This intervention study was designed as a randomized, double-blinded, placebo-controlled trial. As part of the screening process, participants, who were recruited via posters and flyers, underwent an initial interview via telephone according to the inclusion and exclusion criteria, which were as follows. The study included male and female Korean adults who were aged 20–65 yrs, free of chronic diseases (e.g., cardiovascular disease, diabetes mellitus, cancer, hepatic disease, kidney disease, etc.) and other illnesses (e.g., insomnia, food allergy, anemia, and any infection in the preceding 3 months), not pregnant, and not taking any medication or dietary supplements (including probiotics) regularly.

Fig. 1 shows the number of participants who were enrolled, withdrew, and were followed up. The study procedures were scheduled as follows; recruitment and enrollment of participants (from November to December 2018), intervention (from January to April 2019), and assays of biomarkers and microbiome sequencing (from May to December 2019). A total of 48 participants were enrolled after the screening process and randomly and evenly assigned to control and intervention groups. Amongst them, 39 participants (18 and 21 participants in the control and intervention groups, respectively) signed an informed consent form, which was approved by the Kookmin University Institutional Review Board (approval number: KMU-201808-HR-184), and they were subsequently engaged in a 4-week intervention study. Finally, 17 (follow-up rate: 81%) and 13 (follow-up rate: 72%) participants from the intervention and control groups, respectively, completed the intervention and provided urine samples at baseline and endpoint. Reasons of dropouts included time and scheduling conflicts and inconvenience related to study site visits. According to an earlier study [7], a sample size of 20 participants (10 in each group) provides 90% power at a two–sided significance level of 5% for the non–parametric test, as calculated using the G*Power program (version 3.1.9.2; Heinrich-Heine-Universität, Düsseldorf Universität, Germany). Therefore, the sample size of our intervention study was assumed to yield statistically significant results.

γ-PGA (powder form; Vedan Enterprise Corporation, Taichung City, Taiwan) and vitamin B6 (tablet form; Junsei Chemical Co., Tokyo, Japan), which was ground and mixed with the γ-PGA powder, were purchased for the study. The daily dose of vitamin B6 was 100 mg, which corresponds to the tolerable upper intake for female and male adults according to the Dietary Reference Intakes for Koreans [14]. The daily dose of γ-PGA was 600 mg, which is below the level approved by the Ministry of Food and Drug Safety [15], although no studies have reported any toxic side effect of γ-PGA, rendering its use potentially safe. The daily consumption of cornstarch powder, which was selected as a placebo because of its similar color and texture and purchased at a local supermarket, was 700 mg, the same quantity as that of the supplementation dose.

Every participant received a calendar with 30 pockets, each of which contained an envelope with the daily amount of γ-PGA-vitamin B6 supplement or cornstarch powder, and they were instructed to take one envelope every night 1 h before going to sleep, as instructed in the prior study [6], which focused on the sleep-related outcomes.

A questionnaire-based survey was administered at baseline to acquire sociodemographic and behavioral information including smoking status, alcohol consumption, and physical activity data. Smoking and alcohol consumption statuses were categorized into “abstainers” and “current/former consumers.” Physical activity was classified into five categories based on its intensity, and metabolic equivalents were subsequently calculated.

Anthropometric measurements were conducted at baseline, with participants wearing light indoor clothes. Body weight and height were measured to the nearest 0.1 kg and 0.1 cm, respectively, and body mass index was calculated as the weight (kg) divided by the height (m) squared.

In both the baseline and endpoint examinations, clean-catch midstream urine (≥ 2 mL) was collected in the morning from participants who had fasted for ≥ 8 h, and sample tubes were subsequently stored in a deep freezer (−80°C). The frozen samples were transferred to a commercial laboratory for bacterial EV assays, which were conducted according to the laboratory’s standard protocol. The methods of isolating microbe-derived EVs, extracting DNA, sequencing bacterial 16S ribosomal DNA, and assigning taxonomy were similar to those reported in an earlier study [16].

