C L I N I C A L R E S E A R C H A R T I C L E A Genome-Wide Pharmacogenetic Study of Growth Hormone Responsiveness Andrew Dauber,1 Yan Meng,2 Laura Audi,3 Sailaja Vedantam,2 Benjamin Weaver,2 Antonio Carrascosa,3 Kerstin Albertsson-Wikland,4 Michael B. Ranke,5 Alexander A. L. Jorge,6 Jose Cara,7 Michael P. Wajnrajch,7 Anders Lindberg,8 Cecilia Camacho-Hübner,7 and Joel N. Hirschhorn2 1 Division of Endocrinology, Children’s National Hospital, Washington, DC 20010; 2Division of Downloaded from https://academic.oup.com/jcem/article/105/10/3203/5870346 by guest on 04 May 2021 Endocrinology, Boston Children’s Hospital, and Program in Medical and Population Genetics, Broad Institute, Harvard Medical School, Boston, Massachusetts 02115; 3Department of Pediatrics, Institut de Recerca (VHIR), Hospital Vall d’Hebron, Centre for Biomedical Research on Rare Diseases (CIBERER), Autonomous University, Barcelona, Spain; 4Department of Physiology/Endocrinology, Institute of Neuroscience and Physiology, Sahlgrenska Academy, University of Gothenburg, Gothenburg, Sweden; 5University Children´s Hospital, Paediatric Endocrinology, Tübingen, Germany; 6Unidade de Endocrinologia do Desenvolvimento (LIM42), Hospital das Clinicas da Faculdade de Medicina da Universidade de Sao Paulo, Sao Paulo, Brazil; 7Pfizer Inc, Rare Disease, New York 10017; and 8Pfizer, Data Management, Sollentuna, Sweden ORCiD numbers: 0000-0003-4890-0262 (A. Dauber); 0000-0002-8211-9237 (M. P. Wajnrajch). Context: Individual patients vary in their response to growth hormone (GH). No large-scale genome-wide studies have looked for genetic predictors of GH responsiveness. Objective: To identify genetic variants associated with GH responsiveness. Design: Genome-wide association study (GWAS). Setting: Cohorts from multiple academic centers and a clinical trial. Patients: A total of 614 individuals from 5 short stature cohorts receiving GH: 297 with idiopathic short stature, 276 with isolated GH deficiency, and 65 born small for gestational age. Intervention: Association of more than 2 million variants was tested. Main Outcome Measures: Primary analysis: individual single nucleotide polymorphism (SNP) association with first-year change in height standard deviation scores. Secondary analyses: SNP associations in clinical subgroups adjusted for clinical variables; association of polygenic score calculated from 697 genome-wide significant height SNPs with GH responsiveness. Results: No common variant associations reached genome-wide significance in the primary analysis. The strongest suggestive signals were found near the B4GALT4 and TBCE genes. After meta-analysis including replication data, signals at several loci reached or retained genome- wide significance in secondary analyses, including variants near ST3GAL6. There was no significant association with variants previously reported to be associated with GH response nor with a polygenic predicted height score. ISSN Print 0021-972X ISSN Online 1945-7197 Printed in USA © Endocrine Society 2020. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits un- Abbreviations: beta4GalT, beta-1,4-galactosyltransferase; GH, growth hormone; restricted reuse, distribution, and reproduction in any medium, provided the original GHD, GH deficiency; GWAS, genome-wide association study; HE, HumanExome; HO, work is properly cited. Illumina HumanOmni1-Quad BeadChip; HOEE, HumanOmniExpressExome; IGF, insulin- Received 20 March 2020. Accepted 9 July 2020. like growth factor; ISS, idiopathic short stature; MAF, minor allele frequency; MPH, First Published Online 11 July 2020. midparental height; rhGH, recombinant GH; SDS, standard deviation score; SGA, small Corrected and Typeset 25 August 2020. for gestational age; SNP, single nucleotide polymorphism. doi:10.1210/clinem/dgaa443 J Clin Endocrinol Metab, October 2020, 105(10):3203–3214   https://academic.oup.com/jcem   3203 3204  Dauber et al    GWAS of Growth Hormone Responsiveness J Clin Endocrinol Metab, October 2020, 105(10):3203–3214 Conclusions: We performed the largest GWAS of GH responsiveness to date. We identified 2 loci with a suggestive effect on GH responsiveness in our primary analysis and several genome- wide significant associations in secondary analyses that require further replication. Our results are consistent with a polygenic component to GH responsiveness, likely distinct from the genetic regulators of adult height. (J Clin Endocrinol Metab 105: 3203–3214, 2020) Key Words: growth hormone, pharmacogenetics, short stature, genome-wide association G rowth disorders resulting in short stature have therapy in children with short stature. However, this numerous etiologies, including many different association has only been seen in small samples and inherited clinical syndromes. Specific genetic etiolo- has not been consistently replicated (8-12). Similarly, a gies include disorders of the growth hormone–insulin- polymorphism in the IGFBP-3 promoter has been asso- Downloaded from https://academic.oup.com/jcem/article/105/10/3203/5870346 by guest on 04 May 2021 like growth factor-1 (GH/IGF-1) axis—GH deficiency ciated with GH response, but this has not been robustly (GHD) and hormone resistance—and a wide variety of replicated (13, 14). The PREDICT study was designed additional genetic syndromes such as Turner syndrome, to take a more comprehensive approach to examining Noonan syndrome, and skeletal dysplasias (1-3). Other genetic predictors of GH response (15). In that study, children with short stature are born small for gesta- a more extensive candidate gene study was performed: tional age (SGA) and do not have “catch-up” growth. 