Association between life course adiposity and diabetic microvascular complications: A univariable and multivariable Mendelian randomization study

Abstract

Our study aims to investigate the causal effects of life course adiposity on diabetic microvascular complications. Life course adiposity was considered as the exposure, including adult body mass index (BMI), childhood BMI, and birth weight, and diabetic microvascular complications were considered as the outcome, including diabetic kidney disease (DKD), diabetic retinopathy (DR), and diabetic neuropathy (DN). We utilized large-scale genome-wide association study datasets of European ancestry to conduct univariable Mendelian randomization and multivariable Mendelian randomization analyses to estimate the independent causal effects of life course adiposity on diabetic microvascular complications. Univariable Mendelian randomization analyses revealed a causal effect of high adiposity in adults and children on an increased risk of DKD, DR, and DN. Similarly, there was a causal effect of low birth weight on an increased risk of DKD and DR, but no causal relationship was found between birth weight and DN. In multivariable Mendelian randomization analyses, the causal effect of adult BMI on DN became nonsignificant, while it remained significant for both DKD and DR (DKD: inverse-variance weighted (IVW) odds ratio (OR) = 1.142, 95% confidence interval (CI) 1.029–1.829, P = .012; DR: IVW OR = 1.108, 95% CI 1.022–1.200, P = .012). Birth weight continued to demonstrate an independent causal effect on both DKD and DR (DKD: IVW OR = 0.583, 95% CI 0.430–0.790, P < .001; DR: IVW OR = 0.614, 95% CI 0.486–0.777, P < .001). Genetically predicted life course adiposity demonstrates a causal relationship with diabetic microvascular complications. Low birth weight and high adult BMI independently increase the risk of developing DKD and DR.

1. Introduction

Diabetes mellitus (DM) is now recognized as the third major noncommunicable disease threatening human health, following cardiovascular diseases and malignancies.^[1] Diabetic microvascular complications including diabetic kidney disease (DKD), diabetic retinopathy (DR), and diabetic neuropathy (DN) are severe consequences of DM. The rising global prevalence of DM has also led to an increase in cases of DKD, DR, and DN. DKD is a leading cause of end-stage renal disease worldwide, with its proportion increasing to 31.3%.^[2] DR is a leading cause of blindness and visual impairment in adults.^[3] DN is the most common diabetic microvascular complication, primarily affecting the peripheral nerves. Diabetic peripheral neuropathy increases the risk of falls and foot ulcers.^[4]

Several potential risk factors for diabetic microvascular complications have been identified, primarily including long duration of diabetes, poor glycemic control, dyslipidemia, hypertension, and smoking.^[5] Previous cohort studies have provided supportive evidence for an association between adult body mass index (BMI) and diabetic microvascular complications.^[6] Conversely, another study conducted in Asia observed a negative correlation between BMI and DR.^[7] Research on the relationship between childhood BMI and birth weight with diabetic microvascular complications is scarce, often involving small studies with insufficient evidence levels and contradictory conclusions.^[8,9] These issues may be attributed to the limitations of observational studies or ethical issues. Current clinical research lacks sufficient evidence to definitively establish a direct causal link between life course adiposity and the development of diabetic microvascular complications.

Mendelian randomization (MR) is a powerful analytical method used to test for causal relationships. In MR, genetic variants are treated as instrumental variables (IVs). Compared to traditional observational analyses, MR offers unique advantages.^[10] It reduces confounding bias through randomly assigned genetic variants and avoids reverse causation as genetic traits are determined before birth.^[11] Recently, MR has been extensively used to elucidate the etiology of diabetic microvascular complications, playing a significant role. Our study aims to analyze the causal relationship of life course adiposity on diabetic microvascular complications.

2. Materials and methods

2.1. Study design

Figure 1 illustrates the design of our MR study. Our study is based on genome-wide association study (GWAS) datasets, with exposure defined as life course adiposity, including birth weight, childhood BMI, and adult BMI. The outcomes are diabetic microvascular complications, comprising DKD, DR, and DN. Initially, univariable Mendelian randomization (UVMR) is used to analyze the overall causal relationship of life course adiposity on diabetic microvascular complications, followed by multivariable Mendelian randomization (MVMR) to investigate potential independent causal relationships. Our study was conducted in accordance with the “Strengthening the Reporting of Observational Studies in Epidemiology using Mendelian Randomization” guidelines.^[12]

Figure 1.

Overview of the MR analysis design. MR = Mendelian randomization, MVMR = multivariable Mendelian randomization, UVMR = Univariable Mendelian randomization.

