Adiponectin, Adiposity, and Insulin Resistance in Children and Adolescents

Abstract

Context: Determinants of adiponectin and its association with insulin resistance (IR) are less well studied in youth than in adults.

Objectives: The objective of the study was to describe, in youth, the age- and sex-specific distribution of adiponectin concentrations and the association with demographic, anthropometric, and lifestyle factors, parental diabetes, and markers of IR.

Design, Setting, Participants: We studied 1632 French Canadian youth aged 9, 13, and 16 yr who participated in the Québec Child and Adolescent Health and Social Survey, a province-wide, school-based survey conducted in 1999.

Results: Boys had lower adiponectin concentrations than girls by 17% (P < 0.0001). At age 16 yr, mean adiponectin concentrations were 27.7% (boys, P < 0.0001) and 13.3% (girls, P < 0.0001) lower than at age 9 yr (p_interaction = 0.009). Mean adiponectin decreased for every unit increase in body mass index (BMI) Z-score by 8.1% in boys and 11.2% in girls (P < 0.0001). Growth-related change in BMI explained half the age effect in boys and all the age effect in girls. Self-reported pubertal status, physical activity, smoking, and parental diabetes were not independently associated with adiponectin. Fasting insulin and homeostasis model assessment-IR were not associated with adiponectin concentration. However, the interaction of adiponectin and BMI Z-score was significant in a multiple regression model of fasting insulin.

Conclusions: Male sex and changes in body fat may be major determinants of the decreasing adiponectin concentrations of growing youth, which are accompanied by a dissociation of adiponectin and markers of IR. The relationship between adiposity and markers of IR is attenuated in those with higher adiponectin concentrations, making adiponectin a potential intervention target or risk marker.

ADIPONECTIN IS AN adipokine that is secreted solely by adipose tissue, and its concentration is inversely related to the degree of adiposity. It circulates in high concentrations as a multimer. Adiponectin has structural homology to TNFα (1), its expression is regulated by peroxisomal proliferator-activated receptor (PPAR)γ (2), and its receptors are highly expressed in liver and muscle (3). These features make it a candidate mediator in the development of obesity-related insulin resistance (IR). Longitudinal studies in adults suggest that low adiponectin is a risk factor for type 2 diabetes: adiponectin concentrations are 22% lower in adults who subsequently develop diabetes than in those who do not (4). In addition, the negative association between adiponectin and IR appears to be independent of adiposity (5, 6).

Although the distribution of adiponectin concentrations has been reported in adults (7), there is no information about its distribution in representative samples of youth. Limited data suggest that adiponectin is 33% lower in obese children with diabetes than those without diabetes (8). Nine pediatric studies (8–16) have shown a negative association between adiponectin and various measures of IR, but the association was independent of adiposity in only five of these reports (8–12). In youth, puberty leads to changes in body fat, contributes to IR (17), and may modify the association between adiponectin and IR.

The objectives of this report were: 1) to describe the age- and sex-specific distribution of adiponectin concentrations in a large, representative sample of children and adolescents; 2) to examine the relationship of adiponectin with age, pubertal status, adiposity, and lifestyle factors; and 3) to describe the association between adiponectin and surrogate measures of IR.

Subjects and Methods

Subjects

The study population included a sample of French Canadian children and adolescents who participated in the Québec Child and Adolescent Health and Social Survey (QCAHS), a school-based survey conducted between January and May 1999 in the province of Québec, Canada. The survey design and methods, reported previously (18), are only summarized here. The QCAHS used a cluster sampling design to draw three provincially representative, independent samples of youth aged 9, 13, or 16 yr (one sample per age).

