Dysbiosis of oral bacteriome and mycobiome associated with the severity of heart failure

Abstract

Background

Heart failure (HF) is accompanied by chronic inflammation and metabolic stress, but the relationship between HF severity and the oral microbial ecosystem remains incompletely understood. This study aimed to investigate bacterial, fungal, and predicted functional alterations in supragingival plaque from patients with HF.

Methods

This case–control study enrolled 63 patients with HF and 31 healthy controls (HC). Supragingival plaque samples were profiled by 16S rRNA and ITS sequencing. Community structure, differential amplicon sequence variants (ASVs), bacterial-fungal co-abundance networks, HF severity-associated ASVs and predicted MetaCyc pathways were analyzed using QIIME2, SparCC, weighted LASSO regression, and PICRUSt2. Pathway-ASV correlations were further assessed to explore links between taxonomic and predicted functional shifts.

Results

HF status and NYHA class were among the strongest explanatory factors for oral bacterial and fungal community variation. Differential abundance analysis identified 102 bacterial and 68 fungal ASVs between HF and control groups. Weighted LASSO analysis retained six ASVs associated with the NYHA III-IV phenotype: ASV586 (Geotrichum candidum), ASV238 (Nectriaceae), and ASV182 (Neisseria bacilliformis) showed positive coefficients, whereas Haemophilus (ASV1), Streptococcus (ASV0), and Pseudopropionibacterium (ASV148) showed negative coefficients. A combined 6-ASV score discriminated HF from controls with an AUC of 0.804 and NYHA III-IV from the remaining cohort with an AUC of 0.842. PICRUSt2 identified 165 pathways differing between HF and controls. TCA cycle I was enriched in HF and positively correlated with NYHA class, whereas pyruvate fermentation to butanoate, a butyrate-related fermentative pathway, was enriched in controls and negatively correlated with NYHA class. Robust pathway-ASV correlations linked HC-enriched oral biofilm taxa with HC-enriched functional modules, suggesting functional uncoupling of the supragingival plaque ecosystem in more severe HF.

Conclusions

HF severity is associated with oral bacterial-fungal dysbiosis, characterized by enrichment of opportunistic pathogens, depletion of commensal biofilm organisms and predicted functional shifts involving microbial TCA cycle and fermentative capacity. These findings support an “oral-heart axis” in HF and warrant validation by longitudinal multi-omics studies.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12903-026-08977-1.

Background

Heart failure (HF) is a complex clinical syndrome characterized by impaired exercise tolerance resulting from structural and functional cardiac abnormalities, which trigger fluid overload and compensatory neurohormonal activation [1]. Driven by an aging global population and the rising prevalence of risk factors such as coronary artery disease, hypertension, and diabetes mellitus, the incidence of HF continues to escalate [2, 3]. Despite advancements in guideline-directed medical therapy, the prognosis remains poor: the one-year mortality rate following discharge is approximately 16.5%, while the five-year mortality rate for hospitalized patients exceeds 75% [4, 5].

Chronic systemic inflammation is commonly observed in both HF with reduced ejection fraction (HFrEF) and HF with preserved ejection fraction (HFpEF) [6–9]. This persistent, low-grade inflammatory state drives immune dysregulation, leading to maladaptive processes such as cardiac fibroblast activation, which promotes myocardial fibrosis and exacerbates disease progression [10]. Recent evidence highlights the role of microbiota dysbiosis in sustaining this inflammatory state. In HF patients, intestinal hypoperfusion and gut dysbiosis compromise the mucosal barrier, facilitating the translocation of endotoxins and metabolites—such as Trimethylamine N-Oxide (TMAO) and indoxyl-sulfate—into the systemic circulation, thereby fueling systemic inflammation [11–13].

The oral cavity serves as a significant microbial reservoir that can influence systemic health through multiple pathways. Certain acid-resistant oral bacteria can translocate via saliva to the gut, inducing further dysbiosis. [14]. Moreover, local imbalances in oral microbial communities can destroy tooth-supporting tissues, allowing pathogens and their byproducts to enter the bloodstream directly. This contributes to systemic inflammation, a procoagulant state, and atherosclerotic plaque formation [15–18]. A 13-year longitudinal study revealed that periodontal disease is associated with a significantly increased risk of both HFpEF (HR: 1.35) and HFrEF (HR: 2.19), alongside elevated C-reactive protein (CRP) levels [19]. While the oral microbiome has been implicated in various systemic conditions [18, 20, 21], its specific role in heart failure remains poorly understood.

Advancements in next-generation sequencing (NGS), specifically 16S and Internal Transcribed Spacer (ITS) rRNA sequencing, currently allow for high-resolution identification of bacterial and fungal communities. Furthermore, bioinformatics tools like PICRUSt2 enable the prediction of the functional potential of these microbial ecosystems [22]. Supragingival plaque, a biofilm easily accessible through non-invasive swabbing [23], provides an ideal medium for studying these microbial shifts.

By analyzing supragingival plaque through 16S/ITS rRNA sequencing, this study aims to: (1) comprehensively characterize the oral bacteriome and mycobiome of HF patients compared to healthy controls; (2) identify specific microbial signatures and functional alterations associated with the severity of heart failure; and (3) explore novel diagnostic biomarkers and potential therapeutic targets for the management of heart failure.

Methods

Study subjects

In this cross-sectional study, we enrolled patients who were hospitalized at department of cardiology of Peking Union Medical College Hospital between April 2022 and March 2023, and were diagnosed with HF by at least two cardiologists who were qualified and experienced, according to the 2021 European Society of Cardiology guidelines for heart failure [1]. The specific diagnostic algorithm for HF is detailed in the Supplementary Methods. Demographic, medical history, laboratory test results and echocardiography data were obtained through the electronic medical record system of PUMCH.

