Leveraging cross-species transcription factor binding site patterns: from diabetes risk loci to disease mechanisms.

Genome-wide association studies have revealed numerous risk loci associated with diverse diseases. However, identification of disease-causing variants within association loci remains a major challenge. Divergence in gene expression due to cis-regulatory variants in noncoding regions is central to disease susceptibility. We show that integrative computational analysis of phylogenetic conservation with a complexity assessment of co-occurring transcription factor binding sites (TFBS) can identify cis-regulatory variants and elucidate their mechanistic role in disease. Analysis of established type 2 diabetes risk loci revealed a striking clustering of distinct homeobox TFBS. We identified the PRRX1 homeobox factor as a repressor of PPARG2 expression in adipose cells and demonstrate its adverse effect on lipid metabolism and systemic insulin sensitivity, dependent on the rs4684847 risk allele that triggers PRRX1 binding. Thus, cross-species conservation analysis at the level of co-occurring TFBS provides a valuable contribution to the translation of genetic association signals to disease-related molecular mechanisms.

Epigenome-wide association study of body mass index, and the adverse outcomes of adiposity.

Approximately 1.5 billion people worldwide are overweight or affected by obesity, and are at risk of developing type 2 diabetes, cardiovascular disease and related metabolic and inflammatory disturbances. Although the mechanisms linking adiposity to associated clinical conditions are poorly understood, recent studies suggest that adiposity may influence DNA methylation, a key regulator of gene expression and molecular phenotype. Here we use epigenome-wide association to show that body mass index (BMI; a key measure of adiposity) is associated with widespread changes in DNA methylation (187 genetic loci with P < 1 × 10-7, range P = 9.2 × 10-8 to 6.0 × 10-46; n = 10,261 samples). Genetic association analyses demonstrate that the alterations in DNA methylation are predominantly the consequence of adiposity, rather than the cause. We find that methylation loci are enriched for functional genomic features in multiple tissues (P < 0.05), and show that sentinel methylation markers identify gene expression signatures at 38 loci (P < 9.0 × 10-6, range P = 5.5 × 10-6 to 6.1 × 10-35, n = 1,785 samples). The methylation loci identify genes involved in lipid and lipoprotein metabolism, substrate transport and inflammatory pathways. Finally, we show that the disturbances in DNA methylation predict future development of type 2 diabetes (relative risk per 1 standard deviation increase in methylation risk score: 2.3 (2.07-2.56); P = 1.1 × 10-54). Our results provide new insights into the biologic pathways influenced by adiposity, and may enable development of new strategies for prediction and prevention of type 2 diabetes and other adverse clinical consequences of obesity.

Epigenetic upregulation of FKBP5 by aging and stress contributes to NF-κB-driven inflammation and cardiovascular risk.

Aging and psychosocial stress are associated with increased inflammation and disease risk, but the underlying molecular mechanisms are unclear. Because both aging and stress are also associated with lasting epigenetic changes, a plausible hypothesis is that stress along the lifespan could confer disease risk through epigenetic effects on molecules involved in inflammatory processes. Here, by combining large-scale analyses in human cohorts with experiments in cells, we report that FKBP5, a protein implicated in stress physiology, contributes to these relations. Across independent human cohorts (total n > 3,000), aging synergized with stress-related phenotypes, measured with childhood trauma and major depression questionnaires, to epigenetically up-regulate FKBP5 expression. These age/stress-related epigenetic effects were recapitulated in a cellular model of replicative senescence, whereby we exposed replicating human fibroblasts to stress (glucocorticoid) hormones. Unbiased genome-wide analyses in human blood linked higher FKBP5 mRNA with a proinflammatory profile and altered NF-κB-related gene networks. Accordingly, experiments in immune cells showed that higher FKBP5 promotes inflammation by strengthening the interactions of NF-κB regulatory kinases, whereas opposing FKBP5 either by genetic deletion (CRISPR/Cas9-mediated) or selective pharmacological inhibition prevented the effects on NF-κB. Further, the age/stress-related epigenetic signature enhanced FKBP5 response to NF-κB through a positive feedback loop and was present in individuals with a history of acute myocardial infarction, a disease state linked to peripheral inflammation. These findings suggest that aging/stress-driven FKBP5-NF-κB signaling mediates inflammation, potentially contributing to cardiovascular risk, and may thus point to novel biomarker and treatment possibilities.

