Non-Ischemic Heart Failure With Reduced Ejection Fraction Is Associated With Altered Intestinal Microbiota

Themistoklis Katsimichas; Tomohito Ohtani; Daisuke Motooka; Yasumasa Tsukamoto; Hidetaka Kioka; Kei Nakamoto; Shozo Konishi; Misato Chimura; Kaoruko Sengoku; Hiroshi Miyawaki; Taiki Sakaguchi; Ryu Okumura; Konstantinos Theofilis; Tetsuya Iida; Kiyoshi Takeda; Shota Nakamura; Yasushi Sakata

doi:10.1253/circj.CJ-17-1285

Abstract

Background: Research suggests that heart failure with reduced ejection fraction (HFrEF) is a state of systemic inflammation that may be triggered by microbial products passing into the bloodstream through a compromised intestinal barrier. However, whether the intestinal microbiota exhibits dysbiosis in HFrEF patients is largely unknown.

Methods and Results: Twenty eight non-ischemic HFrEF patients and 19 healthy controls were assessed by 16S rRNA analysis of bacterial DNA extracted from stool samples. After processing of sequencing data, bacteria were taxonomically classified, diversity indices were used to examine microbial ecology, and relative abundances of common core genera were compared between groups. Furthermore, we predicted gene carriage for bacterial metabolic pathways and inferred microbial interaction networks on multiple taxonomic levels. Bacterial communities of both groups were dominated by the Firmicutes and Bacteroidetes phyla. The most abundant genus in both groups was Bacteroides. Although α diversity did not differ between groups, ordination by β diversity metrics revealed a separation of the groups across components of variation. Streptococcus and Veillonella were enriched in the common core microbiota of patients, while SMB53 was depleted. Gene families in amino acid, carbohydrate, vitamin, and xenobiotic metabolism showed significant differences between groups. Interaction networks revealed a higher degree of correlations between bacteria in patients.

Conclusions: Non-ischemic HFrEF patients exhibited multidimensional differences in intestinal microbial communities compared with healthy subjects.

The intestinal microbiota, or gut flora, comprises mutualistic microorganisms playing crucial roles in several host-related functions, such as energy harvest from undigested food, vitamin synthesis, and the development of a healthy host immune system.¹ The microbial community composition depends upon many factors, including the host’s geographic origin and diet, but is generally dominated by anaerobic bacteria and tends to remain stable from early childhood to early old age, save significant perturbations, like systemic antibiotic treatment.²^–⁹ Beyond homeostasis and health, research has highlighted compositional and functional alterations of the intestinal microbiota in a variety of human disorders, such as obesity, inflammatory bowel disease, renal failure, and atherosclerosis.¹⁰^–¹³

Editorial p ????

The relationship of the intestinal microbiota to heart disease is largely unknown, with the contribution of the gut to the inflammatory state of heart failure (HF) being an active field of research.¹⁴ More specifically, knowledge of host-microbiota relationships in HF is very limited. Low cardiac output, splanchnic vasoconstriction, and/or venous congestion caused by HF may alter gut physiology and predispose to nonocclusive ischemia of the microvilli and wall edema of the large intestine, with conceivable implications for intestinal microbial communities engaging in constant cross-talk with intact epithelia.¹⁵ An increased concentration of intestinal bacteria in HF patients has been documented and, recently, HF was associated with both an increase in potentially pathogenic intestinal bacteria in an Italian cohort and a depletion of core intestinal microbiota in a population of north European patients.¹⁶^–¹⁹ By using 16S rRNA sequencing and predictive metagenomics, and by inferring microbial interaction networks, we explored possible compositional alterations of intestinal bacterial communities in Japanese patients with non-ischemic HF with reduced ejection fraction (HFrEF), predicted their functional metabolic profiles, and assessed their correlation patterns.

Methods

Study Design and Protocol

This was a study recruiting consecutive patients with non-ischemic HFrEF admitted to Osaka University Hospital from October 2015 to April 2017. Heart disease patients with no HF (termed “non-HF controls”) and healthy volunteers were also recruited as controls. HFrEF was defined as HF with EF <50% and diagnosis was made according to relative signs and symptoms, facilitated by echocardiographic measurements of the left ventricular EF (LVEF). Both chronic stable and acute HF/acute decompensated chronic HF patients were included in the final sample, representing, to the extent possible, a broad spectrum of the HF syndrome. Ischemic etiology for HFrEF patients was excluded based on coronary angiography (CAG) data. All subjects gave written informed consent to participate and the local ethics committee approved the study.

After being properly instructed, subjects gave a stool sample at their earliest convenience, which was collected by appointed staff with a feces tube from a commode specimen collector, immediately stored in a common house freezer until the appointed transporter arrived, and then carried in liquid nitrogen and stored at −80℃ in a dedicated facility. The stool sampling day was considered as baseline for evaluations. Blood count, blood biochemistry, and various clinical and demographic data were collected from patient records or generated for healthy controls to explore differences between groups and correlations with bacterial community composition. Additionally, subjects underwent routine echocardiographic examination, as previously described.²⁰

Study Population

During the recruitment period, a total of 272 HF patients were admitted to Osaka University Hospital. Of them, 95 patients who either had HF with preserved EF (HFpEF) or HFrEF caused by ischemic heart disease were immediately excluded. Another 40 patients were excluded based on age (>70 years old), regardless of underlying disease. A further 109 patients were also excluded for a multitude of reasons, including denying consent, giving consent but failing to provide a sample in the designated commode specimen collector during their hospital stay, having a LV assist device implanted at the time of screening, and other specific exclusion criteria, chief among them the use of antibiotics and/or probiotics within 4 weeks prior to sampling (Figure 1, Table 1). Finally, 28 patients were included for analysis.

