Mediation of Glycemia Exposure in on Time-to-Retinopathy in Type 2 Diabetes

Author

Brendon Chau

Published

August 18, 2025

0.1 Study Objectives

  • Identify differentially methylated CpG sites associated with glycemia exposures
  • Identify differentially methylated CpG sites associated with diabetic complications
    • retinopathy
    • nephropathy
  • Determine whether the identified CpG sites are responsible for the association between glycemia exposures and DM complications

0.2 Cohort Identification

  • Subjects must have
    • \(\ge 1\) DM Dx code or equivalent in pre-period to be included
    • \(\ge 3\) HbA1c measurements in at least two distinct years, in the 5 years prior to blood sample collection
  • In MVP, currently have a sample of 12,037 T2D subjects with methylation data and meet the inclusion criteria

0.3 Outcome Definition

  • Use available patient level EHR data to identify Type 2 diabetic subjects

    • Diagnosis codes alone may fail to capture subjects diagnosed outside of the VA

  • Diabetic Retinopathy identified by algorithm defined in (Breeyear et al. 2023):

0.4 Epigenome-Wide Association Study

  • Hypothesis: High HbA1c has a cumulative effect on DNA methylation levels

  • EWAS covariate of interest: historical HbA1c level 5 years prior to DNA methylation sample collection:

  • EWAS covariates for adjustment (base model): sex, age, cell composition, genetic principal components, potential batch effects, duration of diabetes1(Yang et al. 2024) prior to MVP blood sample collection

    • Duration of diabetes is left-censored, additional analyses were conducted with patients newly diagnosed with T2D within the VA system, with similar findings.

\[ \begin{aligned} M &\sim \text{HbA1c Exposure + sex + age + HARE ancestry}\\ &\qquad+ \text{ Time since Diabetes Dx (years) + cell composition + batch effects} \end{aligned} \]

where \(M = \log_2\left(\frac{\beta}{1 - \beta}\right)\).

0.5 Exposure Definitions

Figure 1: Hypothetical HbA1C trajectory in pre-period, where sample collection occurs nine years following initial DM diagnosis

All exposures are defined using HbA1c measurements collected from five years prior up to and including the MVP sample collection date

  • Mean HbA1c level
  • Excess HbA1c level
    • Threshold set at 6.1% HbA1c (see Miller and Orchard (2020))
  • CV of HbA1c level
Table 1: Summary Statistics for A1c Exposure Definitions
Exposure Mean SD 95%CI lb 95%CI ub
Mean (%) 7.36 1.27 5.60 10.50
Excess (%) 6.59 6.29 0.00 10.66
CV 0.10 0.08 0.02 0.30

HbA1c measurements below 4% and above 18% were excluded

1 Preliminary EWAS Results

1.1 EWAS Results - A1c - Overall

Overall Sample EWAS Manhattan Plot

1.2 EWAS Results - A1c - Overall

Overall Sample EWAS QQ Plot

1.3 EWAS Validation - A1c - Overall

Comparisons to Chen 2020

1.4 EWAS Results - A1c - Annotation

  • GREAT(McLean et al. 2010) was used to identify genes significantly enriched in glycemia associated CpG sites for each exposure at FDR < 0.05
Exposure Signif. CpGs Gene Count
Excess 1158 35
Mean 2224 63
CV 250 8
Any Exposure 2303 74
All Exposures 159 3

1.5 EWAS Results - A1c - Annotation

Gene Chr Name Found in Chen 2020
TXNIP 1 thioredoxin interacting protein (Miller et al. 2023; Tsai et al. 2022; Chen et al. 2016) Y
BRD7 16 bromodomain containing 7 N
ADCY7 16 adenylate cyclase 7 Y

  • TXNIP, BRD7, and ADCY7 were the only genes enriched in CpG sites associated with all three glycemic exposures

  • BRD7 is a regulatory gene, that acts as an activator and binds to the ESR1 promoter, and is related to histone acetylation and chromativn structure regulation

  • ADCY7 is involved in cAMP production, a cellular signaler, and mediates glucagon and incretin hormone responses that regulate blood glucose, insulin secretion, and hepatic glucose production.