Randomized test-treatment studies with an outlook on adaptive designs.

BACKGROUND: Diagnostic accuracy studies aim to examine the diagnostic accuracy of a new experimental test, but do not address the actual merit of the resulting diagnostic information to a patient in clinical practice. In order to assess the impact of diagnostic information on subsequent treatment strategies regarding patient-relevant outcomes, randomized test-treatment studies were introduced. Various designs for randomized test-treatment studies, including an evaluation of biomarkers as part of randomized biomarker-guided treatment studies, are suggested in the literature, but the nomenclature is not consistent. METHODS: The aim was to provide a clear description of the different study designs within a pre-specified framework, considering their underlying assumptions, advantages as well as limitations and derivation of effect sizes required for sample size calculations. Furthermore, an outlook on adaptive designs within randomized test-treatment studies is given. RESULTS: The need to integrate adaptive design procedures in randomized test-treatment studies is apparent. The derivation of effect sizes induces that sample size calculation will always be based on rather vague assumptions resulting in over- or underpowered study results. Therefore, it might be advantageous to conduct a sample size re-estimation based on a nuisance parameter during the ongoing trial. CONCLUSIONS: Due to their increased complexity, compared to common treatment trials, the implementation of randomized test-treatment studies poses practical challenges including a huge uncertainty regarding study parameters like the expected outcome in specific subgroups or disease prevalence which might affect the sample size calculation. Since research on adaptive designs within randomized test-treatment studies is limited so far, further research is recommended.

The Alzheimer's Association international guidelines for handling of cerebrospinal fluid for routine clinical measurements of amyloid ß and tau.

The core cerebrospinal fluid (CSF) Alzheimer's disease (AD) biomarkers amyloid beta (Aß42 and Aß40), total tau, and phosphorylated tau, have been extensively clinically validated, with very high diagnostic performance for AD, including the early phases of the disease. However, between-center differences in pre-analytical procedures may contribute to variability in measurements across laboratories. To resolve this issue, a workgroup was led by the Alzheimer's Association with experts from both academia and industry. The aim of the group was to develop a simplified and standardized pre-analytical protocol for CSF collection and handling before analysis for routine clinical use, and ultimately to ensure high diagnostic performance and minimize patient misclassification rates. Widespread application of the protocol would help minimize variability in measurements, which would facilitate the implementation of unified cut-off levels across laboratories, and foster the use of CSF biomarkers in AD diagnostics for the benefit of the patients.

Blood Leukocyte DNA Methylation Predicts Risk of Future Myocardial Infarction and Coronary Heart Disease.

BACKGROUND: DNA methylation is implicated in coronary heart disease (CHD), but current evidence is based on small, cross-sectional studies. We examined blood DNA methylation in relation to incident CHD across multiple prospective cohorts. METHODS: Nine population-based cohorts from the United States and Europe profiled epigenome-wide blood leukocyte DNA methylation using the Illumina Infinium 450k microarray, and prospectively ascertained CHD events including coronary insufficiency/unstable angina, recognized myocardial infarction, coronary revascularization, and coronary death. Cohorts conducted race-specific analyses adjusted for age, sex, smoking, education, body mass index, blood cell type proportions, and technical variables. We conducted fixed-effect meta-analyses across cohorts. RESULTS: Among 11 461 individuals (mean age 64 years, 67% women, 35% African American) free of CHD at baseline, 1895 developed CHD during a mean follow-up of 11.2 years. Methylation levels at 52 CpG (cytosine-phosphate-guanine) sites were associated with incident CHD or myocardial infarction (false discovery rate<0.05). These CpGs map to genes with key roles in calcium regulation (ATP2B2, CASR, GUCA1B, HPCAL1), and genes identified in genome- and epigenome-wide studies of serum calcium (CASR), serum calcium-related risk of CHD (CASR), coronary artery calcified plaque (PTPRN2), and kidney function (CDH23, HPCAL1), among others. Mendelian randomization analyses supported a causal effect of DNA methylation on incident CHD; these CpGs map to active regulatory regions proximal to long non-coding RNA transcripts. CONCLUSION: Methylation of blood-derived DNA is associated with risk of future CHD across diverse populations and may serve as an informative tool for gaining further insight on the development of CHD.