2.2. Data sources

Summary statistics for adult BMI were obtained from the MRCIEU GWAS database, with the sample including 99,998 individuals. Summary statistics for childhood BMI resulted from a 2-stage meta-analysis conducted by Vogelezang et al on 41 studies involving BMI measurements in European children aged 2 to 10 years,^[13] with a sample size of 39,620. Summary statistics related to newborn birth weight were derived from a meta-analysis by Warrington et al using data from the Early Growth Genetics Consortium and the UK Biobank.^[14] This research collected birth weight information through neonatal records, obstetric medical records, interviews with mothers, or self-reports from individuals when they reached adulthood. Participants with a fetal age of <37 weeks or a birth weight outside the normal range (<2.5 or >4.5 kg) were excluded, resulting in a sample size of 297,356 individuals.

The original data for DKD, DR, and DN were derived from the Finngen database (R9). The summary data used for outcomes encompassed all types of diabetes patients. Disease diagnoses were defined using the International Classification of Diseases (ICD) codes, with DKD identified by ICD-10 code N08.3.DN was classified using ICD-10 codes E1[0-4]4, ICD-9 code 2505, and ICD-8 code 25005. The DKD cohort included 4111 cases and 308,539 controls, the DR cohort consisted of 10,413 cases and 308,633 controls, and the DN cohort comprised 2843 cases and 271,817 controls.

Our study utilized exposure and outcome cohorts composed of individuals of European ancestry, ensuring no sample overlap. Specific data sources are detailed in Table S1, Supplemental Digital Content, https://links.lww.com/MD/R605. Since this study involves a reanalysis of previously collected and published data, no additional ethical approval is required.

2.3. Instruments selection

IVs must satisfy the following 3 core assumptions. First, relevance assumption: IVs must be strongly associated with the exposure under study. Second, independence assumption: IVs are not associated with any other confounding factors that affect the outcome. Third, exclusion restriction: IVs influence the outcome solely through their effect on the exposure and are not otherwise related to the outcome.^[15] Based on these assumptions, genetic variants significantly associated with phenotype were extracted from the exposure GWAS dataset to serve as IVs, with selection criteria set at P < 5 × 10⁻⁸. The linkage disequilibrium coefficient was set to r² < 0.001 and the regional width to 10,000 kb. The proportion of variance in the exposure explained by single nucleotide polymorphisms (SNPs) was quantified using R², calculated as R²=2 × (1-MAF)×MAF×β², where MAF denotes the frequency of the effect allele for a given SNP in the population. The degree of variation explained by all IVs was assessed by summing the R² of all SNPs. Using the F statistic calculated as $F=[R^2\times (n-1)]÷(1-R^2)$, where n represents the sample size of the exposure GWAS, SNPs with F < 10 were excluded to prevent weak IVs did not affect the causal estimates. The exposure IVs were matched with outcome GWAS data, excluding SNPs that had significant associations in the outcome (P < 5 × 10⁻⁸) and finally excluding palindromic SNPs that showed incompatibility due to unclear chain orientation.

2.4. Statistical analyses

We employed UVMR analysis to evaluate the overall causal association between life course adiposity and diabetic microvascular complications. Three statistical methods were utilized. The inverse-variance weighted (IVW) method offers the highest statistical power among all MR methods and served as the primary analytical approach in this study. If there is heterogeneity, random-effects IVW models are applied; otherwise, the fixed-effect IVW model is applied.^[10] Additionally, MR-Egger regression and the weighted median (WM) method were used as complementary MR analytical methods. To ensure the reliability of the results, a series of sensitivity analyses were conducted. These included using Cochran's Q statistic to assess heterogeneity.

Additionally, MR-Egger regression, Mendelian Randomization Pleiotropy RESidual Sum and Outlier (MR-PRESSO), and Leave-one-out analysis were performed to test for horizontal pleiotropy. If MR-PRESSO identifies anomalous outliers, we remove these and repeat the MR analyses.^[16]

To investigate the independent impact of life course adiposity on the risk of microvascular complications in diabetes, we employed MVMR analysis. This analysis considered adult BMI, childhood BMI, and birth weight as risk factors for diabetic microvascular complications, while also accounting for the interactions among these 3 exposures. We implemented the IVW, Mendelian Randomization – Least Absolute Shrinkage and Selection Operator (MR-LASSO), and WM methods to address the issue of collinearity among the variables. The MR-Egger intercept P value indicates horizontal pleiotropy, ensuring the stability and accuracy of our study.

