Recovery of Gut Microbiota of Healthy Adults Following Antibiotic Exposure

Articles
https://doi.org/10.1038/s41564-018-0257-9
Recovery of gut microbiota of healthy adults

following antibiotic exposure
Albert Palleja 1,2,23, Kristian H. Mikkelsen3,23, Sofia K. Forslund4,5,6,7,8,23, Alireza Kashani1,9,
Kristine H. Allin1,10, Trine Nielsen 1, Tue H. Hansen 1, Suisha Liang11,12, Qiang Feng11,12,
Chenchen Zhang11,12, Paul Theodor Pyl1, Luis Pedro Coelho 8, Huanming Yang11,12,13, Jian Wang11,12,13,
Athanasios Typas 8,14, Morten F. Nielsen3, Henrik Bjorn Nielsen2, Peer Bork 5,8,15,16, Jun Wang17,18,19,20,
Tina Vilsbøll3, Torben Hansen 1,21, Filip K. Knop 1,3,22*, Manimozhiyan Arumugam 1* and
Oluf Pedersen1*
To minimize the impact of antibiotics, gut microorganisms harbour and exchange antibiotics resistance genes, collectively
called their resistome. Using shotgun sequencing-based metagenomics, we analysed the partial eradication and subsequent
regrowth of the gut microbiota in 12 healthy men over a 6-month period following a 4-day intervention with a cocktail of 3 last-
resort antibiotics: meropenem, gentamicin and vancomycin. Initial changes included blooms of enterobacteria and other patho-
bionts, such as Enterococcus faecalis and Fusobacterium nucleatum, and the depletion of Bifidobacterium species and butyrate
producers. The gut microbiota of the subjects recovered to near-baseline composition within 1.5 months, although 9 common
species, which were present in all subjects before the treatment, remained undetectable in most of the subjects after 180 days.
Species that harbour β-lactam resistance genes were positively selected for during and after the intervention. Harbouring gly-
copeptide or aminoglycoside resistance genes increased the odds of de novo colonization, however, the former also decreased
the odds of survival. Compositional changes under antibiotic intervention in vivo matched results from in vitro susceptibility
tests. Despite a mild yet long-lasting imprint following antibiotics exposure, the gut microbiota of healthy young adults are
resilient to a short-term broad-spectrum antibiotics intervention and their antibiotics resistance gene carriage modulates their
recovery processes.
H
uman gut microbiota form complex, balanced ecosystems1. optimizing and disseminating antibiotic resistance genes (ARGs),
Perturbations of these biota may drive infections2, obesity3,4, collectively forming resistomes23. The human gut microbiota is con-
diabetes5–7 and inflammatory8 and neurological disorders9. sidered a reservoir for ARGs where members exchange these genes,
Rodent and human studies have shown that fecal transplants10, pre- thereby propagating resistance24. The development and spread of
biotics11, probiotics12 and antibiotics13, for example, can alter the gut microbial antibiotic resistance is a serious health concern given that
microbiota and improve host health. An estimated increase in life previously reliable antibiotics now fail. The intestinal microbiota
expectancy of 2–10 years is attributable to antibiotics14. However, from subjects in different countries harbour different ARG reper-
early-life exposure to antibiotics (especially macrolides) has also toires, reflecting the local usage patterns of antibiotics in healthcare
been associated with metabolic, inflammatory and neurologi- and food production25. However, only a few studies have character-
cal impairments both in animal models15–17 and in observational ized the effects of particular antibiotic regimens on the gut ecosys-
human studies18–22. tems of individuals with respect to the associated resistomes. In a
When exposed to antibiotics, microbial communities respond longitudinal 16S ribosomal RNA gene amplification and shotgun
not only by changing their composition, but also by evolving, metagenomics study of children receiving macrolides or penicillins
1
Novo Nordisk Foundation Center for Basic Metabolic Research, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen,
Denmark. 2Clinical-Microbiomics A/S, Copenhagen, Denmark. 3Center for Diabetes Research, Gentofte Hospital, University of Copenhagen, Hellerup,
Denmark. 4Experimental and Clinical Research Center, Charité-Universitätsmedizin Berlin and Max Delbruck Center for Molecular Medicine, Berlin,
Germany. 5Max Delbruck Center for Molecular Medicine in the Helmholtz Association, Berlin, Germany. 6Charité-Universitätsmedizin Berlin ,
Freie Universität Berlin Humboldt-Universität zu Berlin and Berlin Institute of Health, Berlin, Germany. 7Berlin Institute of Health, Berlin, Germany.
8
Structural and Computational Biology Unit, European Molecular Biology Laboratory, Heidelberg, Germany. 9Danish Diabetes Academy, Odense,
Denmark. 10Department of Clinical Epidemiology, Bispebjerg and Frederiksberg Hospital, Copenhagen, Denmark. 11BGI-Shenzhen, Shenzhen, China.
12
China National GeneBank, BGI-Shenzhen, Shenzhen, China. 13James D. Watson Institute of Genome Sciences, Hangzhou, China. 14Genome Biology Unit,
European Molecular Biology Laboratory, Heidelberg, Germany. 15Molecular Medicine Partnership Unit, University of Heidelberg and European Molecular
Biology Laboratory, Heidelberg, Germany. 16Department of Bioinformatics, Biocenter, University of Würzburg, Würzburg, Germany. 17iCarbonX, Shenzhen,
China. 18Beijing Advanced Innovation Center for Food Nutrition and Human Health, College of Food Science and Nutritional Engineering, China Agricultural
University, Beijing, China. 19Department of Biology, University of Copenhagen, Copenhagen, Denmark. 20State Key Laboratory of Quality Research in
Chinese Medicine/Macau Institute for Applied Research in Medicine and Health, Macau University of Science and Technology, Avenida Wai Long,
Taipa Macau, China. 21Faculty of Health Sciences, University of Southern Denmark, Odense, Denmark. 22Department of Clinical Medicine, Faculty of Health
and Medical Sciences, University of Copenhagen, Copenhagen, Denmark. 23These authors contributed equally: Albert Palleja, Kristian H. Mikkelsen,
Sofia K. Forslund. *e-mail: filipknop@dadlnet.dk; arumugam@sund.ku.dk; oluf@sund.ku.dk
Nature Microbiology | www.nature.com/naturemicrobiology

Articles NaTUre MICrobIology
(β-lactams), we found decreased diversity and enriched ARG car- In line with the alpha diversity measures, principal coordinate
riage from the exposure, more so in the former than the latter18. Two analysis based on Bray–Curtis dissimilarities demonstrated that gut
previous 16S rRNA gene amplification-based studies performed in microbial compositions immediately after treatment (D4 and D8)
one cohort demonstrated how treating healthy individuals with a had profound differences from D0, but gradually returned towards
single common antibiotic (for example, clindamycin, ciprofloxacin, their initial composition (Fig. 2a). To analyse the full composi-
amoxicillin or minocycline) led to an enrichment of ARGs (imputed tional changes after antibiotic perturbation beyond the first two
based on taxonomic composition) and long-lasting compositional principal coordinates, we compared the perturbed samples to their
effects on the microbiota, including depletion of butyrate-produc- corresponding samples at D0 using Bray–Curtis dissimilarities.
ing species26,27. Another 16S rRNA gene study reported that a seven- To verify whether these observed changes were more severe than
day treatment with vancomycin caused profound changes in the expected over time in the absence of antibiotics, we also compared
microbiota, whereas a seven-day amoxicillin treatment did not28. In the Bray–Curtis dissimilarities between same-donor samples in our
contrast, a recent single-antibiotic study reported reduced micro- cohort to Bray–Curtis dissimilarities between same-donor samples
bial diversity after a three-day amoxicillin treatment29. Finally, a with comparable sampling intervals in two control populations35,36.
study using shotgun metagenomic data suggested that the effects of Compared to D0 samples, D4 and D8 showed very high composi-
cefprozil (a β-lactam) on the gut microbiota depended on its initial tional differences (median values 0.98 and 0.95, respectively), indi-
state30, reconciling previously divergent results. Antibiotic resistance cating drastic changes in the microbiome composition immediately
genes undetected at baseline were found after a seven-day treatment post-treatment (Fig. 2b). These differences were significantly larger
with the antibiotic30. However, the effects of a combinatorial multi- than same-donor differences in the control populations (two-sided
ple-antibiotic treatment on the microbiota and the role of ARGs in Wilcoxon rank-sum test, q < 0.001). By D42 and D180, the com-
gut microbial persistence have not been studied so far. munities still exhibited compositional differences from D0, but
Recently, we conducted an intervention in 12 healthy men who at a significantly lower magnitude (median values 0.70 and 0.64,
underwent prospective eradication of the gut microbiota through a respectively). These differences were still significantly larger than
combination of three broad-spectrum bactericidal antibiotics (van- same-donor differences at even larger time separations in control
comycin, gentamicin and meropenem) during a single four-day populations (Fig. 2b; two-sided Wilcoxon rank-sum test, q < 0.05).
course and investigated the effects on host metabolism31. This mul- However, the compositional differences between D42 and D180
tidrug cocktail is a modified version of prophylactic antibiotic pro- samples were not significantly different from the variation seen over
tocols from intensive care units32 and includes antibiotics given to comparable durations in control populations (Fig. 2b, rightmost
patients with infections by multidrug-resistant bacteria. In the pres- panel; two-sided Wilcoxon rank-sum test; US Human Microbiome
ent study, using shotgun metagenomics on stool samples collected Project (HMP) dataset, time interval >100 d, P = 0.10; Voigt dataset,
before and at four different time points over a period of six months time interval >300 d, P = 0.57) suggesting that the gut microbiota
after treatment, we tracked the process of eradication, partial sur- had reached a stable composition by D42.
vival, gradual regrowth and re-establishment of these gut microbial
communities. Using a metagenomic species approach33, we link Specific microbial taxonomic changes during recovery and recol-
individual ARGs of the resistome to their carrier taxa, assessing how onization. We investigated which microorganisms became extinct
the ARG repertoire of intestinal microorganisms helps or hinders or survived after the treatment, which microorganisms colonized
them during survival and recolonization. the gut de novo and which microorganisms were lost permanently
by following the changes in the relative abundance of individual
Results microbial species over time. By D8, 4 days post-intervention, the
Overall composition of gut microbiota of young healthy adults relative abundances of 50 species had significantly changed (two-
recovers from a broad-spectrum antibiotic exposure. We charac- sided Wilcoxon signed-rank test adjusted for compositionality,
terized the gut microbial composition of the 12 study subjects using q < 0.05). There was an enrichment of low-abundance commensals
species-level metagenomic operational taxonomic units (mOTUs)34. like Escherichia coli, Veillonella spp., Klebsiella spp., E. faecalis and
Using samples collected at baseline (D0), immediately after antibi- F. nucleatum (Supplementary Fig. 1 and Supplementary Table 1),
otic treatment (4 d after baseline; D4) and at three further time points indicating a major ecological change. We also observed a deple-
(8, 42 and 180 d after baseline; D8, D42 and D180, respectively), we tion of other commensals, particularly butyrate producers, such
traced the eradication and recovery of gut microbial species by ecologi- as Faecalibacterium prausnitzii, Roseburia hominis, Anaerostipes
cal alpha and beta diversity measures. The gut microbial alpha diversity hadrus, Coprococcus spp. and Eubacterium spp. This replicates
of our subjects recovered within six months of the intervention (Fig. 1). reported decreases of short-chain fatty acid producers under anti-
By D4, immediately after intervention, both the species richness biotic treatment27,28.
and Shannon diversity were dramatically reduced compared to D0 By D42, none of the species whose relative abundance differed
(two-sided Wilcoxon signed-rank test; false discovery rate (FDR)- significantly between D0 and D8 exhibited significant differences,
adjusted P values (q) = 0.0098 and q = 0.0065, respectively). However, suggesting that most of the dominant microorganisms of the human
despite the broad-spectrum nature of the intensive antibiotic treat- gut had regrown. Some of the previously blooming species, includ-
ment, full eradication was not achieved and numerous species were ing Klebsiella spp., Megasphaera micronuciformis, E. faecalis and
detectable by D4 (Fig. 1a). By D8, the gut microbiota still exhibited F. nucleatum, were no longer detectable. Others, such as E. coli and
reduced richness, but Shannon diversity had significantly increased Veillonella spp., remained but returned to D0 levels, so that by the
(Fig. 1b; two-sided Wilcoxon signed-rank test; q = 0.0065), suggest- end of the observational period they were not particularly elevated.
ing that the surviving microorganisms started recovery by regrowing Clostridium spp. were generally undetectable pre-intervention
more evenly. Thereafter, both richness and Shannon diversity were but significantly increased in their relative abundance by D42
gradually regained during the six-month follow-up (as measured at (two-sided Wilcoxon signed-rank test after adjusting for composi-
D42 and D180). Interestingly, although we observed no significant tionality, q < 0.05) (Supplementary Fig. 1 and Supplementary Table 1).
differences in Shannon diversity between D0 and D180, the species These species are known to form endospores under unfavourable
richness remained significantly lower by D180 (two-sided Wilcoxon conditions, reviving when environmental conditions are once again
signed-rank test; q = 0.011), suggesting that some microorganisms that favourable37. This may have provided them with an advantage in
were originally present may have been permanently lost or severely our clinical setup, with spores reviving on sensing an emptier gut
depleted due to the treatment. and effectively colonizing the gut after intervention. Specifically,

