Association between DNA methylation levels of thioredoxin-interacting protein (TXNIP) and changes in glycemic traits: a longitudinal population-based study

Abstract

Thioredoxin-interacting protein (TXNIP) plays an important role in glucose metabolism, and its expression is regulated by DNA methylation (DNAm). Although the association between TXNIP DNAm and type 2 diabetes mellitus has been demonstrated in studies with a cross-sectional design, prospective studies are needed. We therefore examined the association between TXNIP DNAm levels and longitudinal changes in glycemic traits by conducting a longitudinal study involving 169 subjects who underwent two health checkups in 2015 and 2019. We used a pyrosequencing assay to determine TXNIP DNAm levels in leukocytes (cg19693031). Logistic regression analyses were performed to assess the associations between dichotomized TXNIP DNAm levels and marked increases in glycemic traits. At four years, the TXNIP DNA hypomethylation group had a higher percentage of changes in fasting plasma glucose (FPG) and hemoglobin A1c (HbA1c) compared to those in the hypermethylation group. The adjusted odds ratios for FPG and HbA1c levels were significantly higher in the TXNIP DNA hypomethylation group than in the hypermethylation group. We found that TXNIP DNA hypomethylation at baseline was associated with a marked increase in glycemic traits. Leukocyte TXNIP DNAm status could potentially be used as an early biomarker for impaired glucose homeostasis.

CURRENTLY, over 422 million adults globally are afflicted with type 2 diabetes mellitus (T2DM), a figure that has quadrupled in the last three decades [1]. T2DM is a typical multifactorial disease, with both environmental and genetic factors influencing the risk of T2DM onset and development. Recent investigations on the genetic background underlying T2DM have found that identified genetic variants account for 10–20% of phenotypic variance, which is considerably less than expected and is referred to as the problem of “missing heritability” [2]. Limited evidence from genetic association analyses has motivated researchers to explore epigenetic regulation by environmental risk factors. Studies focusing on epigenetic regulation have either focused on biological mechanisms that explain previously reported associations between environmental factors and T2DM, or on novel metabolic and signaling pathways involved in T2DM development.

Thioredoxin-interacting protein (TXNIP) was originally identified as a protein that binds to thioredoxin (TRX) and inhibits its antioxidative activity. In recent decades, other biological functions and roles of TXNIP have been identified in the human body, primary among which is glucose metabolism [3]. The mRNA levels of this protein are modulated by epigenetic changes in its corresponding gene, e.g., TXNIP DNA methylation (DNAm) [4, 5]. Generally, DNAm represents a form of epigenetic regulation that occurs independently of alterations in DNA sequences. The process is affected by various environmental factors, including smoking, alcohol intake, and dietary habits [6-8]. Taken together, TXNIP DNAm levels may mediate the associations between environmental risk factors and various biological functions of TXNIP in disease development, and thus provide a better index with which to predict disease onset.

The associations between glycemic traits and TXNIP DNAm levels have mainly been studied in subjects of European descent [9-11]. In addition, most epigenome-wide association studies (EWAS) conducted to date have adopted a cross-sectional, and relatively few papers have examined the association between TXNIP DNAm levels and changes in glycemic indices or T2DM incidence [9-11]. Consequently, it is still difficult to conclude whether lower TXNIP DNAm levels are the cause or consequence of diabetic conditions. Evidence from prospective studies is therefore needed, in order to clarify the extent of causality between TXNIP DNAm levels and elevated glycemic indices.

Here, we investigated whether leukocyte TXNIP DNAm levels are associated with longitudinal changes in fasting plasma glucose (FPG) and hemoglobin A1c (HbA1c) levels in a population-based study.

Materials and Methods

Study participants

The Yakumo Study was launched in 1982 as a part of a health checkup program for community-dwelling residents in a local area. Every August, interdisciplinary research teams collaboratively conduct a survey in Yakumo Town, Hokkaido, to collect data and samples with the support of municipality staff. The only eligibility criterion for participation is to be at least 40 years old at the time of the health checkup. The selection of eligible participants in this study is shown in Fig. 1. Briefly, we performed this longitudinal study based on two health checkups conducted in 2015 and 2019. A total of 221 residents visited a welfare facility in Yakumo Town and provided informed consent for participation in our research both in 2015 and 2019. Among them, we excluded those who had a medical history of cancer (n = 21), those who had outlier TXNIP DNAm levels (n = 2), those who took medication for diabetes mellitus (n = 19), and those who had a meal within the previous four hours (n = 10). Finally, 169 individuals (72 men and 97 women) were included in our analyses. Written informed consent was obtained from all participants. The protocol used in this study was approved by the Ethics Review Board of Fujita Health University (approval No. HG22-019). Additionally, this study was conducted in accordance with the principles outlined in the Declaration of Helsinki.

