Study groups

An overview of the ten study groups is given in Table 1, details have been described previously [20, 21]. The selection of individuals for the mutation screen was based on genotypes at SNP rs2008514 (proxy of rs7359397) in the vicinity of SH2B1. In total, we analyzed 95 individuals, 90 of whom were likely enriched for the presence of mutations in SH2B1. The other five individuals are heterozygous carriers of a deletion at chr16p11.2 which does not harbor SH2B1[10]. These extremely obese patients (offspring) from the family-based GWAS sample were homozygous for the risk allele T of rs2008514 and had at least one heterozygous parent, thus substantially contributing to the observed over-transmission of the rs2008514 T-allele. Association to obesity of detected variants was analyzed in three steps (Figure 1):

1. i.

   Association testing: All detected variants were genotyped in a sample of 179 extremely obese (age- and sex-adjusted BMI percentile ≥ 99^th; [22]) children or adolescents and 185 lean adult (age- and sex-adjusted BMI percentile 「 15^th; [22]) controls. Basic phenotypical characteristics are given in Table 1. Individuals were independent of the mutation screening sample and were part of our case–control GWAS sample (Genome-Wide Human SNP Array 6.0, see [20, 21]).

    
2. ii.

   Further exploration of non-synonymous variants: The three non-synonymous variants (rs147094247: g.2749C/A - Thr175Asp; rs7498665: g.8164A/G - Thr484Ala; g.9483C/T - βThr656Ile/γPro674Ser) were additionally genotyped in the remaining individuals of the family-based sample and the cases and controls of our GWAS sample, who were not screened for mutations (see Mutation Screen section) and in 988 obese adults [23] as well as in 1,185 independent obese children or adolescents of the ‘Datteln Paediatric Obese Cohort’ (DAPOC [24]; Table 1).

    
3. iii.

   Additional exploration of rare non-synonymous variants: The two coding mutations without previously shown association to obesity βThr656Ile/γPro674Ser and rs147094247 - Thr175Asp - were additionally genotyped in three independent study groups comprising obese children and adolescents (‘Berlin Paediatric Obese Cohort’ — BEPOC; n = 1,046 [25]; Ulm Children's Study 2 and 3; n = 271 and 129, respectively [26]; Table 1) and in two independent population-based cohorts, the ‘Cooperative Health Research in the Region of Augsburg’ (KORA; n = 10,077 [27]) cohort of adults, and the ‘Ulm Children's Study 1’ (n = 782 [28]) study group of children and adolescents (Table 1).

    

Samplea	status	n	% male [%]	age [mean ± SD]	BMI [mean ± SD]	BMI SDSb [mean ± SD]
Mutation Screen	Cases	95	48.42	13.43 ± 3.37	31.87 ± 5.04	4.10 ± 1.71
S I : family-based GWAS	Cases	705	45.11	13.44 ± 3.02	32.02 ± 5.82	4.23 ± 1.96
	Parents	1,410	50.00	42.54 ± 6.02	30.28 ± 6.33	1.65 ± 1.84
S II : case–control GWAS	Cases	453	42.60	14.37 ± 3.75	33.15 ± 6.68	4.51 ± 2.15
	Controls	435	39.08	26.08 ± 5.75	18.09 ± 1.14	−1.45 ± 0.34
S III : subset of case–control GWAS  (for association testing)	Cases	179	49.16	14.27 ± 2.39	35.56 ± 6.13	5.3 ± 2.09
	Controls	185	55.68	25.56 ± 3.94	18.39 ± 1.09	−1.47 ± 0.33
S IV : obese adults	Cases	988	37.25	47.17 ± 14.23	35.70 ± 5.43	3.22 ± 1.66
S V : DAPOC	Cases	1,185	44.22	10.72 ± 2.78	27.69 ± 5.11	3.07 ± 1.56
S VI : KORA	Cases	6.633	56,43	56.62 ± 13.13	29.42 ± 3.70	1.19 ± 1.09
	Controls	3.444	36,76	47.18 ± 13.37	22.72 ± 1.68	−0.53 ± 0.51
S VII : Ulm children’s study 1	Cases	97	57.73	7.57 ± 0.42	20.68 ± 1.71	1.06 ± 0.46
	Controls	685	54.31	7.56 ± 0.42	15.63 ± 1.38	−0.25 ± 0.37
S VIII : BEPOC	Cases	1,046	47.99	10.89 ± 3.57	29.46 ± 5.91	3.54 ± 1.82
S IX : Ulm children’s study 2	Cases	271	51.29	11.07 ± 3.69	29.75 ± 6.04	3.62 ± 1.88
S X : Ulm children’s study 3	Cases	129	43.41	14.90 ± 1.84	40.45 ± 8.00	7.05 ± 2.92