NaTUre MICrobIology Articles
a q = 0.011
b
q = 0.0065 q = 0.0065 q = 0.0065
200 q = 0.0098 q = 0.0024 q = 0.013
175 4
150
Shannon diversity
3
125
Richness
100
2
75
50
1
25
0
0
D0 D4 D8 D42 D180 D0 D4 D8 D42 D180
Time point Time point
Fig. 1 | Gut microbial diversity recovers after a broad-spectrum antibiotic intervention. a,b, Microbial richness (a) and Shannon diversity (b) based on
mOTU relative abundances. The boxplots represent the diversity measures for the 12 volunteers (centre line, median; box limits, first and third quartiles;
whiskers, 1.5 × interquartile range), which are also represented as points connected across time by grey lines. Samples are coloured according to time point
(n = 57). The FDR-adjusted P values are shown between consecutive time points (two-sided Wilcoxon signed-rank test).
Clostridium bolteae abundance was significantly increased by the species level (Supplementary Fig. 1) as well as differences
D42 (two-sided Wilcoxon signed-rank test after compositionality between conspecific strains (Supplementary Fig. 3). For this, we
test, q = 0.026) and remained elevated in all 12 subjects on D180 updated the ARG annotation of IGC genes40 to ARG families in the
(two-sided Wilcoxon signed-rank tests after compositionality test, Comprehensive Antibiotic Resistance Database (CARD)41 and the
q = 0.052; Supplementary Figs. 1,2). Finally, some species detected ResFams42 database. By mapping metagenomic reads to IGC, we esti-
at D0 were not detected again within the study time frame. These mated the abundance of known ARGs against β-lactams (parent class
included: (1) members of genus Bifidobacterium that are considered of meropenem; note that several β -lactamases cannot cleave
pathogen-protective and immunostimulatory; (2) butyrate pro- carbapenems43), aminoglycosides (parent class of gentamicin),
ducers such as Coprococcus eutactus and Eubacterium ventriosum glycopeptides (parent class of vancomycin) and all other antibiotic
and (3) methane-producing Methanobrevibacter smithii associ- classes, which were grouped for comparison. We conducted the
ated with the efficient digestion of polysaccharides. These results analysis at the antibiotic class level as existing ARG annotation is
with regards to compositional changes broadly match expecta- frequently incomplete, whereas cross-resistance between structur-
tions from the relative susceptibility to antibiotics of human gut ally similar antibiotics is common and likewise incompletely anno-
microorganisms from a recent in vitro microbial growth inhibition tated. We further quantified multidrug efflux pumps (MEPs), which
screen38 (see Supplementary Analysis, Supplementary Table 4 and could potentially confer unspecific multidrug resistance. Antibiotic
Supplementary Fig. 12). resistance genes against β-lactams and aminoglycosides were not
We then investigated taxonomic changes at an increased reso- significantly altered from D0 abundances (Fig. 3). Our results were
lution using metagenomic species33 (MGSs) derived from the inte- robust to the choice of ARG database (see Methods). The larger
grated gene catalog (IGC) of 9.9 million gut microbial genes39. MGSs contribution to the ARG abundance changes that we observed
include cultured as well as uncultured gut microbial species and are driven by genes more likely to be chromosomal than located
provide sub-species-level resolution in some taxonomic groups, as on mobile genetic elements (see Methods). Antibiotic resistance
each MGS is probably derived from an independent chromosome33. genes against glycopeptides were significantly depleted by D4 and
MGS-based analysis confirmed many of the microbial changes remained so until D8 (Fig. 3; two-sided Wilcoxon signed-rank test,
observed using mOTUs (Supplementary Fig. 3 and Supplementary q < 0.05). We observed no significant differences when comparing
Table 1) and highlighted additional changes. It corroborated the either D42 or D180 to D0. Abundances of MEPs were significantly
post-intervention significant increase of species usually only found elevated by D8 (two-sided Wilcoxon signed-rank test, q < 0.01) and
at low abundance. Furthermore, it confirmed the depletion of butyr- these were maintained until D42 (two-sided Wilcoxon signed-rank
ate producers (Supplementary Fig. 3) post-treatment and provided test, q < 0.05). By D180, no category showed significant differences
increased sub-species resolution, highlighting eight different MGSs when compared to D0 (Fig. 3). Thus, the overall ARG abundance
belonging to F. prausnitzii, one to R. hominis, three to Eubacterium spp. analysis did not show clear temporal patterns that could explain the
and two to Coprococcus spp. as significantly depleted (two-sided differences between species and conspecific strains.
Wilcoxon signed-rank test after adjusting for compositionality, Unlike previous gut resistome analyses, our use of the MGS
q < 0.05). Remarkably, only two F. prausnitzii strains were able to approach33 enabled us to link ARGs to the MGSs that they belong
recover by D180, another six could not recover through the study to. First, we observed that both the MGSs that survived the treat-
period (Supplementary Fig. 3), reflecting the different recovery ment and the first colonizers often harbour ARGs against multiple
capacities between conspecific strains. antibiotics (Supplementary Figs. 4,5). Second, by using the abun-
dances of MGSs and the abundance of individual ARGs within
Resistance gene carriage impacts survival and colonization poten- them, we noted each MGS as either detected or undetected in
tial of gut microorganisms. We hypothesized that differences in each study sample and, when detected, we marked those as either
antibiotic resistance potential among microorganisms could explain resistant or non-resistant to each of the three antibiotics. We then
the differences in their recovery patterns, which involve trends at assessed how possession of ARGs affected the chance of survival

a b q = 1.7 q=4× q = 1.1
q = 3.8 × 10–4
× 10–4 10–5 × 10–4
q = 0.037
1.0 1.0 1.0 1.0
0.4 0.9 0.9
S6 0.9 0.9
Bray−Curtis dissimilarity
S6 0.8
0.8 0.8 0.8
0.7 0.7
PCo2 (13%)
S6 0.7 0.7
0 0.6 0.6 0.6 0.6
S6
0.5 0.5 0.5 0.5
0.4 0.4 0.4 0.4

