PPAR-γ2 is a nuclear receptor transcription factor, and its codifying gene, on chromosome 3p25, is intensively expressed in adipose tissue(Reference Desvergne and Wahli1). PPAR-γ2 regulates lipid metabolism, adipocyte differentiation, proliferation and insulin sensitivity through regulation of the expression of adipocyte-specific developmental genes(Reference Hegele, Cao and Harris2).
Additionally, there is a missense C-to-G change in codon 12 encoding alanine in substitution for proline in the polypeptide sequence. This polymorphism is relatively common, occurring in 20 % of the Caucasian population(Reference Ek, Andersen and Urhammer3–Reference Radha, Vimaleswaram and Babu5), and seems to be responsible for reduced activity of PPAR-γ2(Reference Groop6).
A few studies have identified an association between the Ala allele and improvement in insulin resistance(Reference Ek, Andersen and Urhammer3, Reference Deeb, Fajas and Nemoto7–Reference Douglas, Erdos and Watanabe9). Because the polymorphism is very close to the amino-terminal-activated independent binding domain, its activity is increased by insulin through phosphorylation. It seems that, while proline prevents the α-helix, alanine favours it, and this amino acid change can exert a profound effect on the structure and, consequently, the function of this protein(Reference Mancini, Vaccaro and Sabatino10).
Given the controversy generated by the studies that report an association between the most frequent polymorphism in the PPAR-γ2 gene and insulin resistance and diabetes, and also considering that insulin resistance is accountable for promoting metabolic alterations that increase cardiovascular risk in subjects, the present study aimed to identify systematically the association of the Pro12Ala polymorphism in the PPAR-γ2 gene with type 2 diabetes mellitus (T2DM), insulinaemia and insulin resistance in Caucasians and non-Caucasians and based on overweight status.
Methods
Research design
The present investigation was carried out within a limited period of 10 years, from 1998 to June 2008, using the keywords: (‘1998’[publication date]:‘2008/06’[publication date]) and ((PPARγ2 polymorphism or peroxisome proliferator-activated receptor) and (‘insulin resistance’ or ‘insulin sensitivity’ or ‘diabetes’ or ‘impaired glucose tolerance’)) and refined by ‘humans’, ‘clinical trial’, ‘editorial’, ‘letter’, ‘meta-analysis’, ‘randomized controlled trial’, adult: 19–44 years, middle aged: 45–64 years, middle aged + aged: 45+ years, aged: 65+ years, 80 and over: 80+ years. The search was performed initially on the PubMed website, which resulted in 152 studies. Other databases were accessed to obtain the full-text articles: ISI Web of Knowledge, MEDLINE, SCIELO, High Wire Press and Science Direct. The reference lists of the original research articles and review articles were used to complement the database search by including additional publications that would not show up in the PubMed search.
Selection of studies was performed independently by two interviewers according to the following exclusion criteria: (1) articles written in languages other than English, Spanish and Portuguese; (2) review articles; (3) missing genotype-specific case numbers or measurement data of fasting insulin or homeostasis model assessment of insulin resistance (HOMA-IR) or diabetes; (4) modified results of fasting insulin or HOMA-IR by log or geometric means; (5) missing measurement of deviation; (6) calculation of HOMA-IR by formulas other than the original represented by Matthews et al. (Reference Matthews, Hosker and Rudenski11); and/or (7) a genotype distribution not in Hardy–Weinberg equilibrium. Thus, four studies were excluded on the basis of the continuous variables, one due to missing deviation data(Reference Kao, Coresh and Shuldiner12), and another as it transformed data into logarithms to normalise their distribution(Reference Hegele, Cao and Harris2); the final excluded studies(Reference Rooij, Painter and Philips13, Reference Stefanski, Majkowska and Ciechanowicz14) calculated insulin resistance by formulas other than HOMA-IR(Reference Matthews, Hosker and Rudenski11).
Forty-one studies were selected for analysis, and all relevant data were extracted individually from each study, including first author, year of publication, country, ethnicity, total number of each genotype of Pro12Ala polymorphism, number of cases representing T2DM and impaired glucose tolerance and controls by normal glucose tolerance and mean and standard deviation for age, BMI, fasting insulin and HOMA-IR.
Data analysis
All analyses were performed on the genotypes Pro12Pro (Pro/Pro) and the sum of Pro12Ala with Ala12Ala (X/Ala). To calculate the mean and standard deviation of X/Ala in some studies presenting separate data, the sum of variances within and between genotypes was used. The same formula was used to group impaired glucose tolerance and diabetes.
