Introduction

Feeding and eating difficulties (FEDs) include avoidant/restrictive food intake disorder (ARFID), anorexia nervosa (AN), rumination disorder (RD), binge eating disorder (BED) and bulimia nervosa (BN)1. In the 5th revision of the American Psychiatric Association’s Diagnostic and Statistical Manual of Mental Disorders (DSM-5)2, ARFID is characterized by low interest in food and eating, underweight and/or nutrition deficiency3, which leads to one or more of the following: significant weight loss (or failure to meet expected weight and height trajectories in children and adolescents); (2) nutritional deficiencies (such as iron deficiency anemia); (3) a dependence on nutritional supplements (i.e., oral or enteral formulas) to meet energy requirements without an underlying condition necessitating this; and (4) and/or significant interference with day-to-day functioning due to the inability to eat appropriately. Although there are no diagnostic age restrictions in ARFID, it is more commonly present in childhood and adolescence4.

The intestinal microecosystem is composed of trillions of microorganisms that maintain a dynamic physiological balance and promote host immunity, metabolism, energy balance, and neural development5, 6. The shaping and multiplication of the gut microbiome start at birth, and the modification of its composition depends on various factors, such as the atmosphere, genetics, diet and lifestyle7. The gut microbiota plays a vital role in digestion8 and correlates the gastrointestinal tract and the central nervous system by the gut–brain–microbiota axis9. Current studies show that the gut microbiota plays an important role in eating disorders10,11,12,13,14, however, most researches focus on anorexia nervosa15. Given the multitude of aspects, complexity and the limited data in ARFID, a multidisciplinary approach seems to be the best option, and researching the characteristics of gut microbes in children with ARFID, may help us obtain a better understanding and provide more help for the treatment of ARFID in the future.

Materials and methods

Patients

Between August 2022 and September 2022, 135 children with ARFID and healthy children from Guangzhou Women and Children Medical Center were recruited through posters and were divided into an ARFID group (ARFID, n = 102) and a healthy control group (HC, n = 33). Each participant was evaluated by our pediatricians who trained in diagnosing ARFID, and the ARFID module of the Eating Disorder Examination (child version)16 was used for the structured clinical interview.

Inclusion criteria

(1) Patients were 3–12 years old; (2) Patients were diagnosed with ARFID; and (3) the disease course was ≥ 3 months.

Exclusion criteria

(1) age < 3 years or > 12 years; (2) treatment with antibiotics, intestinal microbial preparations, or other immunological preparations in the previous month; (3) weight loss or growth retardation due to certain chronic underlying diseases such as chronic heart and lung disease, liver disease, rheumatic disease, kidney disease, or immunodeficiency; and (4) noncooperation with sampling regimens or parental refusal to participate.

Specimen collection

Fresh fecal samples were collected into 10 ml stool containers (Batch No: GC-1022, Bioland) and were immediately frozen and stored at − 80 °C until analysis.

DNA extraction and 16S rDNA amplicon sequencing

The SDS method was used to extract the total genomic DNA from the samples. DNA concentration and purity were monitored on 1% agarose gels. According to the concentration, DNA was diluted to 1 ng/μL with sterile water. The V3–V4 region of bacterial 16S rDNA was then amplified using the primers F: CCTAYGGGRBGCASCAG and R: GGACTACNNGGGTATCTAAT (Novogene Co., Ltd. Beijing, China). All PCR mixtures contained 15 μL of Phusion® High-Fidelity PCR Master Mix (New England Biolabs), each primer at 0.2 μM and 10 ng target DNA, and cycling conditions consisted of a first denaturation step at 98 °C for 1 min, followed by 30 cycles at 98 °C (10 s), 50 °C (30 s) and 72 °C (30 s) and a final 5 min extension at 72 °C. An equal volume of 1X loading buffer (containing SYBR Green) was mixed with the PCR products, and electrophoresis was performed on a 2% agarose gel for DNA detection. The PCR products were mixed in equal proportions, and then a Qiagen Gel Extraction Kit (Qiagen, Germany) was used to purify the mixed PCR products. Following the manufacturer’s recommendations, sequencing libraries were generated with the NEBNext® UltraTM IIDNA Library Prep Kit (Cat No. E7645). The library quality was evaluated on a Qubit@ 2.0 Fluorometer (Thermo Scientific) and Agilent Bioanalyzer 2100 system. Finally, the library was sequenced on an Illumina NovaSeq6000 platform, and 250 bp paired-end reads were generated.