S6 Time point
D0 0.3 0.3 0.3 0.3
−0.4
D4
D8 0.2 0.2 0.2 0.2
Phase
Intervention D42 Cohort
Recovery 0.1 0.1 0.1 0.1 ABX Voigt HMP
D180
−0.2 0 0.2 0.4 0.6
0
us 42
60
d
us 8
us 4
2
d
7
rs 18
18
rs D
rs D
D
M 300
00
rs D
D
D
ve us
D
ve s
PCo1 (17%)
>1
ve s
0 su
>
us
0 su
us
0 s
D ver
D ver
t,
P,
D ers
D ver
ig
ve
Vo
0
0
H
0
D
D
0
42
D
D
Fig. 2 | Gut microbial composition dramatically changes after broad-spectrum antibiotic intervention but recovers progressively. a, Principal coordinate
analysis based on Bray–Curtis dissimilarity between samples, calculated using mOTU relative abundances. Samples are coloured according to time
point (n = 57) and ellipses represent the 95% confidence interval for each time point assuming a multivariate normal distribution. The grey lines connect
samples from the same individual (S6) to denote a representative trajectory followed by the individuals after antibiotic intervention. The variance
explained by each axis is stated as a percentage in parentheses. b, The boxplots show the Bray–Curtis dissimilarities (as a measure of beta diversity)
between same-donor samples from two different time points (centre line, median; box limits, first and third quartiles; whiskers, 1.5 × interquartile range).
The four boxplots in the rightmost panel show that the extent of compositional changes between paired samples during the study time (D0 to D180) were
significantly different compared to the degree of variation over time observed for paired samples in the Voigt and HMP controls (separated by time spans
>300 d and >100 d, respectively) but the compositional changes between D42 and D180 did not differ from the controls. Grey was used to represent
the study distances between paired study samples (n = 57). Orange and brown was used to represent the distances between paired samples for the
Voigt (n = 35) and HMP (n = 82) datasets, respectively. Two-sided Wilcoxon rank-sum tests were used to test for significant differences between sample
distances and FDR-adjusted P values are shown. ABX, antibiotics-treated cohort.
of an MGS (detection at D0 and at a later time point) using Fisher’s selected for, whereas species carrying aminoglycoside ARGs here
exact test (see Methods for details). Metagenomic species with could be negatively selected for. The latter could reflect that amino-
β-lactam ARGs had significantly higher odds of survival (odds ratio glycosides are inactive under anaerobic conditions (as those found
(OR) = 1.64 (1.24–2.17)) by D8, but this effect decreased at later time in the gut) and are usually administered for local or systemic infec-
points (Fig. 4). In contrast, MGSs with glycopeptide ARGs had sig- tions. The trend observed for species carrying glycopeptide ARGs
nificantly decreased chances of survival from D0 to D8 (OR = 0.32 could indicate that, under the present multidrug treatment, carry-
(0.23–0.44)) and D42 (OR = 0.71 (0.60–0.84)), again with the effect ing vancomycin ARGs may not be beneficial for the survival of a
weakening over time. This change may be due to the increase of microorganism. This may indicate that these ARGs are not func-
Gram-negative species in the initial time points after treatment, tional and/or are expensive to make (most known vancomycin
which are naturally resistant to vancomycin without the need for resistance systems are operons of up to nine genes encoding pro-
specific ARGs. Strikingly, MGSs with aminoglycoside ARGs exhib- teins that build an alternative cell-wall precursor less susceptible to
ited a gradual decrease of survival odds from D0 to D180 (OR = 0.67 glycopeptide binding44), or it could reflect that most of the vanco-
(0.48–0.93)). Although we observed a significant enrichment of mycin resistance in the gut microbial population is not encoded by
MEPs by D8 (Fig. 3), MGSs carrying MEPs did not show higher ARGs but is due to the overall cell envelope composition. The same
chances of survival. Interestingly, MGSs that were undetected at trends were observed for carrying glycopeptide ARGs in the control
D0 had significantly higher odds of de novo colonization of the population (HMP; >100 d) but with lower odds (Fig. 4; OR = 0.57
gut ecosystem at a later time point if they carried ARGs against any (0.49–0.65) and OR = 1.36 (1.18–1.56) for survival and de novo
of the three classes used: aminoglycoside ARGs (OR = 4.32 (2.29– colonization, respectively). Thus, glycopeptide resistance capacity
8.57); OR = 1.91 (1.24–2.94); OR = 2.01 (1.39–2.92); for D8, D42 may be found more frequently in taxa with a generally higher gut
and D180, respectively), β-lactam ARGs (OR = 1.64 (1.12–2.40); turnover. Carrying MEPs significantly increased the chance of MGS
OR = 1.40 (1.12–1.77); OR = 1.12 (0.91–1.37), respectively), or gly- survival in the control population (Fig. 4; OR = 1.78 (1.36–2.32)),
copeptide ARGs (OR = 2.45 (1.61–3.74); OR = 2.33 (1.86–2.94); but we did not observe this pattern in our intensive treatment, sug-
OR = 1.33 (1.09–1.62), respectively), although this effect also weak- gesting that MEPs were altogether insufficient to provide resistance
ened over time (Fig. 4). We observed the opposite pattern in the to our antibiotic cocktail.
long term for MGSs carrying MEPs, suggesting that the possession We next tested whether ARG families that changed significantly
of more MEPs does not increase the chances of colonizing the gut in abundance post-intervention were also enriched within the
environment post-treatment (OR = 0.51 (0.35–0.76) and OR = 0.54 genomes of species that were enriched post-intervention. Indeed,
(0.38–0.76) for D42 and D180, respectively). The survival and taxonomic compositional changes predicted the gene-level changes
colonization likelihood trends for MGSs harbouring β-lactam and for 65 of 82 families (Supplementary Analysis and Supplementary
aminoglycoside ARGs were not observed in the control population Table 6). The remaining ARG families may represent cases of genes
(HMP same-donor sample pairs with an interval >100 d; Fig. 4), that are more variably found across conspecific gut strains, or genes
suggesting that MGSs carrying β-lactam ARGs could be positively that have been carried on mobile elements in the recent evolutionary

Aminoglycosides β-Lactams MEPs Glycopeptides Other
2.0 × 10–4 4 × 10–4 6 × 10–3 4 × 10–4 5 × 10–3
9 × 10–5
1 × 10–3
1.5 × 10–4 3 × 10–4 3 × 10–4 4 × 10–3
4 × 10–3
4 × 10–2
ARG abundance
1.0 × 10–4 2 × 10–4 2 × 10–4 3 × 10–3
2 × 10–3
0.5 × 10–4 1 × 10–4 1 × 10–4 2 × 10–3
0 0 0 0 7 × 10–2 1 × 10–3
9 × 10–5
0
4
8
D 2
0
0
4
8
D 2
0
0
4
8
D 2
0
0
4
8
D 2
0
0
4
8
D 2
0
D
D
D
4
18
D
D
D
4
18
D
D
D
4
18
D
D
D
4
18
D
D
D
4
18
D
D
Time point Time point Time point Time point Time point
Fig. 3 | Microbial antibiotic resistance potential changes after broad-spectrum antibiotic intervention. Boxplots show the changes in ARG relative
abundance (centre line, median; box limits, first and third quartiles; whiskers, 1.5 × interquartile range) throughout the intervention (n = 57). These
enrichments/depletions are shown for the three drug classes used in the intervention (aminoglycosides, β-lactams and glycopeptides), MEPs and a general
category where we pooled ARGs that protect against all other antibiotic classes (other). FDR-adjusted P values are shown for the statistically significant
enrichments (two-sided Wilcoxon rank-sum test). The grey horizontal lines represent the median and the 25th and 75th percentile values at D0.
Post antibiotic treatment Control

Survival Colonization Survival Colonization
Aminoglycoside
resistance
. *** ** ***
log2[OR]
.
3
β-Lactam
resistance *** * ** 2
1
0
Glycopeptide
resistance *** *** *** *** ** *** *** –1
–2
–3
MEPs ** ** *** ***
D8 D42 D180 D8 D42 D180 >100 d >100 d
Fig. 4 | Resistance gene carriage impacts the odds of microbial survival and colonization. The left heatmap shows the odds ratio (log2) of survival and
colonization after the intervention of MGSs carrying ARGs to the antibiotics used in the treatment as well as MEPs (for D8, D42 and D180; n = 48).
The right heatmap shows the same for a control population (HMP; n = 82). Asterisks represent statistically significant odds ratios. ***q < 0.001;
**0.001 ≤ q < 0.01; *0.01 ≤ q < 0.05; ▪0.05 ≤ q < 0.1; Fisher’s exact test. Exact FDR P values and other details about the statistical test can be found in
Supplementary Fig. 4.
past. We note among others that the carpapenem-resistance asso- Increase of antibiotic resistance functional potential and enrich-
ciated resistance-nodulation-cell division pump AdeJ, character- ment of virulence factors following antibiotic exposure. We next
istic of Acinetobacter baumannii45,46, is detected and elevated after investigated whether there were microbial functions that increased
intervention in these samples in the absence of its typical host, sug- in abundance alongside the otherwise usually low-abundance spe-
gesting that its genomic carriage in the gut may be wider than previ- cies by D8 and later post-treatment time points. We used previously
ously thought. published annotation of IGC genes40 to the Kyoto Encyclopedia