The New York Heart Association (NYHA) functional classification was used to evaluate exercise tolerance of HF patients [24, 25]. Details of the NYHA classification are provided in the Supplementary Methods. According to the NYHA class, HF patients were divided into the NYHA I-II subgroup and the NYHA III-IV subgroup. Heart failure classification criteria based on left ventricular ejection fraction (LVEF) are also described in the Supplementary Methods. For healthy controls (HC), subjects without any heart disease or any HF-related symptoms or signs were enrolled. Subjects with one or more of the following conditions were excluded: 1) evidence of acute myocardial infarction; 2) cardiac surgery within the previous 6 months; 3) systemic diseases such as systemic lupus erythematosus and malignant tumors; 4) end-stage renal disease (eGFR < 15 ml/kg/1.73m², CKD-EPI); or 5) administration of antibiotics for more than 3 days in the previous 3 months.

A preliminary assessment of dental health, including the presence of dentures, missing teeth and dental caries, was performed for each subject by direct visual inspection and face-to-face medical history inquiry. Visual discovery of tooth decays, such as white spots, brown stains and cavities, was considered evidence of dental caries [26, 27].

Sample collection, DNA Extraction and sequencing

Supragingival plaque samples were collected from all subjects using sterile swabs. None of subjects had consumed food or water, brushed their teeth or used any oral cleanser within two hours before sampling, and all of them rinsed their mouths with sterile 0.9% sodium chloride solution to remove food residue left in the mouth before sampling [28]. Collected plague samples were frozen at −80℃ within 2 h. Genomic DNA were extracted using the CTAB extraction method. We amplified 16S rRNA and ITS rRNA genes, using the 16S (V3-4 region) and the ITS (1-1F region), respectively. Primers for 16S and ITS amplification were listed in Supplementary Methods. A total of 94 samples (63 HFs and 31 controls) were qualified for 16S library preparation, and 92 samples (61 HFs and 31 controls) were qualified for ITS library preparation. Sequencing libraries were generated using the NEB Next® Ultra™ II FS DNA PCR-free Library Prep Kit (New England Biolabs, USA, Catalog #: E7430L), following manufacturer's recommendations. The library was checked with Qubit and real-time PCR for quantification and bioanalyzer for size distribution detection. Quantified libraries were pooled and sequenced on Illumina NovaSeq 6000, based on the PE250 strategy. All the aforementioned procedures were accomplished at Novogene Bioinformatics Technology Co., Ltd.

Preprocessing, denoising and taxonomic assignment

Raw sequence preprocessing and ASV generation were performed by Novogene following the standard amplicon-sequencing bioinformatics workflow. Paired-end reads were assigned to samples based on their unique barcodes. Barcodes and primers were removed using python v3.6.13. Reads shorter than 20 base pairs (bp) after barcode/primer trimming were discarded using cutadapt v3.3 with –minimum-length 20. Paired-end reads were merged using FLASH v1.2.11 with –min-overlap 10 and –max-mismatch-density 0.2 (http://ccb.jhu.edu/software/FLASH/) [29]. Raw tags were quality-filtered using fastp v0.23.1 with -q 19 and -u 15 [30]. Chimeric sequences were removed using VSEARCH v2.16.0, implementing the UCHIME algorithm, by comparison against SILVA v138.1 (https://www.arb-silva.de/) for 16S sequences and UNITE v8.2 (https://unite.ut.ee/) for ITS sequences [31]. The effective clean reads obtained were denoised into amplicon sequence variants (ASVs) via the DADA2 module in QIIME2 (v202202). ASVs with a total count of less than 5 were filtered out [32]. Taxonomic assignment was performed in QIIME2 using feature-classifier classify-sklearn with reference classifiers based on SILVA v138.1 for 16S and UNITE v8.2 for ITS. Multiple sequence alignment and phylogenetic reconstruction were performed using the default QIIME2 phylogeny workflow. The absolute abundance of ASVs was normalized using a standard of sequence number corresponding to the sample with the least sequences. All subsequent bioinformatic analyses were performed based on the output normalized data. A summary of the bioinformatics parameters is provided in the Supplementary Methods.

Microbiome analysis

Alpha diversity (by Chao1 index) was calculated by QIIME2. Comparison of Chao1 between groups was performed using Wilcoxon rank-sum test. For beta diversity, we employed principal coordinate analysis (PCoA) to obtain principal coordinates and visualize differences among samples based on the Bray–Curtis distance matrix generated by R vegan package (v2.7.3). PERMANOVA/Adonis analysis was performed using vegan::adonis2 with method = 'bray', one explanatory factor per model, and 9,999 permutations to estimate the proportion of variance explained by host factors (R2) and the corresponding statistical significance.

For selection of differential features, ASVs detected in less than 20% of all samples were removed first [33, 34]. Differences in the remaining ASVs between the HF and HC groups were then assessed using the Wilcoxon rank-sum test, with nominal P < 0.05 and Benjamini–Hochberg false discovery rate (BH-FDR) < 0.20 used as the selection criteria.

An inter-kingdom co-abundance network between bacterial and fungal ASVs was constructed from differential ASVs using SparCC implemented through the R package SpiecEasi (v0.1.4). The SparCC model was run with iter = 20, inner_iter = 10, and th = 0.1. To estimate empirical significance, 100 bootstrap/randomized correlation matrices were generated with a fixed random seed of 123. Two-sided empirical P values were calculated as the proportion of bootstrap SparCC correlations whose absolute values were greater than or equal to the corresponding observed absolute SparCC correlation. Empirical P values were adjusted using the Benjamini–Hochberg FDR procedure with the p.adjust() function in the R stats package. Network edges were retained only when the absolute SparCC correlation coefficient was > 0.3 and the adjusted FDR was < 0.05. The SparCC network was visualized using the R packages igraph (v2.2.2), ggplot2 (v4.0.2), ggrepel (v0.9.6), readxl (v1.4.5), and scales (v1.4.0). The 10 most connected ASVs in the network were visualized using a bar chart.

