The effects of psyllium husk on gut microbiota composition and function in chronically constipated women of reproductive age using 16S rRNA gene sequencing analysis

Characteristics

A total of 54 patients from both the psyllium husk group and placebo group completed treatment and were available for four weeks. The two treatment groups were commonly matched by age, body mass index (BMI), straining, <3 bowel movements (BMs) / week, hard stool, pain with BMs, feeling of incomplete defecation, digital maneuvering, and feeling of blockage (Table 1). More detailed clinical information is summarized in Supplementary Table 2. The two groups of baseline data were balanced and comparable. Psyllium husk was well tolerated, and no serious adverse events occurred.

Psyllium husk relieves the symptoms of constipated women of reproductive age

As shown in Table 2, there were significant differences between the psyllium husk group and placebo group in terms of straining (2, 8.0% vs. 16, 55.2%, P = 0.000), <3 BMs/week (5, 20.0% vs. 25, 86.2%, P = 0.000), hard stool (8, 32.0% vs. 18, 62.1%, P = 0.027), pain with BMs (7, 28.0% vs. 17, 58.6%, P = 0.024), and feeling of incomplete defecation (4, 16.0% vs. 19, 65.5%, P = 0.000). Digital maneuvering and feeling of blockage were not statistically different between the two groups. Although feeling of blockage did not differ to a statistically significant degree between the groups, the proportion thereof was lower in the psyllium husk group than in the placebo group. These results indicated that psyllium husk as supplementation relieved the symptoms of constipation, and that these therapeutic effects occurred without serious adverse reactions.

Taxonomic profiling of gut microbiota in women of reproductive age receiving placebo and psyllium husk intervention using 16S rRNA gene sequencing

Numerous studies have shown that disorders of intestinal flora are among others associated with chronic constipation [11]. To further explore the potential intestinal flora–related mechanisms of psyllium husk in relieving the clinical symptoms of chronic constipation in women of reproductive age, we determined microbiotal compositions via 16S rRNA sequencing. After filtering, quality control, and chimera removal, a total of 4,658,291 effective sequences were obtained from the 54 fecal samples (29 samples from the placebo group, 25 samples from the psyllium husk group). In addition, 1,335 operational taxonomic units (OTUs), which can be used to further analyze species within sample via rapid visualization of species composition and abundance, were matched at 97% identity, including 21 phyla, 37 classes, 74 orders, 126 families, 261 genera, and 230 species of gut microbes that we annotated for subsequent analyses. Venn diagrams showed that the two groups shared 844 common OTUs (Figure 1A). We used a species-accumulation boxplot to evaluate the sufficiency of sample number and species richness. As shown in Figure 1B, when the number of samples reached 54, the number of species observed was nearly parallel, indicating that the sample size of our experiment was sufficient. Rank abundance was used to reflect the richness and uniformity of species in each sample, while the rarefaction curve was used to reflect the rationality of the amount of sequencing data and to indirectly reflect species abundance in each sample. The rank abundance curve of the psyllium husk group was slightly lower than that of the placebo group, but this difference was not statistically significant (Figure 1C). Furthermore, the rarefaction curve had similar trends in both groups (Figure 1D).

To further evaluate how changes in microbial community structure differed between the two groups, we measured microbial α-diversity using the Shannon, Chao, and ACE diversity indices. Results showed that the between-group difference was not statistically significant (P = 0.938, P = 0.269, and P = 0.469, respectively; Figure 2A–2C). By contrast, β-diversity analysis showed that total diversity was significantly different (P = 1.67e-05 and P = 0.005 for unweighted and weighted UniFrac, respectively; Figure 2D, 2E).