Among eligible youth, questionnaires and anthropometry were completed by 82% (1243 of 1520), 78% (1161 of 1498), and 76% (1133 of 1495) of 9, 13, and 16 yr olds, respectively. The analyses described here were restricted to French Canadian participants because some of the data were collected only in this subgroup as part of a QCAHS substudy of genetic variables. French Canadians comprised 80% (999 of 1243), 78% (908 of 1161), and 82% (923 of 1133) of all 9-, 13-, and 16-yr-old QCAHS participants, respectively. A fasting blood sample was provided by 63% (628 of 999), 69% (629 of 908), and 75% (695 of 923) of 9-, 13-, and 16-yr-old French Canadians, respectively. Twelve percent (235 of 1952) of participants were excluded because parents refused analyses other than glucose and lipids or because blood samples were thawed or of insufficient quantity. Three individuals were excluded because they had diabetes. Those missing pubertal status (n = 59), physical activity scores (n = 6), and 13 and 16 yr olds missing data on tobacco use (n = 21) were excluded. Those who provided blood samples did not differ significantly from those who did not, by sex, body mass index (BMI) Z-score, parental income or education, or parental history of diabetes. The study was approved by the Ethics Review Board of Ste-Justine Hospital. Informed assent and consent were obtained from participants and their legal guardians.

Clinical variables

Height was measured to the nearest 0.1 cm at maximal inspiration using a standard measuring tape and a triangular level. Weight was measured in light indoor clothing with shoes removed. Skinfold thickness was measured to the nearest 0.1 cm with a Lange caliper (Beta Technology, Santa Cruz, CA). BMI was calculated as weight divided by the square of height (kilograms per square meter). Participants and their parents both completed questionnaires. Pubertal status was assessed using a modification of a validated self-report scale (19), by indicating for each of the following body changes if they had not started, barely started, definitely started, or stopped changing: pubic hair, facial hair, and voice change for boys or pubic hair and breast development for girls. Menstruation was categorized as not started or started. Responses were used to categorize participants as prepubertal (all responses “not started”), postpubertal [all responses “stopped changing” (menstruation “started”)], or intrapubertal (any other combination of responses). The physical activity score was based on a 1-wk recall modified from Sallis et al. (20). Participants were asked to indicate which of 18 activities they performed for at least 15 min on each of the previous 7 d. Their responses were summed to create the physical activity score (maximum possible score 126). Smoking was determined by asking 13 and 16 yr olds the number of cigarettes usually smoked per day in the last 30 d. Nine year olds were asked whether they had ever smoked a whole cigarette. For multivariable analyses, all 9 yr olds were categorized as smoking 0 cigarettes per day because only 2.1% reported ever having smoked a whole cigarette. Parental history of diabetes was obtained from the parent questionnaire.

Biochemical variables

After an overnight fast, venous blood was collected between 0800 and 1000 h in a 1 g/liter EDTA collection tube and placed on ice. Samples were centrifuged on site within 45 min of collection, transported on dry ice, and stored at −80 C. Adequacy of the fasting period was checked by nurses before blood was collected. Plasma glucose was measured with the CX7 analyzer (Beckman Coulter, Fullerton, CA) using the glucose oxidase method. Interassay coefficients of variation (CVs) were 3.8, 1.3, and 1.4% at 2.17, 6.38, and 13.69 mmol/liter, respectively (n = 24) (Triad Link levels 1, 2, and 3; Beckman Coulter). Plasma insulin was measured with the ultrasensitive insulin assay on the Access immunoassay system (Beckman Coulter). Interassay CVs were 4.1 and 5.0% at 92 (n = 24) and 285 pmol/liter (n = 23), respectively (Lyphochek Immunoassay Plus Control, levels 1 and 2; Bio-Rad Laboratories, Irvine, CA). Plasma adiponectin was measured by RIA (Linco Research, Inc., St. Charles, MO). Interassay CVs were 10.5 and 12.2% at 3.4 and 37.7 mg/liter, respectively (n = 24). Homeostasis model assessment of IR (HOMA-IR) was calculated as insulin (milliunits per liter) × glucose (millimoles per liter)/22.5 (21).

