Introduction

Western lifestyle has led to a substantial increase in the frequency of obesity and its associated metabolic complications, including insulin resistance. Obesity and type 2 diabetes are complex disorders that are regulated by behavioural, environmental and genetic factors. The identification of genes that predispose individuals to, or protect against, these disorders may highlight important regulatory pathways and suitable markers or targets for intervention. Although numerous susceptibility loci for obesity and related disorders have been identified, the underlying susceptibility genes remain largely unknown [1].

Insulin resistance is central to the pathophysiology of type 2 diabetes. Improving insulin sensitivity offers a promising approach to the intervention and treatment of this disorder. This implies a role for genes that are physiological regulators of insulin receptor function and insulin sensitivity, such as those involved in insulin signalling through the insulin receptor (IR). An early step in insulin signal transduction is the stimulation of tyrosine kinase (TK) activity [2]. Patients with insulin resistance display impaired IR-TK function [3, 4, 5]. Thus, dysregulation of IR-TK function might be an important component of insulin resistance.

The α2-Heremans–Schmid glycoprotein gene (AHSG) is located in a susceptibility locus for type 2 diabetes and the metabolic syndrome on chromosome 3q27 and encodes a protein that is present both in tissue and the circulation [6]. The function of AHSG is only partially understood. AHSG is a natural inhibitor of insulin-stimulated IR-TK [7, 8]. Mice with targeted disruption of AHSG are resistant to weight gain and demonstrate increased insulin sensitivity and glucose clearance in liver and muscle, as well as lower fasting plasma triglycerides [9]. In addition, high AHSG expression in liver has been reported to be associated with susceptibility to diet-induced obesity in rats [10]. Adipose tissue was not investigated in these studies. It has been established that AHSG inhibits mitogenic signals through IR-TK; however, AHSG did not inhibit glucose transport in a rat preadipocyte cell line [8]. In humans, serum levels of AHSG have been reported to be higher in gestational diabetes than in normal pregnancies, and were correlated with indirect measurements of insulin resistance [11]. Although it has been proposed that alleles of human AHSG with high intrinsic activity could be more common among type 2 diabetic patients, to our knowledge, this has not been investigated [8]. Based on the above observations, we propose that AHSG is a potential novel target for for the treatment and/or prevention of human obesity and insulin resistance.

The aim of this study was to investigate the effects of AHSG polymorphisms on obesity and insulin action in human adipocytes.

Subjects and methods

Subjects

In order to study candidate genes for obesity and insulin action in fat cells, we consecutively recruited 364 unrelated, healthy women who were born in Scandinavia and took no regular medication. There was large inter-individual variation in BMI (18–60 kg/m2) among the women, and obesity was defined as a BMI >30 kg/m2. A random subgroup of the women (n=291) underwent an in vitro metabolic investigation on isolated subcutaneous fat cells.

All women came to the laboratory in the morning after an overnight fast. A venous blood sample was obtained for extraction of DNA and determination of plasma glucose, triglycerides, cholesterol, HDL cholesterol and serum insulin. The homeostasis model assessment for insulin resistance (HOMAIR) was calculated as fasting serum insulin (µU/ml) × fasting plasma glucose (mmol/l)/22.5 [12]. In methodological studies on a representative subgroup of this cohort we compared HOMAIR with glucose uptake during euglycaemic–hyperinsulinaemic clamp. A strong correlation was observed between these measurements (r=0.57, p<0.001) (unpublished results). Body fat content was calculated as (age × 0.13) + (BMI × 1.5)−11.5 [13]. In methodological studies on another dataset we compared percent fat as determined by the formula with that determined by dual-energy X-ray absorptiometry (DEXA). An excellent agreement between the two methods was obtained (r=0.92), with a slope not different from 1.0 and intercept not different from zero. In yet further methodological experiments in 2735 subjects we compared percent body fat as calculated using the formula above with that obtained with bioimpedance using the formula: 3+0.83 × % fat (impedance). The correlation coefficient was 0.92.

The hospital’s committee on ethics approved the study, and written informed consent was obtained from each participant.