of Genes and Genomes (KEGG) database47, estimating the abun-

dance of KEGG modules and pathways for each sample. The relative Virulence-associated gene regulation ***
abundance of 192 KEGG modules and 64 KEGG pathways changed
significantly from D0 to D8 (two-sided Wilcoxon signed-rank test, Virulence-associated chaperone activity ***
q < 0.05; Supplementary Table 1a). Among the significant modules,
Virulence-associated cellular metabolism ***
15 represented different efflux pumps, confirming the enrichment
of MEP genes identified in our ARG analysis. Five of these efflux
Virulence-associated capsule formation **
complexes included TolC, which is a central component of multiple
bacterial efflux pumps that are responsible for exporting numer- **
Uptake, iron *
ous and diverse antibiotics, including β-lactams, in both growing
and dormant bacteria48. It is crucial to note that gene enrichment Uptake, haem *
visible under our study design reflects the fitness of bacterial taxa
that carry such genes, which may or may not depend in turn on the Unspecified virulence mechanism **
thus-enriched genes themselves. The involvement of TolC in efflux
pumps extruding bile acids could conceivably impact such fit- Toxin, exotoxin **
ness, although this cannot be concluded from the present data and
remains speculative. Furthermore, we observed an increase in the Secretion system, type III **
abundance of transporters including seven bacterial phosphotrans-
ferase systems, which catalyse phosphorylation and transporta- Secretion system, further unspecified *
tion of a variety of sugars into the cell, allowing microorganisms
to maximize the assimilation of available sugars to grow faster Secretion system, effector *
in a competitive environment. Remarkably, 27 modules encod-
ing two-component systems, which are basic signalling cascades Quorum-sensing system ***
for responding to environmental perturbations, had significant
changes in relative abundance (two-sided Wilcoxon signed-rank Pore formation *
test, q < 0.05; Supplementary Table 1); this included systems known
to respond to glycopeptide (vancomycin/bacitracin) antibiotics in Phagosome escape **
Gram-positive organisms (LiaSR, BceSR, VraSR). Consistent with
the increase in enterobacteria and other Gram-negative bacteria at Phagosome affecting ***
D8, lipopolysaccharide-related features increased, as well as systems
for alternative forms of respiration (for example nitrate, tetrathion- Phagolysosome formation inhibition ***
ate and trimethylamine oxide anaerobic respiration). It has been
recently shown that the use of alternative electron acceptors, often LPS (lipopolysaccharide) ***
produced as part of inflammatory response, help enterobacteria to
bloom in the human gut49–52. Invasion *** *
By D42, only 41 KEGG modules were significantly different in
their relative abundance, including 8 MEPs (6 of which include the
Fimbriae ***
TolC gene) and 4 two-component systems, but none of the KEGG
Contributes to cord formation **
modules related to resistance against specific antibiotics (two-sided
Wilcoxon signed-rank test, q < 0.05; Supplementary Table 1b). By Common pili **
D180, no functional feature from KEGG was significantly differ-
ent in relative abundance, suggesting that the microbiota recovered Biofilm formation *
to near D0 composition, also with regards to functional potential
(Supplementary Table 1c). Anti-apoptosis **
Several virulence factors, such as capsule synthesis and type II,
III and VI secretion systems (T2SS, T3SS and T6SS; Supplementary Adhesion **
Table 1), were also increased by D8, which led us to investigate
them further. For this purpose, we used IGC annotations to the Scaled abundance change from baseline
D4 D8 D42 D180
Pathosystems Resource Integration Center (PATRIC) database –25 –5 –1 0 25 100 225 400
of virulence factors53 and estimated the abundance of PATRIC
functional categories in each sample. Both immediately (D4) and Fig. 5 | Microbial virulence factors are enriched immediately after
four days (D8) after the treatment there was a strong enrichment antibiotic treatment. The heatmap shows differential abundance of genes
of genes involved in quorum sensing. This may be a consequence mapped to each virulence functional category (according to the PATRIC
of selection for bacteria better able to coordinate gene expression virulence specialty gene set). The mean abundance across all samples
under changes in cell density due to significantly reduced micro- (n = 57) at each time point was scaled by subtracting D0 abundance
bial density post-treatment. However, our data cannot be used to and dividing by D0 s.d. The result was then square root-transformed for
conclude this. Activities controlled by this system are involved in visibility. Asterisks show significant differences from D0 levels. ***q < 0.001;
virulence, pathogenic potential and biofilm formation of micro- **0.001 ≤ q < 0.01; *0.01 ≤ q < 0.05; one-sided Wilcoxon rank-sum test.
organisms54. By D8, a set of virulence factors were significantly Exact FDR P values and other details about the statistical test can be found
enriched (Fig. 5), including exotoxins, pore-forming toxins and in Supplementary Fig. 5.
secreted effectors required for invasion or for inhibiting phagosome
formation or maturation and allowing intracellular proliferation.
In contrast, changes in lipopolysaccharide, capsule formation capac-
ity, fimbriae and pili presumably reflect the increase in enterobacteria, may be bacteria with pathogenic potential. Despite this initial
which commonly carry such elements. These observations indicate enrichment of virulence factors after the treatment, these factors
that the earliest gut colonizers, after suppression of D0 commensals, were cleared by D42 and did not seem to cause long-term effects in

healthy individuals of this study, which is consistent with what we post-treatment. Species harbouring β-lactam, but not aminoglyco-
reported in an earlier study31. side ARGs, were possibly positively selected for by our intervention,
which is consistent with the stronger action of β-lactams against
Discussion enteric bacteria and in anaerobic conditions. Strikingly, while
Gut microbiota of young healthy subjects showed resilience and microorganisms carrying glycopeptide ARGs exhibited increased
a capacity for recovery from a combinatorial antibiotic treatment odds of de novo colonization, they had decreased odds of survival
designed to maintain prophylaxis on gut microbiota, although some through the intervention, possibly reflecting that such resistance
species or strains detectable at D0 remained undetected through- mechanisms may be inactive or too energy- and time-consuming
out the rest of the study period. When we compared our metage- to be activated during the short course of a multidrug antibiotic
nomic observations to a recent drug susceptibility screen of diverse challenge. We cannot exclude interactions between the three
gut species, the results were concordant (Supplementary Analysis, administered drugs that may partially mask their effects or
Supplementary Table 4 and Supplementary Fig. 12), confirming render their resistance mechanisms obsolete. Moreover, as we rely on
that we recorded the direct effects of drugs in the gut microbiota. databases of curated ARG families, we cannot rule out the possibility
It is important to note that an accurate estimation of the extent of that the enrichment of ARG families may be biased by the extent to
loss of bacteria would require analysis of absolute count data, which which antibiotic resistance has been characterized in different organ-
could be obtained experimentally through mass-normalized qPCR isms. Many of the gene-level findings that we report are due to the
analysis of samples. Although this was impossible for the present taxonomic shifts (Supplementary Analysis and Supplementary
study due to sample availability, it can be undertaken in future fol- Tables 5,6); that is, they could be considered as gene/functional-
low-up experiments. level projections of taxonomic shifts. Although we hypothesize that
Contrary to expectations, we observed no immediate enrich- these shifts in functional potential may provide a fitness advantage
ment of ARGs targeted to the antibiotics used in the healthy indi- to the microorganisms that harbour these functions, metatranscrip-
viduals in this study (Fig. 3). However, four days post-treatment, a tomic studies are needed to substantiate or falsify this hypothesis.
specialized community of previously low abundant or undetectable In summary, our results demonstrate that the gut microbiota of
species, alongside ARG and virulence factor enrichment (Figs. 3,5 young, healthy adults are resilient to a four-day broad-spectrum
and Supplementary Fig. 4) arose, whereas short-chain fatty acid pro- antibiotic treatment, that most of these communities can recover
ducers, particularly butyrate producers, were depleted as reported to their D0 composition and that the resilience/recovery patterns of
in earlier studies27,28. The increase in the abundance of MEP genes individual species are modulated by ARG carriage. Further studies
and microbial structural elements (for example lipopolysaccharide, are needed to verify whether the human gut microbiota is likewise
pili, fimbriae, among others) presumably reflects the taxonomical resilient to multiple antibiotic exposures over prolonged peri-
changes observed post-treatment, as many of these systems are part ods and whether these results hold in children with an immature
of the core genomes of Enterobacteriaceae spp., which increased gut microbiota or in elderly people with an age-related decline in
after the treatment (for example, E. coli and Klebsiella spp.). Other immune competence and perturbed intestinal microbiota.
virulence elements (for example, T2SS, T3SS, T6SS, capsules and
exotoxins) seemed to exhibit enrichment beyond what these tax- Methods
onomic shifts imply, as they were present only in some virulent Study volunteers and antibiotic exposure. Fecal samples were taken from 12
strains of Enterobacteriaceae spp. and/or other unrelated pathobi- healthy Caucasian males, aged between 18 and 40 yr with an average age of
23.4 yr (s.d. = 5.3 yr) at the start of the study, glycated haemoglobin A1c below
ont species. 43 mmol mol−1 (<6.1%) and with normal bowel function (1–3 bowel movements
Interestingly, Clostridium spp. were particularly likely to colonize per day). The exclusion criteria included: any use of antibiotics 6 months before
after the antibiotic intervention, perhaps having survived as spores. inclusion; BMI below 18.5 kg m−2 or above 25 kg m−2; smoking; abnormal serum/
This is relevant as Clostridium difficile infections are associated with plasma levels of electrolytes, lipids, creatinine, liver enzymes (alanine transaminase,
aspartate aminotransferase, alkaline phosphatase), thyroid stimulating hormone
the use of broad-spectrum antibiotics in acute care settings, which
or haemoglobin; any current or existing disease in the gastrointestinal system or
reduce host resistance to colonization and expansion of this patho- family history of inflammatory bowel disease or diabetes; lactose intolerance or
gen55. Fecal microbiota transplantation successfully treats recur- coeliac disease; allergy against the antibiotics used in this study and the use of
rent C. difficile infections by reconstituting the normal microbiota medication that could not be paused during the study period. In addition to a
homeostasis and thus breaking the infection recurrence cycle56. screening visit, the study design encompassed five study visits (D0, D4, D8, D42
and D180) and a four-day broad-spectrum antibiotic intervention consisting of
Future drug development to minimize antibiotic-induced gut dys- once-daily administration of 500 mg meropenem, 500 mg vancomycin and 40 mg
biosis should target the Clostridium spore-germination process and gentamicin dissolved in apple juice and ingested orally. This cocktail is a modified
increase the abundance of beneficial species with direct coloniza- version of a protocol used in intensive care units32 and is designed to eradicate as
tion resistance such as Bifidobacterium spp., which were depleted by many gut microorganisms as possible without causing direct side effects. None of
our intervention. We validated and extended the results of a previ- the three antibiotics are absorbed by the gut mucosa62–64. Except keeping the diet
unaltered throughout the study period and avoiding yoghurt products on the four
ous report30, where C. bolteae was enriched after a single course of days preceding each visit, no dietary regulations were required. The study was
a β-lactam antibiotic. Thus, we hypothesize that rather than being approved by the Scientific-Ethics Committee of the Capital Region of Denmark,
a direct marker for autism57–59, liver transplant rejection in rats60, or registered with ClinicalTrials.gov (ID: NCT01633762) and conducted according to
low diversity61 as previously suggested, C. bolteae might be a marker the Helsinki Declaration. All the subjects gave informed consent to participate in
of general β-lactam and/or antibiotic-induced dysbiosis, as patients this study. Data regarding glucose metabolism, and gut and pancreatic hormone
secretion have been published previously, alongside further information about the
with these disorders or conditions are likely to be overexposed to protocol and experimental procedures31. The present experiment was conducted
antibiotics. Consistently, C. bolteae is particularly resistant to β-lac- only once on the present cohort, with no separate replication cohort at this point.
tams in vitro, compared to other firmicutes38. This should be further
DNA extraction and shotgun metagenomics sequencing. Stool samples
analysed with an appropriate experimental design that enables dis- were collected before the treatment at D0 and at D4, D8, D42 and D180. The
entangling the effects of antibiotic uptake from those of the disor- participants collected fresh stool samples at home that were immediately frozen
ders themselves on the gut microbiota. in their home freezer at −20 °C and delivered to Gentofte Hospital within 36 h
By combining shotgun metagenomics with a quantitative metage- of sampling. The samples were transported using insulating polystyrene foam
nomic species approach, we tested whether antibiotic resistance containers with a transport time below 3 h and were stored at −80 °C until DNA
extraction. Microbial DNA was extracted from 200 mg frozen stool using the
potential modulates gut microbial recovery following antibiotic International Human Microbiome Standards standard operating procedure
treatment. Microorganisms that harbour β-lactam ARGs exhib- 07 V265. Briefly, bacteria were first subjected to chemical lysis and mechanical
ited initially increased odds of survival and de novo colonization lysis by bead-beating; next debris, aromatic compounds, proteins and RNA

