Arrhythmic Gut Microbiome Signatures Predict Risk of Type 2 Diabetes

•Human gut microbiome exhibits diurnal rhythmicity across populations and individuals•Obese and T2D individuals show disrupted circadian rhythms in the gut microbiome•Arrhytmic bacterial signatures contribute to risk classification and prediction of T2D•These risk signatures show regional differences in applicability across three cohorts Lifestyle, obesity, and the gut microbiome are important risk factors for metabolic disorders. We demonstrate in 1,976 subjects of a German population cohort (KORA) that specific microbiota members show 24-h oscillations in their relative abundance and identified 13 taxa with disrupted rhythmicity in type 2 diabetes (T2D). Cross-validated prediction models based on this signature similarly classified T2D. In an independent cohort (FoCus), disruption of microbial oscillation and the model for T2D classification was confirmed in 1,363 subjects. This arrhythmic risk signature was able to predict T2D in 699 KORA subjects 5 years after initial sampling, being most effective in combination with BMI. Shotgun metagenomic analysis functionally linked 26 metabolic pathways to the diurnal oscillation of gut bacteria. Thus, a cohort-specific risk pattern of arrhythmic taxa enables classification and prediction of T2D, suggesting a functional link between circadian rhythms and the microbiome in metabolic diseases. Lifestyle, obesity, and the gut microbiome are important risk factors for metabolic disorders. We demonstrate in 1,976 subjects of a German population cohort (KORA) that specific microbiota members show 24-h oscillations in their relative abundance and identified 13 taxa with disrupted rhythmicity in type 2 diabetes (T2D). Cross-validated prediction models based on this signature similarly classified T2D. In an independent cohort (FoCus), disruption of microbial oscillation and the model for T2D classification was confirmed in 1,363 subjects. This arrhythmic risk signature was able to predict T2D in 699 KORA subjects 5 years after initial sampling, being most effective in combination with BMI. Shotgun metagenomic analysis functionally linked 26 metabolic pathways to the diurnal oscillation of gut bacteria. Thus, a cohort-specific risk pattern of arrhythmic taxa enables classification and prediction of T2D, suggesting a functional link between circadian rhythms and the microbiome in metabolic diseases. Increasing evidence links the human gut microbiome to metabolic health (Sonnenburg and Bäckhed, 2016Sonnenburg J.L. Bäckhed F. Diet-microbiota interactions as moderators of human metabolism.Nature. 2016; 535: 56-64Crossref PubMed Scopus (854) Google Scholar), and altered microbial profiles are associated with obesity, insulin resistance, and type 2 diabetes (T2D) (Goodrich et al., 2014Goodrich J.K. Waters J.L. Poole A.C. Sutter J.L. Koren O. Blekhman R. Beaumont M. Van Treuren W. Knight R. Bell J.T. et al.Human genetics shape the gut microbiome.Cell. 2014; 159: 789-799Abstract Full Text Full Text PDF PubMed Scopus (1460) Google Scholar; Karlsson et al., 2013aKarlsson F. Tremaroli V. Nielsen J. Bäckhed F. Assessing the human gut microbiota in metabolic diseases.Diabetes. 2013; 62: 3341-3349Crossref PubMed Scopus (259) Google Scholar, Karlsson et al., 2013bKarlsson F.H. Tremaroli V. Nookaew I. Bergström G. Behre C.J. Fagerberg B. Nielsen J. Bäckhed F. Gut metagenome in European women with normal, impaired and diabetic glucose control.Nature. 2013; 498: 99-103Crossref PubMed Scopus (1483) Google Scholar; Pedersen et al., 2016Pedersen H.K. Gudmundsdottir V. Nielsen H.B. Hyotylainen T. Nielsen T. Jensen B.A. Forslund K. Hildebrand F. Prifti E. Falony G. et al.Human gut microbes impact host serum metabolome and insulin sensitivity.Nature. 