Gut microbiome and atrial fibrillation : results from a large population-based study

This is a self-archived version of an original article. This version may differ from the original in pagination and typographic details. Author(s): Title: Year: Version: Copyright: Rights: Rights url: Please cite the original version: CC BY-NC-ND 4.0 https://creativecommons.org/licenses/by-nc-nd/4.0/ Gut microbiome and atrial fibrillation : results from a large population-based study © 2023 The Author(s). Published by Elsevier B.V. Published version Palmu, Joonatan; Börschel, Christin S.; Ortega-Alonso, Alfredo; Markó, Lajos; Inouye, Michael; Jousilahti, Pekka; Salido, Rodolfo A.; Sanders, Karenina; Brennan, Caitriona; Humphrey, Gregory C.; Sanders, Jon G.; Gutmann, Friederike; Linz, Dominik; Salomaa, Veikko; Havulinna, Aki S.; Forslund, Sofia K.; Knight, Rob; Lahti, Leo; Niiranen, Teemu; Schnabel, Renate B. Palmu, J., Börschel, C. S., Ortega-Alonso, A., Markó, L., Inouye, M., Jousilahti, P., Salido, R. A., Sanders, K., Brennan, C., Humphrey, G. C., Sanders, J. G., Gutmann, F., Linz, D., Salomaa, V., Havulinna, A. S., Forslund, S. K., Knight, R., Lahti, L., Niiranen, T., & Schnabel, R. B. (2023). Gut microbiome and atrial fibrillation : results from a large population-based study. EBioMedicine, 91, Article 104583. https://doi.org/10.1016/j.ebiom.2023.104583 2023 Gut microbiome and atrial fibrillation—results from a large population-based study Joonatan Palmu, a , b , x Christin S. Börschel, c , d , x Alfredo Ortega-Alonso, a , e , f Lajos Markó, g , h , i , j Michael Inouye, k , l Pekka Jousilahti, a Rodolfo A. Salido, t Karenina Sanders, t Caitriona Brennan, t Gregory C. Humphrey, t Jon G. Sanders, t , w Friederike Gutmann, g , h , i , j Dominik Linz, m , n , o , p Veikko Salomaa, a Aki S. Havulinna, a , q SofiaK.Forslund, g , h , i , j , r Rob Knight, s , t , u Leo Lahti, v Teemu Niiranen, a , b , x and Renate B. Schnabel c , d , x , ∗ a Department of Public Health and Welfare, Finnish Institute for Health and Welfare, Helsinki, Turku, Finland b Department of Internal Medicine, Turku University Hospital and University of Turku, Finland c Department of Cardiology, University Heart and Vascular Centre Hamburg-Eppendorf, Hamburg, Germany d German Centre for Cardiovascular Research (DZHK), Partner Site Hamburg/Kiel/Lübeck, Hamburg, Germany e Neuroscience Center, University of Helsinki, Helsinki, Finland f Faculty of Sport and Health Sciences, University of Jyväskylä, Jyväskylä, Finland g Experimental and Clinical Research Center, a Cooperation of Charité-Universitätsmedizin and the Max-Delbrück Center, Berlin, Germany h Max Delbrück Center for Molecular Medicine (MDC), Berlin, Germany i Charité –Universitätsmedizin Berlin, Berlin, Germany j DZHK (German Centre for Cardiovascular Research), Partner Site Berlin, Germany k Cambridge Baker Systems Genomics Initiative, Baker Heart and Diabetes Institute, Melbourne, Victoria, Australia l Cambridge Baker Systems Genomics Initiative, Department of Public Health and Primary Care, University of Cambridge, Cambridge, UK m Department of Cardiology, Maastricht University Medical Centre and Cardiovascular Research Institute Maastricht, Maastricht, the Netherlands n Department of Biomedical Sciences, Faculty of Health and Medical Sciences, University of Copenhagen, Copenhagen, Denmark o Centre for Heart Rhythm Disorders, Royal Adelaide Hospital, and University of Adelaide, Adelaide, Australia p Department of Cardiology, Radboud University Medical Centre, Nijmegen, the Netherlands q Institute for Molecular Medicine Finland, FIMM - HiLIFE, Helsinki, Finland r Structural and Computational Biology, European Molecular Biology Laboratory, Heidelberg, Germany s Center for Microbiome Innovation, Jacobs School of Engineering, University of California San Diego, La Jolla, CA, USA t Department of Pediatrics, School of Medicine, University of California San Diego, La Jolla, CA, USA u Department of Computer Science & Engineering, University of California San Diego, La Jolla, CA, USA v Department of Computing, University of Turku, Turku, Finland w Cornell Institute for Host-Microbe Interaction and Disease, Cornell University, Ithaca, NY, USA Summary Background Atrial fibrillation (AF) is an important heart rhythm disorder in aging populations. The gut microbiome composition has been previously related to cardiovascular disease risk factors. Whether the gut microbial profile is also associated with the risk of AF remains unknown. Methods