Investigating the association between fasting insulin, erythrocytosis and HbA1c through Mendelian randomization and observational analyses

Background Insulin resistance (IR) with associated compensatory hyperinsulinemia (HI) are early abnormalities in the etiology of prediabetes (preT2D) and type 2 diabetes (T2D). IR and HI also associate with increased erythrocytosis. Hemoglobin A1c (HbA1c) is commonly used to diagnose and monitor preT2D and T2D, but can be influenced by erythrocytosis independent of glycemia. Methods We undertook bidirectional Mendelian randomization (MR) in individuals of European ancestry to investigate potential causal associations between increased fasting insulin adjusted for BMI (FI), erythrocytosis and its non-glycemic impact on HbA1c. We investigated the association between the triglyceride-glucose index (TGI), a surrogate measure of IR and HI, and glycation gap (difference between measured HbA1c and predicted HbA1c derived from linear regression of fasting glucose) in people with normoglycemia and preT2D. Results Inverse variance weighted MR (IVWMR) suggested that increased FI increases hemoglobin (Hb, b=0.54 ± 0.09, p=2.7 x 10-10), red cell count (RCC, b=0.54 ± 0.12, p=5.38x10-6) and reticulocyte (RETIC, b=0.70 ± 0.15, p=2.18x10-6). Multivariable MR indicated that increased FI did not impact HbA1c (b=0.23 ± 0.16, p=0.162) but reduced HbA1c after adjustment for T2D (b=0.31 ± 0.13, p=0.016). Increased Hb (b=0.03 ± 0.01, p=0.02), RCC (b=0.02 ± 0.01, p=0.04) and RETIC (b=0.03 ± 0.01, p=0.002) might modestly increase FI. In the observational cohort, increased TGI associated with decreased glycation gap, (i.e., measured HbA1c was lower than expected based on fasting glucose, (b=-0.09 ± 0.009, p<0.0001)) in people with preT2D but not in those with normoglycemia (b=0.02 ± 0.007, p<0.0001). Conclusions MR suggests increased FI increases erythrocytosis and might potentially decrease HbA1c by non-glycemic effects. Increased TGI, a surrogate measure of increased FI, associates with lower-than-expected HbA1c in people with preT2D. These findings merit confirmatory studies to evaluate their clinical significance.


Supplementary File 1 -Cohorts
Summary statistics from the largest published genome wide association study (GWAS) in people of European ancestry were used in MR analyses (Supplementary Table 1) (18)(19)(20)(21)(22)(23)(24). Informed consent and institutional approval were previously obtained by the individual cohort investigators. Background Explain the scientific background and rationale for the reported study. What is the exposure? Is a potential causal relationship between exposure and outcome plausible? Justify why MR is a helpful method to address the study question 4 Increasingly, haemoglobin A1c (HbA1c) has replaced fasting glucose and/or the 75 g oral glucose tolerance test to diagnose preT2D, T2D and T2D remission. HbA1c is also used to set glycemic targets for people with diabetes (7-9). Advantages to using HbA1c compared to fasting glucose, include convenience and use of an assay that is standardized, stable, reproducible with limited intraindividual variability (1,10). It provides an average measure of glycemia in the preceding 2 to 3 months (1). However, altered red cell lifespan and erythrocytosis, which is not routinely assessed, can affect HbA1c measurement by nonglycemic pathways, which has implications in patients with red cell disorders and haemoglobinopathies (1,11). In people without T2D, including those with preT2D, non-glycemic parameters are a major predictor of HbA1c: higher haemoglobin associates with lower HbA1c (12,13). Observational studies have also shown an association between IR/HI and increased haemoglobin and red cell count (14-16), but whether this association is causal is not established, nor is it known whether this impacts HbA1c measurement by non-glycemic pathways. 3

Objectives
State specific objectives clearly, including pre-specified causal hypotheses (if any). State that MR is a method that, under specific assumptions, intends to estimate causal effects 5 We undertook bidirectional MR to investigate potential causal associations between fasting insulin (FI) and erythrocytosis (haemoglobin, red cell count and reticulocyte count: primary outcome) in people of European ancestry using summary statistics from the largest genome wide association studies (GWAS). For our secondary outcome, we undertook multivariable MR to assess the nonglycemic effects of FI on HbA1c after adjusting for 2 elevated fasting glucose (FG) and type 2 diabetes (T2D).