In our study, exposures were treated as continuous variables, and low birth weight was defined as a value within the normal birth weight range that is on the lower side. The outcome of the study was a binary variable, and the effect sizes were measured using odds ratio (OR). To control for the false discovery rate (FDR) in multiple comparisons, an FDR threshold of 0.05 was set. Crude P values <.05 but adjusted P values >.05 were considered to provide suggestive evidence of an association. All analyses were conducted using the R software version 4.3.2 (R Foundation for Statistical Computing, Vienna, Austria) .

3. Results

3.1. UVMR results

Following the condition screening and MR-PRESSO tests, we identified SNPs associated with adult BMI, childhood BMI, and birth weight as IVs, as detailed in Tables S2–S4, Supplemental Digital Content, https://links.lww.com/MD/R605.

There was an increased risk of DKD in adults with high adiposity (OR = 1.210, 95% confidence interval (CI) 1.145–1.279, P < .001), with consistent results obtained using both MR-Egger and WM methods, as shown in Figure 2 and Table S5, Supplemental Digital Content, https://links.lww.com/MD/R605. Similarly, the risk of DR was elevated in adults with high adiposity (OR = 1.145, 95% CI 1.082–1.212, P < .001), with consistent findings from the WM method. Although the MR-Egger regression did not indicated a significant causal relationship, its direction aligned with that of IVW. Additionally, the risk of DN was increased in adults with high adiposity (OR = 1.142; 95% CI 1.055–1.237; P = .001), with consistent results from both MR-Egger and WM methods.

Figure 2.

Forest plot depicting UVMR results for the association of life course adiposity with diabetic microvascular complications. CI, confidence intervals, BMI = body mass index, DKD = diabetic kidney disease, DN = diabetic neuropathy, DR = diabetic retinopathy, FDR = false discovery rate, IVW = inverse-variance weighted, No. SNPs = number of SNPs used in MR, OR = odds ratio. WM = weighted median. Blue plots represent P > .05. Green plots and asterisk (*) represent FDR-adjusted P < .1. Orange plots and double asterisk (**) represent FDR-adjusted P < .05.

An increased risk of DKD was observed in children with high adiposity (OR = 1.961, 95% CI 1.599–2.406, P < .001), with consistent findings from the WM method. The direction of the effect of MR-Egger regression was consistent with IVW. For DR, the risk was also elevated in children with high adiposity (OR = 1.505, 95% CI 1.196–1.890, P < .001), with consistent results from the WM method. The direction of MR-Egger regression aligned with IVW. Additionally, the risk of DN in children with high adiposity was increased (OR = 1.759, 95% CI 1.304–2.374, P < .001), with consistent results from WM methods. MR-Egger regression had the consistent direction.

Individuals with low birth weight exhibited an increased risk of DKD (OR = 0.701, 95% CI 0.531–0.926, P = .012). The effects of MR-Egger and WM methods were consistent with IVW. For DR, there was an increased risk associated with low birth weight (OR = 0.831, 95% CI 0.702–0.983, P = .031), but the P value adjusted for FDR is 0.093, which suggests a potential causal relationship. The MR-Egger method and the WM method aligned directionally with IVW, indicating similar trends. No significant causal association was found between low birth weight and the risk of DN (OR = 0.840; 95% CI 0.629–1.122; P = .240).

Some Cochran's Q tests indicated heterogeneity, as shown in Figure 2. However, as the analysis primarily utilized the random effects IVW method, this heterogeneity was considered acceptable.^[10] All MR-Egger intercept tests did not detect pleiotropy. The scatterplot illustrating the causal relationship between life course adiposity and the risk of diabetic microvascular complications is presented in Figure S1, Supplemental Digital Content, https://links.lww.com/MD/R606. The funnel plot exhibited basic symmetry, and the leave-one-out method failed to identify any suggestive biased SNPs, as depicted in Figures S2 and S3, Supplemental Digital Content, https://links.lww.com/MD/R606. These findings indicate that the causal associations inferred from MR analyses are robust.

3.2. MVMR results

Following conditional screening, we compiled SNPs associated with adult BMI, childhood BMI, and birth weight as IVs for the MVMR analysis, as detailed in Tables S6–S8, Supplemental Digital Content, https://links.lww.com/MD/R605.

After adjusting for childhood BMI and birth weight, the causal association between adult BMI and DKD (OR = 1.142; 95% CI 1.029, 1.266; P = .012) and DR (OR = 1.108; 95% CI 1.022, 1.200; P = .012) was attenuated yet remained significant. These findings were supported by WM, MR-Egger, and MR-LASSO methods, as shown in Figure 3 and Table S9, Supplemental Digital Content, https://links.lww.com/MD/R605. The association between adult BMI and DN was no longer significant (OR = 1.074; 95% CI 0.958, 1.204; P = .220).

Figure 3.