Allele-specific quantitative proteomics unravels molecular mechanisms modulated by cis-regulatory PPARG locus variation.

Genome-wide association studies identified numerous disease risk loci. Delineating molecular mechanisms influenced by cis-regulatory variants is essential to understand gene regulation and ultimately disease pathophysiology. Combining bioinformatics and public domain chromatin information with quantitative proteomics supports prediction of cis-regulatory variants and enabled identification of allele-dependent binding of both, transcription factors and coregulators at the type 2 diabetes associated PPARG locus. We found rs7647481A nonrisk allele binding of Yin Yang 1 (YY1), confirmed by allele-specific chromatin immunoprecipitation in primary adipocytes. Quantitative proteomics also found the coregulator RING1 and YY1 binding protein (RYBP) whose mRNA levels correlate with improved insulin sensitivity in primary adipose cells carrying the rs7647481A nonrisk allele. Our findings support a concept with diverse cis-regulatory variants contributing to disease pathophysiology at one locus. Proteome-wide identification of both, transcription factors and coregulators, can profoundly improve understanding of mechanisms underlying genetic associations.

Long-Term Stability of Human Plasma Metabolites during Storage at -80 °C.

Prolonged storage of biospecimen can lead to artificially altered metabolite concentrations and thus bias data analysis in metabolomics experiments. To elucidate the potential impact of long-term storage on the metabolite profile, a pooled human plasma sample was aliquoted and stored at -80 °C. During a time period of five years, 1012 of the aliquots were measured with the Biocrates AbsoluteIDQ p180 targeted-metabolomics assay at 193 time points. Modeling the concentration courses over time revealed that 55 out of 111 metabolites remained stable. The statistically significantly changed metabolites showed on average an increase or decrease of +13.7% or -14.5%, respectively. In detail, increased concentration levels were observed for amino acids (mean: + 15.4%), the sum of hexoses (+7.9%), butyrylcarnitine (+9.4%), and some phospholipids mostly with chain lengths exceeding 40 carbon atoms (mean: +18.0%). Lipids tended to exhibit decreased concentration levels with the following mean concentration changes: acylcarnitines, -12.1%; lysophosphatidylcholines, -15.1%; diacyl-phosphatidylcholines, -17.0%; acyl-alkyl-phosphatidylcholines, -13.3%; sphingomyelins, -14.8%. We conclude that storage of plasma samples at -80 °C for up to five years can lead to altered concentration levels of amino acids, acylcarnitines, glycerophospholipids, sphingomyelins, and the sum of hexoses. These alterations must be considered when analyzing metabolomics data from long-term epidemiological studies.

Metabolite ratios as potential biomarkers for type 2 diabetes: a DIRECT study.