Metagenome sequencing

We selected a subset of specimens after MicroPITA analysis for metagenome sequencing. In brief, a Covaris ultrasonic fragmentation instrument was used to randomly interrupt the 350 bp fragment, and the whole library was prepared by terminal repair, adding an A tail, adding a sequencing joint, purification and PCR amplification. After the completion of library construction, Qubit2.0 was used for preliminary quantification, diluting the library to 2 ng/µl, and then an Agilent 2100 was used to detect the insert size of the library. After the insert size met the expectation, q-PCR was used to accurately quantify the effective concentration of the library (effective concentration > 3 nM) to ensure the quality of the library. After qualified library inspection, Illumina PE150 sequencing was carried out by pooling different libraries according to the requirements of effective concentration and target on-machine data volume.

Bioinformatics and statistical data analyses

16S rDNA

Paired-end reads were assigned to samples based on their unique barcodes and were truncated by cutting off the barcodes and primer sequences. FLASH (Version 1.2.11) was used to merge paired-end reads17. Quality filtering of the raw tags was performed using fastp (Version 0.20.0) software to obtain high-quality clean tags. The clean tags were compared with the Silva database, using Vsearch (Version 2.15.0) to detect the chimera sequences, and then the chimera sequences were removed to obtain the effective tags18. For the effective tags obtained previously, denoising was performed with DADA2 in QIIME2 software (Version QIIME2-202006) to obtain initial amplicon sequence variants (ASVs), and then ASVs with abundances less than 5 were filtered out19. Alpha diversity was calculated from 4 indices in QIIME2, including the Observed_otus, Chao1, Shannon and Simpson indices. Beta diversity was calculated based on unweighted UniFrac distances in QIIME2. LEfSe software (Version 1.0) was used to perform LEfSe analysis.

Metagenome

Readfq was used for preprocessing raw data from the Illumina sequencing platform to obtain clean data for subsequent analysis. Clean data need to be BLASTed to the host database to filter out reads that may come from host origin20, 21. MEGAHIT software (v1.0.4-beta) was used for assembly analysis of clean data, and scaftigs without Ns were obtained by breaking the resulting scaffolds from the N junction22,23,24. MetaGeneMark (V3.05) was used to perform ORF prediction for scaftigs (≥ 500 bp) of each sample, and entries with a length less than 100 nt in the prediction results were filtered out. For the ORF prediction results, CD-HIT software (V4.5.8) was used to eliminate redundancy and obtain the nonredundant initial gene catalog. Clean data of each sample were aligned to the initial gene catalog by using Bowtie2 (Bowtie2.2.4) to calculate the number of reads of the genes on each sample alignment. Based on the abundance of each gene in the gene catalog in each sample, basic information statistics were obtained. The gene sequences were compared with the KEGG, CAZy and CARD functional databases to obtain functional information.

Statistical analysis

SPSS26.0 statistical software was used for data analysis. Measurement data with a normal distribution were represented by (\(\overline{x}\pm s\)), and a t test was used. Measurement data with a nonnormal distribution were represented by M (P25, P75), and the Mann‒Whitney U test was used. For all comparisons, the level of significance was set at 0.05.

Ethics approval

This study was performed in line with the principles of the Declaration of Helsinki. This study was approved by the Ethics Committee of the Guangzhou Women and Children Medical Center, (date 15/03/2021/No.201B01).

Consent to participate

Consent to participate Informed consent was obtained from all parents of the children.

