Florin Grigorescu M.D., Ph.D. Molecular Endocrinology Laboratory, UMR204 NUTRIPASS; Group Nutrition & Genomes, Institut LTniversitaire de Recherche Clinique (RJRC); 641, avenue du Doyen Gaston Giraud, 34093 Montpellier France, tel. : +33 41175 98 95, fax: +33 4 67 54 27 31, e-mail: florin.grigorescu@inserm.fr

This work was supported by: Recurrent budget from University of Montpellier (France); Recurrent budget from University Medicine and Pharmacy „Carol Davila” in Bucharest (Romania). This work was also sustained by the grant MEDIGENE FP7-279171 of European Community (EC).

Introduction

Polycystic ovary syndrome (PCOS) is a complex disease that affects 5 to 10% of women in reproductive age [1, 2]. PCOS is characterised by hyperandrogenism and chronic anovulation [2, 3], but also by a high frequency of insulin resistance, impaired glucose tolerance (IGT), insulin secretion abnormalities, and type 2 diabetes (T2D) [2, 4, 5].

The insulin gene minisatellite (VNTR, variable number of tandem repeats) was associated with PCOS [6, 7], although this has not been confirmed in other studies [8, 9]. VNTR of insulin gene was also associated with type 1 diabetes (T1D) [10, 11], T2D [12], childhood obesity [13], insulin sensitivity, insulin secretion in children [14], and insulin levels in plasma [15].

Insulin gene VNTR is localised 596 bp upstream from the transcription initiation site. It regulates the expression of both insulin and IGF-2 (insulin-like growth factor-2) genes [16, 17]. Insulin gene VNTR is formed by a repetition of a consensus sequence of 14 to 15 bp. It involves two principle classes in the non-African populations: a short and a long one called class I and class III, respectively [10, 18]. VNTR of class I was associated with a higher risk of T1D [10] while VNTR of class III was associated with an increased risk of PCOS [6] and T2D [12]. The SNP (single nucleotide polymorphism) rs689 located upstream from exon 2 of the insulin gene was strongly associated with T1D [11]. This polymorphism has two alleles: A and T, in strong linkage disequilibrium (TD) with VNTR of class I and III, respectively, allowing for the use of this SNP as a surrogate of the minisatellite in association studies. However, it was reported that this polymorphism has a role in the regulation of insulin gene expression through an alternative splicing process [19], although the impact of the phenomenon is minor [20]. On the other hand, a high TD was observed between haplotypes of some SNPs in the insulin gene region and subclasses of the insulin gene VNTR in non-African populations [21]. Note that the subclasses IC and ID (VNTR of class I) were differently correlated with fasting insulin in childhood obesity in the French population [22], while the two other haplotypes in TD with the two subclasses IIIA and IIIB of the minisatellite were differently associated with the protective effect against T1D [23].

The role of the insulin gene and its minisatellite in PCOS and in the metabolic pattern of the disease remains controversial. This controversy was accentuated after the study carried out by Bouatia-Naji et al., who showed no familial association of insulin gene VNTR with childhood obesity in 1023 families or with fasting insulin levels or birth weight in obese children [24], Moreover, genome-wide association studies (GWAS) screening hundred of thousands of SNPs associated the locus of insulin gene VNTR with prostate cancer [25], through the SNP rs7127900, but failed, to our knowledge, to find any association with metabolic disorders, which updates the question concerning loci found associated with complex disorders through the candidate gene approach and missed by GWAS studies.

In this context, the aim of the present study was to investigate the association of insulin gene VNTR with PCOS in a population of women from Central Europe based on haplotyping strategy. Our final aim was to compare the power of haplotypes with that of SNPs taken alone and to get a preliminary idea about the best strategy for studying the genetics of complex disorders.

Material and methods

Population recruitment

The study was approved by the Ethical Committee of The University of Bucharest (Romania), according to Helsinki Declaration (revised in 1983) and after informed consent. Participants in the investigation were not related and were of Caucasian origin (Region of Walachia, Romania). Type 1 genetic design [26] comparing PCOS with control lean healthy women was applied. Note that no patient in this study had diabetes.

Control population

The control population was composed of 105 lean healthy women recruited in the Obstetrics and Gynaecology Department of the University of Bucharest. These women had a normal reproductive status, regular menses, proven fertility (at least one child), and no clinical or biochemical hyperandrogenism.

