The rs1042714 (Glu27Gln) polymorphism
We performed a thorough search of the medical literature in the English, Spanish, Portuguese, French, Italian, German, and Romanian languages. For medical articles in English, we searched PubMed, Embase, CINAHL (Cumulative Index to Nursing and Allied Health literature), the Cochrane database, Current Concepts, Austrom (Meditext), and Medind using the terms Glu27 and obesity. Other MeSH (Medical Subject Heading) terms such as BMI, morbid obesity, overweight, body weight, weight gain, weight loss, body weight changes, body weights, and measures were also used, in combination with Glu27. We used the search engines Google and Scirus to find additional articles, conference abstracts, and government reports suitable for inclusion in our meta-analysis. For Spanish literature, we used Google and the search terms “Glu27” and “obesidad.” For Portuguese literature, we used “Glu27” and “obesidade.” For French literature, we used “Glu27” and “obesité.” For Italian literature, we used “Glu27” and “obesità.” For German literature, we used “Glu27” and “obesitas.” The German terms captured also the Dutch literature. For Romanian medical literature, we used “Glu27” and “obezitate.” We did not search the medical literature in other languages.
As of 18 August 2006, when we finished our search, we identified 39 articles that reported polymorphisms at codon 27 of the ADRB2 gene (rs1042714) and included both obese and nonobese subjects. The English literature search provided 38 of these. The foreign language search provided only a doctoral thesis in Spanish, the results of which also appeared later in English literature. All these studies were selected for meta-analysis, including four with related family pedigrees as well as those using slightly different BMI cut points. Six additional studies performed on three databases (two studies/database in English) and one in Portuguese, that genotyped only obese persons, were excluded.
The broad reach of our terms led us to capture databases used to study other topics, such as the contribution of rs1042714 to various diseases, i.e., diabetes, hypertension, and metabolic disorders of the lipid metabolism, where the authors had calculated the BMI of the patients in addition to genotyping. In some of these databases, the BMI was not reported as a cutoff limit, as needed for our analysis, but rather was reported as a range of values with a standard deviation. In such instances, we corresponded with the authors and asked them to send us their data showing counts of subjects above and below a BMI of 30kg/m2. At this stage, we were able to include 28 of the 39 articles, covering 71.6% of the total number of patients reported. Eleven authors did not respond to our inquiries.
We considered the subjects who were obese (BMI above the cutoff limit) as being the cases, the nonobese subjects as being the controls. The subjects with the mutant allele at codon 27, coding for glutamic acid in the 27th amino acid position of the ADRB2 protein, were considered “exposed” and those with the wild type (Gln27Gln) were considered “nonexposed.” We pooled together the populations of both genders from all studies and compared in three ways those who have the glutamic acid allele at codon 27 (the “exposed”) to the wild type Gln27Gln (the “nonexposed”):
• the first statistical analysis compared the homozygous Glu27Glu to the homozygous Gln27Gln;
• the second statistical analysis compared the heterozygous Glu27Gln to the homozygous Gln27Gln; and
• the third statistical analysis compared the combination of homozygous Glu27Glu and heterozygous Glu27Gln to the homozygous Gln27Gln.
Most of the authors used a cutoff BMI = 30 in their reports (Table 1). Five used BMI = 27 as a cutoff limit, two used BMI = 25, one used BMI = 26, and one a cutoff BMI = 26.4. Two authors characterized the cutoff BMI they used as “extreme” (18,19). We contacted a corresponding author and he stated that the obese subjects had a BMI above the 97th percentile, while the nonobese subjects had a BMI < 20. In the other study, “extreme” is clearly classified as morbid obesity, with a BMI > 40.
Table 1. Studies included in the initial selection for meta-analysis at codon 27
As shown in Table 1, the subjects were from the United States, Europe (Sweden, France, Italy, Spain, Austria, the Netherlands, Denmark, and Poland), Asia (Japan, South Korea, and Taiwan) and two isolated populations: Tongans and Aymara American Indians from Chile.
We calculated the frequency of the Glu27 allele at the global level. Then, due to the wide variation in the frequency of Glu27 in these populations, we separated them by race and we performed further subgroup analyses. The subgroups were:
• whites (all Europeans (6,8,9,10,12,14,16,18,19,20,21,22,23,24,25) and the Anglo-Celtic Australian whites (7)).
• Asians, Pacific Islanders, and American Indians (Japanese (4,5,11,26,27), Taiwanese Chinese (28), Koreans (29), Tongans (30), and Aymara American Indians (31)).
• Americans (mixed races (13,32,33) from United States and Canada).
We separated the data by gender (male, female, and both). Some authors provided data only in subjects of one gender, while others did not tabulate subjects based on gender and BMI combined or failed to report gender at all. The latter studies are included only in the tabulations of both sexes combined.