When necessary, data were transformed from standard error into standard deviation using the specific formula sd = se × √n. In addition, serum insulin values presented in μIU/ml were converted into pmol/l by the conversion factor 6·945(15).
Data were analysed by RevMan (version 5.0; The Cochrane Collaboration, Copenhagen, Denmark)(16). The OR of the Ala allele and T2DM association was calculated by a Mantel–Haenszel test(Reference Coutinho, Medronho, de Carvalho and Bloch17). The inverse of variance with standardised mean difference was used to estimate the association of the Ala allele with serum insulin and HOMA-IR. This measure represents the standardised size effect of polymorphic genotypes (X/Ala) in relation to the wild genotype (Pro12Pro) in fasting serum insulin and HOMA-IR levels. In the course of the analysis, the studies were separated into subgroups to calculate the summary measure in each subgroup and the overall final measure. The subgroups took into account ethnicity, separating Caucasians from non-Caucasians; the subgroups were also divided according to overweight status, separating the studies that showed an average population of normal BMI ( < 25 kg/m2) and increased mean BMI (> 25 kg/m2), classifying them as normal weight and overweight, respectively. The lack of data about the ethnicity of the studies' populations led us to consider as Caucasian those who have ancestry and were born in Europe, or the Middle East, or North Africa, or parts of Central Asia, who share certain genetic and physiological characteristics, beyond white skin.
Statistical analysis of serum insulin in the combination of studies in a population group with normal glucose tolerance and another group with impaired glucose tolerance and T2DM was performed in order to verify that the Ala allele influences the concentration of insulin under different conditions of glucose tolerance.
To assess the statistically significant heterogeneity between studies, a χ2 test with n-1 df was used, where ‘n’ is number of studies. In the case of significant heterogeneity in the global analysis, a random-effects model was calculated; otherwise, a fixed-effects model was calculated. Inconsistency (I 2) was calculated to verify how much of the difference between studies was caused by heterogeneity, with values lower than 25 % considered low, 50 % considered moderate, and values greater than 75 % considered high inconsistency(Reference Higgins, Thompson and Deeks18). A Z test was used to analyse the global effect and the CI. Significance was assumed at P < 0·05.
The outcomes on the left axis that cross the scale (1 or 0) indicate that the corresponding amount is smaller in the X/Ala genotype than the Pro/Pro genotype.
Results
Forty-one eligible studies from the past 10 years were included in the meta-analysis, all of them in Hardy–Weinberg equilibrium. Table 1 describes the main features of each study group while Figs. 1–3 describe the OR and 95 % CI of each group (see below). Citations denoted a, b, c or d represent the same study with different populations.
Table 1 Characteristics of the included studies*
STOP-NIDDM, Study to Prevent Non Insulin Dependent Diabetes Mellitus; T2DM, type 2 diabetes mellitus; IGT, impaired glucose tolerance; NGT, normal glucose tolerance; JHU-WMC, Johns Hopkins University Weight Management Center; BLSA, Baltimore Longitudinal Study on Age; FUSION, Finland-United States Investigation of Non-Dependent Diabetes Mellitus Genetics Study; FCHL, Familial Combined Hyperlipidaemia; DESIR, Data from an Epidemiological Study on Insulin Resistance Syndrome; DPS, Finnish Diabetes Prevention Study; KANWU, a multi-centre study with five participants (Kuopio/Finland, Aarhus/Denmark, Naples/Italy, Wollongong/Australia, Uppsala/Switzerland); ARIC, Atherosclerosis Risk in Communities; WHO-MONICA, Multinational MONItoring of trends and determinants of Cardiovascular diseases; IFG, impaired fasting glucose; CUDAS, Carotid Ultrasound Disease Assessment Study; Busselton, Busselton Population Health Survey; RIAD, Risk factors in Impaired glucose tolerance for Atherosclerosis and Diabetes.
* The same study with different populations is shown by a, b, c or d.
Fig. 1 Type 2 diabetes mellitus in eligible studies (sum of Pro12Ala and Ala12Ala (X/Ala) v. Pro12Pro). Estimated standardised effect sizes and CI are given for the single studies and for global comparison. A negative standardised effect indicates that the corresponding frequency is smaller in X/Ala than in Pro12Pro. M-H, Mantel–Haenszel. The reference numbers for the studies can be found in Table 1. The same study with different populations is shown by a, b, c or d.