PCOS population

A total of 377 women with PCOS were included. Diagnosis of PCOS was established according to Rotterdam criteria [27], Individuals with Cushing Syndrome, non-classical deficiency of adrenal 21-hydroxylase, hyperprolactinaemia, and androgens secreting neoplasms were excluded from the study. PCOS women were stratified as lean (BMI < 30 kg/m2) and obese (BMI > 30 kg/m2). PCOS with more insulin resistance and Acanthosis nigricans (HA-IR-AN syndrome) was distinguished from classical PCOS (Y-PCOS), as previously reported [28].

Clinical and biochemical characteristics

The women were given a standard diet. Then, oral glucose tolerance test (OGTT) was applied to the group of PCOS women. IGT and impaired fasting glucose (IFG) were defined according to American Diabetes Association (ADA) criteria [29], Insulin resistance was assessed by HOMAIR (homeostatic model assessment insulin resistance) index, and insulin secretion by HOMA-β (homeostatic model assessment β-cell function). Note that HOMAIR (mmol/L) = (fasting glucose X fasting insulin)/22.5 while HOMA-β (mmol/L) = (fasting insulin X 20)/(fasting glucose – 3.5).

Measurement of lipid parameters and godanotropins (FSH and LH) in the plasma was carried out in the Department of Endocrinology in Bucharest. Clinical and biochemical characteristics of the population are shown in Table I.

Note that the first portion of the population (160 cases and 90 controls) was used for genotyping seven genetic markers (SNPs). The aim was to determine the LD pattern between these SNPs and select tag-SNPs to be genotyped in the rest of the population.

Genetic exploration

Description of the explored genomic region

The insulin gene (MIM: 176730) is located in the locus 11p15.5, downstream from its minisatellite (VNTR) by about 400 bp. The gene extends on 1431 bp and contains three exons. The region involving SNPs considered in the present study starts from the end of the first intron and extends beyond the 3’UTR side of the gene.

Genotyping of SNPs

All SNPs were genotyped in a portion of the population (160 women with PCOS and 95 controls) by classical sequencing or SSO-PCR. An overview of these techniques was previously reported [30]. Primers and PCR steps used for each SNP are reported in Table IIA. In the rest of the population (377 women with PCOS and 105 controls), only SNPs rs3842754, rs3842748, and rs689 were genotyped by KASPar technology [31], using primers and following the amplification steps reported in Table IIb.

Data and statistical analysis

The LD pattern of the explored genomic region was determined by Haploview 3.31 [32], Haplotype reconstruction was carried out by PHASE 2.1 [33] and cladistics analysis by Arlequin 2.000 [34]. Concerning clinical and biochemical characteristics of the population, continuous variables (expressed as median [25% percentile – 75% percentile]) were analysed using non-parametrical tests. Significance when comparing between continuous variables in phenotypic groups was calculated by Mann-Whitney non-parametrical test (significance was attested for P < 0.05), while Fisher’s exact test was used for comparing between percentages (significance was attested for P(X2) < 0.05). The distribution of continuous parameters in the population was checked by Kolmogorov-Smirnov test. A non-normal distribution was attested for P < 0.05. Variation of SNPs was expressed as allelic frequency (q) or prevalence (%) of their genotypes, while variation of haplotypes was expressed as their prevalence or the prevalence of pairs of haplotypes. The genetic association was determined by logistic regression using StatView 5.0 and SAS (Abascus Concepts, Berkeley, CA), as previously described [35], Results were expressed as nominal P, odds ratio (OR), and 95% confidence interval (Cl). Genotype-phenotype correlations were assessed by Mann-Whitney test. The power of the case-control study was calculated by Sampsize software (http://sampsize.sourceforge.net/iface/s3.html).

Results

Phenotypic data

The explored population involved 482 women (377 with PCOS and 105 controls). Among PCOS women, 148 were obese and 71 were with severe insulin resistance (hyperandrogenism-insulin resistance-Acanthosis nigricans „HA-IR-AN” syndrome) (Table 1). Women with PCOS had higher BMI (P < 0.0001), higher LH/FSH (P < 0.0001), higher testosteronaemia (P < 0.0001), higher fasting insulin (P < 0.0001), and higher insulin secretion (HOMAβ) (P < 0.0001). On the other hand, Acanthosis nigricans was more often observed in PCOS obese women than in PCOS lean (P < 0.0001). Moreover, PCOS obese women had higher fasting insulin than lean PCOS (P < 0.0001), were more insulin resistant (P < 0.0001), and had higher insulin secretion (HOMAß: P < 0.0001). HA-IR-AN women had higher BMI than YPCOS (P < 0.0001) and were more insulin resistant (P < 0.0001) and secretory (P < 0.0095). Note that no patient in this study had diabetes.

