Genetic variations in adiponectin levels and dietary patterns on metabolic health among children with normal weight versus obesity: the BCAMS study

Abstract

BACKGROUND/OBJECTIVES:

Adiponectin represents an important link between adipose tissue dysfunction and cardiometabolic risk in obesity; however, there is a lack of data on the effects of adiponectin-related genetic variations and gene-diet interactions on metabolic disorders in children. We aimed to investigate possible interactions between adiponectin-related genetic variants and habitual dietary patterns on metabolic health among children with normal weight versus overweight/obesity, and whether these effects in childhood longitudinally contribute to metabolic risk at follow-up.

SUBJECTS/METHODS:

In total, 3,317 Chinese children aged 6–18 at baseline and 339 participants at 10-year follow-up from the Beijing Child and Adolescent Metabolic Syndrome study cohort were included. Baseline lifestyle factors, plasma adiponectin levels, and six adiponectin-related genetic variants resulting from GWAS in East Asians (loci in/near ADIPOQ, CDH13, WDR11FGF, CMIP, and PEPD) were assessed for their associations with the metabolic disorders. Being metabolically unhealthy was defined by exhibiting any metabolic syndrome component.

RESULTS:

Among the six loci, ADIPOQ rs6773957 (OR 1.26, 95% CI:1.07–1.47, P = 0.004) and adiponectin receptor CDH13 rs4783244 (0.82, 0.69–0.96, P = 0.017) were correlated with metabolic risks independent of lifestyle factors in normal-weight children, but the associations were less obvious in those with overweight/obesity. A significant interaction between rs6773957 and diet (P_interaction = 0.004) for metabolic health was observed in normal-weight children. The adiponectin-decreasing allele of rs6773957 was associated with greater metabolic risks in individuals with unfavorable diet patterns (P < 0.001), but not in those with healthy patterns (P > 0.1). A similar interaction effect was observed using longitudinal data (P_intearction = 0.029).

CONCLUSIONS:

These findings highlight a novel gene-diet interaction on the susceptibility to cardiometabolic disorders, which has a long-term impact from childhood onward, particularly in those with normal weight. Personalized dietary advice in these individuals may be recommended as an early possible therapeutic measure to improve metabolic health.

INTRODUCTION

Obesity, usually measured by body mass index (BMI), is an established risk factor of type 2 diabetes (T2D), hypertension, dyslipidemia, and ultimately, cardiovascular disease (CVD) [1]. However, the clustering of obesity-related metabolic abnormalities varies widely among individuals with similar BMI [2–5]. Moreover, there is a subset of individuals with obesity that are metabolically healthy, while metabolic abnormalities were also observed in some normal-weight individuals [2–5]. Accurate classifications and mechanistic understandings for metabolic health among individuals with normal weight and overweight/obesity are essential to developing future precise clinical interventions for metabolic disorders in different individuals.

Adipose tissue secretes various adipokines. The secretion of adipokines is altered during adipose tissue dysfunction such as the unhealthy expansion of adipose tissue, and may lead to a spectrum of obesity-associated metabolic abnormalities, despite a normal weight status [3]. Adiponectin is one of the most abundant adipokines with anti-diabetic, anti-oxidant, and anti-atherosclerotic actions through endocrine, autocrine, and paracrine signaling [6]. Paradoxically, although it is produced by adipose tissue, its production varies inversely with the total level of body fatness. In mouse models, the elevation of adiponectin levels contributes to a healthy adipose tissue function with enlarged subcutaneous adipose tissue and enhanced adipogenesis under the challenge of caloric excess, ultimately improving metabolic health [6, 7]. Compelling evidence of human studies has shown that hypoadiponectinemia was associated with obesity and cardiometabolic diseases [8–10]. Recently, reduced adiponectin levels were also shown to differentiate metabolically healthy versus unhealthy in both obese and non-obese adults [11] and children [12]. Moreover, genome-wide association studies (GWAS) identified variants of several genes, such as loci in or near to ADIPOQ, CDH13, WDR11FGF, CMIP, and PEPD, which were associated with circulating adiponectin levels [13–16]. However, the causal relationships between adiponectin levels and cardiometabolic risks inferred by Mendelian randomization studies remain unclear. For example, genetic variants in the ADIPOQ locus associated with high adiponectin levels are associated with improved insulin sensitivity [17] but a higher cardiovascular mortality rate [18], while other studies reported low adiponectin levels did not play a causal role in cardiometabolic diseases [19–22]. Therefore, examining whether adiponectin-related genetic variants affect metabolic health, especially in individuals with different weight status, may help explain the role of adiponectin in cardiometabolic diseases and improve our understanding of metabolic disorders.