Fig. 1

Flow diagram of eligible participants in this study.

Abbreviations: TXNIP, thioredoxin interacting protein.

Biochemical measurements of FPG and HbA1c

Fasting blood samples were collected for laboratory tests and centrifuged within one hour of sample collection, both in 2015 and 2019. Biochemical measurements of FPG and HbA1c were performed using an autoanalyzer in the laboratory at Yakumo General Hospital. HbA1c (%) was measured using the method certified by the National Glycohemoglobin Standardization Program. Based on criteria established by the Japan Diabetes Society, hyperglycemia was defined as FPG ≥126 mg/dL and/or HbA1c ≥6.5% [12]. The percent change in glycemic traits in four years was calculated as follows: percent change of FPG (%) = (FPG₂₀₁₉ – FPG₂₀₁₅)/FPG₂₀₁₅ × 100. We defined the highest interquartile group of the percent change in glycemic traits (% change in FPG: ≥7.1% and HbA1c: ≥5.1%) as having highly increased glycemic traits.

Measurement of TXNIP DNAm

We previously measured DNAm levels at the same cytosine-phosphate-guanine (CpG) site of the TXNIP gene as described elsewhere [13-16]. This CpG site (chr 1: 145,441,552 bp, GRCh37.p13) is located within a 3'-untranslated region (3'-UTR) of the TXNIP gene and has been investigated extensively in previous studies [9-11]. In 2015, genomic DNA was isolated and purified from the buffy coat using a NucleoSpin Tissue kit (Takara, Japan). First, each DNA sample was treated with sodium bisulfite solution using the EpiTect Fast DNA bisulfite kit (QIAGEN, Hilden, Germany). Second, bisulfite-treated DNA was amplified by PCR using Takara EpiTaq HS (Takara). The conditions and primers for PCR are described in our previous study on TXNIP DNAm [13]. Third, sequencing by synthesis assay, i.e., the pyrosequencing method, was performed using a PyroMark Q24 Advanced benchtop sequencer (QIAGEN). Leukocyte TXNIP DNAm levels at this CpG site are expressed as the percentage of methylated cytosine.

Data collection

In 2015 and 2019, prior to the health checkup, the municipality staff distributed a self-administered questionnaire to eligible residents. At the health checkup site, well-trained public health nurses checked the questionnaire for missing data and helped the staff to complete the form with participants. We used the following definitions for lifestyle variables in our analysis: smoking habit (current, ever, or never) and habitual drinking (current, ever, or never). From the results of anthropometric measurements (height and weight) at the baseline survey, we calculated body mass index (BMI).

Statistical analysis

Subjects were divided into TXNIP DNA hypomethylation and hypermethylation groups based on their median TXNIP DNAm levels (77.33%) at baseline. To compare the characteristics between TXNIP hypo- and hyper-methylated groups, we used a paired or unpaired t-test for continuous variables, and a Chi-squared test or Fisher’s exact test for categorical variables. We used multivariable linear regression analysis to assess the associations between baseline TXNIP DNAm levels and baseline glycemic traits. We then calculated changes in glycemic traits between 2015 and 2019. We performed logistic regression analyses to assess the associations between dichotomized TXNIP DNAm levels and highly increased glycemic traits over a four-year period. Additionally, to estimate the risk of a marked increase in glycemic traits with a 1% increase in baseline TXNIP DNAm, we also used baseline TXNIP DNAm levels as a continuous variable. We used sex, age, BMI, serum triglyceride levels, smoking habit, habitual drinking, percentage of neutrophils, and glycemic traits (FPG and/or HbA1c) at baseline as potential confounders in multivariable models, while no covariates were included in crude models. P-values less than 0.05 were considered statistically significant. All statistical analyses were performed using JMP software ver. 14 (SAS Institute Inc., Cary, NC, USA).

