Medical Sequencing at the Extremes of Human Body Mass

Abstract

Body weight is a quantitative trait with significant heritability in humans. To identify potential genetic contributors to this phenotype, we resequenced the coding exons and splice junctions of 58 genes in 379 obese and 378 lean individuals. Our 96-Mb survey included 21 genes associated with monogenic forms of obesity in humans or mice, as well as 37 genes that function in body weight–related pathways. We found that the monogenic obesity–associated gene group was enriched for rare nonsynonymous variants unique to the obese population compared with the lean population. In addition, computational analysis predicted a greater fraction of deleterious variants within the obese cohort. Together, these data suggest that multiple rare alleles contribute to obesity in the population and provide a medical sequencing-based approach to detect them.

Obesity is reaching epidemic proportions in developed countries and represents a significant risk factor for hypertension, heart disease, diabetes, and dyslipidemia.¹ Although the growing prevalence of obesity in the population is thought to be caused by increasing caloric intake and declining energy expenditure,² individual susceptibility to obesity is strongly influenced by heredity. Twin, adoption, and family studies have indicated that 40%–70% of interindividual variation in BMI is heritable.^3,4 In a limited number of cases, single gene defects have been linked to obesity,⁵ but the majority of cases are thought to be attributable to complex genetic and/or environmental interactions. In this study, we sought to explore the relationship between sequence variation in multiple candidate genes and the extremes of human body mass.

Candidate genes for the study included (a) 21 genes strongly associated with obesity that, when disrupted, lead to monogenic forms of obesity in humans and/or that cause obesity when inactivated in mice and (b) 37 genes involved in regulation of food intake,⁶ adipogenesis,⁷ energy expenditure,⁸ or lipid metabolism (table 1). The coding exons and splice junctions of each gene were sequenced in 379 extremely obese (mean BMI 49.0, >95th percentile adjusted for age and sex; BMI was calculated as weight in kilograms divided by square of height in meters) white men and women ascertained through an obesity clinic at the University of Ottawa and in 378 lean (mean BMI 19.4, <10th percentile adjusted for age and sex) apparently healthy white men and women who participated in a study of leanness at the same institution (table 2). A total of 134 kb (60 kb coding and 74 kb noncoding) was sequenced in each individual, representing 96 Mb of high-quality sequence data, with an average coverage of 734 individuals per exon (table 3). Cumulatively, we identified 1,074 genetic variants (see the tab-delimited ASCII file, which can be imported into a spreadsheet, of data set 1), averaging one variant per 125 bp of the reference human genome sequence. Of the variants, 252 were common polymorphisms (minor-allele frequency >1%), whereas the remaining 822 were rare variants, including 400 noncoding, 150 synonymous, and 272 nonsynonymous variants; the nonsynonymous variants included 3 in-frame indels and 8 severe alleles (6 out-of-frame indels and 2 nonsense changes). In accord with previous large-scale gene-centric sequence analyses,^169–171 we observed a paucity of nonsynonymous variants with increasing minor-allele frequency, which is consistent with purifying selection acting on a significant fraction of such DNA sequence changes (fig. 1). Of the 1,074 variants identified in this study, 989 (92%) were not listed in dbSNP (build 124), and, as expected, the majority of these variants (800 [81%] of 989) were rare (i.e., had a minor-allele frequency <1%).

Table 1. .

Summary of Genes and Rare Coding Variants That Are Unique to the Obese or Lean Population

