OSA Is Associated With the Human Gut Microbiota Composition and Functional Potential in the Population-Based Swedish CardioPulmonary bioImage Study
2023; Elsevier BV; Volume: 164; Issue: 2 Linguagem: Inglês
10.1016/j.chest.2023.03.010
ISSN1931-3543
AutoresGabriel Baldanzi, Sergi Sayols-Baixeras, Jenny Theorell‐Haglöw, Koen F. Dekkers, Ulf Hammar, Diem Nguyen, Yi‐Ting Lin, Shafqat Ahmad, Jacob Bak Holm, Henrik Bjørn Nielsen, Louise Brunkwall, Christian Benedict, Jonathan Cedernaes, Sanna Koskiniemi, Mia Phillipson, Lars Lind, Johan Sundström, Göran Bergström, Gunnar Engström, J. G. Smith, Marju Orho‐Melander, Johan Ärnlöv, Beatrice Kennedy, Eva Lindberg, Tove Fall,
Tópico(s)Gut microbiota and health
ResumoBackgroundOSA is a common sleep-breathing disorder linked to increased risk of cardiovascular disease. Intermittent upper airway obstruction and hypoxia, hallmarks of OSA, have been shown in animal models to induce substantial changes to the gut microbiota composition, and subsequent transplantation of fecal matter to other animals induced changes in BP and glucose metabolism.Research QuestionDoes OSA in adults associate with the composition and functional potential of the human gut microbiota?Study Design and MethodsWe used respiratory polygraphy data from up to 3,570 individuals 50 to 64 years of age from the population-based Swedish Cardiopulmonary bioimage Study combined with deep shotgun metagenomics of fecal samples to identify cross-sectional associations between three OSA parameters covering apneas and hypopneas, cumulative sleep time in hypoxia, and number of oxygen desaturation events with gut microbiota composition. Data collection about potential confounders was based on questionnaires, onsite anthropometric measurements, plasma metabolomics, and linkage with the Swedish Prescribed Drug Register.ResultsWe found that all three OSA parameters were associated with lower diversity of species in the gut. Furthermore, in multivariable-adjusted analysis, the OSA-related hypoxia parameters were associated with the relative abundance of 128 gut bacterial species, including higher abundance of Blautia obeum and Collinsella aerofaciens. The latter species was also independently associated with increased systolic BP. Furthermore, the cumulative time in hypoxia during sleep was associated with the abundance of genes involved in nine gut microbiota metabolic pathways, including propionate production from lactate. Finally, we observed two heterogeneous sets of plasma metabolites with opposite association with species positively and negatively associated with hypoxia parameters, respectively.InterpretationOSA-related hypoxia, but not the number of apneas/hypopneas, is associated with specific gut microbiota species and functions. Our findings lay the foundation for future research on the gut microbiota-mediated health effects of OSA. OSA is a common sleep-breathing disorder linked to increased risk of cardiovascular disease. Intermittent upper airway obstruction and hypoxia, hallmarks of OSA, have been shown in animal models to induce substantial changes to the gut microbiota composition, and subsequent transplantation of fecal matter to other animals induced changes in BP and glucose metabolism. Does OSA in adults associate with the composition and functional potential of the human gut microbiota? We used respiratory polygraphy data from up to 3,570 individuals 50 to 64 years of age from the population-based Swedish Cardiopulmonary bioimage Study combined with deep shotgun metagenomics of fecal samples to identify cross-sectional associations between three OSA parameters covering apneas and hypopneas, cumulative sleep time in hypoxia, and number of oxygen desaturation events with gut microbiota composition. Data collection about potential confounders was based on questionnaires, onsite anthropometric measurements, plasma metabolomics, and linkage with the Swedish Prescribed Drug Register. We found that all three OSA parameters were associated with lower diversity of species in the gut. Furthermore, in multivariable-adjusted analysis, the OSA-related hypoxia parameters were associated with the relative abundance of 128 gut bacterial species, including higher abundance of Blautia obeum and Collinsella aerofaciens. The latter species was also independently associated with increased systolic BP. Furthermore, the