Whole-genome assembly of Akkermansia muciniphila sequenced directly from human stool.

BACKGROUND: Alterations in gut microbiota composition under antibiotic pressure have been widely studied, revealing a restricted diversity of gut flora, including colonization by organisms such as Enterococci, while their impact on bacterial load is variable. High-level colonization by Akkermansia muciniphila, ranging from 39% to 84% of the total bacterial population, has been recently reported in two patients being treated with broad-spectrum antibiotics, although attempts to cultivate this microorganism have been unsuccessful.
RESULTS: Here, we propose an original approach of genome sequencing for Akkermansia muciniphila directly from the stool sample collected from one of these patients. We performed and assembly using metagenomic data obtained from the stool sample. We used a mapping method consisting of aligning metagenomic sequencing reads against the reference genome of the Akkermansia muciniphila Muc(T) strain, and a De novo assembly to support this mapping method. We obtained draft genome of the Akkermansia muciniphila strain Urmite with only 56 gaps. The absence of particular metabolic requirement as possible explanation of our inability to culture this microorganism, suggests that the bacterium was dead before the inoculation of the stool sample. Additional antibiotic resistance genes were found following comparison with the reference genome, providing some clues pertaining to its survival and colonization in the gut of a patient treated with broad-spectrum antimicrobial agents. However, no gene coding for imipenem resistance was detected, although this antibiotic was a part of the patient's antibiotic regimen.
CONCLUSIONS: This work highlights the potential of metagenomics to facilitate the assembly of genomes directly from human stool.

Background

The elucidation of the composition of the human gut microbiota, which consists of approximately 100,000 billion bacteria, remains a major challenge for microbiologists. The influences of age, geographic location and dietary habits on physiological variations in the microbiota have been well established [1,2]. Moreover, alterations in the composition of gut flora have been linked with several diseases, including obesity [3], eczema [4], and necrotizing enterocolitis [5]. While culture-dependent methods have been mainly used to elucidate the gut bacterial repertoire, molecular techniques have gradually risen in popularity and are now commonly used for the characterization of the digestive flora because 80% of bacteria in the human gut remain uncultured [6]. Currently, metagenomics is considered the gold standard for human gut studies despite evidence of several biases in this technology [7].

Disturbances induced by antimicrobial agents on the composition of the gut microbiota have been widely explored. Most studies, whether they have been culture-dependent or based on molecular techniques, have agreed that antibiotics restrict the heterogeneity of the gut microbiota [8-11]. Thus, some bacterial populations that are frequently susceptible to antibiotics to which they are exposed [8-12] may be suppressed suggesting population replacement or colonization by resistant microorganisms, such as Enterococci, under antibiotic pressure [13].

Recently, high-level colonization by the Verrucomicrobia phylum of up to 39% and 84% has been reported in two patients receiving a broad-spectrum antibiotic regimen [14]. All reads were assigned to one species, Akkermansia muciniphila, which is an anaerobic Gram-negative bacterium commonly found in the digestive tract that is able to degrade mucin [15]. These data were confirmed by fluorescence in situ hybridization. Despite significant efforts to culture the bacteria from both samples were unsuccessful. An additional recent study has reported a potential connection between Akkermansia muciniphila and obesity [16].

Whole genomes have been previously sequenced directly from samples, such as Chlamydia trachomatis from the vagina [17], uncultured Termite Group 1 bacteria from protist cells [18], and Deltaproteobacteria from ocean samples [19]. Here, we performed a whole-genome assembly for the Akkermansia muciniphila strain Urmite [EMBL: CCDQ000000000], isolated from an atypical stool sample, in which over 80% of the sequences were assigned to the Akkermansia muciniphila type strain ATCC BAA-835 [Genbank:NC_010655.1]. To the best of our knowledge, this is the first report of the whole-genome sequencing of a stool sample in the absence of a cultured isolate.