AIMS/HYPOTHESIS: Circulating metabolites have been shown to reflect metabolic changes during the development of type 2 diabetes. In this study we examined the association of metabolite levels and pairwise metabolite ratios with insulin responses after glucose, glucagon-like peptide-1 (GLP-1) and arginine stimulation. We then investigated if the identified metabolite ratios were associated with measures of OGTT-derived beta cell function and with prevalent and incident type 2 diabetes. METHODS: We measured the levels of 188 metabolites in plasma samples from 130 healthy members of twin families (from the Netherlands Twin Register) at five time points during a modified 3 h hyperglycaemic clamp with glucose, GLP-1 and arginine stimulation. We validated our results in cohorts with OGTT data (n = 340) and epidemiological case-control studies of prevalent (n = 4925) and incident (n = 4277) diabetes. The data were analysed using regression models with adjustment for potential confounders. RESULTS: There were dynamic changes in metabolite levels in response to the different secretagogues. Furthermore, several fasting pairwise metabolite ratios were associated with one or multiple clamp-derived measures of insulin secretion (all p < 9.2 × 10-7). These associations were significantly stronger compared with the individual metabolite components. One of the ratios, valine to phosphatidylcholine acyl-alkyl C32:2 (PC ae C32:2), in addition showed a directionally consistent positive association with OGTT-derived measures of insulin secretion and resistance (p ≤ 5.4 × 10-3) and prevalent type 2 diabetes (ORVal_PC ae C32:2 2.64 [ß 0.97 ± 0.09], p = 1.0 × 10-27). Furthermore, Val_PC ae C32:2 predicted incident diabetes independent of established risk factors in two epidemiological cohort studies (HRVal_PC ae C32:2 1.57 [ß 0.45 ± 0.06]; p = 1.3 × 10-15), leading to modest improvements in the receiver operating characteristics when added to a model containing a set of established risk factors in both cohorts (increases from 0.780 to 0.801 and from 0.862 to 0.865 respectively, when added to the model containing traditional risk factors + glucose). CONCLUSIONS/INTERPRETATION: In this study we have shown that the Val_PC ae C32:2 metabolite ratio is associated with an increased risk of type 2 diabetes and measures of insulin secretion and resistance. The observed effects were stronger than that of the individual metabolites and independent of known risk factors.

Allele-specific methylation of type 1 diabetes susceptibility genes.

The susceptibility to autoimmune diseases is influenced by genes encoding major histocompatibility complex (MHC) proteins. By examining the epigenetic methylation maps of cord blood samples, we found marked differences in the methylation status of CpG sites within the MHC genes (cis-metQTLs) between carriers of the type 1 diabetes risk haplotypes HLA-DRB1*03-DQA1*0501-DQB1*0201 (DR3-DQ2) and HLA-DRB1*04-DQA1*0301-DQB1*0302 (DR4-DQ8). These differences were found in children and adults, and were accompanied by reduced HLA-DR protein expression in immune cells with the HLA-DR3-DQ2 haplotype. Extensive cis-metQTLs were identified in all 45 immune and non-immune type 1 diabetes susceptibility genes analyzed in this study. We observed and validated a novel association between the methylation status of CpG sites within the LDHC gene and the development of insulin autoantibodies in early childhood in children who are carriers of the highest type 1 diabetes risk genotype. Functionally relevant epigenetic changes in susceptibility genes may represent therapeutic targets for type 1 diabetes.

Characterization of missing values in untargeted MS-based metabolomics data and evaluation of missing data handling strategies.

BACKGROUND: Untargeted mass spectrometry (MS)-based metabolomics data often contain missing values that reduce statistical power and can introduce bias in biomedical studies. However, a systematic assessment of the various sources of missing values and strategies to handle these data has received little attention. Missing data can occur systematically, e.g. from run day-dependent effects due to limits of detection (LOD); or it can be random as, for instance, a consequence of sample preparation. METHODS: We investigated patterns of missing data in an MS-based metabolomics experiment of serum samples from the German KORA F4 cohort (n = 1750). We then evaluated 31 imputation methods in a simulation framework and biologically validated the results by applying all imputation approaches to real metabolomics data. We examined the ability of each method to reconstruct biochemical pathways from data-driven correlation networks, and the ability of the method to increase statistical power while preserving the strength of established metabolic quantitative trait loci. RESULTS: Run day-dependent LOD-based missing data accounts for most missing values in the metabolomics dataset. Although multiple imputation by chained equations performed well in many scenarios, it is computationally and statistically challenging. K-nearest neighbors (KNN) imputation on observations with variable pre-selection showed robust performance across all evaluation schemes and is computationally more tractable. CONCLUSION: Missing data in untargeted MS-based metabolomics data occur for various reasons. Based on our results, we recommend that KNN-based imputation is performed on observations with variable pre-selection since it showed robust results in all evaluation schemes.