Before statistical analysis, microbiome data were rarefied to an even depth of 1,069 reads per sample to eliminate the effect of variance in sequencing depth. Alpha-diversity (α-diversity) indices, including observed operational taxonomic units (OTUs) and the Chao1, Shannon, Simpson, and phylogenetic diversity (PD) indices, were calculated and compared between the two groups. These indices provide information on the richness and evenness of a microbial community. Observed OTUs as well as the Chao1, Shannon, and PD indices are used to measure richness, while the Shannon and Simpson indices are used to evaluate evenness. To compare the pairwise dissimilarity of the microbial communities observed in the baseline and endpoint data, beta-diversity (β-diversity) indices, such as the Bray-Curtis and Jaccard indices, were employed. In particular, the Bray-Curtis index indicates microbial abundance, while the Jaccard index provides the presence-absence information of the taxa.

Descriptive statistics were generated and presented as the mean ± standard deviation or proportions. Data comparison between the control and intervention groups was performed using non-parametric analysis methods, such as the Wilcoxon rank-sum test, Wilcoxon signed-rank test, and Fisher's exact test. Permutational multivariate analysis of variance (PERMANOVA) was conducted for β-diversity indices. This statistical method is based on the distance matrix and used to evaluate variability in microbial community composition between two conditions. Principal coordinate analysis (PCoA) of the distance matrix was performed to construct plots illustrating sample distribution; a close distance between symbols indicates the similarity of microbial communities.

Statistical analyses were performed using SAS software (SAS 9.4; SAS Institute, Cary, NC, USA) and R statistical software (v3.5.3; RStudio Inc., Boston, MA, USA). In all tests, a two-sided P-value less than 0.05 (P < 0.05) was considered statistically significant.

The baseline characteristics of the 39 participants who were included in the intention-to-treat analysis as well as those of the 30 participants who completed the intervention study and were included in the per-protocol analysis are summarized in Table 1. On comparing demographic variables and lifestyle factors between the two groups, no significant difference was observed in both analyses.

Table 2 displays the baseline and endpoint α-diversity indices. No significant difference between the baseline and endpoint data was observed in both the control and intervention groups; nevertheless, in the intervention group, all indices tended to increase from baseline, while in the control group, the observed OTUs and Simpson indices remained relatively constant, and the Chao1 index tended to decrease from baseline.

Table 3 presents the results of PERMANOVA, which evaluated the placebo and supplementation effects in the control and intervention groups, respectively, by comparing microbial community composition between the baseline and endpoint data. The supplementation effect was exclusively significant in the β-diversity analysis based on the Bray-Curtis and Jaccard indices (P < 0.05). These results indicate the dissimilarity in microbial community composition between the baseline and endpoint data of the intervention group.

Fig. 2 shows the PCoA plots, which illustrate the sample distribution of the microbial community between the baseline and endpoint data for each group. Baseline data symbols were more likely to be dispersed from those of the endpoint data in the intervention group.

Table 4 compares the mean changes in phylum-level microbial community composition from baseline between the control and intervention groups. The relative abundance values of four dominant phyla, namely, Actinobacteria, Bacteroidetes, Firmicutes, and Proteobacteria, and certain minor phyla, including Cyanobacteria, Deinococcus-Thermus, Saccharibacteria, Verrucomicrobia, and residual miscellaneous phyla, are listed therein. The mean composition changes were not significant within each group, except the Proteobacteria composition of the control group, as well as between the groups. However, the mean changes in the Firmicutes to Bacteroidetes ratio (FB ratio) were significantly different between the control and intervention groups (P < 0.05); the FB ratios decreased from baseline in the intervention group but increased in the control group.

Fig. 3 depicts the relative abundance of the four dominant phyla and a combined phylum group, which included minor phyla, for each participant. In the control group, the relative abundance of Firmicutes tended to increase, whereas that of Proteobacteria and Bacteroidetes decreased from baseline. These results are coherent with those shown in Table 4. In the intervention group, though the mean composition of Bacteroidetes slightly increased from baseline (Table 4), changes in the relative abundance of this phylum changes were quite diverse (Fig. 3). As illustrated in Fig. 3, five participants exhibited a lack of the Bacteroidetes phylum in their baseline data; however, all demonstrated the presence of the Bacteroidetes phylum after supplementation.