Individuals who were enrolled at the Ewha Special Education Research Institute (Seoul, Republic of Korea) or Ewha Womans University MokDong Hospital (Seoul, Republic of Korea) were diagnosed according to the DSM-5 diagnostic criteria by a child and adolescent psychiatrist followed by characterization using the Korean Childhood Autism Rating Scale (K-CARS) as described previously [32]. The K-CARS is a well-established scale for the diagnosis of ASD with good agreement with the DSM-5 diagnostic criteria [33]. This questionnaire contained 15 items, each with 4 symptom scales, and all individual scores on each of the questions were summed to obtain the total score. When the total score was higher than 30 points, the subject was classified as autistic. Individuals who had any associated additional psychiatric and neurological diagnoses, or individuals who were on any antipsychotic medications were excluded from the present study.

Among the characterized ASD individuals, 18 male and 2 female ASD individuals (22.4+/−4.9 years) (Table 1) were joined to this study and their urine was collected during the day. The collected urine samples were frozen and stored at −20℃ until use. Age-matched normal healthy subjects (24 males and 4 females, 21.1+/−9.5 years) (Table 1) were selected from the Inje University Haeundae Paik Hospital (IRB No. 1297992-2015-064) and Seoul National University Hospital Healthcare System Gangnam Center (IRB No. 1502-034-647). The control subjects were not related to ASD and had no clinical findings suggestive of gastrointestinal problems or neuropsychiatric disorders. The control subjects of this study had not taken antibiotics, probiotics or prebiotics in the 3 months prior to the sample collection.

The experimental protocol of human subjects was reviewed and approved by the Institutional Review Board of Ewha Womans University Hospital (IRB No. 2015-08-005-002). All eligible participants had been told about the purpose, procedures, risks and benefits of the present study and informed consent was obtained from all ASD subjects.

Bacteria EVs were isolated from the urine of ASD individuals following the procedure described previously [25,26]. Briefly, each urine sample was centrifuged at 10,000 × g for 10 min at 4℃. The supernatant was taken and passed through a 0.22-µm membrane filter to eliminate foreign particles. Isolated EVs were dissolved in 100 µl PBS, and quantified on the basis of protein.

Bacterial DNA extraction from prepared EVs was performed as described previously [25,26]. Briefly, isolated EVs (1 µg by protein, each sample) were boiled at 100℃ for 40 min, centrifuged at 13,000 g for 30 min, and the supernatants were collected. Collected samples were then subjected to bacterial DNA extraction using a DNA extraction kit (PowerSoil DNA Isolation Kit, MO BIO, USA) following the manufacturer's instructions, Isolated DNA was quantified by using the QIAxpert system (QIAGEN, Germany).

Prepared bacterial DNA was used for PCR amplification of the V3-V4 hypervariable regions of the 16S ribosomal RNA genes using the primer set of 16S_V3_F (5′-TCGTCGGCAGCGTCAGATGTGTATAAGAGACAGCCTACGGGNGGCWGCAG-3′) and 16S_V4_R (5′-GTCTCGTGGGCTCGGAGATGTGTATAAGAGACAGGACTACHVGGGTATCTAATCC-3′). The PCR products were used for the construction of 16S rDNA gene libraries following the MiSeq System guidelines (Illumina Inc., San Diego, CA, USA). The 16S rRNA gene libraries for each sample were quantified using QIAxpert (QIAGEN, Germany), pooled at the equimolar ratio, and used for pyrosequencing with the the MiSeq System (Illumina, USA) according to manufacturer's recommendations.

Obtained raw pyrosequencing reads were filtered on the basis of the barcode and primer sequences using MiSeq Control Software version 1.1.1 (Illumina, USA). Sequence reads were taxonomically assigned using the MDx-Pro ver.1 profiling program (MD Healthcare Inc., Seoul, Korea). Briefly, the quality of sequence reads was retained by controlling an average PHRED score higher than 20 and read length of more than 300 bp. Operational taxonomic units (OTUs) were clustered using CD-HIT sequence clustering algorithms and were assigned using UCLUST [34] and QIIME [35] on the basis of the GreenGenes 8.15.13 16S rRNA sequence database [36]. Based on the sequence similarities, taxonomic assignments were achieved at the following levels: genus, >94% similarity; family, >90% similarity; order, >85% similarity; class, >80% similarity; and phylum, >75% similarity. In cases where clustering was not possible at the genus level due to a lack of sequence information at the database or redundant sequences, the taxon was named based on the higher-level taxonomy with parentheses.

