Microbiome changes in patients with chronic heart failure with preserved ejection fraction correlate with fibrosis markers: Description of a Russian cohort

Novoselov Alexey Lvovich, PhD*, Kaburova Anastasia Nikolaevna, Pokrovskaya Maria Sergeevna, PhD, Zaitchik Boris Tsaler’evich, PhD, Poyarkov Stanislav Vladimirovich, Zagainova Angelika Vladimirovna, PhD, Fedec Zlata Evgenievna, Mania Tamara Rizoevna, Khripach Lyudmila Vasilievna, PhD, Knyazeva Tatyana Dmitrievna, Kirilov Andrey Vadimirovich, Aksenova Marina Gennadevna, PhD, Makovezkaya Anna Kirillovna, PhD, Makarov Valentin Vladimirovich, PhD, Drapkina Oxana Mikhailovna, Prof, and Yudin Sergey Mikhailovich, Prof


Clinical Perspective
WHAT IS NEW?
 Patients with heart failure with preserved ejection fraction have depleted microbiome with impaired biochemical function.
 The most prominent effect on the host physiology could be caused by the significant decrease in species of the genus Faecalibacterium

WHAT ARE THE CLINICAL IMPPLICATIONS?
 Our data support the previously suggested approach of microbiome treatment as one of the possible targets for the development of new therapeutics for patients with heart failure with preserved ejection fraction 6 Chronic heart failure (CHF) is a pathological condition characterized by ventricle remodeling and impaired heart function. 1 It affects approximately 26 million people worldwide, and the 5-year mortality rate after CHF diagnosis remains at approximately 50 %. 2,3 Although the number of CHF patients has been estimated to increase between 2012 and 2030, there is still no effective treatment available. Considering that CHF care is extremely costly, new therapeutic approaches are urgently needed. 4 CHF has two distinct phenotypes: heart failure (HF) with reduced ejection fraction and HF with preserved ejection fraction (HFpEF). 5 These two phenotypes have completely different natural history, pathophysiological mechanisms, and underlying molecular processes 5,6 .
Clinical trials have thus far focused solely on HFpEF. Comorbidities such as arterial hypertension, diabetes mellitus, and obesity, which cause increased left ventricular load, are commonly observed in patients with HFpEF. 7,8 Prolonged ventricular overload generates mechanic stimuli that together with neurohormonal signals promote structural protein production in cardiomyocytes, hypertrophic growth, and ventricle remodeling. 1 Myocardial hypertrophy is accompanied by fibrosis that develops as an adaptation to increased cardiomyocyte stiffness. 9,10 Myofibroblasts are predominantly responsible for cardiac fibrosis. 11,12 However, the origin of the cell types responsible for fibrosis development remains unknown; nonetheless, under hypertrophy, quiescent fibroblasts have been shown to be replaced by proliferative fibroblasts, which transform the myocardial matrix to abnormal structures. 13,14 The emergence of pathological fibroblasts is governed by growth factors and inflammatory cytokines. [15][16][17][18] A recent study showed that the gut microbial community plays a pivotal role in CHF pathogenesis. 19 A new hypothesis of CHF pathogenesis posits that the leaky gut phenotype 7 leads to the translocation of bacterial lipopolysaccharides to the bloodstream and promotes the systemic inflammatory state. 20,21 Furthermore, the gut microbial community has been shown to produce numerous metabolites such as short-chain fatty acids (SCFAs), trimethylamine Noxide (TMAO), and secondary bile acids, which are important metabolic regulators. 22 Several clinical studies have identified differences in the microbiome composition of CHF patients and healthy individuals. [23][24][25] However, how these microbiota changes affect cardiac fibrosis is not yet known. 26 To address these issues, we conducted a clinical trial involving HFpEF patients to identify the changes in the gut microbiome composition and the correlation of fibrosis markers with the microbial taxa.

Data Availability
Source data are available at https://github.com/NovoselovAL/CHF. More data are available upon reasonable request.