Note that the portion of the population where the seven SNPs were genotyped (160 women with PCOS and 95 controls) displayed the following characteristics: the median age of the PCOS group (25 and 75 percentiles) was 24.0 years (21.0–28.0) while it was 32.5 years (27.0–37.0) (P < 0.0001) for controls. PCOS women had higher BMI: 30.00 kg/m2 (24.12–34.4) vs. 22.00 (20.00–23.85) (P < 0.0001) and were more insulin resistant: HOMAIR: 3.31 (1.86–5.58) ns. 1.25 (0.90–1.68) (P < 0.0001). The test for the distribution of continuous parameters showed non-normal distribution of many parameters such as BMI (P < 0.0001), HOMA-β index, and plasmatic testosterone, while dehydroepiandrosterone sulphate (DHEAS) plasmatic levels displayed normal distribution (P).

Genotypic data

Polymorphism analysis in part of the population (160 women with PCOS and 95 controls) showed that SNPs rs3842756, rs3842755, rs3842754, rs3842753, rs3842752, rs3842748, and rs689 were in high LD according to the D’ index (Fig. 1). However, three groups of SNPs were observed according to the level of high LD between them (according to r2 LD index): group 1 (rs3842756, rs3842755, rs3842754, rs3842752), group 2 (rs3842753 and rs689), and group 3 (rs3842748). Thus, we considered SNPs rs3842754, rs689, and rs3842748 to be representative for the first, the second, and the third group, respectively. These SNPs were genotyped in the rest of the population. Frequencies of SNPs rs3842754, rs3842748, and rs689 in tire control group were 0.17, 0.83, and 0.75, respectively, which are very close to their frequencies in other Caucasian populations (http://www.ncbi.nhn.nih.gov/snp/). Reconstruction of haplotypes was done using the three SNPs allowed observing six haplotypes: hi (CCA), h2 (CCT), h3 (CGA), h4 (CGT), h5 (TCA), and h6 (TGT) (Table III). It appeared that the three haplotypes hi, h2, and h3 were frequent while the three others were rare. When we started from the root of the cladogram (h4), the mutation of rs3842754 from C to T allowed us to observe h6 (surrogate of VNTR of subclass IIIA), while mutation of rs3842748 from G to C allowed us to reach the rest of the cladogram (Fig. 2). Mutation of SNP rs689 from T to A led to hi (VNTR of class I). This data is in accordance with literature [21].

Figure 1. Linkage disequilibrium (LD) pattern (estimated by D’ index) in the region of insulin gene between rs3842756 and rs689, in the female population (cases and controls) from Central Europe. The numbers in red squares indicate the D’ index (%). The squares without numbers indicate a perfect LD (D’ = 100%> or 1) between the corresponding SNPs
Rycina 1. Wzór niezrównoważenia sprzężeń (LD) (wyliczony przez indeks D’) w regionie genu insuliny między rs3842756 i rs689 u populacji kobiet (przypadki chorobowe i osoby kontrolne) z Europy Centralnej. Liczby w czernionych kwadratach oznaczają indeks D’ (%). Kwadraty bez liczb oznaczają idealne LD(U = 100% lub 1) między odpowiednimi SNP

Figure 2. Phylogenetic (cladistics) affiliation of SNPs rs3842754, rs3842748, and rs689 illustrating the historical evolution of haplotype diversity in the female population (cases and controls) from Central Europe used in the study. The ancestral (root) haplotype (CGT) is h4. The haplotype hb (TGT) is due to the mutation (C → T) of the SNP rs3842754. The haplotype h2 (CCT) derived from h4 after the mutation (G → C) of the SNP rs3842748, while the haplotype h1 (CCA) is due to the mutation of both rs689 (T → A) and rs3842748 (G → C)
Rycina 2. Związek filogenetyczny (kladystyka) SNP: rs3842754, rs384274S i rs689 ilustrujący ewolucję różnorodności haplotypów w populacji kobiet (przypadki chorobowe i osoby kontrolne) z Europy Centralnej, wykorzystane w badaniu. Haplotyp ancestralny (CGT) to h4. Haplotyp hb (TGT) powstał w wyniku mutacji (C → T) SNP rs3842754. Haplotyp h2 (CCT) uzyskano z h4 po mutacji (G → C) SNP rs3842748, podczas gdy haplotyp h1 (CCA) powstał w wyniku mutacji zarówno rsb89 (T → A), jak i rs3842748 (G → C)

