This study was conducted using healthy people who visited the Hanyang University Hospital in South Korea. The study protocol was approved by the Institutional Review Board (HYUH IRB 2019–09-030) and written informed consent was obtained from all the participants. Written informed consent to participate in this study was provided by the participants’ legal guardian/next of kin.

A total of 114 metagenome samples were used: 53 from healthy infants and children (1–13 years) and 61 from healthy adults (Figure 1). Sequencing data from healthy adults were obtained from our previous study (Kim et al., 2021). Healthy adults aged 30–59 years, with a Charlson comorbidity score of zero, who visited health screening services from June–October 2017 were sampled for this study (Kim et al., 2021). Faecal samples were collected from healthy infants and children aged <13 years enrolled in the national immunization program between October–December 2019. Healthy infants and children were defined as those without any medical conditions other than allergic disease. A questionnaire about personal information was used to collect the data for infants and children. Personal information included delivery mode, birth weight, gestational age, and breastfeeding duration. The questionnaire also contained the medical history of the year prior, such as hospitalization, frequency of hospital visits, and use of antibiotics. About 30–50 g of faecal sample was collected from each individual into a sterile container and stored frozen at −80°C prior to extraction of total DNA content.

Total DNA was extracted using the Fast DNA SPIN Kit for Faeces (MP Biomedicals, #116570200, CA, USA), following the manufacturer’s instructions. The Illumina HiSeqX Platform (Illumina, San Diego, CA, USA) was used to sequence the DNA samples. For every sample, 151 bp paired-end sequences were generated from inserts of 350 bp. Low-quality reads were removed using Sickle (Joshi and Fass, 2011), and reads containing ‘N’ were also removed. Finally, host contamination was removed by discarding reads mapped to the human genome, provided by the National Center for Biotechnology Information (NCBI).

The program MetaPhlAn (version 2.0; Truong et al., 2015) was used to classify microbial community composition using the default option. In the MetaPhlAn process, bowtie2 (Langmead and Salzberg, 2012) was used with the “very sensitive” option, and only bacterial organisms were profiled. The abundance of antibiotic resistance genes was measured as reads per kilobase million (RPKM). Sample reads were aligned to representative sequences using bowtie2 with the sensitive-local option. The representative sequences from CARD version 2.0 (McArthur et al., 2013) were 848 ARG sequences created after clustering using cd-hit (−c 0.9 -n 8). The aligned reads remained when the aligned length was longer than 50% of the ARGs and their similarity was more than 90% matched. RPKM was considered to exist when the aligned length covered >70% of the reference length.

Clustering was performed according to the Dirichlet multinomial mixtures (DMM) in R package (version 4.2.0) to cluster the samples. DMM clustering was based on both genus and ARG abundance. For bacterial proportion, we included 48 genera (average proportion > 0.1% in adult or child group). The optimal number of clusters was calculated by Laplace approximation, which was selected as three and two based on the genus proportion and ARG RPKM, respectively.

Principal coordinate analysis (PCoA) was generated based on Bray–Curtis distances using the vegan package v2.5–7 in R. Alpha diversity was quantified by the Shannon index using genus proportion and ARG proportion. To measure the alpha diversity of the genus proportion, each sample was sampled as 27 M reads (4G bp size). General linear model (GLM) regression analyses were performed to validate the association between age, genus composition, and ARG composition. The PCoA and GLM regression results were compared with various labels such as age group and DMM clusters. The differential abundances of genera and ARG were determined using t-tests and ANOVA tests. Permutational multivariate analysis of variance (PERMANOVA) was also performed to determine the differences between each group for each genus and ARG, using the vegan package in R.

A total of 114 participants (53 children and 61 healthy adults) were analyzed. The children were further divided into infants (≤ 1 year, n = 11), toddlers (1–<4 years, n = 13), and school-aged children (4–13 years, n = 29) according to their growth stages. The age distribution of the enrolled subjects is featured in Figure 1. Clinical factors, such as delivery mode, birth weight, gestational age, and breastfeeding months for infants and toddlers, are described in Table 1. In addition, medical history, such as hospitalization, frequency of hospital visits, and use of antibiotics in the year prior to sample collection, were surveyed (data not shown). There were no statistical differences in the variables among the three groups clustered by bacterial proportion (Table 1).

The gut microbiomes of healthy adults (n = 61) and children (n = 53) were compared. Genus-level composition was quite different between adults and children (Figure 2A). The three most abundant genera in the adult group were Bifidobacterium (12.4%; median proportion), Ruminococcus (7.83%), and Eubacterium (7.33%), whereas the children comprised mostly of Bifidobacterium (24.1%), Faecalibacterium (4.92%), and Bacteroides (3.19%; Figure 2A). The samples from the child group showed two distinctive patterns: Bifidobacterium-dominant and Bacteroides-dominant. Bifidobacterium was the most abundant genus in one group of children and Bacteroides was the most abundant genus in the other group of children. To identify clusters of similar bacterial proportions, Dirichlet multinomial mixture (DMM) clustering was conducted based on the bacterial proportion. Three clusters were identified: cluster G1 was dominant in the adult samples (83.3%), and clusters G2 and G3 were dominant in child samples (85.2% and 90.5%, respectively; Figure 2A; Table 1). G1 was not found in any infants, despite their presence in toddlers and school-aged children.