Figure 1.

Study flow chart. During the recruitment period, patients were screened based on heart failure (HF) syndrome, age, and other exclusion criteria shown in Table 1. HFpEF, HF with preserved ejection fraction; HFrEF, HF with reduced ejection fraction; LVAD, left ventricular assist device.

Table 1. Study Exclusion Criteria

• Age <18 or >70 years

• Systemic use of oral or intravenous antibiotics within 4 weeks prior to stool sampling day

• Topical skin use of antibiotic-containing creams/ointments within 1 week prior to stool sampling day

• Ingestion of probiotic-containing preparations within 4 weeks prior to stool sampling day (common yoghurt excluded)

• Recent major change in dietary habits (e.g., became vegetarian)

• Obesity (body mass index ≥30)

• Chronic kidney disease stages 4/5

• Intestinal disorders, including: inflammatory bowel disease, irritable bowel syndrome, chronic diarrhea, colitis, active
infectious gastroenteritis, untreated Helicobacter pylori infection

• Intestinal surgery in the past 5 years (excluding cholecystectomy and appendectomy)

Extra criterion for HFrEF patients

• Coronary artery disease

Extra criteria for healthy controls

• Chronic illness affecting major organ systems and requiring continuous medical attention

• Excessive alcohol intake, defined as consumption >2 Japanese standard drinks/day (>40 g of ethanol/day)

• Evidence of acute illness at or just prior to stool sampling day, based on history and vital sign measurements

HFrEF, heart failure with reduced ejection fraction.

Eligibility for healthy controls was based on the exclusion criteria listed in Table 1; 20 healthy controls were voluntarily recruited, but 1 withdrew consent prior to sampling and was excluded, resulting in 19 healthy controls included in the study.

Non-HF controls were recruited by a similar process as used for the HFrEF patients to comparatively assess possible effects of age and heart disease-related medication on intestinal bacterial composition and came from a pool of patients scheduled for CAG, coronary computed tomography (CT) angiography, or catheter ablation during the recruitment period. Eligibility was based on the same exclusion criteria as for HFrEF patients, minus the coronary artery disease (CAD) criterion. All CAD patients had documented disease on CAG and/or coronary CT angiography. No patients with acute coronary syndrome were considered. Disease duration for non-HF controls reflected the time since documentation of the disorder with appropriate examination. A total of 18 non-HF controls were finally included in the analysis.

Dietary Information

The nutritional intake of the HFrEF patients and non-HF controls was primarily estimated from dietary information available in hospital records. The nutritional intake for the healthy controls and a few other subjects sampled on their first day of hospital admission was based on the latest edition (2015) of the Japanese National Health and Nutrition Survey, which includes the age- and sex-adjusted daily average nutritional intake of the Japanese population (mhlw.go.jp/stf/houdou/0000142359). Subjects would be assigned estimated values included in the survey, according to their age and sex.

Retrospective insight into dietary habits was provided through administration of a structured questionnaire, which assessed habitual intake in 21 food and drink categories. The answers for consumption fell into 4 categories of frequency (less that once a week/1–2 times a week/3–7 times a week/more than once a day) and were treated as ordinal variables.

DNA Extraction, 16S rRNA Sequencing, and Taxonomic Classification

DNA was extracted from stool samples using a DNeasy PowerSoil Kit (QIAGEN, Hilden, Germany). Libraries were prepared according to the Illumina 16S Metagenomic Sequencing Library Preparation Guide, with a primer set (27Fmod: 5′-AGR GTT TGA TCM TGG CTC AG-3′ and 338R: 5′-TGC TGC CTC CCG TAG GAG T-3′) targeting the V1–V2 regions of the 16S rRNA gene. 251-bp paired-end sequencing of the amplicons was performed on a MiSeq System (Illumina, San Diego, CA, USA), using the MiSeq 500-cycle v2 kit. Paired-end sequences were merged using PEAR (sco.h-its.org/exelixis/web/software/pear/). Merged reads were quality-trimmed with BBtrim (bbmap.sourceforge.net). Using random_sequence_sample.pl (ualberta.ca/~stothard/software.html), 20,000 reads per sample were randomly selected for further analysis. The processed sequences were clustered into operational taxonomic units (OTUs) defined at a 97% similarity cutoff using UCLUST version 1.2.22q. Representative sequences for each OTU were then classified taxonomically by RDP Classifier version 2.2, with the Greengenes 13_8 database. The bioinformatics pipeline QIIME version 1.9.1 was used as the informatics environment for processing of raw sequencing data.²¹

Statistical Analysis

For baseline characteristics, continuous variables were compared by Tukey’s range test when normally distributed and by the Steel-Dwass test when not, to control for multiple comparisons. Categorical variables were compared by Fisher’s exact test with post-hoc correction, as needed. Ordinal variables for dietary information were compared by the Wilcoxon rank-sum test.

Propensity scores were calculated with the R package matchIt.

