Human gut microbiota evaluation and comparation after the separate or combined antibiotics exposure using the simulator of human intestinal microbial ecosystem (SHIME)

Background A cocktail of drugs is an emerging toxic contaminant that has potential public health risk worldwide, which also would cause human intestinal microbial disorder and develop multiple human diseases. However, to date, the combination effects of antibiotics cocktail on human intestinal microbiota dysbiosis and related health risk are not fully understood. Therefore, for the rst time, this study evaluated and compared the in vitro ability of amoxicillin (AMX) and polymyxin E (POL) used separately or combined on antibiotic resistance genes (ARGs) as well as human disease-related pathways in the simulated human gut. Results This study indicated that the combination exposure of POL with AMX reduced the occurrence of drug resistance in the gut microbiota caused by single antibiotic treatment. However, in comparison with the separate use of AMX and POL, the combined treatment exhibited a signicantly higher ability to increase the human disease-related pathways. The combination effects on genetic level might attribute to microbiota shift, as co-occurrence patterns suggested that Bidobacterium attributed to ARGs increasing in the POL treatment group and Enterobacter played a crucial role in human disease-related pathways enrichment after combination treatment. Conclusion These results may open up new perspectives for assessing the direct effects of combination antibiotics on the intestinal microbiota. These suggested side-effects should be considered for a combination of antibiotics prescription.

increase the human disease-related pathways. The combination effects on genetic level might attribute to microbiota shift, as co-occurrence patterns suggested that Bi dobacterium attributed to ARGs increasing in the POL treatment group and Enterobacter played a crucial role in human disease-related pathways enrichment after combination treatment.

Conclusion
These results may open up new perspectives for assessing the direct effects of combination antibiotics on the intestinal microbiota. These suggested side-effects should be considered for a combination of antibiotics prescription.

Background
The antibiotic therapies have been demonstrated paramount importance in the treatment of bacterial infections since its discovery in the 1940s. Nowadays the extensively used antibiotics are considered as toxic emerging contaminants, which pose severe threats to environmental ecosystems [1]. Humans are directly and indirectly exposed to different antibiotics cocktail through inhalation, ingestion of drugs, drinking water, and foods, and their risk assessments have attracted more attention, recently [2][3][4]. As a result, antibiotics may progressively enter into the human gastrointestinal tract and produce a large variety of antibiotic-resistant gut-micro ora [5]. In our previous research, it was found that both human and veterinary antibiotics were mostly detected in the gut of the Chinese population [6]. The stable intestinal microbial ecosystem has been demonstrated not only to provide essential nutrients for human health but also to modulate the immune function by protecting infectious pathogens [7,8]. However, exposure to antibiotics cocktail may lead to the disrupt of the stable ecosystem and promote the spread of antibiotic-resistant bacteria (ARB) and antibiotic resistance genes (ARGs) in the human gut, which may limit the treatment e ciency and resulting from the chronic and relapsing infectious diseases [9][10][11]. And loop forever, host-associated fecal with antibiotic residues as well as ARB and ARGs would contaminate the environment [12][13][14].
Amoxicillin (AMX) is a broad-spectrum oral penicillin-type beta-lactam antibiotic, which kills bacteria by interfering with the synthesis of bacterial cell wall peptidoglycan layers [15,16]. Besides, polymyxin E (POL) is known as a colistin drug, a kind of cationic peptides that binds to anionic lipopolysaccharide of the gram-negative bacterial cell membrane and displaces calcium and magnesium ions from the outer cell, which further leads to change of cell membrane permeability and ultimately cell death [17]. Over the last several years, the effects of AMX or POL on the human intestinal microbiota have been extensively studied [18][19][20][21][22][23][24]. However, only a few research papers have reported their combination effects on intestinal microbial ecosystem by animal models, which would not re ect the actual in uences of antibiotics on human gut microbiota [25][26][27]. Furthermore, these in vivo studies didn't reveal the direct effects of antibiotics on intestinal microbiota because of the disturbances of neurohumoral regulation, the individual differences, dietary habits, and the physiological status. The simulator of the human intestinal microbial ecosystem (SHIME) model is known to be a useful tool for in vitro studies as (i) interactions between the microbiota; and (ii) the effects of prebiotics and other compounds on the microbial communities and metabolic activities [28]. To the best of our knowledge, there are only a few researches that added the antibiotics into this in vitro model, which mainly focused on the bene t of the mucosal environment, high-ber diets, probiotic, and propionate-producing consortium in human intestinal microbiota [29][30][31][32]. Moreover, the combination effect of antibiotics cocktail on human intestinal microbiota dysbiosis and related health risk is not fully understood.
Therefore, this study, for the rst time, evaluated and compared the in vitro ability of AMX and POL (used separately or combined) on ARGs as well as human disease-related pathways in the simulated human gut. Considering the reasonable dosage of AMX and POL for adult human study are about 750 to 1500 mg/day and 1.5 -4 × 10 6 IU/day respectively, and only half volume of the adult gut can be simulated in used SHIME model, 600 mg/day of AMX and 1.5 × 10 6 IU/day of POL were used to examine its impacts on human gut microbiota and associated functional pathways [18,20,23,24]. In this study, the composition of human intestinal microbiota was analyzed by 16S rRNA gene high-throughput sequencing, the human disease-related pathways were predicted by functional predictions and the ARGs were quanti ed by high-throughput quantitative PCR (HT-qPCR). This study achieved a systematic investigation and precise understanding of the direct effects of AMX and POL on the intestinal microbiota, which may be valuable for directing future work.