170 patients (110 with GHD and 60 with Turner Finally, the majority of children with short stature have syndrome) underwent genotyping of approximately no known specific cause and are classified as idiopathic 1500 single nucleotide polymorphisms (SNPs) in a list short stature (ISS). of 103 candidate genes related to growth. Additionally, Recombinant human GH has been used in pediatric baseline gene expression analysis was performed in a populations to increase height for individuals with many subset of patients. The genotypes and gene expression of these conditions. There is substantial variability in profiles were then correlated with first-year GH re- growth response to GH, not only across types of growth sponse. The study identified a total of 11 genes associ- disorders but also between individuals who share the ated with GH responsiveness in patients with GHD and same general etiology for their short stature (4-6). The 10 genes in patients with Turner syndrome. However, variable response seen between different disorders sug- a subsequent replication study was performed by the gests that the sensitivity to treatment is associated with same research group in which none of the original SNP the underlying cause of each condition. However, the signals were reliably replicated (16). There was some additional variability within each disorder indicates that suggestive evidence for signals in the SOS1 and INPPL1 individual factors, including possibly genetic variation, genes. Ultimately, they concluded that genotype infor- also contribute to GH responsiveness. mation did not add much to the predictive capability of Previously, disease-specific models have been de- clinical parameters in determining GH responsiveness. veloped to predict a patient’s response to GH. These The PREDICT study did identify gene expression pro- models include multiple clinical variables such as par- files associated with both growth response (15) as well ental heights, age at GH initiation, degree of short as with increase in IGF-1 levels at 1 month of treatment stature, peak GH levels on stimulation testing, GH (17). Using bioinformatics techniques, they were able dose, and birth parameters (7). However, these clinical to identify pathways, such as glucocorticoid response, parameters only partially predict variation in treat- which are potentially involved in GH responsiveness. ment response. Given the inclusion of parental height To date, there have be no large genome-wide asso- (a proxy for genetic determinants of height) in predic- ciation studies (GWASs) of GH responsiveness, which tion models, and the high heritability of height, it is would provide a comprehensive assessment of the role quite plausible that genetic variation could influence of common genetic variation in GH response. Recent response to GH. Furthermore, identifying genetic vari- GWASs of adult height have identified hundreds of as- ants that influence response could shed light on the sociated common variants (18) and, as suggested by biological mechanisms underlying the variability in re- the correlation between parental height and GH re- sponse to GH therapy. sponse, these may also be candidates for influencing Several studies have proposed candidate variants in GH response and as such merit additional atten- single genes within the GH/IGF-1 axis for influencing tion. Finally, rarer genetic variants influence height, GH response. The most widely studied of these is a including those responsible for single-gene disorders polymorphism in GHR that deletes exon 3 (GHR- (19, 20), and these may also influence GH response. d3), which has been associated with response to GH For example, some genetic syndromes (such as short doi:10.1210/clinem/dgaa443 https://academic.oup.com/jcem  3205 stature homeobox-containing gene deficiency (19)) are between height SDS at the end of 1 year of treatment characterized by better responses to GH than others and the height SDS at start of GH therapy. (such as spondyloepiphyseal dysplasia (20)), consistent Patients with known chromosomal disorders or with the idea that genetic factors may influence GH other significant comorbid conditions were excluded response. from the study. We stratified samples by diagnosis. To search for genetic variation that affects response Patients with a peak GH value below 10 ng/mL were to GH, we carried out what is to our knowledge the lar- classified as GHD. Due to the nature of the cohort, gest GWAS of response to GH, consisting of a total of GH levels were not assayed centrally but rather were 614 children including 276 with idiopathic GHD, 297 based on each center’s reported values. Patients were children with ISS, and 41 children born SGA (with nei- excluded if they had additional pituitary hormone de- ther GHD nor ISS). We identify several suggestive can- ficiencies. Patients who had a birth weight SDS below didate polymorphisms for influencing GH response and –2 were classified as SGA. Patients could be assigned Downloaded from https://academic.oup.com/jcem/article/105/10/3203/5870346 by guest on 04 May 2021 also examine the role of height-associated variants on to both the SGA and GHD categories. Patients who GH response. Finally, we observe no evidence to sup- did not meet either of these criteria were classified port