Results

Basic characteristics of participants at baseline are summarized according to their TXNIP DNAm status (hyper- or hypomethylation) (Table 1). The mean baseline TXNIP DNAm level (standard deviation: SD) in all participants was 76.8% (5.1), which is similar to that observed in other ethnic populations [9-11]. Mean TXNIP DNAm levels in the hypermethylation group were 7.6% higher than those in the hypomethylation group (hyper: 80.7% vs. hypo: 73.1%). The proportion of men in the hypomethylation group was greater than 55%, compared with 29.8% in the hypermethylation group. The mean age of the hypermethylation group (64.3 years) was significantly higher than that of the hypomethylation group (59.0 years). In the TXNIP DNA hypomethylation group, participants had significantly higher serum triglyceride and FPG levels compared to the hypermethylation group. Proportions of current smokers and current drinkers were significantly higher in the TXNIP DNA hypomethylation group than in the hypermethylation group. The associations between potential covariates and baseline glycemic traits are shown in Supplementary Table S1.

Table 1

Participant characteristics according to baseline TXNIP DNA methylation status.

	Hypomethylation (n = 85)	Hypermethylation (n = 84)	p
TXNIP DNA methylation levels (%)	73.1 ± 4.2	80.7 ± 2.3	<0.001a
Men (n, %)	47 (55.3)	25 (29.8)	<0.001b
Age (years)	59.0 ± 8.9	64.3 ± 7.3	<0.001a
Body mass index (kg/m2)	23.3 ± 2.9	23.4 ± 3.4	0.795a
Systolic blood pressure (mmHg)	128.0 ± 18.9	128.0 ± 21.0	0.979a
Diastolic blood pressure (mmHg)	76.5 ± 14.3	74.6 ± 14.5	0.403a
Triglycerides (mg/dL)d	89.0 (68.5–142.5)	76.5 (58.3–108.3)	0.002a
FPG (mg/dL)	88.7 ± 12.3	84.7 ± 6.7	0.011a
HbA1c (%)	5.6 ± 0.3	5.6 ± 0.3	0.436a
% change in FPG (%)	3.1 ± 8.3	0.9 ± 6.2	0.061a
% change in HbA1c (%)	3.9 ± 3.1	2.8 ± 2.7	0.011a
Current smoker (n, %)	18 (21.2)	7 (8.3)	0.004b
Current drinker (n, %)	50 (58.8)	35 (41.7)	0.010b
Hyperglycemia (n, %)	4 (4.7)	1 (3.0)	0.368c

Abbreviations: TXNIP, thioredoxin interacting protein; FPG, fasting plasma glucose; HbA1c, hemoglobin A1c.

% change in FPG (%) = (FPG₂₀₁₉ – FPG₂₀₁₅)/FPG₂₀₁₅ × 100; % change in HbA1c (%) = (HbA1c₂₀₁₉ – HbA1c₂₀₁₅)/HbA1c₂₀₁₅ × 100.

^a Unpaired t test.

^b Chi-squared test.

^c Fisher’s exact test.

^d Data are shown as geometric means (interquartile ranges).

Table 2 shows cross-sectional associations between TXNIP DNAm levels and glycemic traits at the baseline survey. In the crude linear regression analysis, significantly negative associations were found between TXNIP DNAm levels and FPG (β: –0.514; SE: 0.130) and HbA1c (β: –0.012; SE: 0.005). In the multivariable-adjusted analysis, TXNIP DNAm levels were negatively associated with FPG (β: –0.500; SE: 0.136) and HbA1c (β: –0.013; SE: 0.005).

Table 2

Cross-sectional associations between TXNIP DNA methylation levels and glycemic traits.

	Crude		Multivariable adjusteda
	β (SE)b	p	β (SE)b	p
FPG	–0.514 (0.130)	<0.001	–0.500 (0.136)	<0.001
HbA1c	–0.012 (0.005)	0.009	–0.013 (0.005)	0.004

Abbreviations: FPG, fasting plasma glucose; HbA1c, hemoglobin A1c.

^a Adjusted for sex, age, BMI, serum triglyceride levels, smoking habit, habitual drinking, and percentage of neutrophils.

^b β indicates the difference of baseline glycemic traits when baseline TXNIP DNA methylation levels are 1% higher.