PCoA was further conducted using bacterial proportion, which was labeled by the age groups and the DMM clusters of bacterial and ARG composition (Figures 3A–C). For bacterial proportion, the adult and child groups were well separated (PERMANOVA test, p < 0.001). In the child group, infants, toddlers, and school-aged children showed similar patterns (p > 0.1; Figure 3A). When the PCoA result was labeled by DMM clusters of bacterial proportion, two distinctive patterns in children were presented as G2 and G3 clusters (Figure 3B). The two most dominant genera in the G1 group were Bifidobacterium (12.84%, median proportion; p < 1.0e−8, ANOVA test) and Ruminococcus (7.87%; p < 0.005); in the G2 group were Bifidobacterium (47.27%; p < 1.0e⁻⁸) and Escherichia (7.99%; p < 1.0e⁻⁹); and in the G3 group were Bacteroides (18.12%; p < 1.0e⁻⁹) and Faecalibacterium (12.19%; p < 1.0e⁻⁸) (Figure 3B). Statistical analysis revealed significant differences in the proportions of these dominant genera among the three groups.

While the abundance of ARG families was measured by RPKM, the patterns of ARG proportion were compared between the age groups and clusters that were built on bacterial proportion. When DMM clustering was conducted using the ARG composition, two clusters were identified (clusters A1 and A2 in Figure 2). A1 included most adults (n = 56, 91.8% of adult samples; n = 29, 54.7% of child samples), whereas A2 included predominantly children (n = 24, 45.3% of child samples; n = 5, 8.2% of adult samples). Notably, an association was observed between the clusters of bacterial proportion (G1, G2, and G3) and ARG composition (A1 and A2). Most of the samples in G1 (95.5%) belonged to A1, and most of the samples in G2 (81.5%) belonged to A2 (Figure 2). In addition, the samples in G3, which were mostly in children, corresponded to A1. Similar clustering patterns were also observed in the PCoA (Figures 3D–F). No significant differences were observed among infants, toddlers, and school-aged children (PERMANOVA test, p = 0.254; Figure 3D), which was similar to the bacterial proportion (Figure 3A). However, the proportion of school-aged children (62.1%) tended to be higher in the adult-dominant A1 group than in the other child groups (45.5% and 46.2% for infants and toddlers, respectively). The dominant ARGs in A1 were ErmB, tetW, and tetO (p < 0.05, p < 0.001, p < 0.01 for A1 vs. A2, respectively), and in the A2 were pmr transferase and ileS (p < 0.001, p = 0 for A1 vs. A2, respectively; Figure 3F).

When the ARG composition was compared with the clusters based on the bacterial proportion, PCoA on ARG abundance labeled by bacterial clusters showed that the G1 and G3 groups were relatively close (Figure 3E) and mostly corresponded to A1 (Figure 3E,F). Among the children, the samples included in G2 showed a higher abundance of Bifidobacterium intrinsic ileS and pmr transferase compared to the samples in G3 (Figure 3E). These two genes, ileS and pmr transferase, are frequently found in Bifidobacterium and Escherichia, respectively (Lee et al., 2004; Serafini et al., 2011), which were the most abundant genera in G2. In contrast, the samples in G3 showed a higher abundance of ErmB, tetW, tetO, tetQ, and CfxA, which are commonly identified in anaerobic gut flora (Figure 3E)¹ (Parker and Smith, 1993; Scott et al., 2000; Melville et al., 2001).

Bacterial and ARG compositions were compared to determine their association and to examine how microbial and ARG compositions changed with age. Compared to the child samples, bacterial diversity was higher in the adult samples (Figure 4A). However, there was almost no difference among infants, toddlers, and school-aged children (median Shannon index: 1.97, 1.95, and 1.92, respectively; Figure 4A). Accordingly, the adult-dominant cluster G1 had the highest genus diversity, followed by G3 and G2 (2.48, 2. 08, and 1.41 for G1, G3, and G2, respectively; median Shannon index). In addition, alpha diversity was positively correlated with age (R² = 0.3; Figures 4C,D). However, as age increased, the ARG diversity did not change significantly (R² < 0.01; Figure 4E). ARG diversity in infants and adults was slightly higher than that of toddlers and school-aged children without statistical significance (2.36, 2.29, 2.15, and 2.23 in infants, adults, toddlers, and school-aged children; median Shannon index) (Figure 4A). Among the DMM clusters of bacterial proportion, the G2 group showed both the least diversity of ARGs and the most abundant ARGs (Figure 4C). Notably, as ARG diversity increased, bacterial diversity increased (R² = 0.12) (Figure 4B). ARG diversity was not correlated with ARG abundance (R² < 0.01).

Upon comparison of the proportions of specific genera and ARGs to age, we found that the Eubacterium and Ruminococcus, which were the dominant genera in G1 (adult-dominant cluster), showed positive correlations (R² = 0.04 and 0.09, respectively) with age (Figure 5). However, Bifidobacterium, Bacteroides, and Escherichia, which were the major genera in G2 or G3 (child-dominant clusters) (p < 0.05 between adults and children), showed a negative correlation (R² = 0.11, 0.06, and 0.07, respectively) with age (Figure 5). Notably, Bifidobacterium showed the strongest correlation (R² = 0.11) with age.

Overall, nine ARG families showed significant differences among age groups or clusters based on bacterial proportion (p < 0.001 in any pair of groups; Figure 6). Furthermore, three tetracycline resistance genes, tet32, tetO, and tetW, were positively correlated (R² = 0.15, 0,09, and 0.09, respectively) with age (Figure 6). However, six other genes, ampC, TEM, ileS, bacA, pmr transferase, and cepA, showed a negative correlation (R² = 0.08, 0.07, 0.06, 0.09, 0.08, and 0.1, respectively) with age (Figure 6).