Case Report: The effect of intravenous and oral antibiotics on the gut microbiome and breath volatile organic compounds over one year

Background: Antimicrobial resistance (AMR) is a global concern and better understanding of the gut microbiome, a known ‘amplifier’ of AMR, may allow future clinicians to tailor therapy to minimise this risk and offer a personalised medicine approach. To examine the gut microbiome, patients are required to provide faecal samples; more convenient and cheaper solutions need to be found. Methods: As part of a pilot study looking at how routes of administration affect the gut microbiome in NHS patients undergoing routine clinical management for infections, we hypothesised that effects on the gut microbiome varied with the route and metabolism of antibiotic used, and these changes may be reflected in breath metabolites. We present a case report of a patient with an unusual clinical history, alongside breath metabolite and gut microbiome data taken before, during and after antibiotic therapy over a period of one year. Results: We noted a shift in the dominant Bacteroides strain in the patient’s gut microbiome between pre- and post-therapy samples, along with an alteration in the composition of breath metabolites. Conclusions: This study provides a framework for similar future work and highlights the need for further research on the relationships between changes in microbial gut communities and antimicrobial exposure, patient clinical status, and the metabolites of human breath.

Methods: As part of a pilot study looking at how routes of administration affect the gut microbiome in NHS patients undergoing routine clinical management for infections, we hypothesised that effects on the gut microbiome varied with the route and metabolism of antibiotic used, and these changes may be reflected in breath metabolites. We present a case report of a patient with an unusual clinical history, alongside breath metabolite and gut microbiome data taken before, during and after antibiotic therapy over a period of one year.
Results: We noted a shift in the dominant Bacteroides strain in the patient's gut microbiome between pre-and post-therapy samples, along with an alteration in the composition of breath metabolites.
Conclusions: This study provides a framework for similar future work and highlights the need for further research on the relationships between changes in microbial gut communities and antimicrobial exposure, patient clinical status, and the metabolites of human breath.

Introduction
Scientific developments have enabled the possibility of personalised medicine at this time more than any other. In an age where antimicrobial resistance (AMR) is a global concern, insights into the gut microbiome, a known 'amplifier' of AMR 1 , may in the future aid clinicians in tailoring therapy for their patients to reduce this and other health risks. However, for patients, providing faecal samples can be an uncomfortable experience 2 , and the cost of sequencing individual patients' gut microbiomes, both in terms of time and resource, still remains prohibitive. Increasingly, scientists and clinicians are searching for more convenient options for patients, including the use of rectal swabs 3 .
Breath analysis, or breathomics, is an exciting but challenging field of science and medicine. Agreeable to the patient due to its non-invasive nature, it has great potential for individualized point-of-care diagnostics and screening 4 . Volatilomics, the study of Volatile Organic Compounds present in breath, is a systemic analysis, combining metabolic signals from the entire body through the blood-lung interface as well as the gut microbiota 5 .
In a pilot study looking at how routes of administration affect the gut microbiome in NHS patients undergoing routine clinical management for infections, we hypothesised that: i) the effects on the gut microbiome would vary with the spectrum of antibiotic used, the route of delivery, and the metabolism of the drug and ii) related effects may be identifiable in breath metabolites. Seven participants were recruited to assess the feasibility and suitability of such a study for NHS patients. Only patients who had not received antibiotics in the last 12 months were recruited.
Here we present a case report of one of the pilot study participants: a patient with an unusual clinical history, alongside breath metabolite and gut microbiome data taken before, during and after antibiotic therapy over a period of one year.
There are few such case reports in the literature to date.
Case report Patient A, a 57-year-old Caucasian male, presented to clinicians at Hull University Teaching Hospitals NHS Trust with an 11-year history of biliary system infection. In 2006 he had undergone a laparoscopic cholecystectomy for cholecystitis; this was complicated by a 'dropped' gallstone (a stone that remains in the abdominal cavity after surgery), subsequent recurrent hepatic infections and abdominal wall fistulation. Multiple surgical and radiological attempts at stone retrieval and drainage were eventually successful in 2016. However, six months later he continued to suffer recurrent infections and a CT scan in July 2017 revealed a persistent perihepatic abscess and hepaticcutaneous fistula devoid of visible calculi. Consequently, he was referred to the Outpatient Parenteral Antimicrobial Therapy (OPAT) service for consideration of prolonged antimicrobial therapy.
On examination of the patient, a fistula was evident in the right upper quadrant, which correlated with a sinus tract on his CT. Baseline physiological observations were normal. Blood tests showed a C-reactive protein (CRP) of 17 mg/L (normal range 0 to 8 mg/L) and total white blood cell count of 7.9 × 10 9 /L (normal range 4.0 to 11.0 × 10 9 /L) but were otherwise unremarkable. Based on the patient's relevant clinical microbiology results (Table 1), applicability to OPAT usage and longer-term tolerability, a decision was made to start intravenous (IV) Ertapenem 1g once daily. This was continued for 38 days, followed by oral Ciprofloxacin 750mg every 12 hours for 23 weeks.
Prior to starting therapy, Patient A was enrolled into the aforementioned pilot study following informed, written consent. Ethical approval for the study was granted by the HRA, Leicester REC 16/EM/0345 and the University of York Biology Ethics Committee. In total, seven faecal and seven breath samples were provided by Patient A between September 2017 and August 2018. The first samples were taken prior to starting antibiotic therapy; further sampling times are listed in Table 2. Patient A's diet included a daily cod liver oil capsule and regular diets but he did not take probiotic drinks/supplements before, or for the duration of, this study.