Lipolysis and lipogenesis measurements

Abdominal subcutaneous fat biopsies were obtained from 291 women, and lipolysis, lipogenesis and adipocyte cellularity investigated as described previously [14]. Briefly, isolated fat cells were prepared following digestion with collagenase and mean fat cell weight was determined. In lipolysis experiments, diluted cell suspensions were incubated in duplicate for 2 h at 37 °C with air as the gas phase in a medium (pH 7.4) containing albumin (20 g/l), glucose (1 g/l) and ascorbic acid (0.1 g/l). Insulin inhibition of lipolysis was investigated by incubation with adenosine deaminase (1 mU/ml) and the phosphodiesterase-sensitive cyclic AMP analogue 8-bromocyclic AMP (8bcAMP; 10–3 mmol/l) in the presence of increasing concentrations of crystalline human insulin (10–15 to 10–7 mol/l). At the end of the incubation period, an aliquot of the medium was removed for analysis of glycerol release, which was used as an index of lipolysis.

Lipogenesis was investigated by determining the uptake of radiolabelled glucose into lipids. Diluted suspensions of fat cells were incubated in medium (pH 7.4) containing albumin (20 g/l), glucose (5 µmol/l) and 3-3H glucose (5×106 cpm/ml) in the absence or presence of increasing concentrations of crystalline human insulin (10–15 to 10–6 mol/l) for 2 h at 37 °C with air as the gas phase. Incubation was terminated by the addition of sulphuric acid, and the incorporation of radioactivity into lipids determined for each of the samples.

The concentration (log mol/l) of insulin required to produce half maximum effect was determined using logistic conversion of each concentration response curve. This value was converted to its negative form (pD2), which reflects hormone sensitivity. The maximum effect or responsiveness of the hormones was determined as glycerol release or glucose incorporation into lipids at the maximum effective hormone concentration.

Single-nucleotide polymorphism harvesting

There is no published information on polymorphisms in AHSG in the Swedish population. We therefore chose to sequence the seven exons of the AHSG gene, the 3′ untranslated region (UTR), and 1940 bp upstream of the ATG site in 24 obese healthy women (age 47±12 years, BMI 39±5 kg/m2). Coding sequences were obtained from Sequence view for AHSG on contig NT_005962, the 5′ region from AB038689, and the 3′ region from AC068631.16. Primer sequences are provided on request. The aim of sequencing was two-fold: (i) to detect potential functional single-nucleotide polymorphisms (SNPs); and (ii) to obtain a broad overview of the haplotypes in the gene [15, 16]. Only obese subjects were screened so as to increase the likelihood of detecting SNPs predisposing for, as opposed to protecting against, obesity. A study cohort of 24 individuals provides 91% power to detect a rare allele with a frequency of 5% by sequencing, i.e. 1−(0.952)24. Sequencing was performed on an ABI 3100 Genetic Analyzer using the BigDye Terminator v.3 cycle sequencing kit (Applied Biosystems, Foster City, Calif., USA) according to manufacturers instructions. Staden Software was used for sequence analysis [17]. All sequences were scored manually for the presence of SNPs.

Genotyping

The AHSG −469T>G SNP was genotyped by dynamic allele-specific hybridization (DASH) [18]. Briefly, for DASH genotyping, a fragment containing the SNP was amplified using forward (5′-GGAAAGCTAACTAACACTGACA-3′) and reverse (5′-GGAGGATAGTCAGCAACTTGATA-3′) primers. The forward primer was biotinylated at the 5′ end. A genotyping platform (Thermo Hybaid, Ashford, Middlesex, UK) was operated and post-PCR procedures performed according to the manufacturer’s protocol. Alleles were determined using a probe with the sequence 5′-AATTCTCCCATGTCAGG-3′.