2016; 535: 376-381Crossref PubMed Scopus (792) Google Scholar; Qin et al., 2010Qin J. Li R. Raes J. Arumugam M. Burgdorf K.S. Manichanh C. Nielsen T. Pons N. Levenez F. Yamada T. et al.A human gut microbial gene catalogue established by metagenomic sequencing.Nature. 2010; 464: 59-65Crossref PubMed Scopus (6160) Google Scholar, Qin et al., 2012Qin J. Li Y. Cai Z. Li S. Zhu J. Zhang F. Liang S. Zhang W. Guan Y. Shen D. et al.A metagenome-wide association study of gut microbiota in type 2 diabetes.Nature. 2012; 490: 55-60Crossref PubMed Scopus (3147) Google Scholar; Thingholm et al., 2019Thingholm L.B. Rühlemann M.C. Koch M. Fuqua B. Laucke G. Boehm R. Bang C. Franzosa E.A. Hübenthal M. Rahnavard A. et al.Obese individuals with and without type 2 diabetes show different gut microbial functional capacity and composition.Cell Host Microbe. 2019; 26: 252-264.e10Abstract Full Text Full Text PDF PubMed Scopus (66) Google Scholar; Turnbaugh et al., 2006Turnbaugh P.J. Ley R.E. Mahowald M.A. Magrini V. Mardis E.R. Gordon J.I. An obesity-associated gut microbiome with increased capacity for energy harvest.Nature. 2006; 444: 1027-1031Crossref PubMed Scopus (6933) Google Scholar; Zhou et al., 2019Zhou W. Sailani M.R. Contrepois K. Zhou Y. Ahadi S. Leopold S.R. Zhang M.J. Rao V. Avina M. Mishra T. et al.Longitudinal multi-omics of host-microbe dynamics in prediabetes.Nature. 2019; 569: 663-671Crossref PubMed Scopus (134) Google Scholar). Population-based studies highlighted a significant degree of variability in inter-individual microbiomes (Falony et al., 2016Falony G. Joossens M. Vieira-Silva S. Wang J. Darzi Y. Faust K. Kurilshikov A. Bonder M.J. Valles-Colomer M. Vandeputte D. et al.Population-level analysis of gut microbiome variation.Science. 2016; 352: 560-564Crossref PubMed Scopus (850) Google Scholar; Zhernakova et al., 2016Zhernakova A. Kurilshikov A. Bonder M.J. Tigchelaar E.F. Schirmer M. Vatanen T. Mujagic Z. Vila A.V. Falony G. Vieira-Silva S. et al.Population-based metagenomics analysis reveals markers for gut microbiome composition and diversity.Science. 2016; 352: 565-569Crossref PubMed Scopus (769) Google Scholar), regional effects (He et al., 2018He Y. Wu W. Zheng H.M. Li P. McDonald D. Sheng H.-F. Chen M.-X. Chen Z.-H. Ji G.-Y. Zheng Z.-D.-X. et al.Regional variation limits applications of healthy gut microbiome reference ranges and disease models.Nat. Med. 2018; 24: 1532-1535Crossref PubMed Scopus (255) Google Scholar), and drug-associated changes in the gut microbiome (Forslund et al., 2015Forslund K. Hildebrand F. Nielsen T. Falony G. Le Chatelier E. Sunagawa S. Prifti E. Vieira-Silva S. Gudmundsdottir V. Pedersen H.K. et al.Disentangling type 2 diabetes and metformin treatment signatures in the human gut microbiota.Nature. 2015; 528: 262-266Crossref PubMed Scopus (921) Google Scholar; Pryor et al., 2019Pryor R. Norvaisas P. Marinos G. Best L. Thingholm L.B. Quintaneiro L.M. De Haes W. Esser D. Waschina S. Lujan C. et al.Host-microbe-drug-nutrient screen identifies bacterial effectors of Metformin therapy.Cell. 2019; 178: 1299-1312.e29Abstract Full Text Full Text PDF PubMed Scopus (81) Google Scholar). Despite the extensive efforts to define the role of the gut microbiome in metabolic diseases, especially obesity and T2D, limited reproducibility and specificity of disease-associated taxa across cohorts, e.g., members of Christensenellaceae, Collinsella, and Escherichia coli are also associated with Crohn’s disease (Pascal et al., 2017Pascal V. Pozuelo M. Borruel N. Casellas F. Campos D. Santiago A. Martinez X. Varela E. Sarrabayrouse G. Machiels K. et al.A microbial signature for Crohn's disease.Gut. 2017; 66: 813-822Crossref PubMed Scopus (299) Google Scholar), complicates the identification of microbial risk factors. The circadian clock, which synchronizes daily food intake behavior and metabolism with the day and night cycle (Panda, 2019Panda S. The arrival of circadian medicine.Nat. Rev. Endocrinol. 