were eliminated and finally, DNA was precipitated using alcohol. To enable the 3. Set Ci = Ui.
extraction of sufficient DNA from seven of the D4 and two of the D8 samples, 4. For each read-pair that mapped to multiple entries (for example, i1, i2,
we modified the DNA extraction procedure by extracting DNA from 3 × 200 mg i3, …, in) in IGChg38, share it proportionately between these entries based on
feces, adding only one-third of the advised amount of both the phosphate buffer their abundance estimated from uniquely mapped read-pairs (Ui1, Ui2, Ui3, …,
and potassium acetate to each sample, and by adding 6 µl RNase (instead of 2 µl) Uin). For example, consider a read-pair mapping to multiple entries i1, i2,…
after pooling the three samples. The DNA concentration was determined using a , in. For each j in 1, 2,…, n, the respective Cij entry would be incremented by
Qubit Fluorometer (Thermo Scientific) and the quality of the extracted DNA was Uij/ΣkUik.
estimated by agarose gel electrophoresis. Three of the samples from D4 repeatedly 5. Remove the entries corresponding to human chromosomes.
failed during library construction and were therefore discarded. We optimized 6. Normalize the counts by gene lengths to obtain abundances.
the library preparation for six samples at D4 and two samples at D8 (marked 7. Convert them into relative abundances by dividing individual abundances by
by the suffix ‘opt’ in the Sample ID in Supplementary Table 2). Whole-genome the total abundance.
shotgun sequencing was performed on the remaining 57 fecal samples using the
By using the functional annotations of IGChg38 genes provided in the most
Illumina HiSeq 2000 platform under paired-end sequencing (2 × 100 base pairs
recent version of the MOCAT2 software40, we estimated the relative abundance
(bp)). Data generation from the samples through metagenomic sequencing has
of functional pathways, ARGs and virulence gene families, using the following
been shown in a previous report to be fairly robust across multiple replications
databases: KEGG47, CARD41, ResFams42 and PATRIC53. We summed the abundance
when standard protocols are applied36, thus no explicit replication of this step was
of all ARGs providing resistance to any glycopeptide, β-lactam or aminoglycoside
made. We generated 79.4 ± 18.0 million raw metagenomic reads (7.94 ± 1.8 Gb)
(see Results for further reasoning). Genes from the IGChg38 were assigned to a
per sample. The sequencing depths (in millions of reads) for D0, D4, D8, D42 and
CARD model by applying the CARD RGI software, requiring a hit scoring above
D180 samples were 76.5 ± 11.1, 78.1 ± 13.2, 75.6 ± 19.6, 81.2 ± 11.4 and 85.4 ± 26.6,
the family-specific threshold, with the top hit taken if several were achieved.
respectively, suggesting that the read depths were not significantly lower in the
Similarly, ResFams hits were assigned to genes if no CARD hit was assigned and
samples immediately after the intervention. The reads were quality controlled by
the score to a ResFams HMM model exceeded the gathering threshold for that
removing adaptors and by trimming the reads using a quality trimming cut-off
model. Of the three ARG models in CARD version 1.1.5, we excluded target loss
of 20 and a minimum read length of 30 bp66. On average, 6.8 million reads were
models (where loss of a gene confers resistance) and protein variant models (for
filtered out due to adaptor contamination and 0.94 million reads did not pass the
example where known single-nucleotide variants affect antibiotic susceptibility)
trimming criteria. Human DNA reads were removed by screening them against
as ARGs under these models cannot be reliably identified using our analysis
the human genome (version hg19). On average, 0.24 million reads were removed
pipeline. Instead we used only the CARD homolog models, where under
due to human contamination. This resulted in 69.3 ± 8.7, 66.5 ± 13.1, 68.2 ± 15.8,
assumptions of curation of the database, the presence of a member of an ARG
72.0 ± 12.1 and 79.8 ± 22.6 million high-quality non-human reads per sample for
family is considered a reliable indicator for probable ARG potential. When
D0, D4, D8, D42 and D180, respectively. Sample information and quality control
using the homolog models, we assume that metagenomic reads highly similar
summary statistics for the reads are shown in Supplementary Table 2.
to an ARG from a model (having >95% nucleotide similarity) will confer this
functional capacity. Although this cannot be fully guaranteed, 95% nucleotide
Taxonomic characterization and ecosystem diversity measures. To generate
identity is stringent enough to identify genes with very closely related functions,
taxonomic abundance profiles, we used the MOCAT and MOCAT2 software
which makes it unlikely that any exceptions will have a significant impact on
packages40,66. Screened high-quality reads were aligned (alignment length cut-off
our reported results. This is also in line with the current state of the art for
of 45 bp and minimum 97% sequence identity for the option ‘screen’) to a database
metagenomic analyses. We tested whether there is a bias introduced by our choice
consisting of 10 universal single-copy marker genes extracted from 3,496 NCBI
of the ARG database and found that our results on the antibiotics used in this
reference genomes and 263 metagenomes34. We obtained relative abundances for
study and on MEPs hold when using the Antibiotic Resistance Gene Database69,
477 species-level mOTUs. We discarded the low abundance species by removing
which due to its greater age has different discovery biases (Supplementary Fig. 8).
those that were present in less than two study samples and whose average relative
Additionally, we benchmarked our ARG pipeline on three example, well-curated,
abundance across all samples was lower than 0.01%, resulting in a pruned relative
bacterial genomes. Predicted and manually curated gene assignments (taken from
abundance table containing 269 mOTUs. Note that the mOTU annotated here
GenBank) are shown in Supplementary Table 3. In brief, of the 154 ARGs assigned
as C. bolteae (relevant in the discussion of differential abundance analyses—see
by our pipeline to these three genomes, none conflict with the curated annotation
Results) has been reclassified as Lachnoclostridium bolteae recently67 and referred
and 142 are exact matches or synonyms. The remaining 12/154 are all less specific
to by this name in a previous publication30.
assignments than the original annotation (for example ABC efflux pump instead
Based on this table, we calculated species alpha diversity using two measures,
of specifically MsbA). Thus, we conclude that the method used here is very
the richness per sample by counting the number of mOTUs with non-zero
conservative and that although false positives must occur, they will be relatively few
abundance and the Shannon diversity index per sample. The species diversity
and we are confident of the results generated with our ARG pipeline. The virulence
measures can be affected by the amount of DNA available for sequencing, therefore
gene abundances were binned at the level of PATRIC functional descriptors. For
we analysed whether or not library optimization had any effect on the diversity
differential abundance testing at the microbial functional level, we discarded low
measures by plotting the richness and Shannon diversity together with the DNA
abundance KEGG modules and pathways by removing those that were present in
mass per sample (Supplementary Fig. 7). We did not observe any particular pattern
fewer than two samples and whose average relative abundance across all samples
for how high or low amounts of DNA in a sample affected by these metrics.
was lower than 0.01%, resulting in a pruned relative abundance table containing
403 functional modules and 257 pathways.
Functional characterization. The high-quality reads were aligned against a
combined database (IGChg38 hereafter) consisting of the hg38 release of the Identifying metagenomic species. We demonstrated previously how binning
human reference genome and the IGC consisting of 9.9 million non-redundant co-abundant genes across a series of metagenomic samples enables the
microbial genes39. We used bwa mem68 (version 0.7.15-r1140) with default construction of MGSs based on a large gene catalog without the need for a
parameters. The purpose of adding the human genome was to filter out reads that reference genome database33. In the present study we used the IGC that is more
mapped as well or better to some human sequence than to any bacterial gene. The than 2 ×larger than the previously used catalog33. We built a more complete and
alignments were first filtered to only retain alignments longer than 50 bp with accurate collection of such MGSs by grouping co-abundant genes across these
>95% sequence identity. The highest scoring alignment(s) was/were then kept for 1,267 metagenomes39, identifying 1,264 MGSs (each comprising at least 700
each read. Alignments were performed separately for paired-end and single-read genes). In our analyses, to estimate the relative abundance of an MGS in a sample,
libraries (single-reads are generated when a corresponding read from the other we calculated the median relative abundance of all genes belonging to that MGS
direction is removed due to the filtering steps mentioned earlier). As IGChg38 is in that sample. We then considered the MGS detected if the relative abundance
a database predominantly of genes and not genomes, there will be a significant calculated in this way was non-zero (that is, at least half the genes had non-zero
number of read-pairs where one end maps within the gene and the other end maps abundance). The MGSs were annotated to species level if more than 50% of their
outside of the gene—either mapping to an adjacent gene or remaining unmapped genes could be aligned to a RefSeq reference genome at 95% sequence similarity and
by falling in the intergenic region. Therefore, we counted a whole read-pair aligning less than 10% of their genes had an alternative taxonomy. MGSs that could not be
to a gene when (1) both ends from a read-pair mapped to the same gene, (2) only annotated at species level were then annotated at the genus level in a similar manner
one end from a read-pair mapped to the gene, or (3) a read from the single-read using an 85% sequence similarity requirement. Overall, we were able to map 278
library mapped to the gene. The percentages of read-pairs that mapped to the MGSs to the genus and/or species level. We discarded low abundance MGSs by
IGChg38 from D0, D4, D8, D42 and D180 samples were 95.4 ± 1.2, 76.1 ± 21.3, removing those that were present in fewer than two samples and whose average
92.0 ± 5.7, 95.6 ± 1.4 and 95.7 ± 2.4, respectively. In a given metagenomic sample, we relative abundance across all samples was lower than 0.0001%, resulting in a pruned
estimated the relative abundance of each gene in IGC as follows: relative abundance table containing 582 MGSs. Given that we had a sparse relative
abundance table with many zero values, specifically for the differential abundance
1. Initialize two arrays of read-pair counts with 0 for each entry in IGChg38,
analyses, we set the average relative abundance threshold to 0.01% as with the
Ui and Ci.
mOTUs, resulting in a pruned relative abundance table containing 407 MGSs.
2. For each read-pair that maps uniquely to entry i in IGChg38, add 1 to the
corresponding count Ui. This results in a uniquely mapped read-pair count Control populations. To determine whether the changes reported here were really
for each gene. associated with the antibiotic treatment and not simply explained by the passage of