  • TXNIP glucose-sensitive regulator of pancreas function, hypomethylation at TXNIP is strongly associated with glycemic exposure

2 Functional Annotation

2.1 EWAS Functional Annotation - Excess HbA1c

Hypomethylated CpGs Hypermethylated CpGs Heatmap of Chromatin States at CpGs associated with 5 Year Excess HbA1c

2.2 EWAS Functional Annotation - Mean HbA1c

Hypomethylated CpGs Hypermethylated CpGs Heatmap of Chromatin States at CpGs associated with 5 Year Excess HbA1c

2.3 EWAS Functional Annotation - CV HbA1c

Hypomethylated CpGs Hypermethylated CpGs Heatmap of Chromatin States at CpGs associated with 5 Year Excess HbA1c

2.4 EWAS Functional Annotation - Chromatin State

Chromatin State Enrichment in Blood Samples at Glycemia Associated CpGs

3 Mediation Model

3.1 Outcome Model

\[ \begin{aligned} \text{Time to Diabetic Retinopathy} &\sim \text{Exposure + DNA Methylation + sex + age + BMI}\\ &\qquad + \text{Systolic Blood Pressure}\\ &\qquad + \text{Diastolic Blood Pressure}\\ &\qquad + \text{Blood Lipids}\\ &\qquad + \text{ Time since Diabetes Dx (years)} \end{aligned} \]

  • Time to retinopathy was modeled by an Weibull accelerated failure time model

  • Baseline BMI, blood pressure, and blood lipids (Total Cholesterol, HDL-C, Triglycerides) were defined as the nearest measurement to MVP blood sample collection, up to six months post blood-sample collection

    • Subjects missing any baseline biomarkers were excluded

  • Subjects with any prior history of diabetic retinopathy were excluded

3.2 Causal Mediation Model

3.3 Causal Mediation

  • Exposure: \(A\), Glycemic Exposure

  • Mediator(s): \(M\), CpG site methylation

  • Outcome: \(T\), Time-to-Retinopathy following MVP sample collection

\[ \begin{aligned} M &= \beta_0 + \beta_1 A + \boldsymbol{\beta}_2^\top \mathbf{z} + \xi\\ \log(T) &= \theta_0 + \theta_1 A + \theta_2 M + \boldsymbol{\theta}_3^\top \mathbf{z} + \sigma \epsilon \end{aligned} \]

Causal Estimands

\[ \begin{aligned} \text{Natural Direct Effect} &: \operatorname{NDE}(a,\,a^*) = \theta_1 (a - a^*)\\ \text{Natural Indirect Effect} &: \operatorname{NIE}(a,\,a^*) = \theta_2 \beta_1 (a - a^*) \end{aligned} \] - We assume a symmetric one standard deviation difference about the sample mean for each exposure

3.4 Composite Null Hypothesis:

\[ \begin{aligned} \mathrm{H}_{01} &: \beta_1 = 0 \wedge \theta_3 \neq 0\\ \mathrm{H}_{10} &: \beta_1 \neq 0 \wedge \theta_3 = 0\\ \mathrm{H}_{01} &: \beta_1 = 0 \wedge \theta_3 = 0\\ \end{aligned} \]

  • Significance testing performed using HDMT R package (Dai, Stanford, and LeBlanc 2022)

  • Mediation P-value given by the maximum of the p-values associated with testing \(\beta_1 = 0\) and \(\theta_2 = 0\)

  • Max-P statistic: \(p_{\max} = \max\{\mathrm{Pr}[\beta_1 = 0],\,\mathrm{Pr}[\theta_2 = 0]\}\)

  • \(p_{\max}\) is not uniformly distributed under the null, HDMT procedure estimates the proportion of each type of null \(\pi_{01}, \pi_{10}, \pi_{00}\) to control family-wise error and false discovery rate

3.5 Causal Mediation - Indirect Effects

Figure 2: Indirect Effect Manhattan Plot, Max-P

3.6 Causal Mediation - Indirect Effects

Figure 3: Indirect Effect QQ Plot, Max-P
  • Max-P statistic is severely underdispersed relative to a uniform null distribution
Table 2: Summary of Direct Effect and Proportion Mediated by Significant Mediators
Exposure NDE Estimate Proportion Mediated
Excess -0.500 (-0.504, -0.498) 0.0121 (0.00997, 0.0143)
Mean -0.535 (-0.538, -0.531) 0.0108 (0.00990, 0.0118)
CV -0.197 (-0.200, -0.192) 0.0221 (0.01232, 0.0495)

3.7 Causal Mediation - Volcano Plot

Figure 4: Indirect Effect versus Negative \(\log_{10}(p_{\max})\), red indicates signficant mediators at FDR < 0.05
  • At all significant mediators, increased glycemic exposure is associated with a decrease in time-to-retinopathy

3.8 Causal Mediation - Indirect Effects

Figure 5: Indirect Effect Plot, Top 10 Marginally Significant Mediators

  • Expected changes in methylation given changes in glycemic exposure appear to be relatively small

  • Individual CpG sites do not capture much of a mediation effect of glycemia on time-to-retinopathy

