 57 156 Imperial’s FMT Programme. Stool samples from these faecal donors have previously been used to 58 59 157 successfully treat recurrent CDI patients. 60 61 62 7 63 64 65 158 1 2 159 16S rRNA gene sequencing (metataxonomics): 3 4 5 160 DNA extraction is described in the Supplementary Methods. Sample libraries amplifying the V3-V4 6 7 161 region of the 16S rRNA gene were prepared following Illumina’s 16S Metagenomic Sequencing Library 8 9 162 Preparation Protocol,24 with a few modifications. First, we used the SequalPrep Normalization Plate 10 11 12 163 Kit (Life Technologies, Carlsbad, USA) to clean up and normalise the index PCR reactions. Also, we used 13 14 164 the NEBNext Library Quant Kit for Illumina (New England Biolabs, Ipswich, USA) to quantify the sample 15 16 165 libraries. Sequencing was performed on an Illumina MiSeq platform (Illumina Inc, San Diego, USA) 17 18 19 166 using the MiSeq Reagent Kit v3 (Illumina) and paired-end 300 bp chemistry. The resulting data were 20 21 167 pre-processed and analysed as described in the Supplementary Methods. 22 23 24 168 25 26 169 1H-NMR spectroscopy: 27 28 170 Chemostat culture supernatants were prepared for 1H-NMR as described in the Supplementary 29

30 1 31 171 Methods. One-dimensional H-NMR spectra were acquired from chemostat culture supernatants at 32 33 172 300 K on a Bruker DRX 600 MHz NMR spectrometer or a Bruker AVANCE III 600 MHz NMR 34 35 173 spectrometer (Bruker Biospin, Germany). A standard one-dimensional NMR pulse sequence [RD-90°- 36 37 38 174 t1-90°-tm-90°-acq] was used with a recycle delay (4 s) and mixing time (100 ms). The 90° pulse length 39 40 175 was around 10 μs and 32 scans were recorded. Metabolites concentrations were quantified from 41 42 25 43 176 spectra using the Chenomx NMR suite software (Chenomx Inc, Edmonton, Canada). 44 45 177 To confirm the identity of key metabolites in chemostat culture supernatants a series of NMR 46 47 178 spectra including 1D 1H NOESY, 2D 1H−1H TOCSY and 1H−1H COSY of a chemostat culture supernatant 48 49 50 179 and a metabolite standard were recorded (see Supplementary Methods). 51 52 180 53 54 181 Ultra-performance liquid chromatography-mass spectrometry (UPLC-MS) bile acid profiling: 55 56 57 182 Bile acids were extracted from 50 µL of chemostat culture by adding 150 µL of cold methanol, 58 59 183 followed by incubation at -30ᵒC for 2 hours. Tubes were centrifuged at 9500 x g and 4ᵒC for 20 min 60 61 62 8 63 64 65 184 and 120µL of supernatant was loaded into vials. Bile acid analysis was performed using an ACQUITY 1 2 185 UPLC (Waters Ltd, Elstree, UK) coupled to a Xevo G2 Q-ToF mass spectrometer. The MS system was 3 4 5 186 equipped with an electrospray ionization source operating in negative ion mode, using methods 6 7 187 previously described by Sarafian and colleagues.26 Data pre-processing and analysis are described in 8 9 188 the Supplementary Methods. 10 11 12 189 13 14 190 Gas chromatography-mass spectrometry (GC-MS) analysis of human stool samples: 15 16 191 Human stool samples were collected from recurrent CDI patients (n=16) and healthy donors (n=5) 17 18 19 192 as part of a randomised clinical trial comparing the efficacy of capsulized and colonoscopic FMT for 20 21 193 the treatment of recurrent CDI, as previously described.27 Pre-FMT samples were collected from 22 23 24 194 recurrent CDI patients while on suppressive vancomycin. Post-FMT samples were collected 1, 4, and 25 26 195 12 weeks after FMT treatment. All patients were successfully treated following a single FMT. 27 28 196 A targeted GC-MS protocol was used to identify and quantify short chain fatty acids from human 29

30 28 31 197 stool samples as previously-described. Samples were analysed on an Agilent 7890B GC system, 32 33 198 coupled to an Agilent 5977A mass selective detector (Agilent, Santa Clara, CA). Data analysis was 34 35 199 performed using MassHunter software (Agilent). 36 37 38 200 39 40 201 C difficile batch cultures: 41 42 43 202 We tested the effects of valerate (Fisher Scientific) on the vegetative growth of three C difficile 44 45 203 ribotypes (010, 012, and 027) as well as several gut commensal bacteria (Bacteroides uniformis, 46 47 204 Bacteroides vulgatus, and Clostridium scindens) (see Supplementary Methods). We centrifuged an 48 49 50 205 overnight culture of the test isolate at 3000 x g for 10 minutes and resuspended the cells in brain heart 51 52 206 infusion broth (Sigma-Aldrich) (supplemented with 5 mg/mL yeast extract (Sigma-Aldrich), and 0.1% 53 54 207 L-cysteine (Sigma-Aldrich)), containing varying concentrations of valerate (0, 1, 2, 3, 4, 5, 10, and 20 55 56 57 208 mM, pH of broth adjusted to 6.8) in triplicate. The OD600 was measured at time zero and cultures were 58 59 209 incubated at 37ᵒC in an ElectroTek AW 400TG Anaerobic Workstation (ElectroTek, West Yorkshire, 60 61 62 9 63 64 65 210 UK). Additional OD600 measurements were taken at 2, 4, 6, and 8 hours post-inoculation. We plotted 1

2 211 the changes in OD600 (from a time point during the exponential phase) against each concentration of 3 4 5 212 valerate tested. We used ANOVA and Tukey post hoc test to determine whether the concentration of 6 7 213 valerate tested affected the growth of the test isolate compared to batch cultures grown in the 8 9 214 absence of valerate. 10 11 12 215 We also tested the effects of taurocholic acid (TCA) on C difficile germination and vegetative growth 13 14 216 using batch cultures (see Supplementary Methods). 15 16 217 17 18 19 218 Mouse model of CDI: 20 21 219 Mouse experiments were performed under the authority of the UK Home Office outlined in the 22 23 24 220 Animals (Scientific Procedures) Act 1986 after ethical review by Imperial College London Animal 25 26 221 Welfare and Ethical Review Body (PPL 70/7969). We adhered to standards articulated in the Animal 27 28 222 Research: Reporting of In Vivo Experiments (ARRIVE) guidelines. 29 30 31 223 Wild-type C57BL/6 mice (8-10-week-old; female) were purchased from Envigo (UK) and 32 33 224 acclimatised for 1 week prior to use. Mice were housed five per cage in individually ventilated cages 34 35 225 with autoclaved food (RM1, Special Diet Services), bedding (Aspen chip 2 bedding), and water 36 37 38 226 (provided ad libitum). Mice were subjected to a 12 h light and 12 h dark cycle at 20–22°C. 39 40 227 We used a previously published mouse model of C difficile infection as described by Winston and 41 42 29 43 228 colleagues (Figure 6A). Briefly, mice were given 0.5 mg/ml cefoperazone in their drinking water for 44 45 229 5 days (from day -7 to day -2), followed by autoclaved antibiotic-free water for the remainder of the 46 47 230 experiment. C difficile spores were prepared and enumerated (see Supplementary Methods) and mice 48

49 5 50 231 were challenged with 10 C difficile spores by oral gavage on day 0. Mice were orally gavaged with 200 51 52 232 µl of 15 mM glycerol trivalerate (n=5) or 200 µl of PBS (n=5) on days 1, 2, and 3. Faecal samples were 53 54 233 collected on days 1, 2, 3, and 4 and C difficile total viable counts (TVC) were quantified (see 55 56 57 234 Supplementary Methods). Mice were not fasted before oral gavages and all interventions were 58 59 235 performed during the light cycle. 60 61 62 10 63 64 65 236 1 2 237 Statistical analysis and data integration: 3 4 5 238 Statistical tests were performed using IBM SPSS Statistics Software version 23 (paired t-test, 6 7 239 independent t-test, ANOVA) or GraphPad Prism version 7.03 (Mann-Whitney U test, Friedman test). 8 9 240 Short AsyNchronous Time-series Analysis (SANTA) was used to determine whether there were 10 11 12 241 significant changes in data trajectories at several time periods over the course of the chemostat 13 14 242 experiments (see Supplementary Methods).30 Spearman's rho statistic and p-values were calculated 15 16 243 for C difficile TVC and metabolite data using the cor and cor.test functions, respectively, within the 17 18 19 244 stats base library within R. A p-value less than 0.05 was considered significant. We used regularised 20 21 245 Canonical Correlation Analysis (rCCA) to correlate metataxonomic and metabolomic data using the 22 23 31 24 246 mixOmics library within R (see Supplementary Methods). 25 26 247 27 28 248 RESULTS: 29 30 31 249 C difficile total viable counts and spore counts from chemostat culture samples: 32 33 250 Following the addition of C difficile spores to each vessel C difficile TVC and spore counts were 34 35 251 enumerated every other day until the end of the experiment (Figure 1). We found an increase in C 36 37 38 252 difficile TVC during the clindamycin-dosing period (p<0.001). We also found a 94% reduction in C 39 40 253 difficile TVC (p=0.025) and an 86% reduction in C difficile spore counts (p=0.034) in FMT-treated 41 42 43 254 cultures compared to saline-treated cultures. 44 45 255 46 47 256 Bacterial community composition of chemostat culture samples: 48 49 50 257 We found that clindamycin dosing of chemostat cultures altered the composition and the biomass 51 52 258 of the bacterial communities (Figure S1). Clindamycin dosing caused an initial decrease in the total 53 54 259 bacterial biomass one day after starting clindamycin dosing (p<0.001), however the total bacterial 55 56 57 260 biomass increased over time to reach pre-clindamycin levels by the end of the clindamycin dosing 58 59 261 period (p>0.05). Clindamycin dosing also resulted in a decrease in bacterial diversity (p=0.002), 60 61 62 11 63 64 65 262 richness (the number of species present, p=0.001) and evenness (how evenly distributed each species 1 2 263 is, p=0.004) (Figure S2). There was a significant increase in richness following FMT treatment 3 4 5 264 (p=0.035), however diversity (p>0.05) and evenness (p>0.05) remained unchanged. 6 7 265 8 9 266 Global metabolic profiling of chemostat culture samples by 1H-NMR spectroscopy: 10 11 1 12 267 We performed H-NMR spectroscopy as an exploratory technique to generate global metabolite 13 14 268 spectral profiles for samples collected over the course of the chemostat experiments. There was a 15 16 269 significant decrease in valerate (p=0.004), butyrate (p=0.004), and acetate (p=0.013) and a significant 17 18 19 270 increase in 5-aminovalerate (p=0.004), ethanol (p=0.021), succinate (p=0.004), and isobutyrate 20 21 271 (p=0.004) during the clindamycin dosing period compared to the steady state period (Figures 2 and 22 23 24 272 S3). rCCA modelling was used to determine correlations between 16S rRNA gene sequencing data and 25 26 273 metabolite data that correspond to clindamycin dosing. The unit representation plot showed a clear 27 28 274 separation between steady state cultures and cultures sampled during and after the clindamycin 29 30 31 275 dosing period (Figure S4A). The correlation circle plot showed that the separation between the 32 33 276 cultures was due to decreases in the levels of valerate and acetate, and increases in the levels 5- 34 35 277 aminovalerate, succinate, isobutyrate, and ethanol during and after the clindamycin dosing period 36 37 38 278 (Figure S4B). This plot also showed strong correlations between bacterial genera and these 39 40 279 metabolites. During the clindamycin-dosing period there were significant correlations between C 41 42 43 280 difficile TVC and valerate (rS=-0.59, p=0.005), 5-aminovalerate (rS=0.54, p=0.010), and succinate (rS=- 44 45 281 0.49, p=0.022). 46 47 282 Following the end of the clindamycin-dosing period the levels of succinate, butyrate, acetate, and 48 49 50 283 isobutyrate recovered to pre-antibiotic levels, and these levels were not affected by FMT treatment 51 52 284 (Figures 2 and S3). However, after stopping clindamycin dosing the levels of valerate (p=0.009) were 53 54 285 still significantly decreased and the levels of 5-aminovalerate (p=0.009) and ethanol (p=0.013) were 55 56 57 286 still significantly increased compared to pre-antibiotic levels (Figure 2), indicating the levels of these 58 59 287 metabolites did not recover after stopping clindamycin dosing. Moreover, after stopping clindamycin 60 61 62 12 63 64 65 288 dosing the levels of isovalerate decreased (p=0.009) and the levels of propionate increased (p=0.036) 1 2 289 compared to pre-antibiotic levels (Figures 2 and S3). 3 4 5 290 There was a significant increase in valerate (p=0.032), and significant decreases in 5-aminovalerate 6 7 291 (p=0.032), ethanol (p=0.032), propionate (p=0.032), and methanol (p=0.039) in FMT-treated cultures 8 9 292 compared to saline-treated cultures (Figure 2). rCCA modelling was also used to determine 10 11 12 293 correlations between 16S rRNA gene sequencing data and metabolite data that correspond to FMT or 13 14 294 saline treatment. The unit representation plot showed a clear separation between FMT-treated 15 16 295 cultures and saline-treated cultures along the second canonical variate (Figure S4C). The correlation 17 18 19 296 circle plot showed that the separation between FMT- and saline-treated cultures was due to increases 20 21 297 in the levels of valerate and decreases in the levels of 5-aminovalerate, ethanol, and methanol in FMT- 22 23 24 298 treated cultures (Figure S4D). This plot also showed strong correlations between bacterial genera and 25 26 299 these metabolites. Following FMT or saline treatment there was a significant strong negative 27 28 -4 300 correlation between C difficile TVC and valerate (rS=-0.67, p=1.48x10 ), and significant strong positive 29