Statistical analysis

Sample quantiles of adiponectin were used to estimate the population percentiles. Nonparametric 95% confidence intervals (CIs) for the percentiles were constructed by the method of Hutson (22). Percentiles were considered significantly different if their 95% CIs did not overlap.

Variables with skewed distributions (BMI, subscapular and triceps skinfolds, adiponectin, insulin, HOMA-IR) were log_e transformed. Age- and sex-specific Z-scores were estimated for each participant for BMI, subscapular and triceps skinfolds, and adiponectin from the weighted distributions of the study sample. Z-scores were computed as (value − mean)/sd. Hierarchical maximum likelihood linear regression was used to estimate regression coefficients for univariate and multivariate associations. Explanatory variables were treated as fixed effects, and clustering between subjects in the same school was treated as a random effect. Regression coefficients for models of 100Log_eAdiponectin or 100Log_eInsulin represent the percentage difference in adiponectin or insulin concentration, respectively, for 1 U increment in the explanatory variable. Statistical significance was assessed using F tests. Least squares linear regression was used to estimate the proportion of variance (R²) explained by the models. Because of the complex survey design, sampling weights and clustering effects were estimated and incorporated into all computations except correlations and regression models. Statistical analyses were performed with SAS software (SAS Institute, Inc., Cary, NC) and SUDAAN (Research Triangle Institute, Research Triangle Park, NC).

Results

Participant characteristics are shown in Table 1. Parental history of diabetes was reported for 7% (46 of 639) of boys and 5% (31 of 668) of girls (among those whose parents responded to the questions about diabetes).

TABLE 1.

Characteristics of participants

Characteristic	Boys, n = 804	Girls, n = 828
Age, n (%)
9 yr	248 (31)	260 (32)
13 yr	269 (33)	235 (28)
16 yr	287 (36)	333 (40)
Pubertal status, n (%)
Prepubertal	109 (14)	118 (14)
Intrapubertal	685 (85)	581 (70)
Postpubertal	10 (1)	129 (16)
BMI, kg/m2, mean ± sd	20.2 ± 4.2	20.3 ± 4.5
BMI category1, n (%)
Overweight	106 (13)	113 (14)
Obese	80 (10)	66 (8)
Fasting insulin, pmol/liter, mean ± sd	41.1 ± 34.2	45.8 ± 25.4
Fasting glucose, mmol/liter, mean ± sd	5.3 ± 0.4	5.1 ± 0.4
Physical activity score, mean ± sd	9.6 ± 7.7	7.7 ± 6.4
Smoking (cigarettes/d)2, n (%)
0	421 (76)	375 (66)
1–5	81 (14)	123 (22)
>5	54 (10)	70 (12)

Physical activity score was the sum of the number of activities from a list of 18 that were performed for at least 15 min on each of the previous 7 d (maximum score 126).

¹

Defined according to the 2000 Centers for Disease Control and Prevention growth charts (39 ), overweight: ≥85th and <95th percentile; obese: ≥95th percentile.

²

Thirteen- and 16-yr-olds only; boys, n = 556; girls, n = 568.

Adiponectin distribution

Age- and sex-specific distributions of adiponectin concentrations are shown in Table 2. Distributions of adiponectin concentrations were skewed positively. Girls had higher adiponectin concentrations than boys, except at the fifth and 10th percentiles for 9 yr olds. Overall, adiponectin concentrations decreased with age. However, differences were smaller between 13 and 16 yr olds than between 9 and 13 yr olds. Age-related differences were less marked in girls than boys and were not present at the lower percentiles.

TABLE 2.

