Dietary Protein and Fat Intake Affects Diabetes Risk with CDKAL1 Genetic Variants in Korean Adults

Cyclin-dependent kinase 5 regulatory subunit-associated protein 1-like 1 (CDKAL1) is one of the strongest diabetes loci identified to date; evidence suggests that it plays an important role in insulin secretion. Dietary factors that affect insulin demand might enhance the risk of diabetes associated with CDKAL1 variants. Our aim was to examine the interactions between dietary protein and fat intake and CDKAL1 genetic variants in relation to the risk of diabetes in Korean adults. Single nucleotide polymorphisms (SNPs) were selected with a genome-wide association study (GWAS) for diabetes after adjustment for age, gender, and examination site. Using data from the Health Examinees (HEXA) Study of the Korean Genome and Epidemiology Study (KoGES), 3988 middle-aged Korean adults between 40–76 years of age (2034 men and 1954 women) were included in the study. Finally, rs7756992 located within the CDKAL1 gene region was selected from GWAS (p-value < 5 × 10−8). Multivariable logistic regression models were used to evaluate the interactions between genotypes and dietary protein and fat intake in relation to diabetes risk after adjustment for age, gender, BMI, waist circumference, physical activity, smoking status, drinking habits, and examination site. Significant interactions between CDKAL1 rs7756992 and dietary protein and fat intake for the risk of diabetes were observed in men (p-value < 0.05). In women, significant interactions between dietary protein and fat intake and CDKAL1 variants (rs7756992) were associated with increased risk of diabetes (p-value < 0.05). Dietary protein and fat intake interacted differently with CDKAL1 variants in relation to the risk of diabetes in Korean adults of both genders. These findings indicate that CDKAL1 variants play a significant role in diabetes and that dietary protein and fat intake could affect these associations.


Introduction
Diabetes is a rapidly growing public health issue which is associated with morbidity and premature mortality [1]. Prevalence of diabetes is greater among Asians than in other populations from the rest of the world [2]. According to the Korea Health Statistics 2018, prevalence of diabetes in individuals aged 30 years and older (age standardized) was 12.4% (based on fasting blood glucose) and 13.8% (based on fasting blood glucose and glycated hemoglobin). Thus, more than one in 10 adults aged 30 years and older are diagnosed with diabetes mellitus in Korea [3].
Most individuals with type 2 diabetes suffer from serious complications such as nephropathy, neuropathy, retinopathy, cardiovascular disease, and dysfunction of pancreas, skeletal muscle, and liver physiology [4,5]. Diabetes is a complex disease caused by interactions between multiple genetic and

General Characteristics of the Study Population in Cases and Controls
We investigated 3988 participants of the Health Examinees (HEXA) Study of the Korean Genome and Epidemiology Study (KoGES), which included 2034 men (1297 with diabetes, 737 without diabetes) and 1954 women (1207 with diabetes, 747 without diabetes). In men, smoking status, physical activity, and drinking habits did not differ between cases and controls. HDL-cholesterol level was lower in cases (46.4 mg/dL) than in controls (51.6 mg/dL) (p-value < 0.05). Triglyceride (TG) level, BMI, and waist circumference were significantly higher in cases than in controls (p-value < 0.05) ( Table 1).
Smoking, physical activity, and drinking habits did not differ between female cases and controls. Similarly, in men, women had a higher HDL-cholesterol level in the control group than in the case group, while the TG level was significantly higher in cases than in controls (p-value < 0.05). BMI and waist circumference of women were also significantly higher in cases than in controls (p-value < 0.05) ( Table 1). In total population, cases had higher TG level and lower HDL-cholesterol level than controls (p-value < 0.05).

Association of SNPs in CDKAL1 Gene with Fasting Blood Glucose and Glycated Hemoglobin
Among the 83 SNPs, only seven reached the significance threshold p-value (5 × 10 −8 ) and were analyzed for association with fasting blood glucose and glycated hemoglobin (HbA1c) levels. We used linear regression analysis after adjusting for gender and age. The genetic model was based on an additive genetic model. Thus, seven SNPs (rs7756992, rs9368222, rs2206734, rs9465871, rs7747752, rs9356744, and rs6908425) were significantly associated with the risk of diabetes and fasting blood glucose. A total of six SNPs (rs7756992, rs9368222, rs2206734, rs9465871, rs7747752, and rs9356744) showed significant association with fasting blood glucose and HbA1c (all p-values < 0.05); only rs6908425 was not significantly associated with HbA1c (p = 0.227) ( Table 2).
Supplementary Table S1 compares the results of the logistic regression analyses between CDKAL1 SNPs and diabetes after applying either a co-dominant, dominant, or recessive model. Each SNP was significant for diabetes in all three models (p < 0.05). Supplementary Table S2 shows linear regression analysis between SNPs in CDKAL1 and fasting blood glucose and glycated hemoglobin, and all associations were significant (p < 0.05).