^aMutation Screen sample: part of the family-based and the case–control GWAS samples’ cases; Family-based GWAS sample: 705 index patients with early-onset extreme obesity and their biological parent; case–control GWAS sample: GWAS of early-onset extremely obese children and adolescents in comparison to lean, adult controls; case–control sample for association testing: early-onset extremely obese children and adolescents in comparison to lean, adult controls; independent of initial screening sample; subset of cases-control GWAS sample; DAPOC: Datteln Paediatric Obese Cohort (27); KORA: Cooperative Health Research in the Region of Augsburg (30); Ulm children’s study 1: Ulm Research on Metabolism, Exercise and Lifestyle in Children (31); BEPOC: Berlin Paediatric Obese Cohort (28); Ulm children’s study 2 and 3: Ulm Paediatric Obese Cohort A and B (INSULA) (29).

^bCalculation of the BMI SDS values has been based on population reference values following the National Nutrition Survey I (25).

In all samples, body mass index (BMI in kg/m²) was assessed and age- and sex-specific percentile criteria with regard to the German population at the time of sample recruitment (S_I, S_II, S_III, S_IV, S_VI[22]) were used to define overweight or obese cases. Samples were divided into cases (adults: BMI ≥ 25, children and adolescents: ≥ 90^th BMI percentile (http://www.mybmi.de)) and controls (adults: BMI ≤ 25, children and adolescents: BMI ≤ 90^th percentile (http://www.mybmi.de)). Written informed consent was given by all participants and in case of minors by their parents. These studies were approved by the Ethics Committees of the respective Universities and Institutions and were performed in accordance with the Declaration of Helsinki.

Mutation screen

The coding region of SH2B1, located at chr16: 28,858,010 - 28,885,533 (hg18/ NCBI 36), was screened for mutations by denaturating high pressure liquid chromatography (dHPLC, WAVE, Transgenomics) as described previously [29]. Accuracy of dHPLC is similar or even higher than sequencing [30]. To enhance the sensitivity of detection of homozygous mutation carriers, DNA of an individual with wild type genotype (re-sequenced) was added to each sample prior to PCR amplification. All samples with deviant dHPLC patters were re-sequenced. Detailed information regarding used temperatures and primers for PCR and dHPLC analysis can be obtained from the authors. All PCR amplicons with dHPLC patterns deviant from the wild-type pattern were re-sequenced as described previously [31]. At least two experienced individuals independently assigned the genotypes; discrepancies were solved either by reaching consensus or by re-genotyping.

Genotyping

The identified variants in SH2B1 were genotyped in larger study groups using either restriction fragment length polymorphism (RFLP) or TaqMan Assays (detailed information can be obtained from the authors). For rs147094247 (Thr175Asp) and the new coding mutation in fragment 9 (g.9483C/T βThr656Ile/γPro674Ser), custom TaqMan assays were designed (SH2B1_2I_MUT, Assay ID: AHCS0BY, and Frag9_mut1, Assay ID: AHMSHDX, respectively, both Applied Biosystems). At least two experienced individuals independently assigned the genotypes; discrepancies were solved either by reaching consensus or by re-genotyping.

Statistics

Allele and genotype distributions of all detected variants did not deviate from Hardy-Weinberg equilibrium. To analyze the obesity association of all variants, Fisher’s exact test (allelic association) was calculated with PLINK [32]. Population-based samples were divided into cases (BMI ≥ 90^th percentile) and controls (BMI 「 90^th percentile). For rs7498665 an asymptotic, 2-tailed p-value for the transmission disequilibrium test (TDT) was additionally calculated with PLINK. If not stated otherwise, all p-values are asymptotic, two-sided and not corrected for multiple testing.

Functional in silicoanalyzes

To determine the potential alteration in gene expression, all mutations were analyzed for loss or gain of cryptic splice sites, transcription factor binding sites and gain or loss of o-glycosilation sites. Prediction of possible impact of amino acid exchange on structure and function of SH2B1 was done by PolyPhen-2 [33], SNAP [34], PMUT [35], and MutationTaster [36]. Detailed description of used tools can be found in the Additional file 1: Supplementary materials. Conservation was analyzed by aligning sequences of 21 species in total (21 α, eight β and six γ sequences, comprehensive list in Additional file 1: Supplementary materials).