Distribution of adiponectin concentrations (mg/liter) by sex and age

Percentile (95% CI)	Boys			Girls
	9 yr, n = 248	13 yr, n = 269	16 yr, n = 287	9 yr, n = 260	13 yr, n = 235	16 yr, n = 333
5th	5.8 (5.2–6.4)	4.4 (4.1–4.7)	4.0 (3.6–4.6)	5.8 (5.3–6.6)	5.4 (5.1–6.1)	5.2 (4.6–6.2)
10th	6.6 (6.1–6.8)	5.0 (4.6–5.6)	4.8 (4.4–5.1)	6.9 (6.3–7.5)	6.6 (6.0–7.0)	6.4 (5.8–7.1)
25th	8.4 (7.8–9.0)	6.6 (6.2–6.9)	6.1 (5.8–6.7)	9.3 (9.0–9.9)	8.1 (7.6–8.5)	8.0 (7.5–8.3)
50th	10.6 (10.1–11.4)	8.6 (8.1–9.1)	7.9 (7.4–8.3)	11.9 (11.2–12.6)	10.4 (9.8–11.1)	9.9 (9.3–10.3)
75th	13.8 (13.0–14.5)	11.0 (10.2–11.5)	10.0 (9.5–10.5)	15.2 (13.9–15.9)	13.4 (13.0–14.0)	12.3 (11.6–13.0)
90th	16.6 (15.5–17.9)	14.2 (12.7–14.8)	12.4 (11.6–13.0)	18.5 (17.5–20.6)	16.3 (15.1–18.0)	14.8 (14.1–15.9)
95th	18.4 (17.0–19.0)	15.3 (14.4–15.9)	13.5 (12.8–15.4)	21.3 (19.6–23.6)	18.4 (16.8–18.6)	16.7 (15.5–17.9)
Mean (95% CI)	11.2 (10.7–11.7)	9.1 (8.6–9.5)	8.2 (7.8–8.5)	12.5 (11.9–13.0)	10.9 (10.4–11.4)	10.3 (9.9–10.6)

Associations with adiponectin

In univariate linear regression, sex explained 5% of the variation in adiponectin concentrations. Adiponectin concentrations in boys were 17% (95% CI 14–21%) lower than in girls. Due to a significant age sex interaction (described below), all subsequent analyses were conducted separately for boys and girls. Table 3 shows the sex-specific univariate effect sizes of age, pubertal status, anthropometric, and lifestyle factors on adiponectin concentration. Increased age and advanced pubertal status showed substantial negative associations with adiponectin concentration. All measures of adiposity were negatively associated with adiponectin. Neither physical activity nor smoking was significantly associated with adiponectin concentrations. Parental history of diabetes was not associated with adiponectin concentrations among participants whose parents responded (P = 0.7 for boys and girls) (data not shown).

TABLE 3.

Univariate associations between selected variables and adiponectin concentration

Explanatory variable	Boys (n = 804)			Girls (n = 828)
	Beta, %1	se	P	Beta, %1	se	P
Age (yr)
9	Ref.			Ref.
13	−22.1	3.2	<0.0001	−12.7	3.5	0.0004
16	−31.3	3.2	<0.0001	−18.1	3.4	<0.0001
Pubertal status
Prepubertal	Ref.			Ref.
Intrapubertal	−20.1	4.0	<0.0001	−13.9	3.8	0.0003
Postpubertal	−33.5	12.5	0.008	−23.5	4.8	<0.0001
BMI (kg/m2)	−3.1	0.3	<0.0001	−3.0	0.3	<0.0001
BMI Z-score, 1 sd	−8.8	1.3	<0.0001	−11.2	1.2	<0.0001
Subscapular skinfold Z-score, 1 sd	−6.5	1.3	<0.0001	−10.9	1.2	<0.0001
Triceps skinfold Z-score, 1 sd	−2.7	1.4	0.05	−8.0	1.3	<0.0001
Physical activity score	−0.2	0.2	0.2	0.1	0.2	0.5
Smoking (cigarettes/d)2
0	Ref.			Ref.
1–5	−5.7	4.6	0.2	−3.6	3.7	0.3
>5	−3.5	5.5	0.5	−6.0	4.6	0.2

¹

Beta refers to the percent difference of mean adiponectin concentration for 1 U increment of the explanatory variable.