ARGs
There was a total of three batch experiments conducted, and we collected six different group samples (groups A_Con, A_AMX, P_Con, P_POL, Comb_Con, and Comb_AP, each group contain six samples from two-time points and three colon reactors, shown in Fig. 1). In this study, a total of 37 targets ARGs were detected from six different group samples using a high-throughput-qPCR (HT-qPCR) technique. The heatmap showed that the relative abundances of ARGs such as aminoglycoside, beta-lactam, multidrug, and tetracycline resistance genes were noticeably higher in the group P_POL than the group P_Con, while the difference were not comparable between groups Comb_Con and Comb_AP (Fig. 2a). For instance, log relative abundance increment of aac6ib (aminoglycoside) was about 1.7 log units after POL treatment than the control (P_POLA1 vs. P_ConA2), while bl2b_ampc (beta_lactam), mexf (multidrug), and tetra (tetracycline) were about 2.0, 2.4, and 1.9 log units, respectively. However, the increments of log relative abundance of the above mentioned ARGs were just about 0-0.9 log units after AMX treatment than in control (A_AMXA1 vs. A_ConA2), and 0-0.5 log units for combination treatment (Comb_APA1 vs. Comb_ConA2). Additionally, during PCA analysis, the component one explained 29.3% of the variance, and the component two exhibited 15.2% of the variance (Fig. 2b). These data clearly displayed the differences in ARGs abundances of bacteria community between the POL treatment and the control group (P_POL vs. P_Con), while a combination of antibiotics treatment attenuated the discrepancy (Comb_AP vs. Comb_Con).
Human disease-related pathways The metagenomics study of the 16S rRNA gene sequence by PICRUSt revealed the gene numbers of human disease-related functional pathways in the bacterial communities of the six groups, and the genes were presented in the heatmap (Fig. 3a). More importantly, the heatmap showed the gene numbers of human disease-related pathways, including cancer, drug resistance, endocrine and metabolic diseases, infectious diseases, and neurosurgery diseases, which were more abundant in AMX treatment group than that in control group, and the distinction was especially more apparent between groups Comb_Con and Comb_AP. For instance, the gene numbers of colorectal cancer, viral myocarditis, hepatitis B, and toxoplasmosis were about 5 times enriched after AMX treatment than in the control (A_AMXA2 vs. A_ConA2). Moreover, the gene numbers of these human disease-related pathways were about 8 times enriched after POL treatment than in control (P_POLA2 vs. P_ConA2), while gene numbers were about 51 times enriched for combination treatment (Comb_APA2 vs. Comb_ConA2). The rst (PC1) and second principal component (PC2) explained 64.8% and 10.7% of the total variance, respectively (Fig. 3b), and the distinguished feature indicated that the combination of antibiotics treatment had the greatest impact (Comb_AP vs Comb_Con) on the human disease-related pathways of the bacterial community.
Besides, the beta diversity of the microbiota communities and weighted UniFrac distance was affected by antibiotics treatment. As shown in Fig. 3d, three control groups were clustered together in PCA analyses. The beta diversity results showed that all the samples collected after a single antibiotic or combination treatment differed from the control group. Meanwhile, weighted UniFrac distances between a single antibiotic treatment group and the control group (A_AMX vs. A_Con, P_POL vs. P_Con) were all signi cantly higher (P < 0.001, T-test) than that within the control group (A_Con vs. A_Con, P_Con vs. P_Con), and that between the combination group Comb_AP and the control group Comb_Con were slightly higher (P < 0.01, T-test) than that within the control (Fig. S4).