In addition to genetic effects, different foods or dietary patterns have been reported to affect the secretion of adiponectin, thus impacting cardiometabolic risks [23–25]; however, little is known about the interplay between diet and genetic predisposition and pathways involved in hypoadiponectinemia pathogenesis, especially in the longitudinal setting. We hypothesized that dietary patterns might affect the secretion of adiponectin from adipose tissue by modulating the function or activity of the adiponectin-associated genes. Little is known about the mediating role of adiponectin-associated loci in linking together adiponectin levels, dietary patterns, and metabolic abnormalities. Leveraging the large cohort within the Beijing Child and Adolescent Metabolic Syndrome (BCAMS) study, we aimed to examine the role of adiponectin-related genetic variants and habitual dietary patterns on the development of metabolic abnormalities among children with normal weight and overweight/obesity. Besides, we examined whether dietary patterns can modify the associations of adiponectin-related genetic variants on metabolic abnormalities.

METHODS

Subjects

The BCAMS study is a prospective cohort study of obesity and related cardiometabolic abnormalities in a representative sample of Beijing school-aged children which has been previously described elsewhere [26, 27]. Briefly, the baseline survey (n = 19,593 six to 18 years old, 50% boys) was conducted in 2004. Among these participants, 4,500 were recognized as having risk factors defined by the presence of any one of the following: overweight (BMI > 85th percentile), total cholesterol (TC) ≥ 5.2 mmol/L, triglycerides (TG) ≥ 1.7 mmol/L, or fasting blood glucose (FBG) ≥ 5.6 mmol/L based on a capillary blood test. All participants at increased risk for metabolic disorders, together with a parallel reference population of 1,095 children, were invited to undergo medical examinations for verification based on venipuncture blood samples and clinical examination. Among this cohort, 3,317 children (50.7% boys, n = 1683) consented to provide blood samples for adiponectin measurement and other tests, were thus included in the cross-sectional analysis. Follow-up studies of this cohort were performed in 2014, and 559 participants agreed to return to complete the in-depth follow-up examination during the two-year follow-up period [28]. Among these 559 participants, 339 of them had baseline blood samples and adiponectin data and thus were included in the longitudinal analysis. A consort diagram is illustrated in the Supplemental file: Figure S1. The BCAMS study has been registered as a clinical trial at www.clinicaltrials.gov (NCT03421444). The BCAMS was approved by the Ethics Committee at the Capital Institute of Pediatrics in Beijing. The follow-up study was approved by the Ethics Committee at Beijing Chaoyang Hospital. All the phases of the study complied with the Ethical Principles for Medical Research Involving Human Subjects expressed in the Declaration of Helsinki.

Clinical and anthropometric measurements

Height, weight, waist circumference, systolic and diastolic blood pressure, were measured by a standardized protocol [29]. BMI was calculated as weight (kg) divided by height squared (m²). Normal weight, overweight, and obesity were defined by the sex- and age-specific <85th, 85th–95th, and ≥95th percentile of BMI, separately, as recommended by the Working Group on Obesity in China [30], and at follow-up, normal weight, overweight, and obesity were defined as having BMI < 24, 24–28, and ≥28 kg/m², respectively. Percent body fat (FAT%) was assessed by bioelectrical impedance analysis (BIA, TANITA TBF-300A). Pubertal development was assessed by Tanner stages (T1–T5) of breast development in girls and testicular volume in boys [31], by two pediatricians matched by sex to the child.

Laboratory measurements

Blood samples were collected from an antecubital vein after an overnight fast. Within 30 min, plasma and serum aliquots were separated and stored at −80°C until further analysis. A two-hour 75 g oral glucose tolerance test (OGTT) was performed in follow-up individuals. The concentrations of plasma glucose (glucose oxidize method), TG, TC, high-density lipoprotein cholesterol (HDL-C), and low-density lipoprotein cholesterol (LDL-C) were assessed by a Hitachi 7060 C automatic biochemistry analysis system. Fasting insulin and adiponectin were measured by monoclonal antibody-based enzyme-linked immunosorbent assay (ELISA) [32], which was developed in the Key Laboratory of Endocrinology, Peking Union Medical College Hospital. The insulin assay had an inter-assay coefficient of variation (CV) of < 9.0% and no cross-reactivity to proinsulin (<0.05%). The intra- and inter-assay CVs for adiponectin were <5.4% and <8.5%, respectively [33]. The insulin resistance index was calculated by the homeostasis model assessment of insulin resistance (HOMA-IR), HOMA-IR = fasting insulin (mU/L) × FBG (mmol/L)/22.5 [34].