30 -6 31 301 correlations between C difficile TVC and 5-aminovalerate (rS=0.76, p=6.07x10 ), ethanol (rS=0.81, 32 -6 -5 33 302 p=1.72x10 ), and methanol (rS=0.72, p=2.98x10 ). 34 35 303 The identity of the proposed valerate peaks was confirmed in the chemostat culture supernatants 36 37 1 38 304 by performing 1D- H-NMR spectroscopy of a sample before and after valerate spike-in, as well as 2D- 39 40 305 1H-NMR spectroscopy of a valerate-containing sample (Figures S5 and S6, Supplementary Results). 41 42 43 306 The identities of other metabolites were also confirmed in the chemostat culture supernatants by 44 45 307 performing 2D-1H-NMR and STOCSY (Figures S7 and S8). 46 47 308 48 49 50 309 Bile acid UPLC-MS profiling of chemostat culture samples: 51 52 310 Due to the important role that bile acids play in C difficile spore germination and vegetative 53 54 311 growth,5 we also measured changes in bile acids over the course of the chemostat experiments. During 55 56 57 312 the clindamycin dosing period there was a significant increase in the conjugated primary bile acids 58 59 313 taurocholic acid (TCA, p=0.004), glycocholic acid (GCA, p=0.005), and glycochenodeoxycholic acid 60 61 62 13 63 64 65 314 (GCDCA, p=0.005), in the unconjugated primary bile acids cholic acid (CA, p=0.004) and 1 2 315 chenodeoxycholic acid (CDCA, p=0.037), and in the conjugated secondary bile acids taurodeoxycholic 3 4 5 316 acid (TDCA, p=0.045) and glycodeoxycholic acid (GDCA, p=0.004) (Figures 3 and S9). During the 6 7 317 clindamycin dosing period there was a significant decrease in the secondary bile acids deoxycholic acid 8 9 318 (DCA, p=0.006), lithocholic acid (LCA, p=0.005), and ursodeoxycholic acid (UDCA, p=0.037) (Figures 3 10 11 12 319 and S9). Following the end of the clindamycin dosing period the levels of these bile acids recovered to 13 14 320 steady state levels (before clindamycin dosing), and these levels were not affected by FMT treatment. 15 16 321 rCCA modelling was used to determine correlations between 16S rRNA gene sequencing data and 17 18 19 322 bile acid data during the clindamycin-dosing period. The unit representation plot showed separation 20 21 323 between cultures sampled before and during the clindamycin-dosing period along the first canonical 22 23 24 324 variate, but no separation between cultures sampled before and after the clindamycin-dosing period 25 26 325 (Figure S10A). The correlation circle plot showed that the separation between cultures sampled 27 28 326 before and during clindamycin dosing was due to increases in the levels of TCA, CA, CDCA, GCA, GDCA, 29 30 31 327 and GCDCA, and decreases in the levels of DCA, LCA, and UDCA during the clindamycin-dosing period 32 33 328 (Figure S10B). This plot also showed strong correlations between bacterial genera and bile acids. 34 35 329 There were significant strong correlations between C difficile TVC and several bile acids during the 36 37 -4 38 330 clindamycin dosing period, including TCA (rS=0.68, p=6.29x10 ), CA (rS=0.61, p=0.003), DCA (rS=-0.75, 39 -5 -5 40 331 p=8.86x10 ), and LCA (rS=-0.76, p=6.36x10 ). 41 42 43 332 44 45 333 GC-MS analysis of human stool samples: 46 47 334 To confirm the findings from our chemostat experiments we measured the levels of valerate in 48 49 50 335 human stool samples from healthy FMT donors, recurrent CDI patients pre-FMT and at several time 51 52 336 points post-successful FMT (1, 4, and 12 weeks after FMT treatment) (Figure 4). Valerate was depleted 53 54 337 in stool samples from recurrent CDI patients pre-FMT compared to healthy donors (p=0.0075). 55 56 57 338 Valerate levels were significantly increased in CDI patients post-FMT compared to pre-FMT (p=0.0007 58 59 339 at 1 week, 4 weeks, and 12 weeks). There were no significant differences in the levels of valerate in 60 61 62 14 63 64 65 340 stool samples from healthy donors compared to any of the time points from CDI patients collected 1 2 341 post-FMT (p>0.05 for all comparisons). 3 4 5 342 6 7 343 C difficile batch culture experiments with valerate and TCA: 8 9 344 Batch culture experiments were performed to directly study the effects of specific metabolites of 10 11 12 345 interest on C difficile germination and vegetative growth and to confirm the findings from our 13 14 346 chemostat experiments. These experiments showed that valerate inhibited the vegetative growth C 15 16 347 difficile ribotype 010 at concentrations ≥ 4 mM (p=0.008), ribotype 012 at concentrations ≥ 2 mM 17 18 19 348 (p=0.003), and ribotype 027 at concentrations ≥ 2 mM (p=0.008) (Figure 5). The concentration of 20 21 349 valerate in FMT-treated chemostat culture supernatants remained above 4 mM for all samples, 22 23 24 350 whereas the concentration of valerate in saline-treated cultures remained below 2 mM for all samples. 25 26 351 As a control, we also tested the effects of valerate on commensal gut isolates (B uniformis and B 27 28 352 vulgatus, two representatives of Bacteroidetes, and C scindens, a representative of Firmicutes). B 29 30 31 353 uniformis was only inhibited in broth containing 20 mM valerate (p=0.026). B vulgatus was not 32 33 354 inhibited at any concentration of valerate that was tested, and showed more growth in broths 34 35 355 containing 10 mM (p=0.020) or 20 mM (p=0.019) valerate. C scindens was not inhibited at any 36 37 38 356 concentration of valerate tested (p>0.05 for all concentrations of valerate tested). 39 40 357 Batch culture experiments also confirmed previous findings showing that TCA is required for C 41 42 43 358 difficile spore germination but had no effect on vegetative growth (Figure S11). 44 45 359 46 47 360 Glycerol trivalerate intervention study in a CDI mouse model: 48 49 50 361 Next, we determined whether valerate could inhibit C difficile growth when administered as an 51 52 362 intervention in a CDI mouse model (Figure 6A). To avoid the rapid uptake of valerate and obtain 53 54 363 sufficient release in the gastrointestinal tract, valerate was administered to the mice in the form of 55 56 32 57 364 glycerol trivalerate (which is hydrolysed by lipases to release valerate ). One-day post-infection (prior 58 59 365 to glycerol trivalerate or PBS administration) there was no significant difference in the levels of C 60 61 62 15 63 64 65 366 difficile TVC in mouse faeces in each group (p=0.584). Glycerol trivalerate or PBS was administered to 1 2 367 C difficile-infected mice by oral gavage 1, 2, and 3 days post-infection (n=5 per group). There was a 3 4 5 368 significant decrease in C difficile TVC in glycerol trivalerate-treated mice compared to PBS-treated mice 6 7 369 (Figure 6B, p=0.027 on day 2, p=0.007 on day 3, and p=0.024 on day 4). After only 3 doses of glycerol 8 9 370 trivalerate or PBS, glycerol trivalerate-treated mice had an average of 95% less C difficile TVC per gram 10 11 12 371 of faeces compared to PBS-treated mice. No adverse effects were noted in glycerol trivalerate-treated 13 14 372 mice over the course of the experiment. These results corroborate the other findings presented in this 15 16 373 study and support the therapeutic use of valerate to treat CDI. 17 18 19 374 20 21 375 DISCUSSION: 22 23 24 376 We used several ‘omic’ techniques to study the effects of FMT on CDI in a chemostat model under 25 26 377 tightly controlled conditions. The aim of our study was to directly link changes in C difficile counts to 27 28 378 changes in the structure and function of the cultured faecal microbiota. We confirmed our findings by 29 30 31 379 analysing human stool samples, performing C difficile batch culture experiments, and performing an 32 33 380 interventional study in a CDI mouse model. A summary of the key findings and proposed interactions 34 35 381 between C difficile, valerate, and TCA are outlined in Figure S12. 36 37 38 382 In our study we found valerate significantly inhibited the growth of C difficile, both in vitro and in 39 40 383 vivo. Valerate is a short chain fatty acid produced via amino acid fermentation by members of the gut 41 42 33 43 384 microbiota. Valerate was significantly depleted in chemostat culture samples during clindamycin 44 45 385 dosing and did not recover after clindamycin dosing was stopped. Valerate significantly increased with 46 47 386 FMT treatment and had a strong negative correlation with C difficile TVC. These findings were 48 49 50 387 corroborated with human data which showed valerate was depleted in recurrent CDI patient stool but 51 52 388 was restored following successful FMT. Batch culture experiments confirmed that valerate directly 53 54 389 inhibited the vegetative growth of several C difficile ribotypes, but had minimal effects on the growth 55 56 57 390 of other commensal gut bacteria tested. Moreover, we showed that glycerol trivalerate significantly 58 59 391 decreased C difficile TVC in a CDI mouse model. We hypothesise that maintaining or restoring the 60 61 62 16 63 64 65 392 levels of valerate in the gut microbiota of CDI patients will inhibit the vegetative growth of C difficile. 1 2 393 This restoration could be accomplished by directly supplying the gut with valerate (e.g. in the form of 3 4 5 394 glycerol trivalerate) or by administering bacteria capable of transforming valerate precursors into to 6 7 395 valerate. 8 9 396 There are several different metabolic pathways that lead to valerate production. 5-aminovalerate 10 11 12 397 is a product of the anaerobic degradation of protein hydrolysates by members of the gut microbiota, 13 14 398 in particular Clostridium species.34 In this pathway, proline is reduced to 5-aminovalerate by proline 15 16 399 reductase in a Stickland-type fermentation.35-37 5-aminovalerate is then fermented to valerate in a 17 18 19 400 series of reactions mediated by gut bacteria.34,38 We showed that 5-aminovalerate was increased in 20 21 401 chemostat cultures following clindamycin dosing and remained increased after clindamycin dosing 22 23 24 402 was stopped. This finding is supported by data from a patent submitted by Savidge and Dann, who 25 26 403 proposed CDI patients can be distinguished from non-infected subjects by measuring elevated levels 27 28 404 of 5-aminovalerate in the patient’s stool, urine, or blood.39 Moreover, Fletcher and colleagues showed 29