Study Cohort
The present clinical trial focused on HFpEF; confounding effects of other diseases were avoided by including only HFpEF patients in the study cohort. HFpEF patients (n = 47) enrolled in this trial were in the age range of 18 to 80 years and had CHF symptoms, left ventricular ejection fraction (LVEF) of ≥50 %, mitral inflow to annulus ratio (E/e) of ≥13, and N-terminal probrain natriuretic peptide level of >125 pg/ml. In addition, myocardial fibrosis was assessed using T1 myocardial mapping with cardiac magnetic resonance. The control group (n = 43) consisted of healthy individuals without CHF symptoms, in whom echocardiography with tissue Doppler study confirmed LVEF of ≥50 % and E/e of <8. The following were the 8 exclusion criteria: class II obesity (BMI, ≥35 kg/m 2 ), chronic kidney disease (GFR, ˂30 ml/min/1.73 m 2 estimated using CKD-EPI), diabetes mellitus, smoking, history of myocardial infarction and allergic reaction to gadolinium contrast agents, chronic obstructive pulmonary disease, asthma, 2-week history of infection, cardiomyopathy, constrictive pericarditis, heart defect (including congenital heart conditions), autoimmune diseases, inflammatory bowel disease, cancer, 3-month history of probiotics or systemic glucocorticoid use, and pregnancy.
The anthropometric parameters of the study groups are shown in Supplementary materials, S5.
The following biomaterial was collected from study participants for microbiome research: blood samples (n = 90) and stool samples (n = 90). Stool samples were collected at the homes of participants. Any perturbation of gut microbiota was avoided by preserving the stool samples in care medium, the medium composition will be specify somewhere else. All patients read and signed the informed consent form. Ethics committee of each participating institute approved the study protocol. The study was conducted by the Federal State Institution

DNA Extraction, Library Preparation, and 16S rRNA Gene Sequencing
Service company "Genotek" (Russian Federation) extracted DNA from the fecal samples, prepared libraries, and conducted sequencing. In brief, DNA was extracted from the stool samples and subjected to 16S rRNA gene fragment amplification. Sequencing was conducted on an Illumina MiSeq system by using MiSeq Reagent Kit V2 Nano (PE 250), according to manufacturer's protocol. 9

Bioinformatic Analysis
We applied the in-house automated pipeline for data processing. Quality check was conducted using FASTQC and MULTIQC tools. Low-quality reads were removed, and primers were trimmed using QIIME2 cutadapt. For chimer checking and amplicon sequence variant (ASV) picking, the algorithm QIIME2 DADA2 was used with the reference database HITdb v1.00.
The biochemical functions of gut microbiota were predicted by simulating metagenomes by using the PICRUSt2-QIIME2 plug in.

SCFA Measurement
Homogenized feces (3 g) were mixed with 30 ml of chloroform:methanol (2:1) solution. After the sample was dispersed in solution, 3.3 ml of deionized H2O was added, and the mixture was vigorously shaken. The mixture was then placed into a funnel and left at +4ºC for 3 h for phase separation. The clean lower phase was collected and dried on a rotary evaporator. The pellet was dissolved in 1 ml of hexane, followed by hydrolysis with 3 N methanolic-HCl (Supelco) at 90°C for 1 h. The samples were analyzed using GS-MS QP2010 Ultra (Shimadzu), under the following conditions: detector temperature, 200°C; detection range, 45 to 450 m/z.
Chromatographic separation was achieved on capillary column MDN-5 (internal diameter, 30 m × 0.25 mm; Supelco), with helium as the carrier gas. The flow rate was 35.6 cm/s (0.9 ml/min), and split ratio was 4:1.