Lifestyle measurements

Questionnaires were used to obtain information on lifestyle factors [29, 35]. Eight dietary items, including breakfast, beans, seafood, milk, vegetables, fruits, meat, and soft drink intake, were assessed by the consumption frequency. The response options were presented as a five-point Likert scale of the frequency of consumption from ‘seldom or never’ to ‘every day’. Diet scores were calculated by summing up all item scores, in which a lower score indicated more inferior dietary quality. Physical activity was assessed as the number of times the individual engaged in moderate-to-vigorous (MVPA) and classed as low (<3 times/week) and or high (≥3 times/week) [35]. A sedentary lifestyle was evaluated by the time of playing computer games or watching television and defined as ‘screen time ≥2 hours/d’ [35]. Sleep duration was assessed with the question: “How many hours of sleep do you usually get at night?” [36].

Genomic DNA extraction and genotyping

DNA extraction from blood samples was performed using the QIAamp DNA Blood Midi Kits (Qiagen). We selected the six most strongly associated SNPs to adiponectin levels from previously reported GWAS of adiponectin [13–16], ADIPOQ rs10937273, rs6773957, CDH13 rs4783244, WDR11FGF rs3943077, CMIP rs2925979, and PEPD rs889140. Among them, five SNPs (ADIPOQ-rs10937273, CDH13-rs4783244, WDR11-FGFR2-rs3943077, CMIP-rs2925979, and PEPD-rs889140) were previously identified in an East Asian adult population [15, 16], to be most strongly associated with adiponectin levels (P < 10⁻¹⁰ for each of the five SNPs); while another SNP, ADIPOQ-rs6773957, was identified in a European population [15] and was included because it is located at the 3’ UTR of ADIPOQ, which is an important gene regulatory region. All the SNPs were genotyped with the use of the Sequenom Mass Array iPLEX genotyping platform in BioMiao Biological Technology Co, Ltd. [35, 37]. The concordance rate of repeated control samples in each genotyping plate was 100%. All these SNPs had genotyping efficiency of over 0.95 and were in Hardy-Weinberg equilibrium with Bonferroni adjusted P value >0.008 (0.05/6).

Definitions of being metabolically unhealthy

Identification of being metabolic unhealthy at baseline was defined by the presence of any classical components of pediatric metabolic syndrome (MS) according to the modified criteria of Adult Treatment Panel III: (1) elevated systolic blood pressure (SBP) and/or diastolic blood pressure (DBP) ≥ 90th percentile for age, sex and height (according to the BCAMS study); (2) hypertriglyceridemia defined as TG ≥ 1.24 mmol/L; (3) low serum HDL-C defined as ≤1.03 mmol/L, and (4) FBG defined as ≥5.6 mmol/L. Therefore, participants were classified into four groups according to their weight and metabolic status: (1) metabolically healthy normal weight (MHNW); (2) metabolically unhealthy normal weight (MUNW); (3) metabolically healthy with overweight/obesity (MHO); and (4) metabolically unhealthy with overweight/obesity (MUO). Being Metabolically unhealthy at follow-up was defined by the presence of any classical components of MS according to the harmonized definition and metabolic health was defined by the absence of any classical feature of MS [38, 39].

Statistical analysis

All statistical analyses were performed using SPSS version 22.0 software for Windows (SPSS Inc., Chicago, IL, USA). Non-normal distributions were natural logarithm transformed for analysis. Results are expressed as mean ± standard deviation (SD) or percentage (%). Independent sample t-test or χ² test were used for comparison of continuous or categorical variables between children with or without metabolic abnormalities. The associations of individual SNP and metabolic health were examined using additive models. Each SNP was recorded as 0, 1, or 2 according to the number of risk alleles. Given that sex was a differential factor for dietary intake in our study population (Supplementary Table 1) and that children at an advanced pubertal stage generally exhibited a higher BMI, a higher waist circumference, a higher FAT%, and lower adiponectin levels, and some of the metabolic parameters such as SBP, DBP, insulin, and HOMA-IR were transiently elevated during puberty (Supplementary Table 2), we adjusted for sex, pubertal stage, along with age, BMI, and physical activity in multivariable logistic regression models which were used to determine the independent effects of genetic variations and dietary patterns on metabolic abnormalities. Moreover, interactions of individual SNPs with diet patterns on metabolic abnormalities were analyzed using logistic regression by adding the interaction terms in the models (for example, dietary patterns × SNP), with the main effects included in the models as well. Two-sided p values < 0.05 were considered statistically significant, and Bonferroni correction for multiple testing was used when appropriate.