30 40 31 405 that 5-aminovalerate was elevated in the caecal content of C difficile infected mice. In another 32 33 406 metabolic pathway, some Clostridium species can ferment ethanol and propionate to valerate.41 We 34 35 407 found ethanol and propionate increased after clindamycin dosing and decreased following FMT 36 37 38 408 treatment. Taken together, changes in the levels of valerate and valerate precursors (5-aminovalerate, 39 40 409 ethanol, and propionate) over the course of our chemostat experiments supports our hypothesis that 41 42 43 410 disruption of the valerate pathway due to antibiotics results in an environment that permits C difficile 44 45 411 vegetative growth. 46 47 412 In this study we performed an intervention experiment to test the effects of valerate (in the form 48 49 50 413 of glycerol trivalerate) in a CDI mouse model. For the first time we showed that glycerol trivalerate- 51 52 414 treated mice had significantly reduced C difficile TVC compared to PBS-treated control mice (95% 53 54 415 reduction in glycerol-trivalerate treated mice after 3 doses). The results from our study are consistent 55 56 57 416 with the previously published study by Theriot and colleagues, who briefly mention that antibiotic- 58 59 417 exposed mice (who were susceptible to CDI) had a 66-fold decrease in valerate compared to non- 60 61 62 17 63 64 65 418 antibiotic controls (who were resistant to CDI).6 They also showed that antibiotic-exposed mice had 1 2 419 increased levels of amino acids required for C difficile growth (including proline, a precursor to 5- 3 4 5 420 aminovalerate) compared to non-antibiotic controls. Six weeks later, antibiotic-exposed mice were 6 7 421 fully-resistant to CDI following exposure to C difficile spores. As shown in the supplementary material 8 9 422 of their paper, Theriot and colleagues found that valerate levels had partially recovered six weeks 10 11 12 423 later, and were only 4.9-fold decreased in antibiotic-exposed mice compared to non-antibiotic 13 14 424 controls. 15 16 425 C difficile spores can persist following the cessation of antibiotics and germinate to vegetative cells 17 18 19 426 which initiate disease relapse. In our study we found that TCA, a known potent germinant for C difficile 20 21 427 spores,5 was increased during clindamycin dosing, remained elevated for several days after stopping 22 23 24 428 clindamycin, and had a strong positive correlation with C difficile TVC. However, after stopping 25 26 429 clindamycin and allowing chemostat communities to recover and stabilise, we found that bile acid 27 28 430 levels recovered to pre-clindamycin levels and did not change with FMT treatment. We hypothesise 29 30 31 431 that the transient increase in TCA found during and shortly after stopping clindamycin dosing 32 33 432 stimulated C difficile spore germination, resulting in vegetative growth. Once clindamycin dosing 34 35 433 stopped and the levels of TCA decreased to pre-clindamycin levels, C difficile spores had already 36 37 38 434 germinated and were growing in their vegetative state which is no longer affected by the presence of 39 40 435 TCA (as shown by Sorg and Sonnenshein5, and confirmed in our study using batch cultures). These 41 42 43 436 results are also consistent with the previously published study by Theriot and colleagues, who showed 44 45 437 that antibiotic-exposed mice (who were susceptible to CDI) had a 15-fold increase in TCA compared 46 47 438 to non-antibiotic controls (who were resistant to CDI).6 Six weeks later (when antibiotic-exposed mice 48 49 50 439 were fully-resistant to CDI) the levels of TCA in antibiotic-exposed mice recovered to the levels found 51 52 440 in non-antibiotic controls, suggesting there was insufficient levels of TCA to stimulate C difficile 53 54 441 germination. Further discussion of bile acid data is included in the Supplementary Discussion. 55 56 57 442 The first line of therapy for an initial episode of CDI is vancomycin/metronidazole therapy. These 58 59 443 antibiotics kill C difficile vegetative cells, but C difficile spores can persist.42 If 60 61 62 18 63 64 65 444 vancomycin/metronidazole therapy is stopped while these C difficile spores are present, the elevated 1 2 445 levels of TCA (present following the cessation of antibiotics) will cause C difficile spore germination, 3 4 5 446 and low levels of valerate will allow C difficile vegetative growth and therefore recurrent disease. To 6 7 447 prevent CDI initiation and relapse following the cessation of antibiotics, it is also important to maintain 8 9 448 low levels of TCA in the gut. One way to accomplish this would be to ensure the maintenance of bile 10 11 12 449 salt hydrolase enzymes during and after antibiotic exposure. These enzymes are produced by 13 14 450 commensal gut bacteria and are responsible for deconjugating tauro- and glyco-conjugated bile acids. 15 16 451 It has been proposed that antibiotics kill bile salt hydrolase-producing bacteria, resulting in the 17 18 19 452 accumulation of TCA in the gut. Therefore, re-inoculation of bile salt hydrolase-producing bacteria 20 21 453 with FMT may be responsible for degrading TCA present following the cessation of antibiotics, and 22 23 24 454 prevent the germination of C difficile spores. A more controlled method of restoring bile salt hydrolase 25 26 455 activity in the gut microbiomes of CDI patients could be to avoid the administration of live 27 28 456 microorganisms by administering purified bile salt hydrolase enzyme preparations. 29 30 31 457 We also found that succinate transiently increased in chemostat cultures with clindamycin dosing, 32 33 458 but negatively correlated with C difficile TVC. We believe the negative correlation between C difficile 34 35 459 TVC and succinate is because C difficile uses succinate for growth. A study by Ferreyra and colleagues 36 37 38 460 found that succinate was present at low levels in the guts of healthy mice, but transiently increased 39 40 461 with antibiotics or motility disturbance.8 They found that C difficile can metabolise succinate to 41 42 43 462 butyrate, and it exploits the increase in succinate to grow in perturbed gut microbiota. Again, these 44 45 463 findings are consistent with our findings from chemostat experiments. Members of Bacteroidetes and 46 47 464 the Negativicutes class of Firmicutes can degrade succinate to propionate, and as such succinate does 48