Bacterial relative abundance data were compared between groups by the Wilcoxon rank-sum test, followed by false discovery rate (FDR) adjustment at the level of 0.05.²²

The within-subject α diversity of bacterial communities was assessed by richness (S), evenness (Shannon’s E), the Shannon index (H), the Chao 1 estimator, and the Simpson index (1/λ), and compared between groups by Welch’s test.²³^–²⁵ The between-subjects β diversity was assessed with the Bray-Curtis dissimilarity and the unweighted and weighted UniFrac distance metrics.²⁶^,²⁷ The statistical significance of compositional differences between sets of samples was assessed by permutational multivariate analysis of variance (PERMANOVA) and analysis of similarities (ANOSIM).²⁸

Relationships between microbial data and host-related metadata were assessed by multivariate analysis (Supplementary Methods).

Functional metabolic profiles of bacterial communities were predicted with PICRUSt (phylogenetic investigation of communities by reconstruction of unobserved states).²⁹ Variation between groups was assessed by PERMANOVA.

Microbial interaction networks were inferred by SPIEC-EASI (sparse inverse covariance estimation for ecological association inference),³⁰ as detailed in Supplementary Methods. Network differences were assessed by measures of graph theory.

The level of significance for all tests was set at α=0.05 and a probability value of P<0.05 was considered to be statistically significant. The statistical software JMP Pro 13 (SAS, Cary, NC, USA), the statistical computing environment R (r-project.org), the statistical package Primer 7 (Quest Research Ltd, Auckland, New Zealand), the open-source bioinformatics pipeline QIIME version 1.9.1 (qiime.org), and Microsoft Office Excel 2013 (Microsoft Corp., Redmond, WA, USA) were used for database construction and data analysis.

Results

Subjects’ Characteristics

The baseline characteristics of the HFrEF patients, non-HF controls, and healthy controls are shown in Table 2. The median disease duration for HFrEF patients was 3 years, and more than half were diagnosed with idiopathic dilated cardiomyopathy. Their mean LVEF was 25% and 61% were classified as NYHA class III or IV at baseline.

Table 2. Baseline Characteristics

Variables	HFrEF patients (n=28)	Non-HF controls (n=18)	Healthy controls (n=19)
Demographics
Age, years	51±10*^,†	57±9^#	36±6
Males (%)	21 (75)	15 (83)	16 (84)
Diagnosis	DCM 15 (54%)	CAD 16 (89%)	NA
	DHCM 5 (18%)	Arrhythmia 2 (11%)	NA
	Other 8 (28%)	–	NA
Disease duration, years	3 (0.3–11.4)	1 (0.4–3.3)	NA
Clinical variables
NYHA class III–IV (%)	17 (61)	NA	NA
BMI, kg/m²	21.7±3.4^†	25.6±3.7^#	21.8±1.9
Systolic BP, mmHg	95±13*^,†	123±13	116±12
Diastolic BP, mmHg	59±8*^,†	70±10	71±8
Echocardiography
LVEF, %	25±9*^,†	62±5	62±4
LVEDD/m², mm	41±8*^,†	26±2	28±2
LVMI, g/m²	125±45*^,†	81±16	69±10
E/A, m/s	1.75±0.97^†	0.98±0.28^#	1.82±0.44
E/e’ (average), cm/s	13±5*^,†	9±3	6±1
RAP, mmHg	3 (3–8)	8 (3–8)	3 (3–8)
Catheterization data
RAP, mmHg	5 (3–9)	NA	NA
CI (Fick), L/min/m²	2.6±0.7	NA	NA
Blood biochemistry
eGFR, mL/min/1.73 m²	64.4±25.2*	76.0±17.0	87.0±16.9
BNP, ng/L	375 (145–630)*^,†	16 (8–36)^#	2 (2–6)
CRP, nmol/L	6.67 (3.81–35.24)*	3.81 (3.81–10.95)^#	2.86 (1.90–3.81)
Nutritional intake
Carbohydrates, g	260 (260–260)*	260 (260–288.4)^#	292.9 (290.8–292.9)
Lipids, g	58 (58–58)*	58 (51.8–61.2)^#	65.7 (64.7–65.7)
Protein, g	70 (70–70)*	70 (70–76.2)	74.8 (74.8–75.1)
Fiber, g	NA	NA	13.3 (13.3–13.3)
Energy, Kcal [MJoule]	1,800 (1,800–1,800)* [7.53 (7.53–7.53)]	1,800 (1,750–2,171) [7.53 (7.32–9.08)]	2,161 (2,161–2,168) [9.04 (9.04–9.07)]
Medication
β-blockers (%)	27 (96)^†	9 (50)	NA
ACEI or ARB (%)	24 (86)^†	10 (56)	NA
PPI (%)	16 (57)	14 (78)	NA

Values are expressed as mean±standard deviation for normally distributed variables, median and interquartile range for non-normally distributed variables, or number of patients and relative percentage of the group’s total in parentheses. *P<0.05 vs. healthy controls, ^†P<0.05 vs. non-HF controls, ^#P<0.05 vs. healthy controls. Because of missing data, HFrEF patients, non-HF controls, and healthy controls had the following number of subjects, respectively, regarding the following variables: BNP: n=28, n=18, n=17; CRP: n=28, n=17, n=17; E/A: n=21, n=17, n=19; E/e’: n=17, n=14; Estimated GFR: n=28, n=18, n=17; LVMI: n=28, n=17, n=19; n=18; RAP: n=28, n=11, n=19. Disease duration for non-HF controls was not compared with that of HFrEF patients, because the underlying conditions were unrelated. A, peak late diastolic transmitral flow velocity; ACEI, angiotensin-converting enzyme inhibitor; ARB, angiotensin II receptor blocker; BMI, body mass index; BNP, B-type natriuretic peptide; BP, blood pressure; CAD, coronary artery disease; CI, cardiac index; CRP, C-reactive protein; DCM, dilated cardiomyopathy (idiopathic); DHCM, dilated phase of hypertrophic cardiomyopathy; E, peak early diastolic transmitral flow velocity; e’, tissue Doppler velocity of mitral annulus; eGFR, estimated glomerular filtration rate; HF, heart failure; HFrEF, heart failure with reduced ejection fraction; LVEDD, left ventricular end-diastolic diameter; LVEF, left ventricular ejection fraction; LVMI, left ventricular mass index; NA, not applicable or not available; NYHA, New York Heart Association; PPI, proton-pump inhibitor; RAP, right atrial pressure.