Result

Patient demographics

Among the children, 102 were diagnosed with ARFID, and 33 were healthy. There were 50 males and 52 females in the ARFID group and 15 males and 18 females in the healthy control group (HC group). There was no significant difference in age (P = 0.352) or sex composition (P = 0.512). The main characteristics are summarized in Table 1.

Table 1 Patient demographics.
Full size table

16S rDNA sequencing results

After high-throughput sequencing of the fecal samples, we generated 2,676,787,184 valid sequences in total. The average effective length was 415.95 bp, and the average number of sequences was 24,785,066.52.

Composition and diversity of the microbiome in the ARFID group and HC group determined by 16S rDNA sequencing

As shown in Fig. 1, the Goods coverage tended to be relatively stable, and the curve for the rank showed a downward trend with gentle leveling, which proves that the number of samples in this study was basically sufficient, and the sequencing depth could well reflect the complete microbiome composition of fecal samples from children with ARFID.

Figure 1
figure 1

Microbiome composition of the fecal samples. Samples Y.S.1–Y.S.102 were in the ARFID group, and samples K.B.1–K.17 were in the HC group.

Full size image

The Chao1 index values of the HC group were higher than those of the ARFID group, and the differences were statistically significant (P < 0.001) (Fig. 2a). Additionally, the Simpson index values (Fig. 2b) and the Shannon index values (Fig. 2c) of the ARFID group were higher than those of the HC group, and the differences were statistically significant (P = 0.009 and P = 0.02).

Figure 2
figure 2

Comparison of α-diversity between the ARFID group and the HC group. (a) The Chao1 index values of the HC group were higher than those of the ARFID group; (b) the Simpson index values of the ARFID group were higher than those of the HC group; (c) the Shannon index values were significantly higher in the ARFID group than in the HC group.

Full size image

We found two distinct groups of microbes using unweighted UniFrac distance-based principal coordinates analysis (PCoA), which showed that the microbiome of the ARFID group was different from that of the HC group (Fig. 3).

Figure 3
figure 3

Each point in the figure represents a sample, and the points with the same color come from the same group. The closer the two points are, the smaller the difference in community composition is. Principal coordinate analysis (PCoA) plots of individual fecal microbiotas based on unweighted UniFrac distances.

Full size image

At the phylum level, Bacillota were the most abundant bacteria in the ARFID group, followed by Bacteroidota, Proteobacteria and Actinobacteriota, which accounted for 45.7%, 41.2%, 6.8% and 5.4%, respectively. Similar to the ARFID group, Bacillota, Bacteroidota, Proteobacteria and Actinobacteriota were the most abundant bacteria in the HC group, accounting for 48.9%, 34.2%, 6.8% and 8.4%, respectively, but there were no statistically significant differences between the abundance of these microbial taxa in the two groups. At the genus level, the top 4 taxa in the ARFID group and HC group were Bacteroides, Faecalibacterium, Blautia and Bifidobacterium (Fig. 4). The abundance of Bacteroides was higher in the ARFID group than in the HC group, but the difference was not statistically significant (P > 0.05).

Figure 4
figure 4

(a) The top 10 microbes at the phylum level. (b) The top 10 microbes at the genus level. The vertical axis is the relative abundance of each species. Columns with different colors correspond to different species, and the height of the column represents the abundance of the species.

Full size image

To further explore all alterations in the gut microbiotas of the ARFID group and the HC group, we used LEfSe analysis (LDA effect size) to identify the key taxa responsible for the differences in the compositions of the fecal microbiotas between the two groups. The abundance of 5 taxa (the phylum Actinobacteriota and its corresponding class Actinobacteria, the order Bifidobacteriales and corresponding family Bifidobacteriaceae ,the genus Bifidobacterium), were enriched in the HC group. However, the order Enterobacterales and its corresponding family Enterobacteriaceae, the family Bacteroidaceae and corresponding genus Bacteroides, the species Bacteroides vulgatus were enriched in the ARFID group (Fig. 5).