time, we employed three different control populations of previously published gut calculated using the formula OR = ad/bc. Odds ratios are reported in the text with
metagenome samples. We downloaded the metagenomes from two cohorts, HMP35 95% confidence intervals. FDR-adjusted q values and odds of survival/colonization
and another healthy cohort used in a previous report to study the temporal and were represented in a heatmap. To ensure that our findings were not due to ARGs
technical variability of the human microbiome36 (Voigt; antibiotic-treated samples against one class being found in low-abundant, transient gut microbiota, we used
were excluded), from the European Nucleotide Archive. Eighty-two paired HMP HMP samples to test whether MGSs carrying ARGs against different classes
samples from 41 donors sampled at two different time points with time spans were significantly different in their relative abundance, but we did not find any
>100 d and 35 paired Voigt samples from 7 donors with inter-sample time spans of significant differences (Supplementary Fig. 9).
2–773 d were processed using MOCAT as described above to produce comparable To assess the enrichment of virulence factors after antibiotic exposure we
taxonomic and gene abundance profiles. calculated the mean abundance of virulence factors (summed at the level of functional
descriptors) across all samples at each time point and we scaled this number for easier
Statistical tests. To study the overall effects of antibiotic treatment on the visualization by subtracting D0 abundance and dividing by D0 standard deviation
gut microbiota, we determined ecosystem-level metrics such as alpha and (analogous to a Z-score). A one-sided Wilcoxon rank-sum test was used to test
beta diversity and used ordination techniques to plot and inspect the relative whether there was significant enrichment of virulence factors following antibiotic
abundances of the mOTU taxonomic entities resolved. Species richness, Shannon exposure with respect to D0 and FDR-adjusted P values were reported. All statistical
index and Bray–Curtis dissimilarities were computed and principal coordinate tests were performed within the R statistical computing environment.
analyses were conducted with R using the vegan and ape packages. In amplicon-
based studies, where taxonomic resolution varies strongly, a phylogenetic distance Breakdown of ARG signal by taxonomy. To assess whether or not the changes
measure like UniFrac would be required. Here, as we can achieve a uniform and seen under treatment in antibiotic resistance gene abundance across the cohort
high resolution through deep shotgun metagenomic data, we instead made use of can be reduced to contributions from particular clades, we can break down such
the Bray–Curtis distance metric, thereby achieving compatibility with previous contributions by counting abundance separately of Co-abundance gene groups/
shotgun metagenomic studies. Samples were rarefied to 14 million inserts using MGSs linked to different clades. Such taxonomic classification is achieved based
USEARCH software70 before conducting richness and diversity comparisons. on the highest similarity of their constituent IGC genes to genomes from the
Two-sided paired Wilcoxon signed-rank tests were used to test whether species respective clades, under BLASTn analysis. To classify a Co-abundance gene group
richness and Shannon diversity were significantly different between paired samples as belonging to a broader clade (for example a bacterial family), at least 40% of its
at consecutive time points. Two-sided Wilcoxon rank-sum tests were used to test genes must be most similar to gene variants found within that clade, with no more
whether Bray–Curtis dissimilarities between same-donor samples in our study than 15% most similar to gene variants from another clade. Performing such an
were significantly different to those between same-donor samples with similar analysis for bacterial family, based on the high prevalence of many known ARGs in
time separation in control datasets. We also used two-sided paired Wilcoxon Enterobacteriaceae, reveals that the aminoglycoside resistance spike at D8 seems
signed-rank tests to determine whether the microbial composition (mOTU and driven by this family, whereas the effects on other ARG classes traced are not
MGS relative abundances) or KEGG functional features (modules and pathways) (Supplementary Fig. 10).
changed significantly at each sampling time point (D4, D8, D42, D180) compared Moreover, although the classification of groups of genes correlated in
to D0. For better interpretability of results, we removed mOTU linkage groups abundance as either mobile element entities or bacterial chromosomes is still
without taxonomic annotation before performing differential abundance tests, only a partially solved problem, a basic proxy for this classification stems from
which resulted in a relative abundance table containing 106 mOTU species. We whether or not such groups are large enough to be chromosomal (MGSs) or too
adjusted the P values obtained using the Benjamini–Hochberg FDR-control small for this to be the case (Co-abundance gene groups). To assess whether the
procedure and reported all changes significant at q < 0.05. patterns observed here are driven largely by genes found on chromosomes or
Given that the relative abundances of taxa in an ecosystem have to sum to one, on smaller entities such as plasmids, a first check is thus possible by breaking
relative abundances are compositional data. Differential analysis based on relative down the respective contribution of these classes of gene groups. Supplementary
abundances is thus susceptible to compositional effects, as an increase in the Fig. 11 shows this analysis, revealing that the bulk of the ARG abundance from
relative abundance of one taxon will lead to a decrease in the relative abundances each sample traces to groups of correlated genes likely to represent bacterial
of other components71–73. To verify whether the observed microbial shifts were chromosomes/taxa rather than smaller mobile elements like phages or plasmids.
real or spurious differences arising due to such effects, we used a previously
described compositionality test74, which verifies whether differences in the relative Robustness to sequencing depth differences. We note that all our samples had
abundance of a taxon between time points (or groups) were due to a dramatic a minimum sequencing depth of 17.6 million non-human read-pairs and at least
abundance change in another taxon (Supplementary Table 1). We did not test 11.9 million read-pairs were mapped to IGC. Thus, our detection threshold is quite
differences between D0 and D4 because, due to the nature of the intervention, low, on the order of 1 in 10 million. Although we cannot rule out that some specific
there were pronounced changes during this period in most of the microorganisms, species were present in a sample below our detection threshold, the overall trends
dramatically affecting the relative abundances of the surviving species. That reported in our survival and de novo colonization analysis were not affected by
compositional effects will complicate the interpretation of observed changes sequencing depth (see Supplementary Fig. 6).
between these time points can therefore be assumed, even without such a test.
Summing the ARG abundances for each drug, we calculated relative abundances Code availability. For the differential abundance testing, we used a
of ARG abundance against β-lactams, aminoglycosides and glycopeptides from all compositionality test available at: https://github.com/apalleja/compositionality_
samples at each time point. We also considered MEPs, which can provide broad test/. Otherwise standard R packages were used.
resistance to the three study antibiotics and an additional category where we pooled
all other ARGs for comparison (‘others’). Two-sided paired Wilcoxon signed-rank Reporting Summary. Further information on experimental design is available in
tests were performed to test whether there was a significant enrichment in ARGs the Nature Research Reporting Summary linked to this article.
against the antibiotic classes used in the intervention at each time point compared to
D0 and FDR-adjusted P values were reported. Data availability
To investigate whether carrying ARGs against the antibiotics used in the study The high-quality reads have been deposited in the European Nucleotide Archive
is an advantage for a species in terms of its survival through the intervention or with accession number ERP022986. Relative abundances of taxa and functional
de novo colonization after the intervention, we generated contingency tables for features can be downloaded at http://arumugamlab.sund.ku.dk/SuppData/Palleja_
how the survival (or de novo colonization) potential of MGS species changes et_al_2018_ABX/.
depending on whether or not they were carrying appropriate ARGs. We compared
all post-intervention time points to D0 in this analysis. For instance, to investigate
the advantages of carrying aminoglycoside ARGs for survival during the first Received: 30 May 2018; Accepted: 28 August 2018;
eight days of the study, this survival versus non-survival contingency table was Published: xx xx xxxx
determined as follows: a) the number of MGSs that were observed in both D0 and
D8 (that is, that survived), carrying aminoglycoside ARGs, b) the number of MGSs
that were observed in both D0 and D8, not carrying aminoglycoside ARGs, c) References
the number of MGSs that were observed in D0 but not observed in D8 (that is, 1. Lynch, S. V. & Pedersen, O. The human intestinal microbiome in health and
that did not survive), carrying aminoglycoside ARGs and d) the number of MGSs disease. N. Engl. J. Med. 375, 2369–2379 (2016).
that were observed in D0 but not observed in D8, not carrying aminoglycoside 2. Manges, A. R. et al. Comparative metagenomic study of alterations to the
ARGs. An MGS was counted multiple times if it occurred in multiple samples, intestinal microbiota and risk of nosocomial Clostridum difficile-associated
thus we counted the MGS-sample pairs above. Similarly, contingency tables were disease. J. Infect. Dis. 202, 1877–1884 (2010).
constructed to investigate the advantages of carrying aminoglycoside ARGs for 3. Ley, R. E. et al. Obesity alters gut microbial ecology. Proc. Natl Acad. Sci.
de novo colonization by counting MGSs that were not observed on D0 but USA 102, 11070–11075 (2005).
observed on D8 (that is, those that de novo colonized) and MGSs that were 4. Turnbaugh, P. J. et al. An obesity-associated gut microbiome with increased
observed both on D0 and D8 (that is, those that already existed). Fisher’s exact tests capacity for energy harvest. Nature 444, 1027–1031 (2006).
were performed on these contingency tables to test whether the chances of survival 5. Forslund, K. et al. Disentangling type 2 diabetes and metformin treatment
or de novo colonization are affected by observable ARG carriage. The OR was signatures in the human gut microbiota. Nature 528, 262–266 (2015).