Dietary Data

The estimated daily median nutrient and calorie intake showed slight but statistically significant differences between groups (Table 2), but the general nutrient intake pattern was the same in all groups, with approximately two-thirds of intake provided by carbohydrates, one-sixth by protein, and one-sixth by lipids. The questionnaire-based assessment of dietary habits did not reveal significant differences in any of the major food categories (meat, seafood, dairy products, vegetables, and rice) between HFrEF patients and healthy controls after FDR adjustment, but showed that patients, when on their home, long-term diet, were consuming sweets, eggs, fast food, and coffee less frequently than healthy controls.

Taxonomic Classification of Intestinal Bacteria in HFrEF Patients and Healthy Controls

Intestinal bacteria with a relative abundance of at least 0.1% in any subject were classified into 13 phyla, 21 classes, 27 orders, 56 families, and 134 genera. The bacterial communities of both the patient and the healthy control group were dominated by the Firmicutes and Bacteroidetes phyla (Figure 2A). The third most abundant phylum was the Proteobacteria. These results were consistent with the findings of the Human Microbiome Project.³¹

Figure 2.

Taxonomic classification of intestinal microbiota. (A) Stacked bar chart of the phylum-level classification of intestinal bacteria by relative abundance. Bacteroidetes (orange) and Firmicutes (blue) are the 2 most abundant phyla in all subjects, totaling a mean of 93%, 96%, and 96% of the whole communities in HFrEF patients, healthy controls, and non-HF controls, respectively. Vertical bars represent individual samples. (B) Stacked bar chart of the 20 most abundant genera in HFrEF patients, healthy controls, and non-HF controls, revealing that the vast majority of genera in these 3 “top 20” groups were common in all subjects and highlighting the overall dominance of Bacteroides (blue).Vertical bars represent individual samples. HF, heart failure; HFrEF, HF with reduced ejection fraction.

At the genus level, the patient group was characterized by 119 genera, 31 of each were absent from the healthy controls, while the healthy control group included 103 genera, 15 of which were absent from the patient group (Figure 3A, Table S1). Whittaker plots revealed a similar pattern in evenness for both groups (Figure 3B).

Figure 3.

Genus-level overlap and evenness. (A) Two unscaled Venn diagrams showing sets of genera in bacterial communities of HFrEF patients and healthy controls (HC). The left diagram shows an intersection of 88 bacterial genera, 31 genera present only in patients, and 15 genera present only in HC, making a total of 134 classified genera. The right diagram shows the common core set of genera present in ≥50% of patients or HC, 34 of which were shared, 3 of which were present only in the patient core (Eggerthella, Veillonella, and Fusobacterium), and 3 of which were present only in the HC core (Prevotella, Paraprevotella, and SMB53). (B) Whittaker plots, revealing similar evenness (distribution) for bacterial genera in HFrEF patients and HC (interpreted as similar slopes in the lines of the 2 graphs). The Y axis represents the log-transformed relative abundance of genera and the X axis ranks the genera, starting with the genus with the highest relative abundance. HFrEF, heart failure with reduced ejection fraction.

To determine the presence of a common bacterial core, we looked for genera present in at least 50% of subjects in either group. The common core of both groups was characterized by 40 genera, 34 of which were shared (Figure 3A).

Relative Abundance of Bacterial Genera

The most abundant genus for both HFrEF patients and healthy controls was Bacteroides, with a mean relative abundance of 30.5% and 24.2%, respectively. The dominance of Bacteroides was again consistent with previous findings.³¹

The 20 most abundant genera for patients and healthy controls accounted for 79.58% and 84.45%, respectively, of the total genera abundance in the 2 groups (Figure 2B, Table S2); 17 of these 20 genera (85%) were shared by both groups. A comparison of the relative abundances of the common core genera between the HFrEF patient and healthy control groups revealed that Streptococcus and Veillonella were enriched, while SMB53 was depleted in patients (adj. P<0.05, Figure 4, Supplementary Results & Supplementary Discussion).

Figure 4.

Significant differences in common core genera. Genera of the common core microbiota with significantly different relative abundance (Y axis) in HFrEF patients (P) and healthy controls (HC). The boxes represent the range from 1st to 3rd quartile, the lines inside the boxes represent the median, and dots represent outlier values. False discovery rate-adjusted P<0.05 for all. HFrEF, heart failure with reduced ejection fraction.

Microbial Diversity in HFrEF Patients and Healthy Controls

The Shannon index, the Chao 1 estimator, and the Simpson index assessing α diversity were not significantly different between groups (P=0.750, P=0.148, and P=0.354, respectively). The richness and evenness of bacterial genera were also not significantly different (P=0.569 and P=0.961, respectively). Collectively, these data showed that on average, subjects of both groups harbored bacterial communities of similar richness and distribution.