the previously reported association between the as ISS. All female patients were required to remain GHR-d3 or IGFBP-3 promoter polymorphisms and GH prepubertal throughout the study period, which was response. defined as Tanner stage 1 breast development. Male participants were required to have testicular volumes ≤6 cc or be documented as Tanner stage ≤2 at study Participants and Methods completion. The clinical characteristics of the patients Description of study samples are summarized in Table 1. The study included 614 children who were treated with GH on the basis of the diagnosis of GHD, SGA, or Genotyping and quality control ISS. Individuals were recruited from 5 cohorts including Samples were genotyped using either a combination the Boston Children’s Hospital short stature cohort, of the Illumina HumanOmni1-Quad BeadChip (HO) Barcelona, Sweden, a Pfizer USA clinical trial for pa- and HumanExome (HE) BeadChip platforms, or on the tients with ISS, and individuals who were previously HumanOmniExpressExome (HOEE) BeadChip v1.0 or collected by KIGS investigators (Pfizer International v1.2 platforms; all of these platforms cover a similar Growth Database). The majority (504) of these samples genome-wide set of common variants and a more com- are of self-described European ancestry. A replication prehensive set of exonic variants. We used Illumina’s sample from Brazil consisted of 113 individuals treated GenCall algorithm to generate genotype calls for all with GH for GHD, SGA, or ISS. platforms. For the rarer exonic variants, we generated The study was approved by the Institutional a second set of genotype calls using zCall (24), which is Review Board of Boston Children’s Hospital as well specifically designed for calling rare SNPs from array- as by local review boards in Barcelona, Sweden, based platforms. Pfizer USA, and KIGS (Brazil). Written informed con- Quality control was performed separately for each sent was obtained from all patients’ parents or legal of the platforms (HO, HE, and HOEE). The minimum guardians. overall call rate for passing samples was 95%. Samples All clinical information was abstracted from the were also removed for being outliers for amount of available medical records/research databases at the heterozygosity, mismatch between reported gender respective centers. For all patients, we collected in- and genotypic data, or close relationship (estimated formation on birth weight, gestational age, peak GH genomic identity by descent ≥0.2) with other sam- value during pretreatment testing for GHD, age and ples. Self-reported ancestry and gender were compared height at GH start, average GH dose, parents’ heights, with ancestry and sex directly estimated from the gen- and height at the end of 1 year of GH therapy. Height etic data; discordant samples were removed. Standard standard deviation scores (SDSs) were calculated by additional filters were applied to remove poorly geno- adjusting for age and gender according to Prader et al typed SNPs, including variant call rate <95%, Hardy- (21). Birth weight SDSs were calculated according to Weinberg equilibrium P value < 10–6, and evidence of Niklasson et al (22). The midparental height (MPH) nonrandom missingness or batch effects. In total, 588 SDS was calculated by the following: (father’s height individuals had genotype data passing quality control SD score + mother’s height SD score) ÷ 1.61 (23). and the required phenotypic information for associ- Height gain (delta height) was defined by the difference ation analysis. 3206  Dauber et al    GWAS of Growth Hormone Responsiveness J Clin Endocrinol Metab, October 2020, 105(10):3203–3214 Table 1. Clinical Characteristics of Participants models: a minimally adjusted model (age and gender in- cluded as covariates) and a maximally adjusted model GWAS Replication (age; gender; gestational age; birth weight SDS; age at Cohorts Cohort start of GH therapy; GH dosage; height SDS at start of Individuals 614 113 GH therapy; MPH; and whether the patient has mild or Male 437 73 Female 177 40 severe GHD, defined as a peak GH level <5, included as Gestational age 38.6 (2.9) 37.2 (3.1) covariates). In addition, principal components (29) were (weeks) included as covariates to account for ancestry; for the Birth weight (g) 2969 (693) 2754 (809) Birth weight SDS –0.70 (1.14) –0.60 (1.48) multiethnic analysis, 10 principal components were in- Midparental height 164.6 (5.8) 161.9 (5.35) cluded; for European samples, 3 principal components (cm) were included. The primary analysis was considered to Age at start (years) 8.14 (2.71) 8.31 (2.55) be all ancestries, all diagnoses, and a minimally adjusted Downloaded from https://academic.oup.com/jcem/article/105/10/3203/5870346 by guest on 04 May 2021 Average growth 0.28 (0.12) 0.29 (0.06) hormone dose model (Fig. 1). (mg/kg/week) To account for the potential confounding effect due Height at start (Prader –3.00 (0.72) –3.77 (1.18) to different genotyping platforms, the analyses were per- SDS) Δ height SDS during 0.78 (0.40) 0.76 (0.72) formed separately for the Pfizer USA samples genotyped first year of therapy on HO/HE and all other samples genotyped on HOEE, ISS 297 42 and results were meta-analyzed. For meta-analysis, GHD 276 44 SGA 65 30 we used the inverse-variance fixed effects method to SGA & GHD 24 3 combine the results, using METAL (30), weighed by The numbers in parens are SD, standard deviations. The “SDS” listed standard error. Genome-wide significance