Gamma glutamyltransferase and risk of dementia in prediabetes and diabetes.

Diabetes is associated with cognitive impairment and greater risk for dementia, but the role of gamma-glutamyltransferase (γ-GT) in dementia has not been elucidated. We determined incident dementia including Alzheimer’s disease and vascular dementia, analyzing data from participants aged 40 years or older in the National Health Insurance Database, collected by the National Health Insurance Service in Korea, from January 2009 to December 2015. During a median follow-up of 7.6 years, 272,657 participants were diagnosed as having dementia. Higher serum γ-GT was associated with increased risk of dementia (HR = 1.22, 95% CI = 1.20–1.24), and had a strong positive association with early onset dementia (HR = 1.32, 95% CI = 1.24–1.40). An additive impact of higher γ-GT on dementia was observed regardless of glycemic status, and prevalent diabetes with the highest γ-GT quartile had a 1.8-fold increased dementia risk (HR = 1.82, 95% CI = 1.78–1.85). This effect of γ-GT concentration in diabetes was more prominent in individuals with vascular dementia (HR = 1.94, 95% CI = 1.84–2.04). In subgroup analysis, young age, male sex, and relatively healthy subjects with a higher γ-GT quartile had more increased dementia risk. In conclusion, γ-GT concentration as well as glycemic status could be a future risk factor for dementia in the general population.

Risk for dementia by glycemic status. The incidence of dementia was higher as glycemic status was increased (Table 4). Individuals with prevalent diabetes had a 1.6-fold greater risk for dementia compared to those with normal glucose concentration (HR = 1.60, 95% CI = 1.58-1.62) and this phenomenon was more prominent in EOD (HR = 1.73, 95% CI = 1.66-1.81) compared to LOD (HR = 1.55, 95% CI = 1.53-1.57). The risk for VaD (HR = 1.67, 95% CI = 1.62-1.72) was slightly higher compared to that of AD (HR = 1.55, 95% CI = 1.55-1.59). The difference in dementia risk according to glycemic status was also observed in the Kaplan-Meier curve and a similar pattern was shown in both AD and VaD (Fig. 4). As shown in Table 5, the risk for dementia ranked the highest in the treatment failure group. Subjects who fail to treat diabetes had a 2.07-fold greater risk for EOD.
Combined effects of γ-GT levels and glycemic status on dementia risk. Compared to non-diabetic individuals with the lowest γ-GT concentration, those with prevalent diabetes and highest γ-GT concentrations had an approximately 1.8-fold increased risk of dementia (HR = 1.82, 95% CI = 1.78-1.85; Fig. 5A). This additive effect of γ-GT concentration in diabetes was more prominent in VaD (HR = 1.94, 95% CI = 1.84-2.04; Fig. 5B) than AD (HR = 1.76, 95% CI = 1.72-1.80; Fig. 5C). When analyzing the risk of dementia according to γ-GT quartiles among individuals with the same glycemic status (lowest γ-GT quartile as reference in each glycemic status group), the elevated hazard ratio for overall dementia was the most prominent for normoglycemia (HR = 1.18, 95% CI = 1.16-1.20; Fig. 5D) and prevalent diabetes and this pattern was similar in both EOD and LOD. The highest γ-GT quartile had was associated higher dementia risk regardless of glycemic status. In prevalent diabetes and incident diabetes in the younger age group, dementia risk was slightly lower in the second and third γ-GT quartiles and then increased in the highest quartile, representing a J shape (Fig. 5E). The risk for LOD in normoglycemic individuals in the highest γ-GT quartile was lower than EOD risk (HR = 1.14, 95% CI = 1.13-1.16; Fig. 5F).
Association between γ-GT levels and the risk of dementia by other comorbid conditions. The association between γ-GT level and risk of dementia was still significant regardless of other clinical conditions (Fig. 6). The risk for highest γ-GT quartile in EOD was higher than in LOD for both sexes, and male individuals under 60 years old were had the greatest dementia risk, particularly for VaD (HR = 1.40, 95% CI = 1.21-1.61). VaD risk with higher γ-GT values was greater than AD risk regardless of hypertension, cardiovascular disease, and obesity. Regardless of dementia subtype, the impact of γ-GT on incident dementia was stronger in individuals without other comorbidities. In terms of alcohol consumption, heavy alcohol use with a higher γ-GT value had www.nature.com/scientificreports www.nature.com/scientificreports/ an increased dementia risk compared to individuals without heavy alcohol use, and VaD risk was the highest in heavy alcohol use group (HR = 1.44, 95% CI = 1.30-1.61).

