Introduction

Human circulating adiponectin is a well-described adipocytokine exerting anti-inflammatory (Yokota et al. 2000) and insulin-sensitizing effects (Yamauchi et al. 2001; Berg et al. 2001). An inverse relationship to obesity and BMI with beneficial effects for insulin sensitivity was shown (Arita et al. 1999; Cnop et al. 2003). Adiponectin serum levels may be influenced by nutritional compounds as well as physical activity, environmental components and gender (Henneman et al. 2010; Antoniades et al. 2009; Mantzoros et al. 2006). In particular, a 1.5-fold higher concentration in women compared to men was described (Heid 2006). Besides several other loci such as ARL15 (ADP-ribosylation factor-like 15 gene locus) (Richards et al. 2009), the strongest genetic determinants influencing circulating adiponectin concentrations were identified within the ADIPOQ locus (adiponectin gene locus) (Ling et al. 2009; Heid et al. 2010; Richards et al. 2009).

Physiologic effects of adiponectin are mainly mediated by binding to one of its two receptor isoforms (AdipoR1 and AdipoR2) (Zhao et al. 2005) consequently activating signaling cascades such as adenosine monophosphate-activated protein kinase (AMPK) (Yamauchi et al. 2002), and PPARα transcription factor (peroxisome proliferator-activated receptor alpha) or NF-κB (nuclear factor ‘kappa-light-chain-enhancer’ of activated B cells) (Thundyil et al. 2012). AdipoR1 is most abundantly expressed in skeletal muscle and highly affine for globular adiponectin, while AdipoR2 binds full-length and globular adiponectin and is predominantly expressed in liver (Yamauchi et al. 2003). AdipoR1 and AdipoR2 are also expressed in human adipocytes (Rasmussen et al. 2006; Fasshauer et al. 2004; Li et al. 2007). Circulating adiponectin was shown to be involved in enhanced glucose uptake via GLUT4 (glucose transporter 4) (Ceddia et al. 2005; Mao et al. 2006), as well as in enhanced fatty acid uptake and oxidation in skeletal muscle in animal models (Tomas et al. 2002; Yoon 2006). Adiponectin-dependent AMPK activation may demonstrate a link to beneficial effects of this adipokine on metabolic and cardiovascular systems (Kahn et al. 2005). However, beside these important aspects of adiponectin in peripheral tissues, less is known about its central effects in brain. One study demonstrated that adiponectin administration stimulated the AMPK activation in the arcuate hypothalamus influencing food uptake and energy expenditure (Kubota et al. 2007). While elevated adiponectin levels were present in serum and cerebrospinal fluid during fasting state in mice, these levels were normalized after refeeding, suggesting that adiponectin might influence food intake via central mechanism in the brain (Kubota et al. 2007). One may argue whether similar central effects exist in humans (Pan et al. 2006; Kos et al. 2006). Consistent with potential central effects, human adiponectin levels were also shown to be altered in eating disorders such as anorexia nervosa, binge eating disorder and bulimia nervosa (reviewed in, e.g., Bou Khalil and El Hachem 2014). In particular, many studies demonstrated elevated adiponectin serum levels in female patients affected with anorexia nervosa (Modan-Moses et al. 2007; Pannacciulli et al. 2003; Terra et al. 2013), while binge eating disorder was related to decreased circulating adiponectin (Monteleone et al. 2003; Carnier et al. 2012). It is worth noting that inconsistent data exist for patients suffering from bulimia nervosa (Housova et al. 2005; Tagami et al. 2004; Monteleone et al. 2003). Taken together, besides its well-known effects in terms of obesity, type 2 diabetes and related metabolic conditions adiponectin seems to be involved in food intake and energy expenditure as well as in the pathophysiology of eating disorders.

Here, we tested the hypothesis that adiponectin serum levels correlate with human eating behavior factors measured by the German version of the three-factor eating questionnaire (Fragebogen zum Essverhalten—FEV). Further, we analyzed whether genetic variation in the ADIPOQ locus influences adiponectin serum levels and the eating behavior factors restraint, disinhibition and hunger.

Materials and methods

Subjects

The Sorbs cohort is a self-contained population from eastern Germany which was extensively phenotyped for a wide range of anthropometric and metabolic phenotypes including weight, height, waist-to-hip-ratio (WHR) and a 75-g oral glucose tolerance test (OGTT) and standardized questionnaires for individual medical history and family histories (Veeramah et al. 2011). A total of 1,036 subjects with mean age of 48 ± 16 years and mean BMI 26.9 ± 4.9 kg/m2 were included. A total of 548 Sorbs completed the FEV (Pudel and Westenhöfer 1989), the German version of the TFEQ (Stunkard and Messick 1985) as described elsewhere (Gast et al. 2013). Total adiponectin serum levels were measured in the Sorbs population. All subjects gave written informed consent, and the study was approved by the ethics committee of the University of Leipzig. The main characteristics of the Sorbs are summarized in Table 1.