We examined the associations of prevalent and incident AF with gut microbiota in the FINRISK 2002 study, a random population sample of 6763 individuals. We replicated our findings in an independent case–control cohort of 138 individuals in Hamburg, Germany. Findings Multivariable-adjusted regression models revealed that prevalent AF (N = 116) was associated with nine microbial genera. Incident AF (N = 539) over a median follow-up of 15 years was associated with eight microbial genera with false discovery rate (FDR)-corrected P < 0.05. Both prevalent and incident AF were associated with the genera Enorma and Bifidobacterium (FDR-corrected P < 0.001). AF was not significantly associated with bacterial diversity measures. Seventy-five percent of top genera (Enorma,Paraprevotella,Odoribacter,Collinsella, Barnesiella,Alistipes) in Cox regression analyses showed a consistent direction of shifted abundance in an independent AF case–control cohort that was used for replication. Abbreviations: AF, Atrial fibrillation; BMI, Body mass index; FDR, False discovery rate; LPS, Lipopolysaccharides; PCoA, Principal coordinates; SCFA, Short-chain fatty acid; SD, Standard deviation *Corresponding author. Department of Cardiology, University Heart and Vascular Centre Hamburg-Eppendorf, Martinistr. 52, 20246, Hamburg, Germany. E-mail address: r.schnabe[email protected] (R.B. Schnabel). x Both authors contributed equally. eBioMedicine 2023;91: 104583 Published Online xxx https://doi.org/10. 1016/j.ebiom.2023. 104583 www.thelancet.com Vol 91 May, 2023 1 Articles Interpretation Our findings establish the basis for the use of microbiome profiles in AF risk prediction. However, extensive research is still warranted before microbiome sequencing can be used for prevention and targeted treatment of AF. Funding This study was funded by European Research Council, German Ministry of Research and Education, Academy of Finland, Finnish Medical Foundation, and the Finnish Foundation for Cardiovascular Research, the Emil Aaltonen Foundation, and the Paavo Nurmi Foundation. Copyright © 2023 The Author(s). Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Keywords: Atrial fibrillation; Gut microbiome; Metagenomics; Epidemiology Introduction Atrial fibrillation (AF) is a complex disease, and the majority of cases occur after the age of 60 years. 1 The exact mechanisms of AF development and perpetuation remain unclear. Classical cardiovascular risk factors explain only slightly more than 50% of AF risk. Many of these established AF risk factors, including age, sex, hypertension, 2 obesity, 3 prevalent ischemic heart disease 4,5 and heart failure, 6 have been shown to be associated with an altered composition and function of the gut microbiome. Furthermore, diverse bacterial species have been described in atherosclerotic plaques 7 and a reduced diversity of gut microbiome has been observed in heart failure. 8,9 For AF, however, it remains unknown to which extent the gut microbial profile is related to the disease. 10,11 The gut microbiome or its products can act on downstream targets, as has been shown for metabolites such as short-chain fatty acids (SCFA, a major product of microbial dietary fibre degradation), trimethylamine Noxide, and lipopolysaccharides (LPS). 12–14 Recently, a prospective association of trimethylamine N-oxide with AF was reported in Norwegian individuals. 15 In canine experiments, trimethylamine N-oxide injected in ganglionated plexi, which are central components of the autonomic cardiac system, induced enhanced electrical excitability, atrial electrical remodelling, and prolonged induced AF, 16 resulting in fibrosis and cardiac dysfunction. 17 LPS may act through inflammatory pathways. In AF patients, LPS was associated with higher incidence of major adverse cardiovascular events, and adherence to a Mediterranean diet appeared to lower LPS concentrations and outcomes. 18 The literature shows, that to date, most of the research has focused on more easily measurable gut microbiome-produced metabolites. 10 Even without direct measurement of circulating metabolites, a more detailed assessment of the whole microbiome could help the research community to identify potential gut microbial species and metabolic pathways in human AF. A small series of articles based on a case–control study of 50 Chinese patients hospitalized with AF was published recently. 