Discussion
In this large, national population-based study, we demonstrated that serum γ-GT levels had a linear relationship with risk of developing dementia, independent of glycemic status and other important potential confounders. In particular, the influences of higher γ-GT levels and diabetes on developing dementia were much stronger for EOD than for LOD. γ-GT concentrations and glycemic status had an additive impact on dementia incidence. Regarding glycemic categories, the association between higher serum γ-GT concentration and dementia risk was more profound in the normoglycemic group compared with impaired fasting glucose (IFG) or diabetes. In addition, the increased risk of developing VaD according to γ-GT concentration was greater compared to AD risk.
The impact of diabetes on dementia has been established, since other studies demonstrated chronic exposure to hyperglycemia may increase cognitive dysfunction 14,15 . Although there has been no study on the diabetes increases the risk of EOD, the evidence that average for dementia translation was earlier in patients with type 2 diabetes compared to the patients without diabetes 16 accords our result. The remarkably increased EOD risk for treatment failure group might reflect that failure to treat diabetes may accelerate progression to cognitive decline and the importance of glucose control in young adults. Even among individuals without diabetes, the risk for future dementia gradually increased with elevated glucose level, and a lower glucose concentration was inversely associated with dementia risk 14 . In addition, even among individuals without diabetes, dementia risk increased as the glycemic status worsened, to IFG or incident diabetes. Notably, the association of γ-GT levels and dementia outcome was linear in the normoglycemic group in the current study. As to glycemic status, higher γ-GT concentrations in individuals with normal glucose levels had the highest dementia risk. As diabetes and dementia share common features 15 , and diabetes is often co-morbid with dementia 17 , the influence of elevated GGT in the normoglycemic group may be more pronounced.   Table 2. Sex-specific γ-GT and metabolic parameters. Data for continuous variables were expressed as either mean ± standard deviation, or mean (interquartile range) and categorical variables were expressed as number (percent). Abbreviations: BMI, body mass index; SBP, systolic blood pressure; DBP, diastolic blood pressure; FPG, fasting plasma glucose; γ-GT, gamma-glutamyltransferase; AST, aspartate transaminase; ALT, alanine transferase; IFG, impaired fasting glucose. *Obesity was defined as body mass index >25 kg/m 2 per Asian-Pacific definition.
Interestingly, in our study a strong positive association between serum γ-GT levels and dementia risk was observed in the younger age group, particularly in the young male, non-obese group. This is similar to previous studies of γ-GT and cardiovascular disease risk 18,19 . However, the association between γ-GT and dementia is independent of cardiovascular disease, as a stronger association was observed in individuals without previous myocardial infarction or ischemic stroke. Moreover, considering the smaller effect size of diabetes on EOD than LOD 17 , the degree of impact of γ-GT might be relatively greater on EOD than LOD. γ-GT levels are known to reflect vascular and cardiometabolic diseases and γ-GT enzyme activity is independently associated with cardiovascular disease and cardiovascular-related mortality, in a dose-dependent manner 10 , suggesting its causal relationship to vascular damage. In pooled analysis, higher γ-GT enzyme activity was a surrogate marker for diabetes and metabolic syndrome, which contribute to dementia pathogenesis 20 . Likewise, our study showed a relatively prominent impact of γ-GT on VaD compared to AD and a linear association between γ-GT quartiles and impaired glucose status. The pathways involved in γ-GT expression and cardiometabolic disease could be explained by insulin signaling impairment and insulin resistance 21 . Furthermore, γ-GT enzyme activity increased in all-cause and cancer mortality independently of NAFLD 18 , suggesting its essential role in human metabolism. γ-GT is primarily involved in extracellular catabolism of glutathione, the major thiol antioxidant that plays a protective role against oxidants 22 . Therefore, an increased serum γ-GT level reflects systemic oxidation and reactive oxygen species (ROS). We found more increased and stronger dementia risk according to γ-GT groups than NAFLD presence. In addition, one notable finding in our study was the J shape for dementia risk in the diabetes group. Individuals with diabetes have fundamentally increased ROS, and more γ-GT might be needed to  www.nature.com/scientificreports www.nature.com/scientificreports/ compensate for this stress. If genetically or environmentally γ-GT concentration is decreased, the role of γ-GT in this defense mechanism would be impaired, with greater vulnerability to those conditions. We acknowledge some limitations in the current study. First, this study was based on a national general health care dataset, which did not survey genetic factors such as APOE genotype. Second, we could not determine the dietary and lifestyle factors which could affect serum γ-GT concentration. However, we did adjust for alcohol consumption, which is a generally well-established γ-GT level determinant. Third, our analyses focused on dementia type based on diagnosis code rather than imaging or functional cognitive testing. Despite these limitations, our study has several strengths. First, as a large, population-based study, we achieved statistical reliability without selection bias. We were able to categorize into sex-specific γ-GT deciles or quintiles, while previous studies on NAFLD or γ-GT were conducted mostly in male subjects. Second, to our knowledge, this is the first study to demonstrate individual and combined effects of γ-GT and diabetes on development of dementia in the general population. Moreover, our results clearly proved the link between γ-GT concentration and dementia risk, particularly EOD.
In conclusion, our data demonstrated that γ-GT is positively associated with future risk of dementia independent of diabetes and other metabolic conditions. The combined impact of γ-GT and diabetes increased risk for dementia development, implying that γ-GT concentration as well as glycemic status could be a risk factor. Understanding the role of γ-GT in the pathogenesis of dementia could enable clinicians to identify individuals most at risk, especially for EOD, with implications for prevention and treatment.