RNA Extraction and cDNA Synthesis
Total RNA was extracted from the serum by using the TRIzol LS reagent (Invitrogen; Thermo Fisher Scientific, Inc., Waltham, MA, USA), according to manufacturer's protocol. In brief, 250 µl of serum samples were homogenized in 750 µl TRIzol reagent, and 200 µl of chloroform was added to each sample. All samples were centrifuged at 12 000 × g for 15 min at 4°C to separate the mixture. RNA in the aqueous phase was precipitated with 0.5 ml isopropyl alcohol and 1 µl glycogen at room temperature for 10 min. After centrifugation, the pellets (i.e., RNA) were washed with 75 % ethanol. Air-dried RNA pellets were diluted in 100 µl of RNA-free water.
The RT reactions were conducted using the following program: 16°C for 30 min, followed by 42°C for 30 min, and termination by heating at 85°C for 5 min, by using 1 µl MultiScribe™ Reverse Transcriptase.

Quantitative Polymerase Chain Reaction
Quantitative polymerase chain reaction (qPCR) was conducted using StepOnePlus (Applied

Statistical Analysis
Data were analyzed using R software. Whether the relative abundances ASVs differed between the two groups was determined using t-test, with significance level of P < 0.05. Other metrics such as Shannon index, TMAO level, SCFA level, and miRNA content were evaluated using Mann-Whitney U-test. The associations between various parameters such as those between the relative abundances of ASV and SCFA levels, cytokine levels, and miRNA content were assessed using Kendall test with significance level of P < 0.05. Whether metagenomes differed between the two groups was evaluated using t-test, with significance level of P < 0.05. The source code of R scripts is available at https://github.com/NovoselovAL/CHF.