cumulative time in hypoxia during sleep was associated with the abundance of genes involved in nine gut microbiota metabolic pathways, including propionate production from lactate. Finally, we observed two heterogeneous sets of plasma metabolites with opposite association with species positively and negatively associated with hypoxia parameters, respectively. OSA-related hypoxia, but not the number of apneas/hypopneas, is associated with specific gut microbiota species and functions. Our findings lay the foundation for future research on the gut microbiota-mediated health effects of OSA. FOR EDITORIAL COMMENT, SEE PAGE 290Take-home PointsStudy Question: Does OSA in adults associate with the composition and functional potential of the human gut microbiota?Results: OSA-related hypoxia was associated with the relative abundance of 128 gut bacterial species and with the abundance of genes involved in nine gut microbiota metabolic pathways.Interpretation: Our findings suggest a connection between OSA-related hypoxia and alterations of specific species and functional potential of the gut microbiota and lay the foundation for future research on the gut microbiota-mediated health effects of OSA. FOR EDITORIAL COMMENT, SEE PAGE 290 Study Question: Does OSA in adults associate with the composition and functional potential of the human gut microbiota? Results: OSA-related hypoxia was associated with the relative abundance of 128 gut bacterial species and with the abundance of genes involved in nine gut microbiota metabolic pathways. Interpretation: Our findings suggest a connection between OSA-related hypoxia and alterations of specific species and functional potential of the gut microbiota and lay the foundation for future research on the gut microbiota-mediated health effects of OSA. OSA is characterized by upper airway collapse episodes during sleep resulting in complete cessation (apneas) or reduction (hypopneas) of air flow and consequent intermittent hypoxia.1Jordan A.S. McSharry D.G. Malhotra A. Adult obstructive sleep apnoea.Lancet Lond Engl. 2014; 383: 736-747Abstract Full Text Full Text PDF PubMed Scopus (903) Google Scholar The prevalence of OSA has been increasing, partially attributed to the worldwide rising prevalence of obesity,2Benjafield A.V. Ayas N.T. Eastwood P.R. et al.Estimation of the global prevalence and burden of obstructive sleep apnoea: a literature-based analysis.Lancet Respir Med. 2019; 7: 687-698Abstract Full Text Full Text PDF PubMed Scopus (1570) Google Scholar a well-described cause of OSA.3Schwartz A.R. Patil S.P. Laffan A.M. Polotsky V. Schneider H. Smith P.L. Obesity and obstructive sleep apnea: pathogenic mechanisms and therapeutic approaches.Proc Am Thorac Soc. 2008; 5: 185-192Crossref PubMed Scopus (461) Google Scholar Although OSA has been prospectively associated with cardiovascular disease independent of BMI,4Peppard P.E. Young T. Palta M. Skatrud J. Prospective study of the association between sleep-disordered breathing and hypertension.N Engl J Med. 2000; 342: 1378-1384Crossref PubMed Scopus (4083) Google Scholar,5Azarbarzin A. Sands S.A. Stone K.L. et al.The hypoxic burden of sleep apnoea predicts cardiovascular disease-related mortality: the Osteoporotic Fractures in Men Study and the Sleep Heart Health Study.Eur Heart J. 2019; 40: 1149-1157Crossref PubMed Scopus (347) Google Scholar the mechanisms are not yet fully elucidated.6Collen J. Lettieri C. Wickwire E. Holley A. Obstructive sleep apnea and cardiovascular disease, a story of confounders.Sleep Breath. 2020; 24: 1299-1313Crossref PubMed Scopus (48) Google Scholar The most common clinical parameter of OSA severity is the apnea-hypopnea index (AHI), which quantifies the number of apneas and hypopneas during sleep. However, AHI does not differentiate short apnea events with mild oxygen desaturation from prolonged events with severe hypoxia.7Linz D. Colling S. Nußstein W. et al.Nocturnal hypoxemic burden is associated with epicardial fat volume in patients with acute myocardial infarction.Sleep Breath. 2018; 22: 703-711Crossref PubMed Scopus (25) Google Scholar To quantify the time in hypoxia, the percentage of sleep time with oxygen saturation < 90% (T90) is used.8Khoshkish S. Hohl M. Linz B. et al.The association between different features of sleep-disordered breathing and blood pressure: a cross-sectional study.J Clin Hypertens Greenwich Conn. 2018; 20: 575-581Crossref PubMed Scopus (14) Google Scholar Finally, the oxygen desaturation index (ODI) quantifies the number of oxygen desaturation events,9Terrill P.I. A review of approaches for analysing obstructive sleep apnoea-related patterns in pulse oximetry data.Respirology. 