Circulating Levels of Interleukin 1-Receptor Antagonist and Risk of Cardiovascular Disease: Meta-Analysis of Six Population-Based Cohorts.

OBJECTIVE: Interleukin (IL)-1ß represents a key cytokine in the development of cardiovascular disease (CVD). IL-1ß is counter-regulated by IL-1 receptor antagonist (IL-1RA), an endogenous inhibitor. This study aimed to identify population-based studies on circulating IL-1RA and incident CVD in a systematic review, estimate the association between IL-1RA and incident CVD in a meta-analysis, and to test whether the association between IL-1RA and incident CVD is explained by other inflammation-related biomarkers in the MONICA/KORA Augsburg case-cohort study (Multinational Monitoring of Trends and Determinants in Cardiovascular Disease/Cooperative Health Research in the Region of Augsburg). APPROACH AND RESULTS: We performed a systematic literature search and identified 5 cohort studies on IL-1RA and incident CVD in addition to the MONICA/KORA Augsburg case-cohort study for a meta-analysis based on a total of 1855 CVD cases and 18 745 noncases with follow-up times between 5 and 16 years. The pooled standardized hazard ratio (95% confidence interval) for incident CVD was 1.11 (1.06-1.17) after adjustment for age, sex, anthropometric, metabolic, and lifestyle factors (P<0.0001). There was no heterogeneity in effect sizes (I2=0%; P=0.88). More detailed analyses in the MONICA/KORA study showed that the excess risk for CVD was attenuated by ≥10% after additional separate adjustment for serum levels of high-sensitivity C-reactive protein, IL-6, myeloperoxidase, soluble E-selectin, or soluble intercellular adhesion molecule-1. CONCLUSIONS: Serum IL-1RA levels were positively associated with risk of CVD after adjustment for multiple confounders in a meta-analysis of 6 population-based cohorts. This association may at least partially reflect a response to triggers inducing subclinical inflammation, oxidative stress, and endothelial activation.

Cerebrospinal fluid biomarkers measured by Elecsys assays compared to amyloid imaging.

INTRODUCTION: Levels of amyloid ß peptide 42 (Aß42), total tau, and phosphorylated tau-181 are well-established cerebrospinal fluid (CSF) biomarkers of Alzheimer's disease, but variability in manual plate-based assays has limited their use. We examined the relationship between CSF biomarkers, as measured by a novel automated immunoassay platform, and amyloid positron emission tomography. METHODS: CSF samples from 200 individuals underwent separate analysis for Aß42, total tau, and phosphorylated tau-181 with automated Roche Elecsys assays. Aß40 was measured with a commercial plate-based assay. Positron emission tomography with Pittsburgh Compound B was performed less than 1 year from CSF collection. RESULTS: Ratios of CSF biomarkers (total tau/Aß42, phosphorylated tau-181/Aß42, and Aß42/Aß40) best discriminated Pittsburgh Compound B-positive from Pittsburgh Compound B-negative individuals. DISCUSSION: CSF biomarkers and amyloid positron emission tomography reflect different aspects of Alzheimer's disease brain pathology, and therefore, less-than-perfect correspondence is expected. Automated assays are likely to increase the utility of CSF biomarkers.

pulver: an R package for parallel ultra-rapid p-value computation for linear regression interaction terms.