Table 1 Main characteristics of the study populations
Full size table

The replication cohort is an independent German population described elsewhere (Gast et al. 2013). A total of 350 individuals were included in the analysis (mean age of 27 ± 5 years and mean BMI of 27.0 ± 6.2 kg/m2; Table 1). Phenotyping included anthropometric measurements (BMI, weight, height) and human eating behavior factors measured using the FEV (Pudel and Westenhöfer 1989). The local ethics committee of the University of Leipzig approved the study.

Genetic analysis of the ADIPOQ locus

Eleven ADIPOQ SNPs (single nucleotide polymorphisms) within the ADIPOQ gene and the 5′ UTR (5′ untranslated region) were analyzed. Of these, genotypes for seven SNPs (rs864265; rs182052; rs16861205; rs17366568; rs2241767; rs3821799; rs3774261) were extracted from Affymetrix Genome-Wide Human SNP data (Affymetrix Inc., Santa Clara, CA, USA) earlier described by Tönjes et al. (2009). Four additional tagging variants (rs822396; rs1501229; rs2036373; rs17366743) were selected from the HapMap database (r 2 > 0.8, minor allele frequency (MAF) < 0.05) and individually genotyped (Fig. 1). De novo genotyping of rs822396; rs1501229; rs2036373; and rs17366743 was performed using TaqMan® SNP Genotyping Assay (Applied Biosystems by Life-Technologies Carlsbad, CA, USA). Fluorescence was detected by an ABI 7500 Real-Time PCR system. All SNPs were in Hardy–Weinberg equilibrium (all P > 0.05) except rs822396 in the Sorbs. To avoid genotyping errors, a random selection (~5 %) of the sample was re-genotyped; all genotypes matched the initially designated genotypes. Water was used as a no template control (NTC).

Fig. 1
figure 1

Gene structure of the ADIPOQ gene (not scaled) and its location on Chromosome 3. Filled boxes present coding exons, unfilled indicate non-coding exons. ATG: translation start; asterisk SNPs de novo genotyped in this study. Genotype data for the other SNPs were extracted from previous analysis (Tönjes et al. 2009). Underlined SNPs were significantly associated with adiponectin serum levels in the Sorbs. Circle SNPs remain significant with disinhibition and hunger after meta-analysis

Full size image

Measurement of circulating serum adiponectin

Total adiponectin serum levels in the Sorbs were measured using a high-sensitivity human adiponectin ELISA (BioVendor; Heidelberg; Germany) using antibodies specific for human adiponectin according to manufacturers’ instructions. Serum samples were collected after an overnight fast.

Statistics

Non-normally distributed data were logarithmically transformed to approximate a normal distribution. Linear regression models adjusted for age, gender and BMI (except for BMI) were employed to test for genetic association with BMI, eating behavior factors (restraint, disinhibition and hunger) and serum adiponectin levels. Additive model of inheritance was tested. Individuals with type 2 diabetes (T2D) were excluded from the genetic association analysis for eating behavior factors and adiponectin serum levels. Linear regression analyses adjusted for age, gender and BMI were used to assess the relationship between eating behavior and serum adiponectin levels. To correct for multiple testing, we applied Bonferroni correction as suggested at: http://www.quantitativeskills.com/sisa/calculations/bonfer.htm (alpha niveau 0.05; number of tests/phenotypes = 55) and lowered the significance threshold to P < 9.3 × 10−4. All P values > 9.3 × 10−4 but ≤ 0.05 were considered to be of nominal statistical significance. All P values are provided without the Bonferroni correction. SPSS statistics version 20.0.1 (SPSS, Inc.; Chicago, IL) was used for all statistical analyses. Meta-analyses were performed using METAL (Willer et al. 2010).

Results

Circulating adiponectin levels in the Sorbs

As expected, adiponectin serum levels were negatively correlated with BMI (P = 4.5 × 10−5; r = −0.141, Table 2). A strong positive correlation of serum adiponectin with the eating behavior factor restraint was observed (P = 0.001, r = 0.148), which, however, did not withstand adjustment for covariates such as age, gender and BMI (P = 0.083; r = 0.077).