						Obesea		Leana
Gene Group and Gene	OMIM	Mouse Knockouts	Mouse Transgenics	Human Mutations	Associations	NS	S	NS	S
Monogenic obesity:
BRS3	300107	Obese9	None	None	None	1	0	3	0
CARTPT	602606	Obese10	None	Obese11	Yes12–14	1	1	0	1
FABP4	600434	Obese15	None	None	Yes16	1	0	2	0
HTR2C	312861	Obese17	None	None	Yes18,19	1	0	0	0
IL6	147620	Obese20	None	None	Yes21–27	0	1	0	0
LEP	164160	Obese28	Lean29	Obese30	Yes31–38	0	3	0	1
MC3R	155540	Obese39	None	Obese40	Yes41,42	2	0	1	1
MC4R	155541	Obese43	None	Obese44	Yes45–49	8	1	2	1
NHLH2	162361	Obese50	None	None	None	2	0	1	1
NMU	605103	Obese51	None	None	None	1	0	1	0
NPB	607996	Obese52	None	None	None	1	0	2	0
NPBWR1	600730	Obese53	None	None	None	3	0	1	0
NPY1R	162641	Obese54	None	None	None	1	1	2	1
NPY2R	162642	Obese55	None	None	Yes56–58	2	3	2	0
NPY5R	602001	Obese59	None	None	Yes60	1	1	1	0
NR0B2	604630	No apparent phenotype61	None	Obese62,63	Yes63–65	3	0	2	0
PNPLA2	609059	Obese66	None	None	None	3	1	2	2
POMC	176830	Obese67	None	Obese68	Yes69–77	2	3	1	3
PYY	600781	Obese78	None	Obese79	Yes56–58,80	2	0	1	0
SIM1	603128	Obese81	Lean82	Obese83,84	None	6	2	0	2
UCP3	602044	No apparent phenotype85	Lean86	Obese87	Yes88–99	5	1	2	3
Total						46	18	26	16
Obesity candidate:
ADIPOQ	605441	Insulin resistance100	None	None	Yes101–107	2	0	2	0
AGRP	602311	No apparent phenotype108	Obese109	None	Yes110,111	1	1	0	2
APOA5	606368	Hyperlipidemia112	Lipid112	None	Yes113,114	1	0	2	1
ARNT2	606036	Lethal115	None	None	None	2	2	3	0
ASIP	600201	No apparent phenotype116	Obese117	None	None	0	0	0	0
C1QTNF2	…	None	None	None	None	1	2	0	2
C3AR1	605246	Hypoallergic118	None	None	None	4	0	4	3
CCK	118440	No apparent phenotype119	None	None	None	0	0	1	0
CPT1B	601987	None	None	None	None	5	2	7	2
CSF2	138960	Pulmonary anomalies120	None	None	None	0	0	0	1
DGAT1	604900	Lean121	None	None	Yes122,123	5	3	2	2
DGAT2	606983	Lean124	None	None	None	5	0	3	2
GHRL	605353	No apparent phenotype125	None	None	Yes126–129	1	0	0	1
GHSR	601898	No apparent phenotype130	None	None	Yes131–133	1	2	2	1
HSD11B1	600713	Obesity resistance134	Obese135	None	Yes136,137	0	1	1	0
HTR7	182137	Hyperthermia138	None	None	None	1	2	1	3
INSIG1	602055	None	None	None	None	0	2	3	0
INSIG2	608660	None	None	None	None	1	2	2	1
LIPC	151670	Hyperlipidemia139	None	None	Yes140	4	5	2	7
NMUR1	604153	None	None	None	None	4	4	2	1
NMUR2	605108	None	None	None	None	4	0	3	0
NPBWR2	600731	None	None	None	None	1	2	2	5
NPY	162640	No apparent phenotype141	None	None	Yes142–145	0	0	0	0
NTS	162650	No apparent phenotype146	None	None	None	0	0	4	0
PPARGC1A	604517	Lean147	None	None	Yes148–153	3	1	4	1
PPY	167780	None	Lean154	None	None	0	0	1	0
PRKAA1	602739	None	None	None	None	3	1	4	0
PRKAA2	600497	Glucose tolerance155	None	None	Yes156	4	2	3	1
PRKAB1	602740	None	None	None	Yes156	0	1	0	0
PRKAB2	602741	None	None	None	Yes156	2	0	1	0
PRKAG1	602742	None	None	None	None	0	1	0	1
PRKAG2	602743	None	None	Heart157	None	2	0	1	2
PRKAG3	604976	Glycogen metabolism158	Glycogen158	None	None	10	3	4	1
RETN	605565	Gluconeogenesis159	Obese160	None	Yes161–166	0	0	1	0
SIRT1	604479	Insulin sensitivity167	None	None	None	3	2	1	4
TGFBR2	190182	Embryogenesis168	None	None	None	1	1	1	1
WDTC1	…	None	None	None	None	1	0	2	0
Total						72	42	69	45

^aNS = nonsynonymous; S = synonymous.

