A total of 55 HTN patients were enrolled at Beijing Chaoyang Hospital between January and May 2021. HTN was defined as systolic BP (SBP) ≥140 mmHg, and/or diastolic BP (DBP) ≥90 mmHg, and/or previous history of HTN, and/or use of anti-hypertensive medicine before enrollment according to the 2010 Chinese guidelines for the management of HTN (Liu, 2011). The measurement of BP was based on serial data. Such as previous history of HTN and/or the use of anti-hypertensive medicine before enrollment, according to the 2010 Chinese guidelines for the management of HTN (Liu, 2011). The BP of each participating patient was measured in their sitting position, three times before their enrollment, by physicians with a random-zero mercury column sphygmomanometer and the fecal and serum samples collected. The BP of the patients was measured once randomly, and twice in the mornings. The BP was acquired on two consecutive days. The average data of these three times were used as the final BP. The exclusion criteria were: 1) hypertensive patients who received other anti-hypertensive drugs, such as calcium channel blocker and beta blockers; 2) patients suffering from cancer, heart failure, renal failure, stroke, peripheral artery disease, and chronic inflammation; 3) subjects who received statin, aspirin, insulin, metformin, etc.; and 4) individuals who were prescribed antibiotics or were using probiotics over the past two months.

Hypertensive patients were classified into the following categories: untreated group, i.e., newly diagnosed HTN patients who had never received anti-hypertensive treatment (n = 19), and treated group, i.e., those HTN patients who were under ACEI/ARB treatment (n = 36) for at least 4 weeks. In addition, the HTN patients treated with ACEI/ARB were further divided into the well-controlled group (n = 24, SBP <140 mmHg and DBP <90 mmHg) and poor-controlled group (n = 12, SBP ≥140 mmHg or DBP ≥90 mmHg) according to the hosts' BP levels.

The protocol of this study was approved by the Medical Ethics Committee from Beijing Chaoyang Hospital. Written informed consent was obtained from all patients prior to enrollment.

Fresh fecal samples were collected with individual sterile EP tubes, then quickly frozen and stored at −80°C in a cryogenic freezer for cryopreservation. The bacterial DNA was extracted using the TIANGEN kit according to the manufacturer’s instructions. The concentration and quality of the extracted DNA were photometrically determined with a NanoDrop^® ND-2000c UV–vis spectrophotometer (NanoDrop Technologies, Wilmington, DE, United States). The qualified DNA from the fecal samples was diluted to 1 ng/μl and subjected to the amplification of polymerase chain reaction with a universal primer set directed at the hypervariable V4 region of the bacterial 16S rRNA gene (515F 5′-GTGCCAGCMGCCGCGGTAA-3′ and 806R 5′-GGACTACNNGGGTATCTAAT -3′) and Phusion^® High-Fidelity PCR Master Mix with GC Buffer (New England Biolabs Co., Ltd.). PCR products were detected by electrophoresis using 2% agarose gel, and the target bands were recovered using gel recovery kits provided by QIAGEN. Following the 16S rDNA library, preparation, and generation with a TruSeq^® DNA PCR-Free Sample Preparation Kit, the quality of the library was assessed by using the Qubit@2.0 fluorometer (Thermo Scientific) and the Agilent Bioanalyzer 2100 system, and sequenced by Illumina NovaSeq6000.

Based on the unique barcode, raw reads were assigned to different samples, and the assigned paired-end reads of each sample were merged to raw tags by FLASH (Version 1.2.7). The raw tags were filtered, after which the clean tags were obtained using the QIIME (Version 1.9.1) quality control process. Clean tags were aligned to the Gold database, and the chimera sequence was detected by the UCHIME algorithm. Finally, the non-chimera clean tags were considered effective tags, which were clustered into operational taxonomic units (OTUs) with ≥97% sequence identity; chimeras were removed by using Uparse (Version 7.0.1001). Sequences with the highest frequency in the OTUs were selected as representative sequences of the OTUs and applied for taxonomic annotation by the RDP classifier according to the Greengene database.