2019; 15: 67-69Crossref PubMed Scopus (35) Google Scholar), has recently been proposed to influence microbial homeostasis (Thaiss et al., 2014Thaiss C.A. Zeevi D. Levy M. Zilberman-Schapira G. Suez J. Tengeler A.C. Abramson L. Katz M.N. Korem T. Zmora N. et al.Transkingdom control of microbiota diurnal oscillations promotes metabolic homeostasis.Cell. 2014; 159: 514-529Abstract Full Text Full Text PDF PubMed Scopus (558) Google Scholar). Daytime-dependent fluctuations were identified in both the oral and fecal microbiota (Kaczmarek et al., 2017Kaczmarek J.L. Musaad S.M. Holscher H.D. Time of day and eating behaviors are associated with the composition and function of the human gastrointestinal microbiota.Am. J. Clin. Nutr. 2017; 106: 1220-1231PubMed Google Scholar; Thaiss et al., 2014Thaiss C.A. Zeevi D. Levy M. Zilberman-Schapira G. Suez J. Tengeler A.C. Abramson L. Katz M.N. Korem T. Zmora N. et al.Transkingdom control of microbiota diurnal oscillations promotes metabolic homeostasis.Cell. 2014; 159: 514-529Abstract Full Text Full Text PDF PubMed Scopus (558) Google Scholar). In murine models, circadian rhythms in gut microbiota composition and function are sensitive to diet and feeding patterns (Thaiss et al., 2014Thaiss C.A. Zeevi D. Levy M. Zilberman-Schapira G. Suez J. Tengeler A.C. Abramson L. Katz M.N. Korem T. Zmora N. et al.Transkingdom control of microbiota diurnal oscillations promotes metabolic homeostasis.Cell. 2014; 159: 514-529Abstract Full Text Full Text PDF PubMed Scopus (558) Google Scholar; Zarrinpar et al., 2014Zarrinpar A. Chaix A. Yooseph S. Panda S. Diet and feeding pattern affect the diurnal dynamics of the gut microbiome.Cell Metab. 2014; 20: 1006-1017Abstract Full Text Full Text PDF PubMed Scopus (334) Google Scholar). Diet-induced obesity dampens cyclic microbial fluctuations in rodents (Leone et al., 2015Leone V. Gibbons S.M. Martinez K. Hutchison A.L. Huang E.Y. Cham C.M. Pierre J.F. Heneghan A.F. Nadimpalli A. Hubert N. et al.Effects of diurnal variation of gut microbes and high-fat feeding on host circadian clock function and metabolism.Cell Host Microbe. 2015; 17: 681-689Abstract Full Text Full Text PDF PubMed Scopus (365) Google Scholar; Zarrinpar et al., 2014Zarrinpar A. Chaix A. Yooseph S. Panda S. Diet and feeding pattern affect the diurnal dynamics of the gut microbiome.Cell Metab. 2014; 20: 1006-1017Abstract Full Text Full Text PDF PubMed Scopus (334) Google Scholar), and epidemiological studies continue to show associations between circadian clock dysfunction due to modern lifestyle and T2D (reviewed in Onaolapo and Onaolapo, 2018Onaolapo A.Y. Onaolapo O.J. Circadian dysrhythmia-linked diabetes mellitus: examining melatonin's roles in prophylaxis and management.World J. Diabetes. 2018; 9: 99-114Crossref PubMed Google Scholar), supporting the hypothesis that diurnal oscillations in microbiota composition and function may contribute to metabolic health. The lack of documentation of stool sampling time in addition to the well-documented regional and individual differences in microbiota profiles may account for discrepancies between studies. We therefore suggest to consider circadian oscillations to better understand the underlying mechanisms of disease-associated microbiome alterations and to validate risk profiles in prospective cohorts. We provide clear demonstration of robust diurnal oscillations in fecal microbiota composition, using stool across a 24-h sampling period of