Results
Sequencing and assembly of faecal specimens Microbial DNA extracted from stool samples was subjected to long-and short-read DNA sequencing. The final polished assembly comprised of 9,554 contigs, N50 56,726bp and a total length of 159,132,373bp (see 6 for read statistics). 156,292 genes were identified using Prodigal version 2.6.3, within the anvi'o pipeline 7 . 71.7% of contigs were annotated, with 1,295 unique taxon names. CONCOCT annotation resulted in 81 bins from the data, and the manual tree-based approach resulted in 13 bins 8 . The manually curated clusters were favoured over those generated by CONCOCT due to 48 of the 81 bins not being resolved into a genus annotation. This is because annotation is based on the most abundant annotation within the bin, and so the most abundant genera may be unknown, and account for less than 30% of the contigs within the bin. CONCOCT bin annotations are provided 6 . All manually curated bins were annotated, with 8 of 13 being annotated as Bacteroides.
The annotation and semi-quantitative abundance relative to the stage of treatment are provided 6 .
Neither of the previous strategies were able to resolve the data into single genome bins, and both resulted in bins that were either overfilled or highly contaminated with contigs from other genomes, preventing resolution of the data to species level.

Amendments from Version 2
A paragraph has been added to the breath results section describing some preliminary analysis of any relationship between the breath metabolome and the gut microbiome in this patient and provides further context for part of our discussion.
Any further responses from the reviewers can be found at the end of the article A custom analytical pipeline written in-house (CLUSTard, Cansdale et al., in prep) was able to identify 19 metagenome assembled genomes (MAGs) that we assumed represented different species as shown in Figure 1A/B, which constituted over 97% of the patient's microbiome. The abundance of the additional annotated species is summarised within the "Other" bin within Figure 1A/B. These are cumulative abundances taken from all clusters containing the same species annotation, and so some of the species are explained through multiple bins. However, different MAGs with the same species identity have a similar trajectory for the majority of bases within that bin. Contigs that have the same annotation as any of the top 19 MAGs but are not included within Figure 1A/B are minimal, as shown in Figure 1C. These contigs had both a different trajectory and only represented a small percentage of the total bases of that bin. The full results showing the abundance per bin and per contig for the entire assembly are available as extended data 6 .
The two most abundant species throughout the time course are Bacteroides coprophilus and Bacteroides dorei ( Figure 1A). Bacteroides coprophilus is high in abundance during pre-treatment and the course of IV Ertapenem, whereas Bacteroides dorei is low in abundance during IV Ertapenem, but becomes highly abundant during oral Ciprofloxacin therapy, and following treatment.
Comparison of the MAGs used in the assembly for the species in Figure 1A/B compared to the National Center for Biotechnology Information (NCBI) reference genomes for the same species suggests that we do not have complete genomes for all of the top 19 most abundant species. The size of the assembled genomes relative to the NCBI reference for Oscillibacter (160%), Bacteroides dorei (124%), and Faecalibacterium prausnitzii (160%) suggest mis-assemblies that make the genomes larger than the expected size. However, the size of the assembled Bacteroides coprophilus genome is double that expected, which suggests that we could have two strains of Bacteroides coprophilus that have been assembled. There are two high quality MAGs that are singletons (one contig assembled genomes) that are annotated as Bacteroides coprophilus, Cluster_singleton2 and Cluster_singleton4, which are genome sized. However, they have the same trajectory suggesting that they respond to antibiotic treatment in the same way.