To clarify the effect of psyllium on the intestinal flora of constipated women of reproductive age, we conducted a bar-plot analysis. A total of 18 different microbial phyla were detected in all samples; the top 10 gut microbes at the phylum level in both groups are shown in Figure 3A. Firmicutes was the most abundant phylum, accounting for 53.93% and 59.09% in the placebo and psyllium husk groups, respectively. Bacteroidetes constituted the second most common phylum present in the placebo and psyllium husk groups (35.14% vs. 25.78%, respectively). Actinobacteria was the third most common, accounting for 5.14% and 7.04% in the placebo and psyllium husk groups, respectively. Fusobacteria (0.55% vs. 0.02%), Proteobacteria (4.00% vs. 2.29%), Synergistetes (0.14% vs. 0.03%), Verrucomicrobia (0.19% vs. 0.09%), and Acidobacteria (0.04% vs. 0.00%) were more abundant in the placebo group than in the psyllium husk group. Meanwhile, Euryarchaeota (0.51% vs. 1.01%) and Tenericutes (0.30% vs. 0.60%; Figure 3A) were increased in the psyllium husk group. These results indicated that overall microbial composition at the phylum level differed significantly between the two groups.

At the genus level, we detected a total of 261 bacterial taxa, and we found that microbial composition differed significantly between the two groups. Per our analysis of the relative abundance of bacterial genera > 0.1%, the top 10 abundances are shown in Figure 3B, as follows for the placebo and psyllium husk group, respectively: Faecalibacterium (10.96% vs. 14.19%), Bacteroides (23.40% vs. 19.22%), unidentified_Lachnospiraceae (4.79% vs 3.35%), Romboutsia (1.21% vs. 3.20%), Bifidobacterium (4.19% vs. 5.78%), Megamonas (1.90% vs. 0.34%), Streptococcus (1.55% vs. 2.33%), Megasphaera (2.13% vs. 0.65%), Paraprevotella (1.19% vs. 0.31%), and Lactobacillus (0.86% vs. 0.54%). Our genus level analysis revealed that some strains were significantly increased in the control group. We used species-abundance cluster heatmap analysis to visualize the corresponding abundance relationship between samples and bacterial communities at the genus level (Figure 3C, 3D), which confirmed the results of our bar-plot analysis.

To further determine whether specific individual bacterial taxa were differentially enriched between the placebo and psyllium husk groups, we applied LEfSe analysis, which couples LDA with effect-size measurements. As shown in Figure 4A, the following six taxa showed differentiated distributions with LDA scores >4: f_Enterobacteriaceae, o_Enterobacteriales, and s_Bacteroides_vulgatus were significantly more abundant in fecal samples from the placebo group, while we identified significant enrichment in c_Clostridia, o_Clostridiales, and f_Ruminococcaceae in the psyllium husk group.

Differential enrichment of specific bacteria is shown via cladograms based on an LDA = 4 for pairwise comparisons. According to this analysis, there was one family, one class, and one order in the psyllium husk group; and as well as one family and one order in the placebo group, that were differentially abundant between the two groups (Figure 4B). These results indicated that psyllium husk may be a significant factor in shaping gut microbiotal composition, possibly relieving the symptoms of constipation by reshaping intestinal flora composition in constipated women of reproductive age.

Correlation network analysis and random forest analysis of microbiota abundance

We further performed correlation network analysis to evaluate whether psyllium husk was related to changes in the correlation and constructive interaction structures of gut microbiota. The network results showed that samples from the placebo group had a smaller network diameter (ND; 6 vs. 10), lower graph density (GD; 0.014 vs. 0.006), and lower average degree (AD; 1.056 vs. 2.545) than those from the psyllium husk group (Figure 5). This indicated greater significant correlations and clustering of OTUs samples in the psyllium husk group.

Next, we performed random forest analysis using the relative abundances of the microbial populations to predict the potential of bacterial taxa in constipated women of reproductive age. Based on the species abundance at genus levels shown in this analysis, we selected different numbers of species by gradients to construct a random forest model. We screened out important species using MeanDecreaseAccuracy and MeanDecreaseGin (Figure 6A, 6B), cross-validated each model (10-fold), and constructed corresponding ROC curves. Coprobacter and Sellimonas in combination with Enterococcus achieved an area under the curve (AUC) value of 84.42% in the discovery model; Coprobacter and Bacillus in combination with Sellimonas achieved an AUC value of 78.57% in the invalidation model (Figure 6C, 6D).