RESULTS

Baseline and follow-up characteristics according to metabolic health and obesity status

The baseline characteristics of the four groups are shown in Table 1. In the normal-weight group, 36.2% of children were metabolically unhealthy, while 67.4% of children in the overweight/obesity group were classed as unhealthy. The metabolically unhealthy children were older, more mature concerning puberty, had higher BMI and WC, a more unhealthy diet, and lower amounts of physical activity compared with their metabolically healthy counterparts. These trends were apparent in both normal-weight children and children with overweight/obesity (all P < 0.05). As expected, HOMA-IR levels were the lowest in the MHNW group, increased successively in the MUNW and MHO groups, and were highest in the MUO group. In addition, adiponectin levels were significantly higher in the metabolically healthy group than in the metabolically unhealthy group, regardless of obesity status (P < 0.001). These results were significant even after adjustment for age, sex, and BMI (data not shown). The follow-up characteristics of the four groups are shown in Supplementary Table 3.

Table 1.

Baseline characteristics according to weight status and metabolic health categories.

	Normal weight			Overweight/Obesity
	Metabolically healthy (MHNW)	Metabolically unhealthy (MUNW)	P	Metabolically healthy (MHO)	Metabolically unhealthy (MUO)	P
Demography
N	976	553		582	1202
Age (years)	12.2 ± 3.2	13.0 ± 3.1	<0.001	12.0 ± 3.1	12.4 ± 2.9	0.005
Male (%)	42.5	39.3	0.220	54.1	61.5	0.002
Tanner stage (T1/2/3/4/5%)	31.5/15.3/14.3/29.1/9.8	21.5/15.9/13.5/33.4/15.6	<0.001	32.8/15.2/12.8/17.5/21.6	27.4/17.7/12.8/18.3/23.8	0.168
BMI Z-score	−0.31 ± 0.85	−0.14 ± 0.81	<0.001	1.66 ± 0.38	1.83 ± 0.41	<0.001
BMI (kg/m2)	17.52 ± 2.37	18.29 ± 2.45	<0.001	24.29 ± 3.14	26.04 ± 3.70	<0.001
WC (cm)	61.59 ± 7.02	63.51 ± 7.18	<0.001	77.69 ± 9.53	82.77 ± 10.55	<0.001
FAT %	17.41 ± 5.42	19.23 ± 5.84	<0.001	28.59 ± 6.51	30.42 ± 6.62	<0.001
Cardiometabolic risk
SBP (mmHg)	97.6 ± 10.6	107.5 ± 12.8	<0.001	104.8 ± 9.06	117.3 ± 12.2	<0.001
DBP (mmHg)	61.4 ± 8.2	68.7 ± 10.2	<0.001	65.5 ± 6.7	73.7 ± 9.0	<0.001
Total cholesterol (mmol/liter)	4.08 ± 0.78	4.17 ± 0.94	0.043	3.99 ± 0.68	4.10 ± 0.78	0.003
Triglycerides (mmol/liter)a	0.71 (0.56–0.89)	1.03 (0.71–1.46)	<0.001	0.84 (0.65–1.02)	1.20 (0.84–1.59)	<0.001
HDL-C (mmol/liter)	1.58 ± 0.29	1.44 ± 0.34	<0.001	1.40 ± 0.25	1.24 ± 0.27	<0.001
LDL-C (mmol/liter)	2.44 ± 0.73	2.56 ± 0.86	<0.001	2.50 ± 0.62	2.64 ± 0.69	<0.001
FBG (mmol/liter)	4.88 ± 0.38	5.22 ± 0.53	0.003	4.98 ± 0.35	5.25 ± 0.68	<0.001
Insulin (mU/L)a	5.06 (3.15–7.50)	7.38 (4.97–10.32)	<0.001	8.33 (5.53–11.18)	11.97 (8.22–17.22)	<0.001
HOMA-IRa	1.10 (0.67–1.67)	1.68 (1.13–2.42)	<0.001	1.86 (1.23–2.45)	2.72 (1.90–4.01)	<0.001
Adiponectin (ug/mL)a	7.21 (4.81–10.16)	6.38 (4.50–8.96)	0.002	5.25 (3.60–7.19)	4.58 (3.20–6.57)	<0.001
Lifestyle factors
Sleep time (hour/day)	8.59 ± 1.26	8.52 ± 1.24	0.294	8.52 ± 1.15	8.50 ± 1.13	0.316
MVPA (%)	62.3	53.7	0.001	57.1	52.1	0.044
Sedentary time ≥2 h/day (%)	43.1	47.2	0.120	47.4	48.5	0.639
Breakfast	4.5 ± 1.2	4.4 ± 1.2	0.142	4.4 ± 1.3	4.3 ± 1.4	0.105
Bean	2.9 ± 1.5	2.5 ± 1.4	<0.001	2.6 ± 1.4	2.6 ± 1.3	0.968
Meat	3.9 ± 1.4	3.5 ± 1.5	<0.001	3.8 ± 1.4	3.6 ± 1.5	<0.001
Seafood	2.4 ± 1.3	2.1 ± 1.3	<0.001	2.3 ± 1.3	2.1 ± 1.2	0.025
Vegetable	4.8 ± 0.8	4.7 ± 0.8	0.309	4.8 ± 0.7	4.8 ± 0.7	0.984
Fruit	4.3 ± 1.2	4.0 ± 1.4	<0.001	4.1 ± 1.4	4.0 ± 1.4	0.137
Dairy	4.0 ± 1.5	3.5 ± 1.6	<0.001	3.9 ± 1.5	3.7 ± 1.6	0.005
Soft drink	2.5 ± 1.5	2.4 ± 1.5	0.157	2.5 ± 1.4	2.4 ± 1.6	0.287
Diet score	30.2 ± 5.4	28.3 ± 5.5	<0.001	29.4 ± 5.3	28.7 ± 5.4	0.010