Fig. 2 Serum insulin in eligible studies (sum of Pro12Ala and Ala12Ala (X/Ala) v. Pro12Pro). Estimated standardised effect sizes and CI are given for the single studies and for global comparison. A lower insulin concentration was significant for Ala allele carriers excluding Tschritter et al. (Reference Tschritter, Fritche and Stefan21b) (P = 0·02). IV, insulin values. The reference numbers for the studies can be found in Table 1. The same study with different populations is shown by a, b or c.
Fig. 3 Homeostasis model assessment for insulin resistance for eligible studies (sum of Pro12Ala and Ala12Ala (X/Ala) v. Pro12Pro). Estimated standardised effect sizes and CI are given for the single studies and for global comparison. A negative standardised effect size indicates that the corresponding quantity is smaller in X/Ala than in Pro12Pro. IV, insulin values. The reference numbers for the studies can be found in Table 1. The same study with different populations is shown by a, b or c.
The association of the Ala allele with T2DM (Fig. 1) included twenty-four studies that showed heterogeneity (P < 0·00 001). In the overall analysis, the Ala allele had a significant protective association (OR 0·79; 95 % CI 0·66, 0·95). Regarding ethnic differences, the Caucasian subgroup showed a significantly lower risk of developing diabetes for the Ala allele (P = 0·03); however, this protective association was not observed among non-Caucasians (P = 0·21). Both groups proved to be heterogeneous, but this result disappeared when the Caucasian study Soriguer et al. (Reference Soriguer, Morcillo and Cardona19) was deleted (heterogeneity P = 0·10; OR 0·86, 95 % CI 0·75, 0·98, P = 0·02).
Was also calculated the association between the Ala allele and T2DM according to BMI; both subgroups, normal weight (n 3) and overweight (n 16), showed little association (OR 0·40, 95 % CI 0·25, 0·66, P = 0·0004 and OR 0·89, 95 % CI 0·80, 0·99, P = 0·04, respectively). The overweight subgroup presented heterogeneity (P = 0·03), while the normal-weight subgroup was homogeneous (P = 0·47). The total n of this analysis was smaller (n 19) due to failure to provide the average BMI in some studies, so it was not possible to include these studies in one of the subgroups.
The association of serum insulin with the Ala allele showed that insulin concentration is lower in individuals of this allele compared with the wild genotype, but this relationship was not significant (OR − 0·04; 95 % CI − 0·09, 0·01; P = 0·09); the sample proved to be heterogeneous (P = 0·02). The I 2 test was 30 %, indicating moderate inconsistency between studies. Grouping by ethnicity, Caucasians (n 36) and non-Caucasians (n 13), and excluding the multicentre Andrulionytè et al. (Reference Andrulionytè, Zacharova and Chiasson20), produced the following results for Caucasians: OR − 0·04 (95 % CI − 0·11, 0·02, P = 0·22); and non-Caucasians: OR − 0·05 (95 % CI − 0·12, 0·03, P = 0·23). The subgroup of Caucasians showed heterogeneity (P = 0·01), with 39 % of the studies contributing to this, according to results of I 2; however, the subgroup of non-Caucasians was homogeneous (P = 0·49).
As seen in Fig. 2, the study that differed the most was Tschritter et al. (Reference Tschritter, Fritche and Stefan21b), whose sample, after being excluded from the analysis, showed no global heterogeneity (P = 0·17); the meta-effect size calculation using a fixed-effects model proved to be significant (standardised mean difference: − 0·05; 95 % CI − 0·09, − 0·00; P = 0·02). Thus, both subgroups, normal weight and overweight, presented themselves as homogeneous, with a negative association between the Ala allele and the concentration of insulin, although the effects were not significant (P = 0·029 and P = 0·07, respectively).
When separating the diabetic (n 12) and glucose-tolerant (n 30) groups, if the study of Tschritter et al. (Reference Tschritter, Fritche and Stefan21b) study is removed, analysis of the Ala allele is associated with lower insulin concentrations for the diabetic group (standardised mean difference − 0·13, 95 % CI − 0·24, − 0·02, P = 0·02; heterogeneity P = 0·53).
The same protective effect of the Ala allele is observed in the outcome of the association of HOMA-IR with polymorphism (Fig. 3), where the standardised effect size is significantly lower in the Ala allele carriers (P = 0·03) than in Pro12Pro genotype. The heterogeneity result in this analysis was not significant (P = 0·39), and the inconsistency test showed very low inconsistency. According to the subdivisions shown in Fig. 3, the overweight subgroup showed the lowest HOMA-IR values associated with the Ala allele (P = 0·02). However, the test to evaluate the difference between the groups was not significant (P = 0·35).
