Study participants
UK Biobank is a prospective cohort study launched between 2006 and 2010, and has enrolled over 500,000 participants from the UK. There is a comprehensive data set that includes disease diagnosis, circulating metabolites, plasma proteins, genomic sequencing data, and more.
A genome and personalized medical initiative supported by nearby Finnish individuals, Finnen Research aims to advance precision medicine by elucidating the genetic determinants of disease. This project establishes a unique resource for investigating genetic variation and their clinical impacts on the development of disease prevention, diagnosis and treatment.
A total of 333,870 participants from the UK Biobank (n = 115,078) and Finngen Biobank round 5 (n = 218,792) were included in this study, consisting of two specific subpopulations. The workflow for this study is explained in Figure 1.
Overall workflow for research. This diagram illustrates the inclusion criteria and analytical procedures for the study. ROC = receiver operating characteristic curve. NRI = Net Reclassification Index; GWAS = Genome-Wide Association Study. MR = Mendel's randomization. lasso = Minimum absolute contraction and selection operator. MR-Presso = Mendel's randomized multifaceted residuals and outliers. IDI = Integrated Identification Index.
First, metabolite data were available among the 118,002 participants in the UK biobank. Of these, 7,713 participants had diabetes at baseline. Participants with specific diabetic vascular complications at baseline identified through hospitalization records, primary care records, or self-report history during or before the baseline period were pre-excluded for specific analyses. The exclusion criteria consisted of participants with baseline diabetic complications: macrovascular complications (n = 1,957), CHD (n = 1,607), HF (n = 542), stroke (n = 226), microvascular complications (n = 1,955), dkd (n = 432), dn (n = 300), 1,626). Existing diagnosis of all target complications at baseline excluded two participants from all study cohorts. The final analytical sample consisted of 7,711 participants with longitudinal follow-up data. Specifically, eight different cohorts with longitudinal data were included in the observational study. This includes 5,756 participants with large vascular complications, 6,106 participants with CHD, 7,171 participants with HF, 7,487 participants with stroke, 5,758 participants for microvascular complications, 7,758 participants, 7,281 participants, 7,413 participants, and 7,281 participants. All 6,087 DR participants have longitudinal data available for cohort studies.
Phase II analysis included 115,078 participants from the UK biobank and 218,792 participants from Finngen Biobank (Round 5) to establish a causal relationship between selected plasma metabolites and microvascular or microphysiological complications using genome-wide association studies (GWAS) data.
Checking the endpoint
Maternal vascular complications of diabetes were defined as CHD, HF, and stroke, and microvascular complications included DR, DN, and DKD. Endpoints were determined by initially recorded occurrences across four validated sources: self-report history, hospitalization (ICD-9/ICD-10), death registration, or primary care records. At least one criterion is required for diabetes confirmation to improve diagnostic accuracy. Self-reported history, fasting plasma glucose ≥11.1 mmol/L, HBA1C ≥48 mmol/mol, or use of active glucose-lowering medication, hospitalization with self-reported history, or primary care records.
Traditional risk factors
Age, gender, race, smoking status, dietary intake, Townsend detachment index, systolic and diastolic blood pressure, body mass index (BMI), plasma triglycerides, low density lipoprotein (LDL) cholesterol (LDL) cholesterol, plasma creatinine, estimated highly refined filtration rate, plasma creatinine, sex, sex, sex, sexual fusion, AGF, AGF, AGF, AGF, detailed traditional factors and field IDs were provided in supplementary materials 1 Table S3-4 [18].
Metabolite quantification and quality control
Plasma specimens were prepared in 96-well plates containing plasma mimetics to monitor consistency of quantification by UK biobanks. A mixture of two small molecules was also added to serve as a technical reference. These samples were further analyzed at the Finland's Nightingale Health Institute between June 2019 and April 2020. A total of 249 metabolites were analyzed to profile the metabolites of these participants using nuclear magnetic resonance (NMR).
Quality control was applied to eliminate technical and systematic errors. A pre-specified agreement on the protocol was made between the UK Biobank and the Nightingale Health Centre to ensure biomarker consistency throughout the project. To track consistency and eliminate batch effects, two internalized control samples were seeded in 96-well plates equipped with participant plasma samples. Therefore, quantification of NMR differs from other approaches such as spectroscopic measurements. No batch effect provides optimal statistics [19]. Additionally, four sets of internal control samples were used in 1,352 measured 96-well plates. [20].
A total of 168 metabolites were quantified at absolute levels and include amino acids, fatty acids, glycosylated molecules, lipids, and lipoproteins, referring to mmol/L. The remaining 81 were quantified as ratios. Detailed metabolites and field IDs were provided in Supplementary Material 1 Table S5. Of all participants enrolled in this study, a total of 725 (9.4%) showed missing values for all metabolites, with no metabolites per individual, with an average of 0.3 ± 1.50. These missing values were addressed via multiple assignments implemented via R package mouse (v3.16.0).