three large-scaled human populations with a total of 4,131 subjects in Germany (KORA, FoCus, and enable). Most importantly, we demonstrate that loss of circadian rhythmicity affects microbiome features related to the onset and progression of T2D and identified bacterial signatures for metabolic risk profiling in human populations. KORA is a prospective cohort in the region of Augsburg (Germany) designed to understand the role of genetic, lifestyle, and environmental factors in disease progression including metabolic diseases (Table S1). As part of the second follow-up of the S4 KORA cohort, stool was sampled from 1,976 individuals in 2013, for whom we performed high-throughput 16S rRNA gene amplicon sequencing (Tables S2 and S3). Comparing individual microbiota compositions confirmed diverse ecosystems dominated by the two major phyla Firmicutes and Bacteroidetes (cumulative mean relative abundance, 91%) (Figures 1A and 1B ). In comparison with other studies, compositional variations were marginally affected by geography (0.9%), since KORA is restricted to a single city (Augsburg, Bavaria, Germany) and its close surrounding (Arumugam et al., 2011Arumugam M. Raes J. Pelletier E. Le Paslier D. Yamada T. Mende D.R. Fernandes G.R. Tap J. Bruls T. Batto J.M. et al.Enterotypes of the human gut microbiome.Nature. 2011; 473: 174-180Crossref PubMed Scopus (3748) Google Scholar; He et al., 2018He Y. Wu W. Zheng H.M. Li P. McDonald D. Sheng H.-F. Chen M.-X. Chen Z.-H. Ji G.-Y. Zheng Z.-D.-X. et al.Regional variation limits applications of healthy gut microbiome reference ranges and disease models.Nat. Med. 2018; 24: 1532-1535Crossref PubMed Scopus (255) Google Scholar; Yatsunenko et al., 2012Yatsunenko T. Rey F.E. Manary M.J. Trehan I. Dominguez-Bello M.G. Contreras M. Magris M. Hidalgo G. Baldassano R.N. Anokhin A.P. et al.Human gut microbiome viewed across age and geography.Nature. 2012; 486: 222-227Crossref PubMed Scopus (3909) Google Scholar) (Figure 1C). The cohort was characterized by an average individual richness of 348 ± 77 operational taxonomic units (OTUs) and 118 ± 37 Shannon effective number of species (Figure 1D). Unsupervised analysis based on generalized UniFrac distances identified three fecal microbiota clusters (C1, N = 744; C2, N = 981; C3, N = 249) similar to previously reported enterotypes (Arumugam et al., 2011Arumugam M. Raes J. Pelletier E. Le Paslier D. Yamada T. Mende D.R. Fernandes G.R. Tap J. Bruls T. Batto J.M. et al.Enterotypes of the human gut microbiome.Nature. 2011; 473: 174-180Crossref PubMed Scopus (3748) Google Scholar) (Figures 1E, 1F, and S1A). Individuals in C1 had the lowest microbiota richness and showed significantly higher relative abundances of Bacteroides. The most diverse cluster C2 (highest number of subjects) was dominated by members of the genus Ruminococcus, while Prevotella dominated in C3 (Figure 1G). Individuals with obesity (BMI ≥ 30, N = 558), T2D (N = 277), and prediabetic conditions (N = 356) classified according to their oral glucose tolerance (WHO criteria), cancer (N = 200) as well as cardiovascular disease (CVD) (N = 66) were evenly distributed across these clusters (Figure 1E). Multivariate analysis of metadata co-varying with the fecal microbiota profiles identified 40 of 113 features related to physiology (e.g., blood triglyceride levels, body weight, muscle mass, and time of defecation), lifestyle and environment (geographical region, beer/alcohol consumption, and seasons), disease-associated parameters (mostly related to glucose metabolism), and medication, collectively explaining 9.1% of variability (Figure 1H). Time of defecation was among the most significant factors (p value = 0.004; R2 = 0.001) explaining inter-individual