Table 3 shows the characteristics of glycemic traits and the number of participants with hyperglycemia in both 2015 and 2019, as well as longitudinal changes in glycemic traits between 2015 and 2019. Mean levels of glycemic traits and the number of participants with hyperglycemia increased significantly during the four-year period (FPG (SD): 86.7 mg/dL (10.1) vs. 88.3 mg/dL (11.2), p = 0.002; HbA1c (SD): 5.6% (0.3) vs. 5.8% (0.3); p < 0.001; hyperglycemia (%): 5 (3.0) vs. 10 (5.9), p = 0.025). The mean percent changes (SD) of FPG and HbA1c over four years were 2.0% (7.4) and 3.3% (3.0), respectively.

Table 3

Comparison of changes in glycemic traits, percent change in glycemic traits, and number of participants with hyperglycemia between 2015 and 2019 stratified by TXNIP DNA methylation status.

	Binary TXNIP DNA methylation
	Hypomethylation (<77.33%)			Hypermethylation (≥77.33%)
	2015	2019	p	2015	2019	p
FPG (mg/dL)	88.7 ± 12.3	91.2 ± 13.5	0.003a	84.7 ± 6.7	85.4 ± 7.2	0.251a
% change in FPG (%)	3.1 ± 8.3			0.9 ± 6.2
HbA1c (%)	5.6 ± 0.3	5.9 ± 0.4	<0.001a	5.6 ± 0.3	5.8 ± 0.3	<0.001a
% change in HbA1c (%)	3.9 ± 3.1			2.8 ± 2.7
Hyperglycemia (n, %)	4 (4.7)	8 (9.4)	0.046b	1 (1.2)	2 (2.4)	0.317b

Abbreviations: FPG, fasting plasma glucose; HbA1c, hemoglobin A1c.

% change in FPG (%) = (FPG₂₀₁₉ – FPG₂₀₁₅)/FPG₂₀₁₅ × 100; % change in HbA1c (%) = (HbA1c₂₀₁₉ – HbA1c₂₀₁₅)/HbA1c₂₀₁₅ × 100.

^a Paired t-test.

^b McNemar chi-squared test.

Fig. 2 shows the percent changes of FPG and HbA1c levels over four years between hyper- and hypomethylation groups of baseline TXNIP DNAm. The mean percent change (SD) in HbA1c of the hypomethylation group (3.9% (3.1)) was significantly higher compared to that of the hypermethylation group (2.8% (2.7)) (p = 0.011). In addition, the mean percent change (SD) in FPG was higher in the hypomethylation group (3.1% (8.3)) than in the hypermethylation group (0.9% (6.2)), but this association was not statistically significant (p = 0.061) (Table 1).

Fig. 2

Comparison of the percentage changes in glycemic traits over a four-year period between groups with hyper- and hypomethylation of baseline TXNIP DNA methylation.

Abbreviations: TXNIP, thioredoxin interacting protein; FPG, fasting plasma glucose; HbA1c, hemoglobin A1c.

% change in FPG (%) = (FPG₂₀₁₉ – FPG₂₀₁₅)/FPG₂₀₁₅ × 100; % change in HbA1c (%) = (HbA1c₂₀₁₉ – HbA1c₂₀₁₅)/HbA1c₂₀₁₅ × 100.

Boxplots of percent change in glycemic traits according to baseline TXNIP DNA methylation status. The horizontal line represents the mean, the boxes span the interquartile range, and the whiskers indicate the non-outlier ranges.

Table 4 summarizes the results from the logistic regression analyses. In the crude analyses, the hypomethylation group at baseline was positively associated with highly increased levels of FPG (OR: 3.81; 95% CI: 1.80–8.56, p < 0.001) and HbA1c (OR: 2.14; 95% CI: 1.05–4.49, p = 0.039). After adjusting for multiple confounders, significantly positive associations remained between TXNIP DNAm levels and changes in FPG (OR: 3.30; 95% CI: 1.30–8.33, p = 0.012) and HbA1c (OR: 2.77; 95% CI: 1.17–6.56, p = 0.021). The crude and multivariable-adjusted ORs and 95% CIs for highly increased levels of FPG and/or HbA1c are shown according to TXNIP DNAm status. Significantly higher ORs were associated with TXNIP DNA hypomethylation in both unadjusted (crude) analysis (OR: 3.49; 95% CI: 1.82–6.66, p < 0.001) and multivariable analysis (OR: 3.56; 95% CI: 1.64–7.73, p = 0.001). A 1% increase in baseline TXNIP DNAm levels was associated with the decreased likelihood of a marked rise in only FPG (OR: 0.92; 95% CI: 0.84–0.99, p = 0.043) and in either FPG or HbA1c levels (OR: 0.90; 95% CI: 0.83–0.97, p = 0.009). However, this association was not significant for four-year changes in HbA1c (OR: 0.94; 95% CI: 0.87–1.01, p = 0.110). In a sensitivity analysis examining the ORs of hyperglycemia across the three groups categorized by TXNIP DNA methylation rates, the results showed the same trend observed in Table 4, indicating an increased risk in the group with lower TXNIP methylation (Supplementary Table S2).