Three species, B. helcogenes, B. cellulosilyticus and Lactobacillus reuteri, are significantly increased in the post-treatment samples ( Figure 1A/B). Faecalibacterium is strongly suppressed during antibiotic treatment but recovers after treatment stops. Levels of Prevotella dentalis were higher prior to treatment with IV Ertapenem, and continued to reduce during oral Ciprofloxacin, remaining at low levels post-treatment. Parabacteroides is also more abundant during IV Ertapenem use and post-treatment but is suppressed during oral Ciprofloxacin treatment. There are three Alistipes species: two species are suppressed during oral Ciprofloxacin treatment, and displaced by Alistipes shahii, which is the most abundant during this phase. Following treatment there is a more equal mix of all three species, with Alistipes finegoldii and Alistipes sp.
Marseille not recovering to pre-treatment levels but Alistipes shahii remaining higher ( Figure 1A/B). Ruminococcus champanellensis significantly increased during treatment with IV Ertapenem but reduced during oral Ciprofloxacin therapy, remaining low post-treatment. Finally, three species were noted to become particularly abundant in the final sample taken on antibiotics: Lactobacillus reuteri, Akkermansia muciniphila and Bacteroides ovatus.
Resistance genes were identified in both the B. coprophilus and B. dorei reference genomes, and the most abundant MAGs in these two species. As suggested by the assembly size being double the size of the NCBI reference genome, the presence of two strains of B. coprophilus is also suggested by the identification of different resistance genes present within different B. coprophilus MAGs 6 . 75% of all B. coprophilus annotated bases were split between four MAGs, Cluster_contig0015 (15%), Cluster_contig0024 (19%), Cluster_singleton2 (26%) and Cluster_singleton4 (14%). The remaining bases are split across 14 lower abundance and quality clusters. 67% of all B. dorei annotated bases are within the MAG Cluster_contig0017. The remaining bases are split across > 20 other lower abundance and quality clusters. Cluster_contig0017 was used as the representative MAG for B. dorei for gene resistance annotation and the four aforementioned MAGs were used for B. coprophilus.
In B. coprophilus (more abundant during pre-treatment and IV Ertapenem) MAGs Cluster_singleton4 and Cluster_contig0015 have lost/have absent all three resistance genes adeF, tetQ and ErmF present in the NCBI reference. However B. coprophilus MAGs Cluster_singleton2 and Cluster_contig0024 have lost tetQ and ErmF genes, but have maintained an adeF resistance gene. Reaffirming the suggestion that there are two strains of B. coprophilus strains present. In B. dorei (more abundant during oral Ciprofloxacin and post-treatment), we observed an Erm 23S ribosomal RNA methyltransferase and the loss of a tetracycline antibiotic resistance gene 6 .
Blastp searches were performed to identify carbapenem resistance, the presence of gyrA and gyrB plasmids which have previously been identified within the Bacteroides fragilis group and shown to contribute to antibiotic resistance 9,10 , qnr and parC resistance genes. We found no hits that were significant (had an e-value of equal or less than 0.05) in the QNR resistance genes for either B. dorei or B. coprophilus. We did identify carbapenem resistance in B. dorei, with two hits, one 75% of the length of the resistance gene in Cluster_contig0017 on contig3934, with 70.9% percent identity and an e-value of 1.00 e-143 . The other matched 51% of the resistance gene length in the same cluster, but on contig0116 with an e-value of 6.00 e-29 , but only a percentage identity of 38.64%. B. coprophilus had one valid GyrA match, with a 95.4% coverage of the resistance gene length, 48% percentage identity and e-value of 0. However all QNR, GyrB and ParC hits were insignificant, with a low percentage of the resistance gene matching to the contigs, and a high e-value. (Supplementary file added to OSF repository).
Three human gut enterotypes have been identified based on the relative abundance of Bacteroides, Prevotella and Ruminococcus in the gut 11 . This patient identifies as a Type 1 enterotype in both pre and post samples due to the high abundance of Bacteroides. This is shown consistently between different clustering analyses. Despite no change in enterotype, the composition of the pre-and post-treatment microbiota is different. The Bacteroides genus remains the most abundant, however Bacteroides dorei displaces Bacteroides coprophilus and becomes the most abundant species.