Data were shown as mean ± SD or median (interquartile) or percentage.

Diet scores were calculated according to their consumption frequency of dietary records including breakfast, bean, seafood, milk, vegetables, fruits, red meat, and soft drinks, in which a lower score indicated more reduced dietary quality and vice versa.

Boldface type indicates nominally significant values (P < 0.05).

BMI body mass index, WC waist circumference, FAT% fat mass percentage, MVPA moderate-to-vigorous physical activity, SBP systolic blood pressure, DBP diastolic blood pressure, LDL-C low density lipoprotein cholesterol, HDL-C high density lipoprotein cholesterol, FBG fasting blood glucose, HOMA-IR homeostasis model assessment-insulin resistance index.

^aNatural logarithm transformed.

Associations between adiponectin-related variants, dietary patterns and metabolic health in children with different weight status

Multivariable logistic regression analysis was performed to determine the independent effects of six SNPs from previously reported adiponectin-associated loci on childhood metabolic abnormalities (Supplementary Table 4). After adjustment for baseline age, sex, puberty, and BMI, each adiponectin decreasing G allele at the ADIPOQ rs6773957 locus was associated with a 23% higher risk of metabolic unhealthiness (OR = 1.23, 95% CI = 1.06–1.16, P = 0.008) in participants with normal weight. In contrast, each adiponectin decreasing T allele at the CDH13 rs4783244 locus was significantly associated with reduced risk of metabolic unhealthiness (OR = 0.85, 95% CI = 0.73–0.99, P = 0.044) in normal-weight children after adjustment for the above variables. Because CDH13 is known to encode T-cadherin, a novel receptor for high-molecular-weight (HMW) adiponectin, and the adiponectin decreasing T allele at rs4783244 locus is reported to be related to the sensitivity of adiponectin, thereby the association between CDH13 locus and the unhealthy metabolic phenotype became slightly stronger after further adjustment for ln-adiponectin in normal-weight participants (OR = 0.82, 95% CI = 0.69–0.96, P = 0.017) (Model 3, Table 2). There were no significant associations between the other SNPs and metabolic health (all P > 0.05). In addition, unhealthy dietary patterns were also associated with metabolic unhealthiness in both normal-weight and over-weight children (Table 2). In addition, the significant relationship between ADIPOQ rs6773957, CDH13 rs4783244, and ln-adiponectin levels in our data were similar to those observed previously in adults.

Table 2.

Independent associations of adiponectin-related locus and diet with metabolically unhealthy phenotype in different weight status.

	Normal weight			Overweight/Obesity
	OR	95% CI	P value	OR	95% CI	P value
ADIPOQ rs6773957-G
Model 1	1.23	1.06–1.43	0.008	1.04	0.90–1.20	0.612
Model 2	1.26	1.07-1.47	0.004	1.04	0.90–1.20	0.616
CDH13 rs4783244-T
Model 1	0.85	0.73–0.99	0.045	0.96	0.83–1.12	0.614
Model 3	0.82	0.69–0.97	0.017	0.93	0.79–1.09	0.367
Diet scores (per five score increase)
Model 1	0.76	0.69-0.83	<0.001	0.91	0.83–1.01	0.069
Model 4	0.76	0.69–0.84	<0.001	0.90	0.81–1.00	0.045

The data were shown as OR (95% CI).

Model 1: adjusted for gender, age, puberty, and BMI.

Model 2: Model 1 + diet scores and physical activity.

Model 3: Model 1 + dietary scores, physical activity, and ln adiponectin levels.

Model 4: Model 2 + ADIPOQ rs6773957 and CDH13 rs4783244.

Diet scores were calculated according to their consumption frequency of dietary records including breakfast, bean, seafood, milk, vegetables, fruits, red meat, and soft drinks, in which a lower score indicated more reduced dietary quality and vice versa. Boldface type indicates nominally significant values (P < 0.05).