BACKGROUND: Genome-wide association studies allow us to understand the genetics of complex diseases. Human metabolism provides information about the disease-causing mechanisms, so it is usual to investigate the associations between genetic variants and metabolite levels. However, only considering genetic variants and their effects on one trait ignores the possible interplay between different "omics" layers. Existing tools only consider single-nucleotide polymorphism (SNP)-SNP interactions, and no practical tool is available for large-scale investigations of the interactions between pairs of arbitrary quantitative variables. RESULTS: We developed an R package called pulver to compute p-values for the interaction term in a very large number of linear regression models. Comparisons based on simulated data showed that pulver is much faster than the existing tools. This is achieved by using the correlation coefficient to test the null-hypothesis, which avoids the costly computation of inversions. Additional tricks are a rearrangement of the order, when iterating through the different "omics" layers, and implementing this algorithm in the fast programming language C++. Furthermore, we applied our algorithm to data from the German KORA study to investigate a real-world problem involving the interplay among DNA methylation, genetic variants, and metabolite levels. CONCLUSIONS: The pulver package is a convenient and rapid tool for screening huge numbers of linear regression models for significant interaction terms in arbitrary pairs of quantitative variables. pulver is written in R and C++, and can be downloaded freely from CRAN at https://cran.r-project.org/web/packages/pulver/ .

The dynamics of smoking-related disturbed methylation: a two time-point study of methylation change in smokers, non-smokers and former smokers.

BACKGROUND: The evidence for epigenome-wide associations between smoking and DNA methylation continues to grow through cross-sectional studies. However, few large-scale investigations have explored the associations using observations for individuals at multiple time-points. Here, through the use of the Illumina 450K BeadChip and data collected at two time-points separated by approximately 7 years, we investigate changes in methylation over time associated with quitting smoking or remaining a former smoker, and those associated with continued smoking. RESULTS: Our results indicate that after quitting smoking the most rapid reversion of altered methylation occurs within the first two decades, with reversion rates related to the initial differences in methylation. For 52 CpG sites, the change in methylation from baseline to follow-up is significantly different for former smokers relative to the change for never smokers (lowest p-value 3.61 x 10-39 for cg26703534, gene AHRR). Most of these sites' respective regions have been previously implicated in smoking-associated diseases. Despite the early rapid change, dynamism of methylation appears greater in former smokers vs never smokers even four decades after cessation. Furthermore, our study reveals the heterogeneous effect of continued smoking: the methylation levels of some loci further diverge between smokers and non-smokers, while others re-approach. Though intensity of smoking habit appears more significant than duration, results remain inconclusive. CONCLUSIONS: This study improves the understanding of the dynamic link between cigarette smoking and methylation, revealing the continued fluctuation of methylation levels decades after smoking cessation and demonstrating that continuing smoking can have an array of effects. The results can facilitate insights into the molecular mechanisms behind smoking-induced disturbed methylation, improving the possibility for development of biomarkers of past smoking behavior and increasing the understanding of the molecular path from exposure to disease.

Increased lipogenesis in spite of upregulated hepatic 5'AMP-activated protein kinase in human non-alcoholic fatty liver.

AIMS: Molecular adaptations in human non-alcoholic fatty liver disease (NAFLD) are incompletely understood. This study investigated the main gene categories related to hepatic de novo lipogenesis and lipid oxidation capacity. METHODS: Liver specimens of 48 subjects were histologically classified according to steatosis severity. In-depth analyses were undertaken using real-time polymerase chain reaction, immunoblotting, and immunohistochemistry. Lipid profiles were analyzed by gas chromatography/flame ionization detection, and effects of key fatty acids were studied in primary human hepatocytes. RESULTS: Real-time polymerase chain reaction, immunoblotting, and immunohistochemistry indicated 5'AMP-activated protein kinase (AMPK) to be increased with steatosis score ≥ 2 (all P < 0.05), including various markers of de novo lipogenesis and lipid degradation (all P < 0.05). Regarding endoplasmic reticulum stress, X-Box binding protein-1 (XBP1) was upregulated in steatosis score ≥ 2 (P = 0.029) and correlated with plasma palmitate (r = 0.34; P = 0.035). Palmitate incubation of primary human hepatocytes increased XBP1 and downstream stearoyl CoA desaturase-1 mRNA expression (both P < 0.05). Moreover, plasma and liver tissue exposed a NAFLD-related lipid profile with reduced polyunsaturated/saturated fatty acid ratio, increased palmitate and palmitoleate, and elevated lipogenesis and desaturation indices with steatosis score ≥ 2 (all P < 0.05). CONCLUSION: In humans with advanced fatty liver disease, hepatic AMPK protein is upregulated, potentially in a compensatory manner. Moreover, pathways of lipid synthesis and degradation are co-activated in subjects with advanced steatosis. Palmitate may drive lipogenesis by activating XBP1-mediated endoplasmic reticulum stress and represent a target for future dietary or pharmacological intervention.