Table 2. .

Summary of Individuals Included in This Study

Variable	Obese Cohort	Lean Cohort
No. of individuals	379	378
BMIa,b	49.0 ± 8.8	19.4 ± 1.6
BMIa percentile for age and sex	>95th	<10th
Ageb (years)	49.5 ± 10.7	45.5 ± 13.0
Female (%)	63	64
Weightb (kg)	124.8 ± 29.3	56.9 ± 9.0
Heightb (cm)	167.6 ± 10.1	170.5 ± 9.2
Waist circumferenceb (cm)	122.5 ± 20.1	75.8 ± 6.5

^aBMI values are those from the initial visit to the weight-management clinic.

^bData are mean ± SD.

Table 3. .

Sequencing Summary

Measure	Value
No. of genes	58
No. of exons	324
Genomic sequence covered (bp):
Total	134,449
Coding	60,372
Noncoding	74,077
Sequence overall (bp):
Total	96,059,368
Coding	44,254,489
Noncoding	51,804,879
Total no. of variants	1,074
No. of rare variants:
Total	822
Nonsynonymous	272
Synonymous	150
Noncoding	400
No. of common variants:
Total	252
Nonsynonymous	43
Synonymous	44
Noncoding	165
No. of novel SNPs covered	989
No. of known dbSNPs covered	85
No. of dbSNPs not discovered	366

Figure 1. .

The percentage of nonsynonymous, synonymous, and intronic variants for different minor-allele frequencies. Percentages and the actual number (N) of variants are written inside the bars of the graph.

It has been reported elsewhere that multiple rare variants can have a strong effect on complex traits, especially in the population extremes of a given phenotype.^172,173 We therefore examined the frequencies of the nonsynonymous variants in the obese and lean cohorts. Of the 272 rare nonsynonymous changes identified, 213 were unique to one group, with a small excess in the obese population (118 changes) compared with the lean population (95 changes), which did not reach statistical significance. A similar analysis revealed that the prevalence of unique rare synonymous variants, which approximate functionally neutral changes, was essentially identical in the obese and lean cohorts (60 in obese and 61 in lean). We next examined the distributions of nonsynonymous and synonymous variants within each gene individually and found that none of the genes had a statistically significant excess of nonsynonymous variants in the obese or lean group. However, when the genes associated with monogenic forms of obesity were considered together (table 1), unique nonsynonymous variants were significantly more common in the obese group (46 variants in 41 individuals) than in the lean group (26 variants in 27 individuals) (P<.05, by Fisher’s exact test). In contrast, the number of unique synonymous variants in these genes was almost identical among the obese group (18 variants) and lean group (16 variants). It is worth noting that the genes that accounted for the highest difference are MC4R (MIM 155541) (8 variants in obese vs. 2 in lean), SIM1 (MIM 603128) (6 in obese vs. 0 in lean), and UCP3 (MIM 602044) (5 in obese vs. 2 in lean).

The excess of nonsynonymous variants among obese individuals may reflect chance fluctuation in allele frequencies, population stratification, or the accumulation in this group of functional sequence variants that predispose individuals to obesity. Chance fluctuation in allele frequencies seems unlikely, since the excess of nonsynonymous variants in the obese group was not because of an increased number of variants in any single gene, but rather was because of the aggregate contribution of variants at several unlinked loci. Population stratification also seems improbable, since both groups comprised white men and women from the same region (Ottawa, Canada). Furthermore, the number of synonymous variants (table 1) and the allele frequencies of ∼250 common sequence variants (see below) in these genes were similar in the obese and lean groups. Therefore, it seems likely that the excess of rare variants in the obese group represents the accumulation of functional alleles that contribute to the phenotype in these individuals.