Breath metabolites are atypical and vary in this patient
The breath samples from the patient showed substantial modification from ambient air masses, and for some volatiles the patient exhibited significantly different breath outcomes relative to other sampled and reported individuals. Most prominently different in the patient's breath were the methyl halides and thiol, including methyl chloride, methyl bromide, methyl iodide and methane thiol (CH 3 X, where X = Cl, Br, I or SH and shown as MeCl, MeBr, MeI and MeSH, Figure 2).
In urban air masses methyl halides are found at concentrations of approximately 750, 15 and 5 pptv, respectively 12 . Humans metabolise these compounds (both in terms of uptake and production) leading to average exhaled breath containing 3500 ± 2400 parts-per-trillion-by-volume (pptv) MeCl, 6.5 ± 11 pptv MeBr and 1.1 ± 1.8 pptv MeI. In contrast, the patient's average breath profile contained 13,000 ± 2000 pptv MeCl, 73 ± 27 pptv MeBr and 3.3 ± 0.8 pptv MeI. These breath concentrations were significantly different than the average male population for both MeCl and MeBr and consistently more concentrated for all methyl halides. Both methyl chloride and methyl bromide showed consistency from prior to antibiotics and throughout IV Ertapenem therapy (13400 ± 1200 pptv MeCl, 66 ± 9 pptv MeBr) but were significantly reduced during oral Ciprofloxacin (8900 pptv MeCl, 49 pptv MeBr at second sampling, 78 days into oral therapy). Methyl iodide concentrations in the patient's breath did not notably change over time during antibiotic treatment.
Atmospheric data within urban environments is scarce but our previous breath results suggest that methanethiol (MeSH) is common in human breath at concentrations of 1. Other human-metabolised halogenated compounds showed interesting outcomes in Patient A, including dichloromethane (CH 2 Cl 2 ) and chloroform (CHCl 3 ). Ambient urban air contains approximately 5600 and 1300 pptv, respectively, while average healthy human breath is generally less concentrated, although highly variable, at 1000 ± 1500 pptv for dichloromethane and 230 ± 660 pptv for chloroform. The patient exhibited uniform, and depleted, concentrations for dichloromethane throughout the study, 20 ± 6 pptv dichloromethane, while chloroform was uniformly low but highly variable, at 95 ± 71 pptv.
Concentrations of chloroform showed a strong time, and possibly therefore antibiotic dependency, with starting concentrations near 105 pptv that fell throughout IV Ertapenem to a level below the detection limit (<5 pptv) by the start of oral Ciprofloxacin, returning to approximately the same concentrations as the earliest sample post-treatment (~105 pptv).
The sulfur-containing compound, dimethyl sulfide (DMS) was present in the patient's breath, ranging from 900 to 16000 during the various stages of treatment. While the average breath concentration of DMS (4700 ± 4900 pptv) in patient A was similar to human breath averages (2300 ± 2300 pptv DMS), this was primarily due to the high variability of this data across individuals and populations. DMS concentrations appeared to be treatment dependent, with a substantial increase in DMS during IV Ertapenem therapy (from 2900 to 16000 pptv) before falling during oral Ciprofloxacin to near average healthy human concentrations (~2800 pptv).
Isoprene, the most abundant human metabolite in breath, was similar between the patient (125 ± 23 parts-per-billion-by-volume; ppbv) and the healthy human population (185 ± 205 ppbv). Baseline concentrations of isoprene can vary substantially between individuals (from 80 to 300 ppbv within our healthy controls), but within the established baseline it tends not to vary dramatically, which is what we observed in the patient post-antibiotic therapy when isoprene levels rose to 178ppbv.
A PCA plot (not included) was conducted to look at the relationship between breath and faecal data for this patient over time. Only methyl chloride was seen to be a significant contributor to the first and second principal components, and the methyl halides were negatively correlated with Lactobacillus reuteri, Akkermansia muciniphila, and Bacteroides ovatus, suggesting that either: i) methylated sulfur and halogen compounds are more concentrated when these bacteria abundances are low (or that methyl halides and methane thiol concentrations are low when these microbial abundances are high); or ii) that these microbes may be closely identified to other community cohorts that are more directly metabolically connected to these gaseous metabolites.
Clinical follow-up Patient A symptomatically improved during treatment, which resulted in clinical and radiographical resolution of the hepatic-cutaneous fistula. A repeat CT scan seven weeks into antimicrobial therapy suggested the patient's infection had not fully resolved, but a further CT scan at 16 weeks of therapy showed no evidence of ongoing infective changes and resolution of the previous peri-hepatic abscess (see Figure 3). He was discharged from clinical follow up in March 2019 with a CRP of 3.1 mg/L and has not suffered recrudescence since.