²

Smoking data for 13- and 16-yr-olds only (boys, n = 556; girls, n = 568).

BMI Z-score and scapular and triceps skinfold thickness Z-scores were highly correlated (Pearson r ranged from 0.74 to 0.82, P < 0.0001) and collinear, so we included only the BMI Z-score in multivariable analyses because it is the most clinically applicable measure. Parental history of diabetes was not included in multivariable analyses because of the low response rate.

Only age and BMI Z-score were significantly associated with adiponectin in a multivariable model that included age, pubertal status, BMI Z-score, physical activity, and smoking (Table 4, model A). The effect of pubertal status was eliminated after adjusting for age (P = 0.7 and 0.6 for boys and girls, respectively) (data not shown). Boys aged 13 and 16 yr had adiponectin concentrations 19.5 and 27.7% lower than 9 yr olds, respectively; a 1 sd increase in BMI Z-score was associated with an 8.1% decrease in mean adiponectin concentration. Girls aged 13 and 16 yr had adiponectin concentrations 8.8 and 13.3% lower than 9 yr olds, respectively; a 1 sd increase in BMI Z-score was associated with an 11.2% decrease in mean adiponectin concentration. However, these variables explained only 16 and 15% of the variation in adiponectin concentrations in boys and girls, respectively.

TABLE 4.

Multivariate models of adiponectin concentration

Explanatory variable	Boys (n = 804)			Girls (n = 828)
	Beta, %1	se	P	Beta, %1	se	P
Model A
Age (yr)2
9	Ref.			Ref.
13	−19.5	3.7	<0.0001	−8.8	3.9	0.02
16	−27.7	3.9	<0.0001	−13.3	4.2	0.001
BMI Z-score, 1 sd3	−8.1	1.3	<0.0001	−11.2	1.2	<0.0001
Model B
Age (yr)2
9	Ref.			Ref.
13	−12.9	3.9	0.0008	0.3	4.0	0.9
16	−16.7	4.3	<0.0001	−0.1	4.3	0.9
BMI (kg/m2)	−2.3	0.4	<0.0001	−2.8	0.3	<0.0001

All models are adjusted for pubertal status, physical activity score and smoking. For boys, R² = 0.16; for girls, R² = 0.15.

¹

Beta refers to the percent difference of mean adiponectin concentration for 1 U increment of the explanatory variable.

²

Interaction between sex and age, P = 0.009.

³

Interaction between sex and BMI Z-score, P = 0.10.

Adjusting for BMI instead of BMI Z-score markedly reduced the effect size of age in boys and completely eliminated it in girls (Table 4, model B). BMI is a surrogate measure of total body fat. BMI Z-score is a standardized measure that indicates the relative degree of fatness of an individual, compared with sex- and age-matched peers. Conceptually, the difference between BMI and BMI Z-score is the fat mass that is associated with growth and development. Comparison of the age parameters in models A and B suggests that almost half the age effect in boys and all of the age effect in girls is accounted for by growth-related change in BMI. This finding emphasizes the importance of body fat as a determinant of adiponectin concentration.

There was a significant age-sex interaction whereby adiponectin concentration decreased to a greater extent with increased age in boys than in girls (P = 0.009). There was also a significant interaction between BMI Z-score and age in boys, with 13- and 16-yr-old boys showing a greater decrease in adiponectin with increased fatness than 9-yr-old boys (P = 0.007). This was not observed in girls (P = 0.5). Sex did not modify the relationship between BMI Z-score and adiponectin (P = 0.1).