49 43 50 465 not usually accumulate to high levels in the guts of healthy humans. Another strategy to give C 51 52 466 difficile vegetative cells a competitive disadvantage would be to maintain succinate metabolism during 53 54 467 antibiotic exposure by administering succinate-degrading enzymes, so succinate is no longer available 55 56 57 468 for C difficile growth. 58 59 60 61 62 19 63 64 65 469 The findings from our study support the hypothesis that antibiotic exposure causes a depletion of 1 2 470 specific metabolic pathways normally found in healthy gut microbiotas, resulting in a metabolite 3 4 5 471 environment that favours C difficile germination and growth. It also supports our hypothesis that FMT 6 7 472 administration reverses these effects by restoring the bacteria responsible for performing these key 8 9 473 metabolic functions. These findings have the potential to directly impact clinical practice in the 10 11 12 474 foreseeable future by developing targeted treatments for CDI by different routes, alone or in 13 14 475 combinations: (1) directly supplement the gut with valerate (to inhibit C difficile vegetative growth); 15 16 476 (2) directly supplement the gut with bile salt hydrolase enzymes (to degrade taurocholic acid and 17 18 19 477 prevent C difficile spore germination); and (3) directly supplement the gut with succinate-degrading 20 21 478 enzymes (to degrade succinate and give C difficile vegetative cells a competitive disadvantage). These 22 23 24 479 proposed interventions are well-defined and represent safer options that will avoid all the risks 25 26 480 involved with administering live microorganisms to patients and will not promote antimicrobial 27 28 481 resistance. 29 30 31 482 The advantage of using valerate (instead of live microorganisms or relatively undefined FMT 32 33 483 preparations) is that valerate is a well-defined small molecule that is normally present in the healthy 34 35 484 gut. Valerate may be a preferred treatment for CDI in immunocompromised patients where 36 37 38 485 administration of bacteria may increase the risk of a subsequent infection, or in CDI patients that 39 40 486 require further antibiotic treatment for other purposes (where the antibiotic treatment would kill 41 42 43 487 bacteria present in FMT preparations and render the treatment less effective). Valerate could also be 44 45 488 delivered using more patient-friendly methods that do not require the need to preserve live 46 47 489 microorganisms. In this study glycerol trivalerate was orally gavaged to mice, but it has also been 48 49 50 490 included as an additive in animal feed. Therefore, valerate has the potential to be administered to 51 52 491 patients via a less invasive route compared to FMT. This promising new therapy merits further 53 54 492 evaluation in prospective studies in vivo. 55 56 57 493 58 59 494 REFERENCES: 60 61 62 20 63 64 65 495 1. Mitu-Pretorian OM, Forgacs B, Qumruddin A, et al. Outcomes of patients who develop symptomatic 1 2 496 Clostridium difficile infection after solid organ transplantation. Transplant Proc 2010;42:2631- 3 4 5 497 2633. 6 7 498 2. Ma GK, Brensinger CM, Wu Q, et al. Increasing incidence of multiply recurrent Clostridium difficile 8 9 499 infection in the united states: a cohort study. Ann Intern Med 2017;167:152-158. 10 11 12 500 3. van Nood E, Vrieze A, Nieuwdorp M, et al. Duodenal infusion of donor feces for recurrent 13 14 501 Clostridium difficile. N Engl J Med 2013;368:407-415. 15 16 502 4. Petrof EO, Gloor GB, Vanner SJ, et al. Stool substitute transplant therapy for the eradication of 17 18 19 503 Clostridium difficile infection: 'RePOOPulating' the gut. Microbiome 2013;1:3-2618-1-3. 20 21 504 5. Sorg JA, Sonenshein AL. Bile salts and glycine as cogerminants for Clostridium difficile spores. J 22 23 24 505 Bacteriol 2008;190:2505-2512. 25 26 506 6. Theriot CM, Koenigsknecht MJ, Carlson PE,Jr, et al. Antibiotic-induced shifts in the mouse gut 27 28 507 microbiome and metabolome increase susceptibility to Clostridium difficile infection. Nat Commun 29 30 31 508 2014;5:3114. 32 33 509 7. Weingarden AR, Chen C, Bobr A, et al. Microbiota transplantation restores normal fecal bile acid 34 35 510 composition in recurrent Clostridium difficile infection. Am J Physiol Gastrointest Liver Physiol 36 37 38 511 2014;306:G310-9. 39 40 512 8. Ferreyra JA, Wu KJ, Hryckowian AJ, et al. Gut microbiota-produced succinate promotes C difficile 41 42 43 513 infection after antibiotic treatment or motility disturbance. Cell Host Microbe 2014;16:770-777. 44 45 514 9. Ott SJ, Waetzig GH, Rehman A, et al. Efficacy of sterile fecal filtrate transfer for treating patients 46 47 515 with Clostridium difficile infection. Gastroenterology 2017;152:799-811.e7. 48 49 50 516 10. De Filippo C, Cavalieri D, Di Paola M, et al. Impact of diet in shaping gut microbiota revealed by a 51 52 517 comparative study in children from Europe and rural Africa. Proc Natl Acad Sci USA 53 54 518 2010;107:14691-14696. 55 56 57 519 11. Morrison DJ, Preston T. Formation of short chain fatty acids by the gut microbiota and their impact 58 59 520 on human metabolism. Gut Microbes 2016;7:189-200. 60 61 62 21 63 64 65 521 12. McDonald JAK. In vitro models of the human microbiota and microbiome. Emerging Topics in Life 1 2 522 Sciences 2017;1:373-384. 3 4 5 523 13. Macfarlane GT, Macfarlane S. Models for intestinal fermentation: association between food 6 7 524 components, delivery systems, bioavailability and functional interactions in the gut. Curr Opin 8 9 525 Biotechnol 2007;18:156-162. 10 11 12 526 14. McDonald JA, Schroeter K, Fuentes S, et al. Evaluation of microbial community reproducibility, 13 14 527 stability and composition in a human distal gut chemostat model. J Microbiol Methods 15 16 528 2013;95:167-174. 17 18 19 529 15. Van den Abbeele P, Grootaert C, Marzorati M, et al. Microbial community development in a 20 21 530 dynamic gut model is reproducible, colon region specific, and selective for Bacteroidetes and 22 23 24 531 Clostridium cluster IX. Appl Environ Microbiol 2010;76:5237-5246. 25 26 532 16. Freeman J, Baines SD, Jabes D, et al. Comparison of the efficacy of ramoplanin and vancomycin in 27 28 533 both in vitro and in vivo models of clindamycin-induced Clostridium difficile infection. J Antimicrob 29 30 31 534 Chemother 2005;56:717-725. 32 33 535 17. Baines SD, O'Connor R, Saxton K, et al. Comparison of oritavancin versus vancomycin as treatments 34 35 536 for clindamycin-induced Clostridium difficile PCR ribotype 027 infection in a human gut model. J 36 37 38 537 Antimicrob Chemother 2008;62:1078-1085. 39 40 538 18. Chilton CH, Crowther GS, Freeman J, et al. Successful treatment of simulated Clostridium difficile 41 42 43 539 infection in a human gut model by fidaxomicin first line and after vancomycin or metronidazole 44 45 540 failure. J Antimicrob Chemother 2014;69:451-462. 46 47 541 19. Chilton CH, Crowther GS, Baines SD, et al. In vitro activity of cadazolid against clinically relevant 48 49 50 542 Clostridium difficile isolates and in an in vitro gut model of C difficile infection. J Antimicrob 51 52 543 Chemother 2014;69:697-705. 53 54 544 20. Meader E, Mayer MJ, Steverding D, et al. Evaluation of bacteriophage therapy to control 55 56 57 545 Clostridium difficile and toxin production in an in vitro human colon model system. Anaerobe 58 59 546 2013;22:25-30. 60 61 62 22 63 64 65 547 21. Chilton CH, Crowther GS, Spiewak K, et al. Potential of lactoferrin to prevent antibiotic-induced 1 2 548 Clostridium difficile infection. J Antimicrob Chemother 2016;71:975-985. 3 4 5 549 22. Freeman J, Baines SD, Saxton K, et al. Effect of metronidazole on growth and toxin production by 6 7 550 epidemic Clostridium difficile PCR ribotypes 001 and 027 in a human gut model. J Antimicrob 8 9 551 Chemother 2007;60:83-91. 10 11 12 552 23. Mullish BH, Marchesi JR, Thursz MR, et al. Microbiome manipulation with faecal microbiome 13 14 553 transplantation as a therapeutic strategy in Clostridium difficile infection. QJM 2015;108:355-359. 15 16 554 24. Illumina I. 16S Metagenomic Sequencing Library Preparation (Part # 15044223 Rev. B); 2013. 17 18 19 555 Available at: https://support.illumina.com/content/dam/illumina- 20 21 556 support/documents/documentation/chemistry_documentation/16s/16s-metagenomic-library- 22 23 24 557 prep-guide-15044223-b.pdf, 2017. 25 26 558 25. Weljie AM, Newton J, Mercier P, et al. Targeted profiling: quantitative analysis of 1H NMR 27 28 559 metabolomics data. Anal Chem 2006;78:4430-4442. 29 30 31 560 26. Sarafian MH, Lewis MR, Pechlivanis A, et al. Bile acid profiling and quantification in biofluids using 32 33 561 ultra-performance liquid chromatography tandem mass spectrometry. Anal Chem 2015;87:9662- 34 35 562 9670. 36 37 38 563 27. Kao D, Roach B, Silva M, et al. Effect of oral capsule- vs colonoscopy-delivered fecal microbiota 39 40 564 transplantation on recurrent Clostridium difficile infection: a randomized clinical trial. JAMA 41 42 43 565 2017;318:1985-1993. 44 45 566 28. Garcia-Villalba R, Gimenez-Bastida JA, Garcia-Conesa MT, et al. Alternative method for gas 46 47 567 chromatography-mass spectrometry analysis of short-chain fatty acids in faecal samples. J Sep Sci 48 49 50 568 2012;35:1906-1913. 51 52 569 29. Winston JA, Thanissery R, Montgomery SA, et al. Cefoperazone-treated mouse model of clinically- 53 54 570 relevant Clostridium difficile strain R20291. J Vis Exp 2016;(118). doi:10.3791/54850. 55 56 57 571 30. Wolfer A. SANTA-App: Interactive package for Short AsyNchronous Time-series Analysis (SANTA) 58 59 572 in R, implemented in Shiny; 2017. Available at: https://github.com/adwolfer/SANTA-App, 2017. 60 61 62 23 63 64 65 573 31. Le Cao K, Rohart F, Gonzalez I, et al. mixOmics: Omics Data Integration Project. R package version 1 2 574 6.1.2.; 2017. Available at: https://CRAN.R-project.org/package=mixOmics, 2017. 3 4 5 575 32. Schwartz B. The effect of temperature on the rate of hydrolysis of triglycerides by pancreatic lipase. 6 7 576 J Gen Physiol 1943;27:113-118. 8 9 577 33. Neis EP, Dejong CH, Rensen SS. The role of microbial amino acid metabolism in host metabolism. 10 11 12 578 Nutrients 2015;7:2930-2946. 13 14 579 34. Barker HA, D'Ari L, Kahn J. Enzymatic reactions in the degradation of 5-aminovalerate by 15 16 580 Clostridium aminovalericum. J Biol Chem 1987;262:8994-9003. 17 18 19 581 35. Seto B, Stadtman TC. Purification and properties of proline reductase from Clostridium sticklandii. 20 21 582 J Biol Chem 1976;251:2435-2439. 22 23 24 583 36. Hodgins DS, Abeles RH. Studies of the mechanism of action of D-proline reductase: the presence 25 26 584 on covalently bound pyruvate and its role in the catalytic process. Arch Biochem Biophys 27 28 585 1969;130:274-285. 29 30 31 586 37. Seto B. The Stickland reaction. In: Knowles CJ, ed. Diversity of Bacterial Respiratory Systems (Vol. 32 33 587 II). Boca Raton, FL: CRC Press, 1980:49-64. 34 35 588 38. Buckel W. Unusual enzymes involved in five pathways of glutamate fermentation. Appl Microbiol 36 37 38 589 Biotechnol 2001;57:263-273. 39 40 590 39. Savidge T, Dann S. Methods and uses for metabolic profiling for Clostridium difficile infection. 41 42 43 591 2013;PCT/US2012/064218. 44 45 592 40. Fletcher JR, Erwin S, Lanzas C, et al. Shifts in the gut metabolome and Clostridium difficile 46 47 593 transcriptome throughout colonization and infection in a mouse model. mSphere 48 49 50 594 2018;3:10.1128/mSphere.00089-18. eCollection 2018 Mar-Apr. 51 52 595 41. Bornstein BT, Barker HA. The energy metabolism of Clostridium kluyveri and the synthesis of fatty 53 54 596 acids. J Biol Chem 1948;172:659-669. 55 56 57 597 42. Borody TJ, Khoruts A. Fecal microbiota transplantation and emerging applications. Nat Rev 58 59 598 Gastroenterol Hepatol 2011;9:88-96. 60 61 62 24 63 64 65 599 43. Louis P, Flint HJ. Formation of propionate and butyrate by the human colonic microbiota. Environ 1 2 600 Microbiol 2017;19:29-41. 3 4 5 601 6 7 602 Author names in bold designate shared co-first authorship. 8 9 603 10 11 12 604 FIGURE LEGENDS: 13 14 605 Figure 1: Average C difficile plate counts taken from VA (saline-treated cultures, black dashed line) and 15 16 606 VB (FMT-treated cultures, black solid line) over the course of the experiment. (A) Average C difficile 17 18 19 607 total viable counts. (B) Average C difficile spore plate counts. The grey shaded box indicates the 20 21 608 clindamycin-dosing period, while the vertical dotted line indicates the day of FMT or saline dosing. 22 23 24 609 Error bars represent the mean ± standard deviation (*, p<0.05 by SANTA analysis). 25 26 610 Figure 2: 1H-NMR metabolites that changed following clindamycin treatment and with FMT (VA= 27 28 611 saline-treated cultures, dashed line; VB= FMT-treated cultures, solid line). (A) valerate, (B) 5- 29 30 31 612 aminovalerate, (C) ethanol, (D) succinate, (E) propionate, and (F) methanol. The shaded grey box 32 33 613 indicates the clindamycin-dosing time period, while the vertical dotted line indicates the day of FMT 34 35 614 or saline dosing. SANTA analysis with Benjamini-Hochberg FDR was used to compare the following: 36 37 38 615 steady state cultures to clindamycin-treated cultures, steady state cultures to post-clindamycin 39 40 616 cultures, and FMT-treated cultures to saline treated cultures. 41 42 43 617 Figure 3: Bile acids that changed following clindamycin treatment and correlated with C difficile TVC 44 45 618 (VA= saline-treated cultures, dashed line; VB= FMT-treated cultures, solid line). (A) taurocholic acid 46 47 619 (TCA), (B) cholic acid (CA), (C) deoxycholic acid (DCA), and (D) lithocholic acid (LCA). The shaded grey 48 49 50 620 box indicates the clindamycin-dosing period, while the vertical dotted line indicates the day of FMT or 51 52 621 saline dosing. Steady state cultures were compared to clindamycin-treated cultures using SANTA 53 54 622 analysis with Benjamini-Hochberg FDR. 55 56 57 58 59 60 61 62 25 63 64 65 623 Figure 4: Effect of FMT on the concentration of valerate in stool from healthy FMT donors (n=5) and 1 2 624 recurrent CDI patients pre-FMT (n=16) and at several time points post-FMT (n=16). Mann-Whitney U 3 4 5 625 test for donors vs. pre-FMT, Friedman test for pre-FMT vs. post-FMT. ** p<0.01, *** p<0.001. 6 7 626 Figure 5: Valerate inhibits C difficile vegetative growth in batch cultures. Vegetative cells were 8 9 627 inoculated into supplemented brain heart infusion broth containing varying concentrations of valerate 10 11 12 628 (0, 1, 2, 3, 4, 5, 10, and 20 mM) and OD600 measurements were taken at 0, 2, 4, 6, and 8 hours. The 13 14 629 change in OD600 (from a time point during the exponential phase) was plotted against the 15 16 630 concentrations of valerate tested. (A) C difficile ribotype 010, (B) C difficile ribotype 012, (C) C difficile 17 18 19 631 ribotype 027, (D) B. uniformis, (E) B. vulgatus, (F) C. scindens. Error bars represent the mean ± standard 20 21 632 deviation, * p<0.05, ** p<0.01, *** p<0.001. 22 23 24 633 Figure 6: Glycerol trivalerate significantly reduced faecal C. difficle total viable counts in a CDI mouse 25 26 634 model. (A) Experimental design. Mice were given cefoperazone in their drinking water for 5 days, then 27 28 635 switched to antibiotic-free autoclaved water for the remainder of the experiment. On day 0 mice were 29