for our pri- for Birth Weight SDS, Prader SDS and Height SDS are standard devi- mary analysis was set at P < 5 × 10-8. ation scores. Abbreviations: GHD: growth hormone deficiency; GWAS, genome- wide association study; ISS, idiopathic short stature; SDS: standard Association analysis of low-frequency coding deviations above or below the mean; SGA: small for gestational age; variants SGA_SDS, standard deviation score for being small for gestational age. For rarer (minor allele frequency <5%) coding vari- ants represented on the exome chip, 2 sets of ana- Genotype imputation lyses were performed using rare variant tests (31). We To expand the coverage of the genotyping platforms performed association analysis of each variant indi- to include the vast majority of common variants, we vidually (single-variant analysis) and aggregate ana- imputed genotypes for all polymorphic markers in the lysis of missense, splice, and loss-of-function variants 1000 Genomes Phase I Integrated Release Version 3 within each gene (gene-based analysis). Gene-based (March 2012) (25), using the program IMPUTE2 (26, analyses were performed with sequence kernel asso- 27), thereby generating datasets of 36 388 217 auto- ciation tests (32), and meta-analysis of results from somal and 1 250 157 sex chromosomal variants. The the 2 genotyping platforms were performed using data from HO and HE platforms were merged and RAREMETAL (33). then imputed; the remaining samples genotyped on the For single-variant exome analysis, we set the signifi- HOEE platform were imputed separately. We removed cance threshold for single-variant association analysis SNPs with a low quality of imputation (info score <0.4). as P < 6.93 × 10−7 (Bonferroni correction for 72 171 polymorphic variants) for analysis of samples with GH Association analysis responsiveness. The significance threshold for a gene- The primary outcome was the change in height SDS; based test association as P < 2.5 × 10−6 (Bonferroni cor- we performed an inverse normal transformation to en- rection for 20 000 genes). The analyses were carried out sure normality of the outcome variable, but results were under the same phenotypic models as for GWAS. similar if this transformation was omitted. We carried out association testing between this outcome and SNP Association analysis of previously reported height- genotype (coded as allele dosage between 0 and 2), using associated variants an additive model in a regression framework imple- For each individual, we calculated a genetic poly- mented in PLINK (28). We performed analyses stratified gene score (34) as implemented in PLINK (28), using by diagnosis (all diagnoses, ISS only, and GHD only; 697 height-associated loci identified by the GIANT con- the number of SGA samples was too small for reliable sortium (18). Specifically, the reported regression coeffi- analyses) and by ancestry (all ancestries and European cient of each SNP on height and the individual genotype ancestry only). For each analysis, we used 2 different at each SNP is used to create a weighted polygene score doi:10.1210/clinem/dgaa443 https://academic.oup.com/jcem  3207 GHD SGA ISS N=276 N=41 N=297 All GHD ISS N=614 Primary analysis: All Secondary ancestries, minimally Secondary Secondary analyses adjusted model analyses analyses Downloaded from https://academic.oup.com/jcem/article/105/10/3203/5870346 by guest on 04 May 2021 Associaon with GH response Validate and replicate suggesve findings Figure 1. Analysis plan for testing association to growth hormone response. The primary analysis was to combine patients in all 3 categories: growth hormone deficiency (GHD), small for gestational age (SGA) and idiopathic short stature (ISS), including all ancestries and performing a minimally adjusted analysis (age, sex, and principal components of genetic ancestry). Secondary analyses included testing GHD or ISS alone, limiting the analyses to individuals of European ancestry and adjusting for additional covariates. Additional covariates were gestational age in weeks, birthweight standard deviation score (SDS), age at GH start, average GH dose, height SDS (Prader) before treatment, average height of parents, and severity of GHD. For principal components of ancestry, the top 10 eigenvalues were used for all ancestry analyses and the top 3 eigenvalues for analyses of European ancestry samples. For single-variant analysis, associations were considered to have genome-wide significance if they reached P < 5 × 10–8 in the primary analysis; suggestive associations were those that reached P < 5 × 10–7 in the primary or secondary analyses. for each individual. We calculated the variance in GH samples to confirm good agreement with the original response explained by this polygene score (adjusted R2 GWAS data before proceeding to genotyping. In total, estimated from a linear regression model incorporating we obtained genotype data in the Brazilian replication the score as the predictor and the normalized height sample for 26 of these 37 SNPs; 5 of the 28 SNPs were SDS as an outcome, adjusting for age at GH initiation monomorphic in the replication sample and were not and gender). analyzed further. Genotyping was performed using the Sequenom MassArray platform as previously described Association analysis of GHR-d3 and IGFBP3 (35) or, for a subset of variants that could not be geno- variants typed successfully, by Sanger sequencing. We tested for association with variants previously reported