Table 4

Longitudinal associations between baseline TXNIP DNA methylation status and high increases in glycemic traits over four years.

FPG HbA1c	TXNIP DNA methylation	n	High increase, n (%)	Crude			Multivariable adjusted
				OR	(95% CI)	p	OR	(95% CI)	p
Change in FPG	Binary
	Hypermethylation (≥77.33%)	85	11 (13.1)	Reference			Reference
	Hypomethylation (<77.33%)	84	31 (36.5)	3.81	(1.80–8.56)	<0.001	3.30a	(1.30–8.33)	0.012
	Continuousd	169	42 (24.9)	0.89	(0.83–0.96)	0.002	0.92a	(0.84–0.99)	0.043
Change in HbA1c	Binary
	Hypermethylation (≥77.33%)	85	15 (17.9)	Reference			Reference
	Hypomethylation (<77.33%)	84	27 (31.8)	2.14	(1.05–4.49)	0.039	2.77b	(1.17–6.56)	0.021
	Continuousd	169	42 (24.9)	0.95	(0.89–1.02)	0.161	0.94b	(0.87–1.01)	0.110
Change in FPG and/or HbA1c	Binary
	Hypermethylation (≥77.33%)	85	22 (26.2)	Reference			Reference
	Hypomethylation (<77.33%)	84	47 (55.3)	3.49	(1.82–6.66)	<0.001	3.56c	(1.64–7.73)	0.001
	Continuousd	169	69 (40.8)	0.89	(0.83–0.95)	0.001	0.90c	(0.83–0.97)	0.009

Abbreviations: OR, odds ratio; CI, confidence interval; TXNIP, thioredoxin interacting protein; FPG, fasting plasma glucose; HbA1c, hemoglobin A1c.

Highly increased FPG, % change: ≥7.1%; highly increased HbA1c, % change: ≥5.1%.

^a Adjusted for sex, age, BMI, serum triglyceride levels, smoking habit, habitual drinking, percentage of neutrophils, and FPG₂₀₁₅.

^b Adjusted for sex, age, BMI, serum triglyceride levels, smoking habit, habitual drinking, percentage of neutrophils, and HbA1c₂₀₁₅.

^c Adjusted for sex, age, BMI, serum triglyceride levels, smoking habit, habitual drinking, percentage of neutrophils, FPG₂₀₁₅, and HbA1c₂₀₁₅.

^d Odds ratios were calculated when baseline TXNIP DNA methylation levels are 1% higher.

Discussion

In this study, we investigated the associations between leukocyte TXNIP DNAm levels and longitudinal changes in glycemic traits in a population-based study. First, we showed that baseline TXNIP DNAm levels at the CpG site (chr 1: 145,441,552 bp, GRCh37.p13) were negatively associated with baseline FPG and HbA1c levels in a group of Japanese individuals. Second, we found that baseline TXNIP DNA hypomethylation was associated with markedly increased levels of FPG and HbA1c in four years. Third, a 1% increase in the baseline level of TXNIP DNAm was significantly correlated with a reduced risk of marked increases in FPG and/or HbA1c levels. These associations were found to be significant even after adjusting for possible confounders.