The gene-diet interaction on metabolic health during childhood

We next tested for association of the two key variants with dietary patterns, adjusting for age, sex, and other lifestyle factors. No individual variant yielded a significant association with the dietary scores (data not shown). We then examined whether dietary scores modifies the effect of the key variants ADIPOQ rs6773957 and CDH13 rs4783244 on metabolic health in children with different weight statuses. Among normal-weight children, a significant interaction between ADIPOQ rs6773957 and dietary patterns was detected (P = 0.004) after adjustment for age, sex, puberty, physical activity, and BMI (Fig. 1). Notably, each adiponectin decreasing G allele at the ADIPOQ locus was associated with increased risks of metabolic abnormalities by 61% in participants with the lowest diet score (bottom tertile) (95% CI = 1.27–2.05, P = 9.18 × 10⁻⁵, Fig. 1A), compared with no significant association in children with the most healthy diet score (topmost tertile: both P > 0.1). In other words, the protection of a more healthy diet on metabolic health was more prominent in individuals with the adiponectin decreasing G allele (OR per five score = 0.61, 95% CI = 0.48–0.77, P < 0.001 for GG genotype, OR = 0.76, 95% CI = 0.66–0.88, P < 0.001 for GA genotype, Fig. 1B), while participants with the AA genotype showed no significant response to diet score (OR per five score = 0.88, 95% CI = 0.73–1.06, P = 0.163). No interactions were detected in children with overweight/obesity. No significant interactions were identified between the ADIPOQ and CDH13 loci, and other lifestyle factors shown in Table 1.

Fig. 1. The ADIPOQ-diet interaction on metabolic health in normal weight children.

The association of ADIPOQ rs6773957 in the stratification of diet score tertile (A) and the association of every five point increase in diet score (approximately one SD) in the stratification of ADIPOQ rs6773957 genotypes (B) with metabolically unhealthy status in normal-weight children. The data was shown as OR (95% CI). The ORs for every G allele at ADIPOQ rs6773957 or every 5 point increase in diet score were adjusted for age, gender, puberty, physical activity, and BMI. Diet scores were calculated according to their consumption frequency of dietary records including breakfast, bean, seafood, milk, vegetables, fruits, red meat, and soft drinks, in which a lower score indicated more reduced dietary quality and vice versa.

Associations between adiponectin related variants, childhood dietary patterns and metabolic health at 10-year follow-up

To provide longitudinal evidence for the gene-diet interaction, we further examined whether childhood dietary patterns modified the association of the ADIPOQ rs6773957 locus with metabolic abnormalities at follow-up. As shown in Fig. 2, similar to the baseline data, the interaction between childhood diet score and the ADIPOQ rs6773957 locus on metabolic health at follow-up was detected among those subjects who were normal weight at baseline (P = 0.029), although the sample sizes were much smaller in this analysis. In participants with the lowest diet scores (bottom tertile) at baseline, every G allele at the rs6773957 locus significantly increased the risks of metabolic abnormality at follow-up (OR = 3.85, 95% CI = 1.07–13.88, P = 0.039, n = 30), compared to a non-significant association in participants with the most healthy childhood diet scores (topmost combined tertiles: OR = 0.93, 95% CI = 0.47–1.81, P = 0.825, n = 97), after adjustment for baseline age, sex, puberty, BMI, time of follow-up and changes of BMI during follow-up (Fig. 2A). In the stratification of ADIPOQ rs6773957 genotypes, the long-term associations of a more healthy diet in childhood with metabolic health at follow-up was more prominent in individuals with the GG genotype (OR per five score = 0.21, 95% CI = 0.05–0.94, P = 0.042 for GG genotype), while participants with GA and AA genotypes showed no significant response to childhood diet score (OR per five score = 1.31, 95% CI = 0.63–2.72, P = 0.466 for GA genotype;OR = 1.63, 95% CI = 0.51–5.19 for AA genotype, P = 0.409;Fig. 2B). No interaction between childhood diet score and ADIPOQ rs6773957 locus on metabolic abnormalities at follow-up was detected in 208 participants who were overweight/obese at baseline.

Fig. 2. The ADIPOQ-diet interaction on metabolic health after a 10-year follow-up.

The associations of ADIPOQ rs6773957 in the stratification of childhood diet score tertile (A) and the associations of every five-point increase in childhood diet score (approximately one SD) in the stratification of ADIPOQ rs6773957 genotypes (B) with metabolically unhealthy status after 10-year follow-up. The ORs for every G allele at ADIPOQ rs6773957 or every 5 point increase in diet score were adjusted for baseline age, gender, puberty, BMI, time of follow-up, change of BMI during follow-up.