19–21 The results suggested that different gut microbiome signatures in AF exist. Changes in microbial diversity and the predominant microbiome pattern have been seen in paroxysmal AF. Current advances in metagenomics with comprehensive microbiome characterization permit the examination of the gut microbiome in relation to AF at scale. In the population-based FINRISK 2002 cohort of 6763 Research in context Evidence before this study Classical cardiovascular risk factors explain slightly over half of the atrial fibrillation risk. While gut microbiota has been recently linked to cardiovascular health, it remains still unknown to which extent the gut microbiota affects atrial fibrillation risk. However, small case–control study sample (N = 50) of Chinese patients suggest that different gut microbiome signatures in atrial fibrillation exist. Added value of this study Advances in sequencing allows the evaluation of the role gut microbiota to the development of hypertension. We study the risk of hypertension using shallow shotgun metagenomics data for 6763 FINRISK 2002 participants with >15 years register-based follow-up data for atrial fibrillation. Implications of all the available evidence We demonstrate that both prevalent and prospective atrial fibrillation is linked with distinct gut microbial genera. Similar trend was observed for 75% of the top genera in validation cohort. The shift of the bacterial composition in atrial fibrillation towards a spectrum with similarities to the microbiome previously reported in hypertension and heart failure highlights a shared underlying pathophysiology. Extensive research is still required to estimate the significance of these findings to risk prediction and management of atrial fibrillation. Articles 2 www.thelancet.com Vol 91 May, 2023 individuals with >15 years of follow-up, we examined how the prevalence and long-term incidence of AF was associated with the compositional profile and functional potential of the gut microbiome and qualitatively compared our findings with published gut microbiome associations with other cardiovascular diseases. 22 Methods Study sample The Finnish Institute for Health and Welfare has performed population surveys every five years since 1972 to monitor the development of cardiovascular risk factors in the Finnish population. 23 A random population sample of 13,437 individuals aged 25–74 years from six geographic regions was invited to participate in the FINRISK 2002 study. 24 Out of all invited individuals, 8799 (65.5%) participated in FINRISK 2002. In the cross-sectional sample, we excluded 1568 participants who did not provide stool samples, 20 participants due to low total read count (N < 50,000), and 448 participants due to missing relevant covariates resulting in a final sample of 6763 individuals. Of the 448 participants missing relevant covariates, 286 did not provide information on alcohol consumption, 115 did not grant permission to registry follow-up, 30 did not provide smoking status, and 17 had other missing covariates. In the prospective analyses, we additionally excluded 116 individuals with prevalent AF for a final longitudinal sample of 6647 participants. Health examination The participants completed a questionnaire on sociodemographic information, lifestyles, medications, and medical history at home. In the current study questionnaire information was used to define smoking status, alcohol consumption, physical activity, food choices, and as one criterion for the definition of diabetes. Physical examinations were performed at a local study site by trained staff. The participants underwent measurements for height and weight. A nurse drew venous blood samples for analysis of routine biomarkers and measured sitting blood pressure two times on the right arm using a mercury sphygmomanometer and a 14 × 40 cm sized cuff after a 5-min rest. The health examinations were performed in 2002. Stool sampling and storage Stool samples were collected at home after the physical examination in 50 ml Falcon tubes and were mailed to Finnish Institute for Health and Welfare using prepaid packages. The samples were then frozen in −20 ◦C until they underwent metagenomic sequencing in 2017. Stool DNA extraction and library preparation Microbiome analysis was performed at the University of California San Diego using whole-genome untargeted shallow shotgun metagenomic sequencing against mapped reference databases, following a previously published protocol. 