Functional in vitroanalyzes: STAT3 mediated leptin receptor signalling

The effect of SH2B1 harboring the infrequent alleles of Thr484Ala and βThr656Ile/γPro674Ser on leptin receptor activity was determined with a quantitative reporter gene assay (adapted from [37]). HEK293 cells were transiently transfected with the murine long form of the leptin receptor (Lepr-b) in pcDNA3.1, a signal transducer and activator of transcription 3 (STAT3) responsive Photinus luciferase construct (pAH32), a constitutive Renilla luciferase expression vector for data normalization (phRG-b, Promega) and human SH2B1 splice variants beta and gamma with and without the mutations in pCMV-XL5 expression vectors (Lepr-b and pAH32 were kindly provided by Rosenblum et al. [37]). For control empty pcDNA3 expression vector was transfected instead of SH2B1. After stimulation with mouse leptin (concentrations 0, 0.5, 1, 5 , 10, 50, 100, 500 ng/ml), the STAT3 reporter construct led to luciferase expression measured with the Dual–Luciferase Reporter Assay system according to manufactures’ instruction (Promega). Dose–response curves, EC50 and Emax values were calculated by Graph Pad Prism.

New infrequent variant βThr656Ile/γPro674Ser

All three detected βThr656Ile/γPro674Ser mutation carriers are female. The initially identified mutation carrier (a) of the screening sample (height 163 cm, weight 86.2 kg, BMI 32.44 kg/m², age 12.7 years) as well as one mutation carrier (b) from the follow-up samples (height 142 cm, weight 53.2 kg, BMI 26.38 kg/m², age 9.9 years) had a BMI > 99^th percentile. In both cases, the overweight or obese mother (BMI 25.76 kg/m² and 32.61 kg/m², respectively) transmitted the mutation to the extremely obese child. The third mutation carrier (c) had a BMI > 90^th percentile (height 130 cm, weight 31.5 kg, BMI 18.64 kg/m², age 7.2 years). For this mutation carrier, genotypic information about the parents was not available (mother BMI 19.81 kg/m², father BMI 26.7 kg/m²). Insulin levels were only available for one mutation carrier (b), whose level was in the normal range (9.4 mU/l). Additional family members were not available.

The amino acid exchanges in the β and γ splice variants (βThr656Ile/γPro674Ser) are located outside the domain structure (self-dimerization, Pleckstrin-homology and SH2 domain; [34]), which is relevant for the function of SH2B1. We analyzed the impact of the variant rare allele of βThr656Ile/γPro674Ser on leptin signalling in vitro via the STAT3 pathway. In both splice variants, β656Ile or γ674Ser showed unaltered leptin signalling (Figure 2, Additional file 1: Table S2).

While the γPro674Ser exchange in the γ splice variant is predicted to be neutral by in silico programs, the exchange of βThr656Ile in the β splice variant was predicted to be “not neutral” (SNAP), “pathological” (PMUT) or “disease causing” (Mutation taster) in three of four programs (Additional file 1: Table S1). The exchange in the β splice variant would also destroy a predicted O-glycosylation site of the SH2B1 protein. Amino acid conservation was strong on both positions (86% for βThr656Ile and 100% for γPro674Ser, Additional file 1: Table S1).

Previous in vitro data showed functional differences between β and γ SH2B1 variants: (a) It has been shown that the β splice variant is mainly expressed in the hypothalamus [16], a brain region known to be implicated in weight regulation. A Sh2b1β rescue was sufficient to prevent the Sh2b1 knockout phenotype in mice [19]. Leptin signalling is mediated by the interaction of SH2B1β and JAK2 [17]. The β splice variant of SH2B1 recruits insulin receptor substrates 1 and 2 (IRS1 and 2) to the LEPRb/JAK2 complex [38]. SH2B1β enhances JAK2 activity and promotes the activation of several downstream networks like STAT3 and phosphatidylinositol (PI) 3-kinase pathways [18, 39]. Two in silico analyzes predicted an altered folding or function of SH2B1 βThr656Ile (Additional file 1: Table S1). (b) The γ splice variant of SH2B1 is peripherally expressed. It interacts with Tyr1158 in the activation loop of the insulin receptor and prohibits dephosphorylation of IRS1 and IRS2 [40]. This interaction enhances insulin signalling and insulin receptor auto-phosphorylation, leading in turn to activation of downstream pathways [41]. While the three tyrosine motifs in the N-terminal part of SH2B1, which regulate interaction with the insulin receptor [42], are not directly affected by this exchange, it is possible that altered protein folding due to an non-conservative amino acid exchange in a highly conserved position prevents their phosphorylation.