6. Karlsson, F. H. et al. Gut metagenome in European women with normal, 39. Li, J. et al. An integrated catalog of reference genes in the human gut
impaired and diabetic glucose control. Nature 498, 99–103 (2013). microbiome. Nat. Biotechnol. 32, 834–841 (2014).
7. Qin, J. et al. A metagenome-wide association study of gut microbiota in type 40. Kultima, J. R. et al. MOCAT2: a metagenomic assembly, annotation and
2 diabetes. Nature 490, 55–60 (2012). profiling framework. Bioinformatics 32, 2520–2523 (2016).
8. Craven, M. et al. Inflammation drives dysbiosis and bacterial invasion in 41. Jia, B. et al. CARD 2017: expansion and model-centric curation of the
murine models of ileal Crohn’s disease. PLoS ONE 7, e41594 (2012). comprehensive antibiotic resistance database. Nucleic Acids Res. 45,
9. Hsiao, E. Y. et al. Microbiota modulate behavioral and physiological D566–D573 (2017).
abnormalities associated with neurodevelopmental disorders. Cell 155, 42. Gibson, M. K., Forsberg, K. J. & Dantas, G. Improved annotation of antibiotic
1451–1463 (2013). resistance determinants reveals microbial resistomes cluster by ecology.
10. Konig, J. et al. Consensus report: faecal microbiota transfer—clinical ISME J. 9, 207–216 (2015).
applications and procedures. Aliment. Pharmacol. Ther. 45, 222–239 (2017). 43. Queenan, A. M. & Bush, K. Carbapenemases: the versatile β-lactamases.
11. Bindels, L. B., Delzenne, N. M., Cani, P. D. & Walter, J. Towards a more Clin. Microbiol. Rev. 20, 440–458 (2007).
comprehensive concept for prebiotics. Nat. Rev. Gastroenterol. Hepatol. 12, 44. Werner, G., Strommenger, B. & Witte, W. Acquired vancomycin resistance
303–310 (2015). in clinically relevant pathogens. Future Microbiol. 3, 547–562 (2008).
12. Hemarajata, P. & Versalovic, J. Effects of probiotics on gut microbiota: 45. Damier-Piolle, L., Magnet, S., Brémont, S., Lambert, T. & Courvalin, P.
mechanisms of intestinal immunomodulation and neuromodulation. AdeIJK, a resistance-nodulation-cell division pump effluxing multiple
Therap. Adv. Gastroenterol. 6, 39–51 (2013). antibiotics in Acinetobacter baumannii. Antimicrob. Agents Chemother. 52,
13. Mikkelsen, K. H., Allin, K. H. & Knop, F. K. Effect of antibiotics on gut 557–562 (2008).
microbiota, glucose metabolism and body weight regulation: a review of the 46. Hou, P. F., Chen, X. Y., Yan, G. F., Wang, Y. P. & Ying, C. M. Study of the
literature. Diabetes Obes. Metab. 18, 444–453 (2016). correlation of imipenem resistance with efflux pumps AdeABC, AdeIJK,
14. Hollis, A. & Ahmed, Z. Preserving antibiotics, rationally. N. Engl. J. Med. 369, AdeDE and AbeM in clinical isolates of Acinetobacter baumannii.
2474–2476 (2013). Chemotherapy 58, 152–158 (2012).
15. Nobel, Y. R. et al. Metabolic and metagenomic outcomes from early-life 47. Kanehisa, M. et al. Data, information, knowledge and principle: back to
pulsed antibiotic treatment. Nat. Commun. 6, 7486 (2015). metabolism in KEGG. Nucleic Acids Res. 42, D199–D205 (2014).
16. Cho, I. & Blaser, M. J. The human microbiome: at the interface of health and 48. Pu, Y. et al. Enhanced efflux activity facilitates drug tolerance in dormant
disease. Nat. Rev. Genet. 13, 260–270 (2012). bacterial cells. Mol. Cell 62, 284–294 (2016).
17. Frohlich, E. E. et al. Cognitive impairment by antibiotic-induced gut 49. Winter, S. E. et al. Gut inflammation provides a respiratory electron acceptor
dysbiosis: analysis of gut microbiota-brain communication. Brain Behav. for Salmonella. Nature 467, 426–429 (2010).
Immun. 56, 140–155 (2016). 50. Winter, S. E. et al. Host-derived nitrate boosts growth of E. coli in the
18. Korpela, K. et al. Intestinal microbiome is related to lifetime antibiotic use in inflamed gut. Science 339, 708–711 (2013).
Finnish pre-school children. Nat. Commun. 7, 10410 (2016). 51. Rivera-Chávez, F. et al. Depletion of butyrate-producing Clostridia from the
19. Bailey, L. C. et al. Association of antibiotics in infancy with early childhood gut microbiota drives an aerobic luminal expansion of Salmonella. Cell Host
obesity. JAMA Pediatr. 168, 1063–1069 (2014). Microbe 19, 443–454 (2016).
20. Saari, A., Virta, L. J., Sankilampi, U., Dunkel, L. & Saxen, H. Antibiotic 52. Zhu, W. et al. Precision editing of the gut microbiota ameliorates colitis.
exposure in infancy and risk of being overweight in the first 24 months of Nature 553, 208–211 (2018).
life. Pediatrics 135, 617–626 (2015). 53. Driscoll, T. et al. Integration and visualization of host-pathogen data related
21. Shaw, S. Y., Blanchard, J. F. & Bernstein, C. N. Association between the use to infectious diseases. Bioinformatics 27, 2279–2287 (2011).
of antibiotics and new diagnoses of Crohn’s disease and ulcerative colitis. 54. Rutherford, S. T. & Bassler, B. L. Bacterial quorum sensing: its role in
Am. J. Gastroenterol. 106, 2133–2142 (2011). virulence and possibilities for its control. Cold Spring Harb. Perspect. Med. 2,
22. Trasande, L. et al. Infant antibiotic exposures and early-life body mass. a012427 (2012).
Int. J. Obes. 37, 16–23 (2013). 55. Buffie, C. G. & Pamer, E. G. Microbiota-mediated colonization resistance
23. Wright, G. D. The antibiotic resistome: the nexus of chemical and genetic against intestinal pathogens. Nat. Rev. Immunol. 13, 790–801 (2013).
diversity. Nat. Rev. Microbiol. 5, 175–186 (2007). 56. Kassam, Z., Lee, C. H., Yuan, Y. & Hunt, R. H. Fecal microbiota
24. van Schaik, W. The human gut resistome. Phil. Trans. R. Soc. B 370, transplantation for Clostridium difficile infection: systematic review and
20140087 (2015). meta-analysis. Am. J. Gastroenterol. 108, 500–508 (2013).
25. Forslund, K. et al. Country-specific antibiotic use practices impact the human 57. Finegold, S. M. et al. Gastrointestinal microflora studies in late-onset autism.
gut resistome. Genome Res. 23, 1163–1169 (2013). Clin. Infect. Dis. 35, S6–S16 (2002).
26. Rashid, M. U. et al. Determining the long-term effect of antibiotic 58. Song, Y., Liu, C. & Finegold, S. M. Real-time PCR quantitation
administration on the human normal intestinal microbiota using culture and of Clostridia in feces of autistic children. Appl. Environ. Microbiol. 70,
pyrosequencing methods. Clin. Infect. Dis. 60, S77–S84 (2015). 6459–6465 (2004).
27. Zaura, E. et al. Same exposure but two radically different responses to 59. Luna, R. A. et al. Distinct microbiome-neuroimmune signatures correlate
antibiotics: resilience of the salivary microbiome versus long-term microbial with functional abdominal pain in children with autism spectrum disorder.
shifts in feces. mBio 6, e01693-15 (2015).
Cell. Mol. Gastroenterol. Hepatol. 3, 218–230 (2017).
28. Reijnders, D. et al. Effects of gut microbiota manipulation by antibiotics on
60. Ren, Z. et al. Intestinal microbial variation may predict early acute rejection
host metabolism in obese humans: a randomized double-blind placebo-
after liver transplantation in rats. Transplantation 98, 844–852 (2014).
controlled trial. Cell Metab. 24, 63–74 (2016).
61. Le Chatelier, E. et al. Richness of human gut microbiome correlates with
29. Abeles, S. R. et al. Microbial diversity in individuals and their household
metabolic markers. Nature 500, 541–546 (2013).
contacts following typical antibiotic courses. Microbiome 4, 39 (2016).
62. Rao, S., Kupfer, Y., Pagala, M., Chapnick, E. & Tessler, S. Systemic absorption
30. Raymond, F. et al. The initial state of the human gut microbiome determines
its reshaping by antibiotics. ISME J. 10, 707–720 (2016). of oral vancomycin in patients with Clostridium difficile infection. Scand.
31. Mikkelsen, K. H. et al. Effect of antibiotics on gut microbiota, gut hormones J. Infect. Dis. 43, 386–388 (2011).
and glucose metabolism. PLoS ONE 10, e0142352 (2015). 63. Rohrbaugh, T. M., Anolik, R., August, C. S., Serota, F. T. & Koch, P. A.
32. Houben, A. J. et al. Selective decontamination of the oropharynx and the Absorption of oral aminoglycosides following bone marrow transplantation.
digestive tract, and antimicrobial resistance: a 4 year ecological study in Cancer 53, 1502–1506 (1984).
38 intensive care units in the Netherlands. J. Antimicrob. Chemother. 69, 64. Craig, W. A. The pharmacology of meropenem, a new carbapenem antibiotic.
797–804 (2014). Clin. Infect. Dis. 24, S266–S275 (1997).
33. Nielsen, H. B. et al. Identification and assembly of genomes and genetic 65. IHMS_SOP 07 V2: Standard Operating Procedure for Fecal Samples DNA
elements in complex metagenomic samples without using reference genomes. Extraction Protocol H INRA (IHMS, 2015); http://www.microbiome-
Nat. Biotechnol. 32, 822–828 (2014). standards.org/index.php?id=254
34. Sunagawa, S. et al. Metagenomic species profiling using universal 66. Kultima, J. R. et al. MOCAT: a metagenomics assembly and gene prediction
phylogenetic marker genes. Nat. Methods 10, 1196–1199 (2013). toolkit. PLoS ONE 7, e47656 (2012).
35. Consortium, H. M. P. Structure, function and diversity of the healthy human 67. Yutin, N. & Galperin, M. Y. A genomic update on clostridial phylogeny:
microbiome. Nature 486, 207–214 (2012). Gram-negative spore formers and other misplaced clostridia. Environ.
36. Voigt, A. Y. et al. Temporal and technical variability of human gut Microbiol. 15, 2631–2641 (2013).
metagenomes. Genome Biol. 16, 73 (2015). 68. Li, H. & Durbin, R. Fast and accurate short read alignment with Burrows-
37. Talukdar, P. K., Olguin-Araneda, V., Alnoman, M., Paredes-Sabja, D. & Wheeler transform. Bioinformatics 25, 1754–1760 (2009).
Sarker, M. R. Updates on the sporulation process in Clostridium species. 69. Liu, B. & Pop, M. ARDB—antibiotic resistance genes database. Nucleic Acids
Res. Microbiol. 166, 225–235 (2015). Res. 37, D443–D447 (2009).
38. Maier, L. et al. Extensive impact of non-antibiotic drugs on human gut 70. Edgar, R. C. Search and clustering orders of magnitude faster than BLAST.
bacteria. Nature 555, 623–628 (2018). Bioinformatics 26, 2460–2461 (2010).