Ultra-sensitive troponin I is an independent predictor of incident coronary heart disease in the general population.

Troponins are sensitive markers of myocardial injury and predictive of cardiovascular events, but conventional assays fail to detect slightly elevated troponins in a considerable proportion of the general population. Using a novel ultrasensitive assay, we explored the relationship of troponin levels with the incidence of coronary heart disease (CHD) in a case-cohort sample (mean age 52.5 ± 0.2 years, 51.5% women) comprising 803 CHD cases and 1942 non-cases. Ultrasensitive troponin I was detectable in 99.9% of available case-cohort samples. In an age- and sex-adjusted model, individuals in the highest quartile of the troponin distribution had a more than threefold increased risk for CHD events compared to those in the bottom quartile [hazard ratio, HR, 3.11; 95% confidence interval (CI) 2.15-4.49]. In a model adjusting for cardiovascular risk factors including C-reactive protein, cystatin C and N-terminal pro brain natriuretic peptide, individuals in the highest troponin I quartile still showed a hazard ratio of 2.58 (95% CI 1.66-4.00) for incident CHD as compared to those in the lowest quartile. Ultrasensitive troponin I was detectable in almost all individuals of a study sample reflecting middle-aged to elderly European general population. Ultrasensitive troponin concentrations exhibit an independent, graded, positive relation with incident CHD.

A statistical model for the analysis of beta values in DNA methylation studies.

BACKGROUND: The analysis of DNA methylation is a key component in the development of personalized treatment approaches. A common way to measure DNA methylation is the calculation of beta values, which are bounded variables of the form M/(M+U) that are generated by Illumina's 450k BeadChip array. The statistical analysis of beta values is considered to be challenging, as traditional methods for the analysis of bounded variables, such as M-value regression and beta regression, are based on regularity assumptions that are often too strong to adequately describe the distribution of beta values. RESULTS: We develop a statistical model for the analysis of beta values that is derived from a bivariate gamma distribution for the signal intensities M and U. By allowing for possible correlations between M and U, the proposed model explicitly takes into account the data-generating process underlying the calculation of beta values. Using simulated data and a real sample of DNA methylation data from the Heinz Nixdorf Recall cohort study, we demonstrate that the proposed model fits our data significantly better than beta regression and M-value regression. CONCLUSION: The proposed model contributes to an improved identification of associations between beta values and covariates such as clinical variables and lifestyle factors in epigenome-wide association studies. It is as easy to apply to a sample of beta values as beta regression and M-value regression.

Assessment of predictive performance in incomplete data by combining internal validation and multiple imputation.