2020; 25: 475-485Crossref PubMed Scopus (34) Google Scholar and it is the most suitable parameter to measure intermittent hypoxia.10Rashid N.H. Zaghi S. Scapuccin M. Camacho M. Certal V. Capasso R. The value of oxygen desaturation index for diagnosing obstructive sleep apnea: a systematic review.Laryngoscope. 2021; 131: 440-447Crossref PubMed Scopus (50) Google Scholar In sum, the three parameters are complementary to each other because they capture different dimensions of OSA. Studies in animal models of OSA have found that intermittent hypoxia and airway obstruction produce substantial changes in the gut microbiota composition.11Moreno-Indias I. Torres M. Montserrat J.M. et al.Intermittent hypoxia alters gut microbiota diversity in a mouse model of sleep apnoea.Eur Respir J. 2015; 45: 1055-1065Crossref PubMed Scopus (186) Google Scholar, 12Lucking E.F. O'Connor K.M. Strain C.R. et al.Chronic intermittent hypoxia disrupts cardiorespiratory homeostasis and gut microbiota composition in adult male guinea-pigs.EBioMedicine. 2018; 38: 191-205Abstract Full Text Full Text PDF PubMed Scopus (58) Google Scholar, 13Tripathi A. Melnik A.V. Xue J. et al.Intermittent hypoxia and hypercapnia, a hallmark of obstructive sleep apnea, alters the gut microbiome and metabolome.mSystems. 2018; 3: e00020-e00118Crossref PubMed Google Scholar, 14Durgan D.J. Ganesh B.P. Cope J.L. et al.Role of the gut microbiome in obstructive sleep apnea-induced hypertension.Hypertens Dallas Tex 1979. 2016; 67: 469-474Google Scholar In turn, alterations of the gut microbiota induced by OSA may partly mediate the effects of OSA on adverse health outcomes, including hypertension and impaired glucose metabolism.14Durgan D.J. Ganesh B.P. Cope J.L. et al.Role of the gut microbiome in obstructive sleep apnea-induced hypertension.Hypertens Dallas Tex 1979. 2016; 67: 469-474Google Scholar, 15Khalyfa A. Ericsson A. Qiao Z. Almendros I. Farré R. Gozal D. Circulating exosomes and gut microbiome induced insulin resistance in mice exposed to intermittent hypoxia: effects of physical activity.EBioMedicine. 2021; 64103208Abstract Full Text Full Text PDF PubMed Scopus (32) Google Scholar, 16Badran M. Khalyfa A. Ericsson A. Gozal D. Fecal microbiota transplantation from mice exposed to chronic intermittent hypoxia elicits sleep disturbances in naïve mice.Exp Neurol. 2020; 334113439Crossref PubMed Scopus (43) Google Scholar Smaller studies in humans have linked OSA to the microbiota composition in the upper airways (n = 92)17Hong S.N. Kim K.J. Baek M.G. et al.Association of obstructive sleep apnea severity with the composition of the upper airway microbiome.J Clin Sleep Med. 2022; 18: 505-515Crossref PubMed Scopus (2) Google Scholar and the gut.18Li Q. Xu T. Shao C. et al.Obstructive sleep apnea is related to alterations in fecal microbiome and impaired intestinal barrier function.Sci Rep. 2023; 13: 778Crossref PubMed Scopus (8) Google Scholar, 19Bikov A. Szabo H. Piroska M. et al.Gut microbiome in patients with obstructive sleep apnoea.Appl Sci. 2022; 12: 2007Crossref Scopus (7) Google Scholar, 20Ko C.Y. Liu Q.Q. Su H.Z. et al.Gut microbiota in obstructive sleep apnea-hypopnea syndrome: disease-related dysbiosis and metabolic comorbidities.Clin Sci (Lond). 2019; 133: 905-917Crossref PubMed Scopus (77) Google Scholar Bikov et al19Bikov A. Szabo H. Piroska M. et al.Gut microbiome in patients with obstructive sleep apnoea.Appl Sci. 2022; 12: 2007Crossref Scopus (7) Google Scholar reported a lower abundance of the phylum Actinobacteria and a higher abundance of Proteobacteria in patients with OSA (n = 19) compared with control subjects (n = 20), which was not confirmed in other studies of comparable size.18Li Q. Xu T. Shao C. et al.Obstructive sleep apnea is related to alterations in fecal microbiome and impaired intestinal barrier function.Sci Rep. 2023; 13: 778Crossref PubMed Scopus (8) Google Scholar,20Ko C.Y. Liu Q.Q. Su H.Z. et al.Gut microbiota in obstructive sleep apnea-hypopnea syndrome: disease-related dysbiosis and metabolic comorbidities.Clin Sci (Lond). 