Study design and data sources
Present key elements of the study design early in the article. Consider including a table listing sources of data for all phases of the study. For each data source contributing to the analysis, describe the following: Summary statistics from the largest published genome wide association study (GWAS) in people of European ancestry were used in MR analyses a) Setting: Describe the study design and the underlying population, if possible. Describe the setting, locations, and relevant dates, including periods of recruitment, exposure, follow-up, and data collection, when available.

5-6
Summary statistics from the largest published genome wide association study (GWAS) in people of European ancestry were used in MR analyses

Assessment of assumptions
Describe any methods or prior knowledge used to assess the assumptions or justify their validity 4 In people without T2D, including those with preT2D, non-glycemic parameters are a major predictor of HbA1c: higher haemoglobin associates with lower HbA1c (12,13). Observational studies have also shown an association between IR/HI and increased haemoglobin and red cell count (14-16), but whether this association is causal is not established, nor is it known whether this impacts HbA1c measurement by non-glycemic pathways. 8

Sensitivity analyses and additional analyses
Describe any sensitivity analyses or additional analyses performed (e.g. comparison of effect estimates from different approaches, independent replication, bias analytic techniques, validation of instruments, simulations) 6-7 For univariable MR, we used inverse weighted MR (IVWMR) and additional sensitivity analyses including MR-Egger, weighted median, weighted mode and leave-one-out analyses.
Cochrane's Q test was used to assess heterogeneity, while leave-one-out analyses were conducted to assess if any MR estimate was biased by a single SNP potentially with horizontal pleiotropic effect (17) and the F statistic was calculated to assess the strength of the instrument exposure (17,30,31).

9
Software and preregistration 4 a) Name statistical software and package(s), including version and settings used 7 Univariable MR was conducted using the "TwoSampleMR" package in R (R studio® v1.3.1073 and R® v4.0.3). Linkage disequilibrium (LD) pruning was used to select a proxy (r2>0.8) if a SNP was not directly matched from the 1000 Genomes project (Version 0.5.6, Released 2021-03-35). The "ggplot2" and "metaphor" packages in R were used to create plots. We undertook inverse variance weighted multivariable MR (IVW Multivariable MR) to assess the effect of FI on HbA1c after adjustment for FG and T2D as well as Hb (32). Multivariable MR was conducted using both the "TwoSampleMR", "Multivariable MR" and "RMultivariable MR" packages in R (R studio® v1.3.1073 and R® v4.0.3) where the latter two packages assessed heterogeneity via Cochrane's Q test and strength of the instrument via F statistics (30,32). Plots were; generated using "plotobject".
b) State whether the study protocol and details were pre-registered (as well as when and where) The study was not pre-registered.

10
Descriptive data a) Report the numbers of individuals at each stage of included studies and reasons for exclusion. Consider use of a flow diagram 18 12

Assessment of assumptions
a) Report the assessment of the validity of the assumptions 8-9 MR-Egger intercept with p-value was reported as measure of horizontal pleiotropy for all significant associations.
b) Report any additional statistics (e.g., assessments of heterogeneity across genetic variants, such as I 2 , Q statistic or E-value) 8-9 Assessment of heterogeneity across genetic variants were provided for all analyses including I2 and Q statistics. 13

Sensitivity analyses and additional analyses
a) Report any sensitivity analyses to assess the robustness of the main results to violations of the assumptions 8-9 MR-Egger, weighted-median and weighted-mode analyses were completed.
b) Report results from other sensitivity analyses or additional analyses 8-9 Visualization of scatter plots, funnel plots and leave-one-out analyses were completed. c) Report any assessment of direction of causal relationship (e.g., bidirectional MR) 8-9 Bidirectional MR was completed for all univariable MR analyses. d) When relevant, report and compare with estimates from non-MR analyses NA NA e) Consider additional plots to visualize results (e.g., leave-one-out analyses) 8-9 See figures and supplemental files.