to be associated with GH response, using the Results methods described above. For the GHR-d3 variant (10), we used as a proxy the SNP rs6873545 (r2 between To search for genetic determinants of response to GH, GHR-d3 and rs6873545 = 1 in the HapMap CEU we studied 614 children treated with GH and assem- European ancestry sample); we imputed the previously bled from several studies (Table 1; see Methods). The reported IGFBP3 rs2854744 variant. participants were categorized as having a diagnosis of GHD (N = 276), SGA (N = 41), or ISS (N = 297). The Validation and replication of variants with majority (N = 504) of these samples were of European suggestive associations to GH response ancestry. The primary outcome in our study was the in- We used PLINK to “clump” variants into loci based crement in height SDS during the first year of therapy. on linkage disequilibrium (r2 > 0.5), with a single lead We first performed a GWAS and imputed (26, 27) geno- variant in each locus selected based on the best P value. types for all polymorphic variants with frequency >1% We selected for additional genotyping 3 variants that in the 1000 Genomes Phase I Integrated Release Version had associations P < 5 × 10-7 in the primary analysis and 3 (March 2012) (25). Our primary analysis included all 5 additional variants with P < 5 × 10-8 in any secondary samples, regardless of diagnosis, and was adjusted for analysis, as well as an additional 29 variants with as- age, sex, and principal components of genetic ancestry sociations P < 5 × 10-7 in any secondary analysis. We (Fig. 1; see Methods). We also performed secondary ana- performed Sanger sequencing in a subset of the GWAS lyses stratified by ancestry and diagnosis, adjusting for 3208  Dauber et al    GWAS of Growth Hormone Responsiveness J Clin Endocrinol Metab, October 2020, 105(10):3203–3214 a more extensive set of covariates (maximally adjusted; and ST3GAL6 showed its strongest association in the see Methods). The distributions of test statistics for each analysis of individuals with ISS of all ancestries under analysis were not significantly inflated: genomic infla- the minimally adjusted model (P = 6.95 × 10–9). An tion correction factors (λ GC) ranged between 1.004 and intergenic SNP (rs78263566, frequency = 2.5%) be- 1.021, indicating minimal systematic bias (Fig. 2). tween CLEC7A and OLR1 and an intronic SNP (rs74523128, frequency = 1.0%) in LAPTM4B showed GWAS analysis of common variants with GH strongest associations in the analyses of individuals with response GHD under the maximally adjusted model in European We first tested common variants with minor allele ancestry individuals (P = 8.5 × 10–9 and P = 1.5 × 10–8). frequency (MAF) above 1% for association with GH Finally, a low-frequency intronic SNP (rs144751704, response. In our primary analysis (Fig. 1), no vari- MAF = 1.4%) located in the NT5DC1 gene and up- ants achieved genome-wide significance (P < 5 × 10-8; stream of COL10A1 showed the strongest association Downloaded from https://academic.oup.com/jcem/article/105/10/3203/5870346 by guest on 04 May 2021 Table 2). The strongest suggestive (P < 5 × 10–7) associ- to GH response in the analysis of all European an- ations with GH response were observed with a common cestry individuals under the minimally adjusted model variant near the B4GALT4 gene: rs7628585 (intronic, (P = 2.1 × 10–8). We also identified 29 additional vari- MAF = 0.182; P = 5.54 × 10–8); (Fig. 3) and with a ants with P < 5 × 10–7 in secondary analyses, a level of common SNP in the TBCE gene, rs1977748 (intronic, significance suggestive of association. MAF = 0.411; P = 3.75 × 10–7). In secondary analyses, 5 SNPs had association P Attempted replication of potential signals of values below 5 × 10–8 (Table 2); all of these had minor association with response to recombinant GH allele frequencies below 5%. In light of the sample size of We first focused on the 2 signals that reached sug- this study, these associations, especially those specific to gestive significance in our primary analysis: rs7628585 1 diagnosis, should be viewed as tentative in the absence near B4GALT4 and rs1977748 near TBCE. Neither of of additional replication. A low-frequency intronic SNP these signals were replicated nor reached genome-wide (rs55704135, MAF = 3%) in the UBE4B gene showed significance (Table 2). We also attempted to replicate the strongest association to GH response in the analysis the 5 signals that had reached genome-wide significance of all European ancestry individuals under the max- in a secondary analysis, meta-analyzing the discovery imally adjusted model (P = 4.2 × 10–9). An intergenic and replication samples using either the same model for SNP (rs189532746, frequency = 1.3%) between CPOX which we observed the best evidence for association in the discovery sample or using the primary analysis (min- imal covariates, all diagnoses) to maximize power. Two signals for which we were able to obtain replication data retained genome-wide significance after meta-analysis (rs55704135 near UBE4B; meta-analysis P = 2.65 × 10–8 in the European ancestry/all diagnoses/maximum covariate analysis, and rs78263566 near CLEC7A; meta-analysis P = 1.67 × 10–8 in the European ancestry/ GHD/maximum covariate analysis, Table 2). However, for both of these signals, the evidence for association de- creased after replication, so further validation of these associations would be needed. Two other genome-wide significant