Alpha diversity was performed to analyze the diversity and richness of the gut microbial community within the sample. QIIME software (Version 1.9.1) was applied to access bacterial alpha diversity parameters, including observed OTUs, Chao1 values, and Shannon and Simpson indices. Rarefaction curves were carried out by randomly extracting a certain amount of sequencing data from the samples, calculating the number of OTUs, building a curve based on the amount of sequencing data extracted, and the corresponding OTU numbers. The rarefaction curves directly reflected the rationality of the sequencing data and could indirectly reflect the richness of the OTUs in the samples. When the curve tends to be flat, it indicates that the sequencing data amount is progressive and reasonable, and more data would generate few novel OTUs.

Beta diversity is a comparative analysis of the microbial community composition of different samples. According to the OTU abundance of all the samples, the systematic relationship between the OTUs was used to analyze the UniFrac distance, indicating the distance between the samples by the information between the microbial sequences in each sample. The multivariate statistical methods included principal component analysis (PCA) and principal coordinate analysis (PCoA). The PCA was obtained through ade4 and ggplot2 packages and PCoA through the WGCNA, stats, and ggplot2 packages in the R software (version 3.3.3).

The microbial dysbiosis (MD) was determined as previously described (Xu et al., 2020). To access the degree of gut microbial dysbiosis, the MD index was analyzed as the log of [total abundance of organisms increased in untreated patients]/[total abundance of organisms reduced in well-controlled patients] for all the samples.

Prediction for the gut microbial community’s functional capabilities was performed by the phylogenetic investigation of communities by reconstruction of unobserved states (PICRUSt) Version 1.0.0 to generate the Kyoto Encyclopedia of Genes and Genomes (KEGG) ontology profiles. Briefly, based on the tree of the OTUs in the Greengene database and the gene information in the OTUs, the gene function of their shared ancestor was inferred through the KEGG database. Thus, the composition of the bacterial community was mapped to the database, constructing the gene function prediction of the gut bacteria.

In the present study, serum samples from 55 patients were subjected to a metabolomics analysis based on untargeted gas chromatography–time of flight mass spectroscopy (GC-TOF/MS). Overnight fasting venous blood samples were collected the next morning. The whole blood samples were then separated into a serum by centrifugation at 4°C. Then, 50 μl of the sample was mixed with 10 μl of the internal standard, and 175 μl of pre-chilled methanol/chloroform was added. After centrifugation at 14,000 g and 4°C for 20 min, 200 μl of the supernatant was transferred to an autosampler vial and subjected to GC-TOF/MS analysis. The untargeted metabolomics profiling was performed on the XploreMET platform (Metabo-Profile, Shanghai, China) and an untargeted GC-TOF/MS system with Agilent 7890B gas chromatography and a Gerstel multipurpose sample MPS2 (Gerstel, Muehlheim, Germany) were applied. For separation, an Rxi-5 ms capillary column (inner diameter: 30 m × 250 μm, film thickness: 0.25 μm; Restek Corporation, Bellefonte, PA, United States) was used. Helium at a constant flow rate of 1 ml/min was the carrier gas. The injection and transfer interfaces were at a temperature of 270°C, while the temperature for the source was at 220°C. The measurement was performed in the full scan model (m/z 50–500) at electron impact ionization (70 eV).

The raw mass spectral data generated by GC-TOF/MS were processed on the XploreMET software (v3.0, Metabo-Profile, Shanghai, China) to denoise and smoothen, pick peaks, and align metabolite signals. Data sets were transformed into data vectors for statistical analysis. All the measurements were expressed as the mean and standard deviation of the observed peaks. Using the XploreMET software, the retention indices and mass spectral data were compared with those reference standards of known chemicals presented in the JiaLib metabolite database, comprising over 1,200 mammalian metabolites purchased from Sigma-Aldrich (St. Louis, MO, United States), Santa Cruz (Dallas, TX, United States), and Nu-Chek Prep (Elysian, MN, United States). Thus, the precise metabolite annotation in each sample was performed.