Identification of ASVs associated with HF severity by LASSO analysis

Spearman correlations were calculated between differential ASVs and HF severity indicators, including NYHA class and NT-proBNP. ASVs with absolute Spearman rho > 0.2 and P < 0.1 were selected for downstream least absolute shrinkage and selection operator (LASSO) regression analysis using the R package glmnet (v4.1.10). To evaluate microbial signatures associated with a more severe HF phenotype, we constructed a weighted LASSO logistic regression model to discriminate the NYHA III-IV subgroup from the remaining subjects (HC plus NYHA I-II). Relative ASV abundances were used as input features. The model was fitted with family = 'binomial', alpha = 1, standardize = TRUE, and maxit = 1e6. Model performance was assessed using repeated stratified cross-validation with 10 repeats and 5 folds, using random seed 20,260,330. Within each outer training set, class imbalance was addressed using inverse-frequency class weighting. The regularization parameter was selected by inner fivefold cross-validation using the lambda.min criterion with type.measure = 'auc'. For each outer test fold, predicted probabilities were generated and aggregated to estimate cross-validated discrimination. Performance was summarized by the mean cross-validated area under the receiver operating characteristic curve (AUC), 95% confidence interval (CI), sensitivity, specificity, and precision-recall AUC (PR-AUC). ASVs selected in at least 50% of fitted models were defined as stable features.

For ASVs stably selected by the weighted LASSO model, a composite score was constructed to evaluate their combined discriminatory performance. For each ASV, relative abundance values were log-transformed as log10(abundance + pseudocount), where the pseudocount was defined as one-half of the minimum non-zero relative abundance observed for that ASV. The log-transformed abundance of each ASV was then standardized across samples to generate a z score. Directionality was assigned according to the sign of the mean coefficient from the weighted LASSO model. The discriminatory performance of this composite score was assessed using empirical receiver operating characteristic curves (ROCs) without smoothing. ROC analyses were performed for HF versus HC and for NYHA III-IV versus the rest of the cohort using the R package pROC. The AUC and 95% CI were calculated, and the optimal cutoff was determined using the Youden index, with corresponding sensitivity and specificity reported. To visualize score distributions, boxplots were generated for the HC, NYHA I-II, and NYHA III-IV groups. Between-group differences were assessed using two-sided Wilcoxon rank-sum tests with Benjamini–Hochberg FDR correction.

Functional prediction by PICRUSt2

PICRUSt2 (v2.1.2-b) was used to predict metagenomic functional profiles from the 16S sequencing data. The analysis was performed by Novogene using the default pipeline. Pathway relative abundances were compared between the HF and HC groups using two-sided Wilcoxon rank-sum tests, and multiple comparisons were adjusted using the Benjamini–Hochberg FDR procedure. Pathways with FDR < 0.20 were retained for subsequent clinical correlation analysis. Spearman rank correlation was then performed between the relative abundance of retained pathways and clinical indices.

Pathway-ASV correlations were assessed using the 16S dataset. Pathways nominally correlated with NYHA class at Spearman P < 0.05 were selected and correlated with differential ASVs across matched samples. Spearman rank correlation was calculated for each pathway-ASV pair using cor.test(method = 'spearman', exact = FALSE), and P values were adjusted using the Benjamini–Hochberg FDR method across all tested pairs. Statistically robust pathway-ASV associations were defined as absolute Spearman rho > 0.50 and FDR < 0.05. The retained associations were used for downstream visualization of pathway-ASV interaction patterns.

Statistics and visualization

For descriptive statistics of baselines, the Shapiro–Wilk test was employed to determine the normality of continuous data. Continuous normally distributed data are presented as the mean ± standard deviation (SD). Continuous nonnormally distributed data are presented as the median with interquartile range (IQR). Categorical variables are presented as counts and percentages. To compare differences between two groups, Student’s t-test was applied for continuous normally distributed variables, whereas the Mann–Whitney U test was used to compare continuous nonnormally distributed variables. The chi-square test, corrected chi-square test (in case of 1 ≤ T < 5) and Fisher’s exact test (in case of T < 1) were used for categorical variables. For comparisons among ≥ 3 groups, one-way analysis of variance was applied for continuous normally distributed variables. For post hoc comparison, the Bonferroni test was applied in cases of equal variance and the Tamhane test in cases of unequal variance. R v4.3.1 and RStudio were applied. The R packages ggplot2, ggpubr, pheatmap, vegan, SpiecEasi, glmnet, pROC, igraph, ggrepel, readxl, and scales were used for statistical analysis and visualization. All unspecified function arguments were left at default values.

Results

Study population characteristics

A total of 94 subjects were enrolled in this study, with 31 in the control group and 63 in the HF group. There is no significant difference in age (P = 0.274) and sex (P = 0.11) between the HC and the HF groups (Supplementary Table 1), although participants in the NYHA III-IV subgroup were significantly older than those in the other two subgroups (Table 1). Table 1 shows the statistics of baseline characteristics based on subgroups. Compared to the NYHA I-II subgroup, the NYHA III-IV subgroup exhibited a significantly higher level of NT-proBNP (P < 0.001) and hsCRP (P = 0.003). LDH, which reflected the severity of organ damage, was significantly increased in the NYHA III-IV subgroup (P = 0.019). Additionally, subjects in the NYHA III-IV subgroup had worse renal function (assessed by eGFR, P < 0.001) and lower level of hemoglobin (P = 0.017). In terms of comorbidities, more subjects in the NYHA III-IV subgroup had hypertension (P < 0.001), and there were no significant differences in etiology (P = 0.566) or HF type (P = 0.411) between NYHA I-II and NYHA III-IV subgroups. No significant difference in the presence of dental caries, dentures or missing teeth was found between the HF and HC groups (P = 0.689, Supplementary Table 1), or between the NYHA subgroups (P = 0.436).