As a first step to assess the functional significance of the nonsynonymous sequence variants identified in the 21 genes associated with monogenic forms of obesity, we used the computer algorithm PolyPhen¹⁷⁴ to predict the effects of amino acid substitutions on protein function. We observed that variants identified in the monogenic obesity gene group were more likely to be deleterious in the obese cohort than in the lean cohort (19 in the obese vs. 4 in the lean; P<.002, by exact binomial test) (fig. 2A). In comparison, the number of benign variants (25 in the obese and 21 in the lean) and the number of synonymous variants (18 in the obese vs. 16 in lean) in these genes were similar in both cohorts. In contrast, the distribution of synonymous, benign, and deleterious alleles in the 37 candidate genes not associated with monogenic forms of obesity was similar in the obese and lean groups (fig. 2B). This finding is consistent with the notion that the excess of nonsynonymous sequence variants among the monogenic obesity genes in the obese cohort reflects the accumulation of functional variants.

Figure 2. .

PolyPhen distribution analysis of variants unique to the obese and lean cohorts. Data are presented for genes with evidence of monogenic involvement in obesity (A) and for genes with a biological plausibility for a role in obesity (B). The number of variants is indicated above each bar of the graph. A double asterisk (**) indicates P<.002.

To determine whether nucleotide changes within these genes segregate with BMI, we examined familial segregation for 28 rare variants identified in 14 genes (10 monogenic and 4 candidate genes; see data set 1) in obese kindreds, comprising the proband and all first-degree family members who were available and willing to participate. We used MC4R as a test case, since it is the most common cause of monogenic obesity, estimated to account for 1%–6% of cases of severe obesity.⁴⁴ In our study, we identified eight nonsynonymous variants that were unique to the obese cohort, compared with two unique variants in the lean cohort (table 1). We found that the mutant MC4R alleles clearly failed to segregate with obesity in three of the six kindreds with six or more family members available (fig. 3), including an allele with a previously characterized frameshift mutation (L211; 4-bp deletion; fig. 3E)¹⁷⁵ that is almost certainly functional. To corroborate that these MC4R variants were indeed functional, we performed established in vitro functional assays for the novel MC4R variants⁴⁴ identified in our obese population. Of the six putative mutations analyzed for segregation, five displayed impaired MC4R function (table 4). These findings are consistent with previous studies that also show incomplete correlation between MC4R mutations and obesity,^176,177 illustrating the difficulties inherent in determining the correspondence between genotype and phenotype in common complex phenotypes such as obesity. Although several of the kindreds available for study were small, none of the other rare variants examined in 13 additional genes showed significant segregation with BMI in a total of 21 kindreds (data not shown), with the exception of PYY (MIM 600781) Q62P, which we have reported elsewhere.⁷⁹

Figure 3. .

Familial segregation of MC4R variants and BMI. BMI values are based on the subjects' maximum weight. The arrow indicates the individual sequenced in the cohort. y = years.

Table 4. .

Functional Characterization of MC4R Nonsynonymous Variants in the Obese Cohort

				Results of Functional Studies
Variant	Sequence	n	Known or Novel	alpha-MSH Activation (EC50)	Basal Activity	Summary	Family Segregation Data
S30F	tgagt[c/t]ccttg	1	Known185	Not tested alone182	Not tested alone182	…	Not tested
G32E	ccttg[g/a]aaaag	1	Novel	.3 nM	70%	Minor	Figure 3A
E61K	tgttg[g/a]agaat	1	Novel	Low	⩽10%	Severe	Figure 3B
S127L	tgact[c/t]ggtga	1	Known182	29 nM	80%	Intermediate	Figure 3C
L211Dela	ttct[ctct/-]atgt	2	Known175	Truncated receptor	Truncated receptor	Severe	Figure 3D
P299Ha	cgatc[c/a]tctga	2	Known182	Negative	⩽10%	Severe	Figure 3E
A303T	tttat[g/a]cactc	1	Novel	Low	⩽10%	Severe	Figure 3F
C326R	gcctt[t/c]gtgac	1	Novel	.4 nM	150%	Minor	Figure 3G
Wild type	…	…	…	.3 nM	100%	…	…

^aIndividuals who had the L211Del also had the P299H variant.