The PCA serves as an unsupervised model to detect outliers, clustering, and classification trends. The principal components derived from the data set indicate the orthogonality to one another and reflect the reducing levels of the variations. The PCA significantly decreases the number of variables required for data representation and provides visualization of data clusters and outliers based on multiple variables. Partial least-square discriminant analysis (PLS-DA) and orthogonal PLS discriminant analysis (OPLS-DA) were used for multiclass classification, thus providing more valuable information beyond that obtained from the PCA. These are sophisticated multivariate statistical models for mass spectral data and class variables that can detect the differences in the metabolic profiles between groups. Statistical analyses of PCA, PLS-DA, and OPLS-DA were carried out with the software packages in R Studio (R Foundation for Statistical Computing, Vienna, Austria).

Continuous variables were presented as the median with an interquartile range. The categorical variables are shown as numbers and percentages. The Wilcoxon rank test and chi-squared tests were used to examine continuous and categorical variables, respectively. Student’s t-test or the Mann–Whitney Wilcoxon test was used to determine the significant difference between the groups. The linear discriminant analysis (LDA) effect size (LEfSe) version 1.0 was performed to identify taxa and KEGG pathways that are significantly different between the groups, with an LDA score = 2 as the cutoff value. The Z-score in the heat map was obtained by subtracting the mean and dividing the standard deviation. The Z-score was negative when the raw value was lower than the mean and was positive when higher than the mean. A volcano plot was applied for depicting significantly different variables between the groups. The analyses were carried out using SPSS, Version 22.0 (IBM Corp. Armonk, New York). In addition, a different abundance of genera and functions was determined by the Wilcoxon rank-sum test. The presented p values are two-tailed, and a p-value <0.05 was considered statistically significant.

The present study included 19 untreated hypertensive patients with no history of anti-hypertensive treatment and 36 HTN subjects under ACEI/ARB therapy (including 24 well-controlled and 12 poor-controlled). The primary characteristics of the patients (n = 55) are shown in Supplementary Table S1. Compared with untreated or poor-controlled patients, the SBP and DBP levels were significantly lower (p < 0.05) in the subjects showing well-controlled status under ACEI/ARB therapy. Moreover, except for age, there was no difference between groups in sex, cigarette smoking, body mass index, fasting blood glucose, total cholesterol, LDLC, and uric acid and creatinine. Yet, the average age was 49 years in the well-controlled group, and 43.5 years in the poor-controlled group, which is consistent with a previous study (Wang et al., 2014), suggesting that the control rate of BP in HTN patients was higher among the elderly in China.

In total, 6,115,624 high-quality clean tags with an average of 99,219 reads per sample were obtained for the fecal microbiota analysis. A shift in the microbial diversity is considered as a crucial factor indicating the degree of gut dysbiosis associated with diseases (Bäckhed et al., 2015). Moreover, alteration in the fecal microbiota structure has been found in HTN patients when compared to healthy individuals (Li et al., 2017). This study further examined alterations in the microbiome between HTN patients who received ACEI/ARB treatment and those who did not. No differences in OTUs, or alpha diversity indices (diversity within samples), including the Chao1, Shannon, and Simpson indices, were found between the groups (Supplementary Figure S1A). Yet, there was a trend toward elevated microbial alpha diversity in treated patients, but the difference was not significant.

Moreover, hypertensive patients who achieved the desired BP levels by ACEI/ARB treatment or without were further examined. No differences in the OTU numbers, Chao1 richness, Shannon index, or Simpson index were observed among the untreated, poor-controlled, and even well-controlled groups (Supplementary Figure S1B). However, the well-controlled groups showed an increasing tendency in the gut microbial diversity without statistic differences. Thus, it is suggested that ACEI/ARB therapy might exert protection against gut dysbiosis in hypertensive patients to a slight extent but fail to completely restore the disordered situation in alpha diversity.