30 5 31 636 orally gavaged with 10 CFU of C difficile spores. On days 1, 2, and 3 mice were orally gavaged with 32 33 637 glycerol trivalerate (n=5) or PBS (n=5). (B) C difficile TVC were quantified from mouse faecal pellets on 34 35 638 days 1, 2, 3, and 4. The box excludes the upper and lower 25% (quartiles) of data, and the lines go to 36 37 38 639 maximum and minimum values. All data points are marked with “ᴏ” in the plot. Independent t-test of 39 40 640 log10-transformed C difficile TVC, *p<0.05, ** p<0.01. 41 42 43 641 44 45 642 TABLES: 46 47 643 Table 1: Time periods used for CDI chemostat experiments to compare the effects of FMT to saline. 48 49 Vessel Days 0-17 Days 18-24 Day 24 Days 25-31 Days 32-42 Day 42 Days 43-54 50 Added C difficile Control Stabilisation Stable Added Stabilisation 51 spores (day 25) and Added No vessel period communities C difficile period 52 clindamycin every saline intervention (VA) (no intervention) (no intervention) spores (no intervention) 53 12 hrs (days 25-31) Added C difficile 54 Test Stabilisation Stable Added Stabilisation spores (day 25) and Added No 55 vessel period communities C difficile period clindamycin every FMT intervention 56 (VB) (no intervention) (no intervention) spores (no intervention) 12 hrs (days 25-31) 57 58 644 59 60 61 62 26 63 64 65 Figure 1 Click here to download Figure Figure_1.tiff Figure 2 Click here to download Figure Figure_2.tiff Figure 3 Click here to download Figure Figure_3.tiff Figure 4 Click here to download Figure Figure_4.tiff Figure 5 Click here to download Figure Figure_5.tiff Figure 6 Click here to download Figure Figure_6.tiff Figure S1 Click here to download Figure Figure_S1.tiff Figure S2 Click here to download Figure Figure_S2.tiff Figure S3 Click here to download Figure Figure_S3.tiff Figure S4 Click here to download Figure Figure_S4.tiff Figure S5 Click here to download Figure Figure_S5.tiff Figure S6 Click here to download Figure Figure_S6.tiff Figure S7 Click here to download Figure Figure_S7.tiff Figure S8 Click here to download Figure Figure_S8.tiff Figure S9 Click here to download Figure Figure_S9.tiff Figure S10 Click here to download Figure Figure_S10.tiff Figure S11 Click here to download Figure Figure_S11.tiff Figure S12 Click here to download Figure Figure_S12.tiff "Lay Summary" and "Editor's Notes"

LAY SUMMARY: The short chain fatty acid valerate is depleted from the gut following antibiotics and restored with FMT. Valerate significantly decreased C. difficile growth in an interventional mouse model of infection.

EDITOR'S NOTE: BACKGROUND AND CONTEXT: Although FMT is effective at treating recurrent CDI there are concerns regarding its long-term safety. Mechanistic understanding of how FMT works is required to design targeted, well-defined, and safer therapies. NEW FINDINGS: Antibiotics depleted valerate from the gut (permitting C. difficile growth) while FMT restored valerate (decreasing C. difficile growth). Valerate significantly decreased C. difficile growth in an interventional CDI mouse model. LIMITATIONS: We have yet to identify the minimum dose of glycerol trivalerate required to eliminate C. difficile in mouse faeces. IMPACT: Valerate can be used as a safer, microorganism-free method to treat CDI, especially in vulnerable patients (e.g. immunocompromised or paediatric patients) or patients that require further antibiotic treatment.

Supporting Document Click here to download Supporting Document Supporting document no tracked changes.docx

1 SUPPLEMENTARY MATERIAL

2 SUPPLEMENTARY METHODS:

3 Bacterial strains:

4 C difficile DS1684 (ribotype 010, non-toxigenic strain) was used for chemostat experiments and mouse

5 experiments. C difficile DS1684, C difficile CD630 (ribotype 012, virulent multidrug resistant strain), C

6 difficile R20291 (ribotype 027, a hypervirulent strain), Bacteroides uniformis, Bacteroides vulgatus, and

7 Clostridium scindens (DSM 5676) were used in batch culture experiments. C difficile ribotype 027 is a

8 common ribotype in Europe and North America,1-3 while ribotype 012 is one of the common ribotypes in

9 mainland China.4,5 CD630 and R20291 are genetically and phenotypically well-characterised and are good

10 representatives of their ribotypes.6 B uniformis and B vulgatus were isolated from the stool of a healthy

11 male in his 30’s using fastidious anaerobe agar (Lab M, Heywood, UK) or nutrient agar (Sigma-Aldrich, St.

12 Louis, USA), respectively.

13

14 Chemostat model of CDI:

15 The working volume of each vessel was 235 ml and the growth medium feed was set to a retention

16 time of 21 hours.7,8 The composition of the growth medium consisted of a mixture of both soluble and

17 insoluble starches, amino acids, peptides, proteins, vitamins, trace elements, and porcine gastric mucin

18 (type II).9 To mimic the gut environment cultures were maintained at a temperature of 37ᵒC and a pH of

19 6.8, were gently agitated, and kept anaerobic by sparging with oxygen-free nitrogen gas. Chemostat

20 cultures were sampled daily from each vessel and vessels were operated for 54 days post-inoculation.

21 Chemostat culture samples were aliquoted and stored at -80ᵒC for DNA extraction and mass spectrometry

22 analysis. For NMR analysis, fresh chemostat culture was centrifuged at 20,000 x g and 4ᵒC for 10 minutes,

23 and the supernatant was aliquoted and stored at -80ᵒC.

24

1

25 Design of chemostat experiments:

26 We induced CDI in our chemostat gut model following a modified version of the methods previously

27 described by Freeman and colleagues (Table 1).10 Briefly, chemostat cultures were grown for 24 days

28 without experimental manipulation to allow the communities to stabilise. After sampling vessels on day

29 24 we added 7.8x106 C difficile spores to each vessel to achieve an initial concentration of 3.3x104

30 spores/mL.11 On day 25 we added another dose of 7.8x106 C difficile spores to both vessels, and

31 clindamycin was added to both vessels at a final concentration of 33.9 mg/L every 12 hours for 7 days

32 (from days 25-31). After stopping clindamycin dosing chemostat cultures were left to grow for 10 days

33 without experimental manipulation (days 32-42). This was done to allow the perturbed microbial

34 communities to stabilise, so we could more easily determine which bacteria or metabolites were altered

35 by FMT, and which bacteria or metabolites were able to recover after antibiotic treatment in the absence

36 of FMT. After sampling on day 42 we added a single dose of saline to VA (control vessel) and a single dose

37 of FMT to VB (test vessel). Chemostat cultures were then left to grow for a further 12 days without further

38 experimental manipulation to monitor the effects of FMT on the chemostat communities (days 43-54).

39

40 C difficile spore preparation:

41 C difficile spores were prepared using previously described methods.11 C difficile DS1684 was grown

42 anaerobically on fastidious agar plates supplemented with 5% defibrinated horse blood (VWR, Radnor,

43 USA) and incubated at 37ᵒC for 7 days. The growth was removed from the plates using a sterile loop and

44 resuspended in 1 mL sterile water. Next, 1 mL of 95% ethanol was mixed with the cell suspension and was

45 incubated for 1 hour at room temperature. The cell suspension was then centrifuged at 3000 x g and

46 resuspended in 1 mL sterile water. Spores were enumerated by preparing serial 10-fold dilutions in

47 phosphate buffered saline (PBS) (Sigma-Aldrich) and plating the dilutions on Braziers Cycloserine,

48 Cefoxitin Egg Yolk agar plates (containing Braziers CCEY agar base (Lab M), 250 mg/L cycloserine (VWR), 8

2

49 mg/L cefoxitin (Sigma-Aldrich), 8% egg yolk emulsion (SLS, Nottingham UK), 2% lysed defibrinated horse

50 blood (VWR), and 5 mg/L lysozyme (Sigma-Aldrich)).12 Plates were incubated anaerobically at 37ᵒC for 48

51 hours and the number of colonies were enumerated.

52

53 Enumeration of C difficile counts from chemostat culture samples:

54 C difficile total viable counts and spore counts were quantified from fresh chemostat culture samples

55 every other day starting 26 days post-inoculation. C difficile total viable counts (TVC) were enumerated

56 from fresh chemostat culture samples by performing serial 10-fold dilutions in PBS and plating onto

57 Brazier’s Cycloserine, Cefoxitin Egg Yolk agar plates (as described above, with the addition of 2 mg/L

58 moxifloxacin (VWR)) in triplicate using the Miles and Misra method.13 C difficile spore counts were

59 enumerated from alcohol-shocked chemostat culture samples by mixing an equal volume of fresh

60 chemostat culture sample with 95% ethanol and incubating at room temperature for one hour. Samples

61 were then centrifuged at 3000 x g and 4ᵒC for 10 minutes and resuspended in PBS. Spores were then

62 quantified by performing serial 10-fold dilutions in PBS and plating onto Brazier’s Cycloserine, Cefoxitin

63 Egg Yolk agar plates (as described above, without the addition of moxifloxacin) in triplicate using the Miles

64 and Misra method. Plates were incubated anaerobically at 37ᵒC for two days and colonies were

65 enumerated.