Predictive microbial functional profiling

Tax4fun (http://tax4fun.gobics.de/) is a software package that can predict the functions of microbial communities based on 16S rRNA datasets [20]. It has been shown to provide a good approximation of functional profiles for data obtained from the metagenome. Therefore, in our present study, we used Tax4Fun analysis to reveal differential functional profiles of communities between the placebo and psyllium husk groups. At level 1, the proportion of metabolic and human disease-related sequences was significantly decreased (P < 0.05) in the fecal microbiome of the psyllium husk group, while cellular processes were significantly increased (Figure 7A). At level 2, the functional categories associated with glycan biosynthesis and metabolism, enzyme families, metabolism of other amino acids, cellular processes and signaling, drug resistance, and signaling molecules and interaction were enriched in the placebo group, while transcription and genetic information processing were increased in the psyllium husk group (Figure 7B). At level 3, we found that 15 pathways (peptidases, oxidative_phosphorylation, transport, other_glycan_degradation, bata-lactam_resistance, arginine_and_proline_metabolism, lipopolysaccharide_biosynthesis_proteins, folate_biosynthesis, monobacyam_biosynthesis, membrane_and_intracellular_structural_molecules, glycosaminoglycan_degradation, cationic_antimicrobial_peptide_(CAMP)_resistance, lipopolysaccharide_biosynthesis, antimicrobial_resistance_genes, and ubiquinone_and_other_terpenoid-quinone_biosynthesis) were significantly increased in the placebo group, while 7 pathways (replication,_recombination_and_repair_proteins, arginine_biosynthesis, valine,_leucine_and_isoleucine_biosynthesis, thiamine_metabolism, pantothenate_and_CoA_biosynthesis, signal_transduction_mechanisms, and chloroalkane_and_chloroalkene_degradation) were significantly enriched in the psyllium husk group (Figure 7C).

According to the functional annotations and abundance information of samples in the SILVA database, we selected the 35 most abundant features and their abundance information in each sample to draw a heatmap and clustered the functional differences. At level 1 (Supplementary Figure 1A), the functional categories associated with organismal_systems, human_disease, and metabolism were enriched in the placebo group, while genetic_information_processing, environmental_information_processing, and cellular_processes were increased in the psyllium husk group. At level 2 (Supplementary Figure 1B), we found that 21 pathways in the placebo group and 14 pathways in the psyllium husk group were significantly increased. At level 3 (Supplementary Figure 1C), we found significant increases of 23 pathways in the placebo group and 12 pathways in the psyllium husk group.

Patients and trial design

This was a single-blinded, randomized, placebo-controlled trial of 54 women of reproductive age with symptoms of chronic constipation (meeting Rome IV diagnosis standards). Most of the patients were recruited from Shanxi Bethune Hospital. Patients were included if they satisfied the following conditions: (a) had a BM frequency of <3 times/week over the preceding 3 months; (b) were women of reproductive age (18–49 years); and (c) had a BMI within a normal range from 18.5-24 kg/m² [33]. Patients were excluded based on any of the following: (a) were children, men, pregnant, or postmenopausal; (b) had metabolic or neuropsychiatric conditions, cancer, diabetes, Alzheimer disease, or Parkinson disease; (c) had history of intestinal surgeries; or (d) were taking antibiotics, probiotics, prebiotics, nonsteroidal anti-inflammatory drugs (NSAIDs), opioids, traditional Chinese medicine (TCM), proton pump inhibitors (PPIs), or histamine receptor antagonists during the preceding month before sample collection. After participants provided written informed consent, they were advised at the first visit to maintain their daily lifestyles, diets, and physical activities throughout the study. They were asked to take a placebo or psyllium husk (Supplementary Table 1) with a cup of warm water twice a day (30 min before breakfast and dinner). The placebo was consistent with psyllium husk except that it did not contain the seed shell of Plantago rotundus. Participants used a stool diary to record the quantity and characteristics of their stools and to subjectively evaluate the effectiveness of the interventional agent. At the end of week 1, we contacted them to assess compliance with the intervention. Clinicians evaluated the possible side effects of the intervention, as well as efficacy parameters, both at baseline and weekly after the start of treatment. The entire experiment lasted 4 weeks. The 60 participants were randomly assigned to one of two treatment groups. By the end of the trial, three patients had been lost to follow-up; two in the psyllium husk group had withdrawn from the study, while one in the placebo group had withdrawn from the study (Supplementary Figure 2). Please refer to the Supplemental Data table for more clinical characteristic information about the participants.