Furthermore, to evaluate the features and structures of the intestinal microbiota in the hypertensive patients, beta diversity by the PCA and PCoA plots based on the phylum, class, order, family, and genus levels was conducted and compared, respectively. Neither PCA nor PCoA analysis could distinguish the untreated and treated patients into different clusters (Supplementary Figure S2). Well-controlled and poor-controlled subjects in the PCA and PCoA plots could not be distinguished from the other groups (Supplementary Figure S3). Although patients in the well-controlled group showed much lower BP levels compared with the untreated and poor-controlled patients, the impairment in both the gut microbial alpha diversity and beta diversity of HTN could not be reversed by ACEI/ARB therapy. Thus, we speculated that the influence of ACEI/ARB exerted on the microbial diversity was not strong enough to affect gut dysbiosis in HTN patients but may still alleviate dysfunction to some extent.

The rarefaction plots were assessed to characterize the sequencing data obtained from the samples. The rarefaction curves based on all the OTUs were flattened and reached a plateau in each group, implying the adequacy of sequencing depth (Figure 1A). We observed almost overlapping curves and fewer OTUs in the untreated and poor-controlled patients. More OTUs were obtained from the well-controlled patients, suggesting more abundant gut microbiota. A Venn diagram was performed to evaluate whether the pattern of shared and unique OTUs existed among the groups (Figure 1B). There were 279 OTUs overlapping in the 3 groups, and 10 OTUs were shared by the untreated and well-controlled groups. Eleven OTUs were shared by the ACEI/ARB treatment patients, regardless of being poorly controlled or well controlled, and four OTUs were shared by the untreated and poor-controlled groups. Additionally, three OTUs were uniquely presented in the untreated group, ten OTUs in the well-controlled group, and four OTUs in the poor-controlled group.

The compositions and taxonomic constitution of the fecal microbiota were then assessed at different taxonomic levels. Overall, 9 phyla, 15 classes, 21 orders, 41 families, and 60 genera were annotated. Supplementary Figure S4 shows the top 10 most enriched taxon identified in each group. The phylum Firmicutes dominated the intestinal microbiota, followed by Proteobacteria. Clostridia, Gammaproteobacteria, Bacteroidia, and Bacilli were the most predominant classes; Clostridiales and Enterobacteriales were the predominant orders; Lachnospiraceae, Ruminococcaceae, and Enterobacteriaceae were the predominant families; and Blautia, Faecalibacterium, Ruminococcus, Bacteroides, and Bifidobacterium were the predominant genera (Supplementary Figure S4).

For the three most dominant phyla consisting Firmicutes, Proteobacteria, and Bacteroidetes, the ternary plots in Figure 1C show that the average relative abundance of Firmicutes in the well-controlled group (75.255%) and poor-controlled group (77.039%) were higher than that of the untreated group (69.013%). The relative abundance of Proteobacteria and Bacteroidetes in the untreated group (19.706 and 6.298%, respectively) was higher than in the other groups (14.909% in poor-controlled and 12.317% in well-controlled for Proteobacteria; 4.465% in poor-controlled and 5.967% in well-controlled for Bacteroidetes). At the genus level, Klebsiella was predominant in the untreated group, while Clostridium was dominant in the ACEI/ARB group, especially in well-controlled individuals (Figure 1D). To assess the degree of fecal microbial dysbiosis, the MD index evaluating diseased status was included. Intriguingly, comparing the patients without intervention, the MD index was significantly attenuated in hypertensive subjects with well-controlled BP following ACEI/ARB treatment (Figure 1E).