Adiponectin and markers of insulin resistance

Fasting insulin was used as a surrogate measure of IR. Table 5 shows the effect of adiponectin on insulin concentration. Insulin concentration was 6.8% (95% CI 3.3–10.3) lower in boys and 9.3% (95% CI 5.8–12.8) lower in girls for each 1 sd increase in adiponectin, in univariate analysis. Adiponectin explained only 1.6–2.9% of the variation in insulin concentrations. After adjusting for BMI Z-score, adiponectin had no independent effect on insulin concentration. However, higher concentrations of adiponectin did attenuate the detrimental effect of BMI Z-score on insulin concentrations (for interaction term adiponectin Z-score × BMI Z-score, P = 0.02 and P = 0.0007 for boys and girls, respectively) (Table 5, model 4, and Fig. 1).

Fig. 1.

Association between BMI Z-score and insulin concentration at selected adiponectin Z-scores after adjustment for age, pubertal status, physical activity score, and smoking.

TABLE 5.

Models showing the relationship between adiponectin and fasting insulin concentrations

Model	Explanatory variable	Boys (n = 804)				Girls (n = 828)
		Beta, %1	se	P	Model R2	Beta, %1	se	P	Model R2
12	Adiponectin Z-score, 1 sd	−6.8	1.8	0.0002	0.016	−9.3	1.8	<0.0001	0.029
23	Adiponectin Z-score, 1 sd	−7.1	1.8	<0.0001	0.18	−9.4	1.6	<0.0001	0.26
33	Adiponectin Z-score, 1 sd	−1.7	1.6	0.3	0.37	−2.6	1.5	0.1	0.41
	BMI Z-score, 1 sd	25.4	1.6	<0.0001		21.8	1.6	<0.0001
43	Adiponectin Z-score, 1 sd	−1.8	1.6	0.3	0.38	−2.5	1.5	0.1	0.41
	BMI Z-score, 1 sd	24.7	1.6	<0.0001		20.2	1.6	<0.0001
	Adiponectin Z-score × BMI Z-score	−3.5	1.5	0.02		−4.9	1.4	0.0007

¹

Beta refers to the percent difference of mean insulin concentration for an increase of one age and sex-specific sd of adiponectin, or BMI. For the interaction term in model 4, beta may be interpreted as the incremental change in β_{BMI Z-score} for every 1 sd increase in adiponectin or vice versa.

²

Model 1, unadjusted.

³

Models 2, 3, and 4, adjusted for age, pubertal status, physical activity score, and smoking.

The same analyses repeated with HOMA-IR as the surrogate measure of insulin resistance did not change any of the conclusions. Parameter estimates for the effect of adiponectin on HOMA-IR were within 0.5 absolute percentage points of those for its effect on fasting insulin concentration (data not shown).

Discussion

To our knowledge, this is the first report of the distribution of adiponectin concentrations in a representative sample of Caucasian peripubertal youth. It is also the largest study to date addressing the association of adiponectin with age, sex, adiposity, lifestyle factors, and markers of IR in youth. Variations in adiponectin by sex, age, and adiposity were of the same magnitude as those that predict diabetes in adults (4), suggesting that the differences in adiponectin apparent in this group of healthy children may be relevant to future disease risk.

Sex differences in adiponectin concentrations

Both the age- and sex-specific distributions and regression analyses showed a clear sex difference in adiponectin concentrations. Girls had higher mean adiponectin concentrations than boys at all ages. With our large population-based sample, we were able to show this difference even in the youngest age group (most of whom were prepubertal), in whom previous studies had failed to demonstrate sex differences (9, 14, 15, 23, 24). In addition, the progressive decline in adiponectin concentrations with age was more than twice as great in boys as girls. Bottner et al. (24) showed similar patterns in German boys, but in girls, they found no change in adiponectin concentrations through puberty, possibly because of a small sample size or the exclusion of overweight children. Furthermore, in our study, boys and girls had different age-specific relationships between adiposity and adiponectin concentrations. Older boys had a steeper decrease in adiponectin with increasing BMI Z-score than younger boys. In girls, age did not modify the association between BMI Z-score and adiponectin concentrations.