2019; 133: 905-917Crossref PubMed Scopus (77) Google Scholar More recently, Li et al18Li Q. Xu T. Shao C. et al.Obstructive sleep apnea is related to alterations in fecal microbiome and impaired intestinal barrier function.Sci Rep. 2023; 13: 778Crossref PubMed Scopus (8) Google Scholar found that AHI was associated with higher abundance of Fusobacterium species and lower abundance of Peptoclostridium species in 48 individuals with symptoms of OSA. However, these studies did not adjust for important confounders (eg, diet, medications) and were limited in their taxonomic resolution of the microbiota. To overcome these limitations, adequately powered studies combining extensive information on confounders with species-level microbiota data are needed. Here, we used a validated method for population-wide screening for OSA (ApneaLink Air; ResMed)21Ng S.S.S. Chan T.O. To K.W. et al.Validation of a portable recording device (ApneaLink) for identifying patients with suspected obstructive sleep apnoea syndrome.Intern Med J. 2009; 39: 757-762Crossref PubMed Scopus (112) Google Scholar,22Prudon B. Hughes J. West S. A novel postal-based approach to diagnosing obstructive sleep apnoea in a high-risk population.Sleep Med. 2017; 33: 1-5Crossref PubMed Scopus (6) Google Scholar to investigate in a cross-sectional study how the OSA parameters AHI, T90, and ODI are associated with the human gut microbiota analyzed with shotgun metagenomic sequencing in up to 3,570 participants from the large population-based Swedish Cardiopulmonary Bioimage Study (SCAPIS). Moreover, given that animal studies suggested that the OSA-induced alterations are connected to cardiometabolic disturbances, we investigated whether the OSA-associated gut microbiota features were also associated with cardiovascular risk factors independent of OSA severity. From 2013 to 2018, a total of 30,154 people 50 to 64 years of age were randomly invited from the general population across six regions in Sweden to enroll in SCAPIS.23Bergström G. Berglund G. Blomberg A. et al.The Swedish CArdioPulmonary BioImage Study: objectives and design.J Intern Med. 2015; 278: 645-659Crossref PubMed Scopus (201) Google Scholar For the present study, we included 4,045 participants from the Uppsala region with OSA data and gut microbiota data (e-Fig 1), of which 146 (3.6%) self-reported a doctor diagnosis of OSA. We excluded 59 participants who reported treatment with CPAP. The SCAPIS data collection and the present study were approved by the Swedish Ethical Review Authority (DNR 2018-315 B and amendment 2020-06597, and DNR 2010-228-31M, respectively). All participants provided written informed consent. Assessment for OSA was conducted using the ApneaLink Air device22Prudon B. Hughes J. West S. A novel postal-based approach to diagnosing obstructive sleep apnoea in a high-risk population.Sleep Med. 2017; 33: 1-5Crossref PubMed Scopus (6) Google Scholar during one night at home. Apnea was defined as a reduction of breathing flow ≥ 80% for at least 10 s. Hypopnea was defined as a period of at least 10 s with a decrease in the baseline air flow of 30% to 80% combined with a decrease ≥ 4% in oxygen saturation. A desaturation event was defined as a decrease from baseline ≥ 4% in oxygen saturation. At least 4 h of air flow and oxygen saturation recordings were required to compute a valid AHI value, and at least 4 h of oxygen saturation recording was required to compute valid T90 and ODI values. The AHI was calculated as the mean number of apnea and hypopnea events, and ODI was calculated as the mean number of desaturation events per hour of total recording time. The T90 variable was computed by adding the time spent with an oxygen saturation < 90% and dividing by the total recording time. Severity groups were defined as no OSA for AHI < 5, mild for AHI 5 to 14.9, moderate for AHI 15 to 29.9, and severe for AHI ≥ 30.24Flemons W.W. Buysse D. Redline S. et al.Sleep-related breathing disorders in adults: recommendations for syndrome definition and measurement techniques in clinical research.Sleep. 