variabilities in microbiota structure (Figure 1H). Thus, diurnal rhythmicity of fecal microbiota profiles was studied in 1,943 subjects for whom time at sampling was available (Figures 2A–2C ). Community diversity (both species richness and Shannon effective number of species) fluctuated significantly throughout the day (Figure 2A). Diurnal rhythmicity was also evident in relative abundances of the two most dominant phyla, which oscillated in antiphase. Bacteroidetes showed 6% higher mean relative abundance at night, whereas the phylum Firmicutes was higher during the day. Since more than 70% of all samples were collected at morning hours between 5 and 11 am, we re-analyzed the data using 10 random sub-samples of 25 patients for every time point and thereby confirmed initial results including all subjects (Figure S1A). After removal of OTUs low in mean relative abundance (<0.1%) and prevalence (<10% subjects), the heatmap of remaining 422 OTUs illustrated heterogeneous distribution of their peak relative abundances, ranging from early day to late night, suggesting that the microbiota at different times of the day are dominated by different microbial taxa (Figure 2B; Table S7). According to cosine-wave regression analysis, 15.2% of the OTUs were rhythmic (rOTUs) (Figure 2C). Similar proportions of rOTUs were identified using other non-parametric (12.3%) and parametric methods (13.5%; Figure S1B), demonstrating validity of the analysis. Oscillating fluctuations in alpha-diversity and Bacteroidetes as well as similar numbers of significant rhythmic OTUs were confirmed in a smaller, age-matched, and regionally nearby located (Munich/Freising), independent cohort with multiple sampling per person (enable cohort N = 93 subjects with n = 357 fecal samples, 19.9% rOTUs, Figures 2D and 2E). Even in a single individual (subject 1, S1), for whom 58 samples were collected consecutively over 3 years, microbiota oscillations (3% rOTUs) were demonstrated (Figure 2F). Since KORA is best suited for the study of metabolic conditions, we focused on obese, prediabetic, and T2D subjects. Species richness and alpha-diversity were lower in individuals with T2D and obesity (BMI ≥ 30), whereas Firmicutes-to-Bacteroidetes ratios remained unchanged compared with healthy subjects (Figure 3A). Significantly different relative abundances were identified for 30 OTUs in T2D subjects (N = 277) versus all others (N = 1,270) (Figure 3B; Table S7). Robust daily oscillations in alpha-diversity, phyla, and molecular species were observed in KORA subjects without T2D (nonT2D, N = 1,255) and subjects with BMI < 30 (N = 1,393) (Figures 3C–3E, S1C, and S1D). In contrast, rhythmicity in alpha-diversity and phylum proportions (Bacteroidetes and Firmicutes) were absent in subjects with either T2D (N = 401) or a BMI ≥ 30 (N = 545) (Figures 3C–3E). A heatmap showing peak relative abundances of OTUs confirmed the disruption of rhythmicity in subjects with T2D regardless of BMI (Figure S1D). All 10.4% OTUs that oscillated in nonT2D subjects lost rhythmicity in subjects with T2D (Figure S1C). Of note, 3.5% of OTUs gained rhythmicity in T2D cases (Figure S1C). To account for the difference in sample size between subject groups (T2D, N = 269; nonT2D, N = 1,255), the circadian analysis was validated using 10 different randomly selected and sample size-matched groups (Figure S1E). Loss of diurnal oscillations in diabetic subjects was well reflected in the relative abundances of single OTUs (Figures 3D and S1C). OTUs with disrupted rhythmicity in T2D were largely (>60%) not shared with arrhythmic OTUs in obese individuals, indicating a BMI-independent loss of rhythmicity in T2D (Figure 3F). Interestingly, intermediate phenotypes were noted in