Summary statistics for the GWAS datasets for Finngen Biobanks and UK Biobank
Metabolites were further included in confirming causal relationships with specific diabetic complications. The UK biobank cohort was included as exposure data to determine circulating plasma metabolites [21]. The Finngen Biobanks cohort was applied as outcome data to determine the presence of diabetic complications (https://finngen.gitbook.io/documentation/v/r5). Both cohorts consisted of European participants.
Summary statistics for genetic devices related to the outcome of specific diabetic complications were obtained from Finngen Biobank participants. Detailed characteristics are shown in Supplementary Material 1 Table S6. The complete GWAS results for these metabolites were published by the IEU OpenGWAS dataset.
Statistical analysis
Two steps of the analytical approach were carried out to determine the potential relationship between metabolite biomarkers and diabetic vascular complications. Before analysis, all metabolites were first converted by natural logarithm (LN)[x + 1]) and Z transforms scaled. Additionally, metabolite data above the 2.5-97.5 percentile range was identified as outliers and replaced with corresponding boundary values (the lower end 2.5 percentile or the upper end 97.5 percentile) to minimize the impact on subsequent analyses.
In the Phase I analysis, we selected several metabolites that were significantly associated with specific diabetic complications. We selected important plasma metabolites here, including individuals with longitudinal data. Minimum absolute contraction and selection operator-Cox (Lasso-Cox) regression adjusted for traditional covariates was performed to assess the effects of metabolites on complications while taking into account cross-correlations. Results represent the status of complications (present/absence) at the follow-up visit. The key metabolites associated with each complication were identified as non-zero coefficient biomarkers in Lasso-Cox regression and subsequently advanced into Phase II. Multivariate COX proportional hazard regression was performed to estimate hazard ratios (HR). Model 1 (traditional covariates: age, gender, race, smoking status, diet, Townsend detachment index, systolic and diastolic blood pressure, BMI, plasma triglycerides, LDL cholesterol, plasma creatinine and EGFR) and model 2 (combining traditional covariates with layered metabolites with mutually layered metabolites). The identification performance of the prediction algorithm was assessed using regions under the receiver operating characteristic curve (AUC), match index (C-Index), net reclassification index (NRI), and integrated identification index (IDI). NRI can quantify how newly developed models were reclassified as compared to previous models, and IDI can evaluate how models were improved. To ensure robustness of statistical analysis, the statistical power of the overall study was systematically assessed. Our power analysis demonstrated excellent statistical ability to detect associations. Observations were 1.000 for large vascular complications and 0.998 for microvascular complications.
In Phase II analysis, causal relationships were estimated and bidirectional MR was applied between metabolites selected by LassoCox regression and diabetic vascular complications. He was implemented under three assumptions. (1) Genetic instruments (IV) are related to exposure. (2) There are no other mediators between the genetic instrument and the outcome. (3) There is no correlation between genetic equipment and outcomes. Single nucleotide polymorphisms (SNPs) that did not conform to the above assumptions were excluded from the study. The implementation of these assumptions was further tested by heterogeneity tests and multifaceted testing. Heterogeneity tests were performed by applying the Cochran Q test, testing horizontal multiplexing and excluded by the MR-Egger interception test and Mendel's randomized multifaceted multifaceted multifaceted residual amount (MR-Presso). Steiger's filtering analysis was also conducted to ensure the robustness of detected causal relationships.
The following principles apply and equipment was selected: (1) Metabolite-related SNPs were included at a threshold of p <5*10–8. (2) SNPs were filtered using linkage imbalance (LD) tests except those with R2 <0.01 within a 5000 kb aggregation window size. Heterogeneity and horizontal multiplexing were tested to ensure effectiveness at a threshold of P <0.05. (3) MR-Presso tests were conducted to exclude outlier SNPs to ensure robustness of the analysis. (4) Steiger filtering was carried out to exclude SNPs with false causal correlations. Heterogeneity and horizontal multiplexing were tested to ensure effectiveness at a threshold of P <0.05. Five methods of MR were applied, including simple mode, Egger, median inverse weight (IVW), median weighting, and weighting mode. IVW applies Wald estimates to each SNP and derives aggregated genetic association between metabolite biomarkers and diabetic complications. IVW was considered fair and the most statistical value in the absence of horizontal multifaceted objects. [22]. To ensure robustness of the results, sensitivity analyses such as vacation 1-out analysis, forest plots, funnel plots, and scatter plots were performed. False detection rate (FDR) correction was also applied to eliminate bias induced by multiple tests. The selected detailed SNPs were listed in Supplementary Material 2 S1-2.
Two-tailed p-values below 0.05 were considered statistically significant. All statistical analyses were performed using R software version 4.4.2 (Basics of R Statistical Computing).
ethics
Data collection and use of this study was carried out under ethical guidelines. Prior to participation, all involved individuals were given written informed consent. Given the public and anonymous nature of the dataset, institutional review board approval was exempted for this particular analysis.
The role of the funder
This work only represents the author's perspective. Funding sources were not involved in the design or implementation of the research.