25 In brief, Illumina-compatible libraries were prepared from isolated DNA, normalized to 5 ng input per sample, and sequenced using Hi-Seq 4000 for paired-end 150 bp reads. Sequence reads were mapped against taxonomy using SHOGUN v1.0.5 against NCBI RefSeq database (version 82; May 8, 2017). 26 Functional profiles were calculated from a combination of observed and predicted Kyoto Encyclopedia of Genes and Genomes Orthology group (KO) annotations from the RefSeq genomes following the predicted parameters of the SHOGUN tool. 26 Variables and covariates Body mass index (BMI) was calculated as kg/m 2 . Smoking was defined by as current daily smoking. We used self-reported average absolute alcohol consumption (grams per week) during the last 12 months. Information on medication use was retrieved from the Finnish National Drug Purchase Register, which captures all reimbursed prescription drug purchases in Finland. Antihypertensive medication use was defined as a drug purchase occurring during the four months preceding the study baseline under following Anatomical Therapeutic Chemical classification code classes: diuretics (C03), beta-blockers (C07), calcium channel blockers (C08), and renin–angiotensin system inhibitors (C09). Prevalent diabetes was defined as self-reported diabetes, a previous diagnostic code (ICD-10 codes E10-E14 or ICD-8/9 code 250) indicating diabetes in the nationwide Care Register for Health Care, which includes hospital discharges and specialist outpatient visits, three prior diabetes medication purchases (ATC code class A10), or special reimbursement code for diabetes medications in the Drug Reimbursement Register. Heart failure was defined using a previous diagnostic code indicating heart failure in the nationwide Care Register for Health Care (ICD-10 codes I50, I110, I130, I132; ICD-9 codes 4029B, 404, 4148, and 428; ICD8 codes 42,700, 42,710, and 428) or special reimbursement code for heart failure in the Drug Reimbursement Register. AF was defined using ICD-10 code I48, ICD-9 code 4273, ICD-8 code 42,792, in the nationwide Care Register for Health Care, or Causes-ofDeath Registers, or special reimbursement for dronedarone medication in the Drug Reimbursement Register before 31 December 2017. Validation cohort and data collection For external validation, we used a case–control study of patients with AF and limited risk factor burden compared to matched controls specifically collected for the examination of the gut microbiome between October 2019 and March 2020 at the University Clinic Hamburg-Eppendorf, Germany. Cases and controls were matched based on age, sex, cardiovascular risk Articles www.thelancet.com Vol 91 May, 2023 3 factors, and medication. N = 64 of patients with AF and 74 of the controls were finally available for analysis. The OMNIgene.GUT DNA Stabilisation Kit (DNA Genotek) was used. After aliquoting and freezing at −80 ◦C samples were shipped to the Max-Delbrück-Center, Berlin. In total, the cohort comprised 138 individuals with microbiome data available, 64 with an AF diagnosis. Microbial DNA was extracted from stool samples and shotgun sequenced, 27 filtered and quality controlled, then mapped using NGLESS 28 to the mOTU taxonomic space v2.5. 29 Prior to analysis, samples (represented by reads mapping to mOTU marker genes) were rarefied to this count from the smallest sample (1884), calculating alpha diversity in this process, using the RTK software. 30 Features identified in FINRISK were assessed in the replication cohort for (direction of) effects (using the Cliff’s delta nonparametric measure) by comparing rarefied abundances of AF and control samples in this cohort. We did not differentiate between paroxysmal, persistent and permanent forms of AF. Statistical methods We used R version 3.6.3 for all statistical analyses. The source code for the analyses is available at https://doi. org/10.5281/zenodo.4312841. Unless otherwise noted, we adjusted the analyses for age, sex, BMI, systolic blood pressure smoking, alcohol consumption, diabetes mellitus, heart failure, antihypertensive medication use, and total cholesterol. Alcohol consumption was log (x+1)-transformed to reduce the skewness of the lower tail bound distribution. We also assessed the characteristics of the study sample versus those that were excluded due to missing covariates. We calculated alpha diversity (Shannon index as a measure of mean species diversity as variation and richness in the sample) using species-level data with the R package microbiome. 