Although the goal of our study was not to perform an exhaustive genetic association study between common variants and BMI, we identified 252 polymorphisms with a frequency >1% and examined the frequency distributions of the variants in the obese and lean cohorts (see data set 1). We found two variants that displayed a significant frequency difference between the two populations: rs6599571 in DGAT1 (MIM 604900) and rs1800832 in NTS (MIM 162650) (both variants are in the 5′ UTR of their gene) (see data set 1). In an attempt to replicate these findings, we compared their frequencies in a second obese cohort (n=382; mean BMI 38.6) and a second lean cohort (n=381; mean BMI 20.8). For both variants, we observed no significant difference in the allele frequencies between the second cohorts (data not shown), supporting the hypothesis that the initial observation was likely a false-positive discovery or limited to very extreme BMI phenotypes. We should further note that none of the 37 sequenced common variants that were examined elsewhere for their association with BMI displayed a significant frequency difference between our original obese and lean groups (table A1 in appendix A). These results suggest that, in this population, common variants within the coding regions and their proximal exon-intron junctions in this subset of 58 genes are unlikely to contribute appreciably to susceptibility to extreme BMI. However, because we screened primarily the coding sequences and splice junctions of these genes, we cannot exclude the possibility that common sequence variations in noncoding regions that were not sequenced in this study may have significant effects on BMI.

Whereas the heritability of BMI has been firmly established, the identification of genes that contribute to obesity has proved challenging. Genomewide association scans are becoming more feasible, both technologically and economically, and, with them, investigators have begun to systematically explore common variants that influence obesity.¹⁷⁸ However, such studies fail to capture rare variants that have also been shown to influence human phenotypes.^172,173 Resequencing of candidate genes selected for biological plausibility, in an attempt to capture such rare variants, has, in a few instances, resulted in the identification of obesity-associated variants. For instance, the observation that Mc4r-knockout mice are obese⁴³ led to the subsequent finding that mutations in this gene may lead to obesity in humans.^175,179 In the present study, we sought to use a similar approach, using a large-scale sequencing strategy with numerous obesity candidate genes in two cohorts with extreme BMI. We did not uncover a large number of novel genes associated with obesity, an endeavor that may have been obstructed by reasons that range from a partial candidate-gene list (58 genes), a large but still limited collection of only white individuals (n=∼380 in each group), the sequencing of mainly coding regions, and limited power and availability of subject pedigrees. However, we did identify several genes that warrant further investigation. For instance, we observed a noteworthy rare nonsynonymous variant difference between the obese and lean cohorts for SIM1 (6 variants in obese vs. 0 in lean) and PRKAG3 (10 in obese vs. 4 in lean), suggesting that nonsynonymous variants within these genes may influence susceptibility to obesity. SIM1 is of particular interest because of its strong biological plausibility, including evidence that human chromosomal aberrations within the SIM1 region may lead to obesity,^83,84 the observation that Sim1 heterozygous null mice develop obesity,⁸¹ and the absence of reported human obesity–associated rare nonsyndromic variants. In addition, we uncovered a significant difference in the total number of nonsynonymous variants in previously characterized monogenic obesity genes between the obese and lean cohorts, indicating that multiple rare variants may have an incomplete effect on this phenotype. Our familial segregation analysis demonstrated that even thoroughly characterized human monogenic obesity genes, such as MC4R, fail to show consistent linkage with BMI, which further suggests that these variants exhibit variable penetrance. Although our analysis encompassed a modest fraction of candidate BMI genes, it strengthens the hypothesis that the majority of genetic etiology that governs obesity is complex and is likely to be influenced by a combination of multiple susceptibility alleles, the majority of which are not independently causative of extreme BMI.

Subjects.—Unrelated obese white subjects were recruited from the patient population of the University of Ottawa Weight Management Clinic and the Heart Institute Lipid Clinic by use of criteria reported elsewhere.⁷⁹ Briefly, inclusion criteria included a BMI >36; a history of obesity for at least 10 years of adult life; no history of treatment with oral glucocorticoids, antipsychotics, or lithium; and no history of medical conditions including major depression, bipolar affective disorder, or psychosis. Unrelated lean subjects of the same ethnic background, with a BMI ⩽10th percentile for age and sex and with no prior history of a BMI >25th percentile for >2 consecutive years were recruited from the Ottawa community (table 2). Subjects were excluded if they had any medical condition that affects weight gain, such as hypo- or hyperthyroidism, eating disorders, major depression, or malabsorption syndromes. The management of phenotypic data was performed using the SAS statistical package (version 9.1 [SAS Institute]). BMI for obese and lean subjects was categorized according to population percentiles for age and sex by use of the Canadian Heart Health Survey data for subjects aged >18 years (data on file; Health Canada) and the National Health and Nutrition Examination Survey data for children.¹⁸⁰ This study was approved by the institutional review boards of the University of Ottawa Heart Institute and the Ottawa Hospital, and informed written consent was obtained from all participants. Genomic DNA was extracted from white blood cells by standard methods.¹⁸¹