Data were normalized to have a mean of 0 and a standard deviation of 1 by linear normalization. PCA and two-dimensional scatter plots with axis of the first and second principal component were calculated and drawn using Matlab 2011a.

Two-sample comparisons were performed using Student's t-test. Data clustering of pyrosequencing reads were compared using the χ² test or t-test, while the comparisons between phylum compositions were tested by Fisher's exact test using GraphPad Prism 6 (San Diego, CA, USA). All data are presented as the mean±SEM, and a statistical difference was accepted at the 5% level.

Bacteria-derived EVs were isolated from the urine of 20 ASD individuals and 28 normal healthy subjects. The average age of the control and ASD subjects was 21.1+/−9.5 years and 22.4+/−4.9 years, respectively (Table 1). ASD subjects showed mild impairment of social interaction and stereotypies. The average K-CARS values of these ASD individuals was in the range between 31.5 and 33.5 and IQ values were in the range between 65 and 86. The control group was composed of healthy volunteers who had no medical problems including those related to ASD.

After the extraction of bacterial genomic DNA from the isolated EVs, variable regions of the 16S rRNA genes were amplified by PCR, and the libraries were constructed, as described previously [25,26]. Subsequent DNA sequencing analyses led us to identify over 2,000 operational taxonomic units (OTUs) for ASD and normal individuals. There was no significant difference in the alpha diversity between the two groups (Fig. 1A). Among the identified OTUs, we assigned 30 OTUs at the phylum level, 75 OTUs at the class level, 141 OTUs at the order level, 279 OTUs at the family level, and 619 OTUs at the genus level. Among these OTUs, we primarily focused on OTUs that occupied more than 0.1% of the identified taxons in the following analyses.

Sequence readings of EVs-based 16S rDNA indicated that the top five members of the phyla p_Proteobacteria, p_Firmicutes, p_Actinobacteria, p_Bacteroidetes, and p_Cyanobacteria comprised 95.2% of the identified OTUs in healthy subjects, whereas these members covered 90.65% of the total OTUs in ASD individuals, suggesting that ASD individuals have altered phyla composition. More specifically, the occupancy of p_Proteobacteria decreased from 49.12 to 35.30%, p_Cyanobacteria decreased from 4.36 to 1.92%, and p_Armatimonadetes decreased from 0.38 to 0.00% in ASD individuals (Fig. 1B~D). In contrast, the occupancy of p_Firmicutes increased from 24.96 to 33.07% and p_Verrucomicrobia increased from 0.58 to 2.37% in ASD.

The microbiota whose occupancy decreased or increased in ASD individuals were further analyzed at the class, order and family levels (Table 2, Supplemental Fig. S1). The decrease of f_Sphingomonadaceae and f_Rhizobiaceae accounted for the major decrease in p_Proteobacteria in ASD. In p_Cyanobacteria, o_Streptophyta decreased from 3.8 to 1.68%. In p_Verrucomicrobia, f_Verrucomicrobiaceae was the major increase (0.52 to 2.35%). In p_Firmicutes, f_Streptococcaceae, f_Clostridiaceae, an unclassified member of o_Clostridiales and f_Eubacteriaceae increased from 3.43 to 8.09% (Table 2).

The members of the genus occupied by more than 0.1% in either control or ASD individuals are summarized in Table 3 and Supplemental Table 1. Overall, 14 members at the genus level were down-regulated in ASD and their total occupancy in ASD dropped from 34.77 to 14.06%. On the contrary, 17 genus members were up-regulated in ASD and their total occupancy in ASD increased from 6.47 to 22.58%.