Several previous studies have reported associations between TXNIP DNAm and T2DM, as well as various glycemic traits using EWAS [9-11]. Florath et al. reported a relationship between a decrease in TXNIP DNAm levels and increments in fasting glucose and HbA1c concentrations among older Germans [9]. Kulkarni et al. reported that TXNIP DNAm levels are associated with T2DM fasting glucose among Mexican-American individuals [10]. Soriano-Tarraga et al. identified an association between TXNIP DNAm levels and T2DM and HbA1c levels among Caucasian patients recruited in Spain [11]. Overall, these reports indicate that changes in TXNIP DNAm levels are involved in impaired glucose homeostasis and confirm their role in the pathogenesis of T2DM. However, these previous findings were limited to Europeans and Hispanic Americans. Our results indicated that TXNIP DNAm levels are negatively associated with FPG and HbA1c among the general Japanese population. Despite various differences in clinical characteristics (e.g., the proportion of obese participants), we observed that changes in TXNIP DNAm were associated with T2DM through glycemic traits, as in the case of Europeans and Hispanic Americans. A previous study on East Asians, including Japanese, showed higher susceptibility to T2DM compared to Caucasians, even though East Asians are less obese [17]. One possible reason for this association is that East Asians have limited insulin secretion due to genetic factors and are more likely to exhibit insulin resistance due to environmental factors [17, 18]. Therefore, one cause of the high susceptibility to T2DM in East Asians may be related to TXNIP DNAm changes, which is an epigenetic mechanism mediated through environmental factors.

Previous studies reporting associations between TXNIP DNAm and T2DM have been mainly cross-sectional in design. We found that TXNIP DNA hypomethylation at baseline was associated with marked increases in glycemic traits in four years, which suggests that highly increased glycemic traits are a consequence of sustained TXNIP DNA hypomethylation. Previous studies have reported that TXNIP DNAm levels are inversely correlated with TXNIP mRNA expression levels in human peripheral blood leukocytes [4]. In addition, a case-control studies for coronary artery disease (CAD) have shown significantly increased expression levels of TXNIP in peripheral blood leukocytes (relative expression: 0.90 vs. 1.05), even though the difference of TXNIP DNAm levels was only 4% between CAD patients and controls [5]. Therefore, TXNIP DNAm level changes post-transcriptionally influence TXNIP gene expression, and TXNIP DNA hypomethylation is expected to increase the levels of TXNIP. Based on the previous evidence, in the present study, a difference of 8% in TXNIP DNAm levels between the low and high DNAm groups would affect TXNIP gene expression with more than 2-folds under linearity assumption. The TXNIP protein has been reported to inhibit glucose uptake into fat and muscle and to mediate pancreatic β-cell death through apoptosis [3]. High levels of TXNIP can play an important role in impaired glucose homeostasis preceding T2DM. Collectively, our findings suggest that TXNIP DNA hypomethylation increases the levels of TXNIP protein, disrupting glucose homeostasis and pancreatic apoptosis, which in turn precedes the elevation of FPG and HbA1c, potentially leading to the onset of T2DM. TXNIP expression is highly sensitive to glucose homeostasis, and DNAm could serve as a key mechanism for regulating this expression, potentially constituting a new pathological pathway in the development of T2DM.

In general, DNAm status can be modified by lifestyle factors, including smoking habits, alcohol intake, and dietary habits [6-8]. Therefore, we hypothesized that several lifestyle factors would reduce TXNIP DNAm levels, causing a marked increase in glycemic traits. Actually, many environmental or lifestyle-dependent T2DM risk factors have been described in prospective epidemiological studies [19]. We recently reported that smoking is associated with TXNIP DNA hypomethylation in leukocytes [13]. Several previous studies have shown that cigarette smoking is associated with the development of T2DM, and chronic smoking increases the risk of insulin resistance [20, 21]. Our results suggested that TXNIP DNAm changes represent a significant epigenetic mechanism contributing to the increased risk of T2DM due to cigarette smoking. Further, we demonstrated that a 1% higher level of baseline TXNIP DNAm was associated with a lower risk of marked increases in FPG and/or HbA1c levels. Therefore, we can estimate how lifestyle improvement (e.g., quitting smoking) can increase TXNIP DNAm levels and finally change FPG and HbA1c levels in four years. This is valuable information in both clinical settings and for population-based strategies for T2DM prevention.