31 We studied the association between prevalent AF and alpha diversity using logistic regression where prevalent AF was the dependent variable. With N = 6763, 539 incident cases, and alpha set at 0.05, we had 80% and 90% powers to detect odds ratios of 1.13 and 1.16. 32 We calculated the dissimilarity matrix (beta diversity indicating the variation in taxonomic abundance profiles between samples) and Principal Coordinates Analysis (PCoA) using Bray–Curtis dissimilarity on compositional microbial species-level abundance using R packages vegan. 33 We further studied common microbial genera prevalent in at least 1% of the sample population with a relative abundance over 0.1%. We examined associations of prevalent AF or incident AF (prevalent cases excluded) with the common microbial genera using DESeq2 with the Benjamini–Hochberg correction (FDR). 34,35 In DESeq2, microbiome composition is used as the outcome, instead of the exposure variable. The development of AF was assessed in subset of participants without AF at baseline Cox regression models with Breslow approximation for centered log-ratio transformed (CLR) microbial abundances. 36 We also performed functional analyses using log (x+1) transformed KEGG Orthology (KO) groups using Cox regression models with the Benjamini–Hochberg correction. We performed two additional sensitivity analyses. In the first sensitivity analysis, we introduced two additional covariates: leisure time activity and healthy food choices. 37 In the second sensitivity analysis, we limited the follow-up to 7.5 years which is approximately half of the total follow-up time. We also performed sparse Partial Least Squares Discriminant Analysis (sPLS-DA) using mixOmics library under R version 4.1.2. 38 We tuned optimal values for the sparsity parameters using k-fold cross validation. Ethics FINRISK 2002 study and the case–control validation study complies with the Declaration of Helsinki. Helsinki. Informed, written consent was obtained from all participants. The Coordinating Ethics Committee of the Helsinki and Uusimaa University Hospital District approved the FINRISK 2002 study. The case–control validation study was approved by Ärztekammer Hamburg (PV5705). Role of funders Funders had no role in the in the study design, data collection, data analyses nor interpretation or writing of the report. Results The baseline characteristics of the cross-sectional and prospective samples are shown in Table 1. The characteristics of the study sample and of individuals with missing covariates are reported in Supplemental Table S1. With the large study sample size even small between-group differences were significant using chisquare test and ANOVA. However, the number of individuals with missing data was low and the absolute/ clinical between-group differences were small. We defined common genera as genera that were present in the stool samples of at least of 1% of study participants; we used a cut-off value of over 0.1% relative abundance to define the presence of a genus in a stool sample. We observed 91 common microbial genera (Supplementary Table S2). Gut microbiome alpha diversity was not associated with prevalent AF in the ageand sex-adjusted (odds ratio [OR] 1.00; 95% confidence interval [CI] 0.83–1.20; P = 0.98) or in the multivariable-adjusted models (OR 1.04; 95% confidence interval [CI] 0.86–1.26; P = 0.71). Gut microbiome beta diversity was not associated with AF in ageand sex-adjusted (R 2 = 0.024%; P = 0.05) and multivariable-adjusted models (R 2 = 0.020%; P = 0.12). Fig. 1 shows microbial diversity (Bray–Curtis Articles 4 www.thelancet.com Vol 91 May, 2023 dissimilarity) as principal coordinate analysis for species-level bacterial abundances. The first three PcoA axes explained 31.2% of the variation in bacterial abundances. Ecological diversity measures did not accurately discriminate participants with AF status. We observed prevalent AF having nine significant associations with common microbial genera with FDRcorrected P < 0.05 (Fig. 2,Table 2). The associations were positive for Eisenbergiella,Enorma,Enterobacter, and Kluyvera, and negative for Bacteroides,Bifidobacterium,Holdemanella,Parabacteroides, and Turicibacter. Therefore, specific genus level gut microbial abundances have potential to identify individuals with AF. We also performed sPLS-DA analysis to maximize the discrimination potential between general gut microbial composition and AF. The method did not improve discrimination compared to principal coordinate analysis. Therefore, we suggest that the potential link between gut microbiota and AF is mainly driven by specific gut microbial