Key results
Summarize key results with reference to study objectives 10 Our MR analysis suggests that this association is causal and further suggests that increased FI after adjustment for FG and T2D may reduce HbA1c.

6
MR also indicates a bidirectional inverse relationship between Hb and HbA1c. Collectively, these data suggest that increased IR/HI mediated erythrocytosis may potentially lower HbA1c by nonglycemic effects.

15
Limitations Discuss limitations of the study, taking into account the validity of the IV assumptions, other sources of potential bias, and imprecision. Discuss both direction and magnitude of any potential bias and any efforts to address them 11 Our findings may not apply to other ethnic groups given that we used populations with European ancestry only. This may especially be a concern in populations with higher prevalence of hemoglobinopathies and red cell disorders (38-42). Additionally, analyses were not stratified by sex, which is a major determinant of body composition and IR/HI (34). There may also be a relationship between sex hormones such as estradiol and sex hormone binding globulin and HbA1c (43). Finally, there was significant overlap between our exposure and outcome cohorts for our exploratory analyses of the association between waist-hip ratio, erythrocytosis and HbA1c. Our MR analysis suggests that this association is causal and further suggests that increased FI after adjustment for FG and T2D may reduce HbA1c. MR also indicates a bidirectional inverse relationship between Hb and HbA1c. Collectively, these data suggest that increased IR/HI mediated erythrocytosis may potentially lower HbA1c by nonglycemic effects. In exploratory analyses we also report that increased WHR adjusted for BMI, a phenotype associated with IR/HI (26,33) may increase erythrocytosis but we were unable to confirm a significant non-glycemic impact on HbA1c warranting caution in interpreting the data. b) Mechanism: Discuss underlying biological mechanisms that could drive a potential causal relationship between the investigated exposure and the outcome, and whether the gene-environment equivalence assumption is reasonable. Use causal language carefully, clarifying that IV estimates may provide causal effects only under certain assumptions 10 Increased FI is a recognised manifestation of IR. Some features of IR/HI such as increased hepatic glucose production are likely a consequence of reduced insulin action, while others such as hepatic steatosis and dyslipidemia are likely due to increased insulin action via signaling pathways that are not perturbed in IR (35). In vitro studies suggest that insulin can increase erythrocytosis (36), suggesting that increased insulin action may underlie the increased in erythrocytosis with IR/HI.

7
Further studies are needed to confirm these findings and explore underlying mechanisms. c) Clinical relevance: Discuss whether the results have clinical or public policy relevance, and to what extent they inform effect sizes of possible interventions 11 In conclusion, our data suggests that increased FI may increase erythrocytosis and potentially lower HbA1c by non-glycemic effects. These findings might have implications for the diagnoses of preT2D and T2D, its treatment and remission and merits further confirmatory studies. 17

Generalizability
Discuss the generalizability of the study results (a) to other populations, (b) across other exposure periods/timings, and (c) across other levels of exposure 11 In conclusion, our data suggests that increased FI may increase erythrocytosis and potentially lower HbA1c by non-glycemic effects. These findings might have implications for the diagnoses of preT2D and T2D, its treatment and remission and merits further confirmatory studies.
Our findings may not apply to other ethnic groups given that we used populations with European ancestry only. This may especially be a concern in populations with higher prevalence of hemoglobinopathies and red cell disorders (38-42). Additionally, analyses were not stratified by sex, which is a major determinant of body composition and IR/HI (34). There may also be a relationship between sex hormones such as estradiol and sex hormone binding globulin and HbA1c (43). Finally, there was significant overlap between our exposure and outcome cohorts for our exploratory analyses of the association between waist-hip ratio, erythrocytosis and HbA1c. 19

Data and data sharing
Provide the data used to perform all analyses or report where and how the data can be accessed, and reference these sources in the article. Provide the statistical code needed to reproduce the results in the article, or report whether the code is publicly accessible and if so, where All data is public access. Please refer to Table 1 for cohort details.
TwoSampleMR R code and MVMR code is also publicly available.

All authors should declare all potential conflicts of interest 12 None
This checklist is copyrighted by the Equator Network under the Creative Commons Attribution 3.0 Unported (CC BY 3.0) license.