We observed that the genotype GC of rs3842748 was associated with the obese phenotype in PCOS (P = 0.049) and in the classical PCOS (YPCOS) group (P = 0.010) (Table IV). The association with obesity in YPCOS remained after adjustment for age. The pair of haplotypes hl/h6 were associated with the obese phenotype in classical PCOS (P = 0.009) and remained also after adjustment for age. The data of association with obesity was supported by the correlation between the pair of haplotypes I1I/I16 and higher BMI in the YPCOS group (hl/l-16: P = 0.048,28.5 ± 0.9 vs. 26.3 ± 0.4) and a trend for correlation between the GC genotype of rs3842748 and higher levels of BMI in the YPCOS group (P = 0.077).

SNPs and their haplotypes were also correlated with plasmatic levels of androgens in PCOS group. Indeed, we observed the correlation of rs689, through its AA genotype and the pair of haplotypes h1/h1, with higher levels of testosteronaemia, but this was at the limit of significance (rs689 (AA): P = 0.054,87.1 ± 4.2 ng/dL vs. 74.9 ± 2.0 ng/dL; haplotype h i/h i : P = 0.054,86.9 ± 4.1 ng/dL vs. 74.9 ± 2.0 ng/dL). Moreover, rs3842754 and rs3842748, through their alleles C and C, respectively, were correlated with lower plasmatic levels of DHEAS in PCOS women with obesity (rs3842754 (C): P = 0.041, 260.7 ± 16.9 µug/dL ps. 432.9 ± 61.6; rs3842748 (C): P = 0.041, 260.7 ± 16.9 ps. 432.9 ± 61.6) (Fig. 3). Note that the power of the study concerning each positive genetic association is reported in Table IV.

Figure 3. Graphic comparing dehydroepiandrosterone sulpluite (DHEAS) levels between individuals carrying the allele C of the SNP rs3842748 and the others. DHEAS levels are indicated by mean ± standard error The significance of the difference is determined by the Mann-Whitney test. P is considered significant when < 0.05
Rycina 3. Graficzne porównanie stężenia siarczanu dehydroepiandrosteronu (DHEAS) pomiędzy nosicielami allela C SNP rs3842748 i innymi. Stężenie DHEAS oznaczono jako średnią ± błąd standardowy. Istotność różnicy statystycznej ustalana jest na podstawie testu Manna-Whitneya. P uznano za istotne dla wartości P < 0,05

Discussion and conclusions

PCOS is a complex disorder characterised by abnormalities related to reproduction and others with metabolism and glucose homeostasis, including insulin secretion abnormalities. Indeed, women with PCOS were reported as having a decrease in insulin secretion [5], Insulin gene minisatellite (VNTR), which was implicated in the regulation of insulin gene expression, is a potential candidate in PCOS. Different investigations were carried out and gave controversial results [6-9], Some of these studies focused on VNTR itself while others used rs689 as a surrogate since this SNP has two alleles in high LD with VNTR of Class I and III, respectively, in non-African populations. Moreover, studies used different definitions for diagnosing PCOS. One could speculate llial the heterogeneity between investigations is the cause of controversy concerning the association of insulin gene VNTR with PCOS.

In order to get more insight we investigated the association of insulin gene VNTR with PCOS using dense mapping of the genome and reconstruction of haplotypes. The Rotterdam Conference criteria [27] were used for diagnosing PCOS, giving an accurate and precise definition of the disease. All this accuracy was aimed to give an ideal context for determining the relation between the minisatellite and PCOS.

The SNPs taken into account in this study have been demonstrated to combine into haplotypes in high TD with classes and subclasses of VNTR, which allowed them to be used as surrogates. Their genotyping was carried-out in a portion of the population, with the aim of determining the TD pattern and selecting from them those that tag the region (tagSNP = representative of a group of SNPs in high TD). Three tagSNPs were then selected and genotyped in the rest of the population, including rs689. In addition to its ability to mimic the variation of the insulin gene minisatellite, this SNP was implicated in the regulation of insulin gene expression through an alternative splicing process [19, 20].