Table 1.

Characteristics of the study cohort

	NYHA I-II (N = 33)	NYHA III-IV (N = 30)	HC (N = 31)	P value
Age †	50.88 ± 15.12	62.70 ± 16.85	51.87 ± 10.77	0.008ac
Male‡	21(63.6)	15(50.0)	14(45.2)	0.306
SBP, mmHg†	121.06 ± 16.48	119.30 ± 16.46	NA	0.673
BMI, kg/m2†	26.75(23.26,29.86)	24.87(21.22,27.71)	NA	0.228
Primary aetiology				0.566
IHD‡	7(21.2)	5(16.7)	NA
HTN§	2(6.1)	1(3.3)	NA
DCM§	8(24.2)	7(23.3)	NA
RCM§	0(0.0)	0(0.0)	NA
HCM§	5(15.2)	2(6.7)	NA
VHD§	1(3.0)	2(6.7)	NA
Arrythmia§	7(21.2)	5(16.7)	NA
Other§	3(9.1)	8(26.7)	NA
Classification				0.411
HFrEF§	17(51.5)	15(50.0)	NA
HFmrEF§	6(18.2)	3(10.0)	NA
HFimpEF§	3(9.1)	1(3.3)	NA
HFpEF§	7(21.2)	11(36.7)	NA
Complications
CHD§	9(27.3)	9(30.0)	0(0.0)	0.004bc
MI history§	7(21.2)	4(13.3)	0(0.0)	0.029b
HTN§	11(33.3)	21(70)	2(6.5)	< 0.001abc
HLP§	21(63.6)	17(56.7)	3(9.7)	< 0.001bc
DM§	9(27.3)	13(43.3)	0(0.0)	< 0.001bc
Stroke§	3(9.1)	7(23.3)	0(0.0)	0.012c
AF§	10(30.3)	12(40.0)	0(0.0)	0.001bc
ECHO
LVEF, %†	46.40 ± 13.53	47.93 ± 14.51	NA	0.665
LVEDD, mm†	54.85 ± 10.12	55.53 ± 12.87	NA	0.814
Past History
PCI history§	5(15.2)	2(6.7)	0(0.0)	0.068
CABG history§	0(0.0)	2(6.7)	0(0.0)	0.100
Pacemaker history§	0(0.0)	3(10.0)	0(0.0)	0.030
ICD history§	1(3.0)	2(6.7)	0(0.0)	0.416
Medication
ACEI/ARB§	0(0.0)	4(13.3)	NA	0.099
ARNI§	22(66.7)	12(40.0)	NA	0.034a
β blocker§	26(78.8)	22(73.3)	NA	0.612
MRA§	25(75.8)	19(63.3)	NA	0.283
SGLT2i§	20(60.6)	9(30.0)	NA	0.015a
Lab
NT-proBNP, pg/ml†	656.00(272.50,1959.00)	3120(970.25,5741.50)	NA	< 0.001a
eGFR, ml/min/1.73m2†	94.94 ± 20.43	63.87 ± 30.64	NA	< 0.001a
hsCRP, mg/L†	1.99(1.00,3.24)	4.50(2.76,10.23)	NA	0.003
LDH, U/L†	203.00(175.00,233.00)	229.00(206.00,255.00)	NA	0.019

Continuous, normally distributed variable among the three groups were analyzed by a one-way analysis of variance. The Bonferroni test was applied for post hoc comparisons in cases of equal variance, and the Tamhane test was applied in cases of unequal variance. The Kruskal–Wallis H-test was applied for continuous data that were not normally distributed. Continuous, normally distributed variables between two groups were analyzed by Student’s t-test. The Mann–Whitney U test was applied for data of this type that were not normally distributed. Categorical variables were compared by chi-square test, corrected chi-square test (in case of 1 ≤ T < 5) or Fisher’s exact test (in case of T < 1)

ACEI angiotensin-converting enzyme inhibitors, AF atrial fibrillation, ARB angiotensin receptor antagonist, ARNI angiotensin receptor/neprilysin inhibitor, BMI body mass index, CABG coronary artery bypass graft surgery, CHD coronary heart disease, DCM dilated cardiomyopathy, DM diabetes mellitus, eGFR estimated glomerular filtration rate, HCM hypertrophic cardiomyopathy, HLP hyperlipidemia, HTN hypertension, ICD implantable cardioverter defibrillators, IHD ischemic heart disease, LVEF left ventricular ejection fraction, LVEDD left ventricular end diastolic diameter, MI myocardial infarction, MRA aldosterone receptor antagonist, RCM restrictive cardiomyopathy, PCI percutaneous coronary intervention, SBP systolic blood pressure, SGLT2i sodium-glucose cotransporter-2 inhibitor, VHD valvular heart disease

^† mean ± SD; ^‡ n(%); ^§ median(IQR); NA not available

^a: P < 0.05, NYHA I-II vs. NYHA III-IV. ^b: P < 0.05, NYHA I-II vs. control. ^c: P < 0.05, NYHA III-IV vs. control