RESULTS
The sequencing of 90 samples resulted in a library of 10 42 844 high-quality reads, with mean sequencing depth of 11587 ± 3719 reads per sample. We identified 181 ASVs at the genus level. Microbial community richness was estimated using Shannon index and was found to be significantly low (Mann-Whitney U-test) in the samples of patients with CHF ( Figure 1A).
Similar results have been obtained in clinical studies in Germany, China, and Japan. [23][24][25] Hierarchical clustering of the samples resulted in the grouping of the HF-pEF patients into several large groups. Each group had a unique pattern of microbiota ( Figure 1B). Certain microbiota patterns have been associated with some disorder conditions. 19 The microbiomes of patients with CHF were markedly different from those of healthy individuals. We found that the levels of Faecalibacterium, Prevotella, and Pseudomonas were significantly reduced (ttest) in the microbiome of patients with CHF ( Figure 1C and 1D). Metagenomic research of human gut microbiomes has identified Faecalibacterium as the common member of the healthy gut microbiota, and its abundance was suggested as a hallmark for a healthy state. 28,29 We found that the levels of Lachnoclostridium, Blautia, Haemophilus, Dorea, Peptococcus, Tyzzerella, and Peptostreptococcus unclassified ASV2 had statistically significantly increased in patients with CHF ( Figure 1C). The increase of Lachnoclostridium and Peptostreptococcaceae species has been shown to be associated with several disorders such as acquired immune deficiency and cardiovascular diseases, respectively. 26,30 Clinical studies have shown that certain bacterial species such as those belonging to Peptostreptococcaceae and Faecalibacterium are associated with the production of several critical microbial metabolites, for example, TMAO and SCFAs, and the abundance of these bacteria decreased in patients with CHF. 25,31 Our results are in good agreement with those of previous studies. Considering that the changes in the abundance of 13 differential bacterial genera could lead to changes in the levels of SCFAs and TMAO, we measured their levels in the samples from the two groups. The following SCFAs were measured: butyric acid, hydroxide pyruvate, isobutyric acid, isovaleric acid, methylbutyric acid, propionic acid, and valeric acid. Patients with CHF had significantly high level of hydroxide pyruvate (Mann-Whitney U-test) and significantly low level of isovaleric, methylbutyric, and propionic acids (Figure 2A). Experiments on animal models have shown that propionic acid has antiinflammatory effects, and decrease of propionic acid in patients with CHF could be associated with chronic inflammation. 32 The level of TMAO was significantly high in patients with HF-pEF ( Figure 2B). High TMAO levels have been shown to be associated with chronic inflammation. 33 Therefore, we tested the level of cytokines. We detected significant (Mann-Whitney U-test) decrease in TNF-α and TGF-β1 and increase in PICP ( Figure 2C) in the serum of patients with HF-pEF. Furthermore, the level of TNF-α decreased, but not of IL-6 ( Figure 2C). This difference in cytokine profile of the patients could be attributed to the difference in their treatment medications having antiinflammatory properties, such as aspirin or atorvastatin (Table II in the Data Supplement). The gut microbiota is known to influence the homeostasis of other organs through the production of fatty acids.
For example, some studies have shown that certain SCFAs affect the expression of miRNAs. 34 Thus, we tested the levels of miRNAs in the plasma. Several clinical studies have revealed that the expression of miRNAs 221-3p and 361-5p significantly decreased in patients with CHF, and these miRNAs were suggested as the marker for CHF. 35 In this study, we detected statistically significant (Mann-Whitney U-test) decrease in the level of miRNAs 361-5p and 193b-3p in patients with HF-pEF and statistically significant increase in miRNA 183-3p ( Figure 2D).
The changes in the composition of gut microbial community could interfere with the microbial biochemical functions and affect the fermentation of SCFAs and TMAO. 36 We run the PICRUSt2 algorithm on our data to estimate the effect of HF-pEF on the gut microbiome biochemical capacity of patients. PICRUSt2 predicted the abundances for 10 500 genes in the gut metagenomes of patients with HF-pEF and healthy individuals. The comparison of metagenomes of healthy individuals and those of patients with HF-pEF revealed 365 downregulated and 75 upregulated genes ( Figure I in the Data Supplement). Surprisingly, no significant difference was found in the level of genes responsible for TMAO production ( Table   I in the Data Supplement). We found statistically significant (t-test) decrease in certain genes responsible for starch, fatty acid, and secondary bile acid metabolism ( Figure 3A). Decrease in the genes baiI, baiF, and baiA could have the most prominent impact on metabolism since these genes encode enzymes that play the most critical roles in the secondary bile acid biosynthesis pathway ( Figure 3B). The effect of gut microbiome on the gut-heart axis was reported to be mediated via secondary bile acids, SCFAs, and the release of miRNAs into the bloodstream. 36 We tested the hypothesis whether the statistically significant difference in the level of certain cytokines, microbial metabolites (SCFAs and TMAO), and miRNA expression could be responsible for the gut microbiome changes. For this, we conducted correlation analysis (Kendall test) to identify statistically significant correlations between certain bacterial genera and cytokines, microbial metabolites, and miRNAs. We calculated the correlation coefficients between bacterial genera and changes in the microbial metabolites. We found that the increase in Blautia had statistically significant (t-test) positive correlation with the increase in hydroxide pyruvate. Decrease of Pseudomonas had a negative correlation with the decrease in isovaleric acid. Increase of Dorea had a positive correlation with valeric acid, which was used as a constant ( Figure II in the Data Supplement). We found statistically significant correlations between increase in hydroxide pyruvate and bacterial genera Clostridium ASV1 and Clostridiales unclassified ASV2. The decrease in isovaleric acid level was associated with the genus Ruminococcus ( Figure 4A). Our data indicated that significant changes in the 15 microbiome are associated with significant changes in SCFA levels in the gut. These experimental data are in good agreement with previously published data. 19,23 We calculated the correlation coefficients between bacterial genera and cytokine level changes. The differential decrease in bacterial genus Pseudomonas had a strong positive correlation with the decrease in TGF β1 in patients with CHF ( Figure III in the Data Supplement). We found strong statistically significant negative correlation between the decrease of Faecalibacterium and increase of PICP. We also found strong positive correlation between the decrease of TGF-β and Lachnospiraceae unclassified ASV1 ( Figure 4B). Bacteria of Lachnospiraceae have been suggested as a risk factor for cardiovascular disease. 37 We found strong statistically significant correlation between the abundances of different