Discussion
Understanding and mitigating the consequences of antimicrobial exposure on the human gastrointestinal tract microbiome at an individual level is likely to become increasingly important in the era of personalised medicine, given our increasing understanding of the associations between the nature of an individual's gastro-intestinal (GIT) microbiome and the risk of a disease state. The GIT microbiome is also recognised as an 'amplifier' of antimicrobial resistance and changes in an individual's microbiome may contribute to this. In this pilot study, we sought to describe the changes occurring in patients' gut microbiome during IV and oral antibiotic therapy, and assess whether these changes were reflected in breath metabolites.
In Patient A, B. coprophilus, a bacterium that has not been associated with human infection, was in relative abundance throughout IV Ertapenem exposure (suggesting the presence of a resistance mechanism), suppressed during oral ciprofloxacin, and abundant again post-therapy. Low GIT exposure is unlikely to be the cause of abundance during Ertapenem therapy as 10% of a 1g dose is excreted in human faeces (Ertapenem 1g Powder for Concentrate for Solution for Infusion, Electronic Medicines Compendium, 2020). Meropenem exposure, however, has previously been associated with the emergence of carbapenem resistance in Bacteroides fragilis, which is commonly associated with the cfiA metallo-beta-lactamase gene 14 . Fluoroquinolone susceptibility in B. fragilis in the presence of carbapenem resistance has been described in bloodstream infection 15 , although resistance specifically to Ciprofloxacin in B. fragilis strains is generally expected to be high 16 . However, in Patient A, the abundance of the adeF gene throughout, which has been associated with efflux-mediated resistance to fluoroquinolones and tetracyclines in clinical practice, suggests that this gene was not expressed by B. coprophilus during Ciprofloxacin exposure.
In contrast, B. dorei was more abundant during oral Ciprofloxacin exposure and post-therapy suggesting susceptibility to Ertapenem (despite the presence of a carbapenem resistance gene on blastp search) and resistance to Ciprofloxacin, in keeping with potential expression of the adeF gene or another mechanism. Furthermore, the loss of ermF and tetQ from each of the most abundant Bacteroides spp., along with the subsequent gain of ermG in the prevailing most abundant species, B. dorei, highlights the impact of antimicrobial exposure on AMR within the gut microbiome, which may not always be intuitively related to the class of antimicrobial a patient is exposed to 17,18 . The contrasting states of B. coprophilus and B. dorei with their respective genomic data may represent the important differences between phenotypic and genotypic expression by bacteria although without parallel transcriptomic data, it is not possible to comment further.
Faecalibacterium prausnitzii has been proposed as a biomarker of gut health. An important producer of butyrate, studies have demonstrated an association between gut disease and low levels of Faecalibacterium in patients with inflammatory bowel disease 19,20 . Patient A's levels of Faecalibacterium remained suppressed during therapy, almost completely recovering to pre-therapy levels on discontinuation. A recent study in mice treated with Faecalibacterium supplementation and a prebiotic showed clearance of bacterial burden of Clostridium difficile and potential reduction in toxin production compared with untreated mice 21 . Interestingly, Patient A's levels of Clostridium were low before and after antibiotic exposure, but were high at times during exposure to both antibiotics. The patient did not develop diarrhoea while on antibiotic therapy.

Bacteroides ovatus, Lactobacillus reuteri and Akkermansia muciniphila all showed marked increases at time point 5 of antimicrobial therapy (after 78 days of oral Ciprofloxacin).
Akkermansia has been labelled a "sentinel of gut health" due to its apparent effect of decreasing gut permeability and its low levels in patients with diabetes and obesity [22][23][24] . Similarly, studies of Lactobacillus reuteri have proposed possible anti-inflammatory effects 25,26 . Bacteroides ovatus has been shown to reduce to LPS-mediated inflammation in animal studies and its use as a probiotic continues to be investigated 27,28 . Concurrently there were appreciable decreases in the levels of methylated compounds in Patient A's breath sample. These findings may be associated with the recovery in Patient A's clinical status. Further research is warranted to investigate whether markers such as these in patients' breath and/or faecal samples have the potential, in real-time, to indicate whether a review of the need for ongoing antimicrobial therapy is required. This highlights the individuality of each person's results 30 and the importance of large sample sizes of the population in understanding what may or may not be statistically generalisable in personalised medicine. However, the broad pattern of changes may in fact be more universal, as demonstrated through "electronic noses": the use of breath metabolites in the diagnosis of human disease to assess patterns of volatile organic compounds (VOCs) associated with specific illnesses 29,31 . There is increasing evidence that exhaled breath and the gut microbiome are strongly correlated 32-34 . The method of breath analysis we used for Patient A is more amenable to a personalised medicine approach but pattern changes over time may carry more significance than individual VOCs obtained at single timepoints during patient care. The inverse association of chloroform levels and antibiotic therapy may be of interest; if such relationships can be robustly established, it may be possible to monitor aspects of drug therapy at the bedside via the breath, for example, rather than current invasive techniques.
Most VOCs in Patient A's breath did not return to starting-point values on completion of therapy, suggesting that microbial/ human metabolisms within the patient had been substantially modified during illness recovery and/or antimicrobial exposure. Combined with the shift demonstrated in the dominant Bacteroides strain within the bowel, as well as significant differences in the microbial community pre-and post-therapy, this suggests potentially measurable modification of microbial communities in patients receiving prolonged antimicrobial therapy. The relevance of these specific community and metabolism changes to health in the longer-term is yet to be understood, particularly as the shifts may represent a return to a pre-or post-illness healthier state or an adjustment to physiological niches held by certain strains that bear no impact on overall function. After all, Patient A clinically improved and became both subjectively and objectively healthier over time, but it is still possible that persisting changes in Patient A's microbial community as a result of prolonged antimicrobial exposure may have longer-term health implications.
The complex interplay between different bowel flora in vivo in humans remains an important driver for ongoing research of the gut microbiome. Limitations of this pilot project include the likelihood that breath metabolites may be affected by microbiomes or human metabolisms other than those within the gut; for example, the oral and lung microbial communities. However, this study provides a framework for similar future work and highlights the need for further research into the relationships between changes in microbial gut communities during antimicrobial exposure, AMR colonisation and subsequent infection risk, monitoring of patient clinical status, and the metabolites of human breath.