Oral microbiota profile of heart failure patients

We obtained a total of 13,112 denoised ASVs from the 16S procession and 16,874 ASVs from the ITS procession. Figure 1A-B shows the genus-level phylogenetic relationships and phylum-level annotations for the 16S and ITS panels, respectively, and Fig. 1C-D shows the mean phylum-level compositions. In the 16S panel, Firmicutes, Proteobacteria, and Actinobacteria were the three most abundant phyla in both the HC and HF groups. In the ITS panel, Ascomycota was the dominant phylum in both groups, and Basidiomycota was also abundant. The rarefaction curves supported the adequacy of the sequencing depth (Supplementary Fig. 1A-B). Alpha diversity assessed by Chao1 index was relatively increased in HF for 16S, but decreased in HF for ITS, although the difference was not significant (Supplementary Fig. 1C-F). PCoA based on Bray–Curtis distances suggested separation of both bacterial and fungal community compositions between patients with HF and HCs (Supplementary Fig. 2A-D). PERMANOVA/Adonis analysis further indicated that HF status and NYHA class were among the strongest explanatory factors for oral microbiome variation (Fig. 1E-F). Other significant host factors included medication use and comorbidities. Detailed Adonis results are provided in Supplementary Table 2–3.

Fig. 1.

Composition and host-factor associations of the oral microbiome. A-B Phylogenetic trees of the top 100 bacterial (A) and fungal (B) genera ranked by mean relative abundance. Branch colors indicate phylum-level taxonomy. C-D Average relative abundance of bacterial (C) and fungal (D) phyla in healthy controls (HC) and patients with heart failure (HF). E–F Host factors associated with bacterial (E) and fungal (F) community composition, assessed by Adonis using Bray–Curtis distances. Bar length represents R², and bar color denotes the host factor category. Only variables with P < 0.05 are shown. OADs, oral anti-diabetic drugs other than metformin and SGLT2 inhibitors; AVB, atrioventricular block; DM, diabetes mellitus; Af, atrial fibrillation; MRA, mineralocorticoid receptor antagonist

Multi-kingdom co-abundance network of differential ASVs

In the Wilcoxon differential abundance analysis, 102 16S-derived bacterial ASVs and 68 ITS-derived fungal ASVs differed between the HF and HC groups under the prespecified thresholds (Supplementary Tables 4 and 5; Supplementary Fig. 3). We then constructed a SparCC correlation network using differential ASVs with matched 16S and ITS data (Fig. 2A). After filtering at absolute SparCC correlation > 0.3 and FDR < 0.05, 52 bacterial ASVs and 17 fungal ASVs were retained as connected nodes. The network contained 76 edges, including 60 bacterial-bacterial, 8 bacterial-fungal, and 8 fungal-fungal connections. Among the top 10 ASVs ranked by node degree, all were bacterial ASVs (Fig. 2B), suggesting that bacterial taxa occupied the most connected positions in this differential co-abundance network. The strongest cross-kingdom associations included a positive correlation between bacterial ASV127 (Neisseria) and fungal ASV14 (Apiotrichum; SparCC rho = 0.440), and a negative correlation between bacterial ASV127 and fungal ASV30 (Aspergillus reticulatus; SparCC rho = −0.437).

Fig. 2.

SparCC co-occurrence network among differentially abundant bacterial and fungal ASVs. A ASV-level co-occurrence network constructed from differentially abundant ASVs between HF and HC. Circles indicate 16S-derived bacterial ASVs, and triangles indicate ITS-derived fungal ASVs. Node colors indicate phyla. Edges were retained at |SparCC correlation|> 0.3 and FDR < 0.05; edge color indicates the direction and magnitude of the SparCC correlation, and edge width is proportional to the absolute correlation strength. The plotted network contains 69 nodes and 76 edges, including 60 bacterial-bacterial, 8 bacterial-fungal, and 8 fungal-fungal edges. B Top 10 ASVs ranked by node degree in the SparCC network, with taxonomic annotations shown to the right of each bar

Oral Microbes associated with heart failure severity

In the Adonis analysis, NYHA was found to be an important factor influencing the structure of the oral microbiome. We next correlated differential ASVs to NYHA and NT-proBNP to identify oral microbes potentially associated with HF severity. A total of 11 bacterial ASVs and 5 fungal ASVs were screened out (Fig. 3A-B), using the criteria of Spearman |rho| > 0.2 and P < 0.1. These ASVs were entered into a weighted LASSO logistic regression model to discriminate the NYHA III-IV subgroup from the rest of the cohort (HC plus NYHA I-II). The LASSO coefficient paths and lambda selection based on cross-validated AUC are shown in Fig. 3C-D. Under repeated stratified cross-validation (10 repeats × 5 folds), the model showed modest discrimination, with a mean cross-validated AUC of 0.630 (95% CI, 0.598–0.661) and a mean PR-AUC of 0.482 (95% CI, 0.435–0.529). The mean sensitivity and specificity were 0.538 (95% CI, 0.509–0.567) and 0.705 (95% CI, 0.676–0.733), respectively. Using a stability threshold of selection frequency ≥ 0.50, six ASVs were consistently retained across the repeated cross-validation procedure (Fig. 3E): ASV586 (Geotrichum candidum), ASV238 (family Nectriaceae), ASV182 (Neisseria bacilliformis), ASV1 (genus Haemophilus), ASV0 (genus Streptococcus), and ASV148 (genus Pseudopropionibacterium). ASV586, ASV238, and ASV182 had positive mean coefficients, whereas ASV1, ASV0, and ASV148 had negative mean coefficients, indicating opposite directions of association with the NYHA III-IV phenotype. A combined 6-ASV score was then constructed as the signed sum of standardized log-transformed abundances: -z_ASV0—z_ASV1—z_ASV148 + z_ASV182 + z_ASV238 + z_ASV586. This composite score discriminated HF from HC with an AUC of 0.804 (95% CI, 0.716–0.893), sensitivity of 0.72, and specificity of 0.77 at the Youden cutoff of −1.177 (Fig. 3F). It also discriminated NYHA III-IV from the rest of the cohort with an AUC of 0.842 (95% CI, 0.762–0.922), sensitivity of 0.93, and specificity of 0.65 at the Youden cutoff of −1.090. Consistently, the combined 6-ASV score was higher in HF than in HC (P = 2.04e-06, FDR = 4.07e-06), in NYHA I-II than in HC (P = 0.007, FDR = 0.007), and in NYHA III-IV than in HC (P = 2.07e-08, FDR = 8.26e-08). The score was also higher in NYHA III-IV than in NYHA I-II (P = 4.14e-04, FDR = 5.52e-04), suggesting that it captured both HF status and functional severity (Fig. 3G-H).