To further identify the driving force for the observed alterations in dysbiosis of microbial communities among groups, differences in taxonomical composition between the groups were determined. A comparison was preliminarily performed between the treated and untreated groups. Firmicutes were predominant in both groups, with a higher relative abundance in the treated patients (Figure 2A). By contrast, the proportion of Proteobacteria was lower in the treated group. Moreover, Ruminococcus, Clostridium, and Megamonas, which belong to Firmicutes, were predominant in the treated group, while Coprococcus and Staphylococcus, Enterobacter and Klebsiella, along with Akkermansia were lower compared to the untreated group (Figure 2B). Using LEfSe analyses, the discriminant taxa between the untreated and treated groups were represented (Figure 2C). Odoribacter and Clostridium were clearly increased, while Staphylococcus, Enterobacter, and Klebsiella were drastically decreased in the treated patients. Figure 2D shows the heat map of the bacterial correlation in the treated patients and the untreated controls.

By subdividing patients into the well-controlled and poor-controlled groups, we further identified differences in relative abundance changes based on therapeutic effects. An LEfSe analysis demonstrated that 17 biomarkers were significantly different between the untreated and poor-controlled groups (Figure 3A). Enterobacter and Paraprevotella were lower in the poor-controlled group. Enterobacter and Paraprevotella have been previously confirmed to be extremely elevated in hypertensive patients or animals (Chang et al., 2020; Gutierrez-Calabres et al., 2020). In the well-controlled groups, the LDA scores revealed 19 biomarkers that were significantly different from the other groups (Figure 3B). Five bacterial taxa, including Clostridium, Odoribacter, Ruminococcaceae, Clostridiaceae, and Odoribacteraceae were clearly enriched in the well-controlled group. When examining the distinctions within the pharmaceuticals treated group, between well-controlled and poor-controlled groups, we found a spot of difference (Figure 3C). These gut bacteria altered among groups and showed apparent co-occurrence correlation with each other, indicating that the global shifts by ACEI/ARB treatment were collaborative (Supplementary Figure S5).

In addition, taxa discrimination between untreated and poor-controlled patients were regarded as not correlated with BP regulation due to indiscriminatingly high BP in both the groups. Likewise, the most important distinction between poor and well-controlled HTN groups was believed to be not ascribed to ACEI/ARB consumption. On the contrary, intestinal bacteria between well-controlled and untreated patients were indicated to be possibly linked to BP improvement under ACEI/ARB therapy. By excluding taxa that were neither associated with BP nor medical therapy, we obtained 11 shared taxa with those related to drug usage (untreated vs. treated) (Supplementary Figure S6). Proteobacteria, Gammaproteobacteria, Enterobacteriales, Enterobacteriaceae, Serratia, Phascolarctobacterium, Ruminococcus, and Klebsiella were progressively reduced from untreated to poor-controlled groups and ultimately to the well-controlled group, while Ruminococcaceae, Clostridium, and Odoribacter were found increased in well-controlled patients when compared to the untreated group.

The potentials of metabolic and functional pathways in the gut microbiome of hypertensive patients were examined and analyzed through PICRUSt. We compared the KEGG pathways at level 3 and identified 74 KEGG categories with clearly differential abundance between the untreated and well-controlled groups ( Figure 4A). The KEGG categories, including the bacterial secretion system, lipopolysaccharide biosynthesis, tryptophan metabolism, fatty acid metabolism, and phenylalanine metabolism inositol metabolism were prominently decreased in well-controlled patients. Comparisons between the untreated and poor-controlled groups for each KEGG functional category were further performed and are shown by the LDA score in Figure 4B. Most of the aforementioned potential microbial capacities suppressed in well-controlled patients were not found to be significantly improved in the poor-controlled groups. The terracomunal of well-controlled and poor-controlled groups mainly concentrated on promoting potential functions in peptidoglycan biosynthesis, replication and repair, nucleotide metabolism DNA repair, and recombination proteins.