Then we calculated the Hardy-Weinberg equilibrium for every study both in the main group and in the gender-based subgroups. The procedure, including the formulas, is explained elsewhere (34). In order to perform the calculations, we used the Hardy-Weinberg equilibrium calculator software posted at http:www.changbioscience.comgeneticshardy.html. This calculator allowed us to introduce the frequencies of both alleles and their total number at the top and it gave the expected frequencies of genotypes at the bottom. Furthermore, we compared the real and the expected genotype frequencies using the χ2-test. We performed the χ2-test for the whole study and, separately, for both the experimental group and the control group. Also, we performed the test for both genders and then for male and female subgroups separately, if gender-based data were reported. If the χ2-test showed a significant departure from the Hardy-Weinberg equilibrium in the control group, the study was excluded from further analyses. If the departure was significant for the whole study population and/or for the experimental group, but not for the control group, the study was included. Following these calculations, we excluded from the main analyses four studies that have a significant departure from the Hardy-Weinberg equilibrium in the control group. The first study did report gender-based data, but, based on the χ2-test, it was excluded from both the main analyses and the gender-based subgroup analyses in males as well as in females. The second study did not report gender-based data, the third study reported only data in women and a fourth one reported data only in men. A fifth study was excluded because it provided data only for the heterozygotes and for the Gln27 homozygotes, while data for the Glu27 homozygotes were not reported and the Hardy-Weinberg equilibrium could not be calculated. In addition, a study reporting data for both genders was excluded only from the male subgroup, as long as the control group for the male population slightly departed from the Hardy-Weinberg equilibrium. Eventually, 23 studies were included in the main analyses, covering 51.55% from the subjects reported in the medical literature.
For our meta-analysis, we used the “Comprehensive Meta-analysis” software offered by Biostat (Englewood, NJ), at http:www.meta-analysis.com (35). The program provides summary results for both random effects and fixed effects models.
Using these genotype comparisons, we performed subgroup analyses based on race and, separately, on gender, where we included all suitable studies, irrespective of BMI. This way, we performed nine race-based subgroup analyses (three race categories times three categories for the polymorphism) and nine gender-based subgroup analyses (three gender categories—males, females, and both—times three categories for the polymorphism).
We then removed the studies with a BMI different from 30 and, based on race and gender, respectively, we performed two sets of nine subgroup statistical analyses each, that included only the studies that used a cutoff value of BMI = 30.
We performed sensitivity analyses, by first removing the studies that had at least one cell with zero subjects included, and then further removing studies that had at least one cell with less than five subjects included. Separately, by using the CMA software, we performed sensitivity analyses with one study removed.
After the sensitivity analyses, we performed the same operations with all 28 studies included, to detect if there is a loss of power or statistical significance after removing the studies that significantly departed from the Hardy-Weinberg equilibrium or did not provide sufficient data to calculate it. Then, we tested for heterogeneity using the CMA software (35).
In the last step, we ascertained publication bias, using a combination of graphical and numerical methods. As graphical methods, we chose to ascertain the publication bias using a combination of two funnel plots (the funnel plot of precision by log odds ratio (OR) and funnel plot of standard error by log OR, in both fixed effects and random effects model). As numerical methods, we chose to use the Egger's test.
The rs1042713 (Arg16Gly) polymorphism
For this polymorphism we used the same search strategy, by replacing Glu27 with Arg16. As of 18 August 2006, when we finished our search, we identified 27 articles that reported patients genotyped at codon 16 of the ADRB2 gene and included both obese and nonobese subjects, all of them in the English literature. All these studies were selected for meta-analysis, including those with related family pedigrees as well as those using slightly different BMI cut points. Four additional studies, which genotyped only obese persons, were excluded.
As shown in Table 2, the patients were from the United States, Europe (Sweden, Germany, Austria, and Italy), and East Asia (Japan, South Korea, and Taiwan).
Table 2. Studies included in the initial selection for meta-analysis at codon 16
There were four studies reporting data only in men and two studies reporting data only in women. Seven studies reported data in both genders, of which only two reported data for men and women separately and could be included in the gender specific analyses.
Most of the authors used a cutoff BMI = 30 in their reports (see Table 2). One author characterized the BMI cutoff as “extreme” (the obese subjects had a BMI above the 97th percentile), while three others used as a cutoff limit BMI = 27. One other author used BMI = 25 and another author a cutoff BMI = 26.
When the BMI was not reported as a cutoff limit, as needed for our analysis, we corresponded with the authors and asked them to send us their data showing counts of subjects above and below a BMI of 30 kg/m2. Eventually, we were able to include 13 of the 27 articles, covering 36.6% of the total number of patients reported. Fourteen authors did not respond to our inquiries (some of them were asked to send data for both polymorphisms, while others only for rs1042713).
We considered the subjects who were obese (BMI above the cutoff limit) as being the cases and the nonobese subjects as being the controls. The subjects with the mutant allele at codon 16, coding for arginine in the 16th amino acid position of the ADRB2 protein, were considered “exposed” and those with the wild type (Gly16Gly) were considered “nonexposed.” The statistical analysis was conducted in the same way for rs1042713 and rs1042714. When we calculated the Hardy-Weinberg equilibrium for every study, both in the main group and in the gender-based subgroups, we removed two studies (both performed in Europeans, one in men and one performed only in women) that did not have the control group in Hardy-Weinberg equilibrium. Thus, from 13 studies initially selected, only 11 studies underwent further analysis. These two studies were also removed from gender-based and race-based analyses. The sensitivity analyses were also performed and posttest power was calculated the same way as for the B27 meta-analysis. Heterogeneity was studied and the publication bias was ascertained by using the CMA software.