Association of SNPs in CDKAL1 Gene with Fasting Blood Glucose and Glycated Hemoglobin
Among the 83 SNPs, only seven reached the significance threshold p-value (5 × 10 −8 ) and were analyzed for association with fasting blood glucose and glycated hemoglobin (HbA1c) levels. We used linear regression analysis after adjusting for gender and age. The genetic model was based on an additive genetic model. Thus, seven SNPs (rs7756992, rs9368222, rs2206734, rs9465871, rs7747752, rs9356744, and rs6908425) were significantly associated with the risk of diabetes and fasting blood glucose. A total of six SNPs (rs7756992, rs9368222, rs2206734, rs9465871, rs7747752, and rs9356744) showed significant association with fasting blood glucose and HbA1c (all p-values < 0.05); only rs6908425 was not significantly associated with HbA1c (p = 0.227) ( Table 2).
Supplementary Table S1 compares the results of the logistic regression analyses between CDKAL1 SNPs and diabetes after applying either a co-dominant, dominant, or recessive model. Each SNP was significant for diabetes in all three models (p < 0.05). Supplementary Table S2 shows linear regression analysis between SNPs in CDKAL1 and fasting blood glucose and glycated hemoglobin, and all associations were significant (p < 0.05).

Interactions of SNP and Protein Intake (% Energy) in Relation to the Risk of Diabetes
Men in the first tertile of protein intake (% energy) after multivariable adjustment (age, examination site, BMI, smoking status, drinking habits, and physical activity) showed diabetes association with rs7756992 (p for trend = 0.034); the SNP had significantly higher adjusted odds ratio (AOR) when possessing a risk allele (AOR 2.04, 95% confidence intervals [CI] 1.16-3.60). Men in the second tertile group having the rs7756992 GG allele had higher AOR 1.91 (95% CI 1.02-3.58) compared to those with the AA allele (Table 3).
Women in the third tertile of protein intake (% energy) after multivariate adjustment showed diabetes association with rs7756992 (p for trend = 0.011). Women in the first tertile group with the rs7756992 GG allele had higher AOR 1.88 (95% CI: 1.01−3.50) compared to those with the AA allele. In the third tertile group with the rs7756992, AG allele and GG allele had higher AORs, 2.08 (95% CI: 1.12-3.88) and 2.29 (95% CI: 1.16-4.53), compared to those with AA allele ( Table 3).
The total in the first tertile of protein intake (% energy) after multivariate adjustment showed diabetes association with rs7756992 (p for trend = 0.001). The total in the first tertile group with the rs7756992 GG allele had higher AOR 2.05 (95% CI: 1.35-3.09) compared to those with the AA allele (Table 3).

Interactions of SNP and Fat Intake (% Energy) in Relation to the Risk of Diabetes
Men in the first tertile of fat intake (% energy) after multivariate adjustment showed diabetes association with rs7756992 (p for trend = 0.009). Men in the first tertile group with the GG allele had higher AOR 2.33 (95% CI: 1.34-4.05) compared to those with AA allele (Table 4).
Women in the second tertile of fat intake (% energy) after multivariate adjustment did not show diabetes association with rs7756992 (p for trend = 0.089). Women with the rs7756992 GG allele in the second and third tertiles did not have significant association with diabetes risk; however, they had high AORs, 1.86 (95% CI: 1.00-3.44) and 2.26 (95% CI: 1.11-4.59), compared to those with the AA allele (Table 4).