More specifically, an unclassified member of f_Enterobacteriaceae decreased from 8.85 to 6.08%, g_Pseudomonas decreased from 7.48 to 5.10%, g_Sphingomonas decreased from 4.17 to 0.71%, g_Agrobacterium decreased from 3.83 to 0.11%, an unclassified member of o_Streptophyta decreased from 3.80 to 1.68%, g_Achromobacter decreased from 2.42 to 0.05%, g_Roseateles decreased from 1.12 to 0.02%, and an unclassified member of f_mitochondria decreased from 0.84 to 0.10% (Table 3, Fig. 2).

On the other hand, g_Streptococcus increased from 1.58 to 4.77%, an unclassified member of o_Clostridiales increased from 1.06 to 1.87%, an unclassified member of f_Comamonadaceae increased from 0.92 to 3.79%, an unclassified member of f_S24-7 increased from 0.84 to 2.02%, g_Akkermansia increased from 0.52 to 2.35%, g_Rhodococcus increased from 0.40% to 1.56%, and g_Halomonas increased from 0.12 to 1.72% (Table 3, Fig. 2).

In the present study, we demonstrated that bacteria-derived EVs in urine were useful for the rapid assessment of microbiota profiles in ASD. The metagenome analysis of urine EVs indicated that p_Verrucomicrobia (0.58 to 2.37%, p=0.02) and p_Firmicutes (24.96 to 33.07%, p=0.03) increased in ASD, whereas p_Cyanobacteria (4.36 to 1.92%, p=0.01) and p_Proteobacterium (49.12 to 35.3%, p=0.01) decreased. There was no significant change in p_Bacteroidetes (5.85 to 8.62%, p=0.11) and p_Actinobacterium (10.91 to 11.74%, p=0.56). The altered microbiota compositions identified from urine EVs of ASD were partially consistent with microbiota compositions assessed from fecal samples reported in recent studies. The analyses of fecal microbiota compositions in previous studies reported that p_Actinobacteria, p_Verrucomicrobia and p_Cyanobacteria decreased or tended to decrease in ASD, but there were conflicting results for p_Firmicutes, p_Bacteroidetes and p_Proteobacteria (Table 4) [13,15,16,19].

In p_Firmicutes, g_Streptococcus (1.58 to 4.77%, p=0.02), g_Jeotgalicoccus (0.03 to 0.5%, p<0.01), g_Oscillospira (0.1 to 0.47%, p=0.02), and an unclassified member of f_Clostridiaceae (0.27 to 0.7%, p=0.04) significantly increased in ASD by more than two-fold. The genera g_Akkermansia occupied the greatest proportion of the increased phylum p_Verrucomicrobia (0.52 to 2.35%, p=0.02), whereas in p_Proteobacterium, 12 genera decreased and 5 genera increased in ASD. The genera g_Pseudomonas, g_Citrobacter, an unclassified member of f_Enterobacteriaceae, g_Sphingomonas, g_Agrobacterium, unclassified members of f_Rhizobiaceae, an unclassified member of f_Bradyrhizobiaceae, and an unclassified member of f_mitochondria, g_Roseateles, g_Delftia and g_Achromobacter were decreased (total 30.58 to 12.37%), whereas g_Halomonas, g_Erwinia, g_Rhodobacter, g_Comamonas, and an unclassified member of f_Comamonadaceae were increased (total 1.39 to 6.74%).

At the genus level, the present study identified g_Sphingomonas, g_Agrobacterium, an unclassified member of o_Streptophyta, g_Achromobacter, and g_Roseateles as being decreased in ASD by more than two-fold among the total OTUs having occupancy 0.1% or higher in either healthy control or ASD subjects, and g_Streptococcus, an unclassified member of f_Comamonadaceae, an unclassified member of f_S24-7, g_Akkermansia, g_Rhodococcus, and g_Halomonas were increased in ASD. Particularly, g_Desulfovibrio increased in ASD (0.04 to 0.48%, p<0.01), g_Lactobacillus tended to increase (2.56 to 5.45%, p=0.08), g_Bifidobacterium tended to decrease (1.9 to 0.8%, p=0.06), and g_Turicibacter tended to decrease (0.19 to 0.02%, p=0.07) in ASD, whereas g_Enterococcus (0.4 to 0.78%, p=0.06), g_Enterobacter (0.03 to 0.37, p=0.08), and g_Clostridium (0.11 to 0.31%, p=0.09) tended to increase in ASD. The changes in these genus members are broadly consistent with the results of previous reports assessed for fecal microbiota compositions (Table 4) [11,13,17,18,21].