Methods

Stool sample

The patient was a 62-year-old man admitted to the intensive care unit and treated with a 10-day course of imipenem (3 g/day) at the time of stool collection [14]. He did not show with any gastrointestinal manifestations. We did not obtain written informed consent for the stool collection due to the death of the patient. Approval from the local ethics committee of the Institut Fédératif de Recherche IFR48 (Marseille, France) was obtained under agreement 09–022. This agreement allows, according to French legislation, the use of stool samples because they are considered to be waste of human origin and do not involve additional sample collection from the patient.

Culture

Each gram of stool was diluted in 9 ml of Dulbecco’s Phosphate-Buffered Saline (DPBS) (Life technolgies, Saint Aubin, France) and inoculated in serial dilutions ranging from 1/10 to 1/1010 using different culture media and variable conditions. Previous culturomics studies [20] have established 70 culture conditions that produce a large diversity of bacteria. Considering the large proportion of Verrucomicrobia found in the sample by metagenomic analysis, we focused our attention on culturing the Gram-negative Akkermansia muciniphila using selective medium containing the antibiotic vancomycin (to inhibit predominant bacterial populations) or imipenem, which was the antibiotic administered to the patient. Previous reports in the literature [15] prompted us to use media containing mucin or to strengthen the anaerobic conditions. The culture conditions used are summarized in Table 1. A MALDI-TOF database was also amended with the Akkermansia muciniphila MucT strain spectra. Attempts to isolate Akkermansia muciniphila from the stool sample were unsuccessful (Additional file 1).

Table 1

prediction of antibiotic resistance genes in our consensus genome

Class	Best match	Length	GC content	Best hits with organism	Similarity	Coverage	Accession number
		(aa)	(%)		(%)	(%)
Beta-Lactamases	cfxA	332	51	Akkermansia muciniphila ATTCBAA-835	100	100	WP031930069
	tlaA	258	60	Akkermansia muciniphila ATTCBAA-835	97	84	YP001876808
	Beta-lactamase domain protein	298	60,7	Akkermansia muciniphila ATTCBAA-835	100	87	YP001877266
	CphA2	403	60,8	Akkermansia muciniphila ATTCBAA-835	99	95	YP297581
	Act	323	60,8	Akkermansia muciniphila ATTCBAA-835	99	85	YP001876732
	Metal-dependent hydrolases of the beta-lactamase superfamily	348	63,6	Akkermansia muciniphila ATTCBAA-835	99	75	YP001877492
	Metall o-beta-1actamase family protein	468	56,5	Akkermansia muciniphila ATTCBAA-835	99	100	YP001876862
	Zn-dependent hydro1ase	274	59,8	Akkermansia muciniphila ATTCBAA-835	96	98	YP001877763
Glycopeptides	vanX; D-ala D-ala dipeptidase	234	56.6	Akkermansia muciniphila ATTCBAA-835	99	95	YP001878228
MLS	mefA; macrolid efflux pump	401	49,2	Akkermansia muciniphila ATTCBAA-835	100	100	WP031931063
	ermB; erythromycin ribosome methy1ase	245	45	Akkermansia muciniphila ATTCBAA-835	100	100	WP012420167
	o1ec; macrolid ABC transporter protein	702	61,9	Akkermansia muciniphila ATTCBAA-835	65	100	WP012419164
Phenicol	catA3; chloramphenica1 acetyltransferase	211	56	Akkermansia muciniphila ATTCBAA-835	99	100	YP001876953
Su1phonamide	sulII; dihydropteroate synthase	279	65,7	Akkermansia muciniphila ATTCBAA-835	49	100	YP001877991
Tetracyclin	tetO	639	51,2	Akkermansia muciniphila ATTCBAA-835	100	100	WP012419363
Trimethoprim	dfrA3: dihydrofolate reductase	122	57,8	Akkermansia muciniphila ATTCBAA-835	99	75	YP001878622