  • Mean and excess HbA1c effect on retinopathy onset is marginally mediated by methylation at two and four CpG sites, respectively

    • Mean and excess HbA1c are associated withcg04418434 is hypermethylated in the 5’UTR region of RREB1
    • cg04418434 is hypermethylated in the 5’UTR region of RREB1

  • CV HbA1c’s effect on DR onset is mediated at 87 CpG sites

3.9 Causal Mediation Annotation

  • Top genes with CpG sites that mediate glycemic exposure on time to retinopathy
Gene Name CHR Mean Excess CV Gene Region
ARF1 ADP Ribosylation Factor 1 1 X 5’UTR
MALAT1 Metastasis Associated Lung Adenocarcinoma Transcript 1 11 X Body
ELF1 E74 Like ETS Transcription Factor 1 13 X 5’UTR
XYLT1 Xylosyltransferase 1 16 X X Body
HK2 Hexokinase 2 2 X Body
DGUOK-AS1 DGUOK Antisense RNA 1 2 X X Body
RREB1 Ras Responsive Element Binding Protein 1 6 X X 5’UTR
KLF9 KLF Transcription Factor 9 9 X Body

  • ARF1, MALAT1, ELF1, RREB1, KLF9 all have either been found to be expressed in retinal tissue or have also been shown to be associated with retinal disease or macular degeneration

  • XYLT1 has been previously been associated with age-related macular degeneration

  • DGUOK-AS1 has been previously been found to be associated with several cancers, but association with retinopathy appears to be novel

3.10 Further Work

  • Mediation analysis of time to renal disease onset in T2D is ongoing

  • Joint mediation analysis to be performed considering all marginally significant mediators together, to capture the overall mediaiton effect

  • Mediation analysis of DR progression: 16-17% of T2D patients with retinopathy in MVP prior to blood sample collection later develop diabetic macular edema, an advanced complication of retinopathy

References

Breeyear, Joseph H, Sabrina L Mitchell, Cari L Nealon, Jacklyn N Hellwege, Brian Charest, Anjali Khakaria, Christopher W Halladay, et al. 2023. “Development of Portable Electronic Health Record Based Algorithms to Identify Individuals with Diabetic Retinopathy.” medRxiv, 2023–11.
Chen, Zhuo, Feng Miao, Andrew D Paterson, John M Lachin, Lingxiao Zhang, Dustin E Schones, Xiwei Wu, et al. 2016. “Epigenomic Profiling Reveals an Association Between Persistence of DNA Methylation and Metabolic Memory in the DCCT/EDIC Type 1 Diabetes Cohort.” Proceedings of the National Academy of Sciences 113 (21): E3002–11.
Dai, James Y, Janet L Stanford, and Michael LeBlanc. 2022. “A Multiple-Testing Procedure for High-Dimensional Mediation Hypotheses.” Journal of the American Statistical Association 117 (537): 198–213.
McLean, Cory Y, Dave Bristor, Michael Hiller, Shoa L Clarke, Bruce T Schaar, Craig B Lowe, Aaron M Wenger, and Gill Bejerano. 2010. “GREAT Improves Functional Interpretation of Cis-Regulatory Regions.” Nature Biotechnology 28 (5): 495–501.
Miller, Rachel G, Josyf C Mychaleckyj, Suna Onengut-Gumuscu, Trevor J Orchard, and Tina Costacou. 2023. “TXNIP DNA Methylation Is Associated with Glycemic Control over 28 Years in Type 1 Diabetes: Findings from the Pittsburgh Epidemiology of Diabetes Complications (EDC) Study.” BMJ Open Diabetes Research and Care 11 (1): e003068.
Tsai, Hao-Hung, Chao-Yu Shen, Chien-Chang Ho, Shu-Yi Hsu, Disline Manli Tantoh, Oswald Ndi Nfor, Shin-Lin Chiu, Ying-Hsiang Chou, and Yung-Po Liaw. 2022. “Interaction Between a Diabetes-Related Methylation Site (TXNIP Cg19693031) and Variant (GLUT1 Rs841853) on Fasting Blood Glucose Levels Among Non-Diabetics.” Journal of Translational Medicine 20 (1): 87.
Yang, Peter K, Sandra L Jackson, Brian R Charest, Yiling J Cheng, Yan V Sun, Sridharan Raghavan, Elizabeth M Litkowski, et al. 2024. “Type 1 Diabetes Genetic Risk in 109,954 Veterans with Adult-Onset Diabetes: The Million Veteran Program (MVP).” Diabetes Care 47 (6): 1032–41.

Footnotes

  1. interval between the date of the first DM diagnosis code to MVP sample collection date↩︎