Sequence assembly
Sequences from all timepoint samples were pooled for assembly, using long reads for the original assembly, and short reads for polishing the assembly. Sequencing data was basecalled with Guppy 1.8.3, and demultiplexed with Porechop 0.2.3. Nanopore reads were assembled with canu 36 1.8 using parameters genomeSize=100m, corMinCoverage=0, corOutCoverage=all, corMhapSensitivity=high. The canu assembly was polished with the nanopore reads using medaka 0.7.1 and with the Illumina reads using Pilon 37 1.23, running Pilon three times, aligning reads to the assembly with bwa 38 mem 0.7.17 and indexing and sorting read alignments with SAMtools 39 1.9. All tools were run with default options unless otherwise stated.

Sequence analysis
The assembly was filtered to remove contigs less than 1000bp, (see extended data 6 for contig length distribution) before being annotated and clustered using the anvi'o metagenomic workflow using the centrifuge nt database 7,40 . The short reads were used for clustering and taxonomic assignment and abundance was calculated based on the depth of coverage of the short reads mapped to the assembly. Clustering was performed using both an automated method using CONCOCT and manually using hierarchical clustering of contigs 8 . Taxonomy was generated based on the contigs, and the abundance of these taxon through the coverage of the short reads mapped to the assembly using bwa mem 0.7.17 38 .
The metagenomic clustering pipeline CLUSTard (public publication pending) was used to generate clusters of contigs based on mapped short read abundance, using a kraken2 database. Pearson coefficient correlation values were varied to determine percentage similarity in trajectory between contigs within a bin over time. Values between 0.997 and 0.8 were used to assess clustering, with 0.997 used in further analyses.
Contigs from both highly abundant clusters in Bacteroides coprophilus (Cluster_singleton2, Cluster_singleton4, Cluster_ contig0024 and Cluster_contig0015) and the most abundant cluster in Bacteroides dorei (Cluster_contig0017), were annotated using the CARD: RGI database 41 to identify antimicrobial resistance genes that could be responsible for their trajectories throughout treatment. The NCBI reference genomes for Bacteroides salanitronis (CP002530.1) and Bacteroides dorei (CP008741.1) were used to identify expected resistance genes in the assembly and identify subsequent losses or gains in the contigs within these MAGs. Bacteroides salanitronis was used as a reference because it is closely related, and Bacteroides coprophilus is not present currently in this database. These MAGs were mapped to the NCBI references of both Bacteroides Coprophilus and Bacteroides Salanitronis using default options in minimap2 42 . There was no significant difference in the percentage of sequence mapping to these references, which we were satisfied meant that the genomes were similar enough that mapping to Bacteroides Salanitronis using the CARD RGI database would not impact on the resistance genes identified.
Seqkit version v0.11.0 43 was used to convert contig nucleotide sequences to amino acid sequences in order to perform blastp searches with blast+ 2.2.31+ against additional resistance genes. The contigs from these high abundant Bacteroides coprophilus and Bacteroies dorei were compared against gyrA, gyrB and parC sequences from biocyc accessions EG10423 and EG10424 and EG10686 respectively. Carbapenem resistance was identified using accession EEF78324.1 from NCBI as previously described by Goto et al. 43 . We also searched for QNR resistance, by using the QNRA alleles QNRA1-6 which have been identified in the Bacteroides fragilius group 44,45 . Hits were filtered so that only a e-value of less than 0.05 was used. The percentage identity and length of resistance gene covered were also used to determine significance.
Breath volatile metabolite analysis Tedlar bags provide ease of sampling for patients but are unsuitable for long term storage of volatiles. To ensure stable long-term storage we transferred the breath sample from the Tedlar bag to high-quality stainless steel canisters (electropolished prior to single welding) on the same day. Prior to sample transfer from the Tedlar bags the canisters were evacuated (<0.1mbar pressure). To ensure stability and sensitivity of analysis each sample passed through an Ascarite trap, driven by the pressure differential between the Tedlar bag and canister. The volatiles within this study have been shown previously 46 to be stable in these canisters for over two weeks.
Canister-bound breath samples were concentrated onto a liquid nitrogen condensation trap through pressure differential, driven by a two-stage rotary pump, then run on a Gas Chromatograph-Mass Spectrometer (GC-MS) consisting of a Restek© PoraBond Q column (30m, 0.32mm ID, 0.5-μm thickness) within a HP 5972 MSD, running in selective ion mode, within six days of the date of collection 46 . We used CFC-11 as an internal standard since it is ubiquitous and constant in the atmosphere and there are no known metabolisms that produce or consume it within the human body.
Concentration calibrations were performed using purchased, low-concentration (ppbv) standards (BOC Specialty Gases; CFC-11, CFC-12, CFC-113, methyl halides, DMS) directly injected into the condensation system. Within our controlled temperature and volume trap, injection of a limited pressure of high concentration gas (the standards) mimics the same total amount of a greater pressure injection of a low concentration sample (breath samples). Accuracy of these standards for breath sample concentrations were ±10%. Alternatively, serial dilutions of headspace volatiles from neat standards were used (as in isoprene, methane thiol). These standards were less reliable than the prepared standards because of the small scale on which the system is operated as well as differences in reported vapor pressures for each compound. We calculated accuracy for quantification of samples based on these standards to be ±20%.