71. Lovell, D., Pawlowsky-Glahn, V., Egozcue, J. J., Marguerat, S. & Bähler, J. performed sample collections and carried out patient phenotyping. K.H.A. supervised
Proportionality: a valid alternative to correlation for relative data. PLoS the microbial DNA extraction. S.L., C.Z., J.W., Q.F. and H.Y. performed shotgun
Comput. Biol. 11, e1004075 (2015). metagenomics sequencing and taxonomic profiling. P.T.P., L.P.C. and M.A. estimated
72. Weiss, S. et al. Normalization and microbial differential abundance strategies IGC gene profiles. H.B.N. generated the MGS groups based on IGC. A.P., S.K.F. and
depend upon data characteristics. Microbiome 5, 27 (2017). A.K. designed and performed the data analysis. M.A., T.H., P.B. and O.P. supervised
73. Aitchison, J. The statistical analysis of compositional data. J. R. Stat. the data analysis. A.P., S.K.F., K.H.M. and M.A. wrote the paper. K.H.A., T.N., T.H.H.,
Soc. Ser. B 44, 139–177 (1982). A.K., H.B.N., J.W., A.T., P.B., T.V., F.K.K., T.H. and O.P. revised the paper. All authors
74. Palleja, A. et al. Roux-en-Y gastric bypass surgery of morbidly obese patients contributed to data interpretation, discussion and editing of the paper. All authors read
induces swift and persistent changes of the individual gut microbiota. and approved the final manuscript.
Genome Med. 8, 67 (2016).
Acknowledgements Competing interests

The authors declare no competing interests.
This work was funded by an international alliance grant from The Novo Nordisk
Foundation Center for Basic Metabolic Research, which is an independent Research
Center at the University of Copenhagen partially funded by an unrestricted donation
from the Novo Nordisk Foundation (grant no. NNF10CC1016515). Our work was also Additional information
funded by the TARGET research initiative (Danish Strategic Research Council [0603– Supplementary information is available for this paper at https://doi.org/10.1038/
00484B]), the Danish Diabetes Academy supported by the Novo Nordisk Foundation, the s41564-018-0257-9.
Danish Council for Independent Research (Medical Sciences), and the Danish Diabetes
Reprints and permissions information is available at www.nature.com/reprints.
Association. S.K.F. was funded by FP7 METACARDIS HEALTH-F4-2012-305312.
Correspondence and requests for materials should be addressed to F.K.K. or M.A. or O.P.
Author contributions Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in
O.P., T.H. and F.K.K. devised the study protocol. M.F.N. participated in the protocol published maps and institutional affiliations.
design and application and in the participant recruitment and selection. K.H.M. © The Author(s), under exclusive licence to Springer Nature Limited 2018

nature research | life sciences reporting summary
Corresponding author(s): Albert Palleja
Life Sciences Reporting Summary

Nature Research wishes to improve the reproducibility of the work that we publish. This form is intended for publication with all accepted life
science papers and provides structure for consistency and transparency in reporting. Every life science submission will use this form; some list
items might not apply to an individual manuscript, but all fields must be completed for clarity.
For further information on the points included in this form, see Reporting Life Sciences Research. For further information on Nature Research
policies, including our data availability policy, see Authors & Referees and the Editorial Policy Checklist.
Please do not complete any field with "not applicable" or n/a. Refer to the help text for what text to use if an item is not relevant to your study.
For final submission: please carefully check your responses for accuracy; you will not be able to make changes later.
` Experimental design
1. Sample size
Describe how sample size was determined. This is the first metagenomic study characterizing the effects of a prospective eradication of
the gut microbiota of young healthy adults using a multiantibiotic treatment. No formal
sample size calculations were employed for this prospective analysis, but the cohort size was
based on what has been sufficient to test host phenotype effects in previous clinical studies
as well as in other metagenomic studies of comprehensive intervention effects.
2. Data exclusions
Describe any data exclusions. Three samples from day 4 repeatedly failed during library construction and were therefore
discarded.
3. Replication
Describe the measures taken to verify the reproducibility For the experimental findings, these are the first of its kind, analyzing the effects of
of the experimental findings. multiantibiotic treatment on the gut microbial communities of healthy young adults. No
additional replication cohort were recruited for this study and we suggest such replication be
undertaken in discussing the results. However we validated observations of previous single
antibiotic treatments in humans.
4. Randomization
Describe how samples/organisms/participants were We followed only one group (as per a longitudinal study of a single condition, multiantibiotic
allocated into experimental groups. treatment) of 12 healthy young Danish men with specific inclusion criteria described in
doi:10.1371/journal.pone.0142352, therefore there was no need to randomize the study
subjects. Public data previously generated using equivalent protocols were used for non-
intervention controls, compared in cross-section and over time intervals.
5. Blinding
Describe whether the investigators were blinded to Under longitudinal analysis of a single condition, no blinding was either possible or necessary.
group allocation during data collection and/or analysis. All sample processing and statistical analysis was done in an automated computational
framework which does not make use of the time point or intervention status information
except when tested for, making effective blinding occur by default.
Note: all in vivo studies must report how sample size was determined and whether blinding and randomization were used.
November 2017
1
6. Statistical parameters

For all figures and tables that use statistical methods, confirm that the following items are present in relevant figure legends (or in the
Methods section if additional space is needed).
n/a Confirmed
The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement (animals, litters, cultures, etc.)
A description of how samples were collected, noting whether measurements were taken from distinct samples or whether the same
sample was measured repeatedly
A statement indicating how many times each experiment was replicated
The statistical test(s) used and whether they are one- or two-sided
Only common tests should be described solely by name; describe more complex techniques in the Methods section.
A description of any assumptions or corrections, such as an adjustment for multiple comparisons

Test values indicating whether an effect is present
Provide confidence intervals or give results of significance tests (e.g. P values) as exact values whenever appropriate and with effect sizes noted.
A clear description of statistics including central tendency (e.g. median, mean) and variation (e.g. standard deviation, interquartile range)
Clearly defined error bars in all relevant figure captions (with explicit mention of central tendency and variation)
See the web collection on statistics for biologists for further resources and guidance.
` Software
Policy information about availability of computer code
7. Software
Describe the software used to analyze the data in this We used the MOCAT and MOCAT2 software to process and annotate the sequenced reads
study. (doi:10.1371/journal.pone.0047656; doi:10.1093/bioinformatics/btw183). For the differential
abundance testing we used a compositionality test available at: https://github.com/apalleja/
compositionality_test. Otherwise standard R packages were used, which are cited and
publicly available.
For manuscripts utilizing custom algorithms or software that are central to the paper but not yet described in the published literature, software must be made
available to editors and reviewers upon request. We strongly encourage code deposition in a community repository (e.g. GitHub). Nature Methods guidance for
providing algorithms and software for publication provides further information on this topic.
` Materials and reagents

Policy information about availability of materials
8. Materials availability
Indicate whether there are restrictions on availability of The high quality reads have been deposited in the European Nucleotide Archive with
unique materials or if these materials are only available accession number ERP022986. Relative abundances of taxa and functional features can be
for distribution by a third party. downloaded at http://arumugamlab.sund.ku.dk/SuppData/Palleja_et_al_2017_ABX/.
9. Antibodies
Describe the antibodies used and how they were validated n/a
for use in the system under study (i.e. assay and species).
10. Eukaryotic cell lines
a. State the source of each eukaryotic cell line used. n/a
b. Describe the method of cell line authentication used. n/a
c. Report whether the cell lines were tested for n/a

mycoplasma contamination.
d. If any of the cell lines used are listed in the database n/a
of commonly misidentified cell lines maintained by
November 2017
ICLAC, provide a scientific rationale for their use.
2
` Animals and human research participants

Policy information about studies involving animals; when reporting animal research, follow the ARRIVE guidelines
11. Description of research animals
Provide all relevant details on animals and/or n/a
animal-derived materials used in the study.
Policy information about studies involving human research participants

12. Description of human research participants
Describe the covariate-relevant population This study includes 12 young healthy men phenotypically and metabolically characterized in a
characteristics of the human research participants. previous study (doi:10.1371/journal.pone.0142352). The inclusion criteria is detailed in this
previous publication. Subjects had a mean age of 23.4 years (standard deviation 5.3 years) at
the start of the study.
November 2017

Recovery of Gut Microbiota of Healthy Adults Following Antibiotic Exposure

Uploaded by

Copyright:

Available Formats

Recovery of Gut Microbiota of Healthy Adults Following Antibiotic Exposure

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Recovery of Gut Microbiota of Healthy Adults Following Antibiotic Exposure

Uploaded by

Copyright:

Available Formats

Articles

Recovery of gut microbiota of healthy adults

Nature Microbiology | www.nature.com/naturemicrobiology

Nature Microbiology | www.nature.com/naturemicrobiology

Nature Microbiology | www.nature.com/naturemicrobiology

0.4 0.4 0.4 0.4

−0.2 0 0.2 0.4 0.6

Nature Microbiology | www.nature.com/naturemicrobiology

2.0 × 10–4 4 × 10–4 6 × 10–3 4 × 10–4 5 × 10–3

1.5 × 10–4 3 × 10–4 3 × 10–4 4 × 10–3

1.0 × 10–4 2 × 10–4 2 × 10–4 3 × 10–3

0.5 × 10–4 1 × 10–4 1 × 10–4 2 × 10–3

Post antibiotic treatment Control

MEPs ** ** *** ***

D8 D42 D180 D8 D42 D180 >100 d >100 d

Nature Microbiology | www.nature.com/naturemicrobiology

of Genes and Genomes (KEGG) database47, estimating the abun-

Nature Microbiology | www.nature.com/naturemicrobiology

Nature Microbiology | www.nature.com/naturemicrobiology

Nature Microbiology | www.nature.com/naturemicrobiology

Nature Microbiology | www.nature.com/naturemicrobiology

Nature Microbiology | www.nature.com/naturemicrobiology

Acknowledgements Competing interests

Nature Microbiology | www.nature.com/naturemicrobiology

Life Sciences Reporting Summary

nature research | life sciences reporting summary

A description of any assumptions or corrections, such as an adjustment for multiple comparisons

` Materials and reagents

b. Describe the method of cell line authentication used. n/a

c. Report whether the cell lines were tested for n/a

ICLAC, provide a scientific rationale for their use.

nature research | life sciences reporting summary

Policy information about studies involving human research participants

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

MEPs * *