In contrast, ordination by Principal Coordinates Analysis of the unweighted UniFrac distance metric-based results for β diversity, which represents interindividual variation that accounts for the overall change in ecological communities, revealed a pattern of group separation in 3D space (Figure 5). This compositional variation that apparently distinguished patients and healthy controls was verified by ANOSIM based on the UniFrac data (999 permutations, R=0.09, P=0.029), showing that, although the size of this variation was not large, the bacterial communities of each group were more similar to each other than they were to communities of the other group. Similarity percentage (SIMPER) one-way analysis on a genus level Bray-Curtis dissimilarity matrix showed that the average dissimilarity between patients and healthy controls was 53.54%, with 9 genera explaining half of the cumulative difference (Table S3). PERMANOVA on Bray-Curtis similarities based on 4th root transformed relative abundances of bacterial genera also uncovered the difference between groups (9,999 permutations, P=0.040).

Figure 5.

Beta diversity. Principal Coordinates Analysis ordination of the unweighted UniFrac distance metric-based results for β diversity between HFrEF patients and healthy controls. The more similar the samples, the closer they are plotted to each other. An incomplete pattern of group separation in the first 2 components of variation is apparent. HFrEF, heart failure with reduced ejection fraction.

Multivariate Analysis of Bacterial Community Composition and Host-Related Metadata

Multivariate analysis using Spearman’s rank correlation and performing 9,999 permutations assessed how well the HFrEF and healthy host-related data variables described in Supplementary Methods, such as age, nutrients, cardiac index, right atrial pressure, estimated glomerular filtration rate (eGFR), and medication, explained the variation in microbial data. The analysis did not show any significant correlations.

Functional Metagenomics

To extend the insight into metabolic pathway information carried by intestinal bacteria in the HFrEF patient and healthy control samples, we constructed functional metagenomic datasets of Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways and KEGG Ortholog (KO) abundances using PICRUSt. We predicted a total of 136 bacterial metabolic functions, categorized into 12 classes (Tables S4,S5). Gene families related to metabolism accounted for 48.9% and 48.4% of the total predicted gene count for patients and healthy controls, respectively. The 10 most abundant metabolic pathways were the same for bacterial communities of both groups, depicting even distribution among subjects (Figure 6 Top). Correlations-based Principal Components Analysis (PCA) of relative KO abundances of predicted metabolic functions showed adequate separation of subjects based on health status across the first 2 components (Figure 6 Bottom), highlighting differences in pathway variation between groups. The overall difference in metabolic pathways between groups was verified by PERMANOVA on Bray-Curtis similarities of 4th root transformed relative gene family counts between samples (9,999 permutations, P=0.009). An FDR-adjusted comparison of relative KO abundances between HFrEF patients and healthy controls revealed numerous differentially abundant pathways in amino acid, carbohydrate, vitamin, and xenobiotic metabolism (Table S6).

Figure 6.

Metabolic pathway variation. (Top) Stacked bar chart of the 10 most abundant bacterial metabolic pathways in HFrEF patients and healthy controls, based on PICRUSt-predicted KO relative abundances. Their similar prevalence and even distribution are more than apparent, in stark contrast with the relative abundances of bacterial genera seen in Figure 2B. Beyond these pathways, however, the analysis revealed significant variation between patients and healthy controls. The scale of relative abundance was adjusted to reflect only the 10 most abundant pathways, excluding all others. Vertical bars represent individual samples. (Bottom) Principal Components Analysis ordination of KO relative abundances of all PICRUSt-predicted metabolic pathways in bacterial communities of HFrEF patients and healthy controls, showing a separation of samples by health status in the first 2 components of variation, which was verified by PERMANOVA. HFrEF, heart failure with reduced ejection fraction; KO, Kyoto Encyclopedia of Genes and Genomes (KEGG) Ortholog; PERMANOVA, permutational multivariate analysis of variance; PICRUSt, phylogenetic investigation of communities by reconstruction of unobserved states.

Microbial interaction Networks

Networks of microbial associations, inferred by SPIEC-EASI, revealed differential correlation patterns in the HFrEF and healthy control groups (Supplementary Results & Supplementary Discussion, Figure S1).

Comparison With Non-HF Controls

To control for the differences in age, medication, and eGFR between HFrEF patients and healthy controls, we also analyzed available data from a total of 18 consecutive patients with CAD or arrhythmia that were not in HF status (non-HF controls). We calculated propensity scores and matched HFrEF patients and non-HF controls by age, eGFR, and β-blocker, angiotensin-converting enzyme inhibitor (ACEI)/angiotensin II receptor blocker (ARB), and proton-pump inhibitor (PPI) administration, on a 1:1 ratio (Table S7). The analysis resulted in a subset of 18 HFrEF patients and the whole group of 18 non-HF controls, who were matched for age, eGFR, ACEI/ARB, and PPI frequencies, and partially matched for β-blockers. Moreover, there were no differences in frequencies of consumption among the food categories between matched HFrEF patients and non-HF controls, and the nutrient intake patterns estimated from food records were the same. Alpha diversity indices were not different between groups (H, P=0.504, Chao 1, P=0.613), but PERMANOVA on Bray-Curtis similarities based on 4th root transformed relative abundances of the bacterial genera of these 2 groups showed a significant difference (9,999 permutations, P=0.045). This indicated that even after adjustment for age, eGFR, and medication, there were still differences in β diversity between subjects with and without HF.