Coding GWAS derived SNP rs7498665

We confirmed that the described risk allele G at SNP rs7498665 [2, 3] is associated with obesity in our 705 obesity trios (p = 0.009) and in a total of 3,139 independent cases and 434 controls (p = 0.007, odds ratio (OR) = 1.22, 95% confidence interval (CI) 1.06-1.42; Table 1). This coding SNP results in the non-synonymous, non-conservative amino acid exchange Thr484Ala in a slpice variant independent position with low conservation (5%, Additional file 1: Table S1). As the association with obesity was previously described for this SNP, we did not analyze further study groups (2–6). In vitro analyses revealed unaltered leptin signalling via STAT3 in both splice variants for the obesity risk allele at Thr484Ala (Figure 2). Both Emax and EC50 were non-significantly reduced (Figure 2, Additional file 1: Table S2).

Since the obesity risk allele at the GWAS derived polymorphism rs7498665 increases BMI by only approximately 0.15 BMI units (kg/m²) as calculated in a population of 125,931 European individuals [1], we expected only subtle functional alterations associated with the minor allele of this variant.

Other genetic variants in SH2B1

The second newly detected mutation is located in the 3’ UTR at base pair position g.10182 (C/A). This non-coding variant was detected twice within the screening sample, and once in an obese case in the association testing step (Figure 1). The variant showed no association to obesity in a small case control comparison (p = 0.49; Table 2); in silico analyzes predicted a possible change in splice sites for this variant (Table 3).

Results for the four other identified SNPs were as follows: The third coding SNP rs147094247 leads to a non-synonymous, conservative exchange (Thr175Asp) at a conserved position (71%, Additional file 1: Table S1). No association to obesity (p = 0.199, odds ratio (OR) = 4.4, 95% confidence interval (CI) 0.57 - 34.13; Table 1) was found for this SNP in a sample of 11,268 obese and overweight cases and 4,512 lean or normal weight controls (mostly population based). For the non-synonymous SNP rs147094247 (Thr175Asp), in silico analyzes predict a neutral outcome for the altered amino acid (Additional file 1: Table S1).

For SNPs rs60604881 (p = 0.323, OR = 1.17, 95% CI = 0.87-1.58), rs62037368 (p = 1.00, OR = 1.45, 95% CI = 0.24-8.71) and rs62037369 (p = 0.107, OR = 1.29, 95% CI = 0.95-1.74) evidence for association to early onset extreme obesity could not be detected (Table 2). For all SNPs in silico analyzes predict splice site or transcription factor binding site changes (Table 3).

Leptin signalling

The assay that measured STAT3 mediated leptin response successfully showed increased leptin response after co-transfection with wild type SH2B1 splice variants β and γ. This indicates that HEK293 cells and the STAT3 assay allow functional characterization of SH2B1. While in mice only the alpha splice variant was tested for leptin signalling [43], we observe an effect on leptin signalling for both other splice variants (β and γ) in our human cell system (Figure 2, Additional file 1: Table S2). The analysis of the impact of both SH2B1 variants on leptin receptor activity showed no significant reduction of STAT3 mediated signalling by the risk alleles at rs7498665 and βThr656Ile/γPro674Ser. The non-significant decrease in EC50 and Emax for both tested variants in splice variants β and γ could indicate both gain of function and reduced function; the biological impact of both remains to be solved. If indeed a minor functional effect would be present, a much larger number of replicates would be necessary to establish a significant effect (e.g. about 2x270 replicates when using the Emax point estimates of SH2B1γ vs. SH2B1γP674S and their variances when applying Satterthwaite/Welch t-Test aiming at 80% power for two-sided α = 5%). Our results could, of course, indicate that the two variants are not functionally relevant. However, for a polygenic variant a large functional effect is rather unlikely. For example, the melanocortin 4 receptor gene (MC4R), a well known obesity gene, harbors two polymorphisms (Val103Ile and Ile251Leu) that are negatively associated with obesity [44, 45]. Carriers of the minor alleles have a BMI approx. 0.5 BMI units lower than wild type carriers [44, 45]. Initial in vitro assays did not show functional implications for the minor alleles of these SNPs (e.g. [44]), but when the number of different assays was increased, in vitro tests showed potential small gain of function for both minor alleles [46], which could explain the weight lowering effect of the variant. Hence we speculate that the functional effect of the analyzed SH2B1 variants might become detectable when a battery of different functional tests is applied. Currently we have first hints that both variants are compatible with a slightly reduced function. In addition, with STAT3 mediated leptin signalling, we only tested one of the many potential interaction partners of SH2B1 in regulation of energy homeostasis. A potential additive effect of small functional changes in leptinergic and insulinergic signalling could result in stronger impact on body weight maintenance.