1999; 22: 667-689Crossref PubMed Scopus (4911) Google Scholar For grouping based on T90, one group was composed of participants with a T90 = 0, whereas the remaining participants were grouped by tertiles (ie, groups T90 = 0, t1, t2, and t3). For the grouping based on ODI, the participants were divided by quartiles of ODI (ie, groups q1, q2, q3, and q4). Detailed information on the fecal metagenomic analysis can be found in Dekkers et al.25Dekkers K.F. Sayols-Baixeras S. Baldanzi G. et al.An online atlas of human plasma metabolite signatures of gut microbiome composition.Nat Commun. 2022; 13: 5370Crossref PubMed Scopus (45) Google Scholar Briefly, participants were instructed at the first study site visit to collect fecal samples at home and store them in the freezer until the second study site visit. The average interval between visits was 15 days. At the study site, the samples were then kept at −20 °C until they were shipped 0 to 7 days later to the central biobank to be kept at −80 °C. Samples were sent to Clinical Microbiomics A/S (Copenhagen, Denmark) for DNA extraction, shotgun metagenomic sequencing with Illumina Novaseq 6000 system (Illumina), and taxonomic annotation. Each extraction round contained a negative and positive control (ZymoBIOMICS Microbial Community Standard, D6300; Zymo Research). All the negative control subjects showed no detectable DNA. All extractions of positive control subjects had a positive DNA signal. For the positive control samples, the coefficient of variation estimated by the Shannon diversity index was 3.05%. For 158 pairs of biologica replicates, the coefficient of variation was 1.49%. Metagenomic species were defined as coabundant genes as described in Nielsen et al26Nielsen H.B. Almeida M. Juncker A.S. et al.Identification and assembly of genomes and genetic elements in complex metagenomic samples without using reference genomes.Nat Biotechnol. 2014; 32: 822-828Crossref PubMed Scopus (689) Google Scholar and reported as relative abundances. Species that were present in ≤ 1% of the participants were removed, resulting in 1,602 species for subsequent analyses. The taxonomic annotation of the metagenomic species was performed by mapping to National Center for Biotechnology Information's RefSeq27Haft D.H. DiCuccio M. Badretdin A. et al.RefSeq: an update on prokaryotic genome annotation and curation.Nucleic Acids Res. 2018; 46: D851-D860Crossref PubMed Scopus (598) Google Scholar database (downloaded on May 2, 2021). The putative metabolic profile of the species was defined in terms of gut metabolic modules (GMMs).28Vieira-Silva S. Falony G. Darzi Y. et al.Species-function relationships shape ecological properties of the human gut microbiome.Nat Microbiol. 2016; 116088Crossref PubMed Scopus (207) Google Scholar The GMMs are 103 metabolic pathways, defined as a series of enzymatic steps represented by the Kyoto Encyclopedia of Genes and Genomes orthology identifiers.28Vieira-Silva S. Falony G. Darzi Y. et al.Species-function relationships shape ecological properties of the human gut microbiome.Nat Microbiol. 2016; 116088Crossref PubMed Scopus (207) Google Scholar A species was considered to contain a GMM if it contained at least two-thirds of the Kyoto Encyclopedia of Genes and Genomes orthology of a module. For modules with three or fewer steps, all steps were required. For modules with alternative paths, only one path had to fulfil the criterion. Participants answered an extensive questionnaire on demographic information, lifestyle, self-reported health, and diet.23Bergström G. Berglund G. Blomberg A. et al.The Swedish CArdioPulmonary BioImage Study: objectives and design.J Intern Med. 2015; 278: 645-659Crossref PubMed Scopus (201) Google Scholar Smoking was categorized as no tobacco use, former tobacco use, or tobacco use. Education was categorized based on the highest level achieved: incomplete compulsory education, compulsory education, upper secondary education, or university education. Leisure time physical activity was self-reported as follows: mostly sedentary, moderate activity, regular and moderate activity, or regular exercise or training. According to birth country, participants were categorized into the following: born in Scandinavia (Sweden, Denmark, Norway or Finland), Europe, Asia, or other countries. Dietary information was assigned as missing for participants whose ln(total energy intake) was greater than the mean of ln(total energy intake) ± 3 SD in the study sample. From the food frequency questionnaires, variables were calculated to estimate alcohol intake (g/d),29Christensen S.E. Möller E. Bonn S.E. et al.Two new meal- and web-based