Discussion
Significant relationships between CDKAL1 genetic variants (rs7756992) and dietary protein and fat intake in relation to diabetes were observed in the Korean population from the KoGES-HEXA study. Diabetes is caused by genetic and environmental factors, and we found a significant interaction between dietary protein and fat intake and CDKAL1 genetic variants in relation to the risk of diabetes. In men, significant associations between rs7756992 and lower dietary protein and fat intake were observed. In women, significant associations between rs7756992 and higher dietary protein and fat intake were observed.
In the present study, women with risk alleles in the third tertile of protein intake were increasingly at risk for diabetes. The interaction between high protein intake and diabetes risk has been well documented in previous studies [24][25][26][27][28]. In a study on Europeans, high total dietary protein (protein substitution for fat and carbohydrates) increased diabetes development rates [24]. In addition, the incidence of type 2 diabetes was higher in those with high total protein intake (hazard ratio [HR] 1.06 [95% CI 1.02-1.09], p for trend < 0.001). Based on another study on European populations, a higher total protein intake is associated with type 2 diabetes development; all associations were stronger in women with obesity [25]. In men and women from the US, groups with higher total protein intakes were associated with increased diabetes risk [26]. Consistently with this finding, high protein intake was related to increased risk of diabetes in South Asian Indians (OR: 1.47-1.85, 95% CI: 1.02-2.84) [27]. Linn et al. [28] reported that high protein diets increase glucose-stimulated insulin release (p = 0.012) because of decreased glucose limit point of the incretion β-cells (p = 0.031) [28].
Our findings indicate that women in the third tertile fat intake with risk alleles had increased diabetes risk. Previous findings on diabetes risk related to increased fat intake are also well documented. In a multinational study from 2003, including six countries (Greece, Italy, Algeria, Bulgaria, Egypt, and Yugoslavia), animal fat intake was associated with increased incidence of diabetes [29]. In US populations, total fat intake was significantly related to increased risk of diabetes (AOR 1.27, 95% CI 1.04-1.55, p for trend = 0.02) [30]. In particular, high meat consumption and a high saturated fat diet both predisposed to diabetes [30]. Parallel with previous findings, fat ingestion was significantly associated with type 2 diabetes mellitus risk in subjects from the rural populations of San Luis Valley in Colorado with disturbed glucose tolerance [31]. Fasting and postprandial glucose, and glycated hemoglobin levels were studied corresponding to different proportions of dietary carbohydrate and fat intake in Korean patients with non-insulin-dependent diabetes mellitus [32]. The studies showed that those with the lower carbohydrate intake and the higher fat intake had the lower fasting glucose, postprandial glucose and glycated hemoglobin levels in both men and women [32]. A previous study has also shown that maintaining a dietary pattern based on less carbohydrates and more fat may reduce the risk of diabetes in the US population [33].
CDKAL1 is a strong gene identified so far as being significantly associated with diabetes. CDKAL1 variants have been shown to predict the development of diabetes in individuals with impaired insulin secretion, which suggests potential synergistic effects between different risk factors [16,23]. SNP rs7756992 observed in the present study was associated with diabetes-associated indicators. A significant association was observed between the rs7756992 CDKAL1 gene variant and the risk of diabetes. Similar to our finding, individuals with rs7756992 G allele polymorphism were susceptible to diabetes [34,35]. In the Russian population, those who retained G/G allele of rs7756992 had higher OR 1.59 (95% CI 1.10-2.29, p-value < 0.05) compared to those who retained A/A allele [17]. Steinthorsdottir et al. [16] reported that homozygous carriers of the risk allele G of rs7756992 had a 22% decreased insulin response to glucose load than A/A allele carriers. Previous findings suggest significant associations between CDKAL1, alcohol intake, dietary fat, and energy intake in relation to diabetes [23,36,37]. In the Ansan/Ansung Korean cohort, diabetes risk increased by 1.549 (95% CI 1.207-1.720) when high alcohol intake was combined with the presence of CDKAL1 risk alleles [36]. In Japanese men, the interaction between CDKAL1 gene variants and excessive energy intake increased glycated hemoglobin (p = 0.037) [37].
The mechanism on how CDKAL1 interacts with dietary fat and protein intake for type 2 diabetes remains unclear. Dietary factors play a role in the relationship between CDKAL1 polymorphism and diabetes. CDKAL1 polymorphisms may modulate insulin resistance in response to the different levels of dietary fat and protein intake. For example, previous research found that perilipin (PLIN) genetic variants (11482GA and 14995AT) modulated the effects of dietary fat and carbohydrate consumption on insulin resistance in a large sample of Asian female population, indicating a significant gene-diet interaction [38].
To the best of our knowledge, this is the first study to examine CDKAL1-dietary interactions using data from KoGES-HEXA. The present study has several strengths such as the large sample size and inclusion of several potential covariates that affect the relationships between dietary factors, genetic variants, and diabetes. However, several limitations need to be considered when interpreting the results. First, as we conducted cross-sectional analyses using baseline data from KoGES-HEXA, any conclusion about strict cause-effect relationships between dietary factors, genetic variants, and diabetes risk cannot be drawn. Second, we set the age condition of controls (no diabetes) to be 60 years old or older. According to a study, acquired diabetes patients account for more than 50% diabetes patients over 60 years of age [39]. In our study, the age of controls was set at 60 or older, assuming those without diabetes in their 60s and older are more likely to have a genetic background that would not predispose to diabetes. Third, there are only two clinical criteria for determining diabetes (fasting blood glucose and glycated hemoglobin). However, in cohort documents, two more useful clinical indicators were employed to diagnose diabetes. Moreover, according to the Korea Diabetes Association, fasting blood glucose and glycated hemoglobin tests are primarily conducted during diabetes screening [40], so we conducted the study by combining both diabetes history and the two clinical indicators together.