Prepubertal sexual dimorphism has also been reported for body composition, leptin, IGF-I concentrations (25), and HOMA-IR (26), suggesting that unknown mechanisms other than sex hormones contribute to sex-related differences in metabolism and energy homeostasis, even before puberty begins. However, testosterone does seem to influence adiponectin concentrations at older ages. Two studies of pubertal children reported that adiponectin concentrations were negatively correlated with androgen concentrations (particularly testosterone in boys) (15, 24). These studies support the observed interaction between sex and age in our study. Additional evidence for testosterone-suppressing adiponectin concentrations comes from studies in vitro and in eugonadal and castrated mice treated with testosterone (27) and in hypogonadal men treated with testosterone (28). Furthermore, our results suggest that during puberty, boys may develop an increased susceptibility to obesity-induced hypoadiponectinemia. This may be due to direct hormonal effects or sex-specific changes in fat distribution that develop during puberty (29). The tendency of men to accumulate visceral fat (30) and the relatively greater contribution of visceral fat than sc fat in determining adiponectin concentration (11) have been well documented.

Adiposity and adiponectin concentrations

The inverse association between several measures of fatness and adiponectin was consistent with previous reports in youth (8–16, 23, 24). In this study, we were able to further distinguish the effects of relative fatness and the normal accrual of fat with pubertal development. We used BMI Z-score as a measure of relative degree of fatness, compared with sex- and age-matched peers. We used BMI as a measure of total fat. In growing school-age children, the BMI measure incorporates both relative fatness (BMI Z-score) and a unique fat mass that increases directly with age. This fat mass is not anatomically distinct, but defining it allowed us to determine how much of the association between age and adiponectin was due to normal age-related changes in body composition. In girls, all of the age effect on adiponectin could be explained by BMI. In boys, a large proportion of the age effect could be explained by BMI (presumably the remainder is the androgen effect discussed above). It seems that even among growing youth, total fat mass is the major determinant of adiponectin concentrations, and the age effect is largely a result of increased fat mass with increased age. One interpretation of this finding is that not all decreases in adiponectin are unfavorable. There may be an unidentified adaptive physiological reason that normal accrual of body fat is associated with decreasing adiponectin during the pubertal years.

Adiponectin and markers of insulin resistance

Our results do not support an independent effect of adiponectin on markers of IR in youth at a population level. In combination with the association between normal fat accrual and lower adiponectin during the pubertal years, this suggests that there may be mechanisms to reduce the effects of low adiponectin on IR in this age group. Previous reports support a role for adiponectin in the pathogenesis of IR (5, 6) and type 2 diabetes (4) in adults. However, only five (8–12) of nine pediatric studies (8–16) demonstrated an independent association between adiponectin and measures of IR. All three studies that used direct measures of IR showed an independent association (10–12). Studies using surrogate measures of IR that showed an independent association sampled a higher proportion of obese individuals. The existence of a stronger association between adiponectin and IR with increasing adiposity could partly explain the discrepant findings. Our demonstration of an interaction between BMI Z-score and adiponectin concentration on insulin concentration supports this hypothesis.

In the QCAHS, adiponectin concentrations modified the relationship between BMI Z-score and fasting insulin concentration, suggesting that high adiponectin concentrations protect against obesity-induced IR. Because BMI is a rough index of body fat, such a hypothesis can only be tested with more rigorous measures of body composition. However, mouse models treated with adiponectin showed that adiponectin protects against IR to a greater degree in obese than normal-weight mice (31). The mechanism is not known, but TNFα, which correlates with obesity and obesity-induced IR (32), may be involved. TNFα expression in adipocytes is suppressed by adiponectin (33). This may explain how high adiponectin concentrations in overweight individuals protect against IR. PPARγ may also have a role. PPARγ agonists increase adiponectin concentrations in adults with IR (2) and type 2 diabetes (34). These drugs are known to improve insulin sensitivity while causing weight gain. This apparent paradox may be partly explained by the attenuating effect of high adiponectin concentrations on obesity-related IR, a hypothesis supported by our results. If this attenuating effect of higher adiponectin concentrations on obesity-induced IR is replicated in prospective studies, adiponectin may prove to be an intervention target. Interventions directed at this mechanism specifically may include PPARγ agonists.