Discussion

In the present study, we employed 16S rRNA sequencing, a well-established, accurate, reliable, and reproducible method when properly used,³²^–³⁴ to show that the intestinal bacterial communities of non-ischemic HFrEF patients differed from those in subjects without HF. This difference was based on a phylogenetic shift in community composition that extended to predicted information on metabolic pathways and differential network interactions of examined bacterial taxa.

Microbial Dysbiosis

Dysbiosis describes structurally and/or functionally altered states of microbiota as compared with homeostasis. To our knowledge, only 2 other studies using 16S rRNA analysis of intestinal microbiota in HF patients have been indexed in PubMed,¹⁹^,³⁵ and no other predicting functional information or examining network interactions. As with previous reports, our study revealed microbial dysbiosis in HFrEF.

Microbial variation between our groups was primarily assessed with indices of α and β diversity. Whereas assessment of α diversity is roughly the equivalent of counting animal species in areas of a large forest, averaging results, and then comparing the results to those from a different forest (a type of univariate analysis), assessment of β diversity is roughly the equivalent of counting numbers of animal species in different areas of a forest, assessing what is common, and directly comparing the areas to each other and to areas of a different forest (a type of multivariate analysis). Beta diversity assessment was based on the presence or absence of taxa, as shown by the unweighted UniFrac distance metric, which importantly accounts for bacterial phylogenetic relationships among taxa and substantiates phylogenetic shifts. It also reflected taxa abundance, as shown by the Bray-Curtis dissimilarity. Common core microbiota displayed few and subtle differences between groups, suggesting that variation in key bacterial abundances is downplayed in HFrEF-related dysbiosis, and underlies the importance of a systems approach rather than a reductionist approach in addressing ecological variation, focusing on diversity and multivariate analysis rather than individual alterations.

Interestingly, Akkermansia, a common bacterial genus in the human colon, was absent from the healthy controls (Table S1). Although the universal presence of Akkermansia in all human hosts is not the case, it is also possible that this genus was present in at least some healthy controls, but in very low relative abundance, below the 0.1% threshold that was used for taxonomic classification in our study.

At present, no causal effects of dysbiosis on HFrEF pathophysiology can be supported, although evidence suggests that microbial products are associated with HF mortality,³⁶ justifying future options, such as fecal transplantations in laboratory animals, to examine functional effects of HFrEF-associated variation.

Correlations of Bacterial Composition With Host Metadata

Intriguingly, multivariate analysis of the relationship of host metadata and microbiotic data did not uncover significant correlations that would explain part or even the whole of the variation in bacterial communities. This is consistent with previous work,¹⁹ and implies that factors as yet undiscovered may play a role in shaping microbial communities in HFrEF patients, and that proxy variables of tissue hypoxia and congestion including LVEF, cardiac index, and right atrial pressure, may not be good correlates with bacterial abundance patterns.

Measurements of blood flow in the mesenteric arteries, as well as intestinal edema markers, such as bowel wall thickness, in ultrasound examinations, represent interesting tools for further examination of how host-related pathophysiology may be linked to 16S rRNA and metagenomic explorations of the intestinal microbiota.

Bacterial Metabolic Profiles

The metabolic functions of intestinal bacteria represent a major target in microbial research. Bacterial products, such as short-chain fatty acids and vitamins, affect the human host in multiple ways and bacterial metabolic functions may contribute to human health or predispose to disease. In the absence of shotgun sequencing data we used PICRUSt, a technique that adequately and cost-effectively predicts metagenomes (i.e., all genes from all organisms present in a sample) from 16S rRNA data and a reference genome database to evaluate the functional pathways of bacterial communities and assess their metabolic capabilities. The analysis is based on a complex algorithm that first computes gene content for each microorganism in a reference phylogenetic tree and then combines this result with the abundance of 16S rRNA genes of the microorganisms found in each sample, to arrive at the expected abundance of gene families in the bacterial community. In simple terms, the measured bacterial 16S rRNA gene abundances are transformed into predicted abundances of orthologous gene families attributed to different functional pathways. The final output is a table of gene counts per sample per predicted functional pathway, which can be standardized (i.e., divided by the total counts in each sample) to finally get the relative gene counts per sample for each of the predicted pathways. The predicted pathways are derived from the KEGG database, which can be used to link genes included in an organism’s genome to cellular metabolic pathways. This enables comparison of metabolic capacities between different organisms and different environments.

Although the present results should be considered suggestive, with an accuracy of approximately 85%, and do not constitute gene expression data, the discovered variation between patients and healthy controls hints at altered bacterial metabolic capacities in the gut of HFrEF patients. More specifically, the overall pattern revealed by PCA and verified by PERMANOVA provided evidence that HFrEF is associated with distinctions in relative gene abundances linked to bacterial metabolism, as compared with health. Metabolic pathway variation, of which the relative gene pool was significantly different in statistical analysis after FDR adjustment (Table S6), did not seem associated with marked adverse change that could affect host functions, with 1 notable exception: the depletion in patients of gene families associated with pantothenate synthesis. Pantothenate is the anion form of pantothenic acid (vitamin B5), an essential element of life across animal organisms, being a component of coenzyme A. As its Greek name implies, it is available in all sources of food; however, its content in food is reduced with food processing, freezing, cooking, and exposure to vinegar, the use of which is widespread in culinary preparations, especially in Japan. Although speculative, a diminished synthesis by intestinal bacteria in HFrEF may have implications for subjects already at risk for deficiency (e.g., the more elderly patients or patients with unbalanced dietary habits). Future work in this field may help uncover additional differences and verify metabolic distinctions in larger samples. With PICRUSt providing evidence for bacterial metabolic pathway variation in HFrEF, further research with admittedly more expensive techniques of deeper sequencing levels may be reasonable.