Our laboratory subsequently changed genotyping platform, and the other SNPs were genotyped using matrix-assisted laser desorption/ionisation time-of-flight (MALDI-TOF) mass spectrometry (Sequenom, San Diego, Calif., USA) [19]. PCR assays and associated extension reactions were designed using SpectroDESIGNER software (Sequenom), and primers (the sequences of which are available on request) were obtained from Metabion (Planegg-Martinsried, Germany). All amplification reactions were run under the same conditions in a 5-µl reaction containing 2.5 ng of genomic DNA, 1 pmol of each amplification primer, 0.2 mmol/l of each dNTP, 2.5 mmol/l MgCl2 and 0.2 U of HotStarTaq DNA Polymerase (Qiagen, Hilden, Germany). Reactions were heated at 95 °C for 15 min and subjected to 45 cycles of amplification (20 s at 94 °C, 30 s at 60 °C, 30 s at 72 °C) before a final extension at 72 °C for 7 min. Extension reactions were conducted in a total volume of 9 µl using 5 pmol of allele-specific extension primer and the Mass EXTEND Reagents Kit before being cleaned with SpectroCLEANER (Sequenom) using a MULTIMEK 96 automated 96-channel robot (Beckman Coulter, Fullerton, Calif., USA). Clean primer extension products were loaded onto a 384-element chip (SpectroCHIP, Sequenom) with a nanolitre pipetting system (SpectroJet, Sequenom) and analysed by a MassARRAY mass spectrometer (Bruker Daltonik, Bremen, Germany). Peaks were identified on the resulting mass spectra using SpectroTYPER RT 2.0 software (Sequenom). Two independent scorers confirmed all genotypes for each SNP. Phase software was used to determine haplotypes [20].

Statistical analysis

Regression analysis was used to determine the additive effects of alleles on quantitative phenotypes. Similar p values were obtained with ANOVA and ANCOVA including BMI and age as covariates. The chi square test was used to determine whether allele frequencies differed between affected and non-affected individuals. Data are shown as means ± SD. The p values shown are nominal, i.e. they are not corrected for multiple comparisons.

Results

Screening AHSG for polymorphisms

Sequencing the seven exons of AHSG, the 5′ region of the gene, and the 3′ UTR in 24 obese women identified 20 polymorphisms (Table 1). Twelve of these SNPs were previously published in the dbSNP database (http://www.ncbi.nlm.nih.gov/SNP/index.html). Six SNPs (located upstream of the ATG site, in exons, and in the 3′ UTR) were chosen for genotyping. The remaining SNPs were excluded from genotyping for various reasons. Five SNPs had rare allele frequency (<5%), and the study had low power to detect disease association with alleles with such low frequency. Seven SNPs displayed perfect linkage disequilibrium with at least one genotyped SNP according to SNPHAP and attached linkage disequilibirum software (http://archimedes.well.ox.ac.uk/cgi-bin/pise/snphap.pl). Two SNPs were intronic and therefore less likely to be functional.

Table 1 Single-nucleotide polymorphisms in AHSG
Full size table

Relationship between AHSG single-nucleotide polymorphisms and phenotypes

In total, 188 non-obese and 176 obese (BMI >30 kg/m2) age-matched women were genotyped for the six chosen AHSG SNPs. The phenotypes of these women are displayed in Table 2, and their genotypes are displayed in Table 3. The AHSG 39C>G SNP was successfully genotyped in 81% of subjects. The success rate was >98% for all other SNPs. We chose not to repeat 39C>G genotyping since it was a rare SNP, and 39C>G and 7939C>T allelic distribution was in strong linkage disequilibrium in the common haplotypes (Table 4). The 39C>G SNP did not therefore provide much additional information. All SNPs were in Hardy–Weinberg equilibrium (data not shown). Our sample size provides 80% power to detect a susceptibility allele with a rare allele frequency of 35% among controls, approximately what was observed for three of our SNPs (odds ratio 1.5, threshold p value 0.05). For an SNP with a rare allele frequency of 5%, the power is reduced to 28%. Haplotype analysis identified four common haplotypes (Table 4).

Table 2 Phenotypes of study subjects
Full size table
Table 3 Genotyping results
Full size table
Table 4 AHSG haplotypes
Full size table

There were no significant differences between obese cases and non-obese control subjects with respect to allele or haplotype frequencies (data not shown). None of the investigated SNPs were associated with BMI, per cent body fat, waist circumference, fat cell weight, HOMAIR, plasma levels of glucose or triglycerides, or serum insulin (data not shown). Bioimpedance measurements of body fat were available for 272 subjects; SNPs were not associated with these values. Three SNPs were associated with levels of plasma cholesterol (Table 5, Fig. 1). The AHSG −469T>G genotype was associated with responsiveness to insulin-mediated inhibition of lipolysis (p=0.004), responsiveness to 8bcAMP-stimulated lipolysis (p=0.006), and pD2 for insulin stimulation of lipogenesis (p=0.05), and was marginally associated with basal lipolysis (p=0.06). Similar p values were obtained with ANOVA and ANCOVA including BMI and age as covariates. The G allele of −469T>G is specific for AHSG haplotype 1. Unfortunately, the number of chromosomes carrying haplotype 5, which is identical to haplotype 1 at all loci except −469T>G, is too small to allow determination of whether the −469T>G SNP or haplotype 1 is primarily associated with adipocyte phenotypes. No additional associations were observed on analysis of the haplotypes.