Correlations between microbial taxa and human disease-related pathways or ARGs
The network analysis of co-occurrence patterns between the microbial taxa and the ARG subtypes is shown in Fig. 5a. It was seen that Bi dobacterium (signi cantly enriched bacteria after POL treatment) was positively associated with aminoglycoside, macrolide-lincosamide-streptogramin B (MLSB), multidrug, and tetracycline resistance genes. For example, the correlation coe cient of Bi dobacterium with qacedelta1 and tetw was about 0.8 (P < 0.001), and the correlation coe cient with aac3iia and mpha was about 0.7 (P < 0.001). Pseudomonas, another signi cantly increased bacterial genus in either POL treatment or combination treatment group, was also positively associated with several ARGs. For example, the strong correlations were found in Pseudomonas with mexf and bl1_ampc (about 0.8, P < 0.001), and with tetpa and aac6ib (about 0.7, P < 0.001). Fig. 5b shows the results of co-occurrence patterns between the microbial taxa and human diseaserelated pathways. A very similar pattern of results was observed that signi cantly increased bacteria Enterobacter (after in either POL treatment or combination treatment group) was positively associated with almost all of those human disease-related pathways, such as cancer, drug resistance, endocrine, and metabolic diseases, infectious diseases, and neurosurgery diseases. Speci cally, the correlation coe cient of Enterobacter with bladder cancer, pertussis, and Salmonella infection was about 0.95 (P < 0.001) and the correlation coe cient with Prion diseases and cationic antimicrobial peptide (CAMP) resistance was 0.94 (P < 0.001).