Metagenomic sequencing

To extract DNA from the fecal samples, a modified version of the protocol described by Zoetendal et al. was used [21]. A shotgun and a 5-kb paired-end library were pyrosequenced on a Roche 454 Titanium sequencer. This project was loaded on a 1/4 region for each application on a PTP PicoTiterPlate (PicoTiterPlate PTP Kit; Roche), and DNA was extracted twice. The first set of DNA was resuspended in 50 μl TE buffer and used to construct a shotgun library. The DNA concentration was measured using a Quant-it Picogreen Kit (Invitrogen) and a Genios Tecan fluorometer and was calculated to be 37 ng/μl. A second set of DNA was later extracted in an attempt to construct a paired-end library. DNA was resuspended in 120 μl TE buffer, and the concentration was measured as above and calculated to be 11.5 ng/μl. The shotgun library was constructed with 500 ng of DNA as described by the manufacturer with a GS Rapid Library Prep Kit (Roche). The concentration of the shotgun library was measured using a TBS fluorometer and determined to be 1.05x109 molecules/μl. The paired-end library was constructed from a mix of the 2 sets of DNA, but only 2.7 μg of DNA from each set was used instead of the 5 μg recommended by the manufacturer. The DNA was mechanically fragmented to 5 kb with a Covaris device (KBioScience-LGC Genomics, Teddington, UK) and a miniTUBE-Red. The DNA fragments were visualized using an Agilent 2100 BioAnalyzer on a DNA LabChip 7500 with an optimal size of 5 kb. The library was constructed according to the 454 Titanium manufacturer’s paired-end protocol. Circularization and nebulization were performed, generating a pattern with an optimal length of 549 bp. After 17 cycles of PCR amplification followed by double-size selection, the single-stranded paired-end library was then quantified by an Agilent 2100 BioAnalyzer on an RNA 6000 Pico LabChip and was measured to be 549 pg/μl. The library concentration equivalence was calculated to be 1.87×109 molecules/μl. The library was stored at −20°C until use. The shotgun library was clonally amplified with 2 cpb in 4 emPCR reactions, and the 5-kb paired-end library was amplified with lower cpb values (0.25 and 0.5 cpb) in 2 emPCR reactions per condition with a GS Titanium SV emPCR Kit (Lib-L) v2. The emPCR yield was 10.24% for the shotgun library and between 6.4% and 7.8% for the clonal amplification of the 5-kb paired-end library. These percentages were within the quality range of 5% to 20% expected for the Roche procedure. A total of (70,000 beads from the shotgun library were loaded on a ¼ region of a GS Titanium PicoTiterPlate, whereas only 686,598 beads from the paired-end library were loaded on another ¼ region of the PicoTiterPlate with a GS Titanium Sequencing Kit XLR70. Runs were performed overnight and then analyzed with gsRunBrowser and Roche gsAssembler.

Metagenomic alignment

Our metagenomic alignment and the following method are shown in Figure 1.

Figure 1

Schematic of the two-assembly method using metagenomic data performed in this study.

In this study, we generated 1.4 gigabases of metagenomic sequence data from the stool sample. Reads were generated from short-read shotgun and paired-end runs on a 454 sequencer and a SOLiD sequencer. These reads were aligned to a database containing most known human genomes using Deconseq [22]. Only 0.2% of the reads were identified as human, and these were removed from the dataset. The 454 shotgun (122,354 reads) and paired-end sequencing (268,104 reads) data were mapped against the Akkermansia muciniphila (ATCC BAA-835) genome using CLC workbench software (CLC bio, Aarhus, Denmark). An identity of 90% was used as the threshold for the alignment of a read to the reference genome.

Sequence data obtained with the SOLiD sequencer (3,844,884 reads) were mapped against the previously created consensus using CLC Workbench software. The low setting was used for the largest proportion of the data. The parameters used were 70% identity and 40 bp length fraction.

Final mapping was conducted with the 454 shotgun and paired-end data against the previously created consensus genome using CLC Workbench software with the default parameters (80% identity and 50 pb for the length fraction).