variants, (rs189532746, between CPOX and ST3GAL6, and rs78263566, between CLEC7A and OLR1) were not polymorphic in the replication ana- lysis, and we were unable to successfully genotype the fifth signal (rs144751704, located in NT5DC1 and up- stream of COL10A1) in the replication sample. Figure 2. Quantile-quantile (QQ) plot for results from the primary We also attempted to replicate 29 signals with sug- association analysis of common variants with growth hormone response. Each point represents the result from a single variant gestive signals in secondary analyses. Of the 19 signals (genotyped or imputed), ranked by observed P value. The x-axis that were successfully genotyped, 12 were polymorphic indicates the expected P value and the y-axis indicates the observed P in the replication sample, and 2 achieved genome- value, both in negative log scales. The gray line is the expected result under the null, and the red lines indicate 95% confidence intervals wide significance in meta-analysis with the replica- around the expectation. tion data when analyzed under the same model as was doi:10.1210/clinem/dgaa443 Table 2. Top Associations with GH response from Primary and Secondary GWAS Analyses GWAS Ancestries/ Effect GWAS SE, GWAS P Diagnosis/ Repli- Replica- Meta- Meta- Meta- Chromosome Minor Size, Best Best Ana- value, Best Model for Best Nearest Replication cation tion P analysis analysis analysis (position) SNP ID MAF allele Analysis lysis Analysis Analysis Gene(s) Beta SE Value Beta SE P Value Associations from primary analysis with P < 5 × 10–7 3 (118950499) rs7628585 0.182 A 0.338 0.0621 5.54E-08 Primary analysis B4GALT4 –0.091 0.181 6.17E-01 0.292 0.059 6.47E-07 1 (235531451) rs1977748 0.411 T –0.286 0.0563 3.75E-07 Primary analysis TBCE 0.001 0.159 9.96E-01 –0.254 0.053 1.72E-06 Associations from secondary analyses with P < 5 × 10–8 1 (10194439) rs55704135 0.03 A –0.934 0.159 4.24E-09 European/all/ UBE4B –0.231 0.308 4.54E-01 –0.786 0.141 2.65E-08 maximum 3 (98383646) rs189532746 0.013 A –1.635 0.282 6.95E-09 All/ISS/ CPOX, NP NP NP NP NP NP minimum ST3GAL6 12 (10300486) rs78263566 0.025 G 1.685 0.283 8.50E-09 European/ CLEC7A, 0.182 0.721 8.01E-01 1.485 0.2631 1.67E-08 GHD/ OLR1 maximum 8 (98853066) rs74523128 0.01 A 2.516 0.430 1.52E-08 European/ LAPTM4B ND ND ND ND ND ND GHD/ maximum 6 (116512418) rs144751704 0.014 T –1.299 0.232 2.12E-08 European/all/ NT5DC1/ NP NP NP NP NP NP minimum COL10A1 Chromosome and position are based on the hg37 build of the human genome. Best analysis refers to the ancestry/diagnosis/model with the best P value in the GWAS for that variant. Replication and meta-analysis results are shown for that best analysis. Abbreviations: GWAS, genome-wide association study; MAF, minor allele frequency; ND, not done; NP, not polymorphic; SNP, single nucleotide polymorphism. https://academic.oup.com/jcem  3209 Downloaded from https://academic.oup.com/jcem/article/105/10/3203/5870346 by guest on 04 May 2021 3210  Dauber et al    GWAS of Growth Hormone Responsiveness J Clin Endocrinol Metab, October 2020, 105(10):3203–3214 rs7628585 Plotted SNPs 10 r2 rs7628585 100 0.8 8 0.6 80 Recombination rate (cM/Mb) 0.4 0.2 - log10(p−value) 6 60 4 40 Downloaded from https://academic.oup.com/jcem/article/105/10/3203/5870346 by guest on 04 May 2021 2 20 0 0 IGSF11 UPK1B ARHGAP31 POGLUT1 ADPRH IGSF11−AS1 C3orf30 TMEM39A CD80 B4GALT4 TIMMDC1 PLA1A 118.6 118.8 119 119.2 Position on chr3 (Mb) Figure 3. LocusZoom plot for the B4GALT4 locus. The x-axis indicates position on chromosome 3 surrounding the lead signal of association near B4GALT4, rs7628585. Genes depicted in their respective locations. Each dot is a variant in the region tested and is colored by its correlation (r2) with rs7628585 according to the color key. Blue lines indicate recombination rates. The arrow and circle indicate the location and association signal of the second signal at B4GALT4, rs35583194; this insertion/deletion variant was not included in the LocusZoom plot because it is not present in the 1000 Genomes reference panel. SNP, single nucleotide polymorphism. highlighted by the discovery sample. These 2 signals were variants within each gene (see Methods), combining in- rs75922185 near MGAT5; meta-analysis P = 4.93 × 10–8 formation from the different variants in each gene using in the European ancestry/all diagnoses/maximum the sequence kernel association test (32), and limiting covariate analysis, and rs151058087 near AEBP2; analysis to variants below 1%. No genes were signifi- meta-analysis P = 1.67 × 10–8 in the all ancestry/ISS/ cant after correcting for multiple testing (see Methods). maximum covariate analysis (Online Supplementary A full set of association results for our primary analysis Table 1) (36). The variant near AEBP2 was the only for all variants with MAF above 1% and all directly one to achieve nominal significance in the replica- genotyped coding variants will be made available. tion sample. Of potential interest, rs115307564 near RBMS3, strongly replicated under the primary ana- GWAS analysis of variants previously associated lysis model (replication P = 7.96 × 10–4, meta-analysis with GH response or adult height P = 1.67 × 10–6; Online Supplementary Table 1) (36). We also focused specifically on candidate This variant therefore represents another promising polymorphisms previously associated with response signal of association. to GH, including the GHR d3 variant (10-12), which is represented in