DISCUSSION

In this study, we evaluated the relationships of adiponectin-related genetic variations and dietary patterns with metabolic health in children with normal weight and obesity. In addition to replicating the associations between adiponectin levels and genetic variations, we found that SNPs rs6773957 at the adiponectin gene (ADIPOQ locus) and rs4783244 at the adiponectin receptor CDH13 were correlated with metabolic abnormalities, independent of BMI at baseline. Moreover, these loci were better related to metabolic health in normal weight than in children with overweight/obesity. Furthermore, we found diet can modify the genetic associations of the ADIPOQ locus with metabolic health in normal-weight children. That is, unhealthy dietary patterns may strengthen the metabolic risks conferred by the adiponectin-decreasing alleles at the ADIPOQ locus. Our longitudinal analysis further supported the ADIPOQ-diet interaction on metabolic abnormalities after a 10-year follow-up.

Our results were in line with the well-known biological roles of adiponectin in regulating metabolic homeostasis, such as improving adipose tissue function, insulin sensitivity, and lipid metabolism in experimental studies [6]. In humans, the studies of adiponectin have yielded a complicated picture. Most adult studies have shown that decreased adiponectin levels were associated with obesity and cardiometabolic diseases [8–10]. In the stratification of obesity, decreased adiponectin levels can differentiate metabolic disorders in adult individuals with and without obesity [11]. However, the Mendelian randomization studies using variation at the ADIPOQ locus as an instrumental variable to indicate the causal roles of adiponectin on cardiometabolic diseases remain unclear [17–22]. For example, in adults, both positive and negative associations between genetic variants at the ADIPOQ locus have been reported in relation to T2D, MS, and CVD [40–42]. These discrepancies could be driven by factors such as pleiotropic mechanisms and gene-environment interactions. In the present study, the G allele at the ADIPOQ rs6773957 locus was associated with decreased adiponectin levels and increased cardiometabolic risks in normal-weight children, further supporting the critical metabolic effects of adiponectin. It is possible that the ADIPOQ rs6773957 locus reduces adiponectin expression of local white adipose tissue and impair its function, ultimately increasing cardiometabolic risk. Indeed, the SNP rs6773957 is located at the 3’ regulatory region (3’ UTR) of the ADIPOQ gene, and increasing evidence shows that variants in 3’ UTR influence gene expression through mechanisms including interaction with microRNAs [43, 44]. Further studies are needed to explore whether this variation at the ADIPOQ locus in adipose tissue affects adiponectin expression. Of note, the associations between variation in the ADIPOQ locus and metabolic abnormalities were more evident in children with normal weight than those with overweight/obesity. One explanation for this different association may depend on the different genetic basis of adipose tissue dysfunction underlying the normal weight metabolically “obese” phenotype versus the MUO phenotype. According to the “adipose tissue expandability” hypothesis metabolic complications are attributed, not to obesity, but an individual’s capacity for adipose tissue expandability and expansion of adipose tissue could be of two causes, enlargement of adipocytes (unhealthy) or active adipogenesis (healthy expansion) [45, 46], we speculated that subjects who are of normal weight but metabolically unhealthy may have less adipose tissue expandability than those with obesity but metabolically healthy. The poor capacity for active adipogenesis can impair energy storage under the challenge of caloric excess, forcing lipids to be stored in nonadipose ectopic depots, and ultimately, resulting in insulin resistance and metabolic disorders even in the normal body weight. Given that ADIPOQ gene expression is a marker of healthy expansion of adipose tissue [6, 7], it is not difficult to understand why the rs6773957 locus at the regulatory region of ADIPOQ gene, interacting with diet, was more related to a metabolically unhealthy phenotype in children with normal weight than those with overweight/obesity. However, further mechanistic studies are warranted.

In addition to the ADIPOQ locus, we also evaluated the associations of metabolic abnormalities with other strong adiponectin-related loci reported from previous GWAS. Similar to previous studies in adults [13, 47], we also found that the T alleles at CDH13 rs4783244 were associated with lower adiponectin levels, but paradoxically with better metabolic profiles in children. Like previous findings from East Asian adults, our data also suggest a complex relationship between a variant at the CDH13 locus and metabolic traits that is more evident after controlling for its effect on blood adiponectin levels. The CDH13 gene encodes T-cadherin, a novel receptor for high molecular weight adiponectin that is widely expressed in cardiovascular tissues. The binding of high molecular weight adiponectin to T-cadherin has a vital role in metabolic homeostasis [48]. The influence of the CDH13 locus on metabolic health may be mediated by higher T-cadherin expression, and thereby higher adiponectin sensitivity [47]. Although the mechanisms warrant further investigation, these genetic associations highlight the important roles of adiponectin and related genetic variations in metabolic health. Notably, other adiponectin GWAS loci including WDR11FGF rs3943077, CMIP rs2925979, and PEPD rs889140, which have been reported to be associated with lipids, central obesity, and/or glucose traits in the adult population in previous studies [16, 49], did not show significant associations with metabolic health in current pediatric study. The absence of associations in our data might be attributed to age differences and/or the relatively smaller sample size.