This study has both strengths and limitations. The strength of our study lies in investigating TXNIP DNAm levels and changes in glycemic traits in peripheral blood, which is easy to sample in a relatively large population. In previous studies, there has been limited insights into the longitudinal relationship between TXNIP DNAm levels and elevated glycemic indices, and the present study provides some of the first available data elucidating this relationship. Demonstration of a longitudinal relationship between TXNIP DNAm levels and higher glycemic indices may be of clinical use in T2DM. For example, novel blood-based epigenetic biomarkers may be used to estimate the risk for T2DM. Moreover, inhibitors targeting epigenetic changes in the TXNIP gene may be potential future therapies for T2DM and related complications. Our study has several limitations. First, although we were able to adjust for potential confounders in our multivariable analyses, the possibility of residual confounding cannot be completely ruled out. In particular, we could not adjust for dynamic confounders that may have emerged during the follow-up period. Second, although we found that TXNIP DNA hypomethylation at baseline was associated with a marked increase in glycemic traits, the difference in mean DNAm levels between hypomethylation and hypermethylation was relatively small (7.6%). However, a large meta-analysis of prevalent T2DM employing epigenome-wide DNAm from blood samples identified a 1.6% decrease in TXNIP DNAm levels compared with controls, which represents a pooled variance of approximately 3.2% [22]. Therefore, even small changes in TXNIP DNAm status may manifest as clinically significant differences in glycemic traits. Third, the type of white blood cells (WBCs) used in our analysis should be considered, because their type may affect TXNIP DNAm. We determined the type of WBCs from each blood sample using an automated hematology analyzer and attempted to address this issue by adjusting for the percentage of neutrophils in our multivariable analyses. Similar statistical analyses have been performed in several leukocyte DNA methylation studies [23, 24]. Fourth, we did not examine associations between global DNA methylation levels (i.e., LINE-1) and changes in glycemic levels in this study. However, future investigations comparing the results of TXNIP DNA methylation findings with LINE-1 DNA methylation analyses may offer valuable insights into the distinct roles of specific and global epigenetic patterns in glycemic regulation.

In summary, our study revealed that baseline TXNIP DNAm levels were negatively associated with baseline glycemic traits in Japanese individuals. Furthermore, we found that TXNIP DNA hypomethylation at baseline was associated with a substantial increase in glycemic traits. These findings indicate that TXNIP DNA hypomethylation causes a substantial increase in glycemic traits, increasing the risk of T2DM onset over several years. Therefore, leukocyte TXNIP DNA hypomethylation can potentially be used as an early biomarker for impaired glucose homeostasis as well as T2DM development. Future studies are anticipated to explore other biological pathways underlying this association, such as insulin metabolism.

Acknowledgments

We thank the participants and the staff of the Health Examination Program for Residents of Yakumo, Hokkaido, Japan. This work was supported by the Japan Society for the Promotion of Science (JSPS) under Grants-in-Aid for Scientific Research (Grant numbers: 26293144, 17K09139, and 16H06277).

Disclosure

None of the authors have any potential conflicts of interest associated with this research. This study was performed in strict accordance with the principles outlined in the Declaration of Helsinki. Approval was granted by the Ethics Review Board of Fujita Health University (Approval No. HG19-069).