our study by the highly correlated Association analysis of rarer coding variation with SNP rs6873545 (37), as well as the rs2854744 poly- response to recombinant GH using exome array morphism in the IGFBP-3 promoter (13, 38). We con- genotypes firmed that the rs6873545 was a good proxy for the To determine the role of low-frequency coding vari- GHR d3 variant in our study (see Methods). We ob- ants, we also analyzed directly genotyped coding SNPs served no evidence at either of these variants for associ- present on the exome array platforms. In single-variant ation with response to GH (P > 0.05 under all models). analysis, no SNPs were significant after correcting for Thus, we could not replicate the previous associations. multiple testing (see Methods). Because single-variant We also tested a hypothesis that genetic predispos- tests are often underpowered to detect associations with ition to tall or short stature could influence response rare variants, we also carried out gene-based associ- to GH. For each individual, we constructed a genetic ation tests using missense, splice, and loss-of-function polygene score based on the previously established 697 doi:10.1210/clinem/dgaa443 https://academic.oup.com/jcem  3211 height variants and tested for association between this Another low-frequency variant, near RBMS3, was polygene score and the response to GH. We did not see the only one to show a strong signal in the replication even nominally significant evidence of association, sug- cohort primary analysis but did not reach genome- gesting that genetic control of final height and response wide significance. RBMS3 encodes a long noncoding to GH are not strongly connected. ribonucleic acid that may modulate signaling by trans- forming growth factor beta, a well-known modulator of growth; of note, the TGFBR2 gene is adjacent to Discussion RBMS3. This variant, as well as the variants described We report what is, to our knowledge, the largest GWAS in the previous paragraphs, represents signals that of response to GH to date. We leveraged the large would benefit from further validation to see if they are sample size to interrogate both a primary analysis (all robustly associated with GH response. samples, with a minimal set of covariates) as well as We also tested variants previously associated with re- Downloaded from https://academic.oup.com/jcem/article/105/10/3203/5870346 by guest on 04 May 2021 several secondary analyses, including testing for associ- sponse to recombinant GH (rhGH), including the exon ation in all samples and in European samples only, with 3 deletion variant in GHR (10-12) and the rs2854744 specific diagnoses (ISS and GHD), and adjusting for a and rs924140 variants in IGFBP3 (13, 38). We found larger set of known clinical covariates (gestational age, no evidence of association with response to rhGH for age at start of GH therapy, GH dosage, height SDS at any of these variants (all P > 0.05). One possibility why start of GH therapy, MPH). we failed to see an association in our large sample is that We detected 37 suggestive associations (P < 5 × 10–7) the findings from the prior studies are false-positives. We from this GWAS, including 3 suggestive associations note that our study population is a heterogeneous col- from our primary analysis and 5 that had P values lection of patients with short stature, so differences in below 5 × 10–8 in at least 1 secondary analysis. The patient population could also explain these discrepant strongest of the associations in the primary analysis was findings. However, we also saw no evidence of associ- with common variation near the B4GALT4 and TBCE ation when we restricted our analysis to patients with genes. However, despite these strongly suggestive genetic ISS or to those with GHD. data, we did not observe associations with GH response We also tested whether the 697 height-associated in our replication sample. Possible explanations for the loci are associated with response to rhGH. We found lack of replication include lack of power in the replica- no strong association with GH response for any height- tion sample, differences in ethnicity (and hence different associated loci, either individually or jointly. One pos- levels of linkage disequilibrium with an untyped causal sible explanation is that the GH response is regulated variant), or that both of these suggestive associations by different biological mechanisms than for height; are statistical fluctuations rather than true associations. another possible explanation is that variation in GH Interestingly, the B4GALTT4 gene is 1 of 7 beta-1,4- response is associated with variants that are rare and galactosyltransferase (beta4GalT) genes, and mutations have not been identified from previous GWAS analyses. in B4GALT7 lead to abnormal skeletal growth (39). Comprehensive sequencing of samples with data on GH However, mutations in B4GALT4 have not yet been response could help address these possibilities. For ex- reported to be associated with any specific disease. Of ample, if a fraction of patients treated with GH have interest, a low-frequency variant that reached genome- rare, nearly mendelian causes of short stature, then the wide significance in a secondary analysis is located near variable response to GH could be partly dictated by the ST3GAL6, a sialyltransferase that acts in the same variable underlying etiologies in these patients. pathway as the beta4GalT genes. The frequency of this Limitations