Figure 5
figure 5

(a) Biomarkers associated with the ARFID group and HC group discovered by LEfSe analysis (logarithmic LDA score threshold = 4.0) in ARFID samples. (b) Cladogram representing the phylogenetic relationship of biomarkers associated with the ARFID group and HC group identified through linear discriminant effect size (LEfSe) analysis in samples.

Full size image

Composition and diversity of the microbiome in Y1 group and Y2 group determined by 16S rDNA sequencing

To further verify the difference in the gut microbiota of children with ARFID at different ages, we divided the fecal samples of 102 children with ARFID into two groups according to age. Among them, the Y1 group contained a total of 59 children aged between 3 and 6 years, including 25 males and 24 females, and the Y2 group contained a total of 43 children aged between 7 and 12 years, including 12 males and 21 females. As shown in Fig. 6, there were no statistically significant differences between the Y1 group and Y2 group in the Chao1 index, Shannon index and Simpson index (P = 0.1, P = 0.06, P = 0.06).

Figure 6
figure 6

Comparison of the Chao1 index (a), the Simpson index (b) and the Shannon index (c) between the Y1 group and Y2 group.

Full size image

PCoA showed that the structural differences between the Y1 group and Y2 group were not significant (Fig. 7).

Figure 7
figure 7

Principal coordinate analysis (PCoA) plots of individual fecal microbiotas based on weighted UniFrac (a) and unweighted UniFrac distances (b).

Full size image

Moreover, Anosim showed no significant differences in bacterial community structures between the two groups (R-value = − 0.014, P = 0.7) (Fig. 8), so we concluded that there was no apparent difference in gut microbiota diversity between the Y1 group and Y2 group.

Figure 8
figure 8

Anosim between the Y1 group and Y2 group. The y-coordinate is the distance, and the x-coordinate is the number of distances. R-values are between (− 1, 1), and R-values > 0 indicate differences between groups; R value < 0 indicates that the difference within the group is greater than the difference between the groups. P < 0.05 indicates statistical significance.

Full size image

At the phylum level, Bacillota, Bacteroidota, Proteobacteria and Actinobacteria were the most abundant bacterial taxa in the two groups, and there were no statistically significant differences in the abundance of these taxa (P = 0.958, P = 0.456, P = 0.473, P = 0.065). Notably, at the genus level, the abundance of Faecalibacterium was higher in the Y2 group than in the Y1 group, and the difference was statistically significant (P = 0.037) (Fig. 9).

Figure 9
figure 9

Differences between the Y1 group and Y2 group at the genus level as determined by T test analysis.

Full size image

Composition of the microbiome in the ARFID group and HC group as determined by metagenome sequencing

We selected 12 samples through Distinct mode in the microPITA analysis for metagenome sequencing. After filtering and assembly, 952,862 unigenes were obtained, and the total length was 709.59 Mbp, with an average length of 744.7 bp. The dilution curve of the core genes (Fig. 10a) and pan genes (Fig. 10b) showed that the samples we measured could cover the real genes in the tested samples and meet the requirements of subsequent analyses.

Figure 10
figure 10

Rarefaction curves of the core genes (a) and pan genes (b).

Full size image

As shown in Fig. 11, there were 459,055 genes in common between the two groups, and the ARFID group had more gene entries, with 215,800 genes.

Figure 11
figure 11

Venn diagram between the ARFID group and the HC group.

Full size image

The most abundant phyla in both groups were Bacillota, Bacteroidota, Pseudomonadota and Actinomycetota; the abundances of Bacillota and Actinomycetota in the HC group (45.1%, 14.3%) were higher than those in the ARFID group (37.6%, 6.3%), while the abundance of Bacteroidetes in the ARFID group (25.9%) was higher than that in the HC group (16.2%), although the difference was not statistically significant (P = 0.394) (Fig. 12a). At the genus level, different from the result of 16S rDNA sequencing, the abundance of Bacteroides in the ARFID group was much higher than that in the HC group (P = 0.041) (Fig. 12b); Escherichia coli, Streptococcus thermophilus and Lachnospira eligens were the most abundant taxa in the ARFID group at the species level, and Prevotella copri, Bifidobacterium pseudocatenulatum, and Ruminococcus gnavus were the top three microbial taxa in the HC group (Fig. 12c).