66

67 Preparation and instillation of FMT:

68 Fresh faecal samples were placed into an anaerobic chamber within 5 minutes of defecation. FMT

69 preparations were prepared by homogenising 10 g of stool in 100 mL of anaerobic 0.9% saline in a strainer

70 stomacher bag (250 rpm for 1 min). We added 50 mL of anaerobic saline to VA (control vessel) and 50 mL

71 of homogenised stool to VB (test vessel). For Run 1 and Run 2 the stool transplant was prepared from the

72 stool of a healthy male donor in his 30’s, and for Run 3 the stool transplant was prepared from the stool

3

73 of a healthy female donor in her 20’s. Both individuals have been used as FMT donors to treat CDI patients

74 in Imperial’s FMT Programme (and therefore undergone the appropriate donor screening protocols), and

75 had not taken antibiotics for at least 3 months prior to providing the stool sample.

76

77 DNA extraction:

78 DNA was extracted from 250 µL of chemostat culture using the PowerLyzer PowerSoil DNA Isolation

79 Kit (Mo Bio, Carlsbad, USA) following the manufacturer’s protocol, except that samples were lysed by

80 bead beating for 3 min at speed 8 using a Bullet Blender Storm instrument (Chembio Ltd, St. Albans, UK).

81 DNA was aliquoted and stored at -80ᵒC until it was ready to be used.

82

83 16S rRNA gene qPCR:

84 16S rRNA gene qPCR data was used to determine the total bacterial biomass within each sample and

85 was performed using extracted chemostat culture DNA to following a previously published protocol.14 A

86 total volume of 20 µL was used for each reaction and consisted of the following: 1x Platinum Supermix

87 with ROX (Life Technologies, Carlsbad, USA), 1.8 µM BactQUANT forward primer (5’-

88 CCTACGGGAGGCAGCA-3’), 1.8 µM BactQUANT reverse primer (5’-GGACTACCGGGTATCTAATC-3’), 225 nM

89 probe ((6FAM) 5’-CAGCAGCCGCGGTA-3’ (MGBNFQ)), PCR grade water (Roche, Penzberg, Germany), and

90 5 µL DNA. Each PCR plate included a standard curve using E. coli DNA (Sigma-Aldrich) (3-300,000 copies

91 per reaction in 10-fold serial dilutions) as well as no template negative controls. All samples, standards,

92 and controls were amplified in triplicate. Extracted DNA samples were diluted to ensure they fell within

93 the standard curve. Amplification and real-time fluorescence detections were performed using the

94 Applied Biosystems StepOnePlus Real-Time PCR System using the following PCR cycling conditions: 50 °C

95 for 3 min, 95 °C for 10 min, and 40 cycles of 95 °C for 15 sec and 60 °C for 1 min. We used a paired t-test

4

96 to compare changes in log-transformed 16S rRNA gene copy number between samples at specific time

97 points.

98

99 Pre-processing and analysis of 16S rRNA gene sequencing data:

100 We used the Mothur package (v1.35.1) to preprocess and analyse the resulting sequencing data

101 following the MiSeq SOP Pipeline.15 We used the Silva bacterial database for sequence alignments

102 (www.arb-silva.de/) and the RDP database reference sequence files for classification of sequences using

103 the Wang method.16 We determined the Operational Taxonomic Unit (OTU) taxonomies (phylum to

104 genus) using the RDP MultiClassifier script. We resampled and normalised data to the lowest read count

105 in Mothur (9527 reads per sample), which resulted in greater than 99.4% coverage within each sample.

106 We used 16S rRNA gene qPCR data and the following formula to express our 16S rRNA gene sequencing

107 data as absolute abundances (instead of relative abundances):

108 퐴푏푠표푙푢푡푒 푎푏푢푛푑푎푛푐푒 표푓 푡푎푥푎

16푆 푟푅푁퐴 푔푒푛푒 푐표푝푦 푛푢푚푏푒푟 𝑖푛 푠푎푚푝푙푒 109 = 푟푒푙푎푡𝑖푣푒 푎푏푢푛푑푎푛푐푒 표푓 푡푎푥푎 × ( ) ℎ𝑖푔ℎ푒푠푡 16푆 푟푅푁퐴 푔푒푛푒 푐표푝푦 푛푢푚푏푒푟 𝑖푛 푠푎푚푝푙푒 푠푒푡

110 The Shannon diversity index (H’), Pielou evenness index (J’), and richness (total number of bacterial

17 18 111 taxa observed, Sobs) were calculated using the vegan library within the R statistical package.

112 Stream plots were prepared by plotting the absolute abundance of 16S rRNA gene sequencing data

113 (biomass-corrected) over time (OTU-level, coloured by phylum). This was accomplished using the

114 streamgraph function within the streamgraph library (v0.8.1) within R.19

115

116 1H-NMR spectroscopy sample preparation:

117 Chemostat culture supernatants were randomized and defrosted at room temperature for 1 hour.

118 Once samples were defrosted supernatants were centrifuged at 20,000 x g and 4ᵒC for 10 minutes. Next,

119 400 µL of chemostat culture supernatant was mixed with 250 µL of sodium phosphate buffer solution

5

120 (28.85 g Na2HPO4 (Sigma-Aldrich), 5.25 g NaH2PO4 (Sigma-Aldrich), 1 mM TSP (Sigma-Aldrich), 3 mM NaN3

121 (Sigma-Aldrich), deuterium oxide (Goss Scientific Instruments, Crewe, UK) to 1 L, pH 7.4)20 and 600 µL was

122 pipetted into a 5 mm NMR tube.

123

124 Confirmation of NMR metabolite identities using 1D-NMR with spike-in and 2D-NMR spectroscopy:

125 We used the statistical total correlation spectroscopy (STOCSY) analysis method to aid in the

126 identification of metabolites in NMR spectra by determining correlations between intensities of the

127 various peaks across the whole sample.21 To further confirm if the peaks assigned to valerate and other

128 metabolites were correct, we also conducted a two-dimensional NMR spectra (including 1H−1H TOCSY and

129 1H−1H COSY) for the chemostat culture supernatant and valerate standard using typical parameters to

130 confirm the connectivity of the proton in the metabolites.22,23

131 For the valerate spike-in experiment one-dimensional 1H NMR spectra were acquired as described in

132 the 1H-NMR spectroscopy methods section from the main text, except 64 scans were recorded into 65536

133 data points with a spectral width of 20 ppm. After normal 1D 1H NOESY NMR acquisition, 10 µL of valerate

134 standard (99%, 0.9 M in PBS buffer) (Fisher Scientific, Hampton, USA) was added into the sample. A one-

135 dimensional spectrum was recorded again to see if the relevant peaks of valerate increased.

136

137 Data pre-processing and analysis of UPLC-MS bile acid data:

138 Quality control samples were prepared using a mixture of equal parts of the chemostat culture

139 supernatants. We used the quality control samples as an assay performance monitor and to guide the

140 removal of features with high variation.24 We also spiked quality control samples with defined mixtures

141 of bile acids to determine the chromatographic retention times of specific bile acids and to aid in

142 metabolite identification (55 bile acid standards, including 36 non-conjugated bile acids, 12 tauro-

143 conjugated bile acids, and 7 glyco-conjugated bile acids) (Steraloids, Newport, USA).

6

144 We converted the Waters raw data files to NetCDF format and extracted the data using XCMS (v1.50)

145 package implemented within the R (v3.3.1) software. Dilution effects were corrected for using

146 probabilistic quotient normalisation25 and chromatographic features with high coefficient of variation

147 (higher than 30% in the quality control samples) were excluded from further analysis.

148

149 Short AsyNchronous Time-series Analysis (SANTA):

150 SANTA is an automated pipeline that is implemented within R and controlled through a graphical user

151 interface developed with Shiny. 26,27 This method analyses short time series by estimating trajectories as

152 a smooth spline, and calculates whether time trajectories are significantly altered between different

153 groups or over different time periods. SANTA was used to make the following comparisons: stabilisation

154 period vs. clindamycin-dosing period, stabilisation period vs. post-clindamycin stabilisation period, and

155 FMT-treated vs. saline-treated cultures during the treatment period. We used mean subtraction to

156 eliminate between-run differences in metabolite concentrations that arose from differences in the stool

157 used to seed the chemostat vessels. For each metabolite, we calculated the mean for all samples within

158 the same chemostat run, then we subtracted the mean from all its values within the run.28 Depending

159 upon the time series being analysed, the number of degrees of freedom (df) to fit the spline model was

160 chosen to avoid overfitting the data (df = 3-5). We report the pDist values, which uses the area between

161 the mean group fitted curves to determine whether there is a difference between the two groups over

162 time. Analysis used 1000 permutation rounds to calculate p-values and 1000 bootstrap rounds to calculate

163 the 95% confidence bands. Reported pDist values are with Benjamini-Hochberg FDR correction, and p <

164 0.05 was considered significant.

165

166 Integration of 16S rRNA gene sequencing data and metabolite data:

7

167 We used regularised Canonical Correlation Analysis (rCCA) to correlate 16S rRNA gene sequencing data

168 (genus level) with bile acid mass spectrometry or 1H-NMR data from the same set of samples using the

169 mixOmics library within R.29 rCCA is an unsupervised method that maximises the correlation between the

170 two data sets X and Y (information on the treatment groups is not taken into account in the analysis). We

171 used the shrinkage method to determine the regularisation parameters. The plotIndiv function was used

172 to generate unit representation plots, where each point on the scatter plot represents a single chemostat

173 culture sample, and samples were projected into the XY-variate space. The plotVar function was used to

174 generate correlation circle plots, where strong correlations between variables (correlations greater than

175 0.5) are plotted outside of the inner circle. Variables are represented through their projections onto the

176 planes defined by their respective canonical variates. In this plot the variables projected in the same

177 direction from the origin have a strong positive correlation, and variables projected in opposite directions

178 form the origin have strong negative correlations. Variables with stronger correlations sit at farther

179 distances from the origin.

180

181 C difficile germination batch cultures with taurocholic acid (TCA):

182 To test the effects of TCA on C difficile germination we resuspended C difficile DS1684 spores in

30 183 supplemented brain heart infusion broth with or without 1% TCA (Sigma-Aldrich) in triplicate. The OD600

184 was measured immediately after inoculation of broths (time zero) and after an overnight incubation at

185 37ᵒC in anaerobic chamber. A paired t-test was used to determine whether TCA affected C difficile

186 germination.

187

188 C difficile vegetative growth batch cultures with TCA:

189 To test the effects of TCA on C difficile vegetative growth we centrifuged an overnight culture of C

190 difficile DS1684 at 3000 x g for 10 minutes and resuspended the cells in supplemented brain heart infusion

8

191 broth with or without 1% TCA (in triplicate). The OD600 was measured at time zero and cultures were

192 incubated at 37ᵒC in an anaerobic chamber. Additional OD600 measurements were taken at 2, 4, 6, and 8

193 hours post-inoculation, and the change in OD600 was plotted against time. A paired t-test was used to

194 determine whether TCA affected vegetative growth during the exponential phase.