species rather than general gut microbial composition. A total of 539 individuals developed AF over a median follow-up of 14.8 ± 3.0 years. There were no statistically significant differences in gut microbiome alpha or beta diversity between individuals who did and did not develop AF. We observed eight associations between incident AF and baseline common microbial genera with FDR-corrected P < 0.05 using DESeq2 (Table 3, Fig. 2). These associations were positive for Bifidobacterium,Enorma,Lactococcus,Mitsuokella, and Sellimonas, and negative for Tyzzerella,Hungatella, and Sanguibacteroides. Comparisons between top bacterial genera in each analysis were evaluated in the validation cohort (baseline characteristics see Supplementary Table S3) as the direction of effect contrasting AF cases and controls (Supplementary Table S4). Regression analysis revealed that 6 out of 8 shared top bacterial genera detected in both cohorts were shifted in the same direction (Fig. 3). The top hits DEseq2 models overlapped for Enorma, Eisenbergiella, and Bifidobacterium (Fig. 4). While ecological diversity measures did not accurately discriminate participants that developed AF Variable Overall (N = 6763) No prevalent AF (N = 6647) Prevalent AF (N = 116) Age, years (SD) 49.2 (12.9) 48.9 (12.8) 62.9 (8.5) Women, N (%) 3680 (54.4) 3646 (54.9) 34 (29.3) Body mass index, kg/m 2 (SD) 26.9 (4.6) 26.9 (4.6) 29.2 (5.2) Systolic blood pressure, mm Hg (SD) 135.6 (20.2) 135.5 (20.2) 144.2 (21.3) Diabetes mellitus, N (%) 371 (5.5) 357 (5.4) 14 (12.1) Current smoker, N (%) 1594 (23.6) 1580 (23.8) 14 (12.1) Antihypertensive medication, N (%) 1216 (18.0) 1134 (17.1) 82 (70.7) Total cholesterol, mmol/l (SD) 5.6 (1.1) 5.6 (1.1) 5.6 (1.0) Alcohol consumption, g (SD) 80.4 (122.0) 80.7 (122.1) 62.8 (113.8) Prevalent atrial fibrillation, N (%) 116 (1.7) 0 (0.0) 116 (100.0) Incident atrial fibrillation, N (%) 539 (8.0) 539 (8.1) 0 (0.0) Heart failure, N (%) 94 (1.4) 70 (1.1) 24 (20.7) Alcohol consumption reported before log(x+1)-transformation. Data are provided as mean (standard deviation [SD]) number (%). Table 1: Baseline characteristics of the study cohort. Fig. 1: Microbial diversity (Bray–Curtis dissimilarity) shown using principal coordinate analysis of species-level microbial abundances with prevalent atrial fibrillation cases denoted using red. Articles www.thelancet.com Vol 91 May, 2023 5 during follow-up, baseline genus level abundances have potential to identify individuals with extended follow-up. In sensitivity analyses we observed that half of the genera associated with incident AF remained significant and had relatively unchanged effect sizes when follow up was limited to 7.5 years (Supplementary Table S5). The eight genera associated with incident AF remained significant with similar effect sizes in sensitivity analyses when additional covariates for exercise and healthy food choices were included in the model (Supplementary Table S5). We also studied the association between the common microbial genera and incident AF using Cox proportional hazards models but observed no significant associations after FDR correction (Table 4). We then studied the functional associations between KO groups and incident AF using Cox proportional hazard models. Based on prior literature, we focused on 479 KO groups that are associated with the production of SCFAs and 14 trimethylamines, both well-known gut microbiome products. 39,40 In total, 288 of these 493 KO groups were detected in our baseline sample. We observed positive associations of AF with two KO groups (K15896, K15913) related to amino sugar and nucleotide sugar metabolism and with one KO group (K07271) linked to lipopolysaccharide biosynthesis (uncorrected P < 0.05; Supplementary Table S6). We also studied the associations between all available 6843 KO groups and incident AF observing FDR-corrected P > 0.05 for all associations (Supplementary Table S6). Discussion In a large, well-established population-based cohort we identified modest associations of prevalent and incident AF with the gut microbiome. The proportion of variance in microbial diversity measures explained by AF was low. We observed that prevalent AF was associated with nine genera and incident AF with eight genera using an FDR-corrected P value threshold of 0.05; Enorma,Bifidobacterium, and Eisenbergiella were among the top associations of prevalent and incident AF. Enorma also appeared among the top hits in Cox regression models for AF. Some plausible species