Forest plot depicting MVMR results for the association of life course adiposity with diabetic microvascular complications. BMI = body mass index, CI = confidence intervals, DKD = diabetic kidney disease, DN = diabetic neuropathy, DR = diabetic retinopathy, FDR = false discovery rate, IVW = inverse-variance weighted, No. SNPs = number of SNPs used in MR, OR = odds ratio WM = weighted median. Blue plots represent P > .05. Orange plots and double asterisk (**) represent FDR-adjusted P < .05.

After adjustments for adult BMI and birth weight, the causal associations between childhood BMI and DKD (OR = 1.247; 95% CI 0.850, 1.829; P = .259), DR (OR = 1.225; 95% CI 0.933, 1.686; P = .133), and DN (OR = 1.359; 95% CI 0.892, 2.069; P = .153) were no longer significant.

Upon adjusting for adult BMI and childhood BMI, the causal associations between birth weight and DKD (OR = 0.583; 95% CI 0.430, 0.790; P < .001) and DR (OR = 0.614; 95% CI 0.486, 0.777; P < .001) remained significant, with both statistical power and effect size showing marked increases. The findings from WM, MR-Egger, and MR-LASSO were consistent with the IVW results and all showed significance. The MVMR results underscore the independent and substantial impact of birth weight on DKD and DR, indicating that birth weight has the most pronounced effect. No causal relationship was found between birth weight and DN (OR = 0.764; 95% CI 0.547, 1.067; P = .114).

Cochran's Q test indicated heterogeneity in the results, as shown in Figure 3. The MR-Egger intercept test did not reveal pleiotropy, suggesting that the findings are not influenced by horizontal pleiotropy.

4. Discussion

Our MR study evaluated the risk associations between life course adiposity and diabetic microvascular complications. UVMR analysis showed that genetically predicted adiposity in both adulthood and childhood had an overall detrimental impact on the risk of DKD, DR, and DN. Within the normal birth weight range, genetically predicted low birth weight increased the risk of DKD and DR. Further MVMR analysis revealed that the causal associations between adult adiposity and the risk of DKD and DR remained significant. However, the association with DN became nonsignificant. Adjusted childhood BMI showed no significant causal effects on DKD, DR, or DN. After accounting for the confounding effects of adult and childhood adiposity, the impact of birth weight on DKD and DR remained significant, reinforcing the independent and substantial role of birth weight in the risks of these complications. These findings suggest that low birth weight and high adult BMI may independently contribute to an increased risk of developing DKD and DR.

Despite obesity being a well-known risk factor for diabetes, its relationship with diabetic microvascular complications remains unclear. Previous studies showed conflicting results. Analysis of the Korea National Health and Nutrition Examination Survey data showed a negative correlation between BMI and DR, suggesting that high fat may protect against DR.^[17] A cross-sectional study of multi-ethnic Asian adults found that both BMI and obesity showed an inverse association with DR.^[7] However, a retrospective study considered the protective effect of obesity on DR was not independent and might be confounded by multiple factors.^[18] A meta-analysis incorporating 13 prospective cohort studies showed that obesity was associated with a significant increase in DR incidenc.^[19] Previous research has generally supported the view that obesity is a risk factor for DKD and DN. A post hoc analysis of the Action to Control Cardiovascular Risk in Diabetes study found that obesity was associated with the progression of chronic kidney disease and neuropathy in type 2 diabetes mellitus participants.^[20] A systematic review and meta-analysis of 20 cohorts included BMI in DKD risk prediction models.^[21] Studies on childhood BMI and microvascular complications are limited. A small study with 65 children found no significant increase in retinal microvascular damage in obese children.^[22] The impact of birth weight on DKD and DR remains uncertain. A Finnish cross-sectional study showed that low birth weight does not impact DKD and DR in Caucasians with type 1 diabetes.^[23] Two case-control studies showed no significant difference in birth weight between diabetic nephropathy patients and controls.^[24,25] The Atherosclerosis Risk in Communities study found no association between birth weight and the risk of DR in type 2 diabetes.^[9] However, a study by Preston involving 236 nondiabetic and diabetic subjects found that individuals with microalbuminuria had lower birth weights.^[8] A German cohort study found that diabetic patients with low birth weight had the highest risk of DR.^[6] A meta-analysis indicated that low birth weight increased the risk of chronic kidney disease, while high birth weight had no significant impact.^[26] Our research used MR analyses to provide new evidence of a causal association between life course adiposity and diabetic microvascular complications.