195

196 SUPPLEMENTARY RESULTS:

197 C difficile total viable counts and spore counts:

198 There was no significant difference in C difficile TVC at the end of the clindamycin-dosing period

199 compared to TVC immediately prior to administering FMT or saline treatment (p>0.05). There was also no

200 significant difference in C difficile TVC in vessels assigned to receive FMT or saline treatment immediately

201 prior to administering the treatment (p>0.05).

202

203 1H-NMR spectroscopy:

204 Following FMT or saline treatment there were significant strong negative correlations between

-6 -5 205 valerate and 5-aminovalerate (rS=-0.76, p=5.27x10 ), ethanol (rS=-0.69, p=6.53x10 ), and methanol (rS=-

206 0.78, p=3.11x10-6).

207

208 Confirmation of valerate in chemostat culture supernatant by 1D- and 2D-NMR:

209 The chemical shifts for the 1D 1H-NMR spectrum of 99% valerate standard were: 0.9 (t), 1.3 (dt), 1.46

210 (m), 2.2 (t) (Figure S5A). Overlay of the 1D 1H-NMR spectrum of the valerate standard with the sample

211 showed that each peak of the valerate standard is visible in the sample (Figure S5B). Overlay of 1D 1H-

212 NMR spectra of the sample before and after valerate spike-in showed that all the valerate peaks increased

213 after spike-in (Figure S5C). For 2D-NMR analysis overlay of the 1H-1H COSY spectrum of the valerate

214 standard with the sample showed that each peak of the valerate standard was present in the sample

9

215 spectrum (Figure S6A). Overlay of 1H-1H TOCSY spectrum of the valerate standard with the sample showed

216 that each peak of the valerate standard was present in the sample spectrum (Figure S6B).

217

218 SUPPLEMENTARY DISCUSSION:

219 Valerate is not expected to be harmful to host gut cells, as exposure of gut organoids to 5 mM valerate

220 did not result in cell death or cause significant alterations in gene expression (Drs. Lee Parry and Richard

221 Brown of the European Cancer Stem Cell Research Institute, personal communication, 19 Dec. 2017).

222 Moreover, mice that received 15 mM glycerol trivalerate in our study did not show any adverse reactions.

223 The bacterial enzyme 7-α-dehydroxylase is responsible for converting unconjugated primary bile acids

224 CA and CDCA to the secondary bile acids DCA and LCA, respectively. DCA and LCA have been shown to

225 inhibit C difficile vegetative growth at specific concentrations,30,31 and a previous study has shown that

226 DCA and LCA were depleted in pre-FMT samples from recurrent CDI patients, but were restored in post-

227 FMT samples.32 These findings led researchers to propose antibiotic exposure results in the loss of bacteria

228 with 7-α-dehydroxylase activity, reducing DCA and LCA production and permitting C difficile vegetative

229 growth. A study by Buffie and colleagues found that administration of Clostridium scindens (a bacterium

230 with 7-α-dehydroxylase activity) was associated with resistance to C difficile by restoring the production

231 of the secondary bile acids DCA and LCA.33 However, in our study we found that the levels of DCA and LCA

232 recovered in chemostat cultures following the cessation of clindamycin. While we did find strong negative

233 correlations between C difficile TVC and the secondary bile acids DCA and LCA, recovery of these bile acids

234 to pre-clindamycin levels was not enough to decrease vegetative C difficile counts in chemostat cultures.

235 Indeed, while DCA can inhibit C difficile vegetative growth, it appears that DCA can also encourage spore

236 germination at specified concentrations.30 A better strategy to prevent CDI prior to antibiotic exposure

237 would be to prevent germination altogether by degrading TCA, a potent pro-germinant, using bile salt

238 hydrolase enzymes.

10

239 Our chemostat experiments more closely modelled the first episode of CDI and not recurrent CDI. In

240 first episodes of CDI, human patients are exposed to C difficile spores while taking an inciting antibiotic

241 (e.g. clindamycin). In our study clindamycin exposure was sufficient to deplete valerate and elevate levels

242 of TCA, allowing C difficile spore germination and vegetative growth. We waited 10 days after stopping

243 clindamycin dosing before administering the FMT preparation to allow the perturbed microbial

244 communities to stabilise following the cessation of antibiotics. This delay allowed us to more easily

245 determine which metabolites were altered only by FMT, and which metabolites were able to recover after

246 antibiotic treatment (in the absence of FMT). This feature is a major advantage of performing chemostat

247 studies, as it would be unethical to withhold treatment from recurrent CDI patients in human studies to

248 determine which metabolites would recover in the absence of FMT. We found no significant difference in

249 C difficile TVC at the end of the clindamycin-dosing period compared to TVC immediately prior to

250 administering FMT or saline treatment (see Supplementary Results). This means the metabolites that

251 recovered following the cessation of clindamycin dosing, but before FMT, did not affect the vegetative

252 growth of C difficile (i.e. bile acids). However, we did see a significant decrease in C difficile TVC and spore

253 counts after FMT. This decrease suggests that bacterial metabolites that decreased C difficile counts did

254 not recover after stopping clindamycin, but only recovered with FMT (i.e. valerate).

255 Most human studies of CDI have focused on recurrent CDI, not first episode of CDI. In this study we

256 showed that valerate was depleted in recurrent CDI patients pre-FMT, but was restored post-FMT. To our

257 knowledge this is the first study to measure valerate in the stool of recurrent CDI patients pre- and post-

258 FMT. A previous study by Weingarden and colleagues found that TCA was elevated in the stool of

259 recurrent CDI patients pre-FMT, but was decreased post-FMT.32 It is important to note that in these human

260 studies recurrent CDI patients were taking vancomycin when pre-FMT samples were collected, and FMT

261 was administered to recurrent CDI patients within 1-2 days of stopping vancomycin therapy. While we

262 could have designed our chemostat experiments to also include a vancomycin dosing regimen, followed

11

263 by FMT administration 1-2 days later, broad-spectrum vancomycin therapy would have killed more gut

264 bacteria and depleted the chemostat cultures of additional ecosystem functions that were not important

265 for the establishment of the infection, leading to false positives once these functionalities were restored

266 following FMT. In our chemostat experiments FMT was administered 10 days after stopping clindamycin

267 and pre-FMT samples were collected immediately prior to FMT administration. Had we chosen to

268 administer the FMT preparation within 1-2 days after stopping clindamycin we would expect TCA levels

269 to decrease with FMT.

270

271 REFERENCES:

272 1. Wilcox MH, Shetty N, Fawley WN, et al. Changing epidemiology of Clostridium difficile infection

273 following the introduction of a national ribotyping-based surveillance scheme in England. Clin Infect

274 Dis 2012;55:1056-1063.

275 2. Waslawski S, Lo ES, Ewing SA, et al. Clostridium difficile ribotype diversity at six health care institutions

276 in the United States. J Clin Microbiol 2013;51:1938-1941.

277 3. Davies KA, Longshaw CM, Davis GL, et al. Underdiagnosis of Clostridium difficile across Europe: the

278 European, multicentre, prospective, biannual, point-prevalence study of Clostridium difficile infection

279 in hospitalised patients with diarrhoea (EUCLID). Lancet Infect Dis 2014;14:1208-1219.

280 4. Huang H, Fang H, Weintraub A, et al. Distinct ribotypes and rates of antimicrobial drug resistance in

281 Clostridium difficile from Shanghai and Stockholm. Clin Microbiol Infect 2009;15:1170-1173.

282 5. Hawkey PM, Marriott C, Liu WE, et al. Molecular epidemiology of Clostridium difficile infection in a

283 major Chinese hospital: an underrecognized problem in Asia? J Clin Microbiol 2013;51:3308-3313.

284 6. Dawson LF, Valiente E, Donahue EH, et al. Hypervirulent Clostridium difficile PCR-ribotypes exhibit

285 resistance to widely used disinfectants. PLoS One 2011;6:e25754.

12

286 7. Allison C, McFarlan C, MacFarlane GT. Studies on mixed populations of human intestinal bacteria grown

287 in single-stage and multistage continuous culture systems. Appl Environ Microbiol 1989;55:672-678.

288 8. Duncan SH, Scott KP, Ramsay AG, et al. Effects of alternative dietary substrates on competition between

289 human colonic bacteria in an anaerobic fermentor system. Appl Environ Microbiol 2003;69:1136-1142.

290 9. McDonald JA, Schroeter K, Fuentes S, et al. Evaluation of microbial community reproducibility, stability

291 and composition in a human distal gut chemostat model. J Microbiol Methods 2013;95:167-174.

292 10. Freeman J, Baines SD, Saxton K, et al. Effect of metronidazole on growth and toxin production by

293 epidemic Clostridium difficile PCR ribotypes 001 and 027 in a human gut model. J Antimicrob

294 Chemother 2007;60:83-91.

295 11. Freeman J, O'Neill FJ, Wilcox MH. Effects of cefotaxime and desacetylcefotaxime upon Clostridium

296 difficile proliferation and toxin production in a triple-stage chemostat model of the human gut. J

297 Antimicrob Chemother 2003;52:96-102.

298 12. Crowther GS, Chilton CH, Todhunter SL, et al. Comparison of planktonic and biofilm-associated

299 communities of Clostridium difficile and indigenous gut microbiota in a triple-stage chemostat gut

300 model. J Antimicrob Chemother 2014;69:2137-2147.

301 13. Miles AA, Misra SS, Irwin JO. The estimation of the bactericidal power of the blood. J Hyg (Lond)

302 1938;38:732-749.

303 14. Liu CM, Aziz M, Kachur S, et al. BactQuant: an enhanced broad-coverage bacterial quantitative real-

304 time PCR assay. BMC Microbiol 2012;12:56-2180-12-56.

305 15. Kozich JJ, Westcott SL, Baxter NT, et al. Development of a dual-index sequencing strategy and curation

306 pipeline for analyzing amplicon sequence data on the MiSeq Illumina sequencing platform. Appl

307 Environ Microbiol 2013;79:5112-5120.

308 16. Wang Q, Garrity GM, Tiedje JM, et al. Naive Bayesian classifier for rapid assignment of rRNA sequences

309 into the new bacterial taxonomy. Appl Environ Microbiol 2007;73:5261-5267.

13

310 17. Oksanen J, Blanchet FG, Friendly M, et al. vegan: Community Ecology Package. R package version 2.4-

311 3.; 2017. Available at: https://CRAN.R-project.org/package=vegan, 2017.

312 18. R Core Team. R: A language and environment for statistical computing. R Foundation for Statistical

313 Computing, Vienna, Austria.; 2015. Available at: https://www.R-project.org/, 2017.

314 19. Rudis B. streamgraph: streamgraph is an htmlwidget for building streamgraph visualizations. R

315 package version 0.8.1.; 2015. Available at: http://github.com/hrbrmstr/streamgraph, 2017.