prediabetic subjects (N = 352) with a loss of rhythmicity for the two major phyla but not alpha-diversity (Figure 3C). In prediabetes, the proportion of rOTUs was reduced from 10.4% to 7.6% (Figures 3D and S1C). Similar results were obtained using JTK_CYCLE or harmonic cosine-wave regression, demonstrating robustness of the findings (Figure S1C). We identified 87 OTUs that oscillated in controls but lacked rhythmicity in T2D. They belonged to the genera Akkermansia, Bacteroides, Bifidobacterium, Blautia, Clostridium, Coprococcus, Dorea, Prevotella, Roseburia, and Ruminococcus (Figure 3G; Tables S2 and S7), which accords with recently published data describing oscillations in two subjects (Thaiss et al., 2014Thaiss C.A. Zeevi D. Levy M. Zilberman-Schapira G. Suez J. Tengeler A.C. Abramson L. Katz M.N. Korem T. Zmora N. et al.Transkingdom control of microbiota diurnal oscillations promotes metabolic homeostasis.Cell. 2014; 159: 514-529Abstract Full Text Full Text PDF PubMed Scopus (558) Google Scholar). Interestingly, the majority of these arrhythmic OTUs (66 from 87) identified in T2D also lost rhythmicity in prediabetes. In addition, the comparison of two paired stool sampling times with more than 8-h distance in the prospective subcohort of KORA confirmed the presence of daytime-related differences in nonT2D and, most importantly, also confirmed their absence in T2D (Figure S3B), supporting at least to some extent the population data at individual levels. Altogether, these population-based findings clearly indicate that rhythmicity of the fecal microbiota is disrupted in subjects with obesity and T2D. Analysis of eating behavior and dietary intake as influencing factors for time related microbial shifts (Collado et al., 2018Collado M.C. Engen P.A. Bandín C. Cabrera-Rubio R. Voigt R.M. Green S.J. Naqib A. Keshavarzian A. Scheer F.A.J.L. Garaulet M. Timing of food intake impacts daily rhythms of human salivary microbiota: a randomized, crossover study.FASEB J. 2018; 32: 2060-2072Crossref PubMed Scopus (40) Google Scholar) showed no significant differences between nonT2D, prediabetes, and T2D subjects. The number of meals individuals were consuming over one day was equally distributed among the groups with a similar total caloric intake for the groups. No difference was found in the type of consumed meals, e.g., no preferences of late-night eating within one group (Figure 4A), suggesting that eating habits and dietary intake are most likely not the underlying reasons to explain arrhythmicity of microbiota composition observed in T2D cases. Of note, percentage of missing or incomplete information varied between the groups: 7.1% of nonT2D and 20.2% of subjects with T2D provided no or insufficient dietary information. This highlights the problems of dietary assessment, including validity and misreporting (King et al., 2016King B.M. Ivester A.N. Burgess P.D. Shappell K.M. Coleman K.L. Cespedes V.M. Pruitt H.S. Burden G.K. Bour E.S. Adults with obesity underreport high-calorie foods in the home.Health Behav. Policy Rev. 2016; 3: 439-443Crossref Google Scholar) and argues for the combination of food questionnaires with objective biomarkers (Figure 4B). We then sought to identify diagnostic biomarkers for T2D development using microbiota profiles of 1,340 subjects sampled in 2013 as training data and another 699 subjects for whom matched samples at the 5-year follow-up (2018) were available as independent test data (Figure 5A; Table S3). Among the 87 arrhythmic OTUs (Figure 3G), we selected 13 arrhythmic OTUs (s-arOTUs) with differential 24-h time-of-day patterns using the detection of differential rhythmicity (DODR) R packages (Thaben and Westermark, 2016Thaben P.F. Westermark P.O. Differential rhythmicity: detecting altered rhythmicity in biological data.Bioinformatics. 