Consent
Written informed consent for publication of their clinical details and images was obtained from the patient.

Data availability
Underlying data Whole metagenome sequencing, using long and short reads for the assembly of a gut microbiome. Many thanks for your comments and your time, and apologies for the delay in responding. Primarily this pilot study was to look at how routes of of administration affect the gut microbiome in NHS patients undergoing clinical management for infections, and to assess the feasibility and suitability of such a study in these patients. As part of this work we wanted to explore our hypothesis that related effects may be seen in breath metabolites. Unfortunately, answering this question fully and completing a full functional analysis remains outside the scope of this small project and case study, particularly as this patient's breath metabolites appear atypical. However, we accept your point that our exploration of these data is not explicitly stated within the analysis/discussion. We have therefore made some changes to the end of the breath results section that further explains some of our inferences in the discussion as well. For your information, we have also included some additional thoughts below: During our analysis, we conducted a PCA (not included for the reasons discussed forthwith) and other than what we see and describe in the text already (that dorei and coprophilus are replacing each other), we were mostly unable to see direct correlations between the metabolites and the metagenome. However, we did see a negative correlation between methyl chloride and a number of bacteria (see updated end of breath results section).
There may be several reasons for a relationship between the breath and faecal metabolic data not being more clear. Firstly, the analysis was metagenomic, not transcriptomic, so there may have been dead biomass seen in the analysis that shrouds the actual changes occurring; secondly, faecal samples may not be fully representative of the entire gut microbiome and there may therefore be alterations seen in breath that add to aspects of the overall picture but that do not directly correlate with faecal data; and thirdly, we measured several metabolites but not a whole metabolome and we therefore might not have seen more obvious trajectories in line with the changing of the organisms present that may occur if the whole metabolome was analysed. Within the discussion we have attempted to draw inferences regarding this patient's breath data and their microbiome composition over the course of therapy, for example regarding Bacteroides (ovatus), but we have now clarified this further in the results section. Some points are needed to be clarify to improve this piece of work:

Pisut Pongchaikul
In Figure 1A, are these percentage shown with respect to the total abundance of each species or to the total bacterial species at each time point? ○ Stack plot showing the abundance of bacterial species at each time point should be shown to illustrate the dynamic of the species.
○ metagenome sequencing was done it should be possible to do functional analysis to interrogate this question.
My second major concern relates to conclusions about absolute abundance made on the basis of relative abundance measures, see below.

INTRODUCTION
It probably not possible to test the proposed hypotheses, viz. "i) the effects on the gut microbiome would vary with the spectrum of antibiotic used, the route of delivery, and the metabolism of the drug, ii) related effects may be identifiable in breath metabolites" with a single case report. It may be better to reframe the rationale for this study, which was presumably, as a pilot study?

INTRODUCTION
It probably not possible to test the proposed hypotheses, viz. "i) the effects on the gut microbiome would vary with the spectrum of antibiotic used, the route of delivery, and the ○ metabolism of the drug, ii) related effects may be identifiable in breath metabolites" with a single case report. It may be better to reframe the rationale for this study, which was presumably, as a pilot study? The article text has been updated to clarify this.
It is stated that "Based on the patient's relevant clinical microbiology results, applicability to OPAT usage and longer-term tolerability, a decision was made to start intravenous (IV) Ertapenem 1g once daily." It would be useful to include these results in the text.

○
The table from the extended dataset has been moved into the main text as per your suggestion, thank you.
Had no antibiotics been given in the period leading up to the baseline sample collection? This would seem unlikely, given the lengthy delay involved.
○ Not within 12 months of starting the study. The article text has been updated to clarify this, thanks.