Alternative methods for assembling Akkermansia muciniphila genome

The generation of an assembly by the mapping of reads to a reference genome requires a reference with a high level of quality and sufficiently similar sequences. Indeed, if the reference genome contains additional or highly divergent genes, the assembly may include many gaps or poorly assembled regions. De novo assembly remains the best solution, but in our study, it was impossible to achieve because we used a metagenomic sample. An alternative method involves first creating a de novo assembly to obtain a set of contigs and then selecting only those contigs that are highly similar to the reference. In addition, the contigs are ordered with respect to each other using the reference. Thereafter, the sequences can be joined using conventional finishing procedures, and the remaining reads can be used to fill the gaps.

These methods were performed to assemble the Akkermansia muciniphila genome using several tools. The assembly step was conducted using Newbler 2.8 [23] and Mira 3.2 [24]. The contigs obtained were combined by Cisa to reduce the set [25]. The contig mapping step was carried out with ReviSeq algorithm (under development, unpublished). Finally, finishing was performed using Gapfiller and CLC Genomics. The genome obtained using this method contained a single 2.72 Mb chromosome without gaps.

In silico antibiotic resistance gene prediction

The ARG-ANNOT database for acquired antibiotic resistance genes (ARGs) was employed for a BLAST search using the Bio-Edit interface [26]. The assembled sequences were searched against the ARG database under moderately stringent conditions (e-value of 10−5) for the in silico ARG prediction. These sequences were also submitted to Rapid Annotation using Subsystem Technology (RAST) [27] for annotation, additional putative ARG annotations are listed in Table 1. These putative ARGs were further verified through a web-enabled NCBI GenBank BLAST search.

Results

Mapping

The 454 shotgun and paired-end sequencing data were mapped against the Akkermansia muciniphila genome. The total number of mapped reads represented 44% (171,593) of the total reads, and the mean length of these reads was 199 bp. The paired-end reads represented 28% (107,582) of the total reads. The average coverage of the consensus was 13-fold. The consensus length generated from this mapping, including the gaps, was 2,664,714 bp, which was used for the remainder of the experiment. We performed a Mauve genome alignment [28], which showed a high level of similarity between the reference and consensus genomes (Figure 2). The differences were due to 519 gaps in the consensus.

Figure 2

Mauve alignment of the reference genome and the consensus created by mapping.

Sequence data from the SOLiD sequencer were mapped against the previously created consensus. A total of 791,434 reads were mapped, and the mean read length was 43 bp. The maximal coverage achieved by this mapping was 636-fold due to the large amount of data produced from SOLiD technology, and 95% of the previous consensus was covered. The consensus length generated from this mapping, including the gaps, was 2,664,713 bp, which was used for the remainder of the experiment. The second mapping allowed for the reduction in the number of gaps to 446. This consensus sequence produced thanks to this previous mapping was used for the next mapping.

For the final mapping, the mapped reads represented 45% (174,209) of the total, and the mean read length was 197.50 bp. The paired-end reads represented 28% (109,000) of the total reads. The average coverage of the consensus was 13-fold, and the consensus length generated from this mapping, including the gaps, was 2,664,704 bp. As a result of this mapping, only 392 gaps remained.

A large number of short sequences were inserted into gaps in the consensus sequence during these mapping steps. When we finished using all of the metagenomics data, we removed these sequences because they were not useful for the remaining analyses. In the end, we were left with 73 gaps. Analysis of metagenomic data from the trash reads allowed for the collection of 1189 reads corresponding to the genome of the Akkermansia muciniphila type strain (ATCC BAA-835), which has a G + C content of 55.8%. We carried out a BLASTX [29] search of the trash reads against the non-redundant protein sequence database (Nr), collecting only those reads with G + C contents of between 54% and 57%. Only 28 reads remained with a maximum size of 416 bp. However, we mapped these 28 reads against our genome assembly and closed two gaps of 31 bp and 296 bp in size.