Fig. 3.

Identification and evaluation of ASVs associated with HF severity. A-B Spearman correlations of bacterial (A) and fungal (B) ASVs with NYHA class and NT-proBNP among HF patients. ASVs with |rho|> 0.2 and P < 0.1 are displayed. ASV labels are colored according to their enrichment direction in HF versus HC: red, enriched in HF; blue, depleted in HF. + P < 0.1; * P < 0.05; ** P < 0.01. C Coefficient paths from weighted LASSO logistic regression for discriminating NYHA III-IV from the rest of the cohort (HC plus NYHA I-II). (D) Cross-validated AUC across the lambda sequence; dashed vertical lines indicate lambda.min and lambda.1se. (E) Stable ASVs selected by weighted LASSO across repeated stratified cross-validation (10 repeats × 5 folds). Bars show mean coefficients across models, and labels indicate the mean coefficient and selection frequency. Stable features were defined as ASVs selected in at least 50% of models. (F) ROC curves for the combined 6-ASV score distinguishing HF from HC and NYHA III-IV from the rest of the cohort. AUCs, 95% confidence intervals, Youden cutoffs, sensitivity (Se), and specificity (Sp) are shown in the plot. G-H Distribution of the combined 6-ASV score across HC, NYHA I-II, and NYHA III-IV groups (G), and between HC and HF groups (H). Between-group comparisons were performed using two-sided Wilcoxon rank-sum tests with Benjamini–Hochberg FDR correction

Metagenomic pathways predicted by PICRUSt2 associated with heart failure

We predicted metagenomic pathways using PICRUSt2 based on the MetaCyc database (v25.0) and compared pathway relative abundances between the HF and HC groups using the Wilcoxon rank-sum test. A total of 165 pathways showed differential abundance between groups after multiple-testing correction (FDR < 0.20), including 117 pathways enriched in HF and 48 pathways enriched in HC (Supplementary Table 6). We next examined the associations between these differential pathways and HF-related clinical parameters using Spearman correlation. Eight pathways were associated with NYHA class at nominal P < 0.05 (Fig. 4A). TCA cycle I (prokaryotic) was enriched in HF according to the predefined FDR < 0.20 threshold and was positively correlated with NYHA class (rho = 0.279, P = 0.0269), suggesting that increased microbial TCA cycle potential may be linked to greater HF severity. In contrast, seven HC-enriched pathways were negatively correlated with NYHA class, including glycerol degradation to butanol (PWY-7003; rho = −0.309, P = 0.0138), galactose degradation I (PWY-6317; rho = −0.297, P = 0.0179), superpathway of polyamine biosynthesis II (POLYAMINSYN3-PWY; rho = −0.289, P = 0.0215), pyruvate fermentation to butanoate (CENTFERM-PWY; rho = −0.278, P = 0.0274), superpathway of Clostridium acetobutylicum acidogenic fermentation (PWY-6590; rho = −0.277, P = 0.0279), urate biosynthesis/inosine 5'-phosphate degradation (PWY-5695; rho = −0.257, P = 0.0421), and adenine and adenosine salvage III (PWY-6609; rho = −0.252, P = 0.0465). These exploratory findings indicate that HF severity was associated with higher predicted TCA cycle I abundance and lower abundance of several fermentation-, nucleotide salvage-, and carbohydrate metabolism-related pathways.

Fig. 4.

PICRUSt2-predicted functional pathways associated with HF clinical indicators and bacterial ASVs. A Left, Spearman correlations between eight HF-associated pathways and clinical indicators among HF patients. The eight pathways were differentially abundant between HF and HC by Wilcoxon rank-sum test with FDR < 0.2 and were correlated with NYHA class at P < 0.05. Bubble color indicates Spearman rho, bubble size indicates |rho|, and asterisks indicate P < 0.05. Right, median relative abundance of the same pathways in HF and HC, with Wilcoxon FDR values shown for each pathway. B Left, pathway-ASV correlation network based on Spearman analysis across matched samples. Edges were retained at FDR < 0.05 and |rho|> 0.5; edge color indicates the sign and magnitude of rho, and edge width is proportional to |rho|. The network contains 20 robust associations involving seven pathways and eight ASVs. Right, median relative abundance of the retained ASVs in HF and HC, with Wilcoxon FDR values shown. ASV0, g_Streptococcus; ASV1, g_Haemophilus; ASV11, g_Gemella; ASV25, g_Haemophilus; ASV46, o_Lactobacillales; ASV86, s_Veillonella_sp.; ASV109, g_Alloprevotella; ASV164, s_Dialister_invisus

To further investigate microbial contributors potentially linked to these HF severity-associated pathways, we correlated the relative abundances of the eight key pathways with differential ASVs using 94 samples with matched pathway and ASV profiles. Among 840 pathway-ASV correlation tests, 351 associations reached Spearman FDR < 0.05. After applying a stronger threshold of absolute rho > 0.5, 20 robust pathway-ASV associations remained, involving seven pathways and eight ASVs (Fig. 4B). The strongest positive correlations were observed between galactose degradation I and ASV0 (rho = 0.722, FDR = 9.72e-14) and between polyamine biosynthesis II and ASV0 (rho = 0.722, FDR = 9.72e-14). By contrast, TCA cycle I was negatively correlated with ASV1 and ASV109 (rho = −0.505 and −0.505, respectively; FDR = 9.55e-06). Together, these results suggest that HF severity-associated functional alterations were linked to specific differential ASVs, highlighting potential microbial contributors to pathway-level functional shifts in HF.