and genera were identified in relation to AF, which are known in the context of established AF risk factors such as blood pressure control and heart failure. In a validation analysis, 75% of top genera (Paraprevotella,Odoribacter,Collinsella, Enorma,Barnesiella, Alistipes) in our Cox regression analyses showed a consistent direction of shifted abundance in an independent AF case–control cohort. Differentially abundant genera in atrial fibrillation At the genera level, the top association of prevalent AF was observed with Enorma.Enorma belongs to the family of Coriobacteriaceae, which were among the core Fig. 2: Heatmap showing log-fold change associated with atrial fibrillation in common microbial genera for nominally significant associations after FDR-corrected P value. Asterisk denotes association with bacterial plasmid. DEseq2 models were adjusted for age, sex, body mass index, systolic blood pressure, smoking, alcohol consumption, diabetes mellitus, heart failure, antihypertensive medication use, and total cholesterol. Bacterial genus Log2-fold change SE FDR-corrected P value Enorma 0.91 0.15 <0.001 Holdemanella −0.80 0.18 0.001 Eisenbergiella 0.81 0.19 0.001 Kluyvera 1.07 0.30 0.008 Parabacteroides −0.47 0.16 0.039 Turicibacter −0.74 0.25 0.039 Enterobacter 0.88 0.29 0.039 Bacteroides −0.79 0.27 0.039 Bifidobacterium −0.49 0.17 0.041 Desulfovibrio 0.56 0.20 0.051 The estimates are adjusted for age, sex, body mass index, systolic blood pressure, smoking, alcohol consumption, diabetes mellitus, heart failure, antihypertensive medication use, and total cholesterol. P values shown are adjusted for multiple testing using the Benjamini–Hochberg correction (FDR). Log2-fold changes were estimated using DESeq2. SE stands for standard error. Table 2: Top ten associations of prevalent atrial fibrillation (N = 116) with common genera using DESeq2 (N = 6763). Articles 6 www.thelancet.com Vol 91 May, 2023 families related to heart failure. 6 In a small, clinical case–control study with 20 heart failure patients the authors showed a significantly lower abundance of Coriobacteriaceae in diseased individuals. 6 Enorma species also appeared among the top associations with incident AF in both, DESeq2 and Cox regression analyses even after adjustment for prevalent heart failure. An association between Enorma and hypertension, which is strongly related to AF, has also been reported. 41 Further, Coriobacteriaceae have been positively correlated with total cholesterol, low-density cholesterol, and body mass index in healthy humans. 42 A phase II study evaluating the safety and efficacy of a non-steroidal farnesoid X receptor agonist in non-alcoholic fatty liver disease was terminated early because short intervals of cardiac arrhythmia were recorded during Holter monitoring. In this study, the authors had measured a relative decrease in abundance of Coriobacteriaceae in the participants’gut microbiome. 43 In DESeq2 analyses we found Eisenbergiella borderline differentially abundant in incident AF and significantly related to prevalent AF. This finding is in line with reports showing how Eisenbergiella is more abundant in normotensive persons. 44 In addition, the abundance of this genus is different in coronary artery disease patients. 45 Which is a strong predictor and established risk factor of AF. We observed a possible negative association of prevalent AF with Bifidobacterium, and a positive relation for incident AF. This observation may reflect different disease stages with higher impact of concurrent conditions such as heart failure in individuals with AF at baseline. In heart failure patients, the genus Bifidobacterium is depleted. 46 Bifidobacterium has been positively correlated with ejection fraction and negatively with the cardiac stress marker N-terminal pro Btype natriuretic peptide. 47 This prior knowledge may help explain the lower abundance observed in prevalent AF with a link to heart failure. Bifidobacterium belongs to the most abundant intestinal bacteria and is an integral part of most probiotics. Treatments with probiotics containing Bifidobacterium have been suggested to improve the atherogenic lipid profile. 48,49 Furthermore, Bifidobacterium has been related to favourable modulation of blood pressure. 50 In Cox regression analysis for incident AF, the genus Odoribacter was among the most differentially abundant bacteria. It comprises common species of the human intestinal microbiome isolated from faeces. 