Sequencing and data analysis.—Primers were designed to give a maximum product size of 500 bp and a minimum of 40 bp flanking the splice sites, by use of the Exon Locator and eXtractor for Resequencing program (ELXR Web site). An M13F tag (gttttcccagtcacgacgttgta) and an M13R tag (aggaaacagctatgaccat) were added to forward and reverse primers, respectively. From each sample, 10 ng of DNA was amplified in a 10-μl PCR by use of AmpliTaq Gold (Applied Biosystems) and was cleaned using the PCR product presequencing kit (USB Corporation). Bidirectional sequencing was performed using both of the M13 primers and BigDye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems) (JGI Web site), and cleaning was done with tetraethylene glycol before separation on a 3730xl DNA Analyzer (Applied Biosystems). Base calling, quality assessment, and assembly were performed using the Phred, Phrap, Consed (Green Group Web site), and PolyPhred (PolyPhred Web site) software suite. To filter out low-quality sequence, only sequences that had a Phred score ⩾27 were included in the analysis. To minimize false-negative results, we manually analyzed sequence data after PolyPhred analysis at a rank of 5. In addition, every low-quality read was visually examined for indels. All sequence variants identified were verified by manual inspection of the chromatograms, and 156 (99%) of the 157 nonsynonymous variants were verified by a second independent sequencing reaction. All variants were examined by Arlequin (Arlequin Web site), to test for Hardy-Weinberg equilibrium (table A1 in appendix A). Of the 1,074 genetic variants identified, 12 (4 coding and 8 noncoding) had >50% of the data missing in either the lean or the obese panel and thus were removed from further analysis.

PolyPhen analysis.—All coding SNPs were subdivided into groups of frameshift/nonsense variants, synonymous variants, and missense variants. Missense variants were further classified with respect to their potential impact on protein structure or function, on the basis of sequence conservation analyzed using a new version of the PolyPhen method.¹⁷⁴ PolyPhen relies on the analysis of multiple sequence alignments of homologous proteins, together with functional annotation and structural information if available. The new version of PolyPhen constructs multiple sequence alignments by using a pipeline of several existing programs for alignment of sequences, alignment quality control, and clustering of sequences. Computational prediction methods are statistical in nature; therefore, certain percentages of false-positive (∼10%) and false-negative (20%–30%) predictions are expected. However, application of computational predictions increases power to detect differences in the number of rare functional nonsynonymous variants in candidate genes between populations with different phenotypes.

MC4R functional analysis.—Cloning and functional studies of the MC4R mutations were performed as described elsewhere.^176,182,183 Briefly, since MC4R is a single-exon gene, mutated alleles were amplified and cloned directly from the genomic DNA of the patient. This also allowed for confirmation of the presence and the nature of the mutations. Human MC4R alleles were cloned into the pcDNA3 expression vector (Invitrogen), to express the native form and the N-terminal FLAG-tagged and/or C-terminal V5His-tagged form of the receptor. All expression vectors were sequenced, to establish the presence of the mutation and the absence of PCR-induced mutations.

For alpha–melanocyte stimulating hormone (alpha-MSH) activation studies, receptors were transiently transfected into an human embryonic kidney (HEK) 293 cell line stably expressing luciferase under the control of a cAMP-responsive promoter.¹⁸² Cells were split into 96-well plates 24 h after transfection, and, 36 h after transfection, they were washed and incubated in stimulation medium (Minimum Essential Medium–alpha containing 0.1 mg/ml BSA and 0.25 mM isobutylmethylxanthine) and were stimulated with different concentrations of alpha-MSH (Sigma) for 6 h at 37°C in a 5% CO₂ incubator. Luciferase activity, representing cAMP produced in response to alpha-MSH, was assessed using the Steady-Glo Luciferase Assay System (Promega) and a microplate luminescence counter (Packard Instrument). Results were normalized to the maximal stimulation by 8Br-cAMP. Basal activity of the receptors was determined by transient cotransfection with the cAMP-dependent luciferase–expressing plasmid. All experiments were normalized for transfection efficiency by cotransfection of a plasmid encoding the Renilla luciferase–expression plasmid pRL-RSV, to control for transfection efficiency. Data were analyzed using the GraphPad Prism software (GraphPad Software).