316 20. Beckonert O, Keun HC, Ebbels TM, et al. Metabolic profiling, metabolomic and metabonomic

317 procedures for NMR spectroscopy of urine, plasma, serum and tissue extracts. Nat Protoc

318 2007;2:2692-2703.

319 21. Cloarec O, Dumas ME, Craig A, et al. Statistical total correlation spectroscopy: an exploratory approach

320 for latent biomarker identification from metabolic 1H NMR data sets. Anal Chem 2005;77:1282-1289.

321 22. Liu Z, Wang L, Zhang L, et al. Metabolic Characteristics of 16HBE and A549 Cells Exposed to Different

322 Surface Modified Gold Nanorods. Adv Healthc Mater 2016;5:2363-2375.

323 23. Dai H, Xiao C, Liu H, et al. Combined NMR and LC-DAD-MS analysis reveals comprehensive

324 metabonomic variations for three phenotypic cultivars of Salvia Miltiorrhiza Bunge. J Proteome Res

325 2010;9:1565-1578.

326 24. Sangster T, Major H, Plumb R, et al. A pragmatic and readily implemented quality control strategy for

327 HPLC-MS and GC-MS-based metabonomic analysis. Analyst 2006;131:1075-1078.

328 25. Veselkov KA, Vingara LK, Masson P, et al. Optimized preprocessing of ultra-performance liquid

329 chromatography/mass spectrometry urinary metabolic profiles for improved information recovery.

330 Anal Chem 2011;83:5864-5872.

331 26. Chang W, Cheng J, Allaire JJ, et al. shiny: Web Application Framework for R. R package version 1.0.3.;

332 2017. Available at: https://CRAN.R-project.org/package=shiny, 2017.

14

333 27. Wolfer A. SANTA-App: Interactive package for Short AsyNchronous Time-series Analysis (SANTA) in R,

334 implemented in Shiny; 2017. Available at: https://github.com/adwolfer/SANTA-App, 2017.

335 28. Gratton J, Phetcharaburanin J, Mullish BH, et al. Optimized Sample Handling Strategy for Metabolic

336 Profiling of Human Feces. Anal Chem 2016;88:4661-4668.

337 29. Le Cao K, Rohart F, Gonzalez I, et al. mixOmics: Omics Data Integration Project. R package version

338 6.1.2.; 2017. Available at: https://CRAN.R-project.org/package=mixOmics, 2017.

339 30. Sorg JA, Sonenshein AL. Bile salts and glycine as cogerminants for Clostridium difficile spores. J

340 Bacteriol 2008;190:2505-2512.

341 31. Thanissery R, Winston JA, Theriot CM. Inhibition of spore germination, growth, and toxin activity of

342 clinically relevant C difficile strains by gut microbiota derived secondary bile acids. Anaerobe

343 2017;45:86-100.

344 32. Weingarden AR, Chen C, Bobr A, et al. Microbiota transplantation restores normal fecal bile acid

345 composition in recurrent Clostridium difficile infection. Am J Physiol Gastrointest Liver Physiol

346 2014;306:G310-9.

347 33. Buffie CG, Bucci V, Stein RR, et al. Precision microbiome reconstitution restores bile acid mediated

348 resistance to Clostridium difficile. Nature 2015;517:205-208.

349

350 Author names in bold designate shared co-first authorship

351

352 SUPPLEMENTARY FIGURE LEGENDS:

353 Figure S1: Stream plots showing the OTU abundances in each chemostat culture over time. Each stream

354 of colour represents an OTU, and streams are grouped by phylum: Bacteroidetes (blue), Firmicutes

355 (green), Proteobacteria (orange), Verrucomicrobia (purple), unclassified (grey), and C difficile (red). The

15

356 width of the stream represents the OTU abundance at each time point. The dotted box indicates the

357 clindamycin-dosing period, while the dotted vertical line indicates the day of FMT or saline dosing.

358 Figure S2: Diversity of bacterial communities cultured in chemostat vessels (VA= saline-treated cultures,

359 dashed line; VB= FMT-treated cultures, solid line). (A) Shannon diversity index (H’), (B) Richness (Sobs), (C)

360 Pielou’s evenness index (J’). The shaded grey box indicates the clindamycin-dosing period, while the

361 vertical dotted line indicates the day of FMT or saline dosing. SANTA analysis with Benjamini-Hochberg

362 FDR was used to compare the following: steady state cultures to clindamycin-treated cultures, steady

363 state cultures to post-clindamycin cultures, and FMT-treated cultures to saline treated cultures.

364 Figure S3: 1H-NMR metabolites that changed following clindamycin treatment and with FMT (VA= saline-

365 treated cultures, dashed line; VB= FMT-treated cultures, solid line). (A) butyrate, (B) acetate, (C)

366 isobutyrate, and (D) isovalerate. The shaded grey box indicates the clindamycin-dosing time period, while

367 the vertical dotted line indicates the day of FMT or saline dosing. SANTA analysis with Benjamini-Hochberg

368 FDR was used to compare the following: steady state cultures to clindamycin-treated cultures, steady

369 state cultures to post-clindamycin cultures, and FMT-treated cultures to saline treated cultures.

370 Figure S4: Regularized CCA (rCCA) model correlating 16S rRNA gene sequencing data (genus-level) and 1H-

371 NMR metabolite data. (A) The representation of units (a.k.a. samples) for the first two canonical variates

372 showing the correlations between variables before (grey), during (blue), and after (orange) the

373 clindamycin-dosing period. “A” represents samples collected from VA and “B” represents samples from

374 VB. (B) Correlation circle plot showing strong correlations between variables before, during, and after the

375 clindamycin-dosing period. Metabolites are shown in blue and bacterial genera are shown in orange.

376 Clostridium cluster XI (the clostridial cluster that includes C difficile) is shown in a black box. (C) The

377 representation of units (a.k.a. samples) for the first two canonical variates showing the correlations

378 between variables following FMT (blue) or saline (orange) treatment. “A” represents samples collected

379 from VA (saline-treated cultures) and “B” represents samples from VB (FMT-treated cultures). (D)

16

380 Correlation circle plot showing strong correlations between variables following FMT or saline treatment.

381 Metabolites are shown in blue and bacterial genera are shown in orange. Clostridium cluster XI (the

382 clostridial cluster that includes C difficile) is shown in a black box.

383 Figure S5: 1D 1H-NMR to confirm the identity of valerate in chemostat culture supernatants. (A) 1D 1H-

384 NMR spectrum of valerate standard (blue). (B) Overlay of 1D 1H-NMR spectrum of valerate standard (blue)

385 with sample spectrum (red). Each peak of the valerate standard is visible in the sample spectrum. (C)

386 Overlay of 1D 1H-NMR spectra of sample before (blue) and after (red) valerate spike-in. All the peaks

387 proposed to belong to valerate increased following spike in with valerate standard (green).

388 Figure S6: 2D 1H-NMR to confirm the identity of valerate in chemostat culture supernatants. (A) Overlay

389 of the 1H-1H COSY spectrum of valerate standard (blue) with sample spectrum (red). Each peak of the

390 valerate standard is visible in the sample spectrum. (B) Overlay of the 1H-1H TOCSY spectrum of valerate

391 standard (blue) with sample spectrum (red). Again, each peak of the valerate standard is visible in the

392 sample spectrum.

393 Figure S7: Overlay of 1H-1H COSY sample spectrum (blue) and 1H-1H TOCSY sample spectrum (red) to

394 confirm the identity of other metabolites found in chemostat culture supernatants.

395 Figure S8: Statistical total correlation spectroscopy (STOCSY). (A) 5-aminovalerate STOCSY spectrum

396 obtained by correlating all points in the spectra with the 5-aminovalerate resonance at 3.019 ppm. Peak

397 clusters with high correlations (*) correspond to positions where we expected to see peaks for 5-

398 aminovalerate. (B) Succinate STOCSY spectrum obtained by correlating all points in the spectra with the

399 succinate resonance at 2.408 ppm. No other peaks had high correlations with the peak at 2.408,

400 confirming this peak belonged to succinate.

401 Figure S9: Bile acids that changed following clindamycin treatment (VA= saline-treated cultures, dashed

402 line; VB= FMT-treated cultures, solid line). (A) taurodeoxycholic acid (TDCA), (B) glycocholic acid (GCA),

403 (C) glycodeoxycholic acid (GDCA), (D) glycochenodeoxycholic acid (GCDCA), (E) chenodeoxycholic acid

17

404 (CDCA), and (F) ursodeoxycholic acid (UDCA). The shaded grey box indicates the clindamycin-dosing

405 period, while the vertical dotted line indicates the day of FMT or saline dosing. Steady state cultures were

406 compared to clindamycin-treated cultures using SANTA analysis with Benjamini-Hochberg FDR.

407 Figure S10: Regularized CCA (rCCA) model correlating 16S rRNA gene sequencing data (genus-level) and

408 bile acid data. (A) The representation of units (a.k.a. samples) for the first two canonical variates showing

409 the correlations between variables before (grey), during (blue), and after (orange) the clindamycin-dosing

410 period. “A” represents samples collected from VA and “B” represents samples from VB. (B) Correlation

411 circle plot showing strong correlations between variables before, during, and after the clindamycin-dosing

412 period. Bile acids are shown in blue and bacterial genera are shown in orange. Clostridium cluster XI (the

413 clostridial cluster that includes C difficile) is shown in a black box.

414 Figure S11: TCA is required for C difficile spore germination, but has no effect on C difficile vegetative

415 growth. (A) C difficile spores were incubated supplemented brain heart infusion broth in the presence and

416 absence of 1% TCA and grown overnight. There was a significant increase in C difficile germination in the

417 presence of TCA (*** p<0.001). (B) C difficile vegetative cells were inoculated into supplemented brain

418 heart infusion broth in the presence and absence of 1% TCA. There were no significant differences in the

419 growth of C difficile in the presence or absence of TCA at any time point in the growth curve. Growth of C

420 difficile in the broths was quantified by taking OD600 measurements using a plate spectrometer. Error bars

421 represent the mean ± standard deviation.

422 Figure S12: Schematic of key metabolite interactions with C difficile during health, CDI, and recurrent CDI.

423 Initial antibiotic treatment decreases the diversity of the gut microbiota, killing bacteria that produce

424 valerate and bile salt hydrolase (an enzyme that degrades TCA). This results in increased levels of TCA and

425 decreased levels of valerate, allowing for C difficile spore germination and vegetative growth

426 (respectively). Treatment for CDI (vancomycin/metronidazole) decreases C difficile vegetative cells, but C

427 difficile spores remain and microbial community diversity remains low. Again, this antibiotic exposure

18

428 results in an environment with high TCA and low valerate, allowing the remaining C difficile spores to

429 germinate and grow once vancomycin/metronidazole is stopped. When a patient receives FMT for

430 recurrent CDI (usually within 1-2 days of stopping suppressive antibiotics) valerate levels (and valerate

431 producing bacteria) and bile salt hydrolase levels (and bile salt hydrolysing bacteria) are restored, resulting

432 in an environment that inhibits both C difficile germination and vegetative growth.

19

Revised Manuscript in Word or RTF (no changes marked)

Click here to access/download Revised Manuscript in Word or RTF (no changes marked) Manuscript main text no tracked changes.docx