References

• 1   Zhou  B,  Lu  Y,  Hajifathalian  K,  Bentham  J,  di Cesare  M, et al. (2016) Worldwide trends in diabetes since 1980: a pooled analysis of 751 population-based studies with 4·4 million participants. The Lancet 387: 1513–1530.
• 2   Stancakova  A,  Laakso  M (2016) Genetics of type 2 diabetes. Endocr Dev 31: 203–220.
• 3   Parikh  H,  Carlsson  E,  Chutkow  WA,  Johansson  LE,  Storgaard  H, et al. (2007) TXNIP regulates peripheral glucose metabolism in humans. PLoS Med 4: e158.
• 4   Wang  X,  Han  Y,  Zhang  S,  Cui  N,  Liu  Z, et al. (2016) Associations of polymorphisms in TXNIP and gene–environment interactions with the risk of coronary artery disease in a Chinese Han population. J Cell Mol Med 20: 2362–2373.
• 5   Rong  J,  Xu  X,  Xiang  Y,  Yang  G,  Ming  X, et al. (2020) Thioredoxin-interacting protein promotes activation and inflammation of monocytes with DNA demethylation in coronary artery disease. J Cell Mol Med 24: 3560–3571.
• 6   Gao  X,  Zhang  Y,  Breitling  LP,  Brenner  H (2016) Tobacco smoking and methylation of genes related to lung cancer development. Oncotarget 7: 59017–59028.
• 7   Liu  C,  Marioni  RE,  Hedman  ÅK,  Pfeiffer  L,  Tsai  PC, et al. (2018) A DNA methylation biomarker of alcohol consumption. Mol Psychiatry 23: 422–433.
• 8   Fujii  R,  Yamada  H,  Munetsuna  E,  Yamazaki  M,  Ando  Y, et al. (2019) Associations between dietary vitamin intake, ABCA1 gene promoter DNA methylation, and lipid profiles in a Japanese population. Am J Clin Nutr 110: 1213–1219.
• 9   Florath  I,  Butterbach  K,  Heiss  J,  Bewerunge-Hudler  M,  Zhang  Y, et al. (2016) Type 2 diabetes and leucocyte DNA methylation: an epigenome-wide association study in over 1,500 older adults. Diabetologia 59: 130–138.
• 10   Kulkarni  H,  Kos  MZ,  Neary  J,  Dyer  TD,  Kent  JW, et al. (2015) Novel epigenetic determinants of type 2 diabetes in Mexican-American families. Hum Mol Genet 24: 5330–5344.
• 11   Soriano-Tárraga  C,  Jiménez-Conde  J,  Giralt-Steinhauer  E,  Mola-Caminal  M,  Vivanco-Hidalgo  RM, et al. (2016) Epigenome-wide association study identifies TXNIP gene associated with type 2 diabetes mellitus and sustained hyperglycemia. Hum Mol Genet 25: 609–619.
• 12   Seino  Y,  Nanjo  K,  Tajima  N,  Kadowaki  T,  Kashiwagi  A, et al. (2010) Report of the committee on the classification and diagnostic criteria of diabetes mellitus. J Diabetes Investig 1: 212–228.
• 13   Maeda  K,  Yamada  H,  Munetsuna  E,  Fujii  R,  Yamazaki  M, et al. (2020) Association of smoking habits with TXNIP DNA methylation levels in leukocytes among general Japanese population. PLoS One 15: e0235486.
• 14   Yamazaki  M,  Yamada  H,  Munetsuna  E,  Maeda  K,  Ando  Y, et al. (2022) DNA methylation level of the gene encoding thioredoxin-interacting protein in peripheral blood cells is associated with metabolic syndrome in the Japanese general population. Endocr J 69: 319–326.
• 15   Maeda  K,  Yamada  H,  Munetsuna  E,  Fujii  R,  Yamazaki  M, et al. (2022) Association of drinking behaviors with TXNIP DNA methylation levels in leukocytes among the general Japanese population. Am J Drug Alcohol Abuse 48: 302–310.
• 16   Maeda  K,  Yamada  H,  Munetsuna  E,  Fujii  R,  Yamazaki  M, et al. (2023) Serum carotenoid levels are positively associated with DNA methylation of thioredoxin-interacting protein. Int J Vitam Nutr Res doi: 10.1024/0300-9831/a000791. Epub ahead of print.
• 17   Kodama  K,  Tojjar  D,  Yamada  S,  Toda  K,  Patel  CJ, et al. (2013) Ethnic differences in the relationship between insulin sensitivity and insulin response: a systematic review and meta-analysis. Diabetes Care 36: 1789–1796.
• 18   Fujimoto  WY (1996) Overview of Non-insulin-dependent Diabetes Mellitus (NIDDM) in Different Population Groups. Diabet Med 13: 7–10.
• 19   Kolb  H,  Martin  S (2017) Environmental/lifestyle factors in the pathogenesis and prevention of type 2 diabetes. BMC Med 15: 131.
• 20   Willi  C,  Bodenmann  P,  Ghali  WA,  Faris  PD,  Cornuz  J (2007) Active smoking and the risk of type 2 diabetes. JAMA 298: 2654–2664.
• 21   Xie  X,  Liu  Q,  Wu  J,  Wakui  M (2009) Impact of cigarette smoking in type 2 diabetes development. Acta Pharmacol Sin 30: 784–787.
• 22   Juvinao-Quintero  DL,  Marioni  RE,  Ochoa-Rosales  C,  Russ  TC,  Deary  IJ, et al. (2021) DNA methylation of blood cells is associated with prevalent type 2 diabetes in a meta-analysis of four European cohorts. Clin Epigenetics 13: 40.
• 23   Joyce  BT,  Gao  T,  Liu  L,  Zheng  Y,  Liu  S, et al. (2015) Longitudinal study of DNA methylation of inflammatory genes and cancer risk. Cancer Epidemiol Biomarkers Prev 24: 1531–1538.
• 24   Tellez-Plaza  M,  Tang  W,  Shang  Y,  Umans  JG,  Francesconi  KA, et al. (2014) Association of global DNA methylation and global DNA hydroxymethylation with metals and other exposures in human blood DNA samples. Environ Health Perspect 122: 946–954.