interactive food frequency questionnaires: validation of energy and macronutrient intake.J Med Internet Res. 2013; 15: e109Crossref PubMed Scopus (60) Google Scholar fiber intake (g/d),30Christensen S.E. Möller E. Bonn S.E. et al.Relative validity of micronutrient and fiber intake assessed with two new interactive meal- and web-based food frequency questionnaires.J Med Internet Res. 2014; 16: e59Crossref PubMed Scopus (31) Google Scholar and total energy intake (kcal/d).29Christensen S.E. Möller E. Bonn S.E. et al.Two new meal- and web-based interactive food frequency questionnaires: validation of energy and macronutrient intake.J Med Internet Res. 2013; 15: e109Crossref PubMed Scopus (60) Google Scholar The variable season consisted of 11 categories based on the month of the first study site visit. Self-reported hypertension; lung disease (ie, COPD, chronic bronchitis, pulmonary emphysema); and use of medication for hypertension, hyperlipidemia, and/or diabetes were categorized as binary variables. Diabetes was defined as either a self-reported doctor diagnosis or as fasting plasma glucose ≥ 7.0 mM or glycated hemoglobin (HbA1c) ≥ 48 mol/mol. Impaired glucose tolerance was defined as no previous diabetes diagnosis and fasting glucose ≥ 6.1 and < 7.0 mM or HbA1c ≥ 42 and < 48 mol/mol. Individuals who used proton pump inhibitors and metformin users identified through the plasma metabolome. Previous antibiotic use (Anatomical Therapeutical Chemical code J01) was based on the Swedish Prescribed Drug Register. The fasting plasma samples were collected during the first site visit and stored at −80 °C in the central biobank until sent in random order to Metabolon Inc for metabolomics profiling, as previously described.25Dekkers K.F. Sayols-Baixeras S. Baldanzi G. et al.An online atlas of human plasma metabolite signatures of gut microbiome composition.Nat Commun. 2022; 13: 5370Crossref PubMed Scopus (45) Google Scholar We created a directed acyclic graph (DAG) (e-Fig 2) using the application DAGitty 3.031Textor J. van der Zander B. Gilthorpe M.S. Liśkiewicz M. Ellison G.T. Robust causal inference using directed acyclic graphs: the R package 'dagitty'.Int J Epidemiol. 2016; 45: 1887-1894PubMed Google Scholar to identify the minimal set of confounders for adjustment. Therefore, the main model consisted of age, sex, smoking, alcohol intake, BMI, and DNA extraction plate to account for variation between batches. Given the complexity of the DAG and that potential confounders (eg, diet32Bolte L.A. Vich Vila A. Imhann F. et al.Long-term dietary patterns are associated with pro-inflammatory and anti-inflammatory features of the gut microbiome.Gut. 2021; 70: 1287-1298Crossref PubMed Scopus (190) Google Scholar) were not included in the minimal set, we constructed an extended model accounting for fiber intake, total energy intake, physical activity, education, birth country, and season. Sleep duration was not included because we regarded it as a mediator in the DAG. Unless otherwise stated, associations were investigated with partial Spearman correlation. The Shannon index33Spellerberg I.F. Fedor P.J. A tribute to Claude Shannon (1916–2001) and a plea for more rigorous use of species richness, species diversity and the 'Shannon–Wiener' Index.Glob Ecol Biogeogr. 2003; 12: 177-179Crossref Scopus (886) Google Scholar (alpha diversity) was used as a metric of gut microbiota richness and evenness, whereas the Bray-Curtis dissimilarity (beta diversity) was used as a metric of interindividual compositional difference. From the R package vegan,34Oksanen J. Simpson G. Blanchet F. et al._vegan: Community Ecology Package_. R package version 2.6-4. 2022Google Scholar we used the function diversity to calculate the Shannon index and the function vegdist to calculate the Bray-Curtis dissimilarity based on the species relative abundance. To graphically compare beta diversity across groups of OSA severity, we conducted a principal coordinate analysis on the Bray-Curtis dissimilarity matrix, and differences were tested with permutational multivariate analysis of variance. Given that BMI may influence or be influenced by gut microbiota species,35Meijnikman A.S. Gerdes V.E. Nieuwdorp M. Herrema H. Evaluating causality of gut microbiota in obesity and diabetes in humans.Endocr Rev. 2018; 39: 133-153Crossref PubMed Scopus (192) Google Scholar BMI could be considered either a confounder or a source of reverse causation. Therefore, to investigate the association of OSA with individual species, we first applied the main model not including BMI as a screening step. The species identified in the screening were investigated adjusting for all main model covariates including BMI, and adjusting for the extended model. The P values were adjusted for multiple comparisons using the Benjamini-Hochberg method36Benjamini Y. Hochberg Y. Controlling the false discovery rate: a practical and powerful approach to multiple testing.J R Stat Soc Ser B Methodol. 