Finishing

The PCR results allowed us to close 6 gaps of different lengths (ranging from 80 bp to 1168 bp). We then compared the consensus genome created by the mapping with that obtained by the alternative method. This comparison revealed sequence insertions that did not exist in the reference but were confirmed by PCR. We were also able to close the remaining gaps using this method. To ensure that the gaps would be closed using the alternative genome, we mapped the reads against the corresponding regions of the genome, leaving a total of 56 gaps.

Comparison

There were 2192 genes identified in the Akkermansia muciniphila type strain (ATCC BAA-835) genome. In the genome obtained by mapping, we identified 2237 genes. After a BLASTP [29] search and the verification of false positives, 49 Akkermansia muciniphila genes remained that were not present in our genome, and 52 genes remained that were not present in the reference. These sets of genes were analyzed and visualized using a circular map constructed with ACT Artemis software (Wellcome Trust Sanger Institute, Cambridge) [30] (Figure 3). We were thus able to estimate that our genome had lost 49 genes and gained 52 others. For these genes, we have detailed mutations type (stop, frameshift, multiples or replace) in Additional file 2. These loss/gains genes could be explain by the adaptation of this bacterium living in sympatric environment (metagenomic sample) [31] and having the opportunity to exchange genes.

Figure 3

Circular map of the genome and the consensus genome. The red bars correspond to reference genes that are absent in the consensus. The blue bars correspond to consensus genes that are absent in the reference.

Functional analysis

We used the Clusters of Orthologs Groups (COG) database [32] through the WebMGA server [33] to analyze the distribution of the 49 lost and 52 gained genes among the different functional categories. A few of the genes could be annotated with this database. The majority of the 49 lost genes were related to metabolism and the “C” categories (Energy production and conversion). A few were involved in cellular processes as well as signaling and information storage and processing. Some were part of the “J” (Translation, ribosomal structure and biogenesis) and “O” categories (Posttranslational modification, protein turnover, chaperones). The majority of the 52 genes that had been gained were involved in metabolic reactions. Only two of these genes that belonged to the “L” category (Replication, recombination and repair).

Metagenomic data other than that from Akkermansia muciniphila

The mapping against the reference genome revealed a number of reads that could not be aligned with Akkermansia muciniphila (trash reads). The mothur package [34] was used to remove redundant reads (i.e., reads that were 100% identical) from the 218,865 trash reads. This left 175,756 reads, which were compared to GenBank’s Nr database using BLASTX [29]. We kept only the best hits based on the number of thresholds we established. The best hits with an E-value cut-off of 10−4, an identity cut-off of 50% and a score cut-off of 50 were retained. With these results, we classified the metagenomic data by genus (Figure 4) and species (Figure 5).

Figure 4

Classification by genus of the remaining 175,756 reads after mapping. Only abundant reads are represented.

Figure 5

Classification by species of the remaining 175,756 reads after mapping. Only abundant reads are represented.

Antibiotics resistance gene research

An in silico ARG prediction was performed using ARG-ANNOT [26] and RAST [27]. Resistance studies of Akkermansia muciniphila have shown the presence of a range of putative ARGs from different antibiotic classes. The details of the ARG analysis are presented in Table 1. However, eight beta-lactamase genes were detected in this isolate that shared over 96% similarity and belonged to the class 1 and 2 beta-lactamases and metallo-beta-lactamases. Out of the three detected macrolide resistance genes, one was only 65% similar to a known macrolide. However, one gene each was found to be associated with resistance to vancomycin, chloramphenicol, sulfonamide, tetracyclin and trimethoprim.