2016; 32: 2800-2808Crossref PubMed Scopus (32) Google Scholar) (Figure 5B; Table S7), which overlap with the 30 differentially abundant OTUs detected in the whole cohort (Figure 3B). We trained a generalized linear model (GLM) on these 13 s-arOTUs to classify T2D. The model performed significantly better than an equal number of randomly selected control OTUs (rndOTUs), which were not rhythmic in either of the groups (repeated 100 times, mean area under curve [AUC] = 0.79 versus 0.59; p value for 100 permutations = 2.03 × 10−8) (Figure 5C). As a complementary and hypothesis-free approach for identifying T2D biomarkers, we also trained a random forest (RF) model, in which a 5-fold cross validation was applied to 80% of the data, while the remaining 20% were used to assess performance. We repeated this random split 100 times and identified 63 out of 425 OTUs that were consistently selected as being predictive, with a mean AUC of 0.73 on the test set (Figure 5D). BMI, as an additional variable in the model, reduced the number of selected OTUs to 14 (rfOTUs) with significant differences in relative abundance (mean AUC = 0.77, Figures S2I and 3B; Table S7). This signature included Bifidobacterium longum (OTU 37), Clostridium celatum (OTU 101), Intestinibacter bartlettii (OTU 63), Romboutsia ilealis (OTU 76), and several taxa closely related to Fecalibacterium prausnitzii (OTU 1,014, OTU 1,860, OTU 3,247, OTU 11,374, and OTU 34,127) and Escherichia coli (OTU 12 and OTU 34,182). However, a model trained on the outcome obesity was not able to differentiate T2D (mean AUC = 0.68), suggesting that the selected 14 rfOTUs are not merely surrogates of the confounding variable BMI (Figure S2B). In reverse, the selected 14 rfOTUs failed to differentiate obesity (mean AUC = 0.63; Figure S2B), supporting the finding that obesity and T2D differentially affect microbiota profiles. In addition, BMI did not perform well in a mixed effect RF model (AUC = 0.69; 63 selected OTUs; Figures S4F and S4G). Strikingly, 13 of these 14 rfOTUs are the same as the above identified 13 s-arOTUs (Figures 5B and S2I), supporting the importance of arrhythmic OTUs in the diabetic risk signature. We further show robustness of our results when adjusting for compositionality bias (Tsilimigras and Fodor, 2016Tsilimigras M.C. Fodor A.A. Compositional data analysis of the microbiome: fundamentals, tools, and challenges.Ann. Epidemiol. 2016; 26: 330-335Crossref PubMed Scopus (112) Google Scholar) (Figure S4E). A GLM using the selected rfOTUs and BMI (rfOTUs + BMI) classified T2D with an AUC of 0.79, performing significantly better than a set of 14 randomly picked OTUs (rndOTUs, AUC = 0.60; repeated 100 times, Figure 5E). Due to a comparable performance of both models and the extensive overlap of OTUs, all 13 s-arOTUs were used for further analysis. The integration of miscellaneous diabetes risk markers improved the classification up to an AUC of 0.87. A RF model trained on an expanded set of miscellaneous risk markers selected 9 OTUs which are all shared with the 13 s-arOTUs (Figure S2C). In agreement with previous results (Forslund et al., 2015Forslund K. Hildebrand F. Nielsen T. Falony G. Le Chatelier E. Sunagawa S. Prifti E. Vieira-Silva S. Gudmundsdottir V. Pedersen H.K. et al.Disentangling type 2 diabetes and metformin treatment signatures in the human gut microbiota.Nature. 