Table 5 Associations between single-nucleotide polymorphisms in AHSG and plasma cholesterol
Full size table
Fig. 1
figure 1

Associations between −469T>G genotype (black bars: TT, n=91; hatched bars: GT, n=134; white bars: GG, n=63) and lipolysis (a) and lipogenesis (b) phenotypes. See subjects and methods for details of phenotype quantification. The value pD2 is the concentration (log mol/l) of insulin that produces half maximum effect, converted to its negative form Data are shown as means ± SD. The p values were obtained by regression of quantitative phenotypes on additive allele effects. * p=0.06; ** p=0.05, *** p<0.01

Full size image

Discussion

In this study we have demonstrated associations between an AHSG polymorphism and lipid metabolism in fat cells. The −469T>G SNP was strongly associated with insulin inhibition of lipolysis and 8bcAMP-stimulated lipolysis in adipocytes, and was more weakly associated with insulin stimulation of lipogenesis and basal lipolysis.

Insulin resistance and type 2 diabetes are complex phenotypes, with disturbed glucose and lipid homeostasis in adipose tissue, liver and muscle, and changes in insulin production in the pancreas. The identification of individual underlying susceptibility or protective alleles for these complex phenotypes has proved difficult. Attempts have been made to increase power by analysing several loci simultaneously for phenotypic association, analysing a subgroup that display a more homogenous phenotype, or analysing intermediate phenotypes that may display a stronger association with an individual locus. The latter has proven particularly successful in animal models, probably due to access to abundant tissue and the ability to control for exogenous impact on phenotypes. For practical and economic reasons we used rather simple measurements of body fat (calculation) and insulin sensitivity (HOMAIR), since methodological studies had shown that these measures provided valid information.

Our results indicate that AHSG variants have a local impact in adipose tissue on insulin regulation of lipogenesis and lipolysis. In particular, there appears to be a strong association between variants and anti-lipolytic pathways. Whether AHSG polymorphisms also regulate glucose homeostasis is unclear; we did not observe an association with the indirect measurement of insulin sensitivity on glucose metabolism, HOMAIR. This does not exclude an association with direct measurements such as glucose uptake during euglycaemic–hyperinsulinaemic clamp. However, it is difficult to obtain in a large enough dataset on this phenotype to make genetic analysis meaningful. In addition, separate methodological experiments in our laboratory demonstrated that measurements of HOMAIR and glucose uptake during euglycaemic–hyperinsulinaemic clamp were strongly correlated (r=0.57). Our results are in agreement with a threshold model of susceptibility for type 2 diabetes. In this model, individual genetic loci regulate intermediate molecular phenotypes, and when the individual’s set of susceptibility alleles at such loci exceeds a threshold, clinical disease may occur. Insulin action on adipocytes may be a phenotype that is sensitive to variation in AHSG.

We observed a weak association between AHSG allele and plasma cholesterol. However, given that AHSG has no obvious primary role in plasma lipid regulation, we interpret this result with caution.

It would be appropriate to confirm our results using an independent dataset. However, to our knowledge, there is no other large human dataset on adipocyte phenotypes. Instead, additional molecular functional studies on the role of the investigated AHSG polymorphism on insulin signalling in adipocyte are needed. In mice, AHSG has been shown to inhibit insulin signalling in muscle; however, its effect on adipose tissue has not been investigated. The association between AHSG and 8bcAMP-stimulated lipolysis indicates that the gene may also control insulin-independent signalling pathways in lipolysis regulation.

In summary, AHSG is not a strong candidate for obesity and type 2 diabetes, but may control insulin signalling in adipose tissue. AHSG is an attractive candidate gene for disturbed adipocyte lipolytic function in obesity and insulin resistance.