Discussion

This study demonstrated an original approach for obtaining an assembled microbial genome. This approach permitted the assembly of a nearly complete genome from metagenomic data derived from human stool. We demonstrated the feasibility of assembling this genome by mapping reads to a reference genome. We used the genome of Akkermansia muciniphila, which is a representative of the phylum Verrucomicrobia, as the reference. We are confident in our findings, having routinely sequenced whole bacterial genomes [35-37] and mapped the Akkermansia muciniphila reference genome to fill remaining gaps. Whole-genome sequencing of microorganisms has previously been performed directly from human samples, such as Chlamydia trachomatis derived from vaginal swabs [17] (Table 2). However, to the best of our knowledge, this study is the first report of a bacterium that has been entirely sequenced from a human stool sample.

Table 2

Whole-cell sequenced genomes already published

Spcies/Strain sequenced	Specimen origin	Sequencing technology	N reads	N scaffolds/contigs	N gaps
Akkermansia muciniphila strain Amuc	Human stool	-SOLID	4,235,342	1 scaffold	56 gaps
		-454 (shotgun, PE)
Chlamydia trachomatis [17]	Human vaginal swab	Illumina HiSeq 2000	70,201,544	85% aligned/reference (2.6xcov)
				6% aligned (6xcov)
		-Sanger	NA	Complete genome (120 contigs)
Termit Group l StrRs-D17 [18]	Single host protist cell	-454
SAR324 clade of Deltaproteobacteria	Global ocean single cell marine sample	Illumina GA PE 100 pb	67,995,232	646 Contigs

The detected beta-lactamases resistant genes could confer resistance to many beta-lactam antibiotics, such as benzylpenicillin, amoxicillin, cephalothin, ceftriaxone, ceftazidime penicillins, cephalosporins, monobactam aztreonam and imipenem. The detected macrolide genes may confer resistance to erythromycin, azithromycin or clarithromycin due to the high expression of these genes observed during antibiotic treatment; however, the other detected ARGs may confer resistance to their respective antibiotics. Although our sample was collected from a patient being treated with a broad-spectrum antibiotic regimen, the in silico prediction warrants further experimental validation. Moreover, a previous attempt to detect carbapenemase by MALDI-TOF [38] directly from a stool sample has shown negative results [14], suggesting that caution should be used in the interpretation of these findings.

KEGG analysis revealed no apparent variations in metabolic pathways between the two strains [39] (data not shown). These data exclude particular needs for the strain present in the sample, which may explain our failure to culture an isolate despite the use of enriched or selective media with antibiotics. Akkermansia muciniphila is a fastidious and strictly anaerobic bacterium. It is possible that precautions for maintaining the anaerobiosis of the sample from the time of sample collection to aliquoting were unsuccessful, rendering the strain non-viable because of its extreme sensitivity to oxygen.

Conclusions

We have proposed an original approach for sequencing a complete genome directly from human stool samples, which was assembled by mapping reads to a reference genome. If data obtained here did not explain our failure to culture the strain from the sample, resistome analysis provided some clues concerning the colonization and survival of Akkermansia muciniphila in the gastrointestinal tract in a patient treated with a broad-spectrum antibiotic regimen.

Reviewers’ comments

We thank the reviewers for their valuable comments and helpful suggestions. We would like to respond and revise our manuscript in light of the reviews.

Reviewer’s report 1: Dr. Eric Bapteste, UPMC, Institut de Biologie Paris Seine, France

Reviewer 1

This work reports the sequencing of an almost complete bacterial draft genome (Akkermansia muciniphila) from a stool sample collected from a patient. This bacteria was likely resistant to antibiotic treatments, and one of the goals of this analysis was to identify genes potentially involved in the emergence of this medically challenging phenotype by comparing this genome to the genome of a closely related reference.

To this end, the authors used, according to their own words, an original approach to assemble metagenomic reads, producing a genome with 56 gaps left. I am not qualified to evaluate the sampling process, so I will focus my report on the other parts of this work. Although the methodology seems very sound, I would like to encourage the authors to elaborate a bit more on several aspects of their analyses.

Would it be possible to explain in what sense the proposed approach is original (what is new/ different from usual approaches for reads assembly)?