Randomization scheme

A total of 60 participants who met the research criteria were recruited and randomly assigned to two treatment groups. Participants were randomly assigned to a preexisting list generated according to a computer program, and the assignment of each group was concealed in an opaque sealed envelope and kept by a dedicated person.

Sample collection

Subjects’ fecal samples were collected into tubes pre-filled with Stool DNA Stabilizer (PSP^R Spin Stool DNA Plus Kit; Invitek Molecular, Berlin, Germany). Within 3 days, we transferred the tubes to −80° C storage until the next processing step.

16S ribosomal ribonucleic acid (rRNA) gene sequencing analysis

We performed 16S rRNA sequencing analysis of human feces as previously described [34, 35]. Briefly, we extracted fecal microbial DNA from 54 samples (29 samples from the placebo group and 25 samples from the psyllium husk group) using an E.Z.N.A. Stool DNA Kit (Omega Bio-tek, Inc., Norcross, GA, USA) per the manufacturer’s protocols. Subsequently, we conducted polymerase chain reaction (PCR) amplification of bacterial V3–V4 hypervariable regions of the 16S rRNA gene using the primers PF 5′-CCTAYGGGRBGCASCAG-3′ and PR 5′-GGACTACNNGGGTATCTAAT-3′. The barcode sequences and FLX Titanium adaptors were incorporated into the primers. Then, we pooled the purified amplification products in equal molar amounts and sequenced the paired ends on an Illumina MiSeq platform (Illumina, Inc., San Diego, CA, USA). Raw data were demultiplexed, quality filtered, and spliced to obtain clean data. Effective sequences were clustered into OTUs with a 97% identity cutoff, and chimeric sequences were removed via the UCHIME algorithm (https://www.drive5.com/usearch/manual/uchime_algo.html). We classified OTUs using the Ribosomal Database Project (RDP) classifier algorithm (http://rdp.cme.msu.edu/) and the SILVA database (https://www.arb-silva.de/).

Bioinformatic analysis

We analyzed 16S rRNA gene diversity using the Quantitative Insights Into Microbial Ecology (QIIME) tool. To determine α-diversity, we used the Chao, Shannon, and Abundance-based Coverage Estimator (ACE) diversity indices [36]; both unweighted and weighted UniFrac phylogenetic distance matrices were used to evaluate β-diversity [37, 38]. Linear discriminant analysis (LDA) of effect size (LEfSe), Student’s t test, and the Wilcoxon rank-sum test were used to analyze differences in bacterial abundances at different bacterial classification levels. Relationships among microbial communities were visualized by a network [39]. Random forest is a classic machine learning model based on a classification tree algorithm, which we performed as previously described [40]. Briefly, based on the analysis of species richness, we constructed a random forest model by selecting different numbers of species by gradients at different classification levels. The important species were selected by MeanDecreaseAccuracy and MeanDecreaseGin. Then, we cross validated each model and drew ROC curves.

Statistical analyses

For characteristics of constipated patients, data are described as the mean ± standard deviation (SD) or as the median (percentage). We used Wilcoxon rank-sum test, independent sample t test, Pearson’s χ² test, or Fisher’s exact test to evaluate differences between the groups. P < 0.05 was considered statistically significant. We performed all statistical analyses using SPSS version 19.0 (IBM Corp., Armonk, New York, USA).