Thanks -the article text has been updated to clarify this in the DNA extraction section.
Were sequences from all samples pooled for assembly, clustering and taxonomic assignment, or was this done separately for each sample? ○ Thanks -the article text has been updated to clarify this in the Sequence Assembly and Analysis sections.
Were ambient air samples collected as controls at the same time of breath collection? This would seem to be key. In figure 2A, what samples are used for the 'averaged' values?
○ Were breath samples collected in a hospital environment? Are any of the metabolites with high concentrations in this study likely to be found at higher concentration in hospital air vs ambient urban air (which is the comparator used)? ○ Thanks -the legend for Figure 2A has been updated with a reference. For your information though, we did not sample for ambient air at the time of the patient sampling, as sampling for all participants was done at different locations, not always in a hospital environment and often in different parts of the hospital. Also, these compounds change in air rapidly, so we can expect a 30% variability from one moment to the next anyway (Low et al., 2003). We, therefore, have a moving background, even if there had been an ambient air sample taken at the same place each time.
Variability within the local ambient environment, combined with the fact that the individual's blood and body were equilibrating with lung volatiles for ~ the previous 15 minutes or so means that an ambient air sample is not particularly informative in this case, and the sampling strategy was constrained by care.
However, it is worth noting that methyl chloride, methyl bromide, methyl iodide, and dimethyl sulfide are EXTREMELY high (often over an order of magnitude more than average male concentrations), and not in a way that is matched through any ambient background, internal or external that have been reported. Where the patient is low (isoprene), this is known to be human metabolism (~1,000x ambient background anywhere), so this decrease is notable.
Specifically, there are no published records of ambient concentrations of methyl chloride in the parts per billion range. Likewise, dimethyl sulfide concentrations are not seen in the parts per billion range in the ambient atmosphere anywhere. Methyl bromide concentrations do not ever hit hundreds of parts per trillion either, and the key for all of this is that there are no known mechanisms by which they would be released or generated in a hospital or home environment. Normally, methyl iodide appears to be essentially completely utilized within the human body (as a strong methylating agent it reacts away while in transit), so the fact that it is observed here is indicative of VERY different processes in this individual's metabolism and something similar can be said for methane thiol.

RESULTS
Did the patient take probiotic therapy, which may account for the increase in Lactobacillus reuteri following treatment?
○ Thanks -we have addressed this in the introduction now. For your information, participants were asked at baseline regarding their supplement consumption. This participant responded: Cod liver oil capsule every day, orange juice every morning, and regular yoghurts. He did not take probiotics before, or for the duration of, this study.
Given the likely mis-assemblies for several key taxa, do the authors think that a single sample-level assembly might assist in teasing this out? ○ Thank you. We are not convinced this would necessarily help. Metagenomes often contain quite a few misassemblies within the contigs, but given that we are looking at the taxonomy based on contig annotations, small errors such as SNPs caused by errors in the long reads, or chimeric reads, should not contribute to bad annotations because they will still match to the nearest taxonomy. Using all of the timepoints together in the assembly means that low abundance taxa are more likely to be represented Do the different MAGs for Bacteroides coprophilus differ in relative abundance by timepoint? If so, how does this relate to the prevalence of the different resistance genes over time? What about for Bacteroides dorei? ○ Cluster_contig0024 and Cluster_singleton2 which have lost and maintained the same genes have a similar trajectory, with a low abundance pre-treatment, peak abundance between timepoints 2 and 3, and then the low abundance following the beginning of the second course of treatment (timepoint 4). This low abundance is maintained even post-treatment. Cluster singleton4 and Cluster_contig0951 both lost all resistance genes. They do not have the same trajectory in relative abundance between timepoints pre-and up to timepoint 3. However, after timepoint 3, both MAGs become very low abundance and do which is maintained even in the post timepoint. Given that the transition between timepoint 3 and 4 represents the transition between different administration routes, it seems like the absence of these resistance genes is contributing to them becoming low abundance following the second treatment course.
I found this sentence very confusing: "Prevotella is also identified but in low abundance, and low but higher levels, compared to Prevotella, of Ruminococcus is also observed." ○ Thanks, this has been removed as it didn't add to the point we were making overall regarding the patient's enterotype.

DISCUSSION
The increase in B. coprophilus relative abundance during ertapenem therapy is interesting. Of course, this is relative, not absolute abundance, which is an important difference. It would be very useful to have some measure of total bacterial load over the course of therapy. Would 16S qPCR help here (even if an imperfect measure)?
○ It would, however, 16S abundance still isn't perfect, and we know that 16S sequencing often results in more skews. This would be less with qPCR versus sequencing, but we don't think it would add much more to the overall picture. Given the point above, it is probably not correct to make statements such as "B. coprophilus, a bacterium that has not been associated with human infection, was abundant throughout IV Ertapenem exposure" -the persistence (or increase) was in relative abundance and may in fact reflect a decline in absolute abundance if all taxa decrease in abundance, as one might expect during broad spectrum antibiotic therapy.
○ Thanks -we've edited this sentence to clarify it is relative abundance.