Obesity is considered a potential risk factor for the onset of diabetic microvascular complications. This relationship may be influenced by mechanisms such as insulin resistance, low-grade inflammatory responses, and oxidative stress.^[27] Under diabetic conditions, the phenotype of perivascular adipose tissue shifts towards a pro-inflammatory, pro-oxidative, and vasoconstrictive state.^[28] Excessive dysfunctional adipose tissue may release fatty acids and adipokines, contributing to further microvascular endothelial dysfunction.^[29] Our findings indicate that low birth weight is a primary factor increasing the risk of DKD and DR. This is supported by the hypothesis of The Fetal Origins of Adult Disease, which suggests that early development events are associated with the risk of many adult diseases.^[30] Several mechanisms may explain the negative correlation between low birth weight and both DKD and DR. Reduced fetal growth rate may be a predictive adaptation to adverse developmental environments, potentially inducing a set of phenotypes referred to as “thrifty,” affecting organ development and leading to altered physiological and metabolic set points. When mismatched with a nutrient-rich postnatal environment, this may further promote obesity, insulin resistance, and metabolic syndrome.^[31] Early embryonic development is critical for establishing epigenetic patterns, with intrauterine growth shown to cause epigenetic changes in the fetal genome, especially alterations in DNA methylation patterns, which may lead to susceptibility to environmental factors and complex diseases in adulthood.^[32,33] Moreover, slow intrauterine growth can affect hormone secretion and tissue sensitivity to hormones, permanently altering the body hormonal responses. For example, elevated glucocorticoid levels can influence fetal metabolic programming, leading to persistent metabolic pathway changes into adulthood.^[34] Considering the potential correlation between birth weight and diabetic microvascular complications, research in this area helps us better understand the early origins of these complications.

From a public health perspective, our findings hold significant implications. Birth weight may play an early determining role in the development of DKD and DR, and it explains some of the effects of adult obesity. Birth weight can serve as an early screening marker for individuals at high risk of DKD and DR, facilitating the development of personalized prevention and intervention strategies for this population. Furthermore, our study highlights the importance of prenatal health management, suggesting that appropriately increasing birth weight may provide long-term health benefits.^[35,36] However, despite these potential advantages, the possible adverse effects of high birth weight should also be considered. A more cautious approach is needed to balance the risks and benefits, ensuring that public health interventions are more precise.

We gathered large-scale GWAS summary datasets from European populations, offering a much larger sample size than typical epidemiological studies. Our MR analyses minimize reverse causation and confounding factors, providing reliable causal relationship estimates. Our results were evaluated using various MR analysis methods and underwent extensive quality control to ensure consistency and robustness. However, the study had several limitations. Firstly, despite using multiple methods to eliminate pleiotropy from multi-effect SNPs, unobserved pleiotropy may still exist, and heterogeneity may reduce the precision of the results, so caution should be exercised when interpreting the findings. Secondly, biases from nonrandom sample selection or inconsistent GWAS experimental designs could affect our outcomes’ accuracy. Thirdly, our study primarily used data from European populations, so it is uncertain whether these findings can be generalized to other ethnic groups. Additionally, since our study focuses on binary outcomes, this limits the applicability of nonlinear MR methods. Nonlinear MR typically requires individual-level data to model more complex, nonlinear relationships between variables. Since our analysis is based on summary statistics, which do not include individual-level data, we were unable to investigate the nonlinear relationship between life course adiposity and diabetic microvascular complications. While the MR method offers unique advantages in estimating causal relationships, further studies are necessary.

5. Conclusion

Our study provides genetic evidence for a causal relationship between life course adiposity and the risk of diabetic microvascular complications, and further substantiates low birth weight and high adult BMI may independently contribute to an increased risk of developing DKD and DR. This findings provides a new perspective on prevention strategies for diabetic microvascular complications.

Acknowledgments

The authors sincerely acknowledge all the researchers who freely shared their research data, as well as the websites that collected and shared these data.

Author contributions

Conceptualization: Jun Ma, Meng Hao, Jie Yao, Yuwen Li, Ting Liu.

Data curation: Jun Ma, Meng Hao, Jie Yao, Yuwen Li.

Formal analysis: Jun Ma, Meng Hao.

Investigation: Ting Liu.

Methodology: Jun Ma, Ting Liu.

Project administration: Ting Liu.

Resources: Ting Liu.

Software: Jun Ma, Meng Hao, Jie Yao, Yuwen Li, Ting Liu.

Supervision: Ting Liu.

Visualization: Jun Ma, Meng Hao, Yuwen Li.

Validation: Meng Hao, Jie Yao, Yuwen Li, Ting Liu.

Writing – original draft: Jun Ma.

Writing – review & editing: Meng Hao, Jie Yao, Yuwen Li, Ting Liu.