1995; 57: 289-300Google Scholar with a false discovery rate set at 5% and referred to as q values. The species associated with at least one of the OSA parameters in the extended model were further examined in the following four sensitivity analyses: (1) included to the extended model the variables of medication use, more specifically metformin, proton pump inhibitors, and medications for hypertension and/or hyperlipidemia; (2) included waist-to-hip ratio in the extended model to further account for visceral adiposity; (3) excluded participants that had used any antibiotic in the 6 months before the first site visit; and (4) excluded participants with lung disease. The species identified in the extended model were investigated for co-occurrence using a probabilistic co-occurrence analysis implemented in the R package cooccur.37Griffith D.M. Veech J.A. Marsh C.J. cooccur: probabilistic species co-occurrence analysis in R.J Stat Softw. 2016; 69: 1-17Google Scholar We handled missing data using complete case analysis. In a secondary analysis, we imputed the AHI values for the participants who had valid T90 and ODI values but not valid AHI values. Multiple imputation was conducted using predicted mean matching with five nearest neighbors and 10 imputations including all covariates from the extended model, and Shannon index, T90, ODI, waist-to-hip ratio, and species relative abundance. A separate round of multiple imputation was performed for each species, and Rubin38Rubin D.B. Multiple Imputation for Nonresponse in Surveys. John Wiley & Sons, 2004Google Scholar combination rules were applied to account for the uncertainty in the imputation. This analysis was conducted using the software Stata 15.1 (StataCorp). All other analyses were conducted using the R software version 4.1.1. Effect modification by hemoglobin level was explored by categorizing participants into low or high hemoglobin groups based on the sex-specific median hemoglobin level. For every species, the pair of correlation coefficients obtained from the two groups was compared as described in Altman and Bland.39Altman D.G. Bland J.M. Interaction revisited: the difference between two estimates.BMJ. 2003; 326: 219Crossref PubMed Scopus (1464) Google Scholar Standard errors were estimated with bootsrapping for 1,000 iterations. We conducted enrichment analyses for GMMs based on ranked P values from the extended model results, stratified by the direction of the correlation coefficients. From the GUTSY Atlas (https://gutsyatlas.serve.scilifelab.se/), we retrieved information on enrichment for metabolite groups in the species associations with plasma metabolites.25Dekkers K.F. Sayols-Baixeras S. Baldanzi G. et al.An online atlas of human plasma metabolite signatures of gut microbiome composition.Nat Commun. 2022; 13: 5370Crossref PubMed Scopus (45) Google Scholar In a post hoc analysis, we assessed the association of our main gut microbiota findings with HbA1c, and systolic and diastolic BP with adjustment for age, sex, alcohol intake, smoking, fiber intake, total energy intake, physical activity, birth country, ODI, T90, AHI, and DNA extraction plates, as suggested by the DAG (e-Fig 3). We excluded users of medications for hypertension from the analyses with BP, and we excluded users of medications for diabetes from the association analysis with HbA1c. Finally, we added BMI to the model. The study sample consisted of 3,175 participants with valid AHI values (54% female) and 3,570 participants (52% female) with valid T90 and ODI values. The mean age was 57.7 years. Population characteristics are described in Table 1 and in e-Tables 1 and 2. The Spearman correlation coefficient between the OSA parameters was ρ = 0.56 for AHI and T90, ρ = 0.92 for
Referência(s)