Discussion

This study provides a comprehensive characterization of the supragingival plaque microbiome in patients with heart failure (HF), revealing distinct bacterial and fungal signatures compared with healthy controls (HC). The oral cavity is a highly structured microbial ecosystem where early colonizers, fermentative bacteria, nitrate-reducing organisms, and fungi interact within multispecies biofilms on tooth surfaces and mucosal niches [18, 35, 36]. Our findings suggest that HF-related changes should be interpreted not merely as shifts in individual taxa, but as evidence of systemic oral ecological remodeling. In our cohort, disease status was the primary driver of diversity in both bacterial and fungal communities. Furthermore, a weighted LASSO model identified specific features associated with the NYHA III-IV phenotype: fungal ASV586 (Geotrichum candidum), fungal ASV238 (family Nectriaceae), and bacterial ASV182 (Neisseria bacilliformis) exhibited positive coefficients. Conversely, ASV1 (Haemophilus), ASV0 (Streptococcus), and ASV148 (Pseudopropionibacterium) showed negative coefficients. This bidirectional pattern suggests that advanced HF is characterized by the enrichment of opportunistic or environmentally derived bacterial-fungal signatures, alongside the depletion of resident oral taxa that typically maintain biofilm stability.

The positive severity-associated features reflect an oral ecosystem under host stress but necessitate cautious interpretation. Although G. candidum is generally considered a low-virulence yeast-like fungus, oral geotrichosis has been documented primarily in immunocompromised patients, such as those with diabetes, hematologic malignancies, or HIV, it has also been rarely implicated in infective endocarditis [37, 38]. Nectriaceae contains diverse environmental fungi and medically relevant fusarioid members. However, family-level annotation prevents species-level inference [39]. N. bacilliformis, though rare, has been identified as a causative agent of infective endocarditis, reinforcing its potential role as an opportunistic signal with cardiovascular implications [40]. In advanced HF, factors such as chronic inflammation, impaired immunity, malnutrition, xerostomia, medication exposure and mucosal barrier dysfunction may disturb the plaque habitat, allowing low-abundance opportunistic organisms to expand. Thus, these taxa are likely markers of an unstable oral ecosystem rather than direct drivers of HF progression.

The negatively associated ASVs further underscore loss of commensal ecological structure. Streptococcus species are primary colonizers of oral surfaces and help build the early biofilm scaffold, whereas Haemophilus and other facultative organisms participate in oral microbial succession, interspecies metabolic exchange, and nitrate reduction [35, 41]. Oral nitrate-reducing bacteria contribute to the enterosalivary nitrate-nitrite-nitric oxide pathway, which links plaque ecology with vascular tone, endothelial function, blood pressure regulation, and cardiometabolic health [42, 43]. Therefore, depletion of Haemophilus- and Streptococcus-related ASVs in more severe HF may reflect loss of oral functions that support nitric oxide bioavailability. However, as these genera also include HACEK organisms and cariogenic species, they should not be categorized as uniformly protective [44, 45]. Overall, the LASSO-selected ASVs indicate a transition from a commensal-dominated biofilm toward an opportunistic, fungal-enriched dysbiotic state.

Functional predictions via PICRUSt2 further bridge these taxonomic shifts with metabolic potential. Among 165 pathways differing between HF and HC, eight showed nominal associations with NYHA class. TCA cycle I was enriched in HF and positively correlated with NYHA class. This may indicate a shift toward stress-adapted, oxygen-tolerant metabolic configurations within the plaque, influenced by altered saliva composition and local redox status in HF. Conversely, the pyruvate fermentation to butanoate pathway, which is a key route for butyrate production, was depleted in HF and negatively correlated with NYHA class. In oral biofilms, short-chain fatty acids (SCFAs) are produced by fermentative bacteria and may influence local pH, epithelial responses, inflammatory tone, and interspecies metabolic networks [46–49]. Systemically, butyrate and other SCFAs have been proposed as mediators of the gut-heart axis, with potential roles in epithelial barrier maintenance, immune regulation, blood pressure control, vascular relaxation, and cardiac metabolism [50–52]. However, oral butyrate is context-dependent. Periodontal pockets can contain elevated SCFA concentrations, and butyric acid produced by periodontopathic bacteria has been linked to periodontal tissue injury, inflammatory responses, and disease progression [48, 49]. Therefore, the inverse association between NYHA class and predicted pyruvate-to-butanoate capacity likely represents the collapse of a HC-associated fermentative network rather than a direct loss of a protective metabolite.