Statistical analysis of common variants.—Common SNPs were preprocessed to remove triallelic SNPs (one SNP removed) and SNPs for which >50% of the data were missing (three SNPs removed). In addition, we clustered together SNPs that differed in at most three individuals, picking one representative from each such cluster. Standard χ² tests based on a 3×2 contingency were applied for each of the remaining 252 SNPs, on the basis of a contingency table of genotype-phenotype frequencies. The obtained P values were adjusted for multiple SNP testing by use of the false-discovery-rate procedure.¹⁸⁴

Appendix A

Table A1. .

Analysis of Previously Examined Variants for Association with BMI

		No. of Subjects with Genotype		P Value
Gene and Published Variant (dbSNP)	Sequence	Obese Cohort	Lean Cohort	χ2 Testa	Multiple Testingb
ADIPOQ:
G15G102–105,107,186 (rs2241766)	cccgg[t/g]catga	TT = 291; GT = 57; GG = 35	TT = 279; GT = 57; GG = 35	.969638	1
AGRP:
A67T110,111 (rs5030980)	aggag[g/a]ctcag	GG = 299; GA = 28; AA = 0	GG = 323; AG = 26; AA = 0	.593803	.935988
APOA5:
5′ UTR114 (rs651821)	agcag[a/g]taatg	AA = 322; AG = 43; GG = 1	AA = 319; AG = 48; GG = 1	.594544	.935988
S16W114 (rs3135506)	gtttt[c/g]ggcca	CC = 331; CG = 38; GG = 1	CC = 331; CG = 43; GG = 2	.599993	.935988
GHRL:
R51Q126,127,129	gcccc[g/a]agctc	GG = 369; GA = 7; AA = 0	GG = 370; GA = 2; AA = 0	.0967801	.633575
L72M126–129 (rs696217)	atgaa[c/a]tggaa	CC = 319; CA = 57; AA = 2	CC = 323; CA = 48; AA = 1	.382417	.935988
GHSR:
G57G131,133 (rs495225)	gtggg[t/c]atcgc	TT = 189; TC = 158; CC = 32	TT = 143; TC = 118; CC = 33	.478552	.935988
MC3R:
T6K42 (rs3746619)	aaaga[c/a]gtatc	CC = 310; CA = 61; AA = 1	CC = 311; CA = 59; AA = 2	.854573	.980245
V81I41,42 (rs3827103)	ccgag[g/a]ttttc	GG = 314; GA = 62; AA = 1	GG = 312; GA = 58; AA = 2	.762315	.9797
MC4R:
I103V46–49 (rs2229616)	ccatt[a/g]tcatc	GG = 369; GA = 10; AA = 0	GG = 366; GA = 11; AA = 0	.660644	.958137
NPY:
L7P142,145 (rs16139)	gcgac[t/c]ggggc	TT = 343; TC = 34; CC = 0	TT = 354; TC = 19; CC = 1	.0359879	.633575
S50S144 (rs9785023)	tactc[g/a]gcgct	AA = 95; AG = 209; GG = 73	AA = 107; AG = 189; GG = 78	.392291	.935988
IVS2-116delT144 (rs16134)	(tgaacacctgacaataa/-)	−/− = 95; −/+ = 175; +/+ = 65	−/− = 95; −/+ = 155; +/+ = 88	.0974731	.633575
S68S144 (rs5574)	cgatc[c/t]agccc	TT = 102; TC = 184; CC = 78	TT = 99; TC = 177; CC = 94	.44487	.935988
NPY2R:
I195I56–58 (rs1047214)	atcat[t/c]ccgga	TT = 111; TC = 179; CC = 84	TT = 108; TC = 192; CC = 72	.493036	.935988