Study population. This longitudinal cohort study utilized data collected from participants in the National
Health Insurance database maintained by the Korean National Health Insurance Service (NHIS), the single insurer in the Korean public health insurance sector that provides national health examinations for all Koreans. The NHIS database comprises the entire Korean population 23 ; therefore, it can be used as a population-based,   24 . This biannual regular checkup includes anthropometric measurements, blood pressure, social habits, physical activity, and laboratory tests with overnight fasting. All biochemical samples are collected and measured as previously described 25 . Past medical history, alcohol consumption, smoking history, and exercise habits are collected by standardized self-reporting questionnaires. BMI is calculated as weight(kg)/height(m 2 ) and obesity is defined as BMI ≥ 25 kg/m 2 , using the Asian-Pacific criteria 26 . All participants provided written informed consent to participate in the original NHIS. The study was approved by the institutional review board of the Yonsei University College of Medicine (4-2016-0575). All procedures performed in studies involving human participants were in accordance with the ethical standards of the 1964 Helsinki Declaration and its later amendments or comparable ethical standards.   www.nature.com/scientificreports www.nature.com/scientificreports/ Definition of dementia and diabetes. To obtain information on incident dementia, we followed a cohort of 6,595,271 participants who were ≥40 years and were dementia-free NHIS beneficiaries (Fig. 1).    www.nature.com/scientificreports www.nature.com/scientificreports/ dose-response relationship between serum γ-GT and dementia incidence, we categorized sex-specific γ-GT by quartile or decile. In addition, we divided by age at the time of study recruitment (40-59, ≥60 years) to designate as EOD (40-59 years) and LOD ( ≥ 60 years) 27 . NAFLD was estimated by a calculating hepatic steatosis index (HSI) and defined by HSI ≥ 36 28 .
A diagnosis of diabetes based on ICD-10 codes included the principal diagnosis and up to four accompanying diagnoses. To investigate the effect of diabetes and γ-GT, we categorized subjects into 4 groups according to glycemic status: 1) Prevalent diabetes was defined as at least one service claim with a diagnosis of diabetes, either in outpatient or inpatient care and at least one prescription of a hypoglycemic agents during the index period; 2) www.nature.com/scientificreports www.nature.com/scientificreports/ Incident diabetes was defined as individuals without diabetes diagnosis or treatment, but whose fasting glucose concentration was ≥126 mg/dL; 3) IFG was defined as fasting glucose concentration ≥100 mg/dL and <126 mg/ dL; and 4) Others were identified as normoglycemia. In addition, we defined subject who had diabetes diagnosis code and prescribed medication, but did not maintain (less than 3 years) the prescribed medication as treatment failure group whereas maintained the anti-diabetic medication prescription more than 3 years as treatment maintenance group.
Hypertension was defined as ICD-10 codes (I10-13, I15) plus treatment with anti-hypertensive agents, or systolic or diastolic blood pressure ≥140 mmHg/≥90 mmHg; dyslipidemia was ICD-10 code of E78 plus treatment with lipid-lowering agents or total cholesterol ≥240 mg/dL. A previous medical history of ischemic stroke or myocardial infarction was defined as ICD-10 codes of I63-I64 or I21-I22. We identified low social economic status as the lowest quartile for income in the study population 29 . Heavy alcohol consumption was defined as ≥ 30 g per day, and regular exercise was categorized as ≥3 times per week of moderate to vigorous physical activity.
Statistical analysis. Descriptive characteristics were mainly presented as mean ± standard deviation (SD).
As AST, ALT, and γ-GT were not normally distributed, those markers were expressed as median with interquartile range (IQR). As a sex difference in γ-GT level has been documented 30 , we categorized sex-specific cut-off. Data were expressed as numbers and as a frequency percentage. The χ 2 test was used to determine differences in percentages of categorical variables, and the independent Student's t test evaluated differences between the means of two continuous variables. A one-way analysis of variance (ANOVA) was used to compare the baseline characteristics of continuous variables by γ-GT categories. Incidence rates were expressed as events per 1,000 person-years, and were adjusted for age and sex using the direct method. Cox proportional hazards regression analysis was used to identify the association between γ-GT level and dementia after adjustment for other risk factors. The results were presented as HRs and 95% CI. Subgroup analysis was performed according to covariates including sex, age, hypertension, previous cardiovascular disease (ischemic stroke and myocardial infarction) history, obesity, and alcohol consumption. Dementia-free survival according to γ-GT categories and glycemic status was analyzed using the Kaplan-Meier curve and expressed as adjusted HR and 95% CI. A two-sided P value <0.05 was considered statistically significant. Statistical analyses were performed using SAS version 9.4 (SAS Institute, Cary, NC).
Ethics approval and consent to participate. All participants provided written informed consent to participate in the original NHIS. The study was approved by the institutional review board of the Yonsei University College of Medicine (4-2016-0575).

Data availability
This study used the National Health Insurance Service NHIS 2009-2015 data, which were released by the KNHIS. Access to NHIS-NSC data are available from the website of NHIS (https:// nhiss.nhis.or.kr) after completing the application process and receiving approval (http://nhiss.nhis. or.kr/bd/ab/bdaba021eng.do).