Authors’ response: We thank Dr. Bapteste for his comments on our manuscript. Our study presented an original approach because we obtained a genome directly from metagenomic data, without pre-processing. Our work allowed us to sequence the Akkermansia muciniphila strain Urmite genome directly from a stool sample, which has never been published to the best of our knowledge.

Also, the section, entitled Mapping (p.8), could be slightly improved in order to introduce more clearly what are the critical steps of this approach and what is their logical order of application. For example, does the order in which sequences were assembled matter? Here, the mapping started with reads from 454 shotgun data, then further information were obtained by mapping data obtained from a SOLiD sequencer. The former mapping produced a 13-fold average coverage and left 519 gaps, the latter one resulted in a 636-fold average cover and left 446 gaps. It is not very clear how the authors reconciled these two results to obtain their next mapping (the one with 392 gaps left and a 13-fold coverage). Likewise, the logic and order of steps of gaps reduction could be explained a bit more. That way, future studies may directly use the protocol proposed in this work.

Authors’ response: There is a logical order of applications for the mapping approach. This approach involves aligning reads against a reference genome. We started with the longer reads from the 454 shotgun data to obtain as long a consensus sequence as possible. Then, we used the shorter reads from the SOLiD sequencer, which allowed us to close gaps and to obtain a higher-quality sequence. Indeed, the SOLiD technology produced short reads in large quantities, from which greater coverage was obtained. For each step, we used the previously generated consensus sequence for the next mapping method. We added an additional explanation on page 8, l. 180–184.

Would it be possible to give a little bit more of an evolutionary perspective to the results that were found, i.e. the claim that Akkermansia muciniphila lost 49 genes and gained 52 others. How quick were these gains and losses? Maybe, providing the readers with a distance in terms of % identity between the 16S of the reference genome and the 16S of the newly assembled genome might give a better sense of the extent of divergence that occurred in these genomes outside this gold standard marker?

Likewise, did these gains and losses concern limited regions of the genomes, such as genomic islands, or were they widespread? Is there any clue of the mechanisms involved in the genes gains?

Authors’ response: In this study, we did not focus on the evolutionary process. The 16S sequences are identical (100% identity) in the reference and the draft genome. According to Figure3, these genes are widespread in the genomes and are not situated in limited regions. Thanks to your questions, we have clarified our findings. We have created Additional file2 (line 219) to clarify whether these genes have stop codons, frameshifts, or mutations or are replaced with other genes.

This work is based on metagenomic data; the bacteria existed in a sympatric environment, allowing the opportunity to exchange foreign sequences [31]. In this specific environment, the bacteria had the capacity to acquire genes to integrate them into chromosome and the ability to keep them included in the chromosome. The lost or gained genes were linked in the microorganism’s adaptation in a sympatric environment.

Minor points: The abstract indicates that 56 gaps are left in the draft genome, but Table 2 indicates 58 gaps, please reconcile these numbers.

Authors’ response: Thank you for this comment; we corrected this table.

p.7. The authors correctly explain that if the reference genome contains additional or highly divergent genes with respect to the environmental genome that they aimed at reconstructing, their protocol would result in gaps in this latter draft genome. Conversely, it might also be useful to discuss what would happen, in terms of assembly, if the draft genome contains additional or highly divergent genes with respect to the reference genome. In particular, is not there a risk to lose some of the original gene content of Akkermansia muciniphila? For example, could lost genes be fast evolving genes?

Authors’ response: If the draft genome contains additional highly divergent genes with respect to the reference genome, we could lose this information and would not be able to reconstruct the original genetic content entirely. We were able to use this mapping method because the genomes were very similar.

p.8. typo? They sequences?

Authors’ response: We corrected this typographical error.

Figure 1: the authors used CLC for the mapping, did they try other assemblers (and a different range of parameters) to estimate whether the number of gaps could be further reduced? In particular, in a second step of the analysis, could not it help to relax the criterion of % similarity (>90%) to possibly aggregate more divergent genes into the contigs/genome?