In the analysis by ethnicity, both the Caucasian and non-Caucasian subgroups were homogeneous (P = 0·41 and P = 0·29, respectively). Despite the association of each ethnic subgroup showing the same trend of the overall result (Caucasians' standardised mean difference − 0·06, 95 % CI − 0·11, 0·00, P = 0·07; non-Caucasians' standardised mean difference − 0·06, 95 % CI − 0·17, 0·05, P = 0·27), described previously, the associations were not significant.
Discussion
The studies presented in the present meta-analysis showed great variability in their frequency of polymorphism, ranging from 0·04 to 0·55; the samples with the lowest rates included non-Caucasian populations (African(Reference Kao, Coresh and Shuldiner12), Japanese(Reference Deeb, Fajas and Nemoto7c, Reference Hara, Okada and Tobe8a,8b, Reference Yamamoto, Kageyama and Nemoto22, Reference Yamamoto, Hirose and Miyashita23a,23b), Korean(Reference Oh, Min and Chung24, Reference Rhee, Oh and Lee25) and Asian descendants in general(Reference Tai, Corella and Deurenberg-Yap26a)). The lower frequencies of polymorphism coincided with lower average BMI; however, in one population of Asians with a higher mean BMI, the frequency remained low(Reference Radha, Vimaleswaram and Babu5a), similar to the results of previous studies(Reference Deeb, Fajas and Nemoto7c, Reference Hara, Okada and Tobe8a,8b, Reference Kao, Coresh and Shuldiner12, Reference Yamamoto, Kageyama and Nemoto22–Reference Tai, Corella and Deurenberg-Yap26a).
Of the selected studies, twenty investigated Caucasian populations, and eleven European populations; only eight Asian populations, a North American and a South American population were included (Table 1). It should be noted that studies with European populations that did not clarify the ethnicity of the individuals included were then considered Caucasians, due to the low mixture of races of these populations. Interest in investigating the influence of the Pro12Ala polymorphism on the development of T2DM and insulin resistance appears to be higher in Caucasian populations, probably because of the greater frequency of this polymorphism among them.
In the present meta-analysis, twenty-four recent studies were included in the analysis of the association between the Ala allele and T2DM (Fig. 1), resulting in a protective effect of this allele to lower the risk of developing diabetes similar to Altshuler et al. (Reference Altshuler, Hirshhorn and Klannemark27) (n 11; OR 0·79; P = 0·00 007) and Ek et al. (Reference Ek, Andersen and Urhammer3) (n 10; OR 0·81; 95 % CI 0·72, 0·91; P = 0·00 034). Furthermore, the risk of T2DM in Ala allele carriers could be shown to differ among ethnic groups, as the OR was lower in Caucasians and there was no significant risk in non-Caucasians (including Asians, African-Americans and South Americans). However, there was an increase in the overall CI and heterogeneity within Caucasian and non-Caucasian subgroups.
Regarding ethnicity, the meta-analysis performed by Ek et al. (Reference Ek, Andersen and Urhammer3) showed that both Caucasians (n 7) and Asians (n 3) presented a significant negative association between the Ala allele and T2DM. However, there was a statistical difference between the OR of the two ethnic groups because the strength of association was lower for Asians than for Caucasians (Asian OR 0·42, 95 % CI 0·26, 0·67 v. Caucasian OR 0·85, 95 % CI 0·76, 0·96; P = 0·0033). Nevertheless, Radha et al. (Reference Radha, Vimaleswaram and Babu5) found in a study conducted with one Caucasian and two Asian populations that the Ala allele did not protect South Asian populations against T2DM, but did protect the Caucasians. This study found no significant difference between the polymorphism frequency in South Asian diabetics and non-diabetics (20 v. 23 % in the Dallas cohort and 19 v. 19·3 % in the Chennai cohort; P>0·05). Thus, both studies corroborate the present meta-analysis by suggesting that the Ala allele is a protective factor for T2DM in Caucasian populations.
Only four studies(Reference Deeb, Fajas and Nemoto7c, Reference Douglas, Erdos and Watanabe9a, Reference Hara, Okada and Tobe8, Reference Soriguer, Morcillo and Cardona19) observed a significantly inverse association between the Ala allele and T2DM (i.e. found that the Ala allele is not associated with T2DM). Two of these analysed Nissei (second-generation Japanese) populations, one living in the Occident and another in Japan; the first(Reference Deeb, Fajas and Nemoto7) showed a strong association between the wild genotype (Pro12Pro) and T2DM (OR 4·35; 95 % CI 1·24, 15·3; P = 0·028). The other studies were carried out on Caucasians.