Discussion
After two weeks of SHIME operation, the microbiota's composition and diversity were stabilized, and most predominant phyla in the gut microbiome were Bacteriodetes, Proteobacteria, Synergistetes, and Firmicutes, which were also previously demonstrated by Yu's group [33].{Yu, 2016 #1;Li, 2015 #3} These predominant phyla had also been observed in the fecal samples of humans and animals [34,35]. As mentioned above, a large body of data showed that stabilized microbiota might be disturbed by antibiotics exposure, resulting in chronic and relapsing infections [9][10][11]. The effects of AMX or POL on the human intestinal microbiota have been extensively studied [18][19][20][21][22][23][24]. However, the combination effect of antibiotics cocktail on human intestinal microbiota dysbiosis and related health risk is not fully understood. Hence, this study for the rst time evaluated and compared the in vitro ability of AMX and POL (used separately or combined) on ARGs as well as human disease-related pathways in the simulated human gut. As the bacteria-killing mechanism for AMX is interfering synthesis of the bacterial cell wall peptidoglycan [15,16], whereas POL is binding to the anionic lipopolysaccharide of the gram-negative bacterial cell membrane [17]. Therefore, it is supposed that the combination of AMX and POL can cause a synergistic effect on human intestinal microbiota.
Combination treatment reduced occurrence of drug resistance development As mentioned earlier, exposure to antibiotics may promote the spread of ARB and ARGs in the human gut, which may limit the treatment e ciency and resulting in chronic and relapsing infections [9][10][11]. For instance, ARGs from such classes, including aminoglycosides, beta-lactams, and tetracycline have been con rmed to enrich in human gut microbiota following administration of AMX-clavulanic acid for one week [36], while patients who have taken AMX lower the relative abundance of ARGs from 0.81% to 0.14% [37]. The controversial conclusions in their results may attribute to the subjects, antibiotics administration and, the statistical approach of ARGs varied widely between studies. The minor effect of AMX on ARGs found in our study was not totally in agreement with the results. Barraud and associates discovered that AMX exposure to women during labor signi cantly selected bla (TEM)-positive Enterobacteria in their gut microbiota as well as their children's, indicating ARB would cumulate across generation [38]. Similarly, POL was suggested to cause the selection of antimicrobial-resistant bacteria and the exchange of ARGs through horizontal gene transfer (HGT) between bacterial species [23]. Buelow and associates also con rmed that cefotaxime, POL, tobramycin, and amphotericin B combination treatment would select the antibiotic-resistant bacteria and aminoglycoside resistance genes such as aph(2")-Ib and the aadE-like gene [39]. Furthermore, the increment of relative log abundance of ARGs such as aac6ib (aminoglycoside) caused by POL treatment that showed in our research was also consistent with the previous results.
The extensive use and misuse of antibiotics are known as contributors to the development of antibiotic resistance, which is a recent threat to global health. Antibiotics combination therapy is one of the essential ways that can prevent the development of antimicrobial resistance [40][41][42]. For instance, the addition of an aminoglycoside to a beta-lactam therapy regimen has been suggested to have a bene cial effect in delaying or preventing the development of antimicrobial resistance [43]. Combination therapy also suppresses or minimizes the degree of resistance of daptomycin-resistant viridans group streptococci compared with daptomycin monotherapy [44]. This study, for the rst time, revealed that the combination of AMX and POL treatment reduced the occurrence of drug resistance development, which was consistent with the above ndings. However, the combination therapy with beta-lactam ceftazidime and the uoroquinolone cipro oxacin selected for mutants that displayed broad-spectrum resistance, leading to decreased susceptibility to the combination of drugs applied as well as to unrelated antibiotic classes [45]. Therefore, the increased risk of selecting for broad-spectrum resistance antibiotic combinations should also be considered before being implemented in the clinical settings.
Combination treatment promote the increase of the human disease-related genes Antibiotics have been widely used for the prevention and treatment of diseases to humans. For instance, AMX is effective in resistant pneumococcal and Helicobacter pylori eradication therapy [19,46]. Oral POL contributes to intestinal eradication of multidrug-resistant Enterobacteriaceae, ESBL-producing Escherichia coli, and Klebsiella pneumoniae in immunocompromised patients [23,24]. However, the facilitating clearance of targeted infections, antibiotic-induced disruption of the microbiota and immune homeostasis can lead to disease [47]. AMX had been found to promote the selection of human disease category genes, including drug resistance categories in intestinal microbiota, and the increased gene numbers of human disease-related functional pathways shown in our results were in agreement with this [48]. Some studies demonstrated that AMX increased the beta-lactam resistance and antibiotic-resistant bacteria in the gut [19,49]. Other papers reported AMX would activate innate immunity and cause infectious diseases such as urinary tract infections, diarrhea, and liver abscess [50][51][52][53]. AMX was also suggested to affect insulin sensitivity and increase succinate, monosaccharide, and oligosaccharide levels in the fecal samples, and prenatal exposure was also associated with the increase of childhood overweight risk [54][55][56]. AMX was even con rmed to increase the risk of colon cancer [57]. All the above might be explained by our ndings that the human disease-related pathways, including cancer, drug resistance, endocrine, and metabolic diseases, infectious diseases were more abundant in the AMX treatment group. Similarly, POL suggested the effect of immune responses and induce disease incidence, including insulin-dependent diabetes, diarrhea, bacterial infections, and translocation using animal models [58][59][60][61][62].
A combination of antibiotic therapy is frequently used to treat severe infections, which would not only prevent the development of antimicrobial resistance but also improve the treatment e cacy [40][41][42]. For instance, a study con rmed that combination of colistin-polymyxin B or tigecycline with a carbapenem therapy is superior to monotherapy for carbapenemase-producing Klebsiella pneumoniae [63]. Besides, the combination of AMX and POL therapy would effectively attenuate bacterial endotoxin-and Shiga exotoxin-mediated cytotoxicity and reduce mortality from peritonitis [25,64]. However, excessive use of combinations should be avoided because it might be associated with increased risk for the toxicity and super-infections [41]. Combination therapy was also reported to cause a rise in infections triggered by multidrug-resistant gram-negative organisms [65]. Another paper revealed that a combination of an aminoglycoside with beta-lactam is associated with an increased risk for nephrotoxicity [66]. This study, for the rst time, revealed that the combination of AMX and POL treatment promote the increase of the human disease-related genes such as colorectal cancer, viral myocarditis, hepatitis B, and toxoplasmosis. Therefore, these results might indicate that the side effects of combination therapy should be carefully considered.