Confounders of Variation

Age, ethnicity, renal function, and diet are known factors affecting intestinal microbial composition and may confound results if not properly addressed.¹^–³^,¹² Subjects included in this study were all Japanese, belonged to an adult group aged 18–70 years, had preserved renal function, and were consuming foods typically found in a common Japanese diet, with vegetarian diets being an exclusion criterion and estimated nutrient intake being similar in groups. Concerning age, especially, most research shows changes in intestinal microbial composition after a subject has passed the threshold of being elderly and much research shows that intestinal microbial communities remain relatively stable throughout adult life and before the stage of old age, if they are not seriously perturbed by, for example, antibiotic administration.¹^,⁴^,³¹

Study Limitations

Limitations in this study include the small sample sizes, which reflect the strict exclusion criteria and sampling protocol to ensure sample quality and avoid contamination involved in 16S rRNA analysis, and the inherent inability to control for all confounders of variation in observational research. However, previous studies have included even smaller numbers of subjects and a comparison between groups of such sizes in HF and other disorders has previously yielded significant results.¹²^,¹³^,¹⁹ A further limitation of this study was the inability to match HFrEF and healthy controls by age, renal function, and medication. However, a comparison of HFrEF patients to age-, eGFR-, and medication-matched non-HF controls showed that differences in β diversity still remained between subjects with and without HF. Finally, sampling bias during recruitment cannot be completely excluded, particularly because patients who were admitted to the department of cardiovascular medicine but quickly transferred to other departments (e.g., department of cardiovascular surgery, intensive care unit) were generally not screened or sampled for logistic reasons.

Conclusions

We have found significant alterations in the intestinal bacterial communities of non-ischemic HFrEF patients on multiple dimensions of variation, including a basis of structure and extensions of metagenomic carriage of metabolic pathways and interactions of bacterial taxa. These findings may help lay the foundation for further research into HFrEF-related microbial dysbiosis.

Acknowledgments

We thank Dr. Dan Knights, Assistant Professor in the Department of Computer Science and Engineering at the University of Minnesota, and Gavin Douglas, MSc from the Langille Lab at Dalhousie University, for helpful comments on PICRUSt analysis.

Funding

Grant-in-aid from the Japan Society for the Promotion of Science (16K09433), Daiichi-Sankyo research scholarship.

Supplementary Files

Supplementary File 1

Supplementary Methods

Supplementary Results

Supplementary Discussion

Figure S1. Microbial interaction networks showing different correlation patterns at the genus level between bacterial communities of HFrEF patients and healthy controls.

Table S1. Bacterial genera found only in HFrEF patients or only in healthy controls

Table S2. The 20 most abundant genera in HFrEF patients and healthy controls

Table S3. Similarity percentage (SIMPER) analysis of untransformed genus level Bray-Curtis similarities showing the top 9 genera contributing 50% of the average dissimilarity between HFrEF patients and healthy controls

Table S4. Classes of PICRUSt-predicted metabolic functions in sampled bacterial communities

Table S5. Ratio of PICRUSt-predicted counts of metabolism-related gene families by metabolic function and subject group

Table S6. KEGG metabolic pathways differentially abundant in HFrEF patients compared with healthy controls

Table S7. HFrEF patients and non-HF controls matched by age, eGFR, and medications, and summary of propensity score matching

Table S8. Significant differences in baseline characteristics of HFrEF patients in NYHA classes I–II and NYHA classes III–IV

Please find supplementary file(s);