Figure 12
figure 12

The top 10 microbes at the phylum level (a), genus level (b) and species level (c). The vertical axis is the relative abundance of each species. Columns with different colors correspond to different species, and the height of the column represents the abundance of the species.

Full size image

Functional differences between the ARFID and HC groups

The KEGG database (Kyoto Encyclopedia of Genes and Genomes, Version: 2018.01) and the CAZy database (Carbohydrate-Active-enzymes Database, Version: 2018.01) are two commonly used databases. After blasting the filtered genes against the KEGG database, 6948 KOs were obtained. In KEGG Level 1, we found that a total of 44 pathways were related to the fecal microbiome, and the most abundant pathway was metabolism (Fig. 13).

Figure 13
figure 13

The number of genes in KEGG Level 1.

Full size image

In KEGG Level 2, carbohydrate metabolism, amino acid metabolism and environmental information processing: membrane transport were the most abundant pathways in the ARFID group and HC group (Table 2).

Table 2 The top 10 most abundant pathways in KEGG Level 2 (%).
Full size table

In KEGG Level 3, ko02010 (ATP-binding cassette transporter) and ko02020 (Two-component system) were the most abundant pathways in both groups (Table 3), but the differences were not statistically significant (P > 0.05).

Table 3 The top 10 most abundant pathways in KEGG Level 3 (%).
Full size table

Based on the CAZy database, 541 ECs were identified. Six major enzymes were found in the gut microbiota: glycoside hydrolase (GH), glycosyl transferases (GT), carbohydrate binding module (CBM), carbohydrate esterase (CE), polysaccharide lyase (PL) and auxiliary oxidoreductase (AA). Glycoside hydrolase was the most annotated gene in the two groups (Fig. 14).

Figure 14
figure 14

The number of CAZy annotations.

Full size image

As shown by the clustering heatmap, GT26 was significantly more abundant in the ARFID group, while GH51 and GH36 were more abundant in the HC group (Fig. 15).

Figure 15
figure 15

Cluster plot of the abundance of functions in Level 2 between groups based on the CAZy annotations.

Full size image

Based on the CARD database (v2.0.1), we found a total of 195 resistance genes. It can be seen from the species attribution diagram of resistance genes that Bacillota, with the highest content of resistance genes in the both groups, accounted for 38% of all drug-resistant bacteria in the ARFID group and 45% in the HC group (Fig. 16).

Figure 16
figure 16

Double-circle diagram of the relationship between two groups of drug resistance genes and species attribution. (a) is the ARFID group, (b) is the HC group, the inner circle is the species distribution of AROs, and the outer circle is the species distribution of all sample genes.

Full size image

The most abundant antibiotic resistance genes in the ARFID group were vanT (vancomycin resistance gene), followed by tetQ (tetracycline resistance gene), adeF (Acinetobacter baumannii resistance gene), and ermF (macrolide resistance gene), while vanT, tetO, vanW (vancomycin resistance gene) and vanY (vancomycin resistance gene) were the 4 most abundant antibiotic resistance genes in the HC group (Fig. 17). The abundance of ermF in the ARFID group was significantly higher than that in the HC group (P = 0.041).

Figure 17
figure 17

Abundance of antibiotic resistance (AR) genes in the ARFID group and HC group.

Full size image

Discussion

The human body carries a large number of gut microbes, which affect our metabolism, hormonal status, immune system and even behavior25, therefore, the human microbiome is becoming an important research topic in somatic and psychiatric diseases26, 27. Though research into ARFID is increasing28, none research on the gut microbiota of children with AFRID have been found so far. Recently, the most widely used high-throughput sequencing technique for bacterial identification has been 16S rDNA sequencing29, but bacteria can only be identified to the genus level, and specific information at the species level cannot be obtained. However, compared with 16S rDNA sequencing, metagenome sequencing can not only solve this problem but also obtain functional information on the microbiome. Based on 16S rDNA sequencing and metagenome sequencing, this study analyzed the composition of the gut microbiota in children with ARFID and healthy children and found that there were differences in the structure, diversity and functional information of the gut microbiome between the two groups.