A healthy diet, a potentially modifiable lifestyle factor, also showed protective associations on metabolic health in our study. Despite reservations about their utility, particularly for measuring energy intake [50], consistent with our results, several other studies have reported negative associations between healthy dietary patterns and obesity and obesity-related disorders [23–25], but the underlying mechanisms remain unclear. Given that in the present study the protection of healthy dietary patterns was more evident in normal-weight children than in those with overweight/obesity, the effects of healthy dietary patterns may not be simply through modulating the mass of adipose tissue. Some studies have suggested that improving the adipose tissue function, including alteration in adipokine secretion, maybe the potential mechanism underlying the impacts of dietary modification [23–25]. Interestingly, we found that dietary patterns modified the genetic association between variation at the ADIPOQ locus and metabolic health, suggesting the vital role of adiponectin and adiponectin-associated loci in linking dietary patterns and metabolic abnormalities. The metabolic risks associated with allelic variation at the ADIPOQ locus were augmented by about 60% in children who had the worst dietary patterns compared to those with the healthiest dietary patterns at baseline. In addition to this novel finding in a cross-sectional setting, we have also provided longitudinal evidence that the childhood dietary pattern can strengthen the metabolic risks conferred by this locus after ten years while consuming a healthy diet can attenuate these risks. From another point of view, we also found that children with a higher genetic predisposition to cardiometabolic risks seem to be more responsive to the favorable effects of improved dietary quality on metabolic health. Thus, targeting children with a genetic predisposition to lower adiponectin secretion and metabolic health impairment, would be useful to prevent CVD by nutritional and lifestyle interventions, and would be more timely than targeting adults based on traditional cardiometabolic risk factors. Our study gives new insights into potential strategies to reduce the risk of cardiometabolic diseases.

Our study had several strengths. The major strength was the availability of a large well-characteristic group of Chinese children, and particularly a subgroup with a 10-year follow-up that gave a possibility to analyze longitudinal effects of the gene-environment interaction on the risk of development of metabolic complications. Despite these strengths, there are also some limitations. Firstly, we merely assessed the frequency of food consumption at baseline, which may not provide a very representative picture of the diet over time. Secondly, our study cohorts were restricted to Chinese children; future replication studies are needed to evaluate the generalizability of our findings to other ethnic groups and older adults. Thirdly, like most previous studies, we defined the phenotype of metabolic unhealthiness based on the presence of classical MS components [4], however, this definition seems to be practicable but is still imprecise, since the absence of MS alone does not exclude the presence of other health-related problems [4]. Fourth, the main finding is that the investigated genetic and lifestyle factors have a more noticeable impact on the development of metabolic complications in normal-weight individuals than those with obesity. This result is not apparent and requires more work to explain the possibility. Lastly, we found a relationship between ADIPOQ rs6773957 and metabolic health, and the modification of dietary patterns based on both cross-sectional and longitudinal evidence, but we could not avoid the possibility of introducing selection bias, since the number of subjects returning for follow-up evaluation after ten years was relatively small compared to the original cohort. Further studies with larger cohorts are needed to verify these findings, and additional experimental studies are required to establish the mechanistic links involved.

In conclusion, our study gives novel insights into the heterogeneity of normal weight and obesity, and provides suggested mechanisms for pediatric metabolic abnormalities. We found that adiponectin-related genetic variants were correlated with metabolic abnormalities in our large cohort of children. Interestingly, these loci were better related to metabolic health in normal weight than in the overweight/obesity group, suggesting the important role of adipose tissue dysfunction, including alteration in adipokine secretion, on metabolic abnormalities, rather than increased fat mass alone. Furthermore, we provide the first evidence that dietary patterns can modify the genetic effects of ADIPOQ locus on metabolic health in normal-weight children i.e., unhealthy dietary patterns may strengthen the metabolic risks conferred by the ADIPOQ locus compared to the insignificant genetic associations in healthy dietary patterns. Moreover, our longitudinal analysis based on 10-years follow-up data further indicated the ADIPOQ-diet interaction on metabolic abnormalities. Our study has the potential to impact the optimization of intervention options and regimes in the management of pediatric metabolic abnormalities.

DATA AVAILABILITY

All datasets used in the current investigation are available from the corresponding author upon reasonable request.