Combination effects on the genetic level attribute to microbiota shift
The phenomena of microbiota shift shown in our research are identical to several in vivo studies on intestinal microbiota; however, Proteobacteria phylum such as Enterobacter and Bacteroides were most predominantly detected following AMX administration [52,67,68]. Furthermore, the percentage of Escherichia-Shigella was decreased after POL treatment found in this study was further supported by earlier reports that con rmed the eradication ability of POL on Escherichia and its cytotoxicity [23,64]. Some studies have also discovered a trend toward increasing taxonomic diversity after AMX treatment and a signi cant effect on the beta diversity pursued by POL exposure [23,69]. Additionally, this study, for the rst time, revealed that POL treatment caused a bloom of Pseudomonas and Bi dobacterium, while combination treatment (AMX and POL) caused the enrichment of Enterobacter and Pseudomonas. Bi dobacterium intrinsically carries ARGs such as ermx and tetw, and Pseudomonas possesses mexf, aada6, and blap1b, which make them inherently resistant to MLSB, tetracycline, aminoglycosides, and beta-lactam antibiotics, and consequently pose a signi cant therapeutic challenge [70][71][72][73]. All of the above ndings supported the positive correlations between the abundances of Bi dobacterium and Pseudomonas with ARGs in our study. Signi cant Bi dobacterium shift discovered in this study also explained POL exposure caused more severe drug resistance development than combination treatment. Although combination treatment reduced the occurrence of drug resistance development produced by single POL treatment. However, the bloomed multidrug-resistant Pseudomonas was still an immense threat to human health because of its high virulence factor reported in many previous research papers [74][75][76].
Besides, Enterobacter was con rmed to increase after both AMX administration and combine exposure of antibiotics, and the combination treatment caused its more substantial enrichment. Although these bacteria are considered commensal in the human gut, fecal carriage of this opportunistic pathogen con rmed to be a severe risk factor for the infection [77][78][79]. The signi cant increase in opportunistic pathogen associated with human disease-related pathways of human intestinal microbiota following combination of AMX and POL treatment exhibited in this study have not been reported in details yet. More importantly, this study displayed a positive correlation between the opportunistic pathogen Enterobacter and human disease-related pathways, which suggested the bloom of this pathogen caused by antibiotics treatment that might contribute to human diseases. For instance, Enterobacter was reported to be related to human infectious diseases such as colitis, neonatal sepsis, and bloodstream infections [80][81][82]. Furthermore, some studies have also indicated that Enterobacter was enriched in cancer patients and related to cardiovascular diseases, endocrine and metabolic diseases, and neurosurgery diseases [83][84][85][86][87]. All of the ndings, as mentioned above, were linked with positive correlations between the abundance of Enterobacter and human disease-related pathways. And this advocated that the enhanced effect of combination treatment on the increase of the pathways might attribute to apparent Enterobacter enrichment in our study. Therefore, the results of our research demonstrated a severe impact and a negative side-effect of combination antibiotics treatment that enriched pathogens related to health problems, which should be considered as a fundamental aspect of the cost-bene t equation for the antibiotics combination prescription.