Authors’ response: We did not try other mapping software, but we used different ranges of parameters (lines 138, 142, 144–145). We chose to use a high-stringency condition for the first mapping to be sure that we aligned the reads that belonged to Akkermansia muciniphila because we aligned reads from the metagenomic data of a stool sample.

Table 1 typo:? best his? instead of? best hits?.

Authors’ response: We corrected this typographical error.

Quality of written English: Acceptable.

Reviewer’s report 2: Prof. William Martin, Institut of Botanic III, Heinrich-Heine University, Düsseldorf, Germany

Reviewer 2

This is a fine paper reporting a whole genome sequence assembly from human stool, a substantial technical advance. The focus of the paper is methodological, the applications of the method are broad. This is one of the world’s leading genomics groups, which shows in the quality of preparation for this paper. In my view it can be published as is, maybe following one more check regarding the permissions policies of BD and IFR48 with regard to the consent issue.

Authors’ response: We thank reviewer 2 for the comments on our article. As already written in the paper (lines 82–85), French legislation (agreement 09–022) allows the utilization of stool samples without the patient’s consent because these samples are considered to be wastes of human origin.

typo p. 4: did not present with any did not show any.

Authors’ response: We corrected this typographical error.

Quality of written English: Acceptable.

Reviewer’s report 3: Dr. Vivek Anantharaman, NCBI, NLM, NIH, USA

Reviewer 3

The paper presents a novel method of sequencing a bacterial genome from a stool sample. As a methods paper this paper presents the data well. But I have a few concerns in the analysis section. The authors say that 49 Akkermansia muciniphila genes were not present in their genome and 52 genes were not present in the reference set.

Do the genes that are missing fall in regions where the synteny of the genomes is disrupted? If so, have they been replaced by some other gene? Or, are they rapidly diverging genes and hence have escaped the cut-off?

Authors’ response: We thank reviewer 3 for the comments on our article. In this study, we did not focus on the evolutionary process. We have performed other verifications, and we can say that in the 49 genes absent in Akkermansia muciniphila strain Urmite, 22 had mutations involving the appearance of stop codons, 11 were caused by different mutations, 4 had mutations involving frameshifts, and the remainder were replaced by some other gene in the same location on the genome. We have performed the same verifications for the 52 genes present only in the Akkermansia muciniphila strain Urmite genome, and we can say that 4 had mutations involving the appearance of stop codons, 21 were caused by different mutations, 12 had mutations involving frameshifts, and the remainder were replaced. To clarify this point, we have added Additional file2.

A list of the novel genes involved in antibiotics resistance are shown in the Table 1. Some of these genes are shown to have 100% identity to Bacillus subtilis, Enterococcus faecalis and Staphylococcus warneri? all firmicutes, 99% or higher similarity to Bacteroides, and 95% or higher to Clostridium genes. Given that Akkermansia is a Verrucomicrobia, the high identities (especially a 100% identity) of the? novel genes? to those from organisms belonging to a totally different clade would suggest that they are contamination from those genomes and not necessarily novel genes. The authors have to either explain the very high similarity or consider these genes as dubious.

Authors’ response: Table1was based on the sequences present in the ARG-ANNOT database, which allows targeting putative genes. We found putative resistance genes based on sequence homology. We performed verification using BLAST for 9 genes that have high similarity in other bacteria. Among these genes, we found 5 genes that have almost 100% similarity and 100% coverage with Akkermansia but are not annotated as resistances genes, and most are hypothetical. Thanks to this database, we annotated resistance genes in Akkermansia muciniphila strain Urmite. To clarify this point, we have revised Table1. According to this revision, we have modified the text (lines 234–236).

Minor issues

The Additional files are not referred to in the text of the paper. This should be added.

Authors’ response: We have taken this comment into account, and we made this change.

In the? In silico antibiotic resistance gene prediction? paragraph? They sequences were also? should read? These sequences?

Authors’ response: We corrected this typographical error.

Quality of written English: Acceptable.