The pathway-ASV network provides a concise bridge between microbial composition and predicted function. After filtering for robust correlations, 20 pathway-ASV associations remained, involving seven pathways and eight ASVs. These ASVs included Streptococcus (ASV0), Haemophilus (ASV1 and ASV25), Alloprevotella (ASV109), Dialister invisus (ASV164), Veillonella (ASV86), Gemella (ASV11), and an unclassified Lactobacillales (ASV46), all of which showed higher relative abundance in HC than in HF. These taxa are typical members of oral biofilm communities and are plausibly involved in carbohydrate fermentation, nitrate metabolism, lactate cross-feeding, and biofilm maturation [36, 42, 53–55]. Strong positive correlations linked HC-enriched ASVs with HC-enriched functional modules, whereas TCA cycle I was negatively correlated with Haemophilus (ASV1) and Alloprevotella (ASV109). These patterns suggest that the progression of HF is accompanied by both the erosion of taxonomic diversity and the functional uncoupling of the supragingival plaque ecosystem.

Compared with previous research that primarily focused on 16S rRNA bacterial diversity, this study provides a more comprehensive view by integrating bacterial 16S, fungal ITS, and functional prediction analyses. However, several limitations must be acknowledged. First, the single-center design and relatively small sample size may limit the generalizability of our findings. Second, causality between oral dysbiosis and HF progression cannot be determined in this cross-sectional study. Third, although we considered available clinical variables where feasible, residual confounding from age, comorbidities, medication exposure, hospitalization-related management, and oral health status cannot be fully excluded. Fourth, amplicon sequencing lacks the species-, strain-, and gene-level resolution of shotgun metagenomics, and PICRUSt2-based pathway analysis cannot directly measure gene expression or metabolite production [56]. Furthermore, the lack of detailed radiographic dental examinations, periodontal pocket assessment, saliva measurements, and subgingival plaque sampling may provide an incomplete picture of total oral health status, given the strong habitat specificity of oral microbial communities [35]. Future longitudinal studies employing oral clinical phenotyping, shotgun metagenomics, metatranscriptomics, metabolomics, and inflammatory biomarker profiling across saliva, supragingival plaque, and subgingival plaque are required to further elucidate the oral-heart axis and its therapeutic potential.

Conclusions

These findings support a conceptual model in which the progression of HF is associated with oral dysbiosis across taxonomic, functional, and interaction-network levels. Severe HF is marked by the enrichment of opportunistic bacterial-fungal candidates alongside a depletion of commensal taxa critical for biofilm stability and nitrate-nitric oxide (NO) metabolism. Furthermore, we observed an expansion in predicted microbial TCA cycle potential and a concomitant loss of HC-enriched fermentative capacity, specifically the pyruvate-to-butanoate pathway. Collectively, these results suggest that the oral microbiome in HF functions as a dysbiotic biofilm system that may interact with cardiovascular physiology via systemic inflammation, microbial translocation, and altered metabolite bioavailability. Since the study is cross-sectional and the pathway interpretation is based on 16S profiles, these associations should be viewed as hypothesis-generating. Future longitudinal multi-omics studies are warranted to determine whether these oral microbial signatures contribute to HF progression or primarily serve as biomarkers of disease severity.

Abbreviations

HF

Heart failure

HFrEF

Heart failure with reduced ejection fraction

HFmrEF

Heart failure with mid-range ejection fraction

HFpEF

Heart failure with preserved ejection fraction

HFimpEF

Heart failure with improved ejection fraction

TMAO

Trimethylamine N-Oxide

CRP

C-reactive protein

NGS

Next-generation sequencing

ITS

Internal Transcribed Spacer

NYHA

New York Heart Association

NT-proBNP

N-terminal of the prohormone brain natriuretic peptide

BNP

Natriuretic peptide

LVEF

Left ventricular ejection fraction

eGFR

Estimated Glomerular Filtration Rate

CKD-EPI

Chronic Kidney Disease Epidemiology Collaboration Equation

ASV

Amplicon sequence variants

PCoA

Principal coordinate analysis

LASSO

Least absolute shrinkage and selection operator

ROC

Receiver operating characteristic curve

TCA

Tricarboxylic Acid

SCFAs

Short-chain fatty acids

Authors’ contributions

Hanyu Li: conceptualization, investigation, methodology, formal analysis, visualization, data curation, writing-original draft, writing-review and editing. Qingyang Wu: investigation, formal analysis, visualization, data curation, writing-original draft, writing-review and editing. Zeyuan Wang: investigation, data curation. Xiaomin Hu: conceptualization, funding acquisition, project administration, Supervision, Writing-review and editing. Shuyang Zhang: conceptualization, funding acquisition, project administration, supervision, writing-review and editing.

Funding

This study was supported by grants from Chinese Natural Science Foundation, grant number 82270405, Beijing Nova Program (20220484031), CAMS Innovation Fund for Medical Sciences, grant number 2025-I2MXHXX-004, Peking Union Medical College Hospital Talent Cultivation Program Category C UBJ11751, National High Level Hospital Clinical Research Funding [grant numbers 2025-PUMCH-A-100.

Data availability

The 16S amplicon sequencing data have been deposited in the Genome Sequence Archive (GSA) under the BioProject accession number PRJCA023815, and can be accessed via the following web link: https://ngdc.cncb.ac.cn/gsa/browse/CRA015231 The ITS amplicon sequencing data have been deposited in the GSA under the BioProject accession number PRJCA023818, and can be accessed via the following web link: https://ngdc.cncb.ac.cn/gsa/browse/CRA015230

The ITS amplicon sequencing data have been deposited in the GSA under the BioProject accession number PRJCA023818, and can be accessed via the following web link: https://ngdc.cncb.ac.cn/gsa/browse/CRA015230.

Declarations

Ethics approval and consent to participate

The study protocol was approved by the Ethics Committee of Peking Union Medical College Hospital (ID: K23C0687) and executed conforming to the principles of the Declaration of Helsinki. All subjects provided written, informed consents for participation in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.