51 One of the Fig. 3: The trends in top bacterial genera with atrial fibrillation in derivation and validation cohorts. High agreement between bacterial genera scoring highest versus incident atrial fibrillation under a Cox regression model (see Table 4) with the direction (upwards-facing blue: positive association; downwards-facing red: negative association) in FINRISK (Cox estimate) and the validation cohort (Cliff’s Delta nonparametric effect size parameter). Bacterial genus Log2-fold change SE FDR-corrected P value Sellimonas 1.02 0.09 <0.001 Mitsuokella 1.02 0.13 <0.001 Enorma 0.46 0.07 <0.001 Tyzzerella −0.32 0.05 <0.001 Bifidobacterium 0.37 0.08 <0.001 Lactococcus 0.47 0.11 <0.001 Hungatella −0.20 0.06 0.016 Sanguibacteroides −0.26 0.08 0.022 Lactobacillus 0.27 0.10 0.051 Eisenbergiella −0.24 0.09 0.084 The estimates are adjusted for age, sex, body mass index, systolic blood pressure, smoking, alcohol consumption, diabetes mellitus, heart failure, antihypertensive medication use, and total cholesterol. P values shown are adjusted for multiple testing using the Benjamini–Hochberg correction (FDR). Log2-fold changes were estimated using DESeq2. SE stands for standard error. Table 3: Top ten associations of incident atrial fibrillation (N = 539) with common genera using DESeq2 (N = 6647). Articles www.thelancet.com Vol 91 May, 2023 7 top associations in both prospective analyses was observed with a related taxon of Sanguibacteroides. 52 In our cohort, individuals who developed AF had lower abundances of Odoribacter and Sanguibacteroides.A higher relative abundance of the Odoribacter genus has been related to lower blood pressure in pregnant women. 53 Blood pressure is a strong AF risk factor. 54 Bacteria of this genus produce the SCFA acid butyrate, a signalling molecule in blood pressure control. 55 SCFAs belong to the most abundant gut microbiome-derived physiologic modulators. They interact with G-protein coupled receptor pathways including renin secretion and sympathetic activation, 56 which are central to blood pressure regulation. Further, Odoribacter abundance positively correlates with isobutyric acid and an unfavourable lipid profile. 57 On the other hand, abundance of Atherosclerotic disease Myocardial infarction Heart failure Fibrosis Incident AF (N=539) Median follow-up 15 years Hypertension Partial bacterial overlap with risk factors and concomitant diseases Prevalent AF (N=116) 75% replication in AF case-controlstudy (N=148) Cardiovascular risk factorsinthe community Age Sex Hypertension Obesity Diabetes mellitus Overlap Enorma Eisenbergiella Bifidobacterium* Prevalent AF Holdemanella Kluyvera Parabacteroides Turicibacter EnterobactBactero ides* Desulfovibrioer Incident AF Sellimonas Mitsuokella Tyzzerella Lactococcus Hungatella Sanguibacteroides Lactobacillus s k Gut microbiome in the community (N=6763) Fig. 4: The circle illustrates the interaction between cardiovascular risk factors, gut 2 microbiome, and atrial fibrillation (AF) initiation/ perpetuation. The upper dark green box shows 3 the observed associations of prevalent AF with microbial genera, the lower box the top 4 associations of incident AF with genera in DESeq2 analyses. The light green box comprises 5 the top associated genera in both analyses. *Associations have been reported in prior studies. Bacterial genus HR 95% confidence interval P value FDR-corrected P value Odoribacter 0.917 0.851–0.982 0.009 0.697 Solobacterium 1.093 1.021–1.164 0.015 0.697 Sanguibacteroides a 0.943 0.891–0.995 0.026 0.804 Collinsella 1.041 0.982–1.100 0.186 0.818 Enorma a,b 1.069 0.967–1.172 0.201 0.818 Barnesiella 0.96 0.913–1.008 0.095 0.818 Paraprevotella 0.962 0.914–1.011 0.12 0.818 Alistipes 0.942 0.871–1.013 0.101 0.818 Enterococcus 0.941 0.857–1.024 0.15 0.818 Clostridium 1.087 0.967–1.207 0.174 0.818 The hazard ratios (HR) are adjusted for age, sex, body mass index, systolic blood pressure, smoking, alcohol consumption, diabetes mellitus, heart failure, antihypertensive medication use, and total cholesterol. Microbial abundances were transformed using centered log-ratio transformation. We used Breslow approximation for ties in Cox regression model. P values shown are adjusted for multiple testing using the Benjamini–Hochberg correction (FDR). a Overlap with top genera of DESeq2 analyses. b Overlap with top genera in prevalent atrial fibrillation. Table 4: Top ten common genera associated with incident atrial fibrillation in Cox regression analyses (N = 6923). Articles 8 www.thelancet.com Vol 91 May, 2023