Table 2 Adiponectin serum levels in T2D and obesity in the Sorbs
Full size table

ADIPOQ SNPs associated with adiponectin serum levels and human eating behavior phenotypes

We observed nominal associations between four intragenic SNPs (rs1501229; rs17366743; rs17366568; and rs3774261) and circulating adiponectin levels in the Sorbs (all P < 0.05, Table 3). The strongest relationship was detected at rs3774261 (Table 3), an intronic variant with minor allele carriers conferring increased serum adiponectin levels (P = 0.006; β = 0.693).

Table 3 Association analysis of ADIPOQ genetic variants with adiponectin levels and eating behavior factors in the Sorbs
Full size table

Of these four variants, two SNPs were also nominally related to disinhibition (rs1501229 and rs3774261, Table 3). Minor allele carriers of these two variants showed both elevated serum adiponectin levels and increased disinhibition scores. In addition, we observed a third SNP (rs2036373) exerting a nominal association with hunger (Table 3). Two further variants, rs822396 and rs864265, were exclusively related to disinhibition. However, none of these associations withstood Bonferroni corrections for multiple testing (all P > 9.3 × 10−4). No relationship between the variants and BMI was observed.

Replication analysis in an independent German cohort

Two variants (rs3774261 and rs1501229) conferring the strongest relationships to both, adiponectin serum levels and the eating behavior factor disinhibition in the Sorbs, were taken forward to replication analyses in an independent German cohort. None of the variants were related to eating behavior factors (Table 4, all P > 0.05). Nonetheless, we observed similar effect directions for rs1501229 as in the Sorbs showing elevated disinhibition scores in minor allele carriers. We did not find similar effect directions between the Sorbs and the replication cohort for rs3774261.

Table 4 Replication analyses in an independent German cohort
Full size table

Meta-analyses

A sample size-weighted meta-analysis for rs1501229 and rs3774261 including the results from the two study populations (Sorbs and German cohort) resulted in nominal statistical significance at rs1501229 with disinhibition and hunger (combined P disinhibition = 0.0369 (Z score 2.086); P hunger = 0.01798 (Z score 2.366) Table 5).

Table 5 Meta-analysis for Sorbs and German replication cohort
Full size table

Discussion

The present study mainly supports the well-known relationship between genetic variants in the ADIPOQ gene locus and adiponectin serum levels. Beyond this, we observed two variants conferring increased adiponectin levels along with increased eating behavior scores which, however, did not withstand correction for multiple testing. Moreover, we found circulating adiponectin correlated with the eating behavior factor restraint in the German Sorbs.

ADIPOQ locus and adiponectin serum levels

In the Sorbs, we found mean circulating adiponectin levels of 16.62 ± 5.10 µg/ml. In line with others describing significant differences between male and female adiponectin serum levels (Heid 2006) which is most likely caused by enriched testosterone levels in men inhibiting the secretion of high molecular weight (HMW) adiponectin from adipocytes (Xu et al. 2005; Wang et al. 2008), we found ~19 % higher serum levels in women compared to men. Many studies described genetic variants in the ADIPOQ locus to be associated with reduced adiponectin levels in T2D, obesity and impaired insulin sensitivity (Vasseur 2002; Comuzzie et al. 2001; Ramya et al. 2013; Peters et al. 2013; Kadowaki 2006). Consistently, in the Sorbs, adiponectin serum levels are negatively correlated with BMI. The ADIPOQ gene locus was shown to be a major locus influencing plasma adiponectin levels with genome-wide significance (Heid 2006; Ling et al. 2009). Several studies identified rs17366568 upstream the transcription start site showing strongest associations to adiponectin serum concentration (Heid 2006; Peters et al. 2013; Cohen et al. 2011; Mather et al. 2012). In line with this, the same variant was nominally associated with circulating adiponectin in the Sorbs (P value = 0.027). Consistently with other studies, we observed three other markers influencing adiponectin serum levels (Ramya et al. 2013).