2015; 528: 262-266Crossref PubMed Scopus (921) Google Scholar; Pryor et al., 2019Pryor R. Norvaisas P. Marinos G. Best L. Thingholm L.B. Quintaneiro L.M. De Haes W. Esser D. Waschina S. Lujan C. et al.Host-microbe-drug-nutrient screen identifies bacterial effectors of Metformin therapy.Cell. 2019; 178: 1299-1312.e29Abstract Full Text Full Text PDF PubMed Scopus (81) Google Scholar), metformin (MET) intake significantly affected T2D classification (+MET T2D AUC = 0.87 versus −MET T2D AUC = 0.60; Figure S2D), but a risk signature based on +/MET intake was not able to classify T2D in the KORA cohort (AUC = 0.60; Figure S2E). Although MET was found to synchronize peripheral circadian clocks (Barnea et al., 2012Barnea M. Haviv L. Gutman R. Chapnik N. Madar Z. Froy O. Metformin affects the circadian clock and metabolic rhythms in a tissue-specific manner.Biochim. Biophys. Acta. 2012; 1822: 1796-1806Crossref PubMed Scopus (47) Google Scholar), indicating that MET may directly interfere with the circadian analysis, MET did not change rhythmicity in alpha-diversity and phyla nor the overall percentage of rOTUs in subjects with T2D in our study (Figures S2F and S2G). Of note, 9 of 14 OTUs that gained rhythmicity in T2D showed diurnal oscillation in T2D only when taking MET and, thus, may represent rather protective OTUs. Importantly, none of the rOTUs identified in +/−MET-T2D overlap with the 13 s-arOTUs used for the classification of T2D (Figure S2H). Consequently, MET did not affect the predictability of T2D based on the s-arOTU signature. Considering the fact that 5 OTUs gained rhythmicity in T2D, these OTUs may have been chosen alternatively for T2D classification, nevertheless we focused on the larger proportion of arrhythmic OTUs. Disruption of microbial rhythmicity in T2D was confirmed in another large-scaled and KORA-independent cohort from the northern part (Kiel) of Germany (FoCus: N = 1,070 nonT2D, N = 293 T2D) (Relling et al., 2018Relling I. Akcay G. Fangmann D. Knappe C. Schulte D.M. Hartmann K. Müller N. Türk K. Dempfle A. Franke A. et al.Role of wnt5a in metabolic inflammation in humans.J. Clin. Endocrinol. Metab. 2018; 103: 4253-4264Crossref PubMed Scopus (13) Google Scholar). Loss of daily oscillations in richness and alpha-diversity was associated with a significant reduction of rhythmic OTUs in T2D (1.4% rOTUs) compared with nonT2D (8.5% rOTUs) (Figure 5F). Based on a BLAST search we assigned the corresponding s-arOTU and presented the differences in relative abundances between nonT2D and T2D (Figure S2J; Table S8). In combination with BMI the relative abundance values were considered as input for the imported KORA GLM, classifying T2D with an AUC of 0.76 and respective values for sensitivity and specificity of 75% and 69% (Figures 5G and 5H). Interestingly and different from FoCus, the across country validation of the T2D risk signature using the twin cohort from UK (TwinsUK; N = 1,399 including N = 1,259 nonT2D, N = 46 iT2D, and N = 94 pT2D cases) performed substantially worse in classifying T2D (AUC = 0.68 for pT2D) and predicting incident T2D (AUC = 0.69 for iT2D) (Figures S3C–S3E; Table S8). This concords with previously published data from a Chinese study with cohorts from different districts (He et al., 2018He Y. Wu W. Zheng H.M. Li P. McDonald D. Sheng H.-F. Chen M.-X. Chen Z.-H. Ji G.-Y. Zheng Z.-D.-X. et al.Regional variation limits applications of healthy gut microbiome reference ranges and disease models.Nat. Med. 2018; 24: 1532-1535Crossref PubMed Scopus (255) Google Scholar). We next addressed the question of whether these 13 s-arOTUs are also able to predict T2D in the prospective arm of KORA, which included 699 paired samples from 2013 and 2018 with 17 persisting T2D (pT2D) and 20 newly incident T2D (iT2D) cases (Figure 6A; Table S4). Here, rhythmicity was found for alpha-diversity, phyla proportions, and taxa in nonT2D but were lost for T2D cases (Figures 6B and S3A). Similar to the cross-sectional analysis (Figure 5C), the GLM was able to