BACKGROUND: Missing values are a frequent issue in human studies. In many situations, multiple imputation (MI) is an appropriate missing data handling strategy, whereby missing values are imputed multiple times, the analysis is performed in every imputed data set, and the obtained estimates are pooled. If the aim is to estimate (added) predictive performance measures, such as (change in) the area under the receiver-operating characteristic curve (AUC), internal validation strategies become desirable in order to correct for optimism. It is not fully understood how internal validation should be combined with multiple imputation. METHODS: In a comprehensive simulation study and in a real data set based on blood markers as predictors for mortality, we compare three combination strategies: Val-MI, internal validation followed by MI on the training and test parts separately, MI-Val, MI on the full data set followed by internal validation, and MI(-y)-Val, MI on the full data set omitting the outcome followed by internal validation. Different validation strategies, including bootstrap und cross-validation, different (added) performance measures, and various data characteristics are considered, and the strategies are evaluated with regard to bias and mean squared error of the obtained performance estimates. In addition, we elaborate on the number of resamples and imputations to be used, and adopt a strategy for confidence interval construction to incomplete data. RESULTS: Internal validation is essential in order to avoid optimism, with the bootstrap 0.632+ estimate representing a reliable method to correct for optimism. While estimates obtained by MI-Val are optimistically biased, those obtained by MI(-y)-Val tend to be pessimistic in the presence of a true underlying effect. Val-MI provides largely unbiased estimates, with a slight pessimistic bias with increasing true effect size, number of covariates and decreasing sample size. In Val-MI, accuracy of the estimate is more strongly improved by increasing the number of bootstrap draws rather than the number of imputations. With a simple integrated approach, valid confidence intervals for performance estimates can be obtained. CONCLUSIONS: When prognostic models are developed on incomplete data, Val-MI represents a valid strategy to obtain estimates of predictive performance measures.

DNA methylation and body-mass index: a genome-wide analysis.

BACKGROUND: Obesity is a major health problem that is determined by interactions between lifestyle and environmental and genetic factors. Although associations between several genetic variants and body-mass index (BMI) have been identified, little is known about epigenetic changes related to BMI. We undertook a genome-wide analysis of methylation at CpG sites in relation to BMI. METHODS: 479 individuals of European origin recruited by the Cardiogenics Consortium formed our discovery cohort. We typed their whole-blood DNA with the Infinium HumanMethylation450 array. After quality control, methylation levels were tested for association with BMI. Methylation sites showing an association with BMI at a false discovery rate q value of 0·05 or less were taken forward for replication in a cohort of 339 unrelated white patients of northern European origin from the MARTHA cohort. Sites that remained significant in this primary replication cohort were tested in a second replication cohort of 1789 white patients of European origin from the KORA cohort. We examined whether methylation levels at identified sites also showed an association with BMI in DNA from adipose tissue (n=635) and skin (n=395) obtained from white female individuals participating in the MuTHER study. Finally, we examined the association of methylation at BMI-associated sites with genetic variants and with gene expression. FINDINGS: 20 individuals from the discovery cohort were excluded from analyses after quality-control checks, leaving 459 participants. After adjustment for covariates, we identified an association (q value ≤0·05) between methylation at five probes across three different genes and BMI. The associations with three of these probes--cg22891070, cg27146050, and cg16672562, all of which are in intron 1 of HIF3A--were confirmed in both the primary and second replication cohorts. For every 0·1 increase in methylation ß value at cg22891070, BMI was 3·6% (95% CI 2·4-4·9) higher in the discovery cohort, 2·7% (1·2-4·2) higher in the primary replication cohort, and 0·8% (0·2-1·4) higher in the second replication cohort. For the MuTHER cohort, methylation at cg22891070 was associated with BMI in adipose tissue (p=1·72 × 10(-5)) but not in skin (p=0·882). We observed a significant inverse correlation (p=0·005) between methylation at cg22891070 and expression of one HIF3A gene-expression probe in adipose tissue. Two single nucleotide polymorphisms--rs8102595 and rs3826795--had independent associations with methylation at cg22891070 in all cohorts. However, these single nucleotide polymorphisms were not significantly associated with BMI. INTERPRETATION: Increased BMI in adults of European origin is associated with increased methylation at the HIF3A locus in blood cells and in adipose tissue. Our findings suggest that perturbation of hypoxia inducible transcription factor pathways could have an important role in the response to increased weight in people. FUNDING: The European Commission, National Institute for Health Research, British Heart Foundation, and Wellcome Trust.