The EV levels of g_Oscillospira, unclassified members of f_Clostridiaceae and f_Eubacteriaceae, and an unclassified member of o_Clostridiales were increased in ASD subjects. Previous studies have reported that several species of c_Clostridia produced 4-ethylphenyl sulfate (4-EPS) and p-cresol, which were found at high concentrations in the urine of ASD children. Administration of 4-EPS in healthy mice produced myelination deficits in the prefrontal cortex and sociability defects [37,38,39]. It was also reported that c_Clostridia produced propionic acid (PPA) and its related short-chain fatty acids (SCFAs) as fermentation products, and PPA infusions in rats induced ASD-linked neurochemical and behavioral changes [40]. These results suggest that bacteria-derived metabolites induce neurochemical and structural changes and shape behavioural abnormalities.

Oral treatment with Bifidobacteria fragilis ameliorated ASD-related gastrointestinal deficits and associated behavioural abnormalities behavioral abnormalities in the poly (I:C)-injection model [41]. Bifidobacteria infantis attenuated pro-inflammatory immune responses and production of serotonergic precursor, tryptophan, and has potential anti-depressant properties [42,43]. Considering these results, ASD groups with decreased EV levels of g_Bifidobacterium might have benefits by probiotic treatment with g_Bifidobacterium.

The microbiota profile assessed from urine EVs might reflect a large part of the gut microbiota. Nonetheless, we do believe that the microbiota profile assessed from urine EVs is not likely a simple alternative for microbiota profile assessed from stool. Possible sources for metagenome analysis of bodily microbiota may include stool bacteria, stool EVs, gut (ex, stomach and/or specific regions of the small and large intestines) bacteria, respiratory exhale EVs, oral/nasal bacteria and EVs, urinary system bacteria and EVs, and blood EVs. Generally speaking, microbiota in stool represents the intestinal compartment, whereas microbiota in urine or blood reflects the whole body including the intestinal compartments, oral system, respiratory system, and urinary system. Nonetheless, among the body parts, the gut is the major source of bodily microbiota. It was reported that the metabolites of intestinal microbiota activities, including phenyllactate, p-cresol sulfate, concentrations, and serotonin in urine, plasma, and stool of mouse pups undernourished by timed separation from lactating dams, then resumed ad libitum nursing, were different from each other, although they had some correlations [44,45]. Similar to metabolite profiles of intestinal microbiota activities, available information suggests that metagenome analysis assessed from these sources might be closely related, but represent some distinct landscapes. For an example, metagenome analysis of bacteria and bacteria-derived EV in stool of inflammatory bowel disease model mice indicated that the EV composition in stool was more drastically altered compared to that of bacterial composition in stool [25]. Considering that bacteria-derived EV indicates the metabolically or pathologically activated microbiota [25,27], urine EV may be more representative of the host's microbiota activities than stool bacteria.

To the best of our knowledge, this is the first report characterizing microbiota in ASD individuals on the basis of urine EVs. Compared to blood and feces, urine is easily obtained in large volume and is readily available via a non-invasive method. Considering the general difficulties in repeated sampling microbiota sources from ASD individuals, particularly low functioning individuals with ASD or toddlers with ASD, using urine as a sample source would be a great advantage for rapid and repeated assessments of microbiota changes under varying physiological contexts compared to the use of blood and feces. Comparative analyses of EV profiles from urine, blood and stool of ASD individuals will be valuable. Also it will be worth to understand EV profiles of ASD with diverse factors including age, sex, familial history, genetics, and ethnics.

Overall, the present study assessed urine EVs from individuals with mildly autistic subjects. We believe that further systematic and unbiased analyses of male and female subjects with broad ASD spectrums are necessary. This study focused on young adult subjects. Considering that ASD should be diagnosed in young children as early as 1.5~3 yr of age, this analysis should be expanded to toddlers and infants.