Study Population
The data used in this study were collected from the HEXA cohort of the KoGES from 2004 to 2016 (Seoul, Busan, Daegu, Gwangju, Ulsan, Anyang, Gyeonggi province, Chuncheon, Gangwon province, Cheonan, Chungnam province, Masan, and Gyeongnam province). KoGES-HEXA targeted individuals over 40 years old, living in urban cities; all subjects were examined at medical centers to construct the epidemiological infrastructure necessary to correlate environmental and genetic factors with common chronic diseases in urban cities.
This study used KoGES-HEXA data from years 2004-2013 (n = 28,445). Among these HEXA cohort populations, 24,301 people with no data on diabetes were excluded from both cases and/or controls; subjects under 60 years old were excluded from the control group. Forty-eight people who had inadequate energy intake (< 500 kcal or > 5000 kcal), 38 people with missing data (about drinking habits, smoking status, physical activity, anthropometric measurements, and biochemical variables), and 70 people missing genotype data were excluded. Thus, 3988 individuals ( Figure 2) constituted the actual analytic study group. The study protocol was reviewed and approved by the Institutional Review Board (IRB) of Inha University on January 31, 2020 (IRB No. 200129-1A).

General Characteristics, Anthropometric Measurements, and Biochemical Variables
We surveyed general information of subjects, including age, gender, smoking status (current, past, none), drinking habits (current, past, none), and physical activity (yes, no); anthropometric measurements, including waist circumference, height, weight; and biochemical variables (HDL-cholesterol and triglycerides). Body mass index (BMI, kg/m 2 ) was calculated on the basis of the height (m) and weight (kg) of study participants.

General Characteristics, Anthropometric Measurements, and Biochemical Variables
We surveyed general information of subjects, including age, gender, smoking status (current, past, none), drinking habits (current, past, none), and physical activity (yes, no); anthropometric measurements, including waist circumference, height, weight; and biochemical variables (HDLcholesterol and triglycerides). Body mass index (BMI, kg/m 2 ) was calculated on the basis of the height (m) and weight (kg) of study participants.

Diabetes Detection
Patients with diabetes were defined based on a previous diabetes diagnosis. Moreover, individuals who had fasting blood glucose ≥ 126 mg/dL or glycated hemoglobin (HbA1c) ≥ 6.5% were included as per the 2015 treatment guidelines for diabetes, from the Korean Diabetes Association (KDA) [40]. The non-diabetes group (controls) consisted of people with no diabetes diagnosis, fasting blood glucose < 110 mg/dL, glycated hemoglobin < 5.8%, and an age over 60.

Dietary Measurements
Dietary intake was assessed by using a food frequency questionnaire (FFQ), which included data about daily energy intake (kcal/day), daily protein intake (g/day), and daily fat intake (g/day). The percentages (%) of energy intake from protein and fat intake were calculated as follows: 1 g of protein and 1 g of fat was multiplied by 4 kcal and 9 kcal, respectively, to obtain the percentage of energy consumption.