We observed the correlation of the genotype AA of rs689 and the pair of haplotypes h1/h1 with higher levels of testosteronaemia in PCOS and with lower plasmatic levels of DHEAS in PCOS with obesity, which highlights once again the close relation between the reproductive and the metabolic components in PCOS, as already reported many times in literature (Diamanti-Kandarakis. 2008).

Moreover, the correlation between the gene and lower plasmatic levels of DHEAS in our study is interesting and in accordance with literature. Indeed, previous studies reported negative correlation between DHEAS levels in plasma and HOMAIR index (reflects fasting insulin) in PCOS [37], as well as high levels of DHEAS being correlated with lower insulin levels and more favourable metabolic and cardiovascular parameters [38]. Our finding may give more insight into understanding the genetic background that controls variations of DHEAS.

Moreover, the genotypes GC of rs3842748 and h1/h6 (corresponding to the genotype I/III of VNTR) were associated with obesity and correlated with higher BMI in PCOS (especially PCOS with lower insulin resistance), which represents a novelty in our study. Note that this genotype has been previously associated in T1D with glutamic add decarboxylase antibodies (GADA) [39] but never with T2D, PCOS, or obesity. At the same time, the minisatellite of class III, widely associated in literature, with PCOS and T2D [6, 12] displayed weak results in our investigation. Perhaps the situation is more complicated than expected. Indeed, locus involving insulin gene VNTR involves IGF-2 (MIM: 147470) and tyrosine hydroxylase (MIM: 191290) genes. These two genes were assodated, as well as the minisatellite, with metabolic parameters [40]. Combining the variation of these three genes may ameliorate the genetic study and determine subgroups of VNTR of class III other than IIIA and IIIB (determined by our haplotypes). One could speculate that the heterozygote genotype I/III of VNTR in our population encompasses a subgroup of VNTR class III related to obesity. The assodation of VNTR pill with obesity could be due also to VNTR of class I. Here also the context seems to be complex. Indeed, Le Stunff et al. observed an excess of paternal transmission of VNTR of class I to obese children [13] and other studies attested to the role of genetic imprinting occurring in tire INS-IGF2 locus at the age of onset of monogenic diabetes caused by INS gene mutations [41]. Future studies concerning this locus should take into account this phenomenon.

Criticism could be addressed concerning the size of the explored population, especially in the context of current genetic association studies exploiting populations of thousands of individuals, with the aim of getting statistical power. Such studies may display weakness at the phenotypic level in the sense that large recruitment may lead to less stringency, contaminating the phenotypic pattern with inappropriate clinical backgrounds, which may lead to spurious associations. The power of our case-control study was acceptable (72.85, 73.95, 51.19% for the three positive associations obtained, respectively), and frequencies of SNPs in our control population were close to those in Caucasian populations [21], which attests to the reliability of the study.

The present investigation was able to detect association of insulin gene VNTR with metabolic components in PCOS, but it was not able to highlight any consistent advantage for using haplotypes instead of SNPs taken alone. This needs to be deepened by extending the explored region upstream and downstream through the screening of much more SNPs. This may allow us to observe that haplotypes carry more genetic information than SNPs taken alone. It is expected to carry out future studies in the context of the European Project MEDIGENE FP7 (FP7-279171-1) with deeper screening of the 50 kb locus involving VNTR of insulin gene, IGF-2, and TH genes for a better understanding of the relation between the locus and complex metabolic disorders. Indeed, we believe that genes reported as not associated with metabolic complex disorders by GWAS studies (as insulin gene VNTR) even though they were previously associated using the candidate gene approach is due to a lack of genome characterisation by GWAS. Using hundred of thousands of SNPs (six KSNPs to about a million SNPs) for screening the whole genome is not enough, knowing the existence of at least 27 million SNPs and rare variants with potential effects. One could also criticise our study since we focused on the insulin secretion component without taking into account potential genes such as TCF7L2 (MIM: 602228). It is expected in future that genotyping, in our population, of other potential genes with a role in insulin secretion, notably TCF7L2 and study the interaction of the insulin gene locus with this panel of genes will take place.

Acknowledgments

This work was supported by the recurrent budget from the University of Montpellier (France), and University Medicine and Pharmacy „Carol Davila” in Bucharest (Romania). RA, FG, and MC are sustained by the grant MEDIGENE FP7-279171 of the European community (EC). Special thanks are also due to Dr. Mouna Flouari-Barat and Mr. Bruno Dumont for their technical help and also to Prof. Jean-Pierre Daures, Dr. Mohamed Belkacemi and Charles for their advice in statistics.