ADIPOQ locus and eating behavior

Beside its well-described role in endocrine metabolism and cardiovascular function in peripheral tissues as well as the reported autocrine effects on adipocytes (Wu et al. 2003), adiponectin confers also central effects on energy expenditure or food intake (Kubota et al. 2007; Qi et al. 2004; Kadowaki et al. 2008). It was recently shown that adiponectin is also expressed in brain (Rodriguez-Pacheco et al. 2007; Psilopanagioti et al. 2009) and is functionally active (Rodriguez-Pacheco et al. 2007). In addition, several studies reported intracerebral injection (Hoyda et al. 2009a; Iwama et al. 2009; Park et al. 2011) and most importantly expression of adiponectin receptors in the brain (Hoyda et al. 2009b; Dadson et al. 2011). Further, it was shown that peripherally administered adiponectin is able to cross the blood–brain barrier and binds similarly to leptin neuronal targets in the hypothalamus (Thundyil et al. 2012; Qi et al. 2004; Kubota et al. 2007). In the present study, we observed several SNPs in the ADIPOQ gene locus which were nominally associated with human eating behavior factors such as disinhibition and hunger (Pudel and Westenhöfer 1989) as well as with adiponectin levels themselves. Minor allele carrier shows elevated adiponectin serum levels along with increased disinhibition scores which indicate the tendency to frequently overeat. It is of note, however, that none of these nominal associations withstands correction for multiple testing. Moreover, although we observed a positive correlation of restraint with adiponectin serum levels, no significant relationship of eating behavior factors and adiponectin concentrations was observed. Comparing low versus high adiponectin level groups results in significantly higher restraint scores in the high adiponectin group; however, these data do not withstand adjustment for gender (data not shown). Nevertheless, albeit non-significant, our data are in line with the hypothesis that adiponectin activates AMPK-mediated signaling via adiponectin receptor binding in the hypothalamus as demonstrated in animal models (Kubota et al. 2007; Minokoshi et al. 2008). This may result in overeating or increased hunger feelings. Since adiponectin levels can be further influenced by nutritional compounds, increased food intake may in turn serve as a positive feedback process (Mantzoros et al. 2006). However, this mechanism seems to be accompanied by decreased energy expenditure that may ultimately lead together with overeating to increased body weight (Kubota et al. 2007; Minokoshi et al. 2008). In contrast, Qi et al. (2004) reported a temporary loss in body weight after adiponectin injection into lateral-cerebral ventricles without decreased food intake in mice. The authors concluded that higher energy expenditure resulted in the weight loss, which was further supported by increased brown adipose tissue UCP-1 (uncoupling protein-1) mRNA expression. It is still not yet clarified how adiponectin acts in the human brain, because animal studies are not fully comparable with humans and adiponectin in human cerebrospinal fluid is 1000-fold lower than serum levels (Pan et al. 2006; Kos et al. 2006). Further studies are warranted to better understand whether adiponectin acts directly through beneficial effects in peripheral tissues or indirectly by activating AMPK that may lead to altered food intake or energy expenditure.

Our study is limited at several aspects. In particular, in regard to our association results for eating behavior factors, it needs to be acknowledged that the TFEQ from Stunkard and Messick (1985) provides several restrictions. While the questionnaire identifies the three factors restraint, disinhibition and hunger, the subscale disinhibition was most consistently reported to be related with increased BMI and obesity as well as with higher energy intake (Bryant et al. 2008). It was demonstrated (Dykes et al. 2003; French et al. 1994) that disinhibition is strongly related to overeating without hunger feelings in certain situational circumstances which otherwise correlates with a high amount of food intake. Conflicting results, however, were reported for the relationship between restraint and BMI (Dykes et al. 2003; French et al. 1994), while in individuals with high restraint scores both increased and decreased energy intakes were observed (Bellisle et al. 2004; French et al. 1994). Moreover, since the questionnaire does not allow to drawing any conclusions in terms of energy intake, our data can only be interpreted in terms of eating behavior dimensions which are related to energy intake.

Therefore, our data need to be interpreted with caution. Despite the fact that the identified associations did not withstand Bonferroni correction, in concert with the well-known effects of SNPs on adiponectin concentrations, we suggest that genetic variants in ADIPOQ may potentially play a role in eating behavior which may be mediated via influencing the serum adiponectin levels. However, we are well aware that our data would not allow drawing these conclusions without including larger studies necessary to confirm the observed effects. Moreover, the sample size of the independent replication cohort is small, which is most likely one reason for the non-significant association results. Moreover, the small sample size may also lead to false-positive results. Since effects of gender and age on eating behavior are well recognized (Hays and Roberts 2008; Provencher et al. 2003; Jastreboff et al. 2014; Dakanalis et al. 2013), the large difference in age between the two cohorts in concert with the differential gender ratio may be one reason to explain the observed discrepancies in eating behavior scores. In addition to the reported other limitations, differences in eating behavior scores may have prevented us from identifying statistically significant SNP effects. Further, we have no serum adiponectin levels available in the replication cohort.

Taken together, in addition to the known relationship between genetic variation in the ADIPOQ gene locus and adiponectin serum levels, our data suggest a potential correlation with human eating behavior factors. This may indicate potentially regulatory mechanisms in the brain in regard to beneficial effects of adiponectin. Whether the association with eating behavior is mediated by adiponectin levels or vice versa warrants further investigations.