DISCUSSION
Pressure-overload hypertrophy (POH) is typically considered as a cause of hypertrophic growth and myocardial fibrosis. 10,40 The molecular mechanisms underlying myocardial fibrosis are yet not known; however, impaired fibroblast transdifferentiation to pathological hyperproliferative fibroblasts has been commonly observed in animal models. 15,16 Pathological fibroblasts are apoptosis resistant and rapidly replace normal stroma to the pathological matrix, resulting in myocardial stiffness and impaired relaxation. 13,14 Clinical studies indicate that impaired matrix turnover is reflected by the level of the procollagen type I carboxy-terminal propeptide (PICP) and has a strong correlation with fibrosis. 10 The serum of HFpEF patients had an increased level of PICP. 41 Our findings are in line with previously published data, which indicated high level of PICP in HFpEF patients ( Figure 2C). We identified correlations between several bacterial ASVs and fibrosis markers (Table IV in the Data Supplement). Recently, the association of chronic inflammation with HFpEF pathogenesis has been postulated. 42 Clinical studies indicate that patients with CHF have elevated levels of proinflammatory cytokines, particularly TNF-α and IL-6. 43 In our Russian cohort, TNF-α and IL-6 were not elevated ( Figure 2C), probably because the patients had received treatment for HF-pEF. Inflammatory mediators play an important role in LV remodeling, including myocyte hypertrophy and fibrosis. 44 The activation of cardiac immune response is mediated through pathogen-associated molecular pattern (PAMP) or damageassociated molecular pattern (DAMP) receptors, which increase in response to heart tissue injury. PAMP receptors recognize both endogenous signals released by dying or injured myocardial cells, or exogenous signals, which are molecules derived from the gut microbiota. 45 The bacteria-derived pattern of lipopolysaccharides has been proposed as one of the causative agents promoting chronic inflammation. 46 Although the a causative link between gut microbiota and CHF remains illusive, numerous clinical studies have found alterations of gut microbiome in CHF patients. 19,21,[23][24][25] Published data indicate that microbial communities of CHF patients have depleted species richness. [23][24][25] Our Russian cohort of HFpEF patients ( Figure 1A) showed these common features with other studied cohorts. Several metabolites produced by the gut microbiota have been linked to CHF. 19 We observed significantly low level of isovaleric, methylbutyric, and propionic acids in the gut of HFpEF patients ( Figure 2A); however, whether this could contribute to CHF pathogenesis is not yet known. Recently, SCFAs have been mechanistically linked to hypertension, an important condition for CHF development. 48,47 SCFAs alter blood pressure in the host via the FFAR3 and OR51E2 receptors. 47 Clinical studies have revealed a link between CHF and high level of TMAO in American and Norwegian cohorts. , 48 Our observation is consistent with previously published data: TMAO was found to have significantly increased in the Russian cohort ( Figure 2B). Experiments on animal models have shown that TMAO worsened cardiomyocyte contractility. 49 This finding suggests that TMAO may be one of the metabolites contributing to CHF. Host liver and gut microbiota produce TMAO, and evaluating the microbiome contribution to the net TMAO production is important. 19 Analysis of 16S rRNA metagenomic data did not reveal any differences in bacterial genes responsible for TMAO metabolism ( Table I in the Data Supplement); however, we found that the expression of certain genes responsible for secondary bile acid metabolism was significantly decreased ( Figure 3A).
Secondary bile acids are ligands of farnesoid-X-receptors, which mediate apoptosis in cardiomyocytes and contribute to myocardial injury. 50 The presence of fibrosis in HFpEF patients was also confirmed using magnetic resonance imaging studies (Table IV in      The PICRUSt2 annotation included 10 500 genes. A, We identified 440 genes that were differently expressed in the gut metagenomes of heart failure with preserved ejection fraction (HF-pEF) patients. Genes responsible for starch, fatty acid, and secondary bile acid metabolism were significantly decreased (p ˂ 0.05, t-test) in HF-pEF patients. B, The baiI, baiF, and baiA genes responsible for the key steps in secondary bile acid biosynthesis, according to KEGG ontology. Ruminococcus, Clostridium ASV1, Clostridiales unclassified ASV2, and Alistipes and hydroxide pyruvate, isovalerianic acid, and TMAO. There were 15 common bacterial genera.