Perspectives
The ndings in this study suggested several numbers of opportunities for additional study. First, expansion of the analysis to incorporate multiple-omics approaches of the metagenome, metatranscriptome, and metabolome were not taken in this study, which attributed to the costs associated with sequencing and analysis. With multiple-omics data, the con rmed genome composition and expression of the gut microbiota would help us understand how antibiotics induced human intestinal microbiota dysbiosis and related disease. It is also of interest to inoculate the microbiota mixture from several patients to discover the different effects of antibiotics between healthy and sick individuals. However, it is di cult to obtain such samples, and the SHIME operation cost is very high. Further in vivo studies aimed to verify whether the ndings in this in vitro study re ect the reality, which will provide new insights to measure how antibiotic affects the gut microbiota and the associated disease. Moreover, the combination antibiotics' exposure was con rmed to cause the synergistic effect on increasing the human disease-related genes, and their prescription should be considered as an essential aspect of the risk assessment.

Conclusions
In this study, a combination of POL and AMX was con rmed to reduce the occurrence of drug resistance in the gut microbiota caused by single antibiotic treatment. However, the results also revealed that combination treatment would promote the increase of the disease-related pathways compared with the separate use of AMX or POL. Combination effects on the genetic level might attribute to microbiota shift, which were explained well by the phenomenon that Bi dobacterium was positively associated with the ARGs and Enterobacter was positively related to the human disease-related pathways. These results might be valuable to direct the future work and opened up new perspectives to address the direct effects of combine antibiotics on the intestinal microbiota. These suggested side-effects should be considered for a combination of antibiotics prescription.

Antibiotics treatment and samples collection
In this study, the SHIME was constructed using ve double-jacketed reactors designated as the stomach, small intestine, ascending colon, transverse colon, and descending colon, respectively (Fig. 1). The last three reactors were inoculated with a mixture of fecal microbiota from one healthy adult volunteer, who did not suffer by any gastrointestinal disease or take antibiotics in the past six months, on account of the differences between individuals may be alleviated by same culture condition [88]. The details of the SHIME system and the startup process are summarized in the Supplementary material.
There was a total of three batch experiments conducted in which each one maintaining for two weeks of stabilization and afterward one week for the addition of nutritional medium (groups A_Con, P_Con, and Comb_Con, each group contain six samples from two-time points and three colon reactors). Then the SHIME was exposed individually by AMX (600 mg/day; group A_AMX), POL (1.5 x 10 6 IU/day; group P_POL), as well as combination of them (600 mg/day of AMX and 1.5 x 10 6 IU/day of POL; group Comb_AP) for one week in each batch experiment (Fig. 1). Each sample is a mixture of three samples collected at speci c time intervals in a day. The samples were stored at −80 °C for further analyses.

16S rRNA gene sequencing and analysis
Total DNA was extracted from the samples using the E.Z.N.A. stool DNA Kit (Omega, USA) according to the manufacturer's protocols. The V3-V4 region of the bacterial 16S rRNA gene was ampli ed by polymerase chain reaction (PCR). The raw reads of the sequences were deposited into the NCBI Sequence Read Archive (SRA) database under the accession number SRR11249954-11249989. The raw Illumina fastq les were de-multiplexed, quality-ltered, and analyzed using Quantitative Insights Into Microbial Ecology (QIIME) [89]. The 16S rRNA gene sequences were further taxonomically classi ed using the Ribosomal Database Project (RDP) classi er 2.0.1 [90].
The effects of antibiotics on alpha diversity, the taxon richness (Chao1 index), evenness (Simpson index), and diversity (Shannon index) were calculated for all the samples as previous did [91,92]. Linear discriminant analysis effect size (LEfSe) was performed to determine bacterial taxa that were signi cantly differed between the six groups using the Galaxy application tool [93]. Functional predictions of microbial community were performed to visualize the distribution of human disease-related pathways in the six groups using Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt) [94]. The accuracy of PICRUSt for the detection of more challenging functional groups was good (min. accuracy = 0.82), suggesting that their inference of gene abundance across various types of functions was reliable, and PICRUSt predictions had high agreement with metagenome sample abundances across all body sites (Spearman r = 0.82, P < 0.001). These analyses were conducted by BioMarker Technology Co., Ltd (Beijing, China). The details of taxonomical classi cation, LEfSe analysis, and functional predictions are described in the Supplementary material.