In our study, the Chao1index (reflecting the microbe species richness), was higher in the healthy children than in the children with ARFID, and the difference was statistically significant (P < 0.001). And the Shannon index and Simpson index (reflecting microbial species diversity) in the ARFID group, were higher than that in the HC group, and the differences were statistically significant (P = 0.009 and P = 0.02). These results suggesting that intestinal flora community structure and diversity changed compared with those in healthy children and that intestinal flora disorders occurred in children with ARFID. Due to the different databases that used in the process of metagenome and 16S rDNA sequencing, as well as the different depth of sequencing, there will be differences in the annotation results. For example, the most abundant microbes in the two groups at the phylum level were Bacillota, Bacteroidota, Proteobacteria and Actinobacteriota through 16S rDNA sequencing, however, the most abundant phyla in both groups were Bacillota, Bacteroidota, Pseudomonadota and Actinomycetota at the phylum level through metagenome sequencing. Besides, At the genus level, different from the result of 16S rDNA sequencing, the abundance of Bacteroides in the ARFID group was much higher than that in the HC group (P = 0.041), in the result of metagenome sequencing. In order to further understand the differential abundance at the species level between the two groups, we conducted metagenomic analysis of some samples. However, there were no statistically significant differences between the abundance of these microbial taxa in the two groups at the species level. Based on the LEfSe analysis, the microbes that had a strong influence in the ARFID group were the order Enterobacterales and its corresponding family Enterobacteriaceae, the family Bacteroidaceae and corresponding genus Bacteroides, the species Bacteroides vulgatus, while the phylum Actinobacteriota and its corresponding class Actinobacteria, the order Bifidobacteriales and its corresponding family Bifidobacteriaceae, the genus Bifidobacterium influenced the HC group most.

Previous studies revealed that Bacteroides is a dominant genus in human intestines30. They can colonize the host intestines for a long time and have the function of regulating the intestinal microenvironment31, using carbon hydrate, enhancing the host’s adaptability to the environment32, 33 and secreting metabolites such as short-chain fatty acids (SCFAs)34, 35, thus establishing a stable symbiotic relationship with the host36. However, some reports have also shown that Bacteroides have adverse effects on the host37,38,39. Bacteroides vulgaris is widely considered a class of Bacteroides related to inflammatory bowel disease. This conjecture has been verified in a large number of in vivo and in vitro experiments40,41,42,43. In addition, some animal experiments have proven that Bacteroides vulgaris alleviates inflammation in mice44, 45. The family Enterobacteriaceae has had a great medical and public health impact on the global community, as these species are associated with a wide range of clinical syndromes and are major causative agents of foodborne enteritis46. So far, a large number of studies have focused on the epidemiology, pathogenesis, virulence, and/or antibiotic resistance of pathogenic strains of Enterobacteriaceae in humans47. Bifidobacterium, belonging to Actinobacteria, is a genus of gram-positive, pleomorphic, rod-shaped bacteria that are strictly anaerobic, and these bacteria have important immune regulation, anti-tumor, anti-pathogenic, anti-inflammation, anti-aging, and hypolipemic effects in humans48. Therefore, Bifidobacterium species have long been used as probiotics to alleviate various diseases by changing the gut microbiota composition and are significantly associated with human health49. The results of our study suggested that the increased content of Bacteroides or Enterobacteria or the reduced content of Bifidobacterium may be related to ARFID, but the pathogenesis still needs further study. As is known to all, a persistent failure to meet appropriate nutritional is one of the main characteristics of ARFID50, 51. Growing evidence indicates that malnutrition may bring about qualitative changes in the microbiome52 and specific strains of probiotics can potentially address the qualitative shift that occurs in the malnourished microbiome53. Previous study have shown that the depletion in gut Bifidobacterium represents the first step in gut microbiota alteration that associated with severe malnutrition54, suggesting that Bifidobacterium supplementation may be extremely important in the treatment of malnutrition. Although the best therapeutic intervention for eating disorders (including ARFID) is currently family-based intervention, in conjunction with both medical and dietetic monitoring and management, and pharmacologic management is never recommended as a first-line treatment, probiotic preparations may be useful as adjunctive interventions, similar to the previous researches, the live Bifidobacterium preparation may be a good choice through our study.