Based on the serum metabolomics data, discrimination between the patients with and without treatment was assessed using PCA, PLS-DA, and OPLS-DA clustering analyses. As observed in the PCA plots, there was an overlap of the samples derived from distinct groups; treated patients could not be clearly distinguished from untreated ones (Figure 5A). Nonetheless, according to serum metabolic profiles, the PLS-DA scatter plots showed obvious separation between the treated and untreated groups (Figure 5B). In components 1 and 2, the patients under ACEI/ARB therapy showed significantly different scores than the untreated controls (Figure 5B). On the other hand, individuals in the treated group were distinguished from the others, as evidenced by the OPLS-DA score plots (Figure 5C). For a detailed metabolic composition, all the metabolites identified were classified into 18 classes, and the proportion of each is described in Figures 5D,E. The serum samples of individuals in the treated or untreated groups mainly consisted of organic acids, amino acids, carbohydrates, lipids, and fatty acids. Compared with the untreated ones, patients following medication had significantly higher levels of lipids and alcohol. Among all the 154 metabolites detected, the differentially abundant metabolites in the serum of patients were explored. Notably, 14 metabolites were observed to be discriminatory between the groups (Figure 5F). Interestingly, in treated patients, almost all the significantly altered metabolites were elevated.

Similarly, when further comparison was performed between poor-controlled and untreated, well-controlled and untreated, and well-controlled and poor-controlled patients, respectively, the PCA plots again failed to completely distinguish the different groups (Supplementary Figures S7A, 8A, 9A). Consistent with the aforementioned observations, scatter plots of PLS-DA at both components 1 and 2 obviously discriminated patients from different groups (Supplementary Figures S7B, 8B, 9B). OPLS-DA score plots further verified the distinctions of metabolic features between the groups (Supplementary Figures S7C, 8C, 9C). Among the 18 metabolic classes, the abundance of fatty acids and alcohols was different between poor-controlled and untreated ones; lipids and vitamins were unequal between well-controlled and untreated ones; but none was identified to be prominently distinct between well-controlled and poor-controlled patients (Supplementary Figures S7D, 8D, 9D).

Compared to untreated patients, there were 10 reduced metabolites and 2 enhanced in poor-controlled individuals, whereas 3 metabolites were reduced and 12 were elevated in well-controlled ones (Supplementary Figures S7E, 8E). However, five different metabolites were detected between patients of the well-controlled and poor-controlled groups by therapy (Supplementary Figure S9E). The relative abundance of these discriminatory metabolites between the groups is described in Figure 6. Fatty acids such as elaidic acid, oleic acid, linoleic acid, and palmitic acid were depleted in the poor-controlled group, whereas alcohols inositol, organic acids hydroxy propionic acid, oxalic acid, and fatty acid pentadecanoic acid were augmented in the well-controlled group, as compared with HTN patients with no history of treatment (Figures 6A,B). Oxalic acid was observed to be increased in well-controlled patients relative to both untreated and poor-controlled individuals (Figure 6C). The varied serum metabolites identified among groups were confirmed to be significantly correlated with fecal microbiota (Supplementary Figure S10). Inositol was found to be negatively correlated with opportunistic pathogens previously reported, such as Klebsiella, Enterobacter, and Proteobacteria.

Additionally, metabolites those altered between untreated versus poorly controlled hypertensive patients were not associated with BP regulation due to indiscriminate BP levels in these individuals. As both poor and well-controlled HTN patients received medical treatments, the other factors were speculated to be accountable for their distinctions. Among the metabolites found to be different in well-controlled versus untreated patients, which were highly possible to be linked with BP modulation under ACEI/ARB therapy, we excluded those neither associated with BP nor with medical therapy, and we further obtained seven shared ones with untreated versus treated patients (related to drug usage) (Figure 7A). Inositol, 4-deoxypyridoxine, methylphosphate, maleimide, uric acid, hexadecylglycerol, and fructose showed progressively increased in abundance from untreated to poor-controlled groups and finally to the well-controlled group, with marginal statistical significance in poorly controlled patients and more robust statistical difference in well-controlled versus untreated patients (Figures 7B–H).