Genotyping
A total of 28,445 samples were genotyped according to the manufacturer's protocol, which recommended the Axiom ® 2.0 Reagent Kit (Affymetrix Axiom ® 2.0 Assay User Guide; Affymetrix, Santa Clara, CA, USA), and the genotype data were produced using the Korean-Chip, which is available through the Korean-Chip consortium. The Korean-Chip was designed by the Center for Genome Science at the Korea National Institute of Health (4845-301, 3000-3001). The detailed procedure was described in a previous report [41]. Samples that revealed the following features we excluded during the quality control process: sex inconsistency, markers with a high missing rate (>

Diabetes Detection
Patients with diabetes were defined based on a previous diabetes diagnosis. Moreover, individuals who had fasting blood glucose ≥ 126 mg/dL or glycated hemoglobin (HbA1c) ≥ 6.5% were included as per the 2015 treatment guidelines for diabetes, from the Korean Diabetes Association (KDA) [40]. The non-diabetes group (controls) consisted of people with no diabetes diagnosis, fasting blood glucose < 110 mg/dL, glycated hemoglobin < 5.8%, and an age over 60.

Dietary Measurements
Dietary intake was assessed by using a food frequency questionnaire (FFQ), which included data about daily energy intake (kcal/day), daily protein intake (g/day), and daily fat intake (g/day). The percentages (%) of energy intake from protein and fat intake were calculated as follows: 1 g of protein and 1 g of fat was multiplied by 4 kcal and 9 kcal, respectively, to obtain the percentage of energy consumption.

Genotyping
A total of 28,445 samples were genotyped according to the manufacturer's protocol, which recommended the Axiom ® 2.0 Reagent Kit (Affymetrix Axiom ® 2.0 Assay User Guide; Affymetrix, Santa Clara, CA, USA), and the genotype data were produced using the Korean-Chip, which is available through the Korean-Chip consortium. The Korean-Chip was designed by the Center for Genome Science at the Korea National Institute of Health (4845-301, 3000-3001). The detailed procedure was described in a previous report [41]. Samples that revealed the following features we excluded during the quality control process: sex inconsistency, markers with a high missing rate (>5%), individuals with a high missing rate (>10%), minor allele frequency <0.01, and a significant deviation from Hardy-Weinberg equilibrium (HWE) (p < 0.001).
After genotyping and sample quality control, GWAS was performed to select SNPs significantly associated with diabetes in KoGES-HEXA subjects after normalization for age, gender, and examination site (Bonferroni p-value < 5 × 10 −8 ). Finally, rs7756992 located within the CDKAL1 gene region was selected.

Statistical Analyses
PLINK (version 1.90 beta) was used for GWAS, in order to select SNPs associated with diabetes. Association of genetic variants with diabetic or control individuals was based on an additive genetic model and analyzed by logistic regression. Association of CDKAL1 SNPs with fasting blood glucose and glycated hemoglobin levels was analyzed with a linear regression model after adjusting for age, gender and examination site.
We calculated frequency and percentage for each data category from the subjects (gender, smoking status, drinking habits, and physical activity) and conducted a chi-squared test to detect the significant associations between these categorical variables. We calculated mean and standard deviation for all other continuous variables (age, HDL-cholesterol, TG, BMI, waist circumference, daily energy intake, daily protein intake, and daily fat intake) and conducted t-tests to detect differences between cases and controls. Multivariable logistic regression models were evaluated for interactions between CDKAL1 genetic variants and dietary protein and fat intake (% energy/day) in relation to the risk of diabetes after adjusting for age, BMI, examination site, smoking status, drinking habit, and physical activity. Intake (% energy/day) of dietary protein and fat was divided into tertiles. Statistical analyses were performed by using the PLINK and SPSS (Statistical Package for the Social Sciences) software (version 25.0; SPSS Inc., IBM, New York, NY, USA). Statistical significance was determined with two-sided p-value < 0.05. Web-based program Locuszoom version 1.3 (http://csg.sph.umich.edu/locuszoom/) was used to observe regional association plots.

Conclusions
In conclusion, dietary protein and fat intake interacted with CDKAL1 variants in relation to the risk of diabetes, which does vary depending on gender. Patients with diabetes should undertake dietary control despite being treated with medication to manage their blood glucose. The current findings support the role of dietary protein and fat intake as useful indicators for diabetes risk in Korean men and women who have CDKAL1 risk alleles. These findings would help public health professionals to detect high-risk individuals for diabetes with different responses to diet, and this can contribute to the development of more genetic-targeted dietary guideline for specific subpopulations. CDKAL1 variants play a significant role in diabetes, and dietary protein and fat intake could impact their function. Recently, human islet 3D genome maps have been developed and validated to identify target genes for diabetes-relevant regulatory elements [42]. For future studies, polygenic risk score based on the combined set of risk variants through islet hub variants could be used to provide more insights on illuminating the pathophysiology on the genetics of diabetes.