The composition and function of the gut microbiota have been largely overlooked in 3–6-year-old and 6–12-year-old children compared with infants and adolescents55, probably because such investigations have been constrained by ethical and practical considerations, such as difficulties in obtaining fecal samples from individuals in these age groups56. In our results, there seemed to be no difference between preschool children and primary school children with ARFID in terms of gut microbiota diversity. At the phylum level, the most abundant bacteria in the two groups were the same; however, at the genus level, the abundance of Faecalibacterium was significantly higher in primary school children than in preschool children with ARFID, while the abundance of Bifidobacterium and Ruminococcus gnavus were higher in preschool children with ARFID. Faecalibacterium is a genus of strictly anaerobic, extremely oxygen-sensitive (EOS), Gram-positive bacteria and is considered to be ubiquitous in the gastrointestinal tracts of healthy humans57. Numerous types of evidence suggest that Faecalibacterium plays an important role in immune system regulation, gut barrier protection, and microbiota modulation58. Notably, due to Faecalibacterium increased with age from newborns to adults and decreased again at later ages59, the Faecalibacterium was more abundant at the genus level in older children with ARFID in our study, which may related to age gain, however, more researches are needed to support this hypothesis.

Finally, according to the KEGG annotation results, there seemed to be no significant difference in gut microbiota function between children with ARFID and healthy children. However, GT26 was significantly enriched in children with ARFID compared with healthy children in Level 2 based on the CAZy database. It was reported that based on the structural relatedness of glycosyltransferase (GT) catalytic and carbohydrate-binding modules, GTs have been grouped into 115 families in the CAZy database, and only four distinct GT protein folds, termed GT-A through GT-D, have been reported thus far60. The architecture of GT-B enzymes consists of two β/α/β Rossmann-like domains, and the GT26 family was predicted to possess the GT-B fold61, but there have been no other further reports on GT26 to date.

The human gut microbiota is an important reservoir of antibiotic resistance genes (ARGs). Due to improper use of antibiotics, the gut microbiota will be disordered, and the presence of drug-resistant bacteria can affect the intestinal microecological environment, which may accelerate the development of disease. A recent report suggested that tetracycline, multidrug, and macrolide-lincosamide streptogramin (MLS) resistance genes were the top three most abundant ARG types in healthy individuals, and ermF was a representative ARG in the Chinese population62. Unlike findings in adults, our results suggested that the most abundant genes for gut microbiota resistance in healthy children were vancomycin, tetracycline and macrolide resistance genes. ErmF is a subtype of the MLS resistance gene type. In our study, we found that the ermF level was significantly higher in children with ARFID, which may be due to the higher abundance of Bacteroides. Since the use of macrolide antibiotics is the main reason for the increase in the abundance of Bacteroides63, we should be more cautious in the use of macrolide antibiotics in clinical practice. At present, we still know little about the impact of resistance genes on the gut microbiota in children, and there is not enough evidence to prove that resistance genes are directly related to the occurrence and development of ARFID, but researchers cannot ignore the potential threat of resistance genes to children’s health.

This was a preliminary study of the intestinal flora in children with ARFID, however, some limitations were identified. First, because of the high cost of metagenome sequencing, the total sample size was relatively limited. Moreover, due to the lack of healthy volunteers, the number of healthy controls was small. Finally, the sampling area was limited. To overcome these limitations, future studies should control related confounding factors and apply multiomics research strategies to microbial studies of ARFID, which will help reach more complete and in-depth conclusions.>