Potential value of normative data

The possible role of adiponectin as an effect modifier of the adiposity-IR relationship also makes it a potential risk stratification tool. Although we do not yet fully understand the importance of absolute concentrations of adiponectin, the age- and sex-specific percentile norms we have reported may be useful in the future for identifying children at risk of morbidity.

The additional value of these percentile distributions will be in the evaluation of population-wide trends with the evolution of the obesity epidemic and for comparison with groups of children with the metabolic syndrome.

Study limitations

After accounting for the effects of sex, age, puberty, anthropometry, and lifestyle factors, 80% of the variation in adiponectin concentration remained unexplained. Genetic or other unknown metabolic or hormonal factors may be important determinants. However, it may not be possible to explain much more of the variation in adiponectin concentrations because of biological variation; adiponectin shows random pulsatility in its secretion (35). Furthermore, measurement of total adiponectin may not be the appropriate biological measure to assess associations with the variables of interest. The lack of a statistically significant association between adiponectin concentration and measures of IR does not rule out the possibility that the distribution of multimeric complexes or intact vs. globular domain adiponectin are important in determining insulin sensitivity among youth.

Further limitations to this study include the relatively low proportion (roughly 54%) of French Canadian youth sampled who had adiponectin measured, which may have introduced selection bias. However, those who provided blood samples were similar to those who did not. Pubertal status was not independently associated with adiponectin. The large proportion of intrapubertal participants may indicate misclassification of both prepubertal and postpubertal participants, which could have diluted a true effect. We also found no associations between adiponectin and lifestyle factors or family history of diabetes. There is limited evidence of these associations in adults (36–38). Although there may truly be no associations in youth (particularly for tobacco use, given the low cumulative exposure), it is possible that our measures were unable to detect meaningful differences in these variables or that misclassification may have diluted the effect (e.g. up to 2.1% of 9 yr olds may have been incorrectly designated as nonsmokers). Finally, drawing conclusions beyond those of associations should be avoided in cross-sectional studies. Prospective studies in youth are needed to evaluate the role of adiponectin in the obesity-IR pathway.

Conclusions

Low adiponectin was associated with male sex, increased age, and adiposity but not physical activity, smoking, or parental history of diabetes. The association with age was driven mainly by changes in BMI, especially in girls. This suggests that fat mass, no matter what its origin (developmental or energy excess), is a major determinant of adiponectin concentrations in youth. Although adiponectin was not independently associated with markers of IR, it did appear to attenuate the detrimental effect of adiposity. The role of adiponectin in the obesity-IR pathway is less clear in healthy youth than adults. Further interventional or longitudinal studies are needed to determine whether adiponectin can protect against obesity-mediated IR among youth.

Acknowledgments

This work was supported by the Québec Ministry of Health and Social Services and Health Canada. The study was funded by the Canadian Institutes of Health Research (MOP-68983). Z.P. was supported by a fellowship from the Canadian Diabetes Association. J.O. holds a Canada Research Chair in the Childhood Determinants of Adult Chronic Disease.

Disclosure of potential conflicts of interest: Z.P., E.E.D., J.O., G.P., E.L., R.W.P., and M.L. have nothing to declare.

Abbreviations:

 
• BMI,

  Body mass index;

 
• CI,

  confidence interval;

 
• CV,

  coefficient of variation;

 
• HOMA-IR,

  homeostasis model assessment of IR;

 
• IR,

  insulin resistance;

 
• PPAR,

  peroxisomal proliferator-activated receptor;

 
• QCAHS,

  Québec Child and Adolescent Health and Social Survey.