I312I56,58 (rs2880415)	cacat[t/c]atcgc	TT = 108; TC = 175; CC = 90	TT = 109; TC = 192; CC = 73	.277517	.935988
NPY5R:
3′ UTR60	ttttg[t/c]taaca	TT = 335; TC = 26; CC = 0	TT = 317; TC = 41; CC = 1	.0499633	.633575
NR0B2:
G171A63–65 (rs6659176)	gaaag[g/c]gacca	GG = 320; GC = 37; CC = 2	GG = 293; GC = 67; CC = 3	.00666681	.260006
POMC:
C6C76 (rs8192605)	tgctg[c/t]agccg	CC = 364; CT = 7; TT = 0	CC = 357; CT = 5; TT = 0	.589687	.935988
SSG99∧100Ins69,71,73 (rs10654394)	ggc[-/agcagcggc]gca	−/− = 343; −/+ = 31; +/+ = 2	−/− = 329; −/+ = 46; +/+ = 1	.0730815	.633575
A195A76 (rs2071345)	cctgc[c/t]gatga	CC = 374; CT = 0; TT = 0	CC = 367; CT = 1; TT = 0	1	1
E214G72	ggccg[a/g]gaga	AA = 368; AG = 7; GG = 0	AA = 368; AG = 3; GG = 0	.208963	.905505
Y221C77,187	cccct[a/g]cagga	AA = 374; AG = 1; GG = 0	AA = 370; AG = 2; GG = 0	1	1
R236G69,70	acaag[c/g]gctac	CC = 369; CG = 6; GG = 0	CC = 366; CG = 6; GG = 0	.98881	1
3′ UTR74–76 (rs1042571)	gctct[c/t]ccctg	CC = 227; CT = 129; TT = 18	CC = 236; CT = 117; TT = 18	.687893	.958137
PPARGC1A:
T394T153 (rs2970847)	aaaac[g/a]gaaat	GG = 226; GA = 107; AA = 14	GG = 240; GA = 111; AA = 20	.686798	.958137
G482S148–153 (rs8192678)	agacc[g/a]gtgaa	GG = 155; GA = 176; AA = 42	GG = 153; GA = 167; AA = 52	.519019	.935988
PYY:
R72T56,57,80 (rs1058046)	ggaca[g/c]gcttc	GG = 174; GC = 153; CC = 47	GG = 166; GC = 173; CC = 34	.173746	.847014
IVS3-6156,80 (IVS3+68) (rs162430)	catca[c/t]ttaac	CC = 293; CT = 51; TT = 6	CC = 299; CT = 67; TT = 5	.425068	.935988
RETN:
IVS2+39164 (rs3219177)	ggtct[c/t]agaga	CC = 234; CT = 133; TT = 12	CC = 227; CT = 100; TT = 17	.138268	.77035
3′ UTR162,164 (rs3745368)	tgcgg[g/a]ggagc	GG = 350; GA = 24; AA = 1	GG = 352; GA = 22; AA = 1	.760829	.9797
SIM1:
P352T188 (rs3734354)	ccaaa[c/a]cagcc	CC = 277; CA = 96; AA = 8	CC = 279; CA = 92; AA = 6	.821095	.9797
A371V188 (rs3734355)	ggggg[c/t]caaat	CC = 277; CT = 96; TT = 8	CC = 278; CT = 92; TT = 6	.828977	.9797
T653T188	cccac[c/t]gcact	CC = 369; CT = 14; TT = 0	CC = 357; CT = 20; TT = 0	.278331	.935988
UCP3:
Y99Y95,99 (rs1800006)	ctcta[t/c]gactc	TT = 217; TC = 137; CC = 25	TT = 210; TC = 143; CC = 22	.81318	.9797
V102I92,95,96 (rs2734830)	actcc[g/a]tcaag	GG = 378; GA = 1; AA = 0	GG = 375; GA = 0; AA = 0	1	1
Y210Y92,95,96,99 (rs2075577)	gacta[t/c]cacct	TT = 191; TC = 107; CC = 85	TT = 202; TC = 105; CC = 69	.381994	.935988

^aStandard χ² test based on a 3 × 2 contingency.

^bAfter correction for multiple testing by use of the false-discovery-rate procedure.