http://dx.doi.org/10.1253/circj.CJ-17-1285

References

1. Clemente JC, Ursell LK, Parfrey LW, Knight R. The impact of the gut microbiota on human health: An integrative view. Cell 2012; 148: 1258–1270.
2. Turnbaugh PJ, Ridaura VK, Faith JJ, Rey FE, Knight R, Gordon JI. The effect of diet on the human gut microbiome: A metagenomic analysis in humanized gnotobiotic mice. Sci Transl Med 2009; 1: 6ra14.
3. Yatsunenko T, Rey FE, Manary MJ, Trehan I, Dominguez-Bello MG, Contreras M, et al. Human gut microbiome viewed across age and geopraphy. Nature 2012; 486: 222–227.
4. Faith JJ, Guruge JL, Charbonneau M, Subramanian S, Seedorf H, Goodman AL, et al. The long-term stability of the human gut microbiota. Science 2013; 341: 1237439.
5. Qin J, Li R, Raes J, Arumugam M, Burgdorf KS, Manichanh C, et al. A human gut microbial gene catalogue established by metagenomic sequencing. Nature 2010; 464: 59–65.
6. Guarner F, Malagelada JR. Gut flora in health and disease. Lancet 2003; 361: 512–519.
7. Rajilić-Stojanović M, Heilig HG, Molenaar D, Kajander K, Surakka A, Smidt H, et al. Development and application of the human intestinal tract chip, a phylogenetic microarray: Analysis of universally conserved phylotypes in the abundant microbiota of young and elderly adults. Environ Microbiol 2009; 11: 1736–1751.
8. Lynch SV, Pedersen O. The human intestinal microbiome in health and disease. N Engl J Med 2016; 375: 2369–2379.
9. Maurice CF, Haiser HJ, Turnbaugh PJ. Xenobiotics shape the physiology and gene expression of the active human gut microbiome. Cell 2013; 152: 39–50.
10. Ley RE, Turnbaugh PJ, Klein S, Gordon JI. Microbial ecology: Human gut microbes associated with obesity. Nature 2006; 444: 1022–1023.
11. Manichanh C, Borruel N, Casellas F, Guarner F. The gut microbiota in IBD. Nat Rev Gastroenterol Hepatol 2012; 9: 599–608.
12. Vaziri ND, Wong J, Pahl M, Piceno YM, Yuan J, DeSantis TZ, et al. Chronic kidney disease alters intestinal microbial flora. Kidney Int 2013; 83: 308–315.
13. Karlsson FH, Fåk F, Nookaew I, Tremaroli V, Fagerberg B, Petranovic D, et al. Symptomatic atherosclerosis is associated with an altered gut metagenome. Nat Commun 2012; 3: 1245.
14. Rogler G, Rosano G. The heart and the gut. Eur Heart J 2014; 35: 426–430.
15. Sandek A, Rauchhaus M, Anker SD, von Haehling S. The emerging role of the gut in chronic heart failure. Curr Opin Clin Nutr Metab Care 2008; 11: 632–639.
16. Sandek A, Bauditz J, Swidsinski A, Buhner S, Weber-Eibel J, von Haehling S, et al. Altered intestinal function in patients with chronic heart failure. J Am Coll Cardiol 2007; 50: 1561–1569.
17. Sandek A, Swidsinski A, Schroedl W, Watson A, Valentova M, Herrmann R, et al. Intestinal blood flow in patients with chronic heart failure: A link with bacterial growth, gastrointestinal symptoms, and cachexia. J Am Coll Cardiol 2014; 64: 1092–1102.
18. Pasini E, Aquilani R, Testa C, Baiardi P, Angioletti S, Boschi F, et al. Pathogenic gut flora in patients with heart failure. JACC Heart Fail 2016; 4: 220–227.
19. Luedde M, Winkler T, Heinsen FA, Rühlemann MC, Spehlmann ME, Bajrovic A, et al. Heart failure is associated with depletion of core intestinal microbiota. ESC Heart Fail 2017; 4: 282–290.
20. Taniguchi T, Sakata Y, Ohtani T, Mizote I, Takeda Y, Asano Y, et al. Usefulness of transient elastography for noninvasive and reliable estimation of right-sided filling pressure in heart failure. Am J Cardiol 2014; 113: 552–558.
21. Caporaso JG, Kuczynski J, Stombaugh J, Bittinger K, Bushman FD, Costello EK, et al. QIIME allows analysis of high-throughput community sequencing data. Nature Methods 2010; 7: 335–336.
22. Benjamini Y, Hochberg Y. Controlling the false discovery rate: A practical and powerful approach to multiple testing. J R Statist Soc Ser B 57 1995; 1: 289–300.
23. Shannon CE. A mathematical theory of communication. Bell Syst Tech J 1948; 27: 379–423.
24. Chao A. Estimating the population size for capture-recapture data with unequal catchability. Biometrics 1987; 43: 783–791.
25. Simpson EH. Measurement of diversity. Nature 1949; 163: 688.
26. Bray JR, Curtis JT. An ordination of upland forest communities of southern Wisconsin. Ecol Monogr 1957; 27: 325–349.
27. Lozupone C, Knight R. UniFrac: A new phylogenetic method for comparing microbial communities. Appl Environ Microbiol 2005; 71: 8228–8235.
28. Clarke KR. Non-parametric multivariate analyses of changes in community structure. Austral Ecol 1993; 18: 117–143.
29. Langille MG, Zaneveld J, Caporaso JG, McDonald D, Knights D, Reyes JA, et al. Predictive functional profiling of microbial communities using 16S rRNA marker gene sequences. Nat Biotechnol 2013; 31: 814–821.
30. Kurtz ZD, Müller CL, Miraldi ER, Littman DR, Blaser MJ, Bonneau RA. Sparse and compositionally robust inference of microbial ecological networks. PLoS Comput Biol 2015; 11: e1004226.
31. Human Microbiome Project Consortium. Structure, function and diversity of the healthy human microbiome. Nature 2012; 486: 207–214.
32. Clarridge JE 3rd. Impact of 16S rRNA gene sequence analysis for identification of bacteria on clinical microbiology and infectious diseases. Clin Microbiol Rev 2004; 17: 840–862.
33. Drancourt M, Bollet C, Carlioz A, Martelin R, Gayral JP, Raoult D. 16S ribosomal DNA sequence analysis of a large collection of environmental and clinical unidentifiable bacterial isolates. J Clin Microbiol 2000; 38: 3623–3630.
34. Wen C, Wu L, Qin Y, Van Nostrand JD, Ning D, Sun B, et al. Evaluation of the reproducibility of amplicon sequencing with Illumina MiSeq platform. PLoS One 2017; 12: e0176716.
35. Kamo T, Akazawa H, Suda W, Saga-Kamo A, Shimizu Y, Yagi H, et al. Dysbiosis and compositional alterations with aging in the gut microbiota of patients with heart failure. PLoS One 2017; 12: e0174099.
36. Tang WH, Wang Z, Fan Y, Levison B, Hazen JE, Donahue LM, et al. Prognostic value of elevated levels of intestinal microbe-generated metabolite trimethylamine-N-oxide in patients with heart failure: Refining the gut hypothesis. J Am Coll Cardiol 2014; 64: 1908–1914.

Corresponding author