High-throughput quantitative PCR (HT-qPCR) and analysis
High-throughput-qPCR reactions were performed using the Wafergen SmartChip Real-time PCR system as previous did [91,92]. The reactions were conducted by Anhui MicroAnaly Gene Technologies Co., Ltd (Anhui, China). A total of 108 primer sets were chosen (Excel S1), which included 102 primer sets to target the almost all major classes of antibiotic resistance genes (ARGs) found in the microbiota of Chinese human gut [95], along with ve mobile genetic elements (MGEs) and one 16S rRNA gene. The results were analyzed using the SmartChip qPCR Software. Data with multiple melting peaks or ampli cation beyond the range (0) were excluded and then screened with conditions that a threshold cycle (CT) must be <31, and positive samples should have three replicates simultaneously. The details of HT-qPCR analysis are described in the Supplementary material.

Data analysis
All the results were expressed as mean values and standard deviations. The statistical analysis was performed with SPSS 17.0 software (SPSS Inc., Chicago, Ill., U.S.A.). The T-test was conducted to compare the differences between the groups, and all the statistical tests were two-tailed. The statistical signi cance was set at three different levels (*P < 0.05, **P < 0.01, and ***P < 0.001). Principal component analysis (PCA) of ARGs, human disease-related pathways, and OTUs were plotted by 'ggplot2', 'vegan', and 'corrplot' packages of R (version 3.5.1). Correlations between the microbiota and human disease-related pathways or ARGs were analyzed using the Spearman test in R with the 'vegan' package. The correlations between the pairs of variables were considered to be signi cant at r > 0.6, and P values were < 0.05. The Gephi (V 0.9.1) software was used to visualize the bipartite network graphs using the Force Atlas algorithm.

Additional Files
Additional le 1: Supplementary information le. Supplementary Methods. Fig. S1. Composition of microbial community at phylum level in groups A_Con and A_AMX (a), P_Con and P_POL (b), Comb_Con and Comb_AP (c). Fig. S2. LDA score and cladogram of LEfSe comparison analysis between the six groups. The colored shading depicts bacterial taxa that were signi cantly higher in each group, as indicated. Selection of discriminative taxa between six groups were based on an LDA score cutoff of 4.0, and differences in the relative abundances of taxa were statistically determined based on a Mann-Whitney test at a signi cance level of 0.05. Fig. S3. Effect of antibiotic treatment on gut microbiota alpha diversity of the six groups (***p < 0.001, **p < 0.01, *p < 0.05). The Chao1 index (a) was used to calculate the community richness, Simpson index (b) was used to calculate the community evenness, and Shannon index (c) was used to calculate the community diversity. Fig. S4. Weighted UniFrac distance within control groups (A_Con vs A_Con, P_Con vs P_Con, Comb_Con vs Comb_Con) and between treatment group and control group (A_AMX vs A_Con, P_POL vs P_Con, Comb_AP vs Comb_Con) (***p < 0.001, **p < 0.01, *p < 0.05).
Additional le 2: Excel S1 Primer sets of HT-qPCR reactions using Wafergen SmartChip Real-time PCR system. Figure 1 Schematic of designed SHIME model and sampling time points setup.

Figure 2
Heatmap (a) and PCA (b) analysis of antibiotic resistance genes (ARGs) in the six groups. Heatmap colors re ect relative abundance of ARGs from low (blue) to high (red). Heatmap (a) and PCA (b) analysis of human disease-related pathways in the six groups. Heatmap colors re ect gene numbers of human disease-related pathways from low (blue) to high (red). Page