Pisabarro et al. (Reference Pisabarro, Sanguinetti and Stoll28) reported that Ala allele carriers developed T2DM at a younger age. Regarding sex, the Pro12Ala polymorphism was found to be strongly associated with T2DM in women but not in men(Reference Hegele, Cao and Harris2). The heterogeneity shown in Fig. 1 demonstrates that there is a high probability that the difference between OR is not due to chance (random error), but rather expresses different effects, probably influenced by age(Reference Pisabarro, Sanguinetti and Stoll28), sex(Reference Hegele, Cao and Harris2) or even lipids in the diet(Reference Luan, Browne and Harding29).
Those with the Ala allele had lower insulin concentrations in the global analysis, but this effect becomes significant when removing Tschritter et al. (Reference Tschritter, Fritche and Stefan21), which used a sample of overweight subjects with impaired glucose tolerance. This result confirms the meta-analysis performed by Tönjes et al. (Reference Tönjes, Scholz and Loefler30), who showed a standardised effect size of 0·168 (P = 0·040) for Ala12Ala's association with lower insulin concentrations compared with the Pro12Pro genotype. However, they verified no significant serum insulin effect in association with the Ala allele. Their study assessed only non-diabetic samples(Reference Tönjes, Scholz and Loefler30) that still showed homogeneity (P = 0·052).
Other studies found an association between the Ala allele and the lowest insulin concentration and increased sensitivity to insulin, regardless of sex(Reference Deeb, Fajas and Nemoto7), BMI(Reference Altshuler, Hirshhorn and Klannemark27), being non-diabetic(Reference Buzzetti, Petroni and Ribaudo31) and age(Reference Temelkova-Kurktschiev, Hanefeld and Chinetti32). Kao et al. (Reference Kao, Coresh and Shuldiner12) showed variations in BMI and fasting insulin depending on Pro12Ala genotype (P = 0·0027). Two other studies showed lower obesity levels associated with insulin genotype Pro12Pro(Reference Clement, Hercberg and Passige33, Reference Valve, Sivenius and Miettinen34), but there was no apparent similarity between the studies that had shifted to the right on the graph of this analysis.
While the Ala allele is associated with a lower risk of developing diabetes, it is interesting to note that even under conditions of abnormal glucose tolerance, this polymorphism was associated with lower concentrations of insulin. This contributes to cardiovascular risk factors because hyperinsulinaemia is involved in several cardiovascular pathophysiological mechanisms(Reference Takahashi, Hasebe and Kawashima35–Reference Adachi, Jacobs and Hashimoto37).
The beneficial effect of the Ala allele on sensitivity to insulin observed in the present study covered nineteen samples that analysed mean HOMA-IR, with one of the two samples composed of Brazilian diabetics being of significantly negative association(Reference Tavares, Hirata and Rodrigues38). Using diabetic and non-diabetic samples in the same meta-analysis enriched the present study because it is assumed that despite the different values found for glucose-tolerant individuals, the difference between the wild genotype and Ala allele averages would influence the final result.
The insulin resistance values were significantly lower in the group carrying the Ala allele (Fig. 3), as in Tönjes et al. (Reference Tönjes, Scholz and Loefler30) in the obesity group, although a significant association in the overall HOMA-IR effect was present, highlighting the power of association in this study through homogeneity among the studies selected for this analysis.
The present study has endeavoured to standardise the measures examined in order to minimise bias and heterogeneity between studies; stratification analysis was also performed to better characterise the subgroups.
Final considerations
These results suggest that the Ala allele is protective against T2DM development in Caucasians and not in some of the other populations. However, the mechanism has not been fully elucidated in the literature. It is important to emphasise that sensitivity to insulin is influenced by multiple factors; Pro12Ala polymorphism of the gene PPAR-γ2 is one of them.
Acknowledgements
The authors thank Professor Paulo Borges of the Institute of Communication and Information Science and Technology in Health/FIOCRUZ for his assistance with statistical analysis and the Conselho Nacional de Pesquisa (CNPq) and Fundação Carlos Chagas Filho de Amparo à Pesquisa do Rio de Janeiro for funding.
The authors certify that they have contributed substantially to the conception and planning and interpretation of the data; we have contributed significantly to the preparation of the draft or to the critical revision of the content; and we participated in the approval of the final version of the manuscript.
We declare that there are no conflicts of interest.