of our study include the collection of variant was too low to robustly perform replication. samples from various locations in the world with Of the remaining variants that we were able to assay various ethnic backgrounds, and a mixture of diag- in the replication sample (see Methods), 4 signals had noses. We addressed this heterogeneity by performing meta-analysis P values that reached or retained genome- secondary analyses of more homogeneous, if smaller, wide significance in secondary analysis (near UBE4B, subgroups of individuals. The sample size is larger CLEC7A, MGAT5, and AEBP2). Remarkably, MGAT5 than previous studies but still quite small as compared also encodes a glycosyltransferase, although not ob- with many GWAS studies and, therefore, has limited viously in the same pathway as the beta4GalT genes. power to detect variants associated with GH response. Thus, our results provide an intriguing suggestion that A sample size of 614 should have 80% power to detect variation in glycosylation pathways may regulate the re- associations that explain, in our heterogeneous popu- sponse to GH. lation, 6.5% of the variance in GH response. Although 3212  Dauber et al    GWAS of Growth Hormone Responsiveness J Clin Endocrinol Metab, October 2020, 105(10):3203–3214 effect sizes this large are unusual for polygenic traits, Spanish patients were recruited in Hospital Universitari large effect sizes have been seen in other pharmacogen- Vall d’Hebron and Vall d’Hebron Institut de Recerca (VHIR). omics scenarios. Thus, any heritable contribution to Swedish patients were recruited from clinical trial TRN GH response is likely polygenic. While our work shows 88-080, which was initiated in 1988 as a Kabi-sponsored study and changed to an Investigator-sponsored study in the 1990s, that there are likely no common variants that on their with K.A.W. as sponsor. The principle investigator was Kerstin own have clinically significant predictive power, larger Albertsson-Wikland and co-investigators included: A. Stefan sample sizes would be needed to identify variants with Aronsson, Jan Gustafsson, Lars Hagenäs, Sten A. Ivarsson, smaller effects that in combination, as part of a poly- Berit Kriström, Claude Marcus, Karl Olof Nilsson, Martin genic risk score, could still provide meaningful predic- Ritzén, Torsten Tuvemo, Otto Westphal, Jan Åman. tion of the response to GH. KIGS Investigators: Jean de Schepper (country coordinator), Our study only examined the relationship between Vrije Universiteit Brussels (VUB), Brussels; Marine Cools, Ghent genotype and growth response. As noted earlier, other University, Ghent; Jan Lebl, Motol University Hospital, Prague; Downloaded from https://academic.oup.com/jcem/article/105/10/3203/5870346 by guest on 04 May 2021 approaches, such as gene expression profiling, can look Jana Černá, FN Ostrava State Hospital, Ostrava; Michael at other molecular predictors of GH response, and these B. Ranke and Gerhard Binder, University Children’s Hospital, alternate approaches have been reviewed elsewhere (40). Pediatric Endocrinology and Diabetology, Tübingen; Helmuth For example, 1 study looked at the methylation status of Dörr, University Children’s Hospital, Pediatric Endocrinology and Diabetology, Erlangen; Berthold Hauffa, Paediatric the IGF-1 promoter and found that approximately 25% Endocrinology, Department Paediatrics, University Hospital of of the first-year GH response may be attributable to Essen, Essen; Heinrich Schmidt, Dr von Haunersches Kinderspital, differences in methylation at the P2 promoter of IGF-1 München; Franziska Lorenzen, Pediatric Endocrinology Practice, (41). Additionally, while our study focused on first-year Hannover; Roland Pfäffle, University of Leipzig Hospitals and height response, consideration must be given to other Clinics, Leipzig; Klaus Mohnike, Otto von Guericke University definitions of GH responsiveness including biochemical, Magdeburg, Magdeburg; Lohtar Reinken, Evangelisches transcriptomic, and proteomic responses, which may Krankenhaus, Hamm; Bernhard Köster, Klinikum, Lüdenscheid; offer more mechanistic insights into GH biology. Ludmila Kostalova, Comenius University in Bratislava and In summary, we have completed the largest genetic University Children’s Hospital Bratislava, Limbova 1, Bratislava. study to date of response to GH. We were unable to replicate previous associations, nor do we identify any Additional Information new variants that are clearly and robustly associated with GH response. However, some associations reached Correspondence and Reprint Requests: Michael Wajnrajch, genome-wide significance in secondary analyses and MD MPA, Endocrine Care & Inborn Errors of Metabolism, merit further investigation, and our data collectively Pfizer Inc, 235 East 42nd Street, MS 235-10-01, New York, NY raise the hypothesis that variation in glycosylation may 10017, USA. E-mail: Michael.Wajnrajch@Pfizer.com. contribute to variation in GH response. Larger sample Disclosure Summary: A.D. has received consulting fees from OPKO Biologics, Sandoz, Novo Nordisk, Ipsen, and Pfizer; sizes will be needed to more definitively identify any gen- A.A.L.J. received speaker fees from Sandoz and NovoNordisk; etic factors that robustly influence the response to GH. K.A.W. received an unrestricted grant from Pfizer; M.B.R. has received speaker fees from Novo Nordisk, Merck Serono, Sandoz and Mediagnost. J.C., M.P.W., and C.C.H. are em- Acknowledgments ployees of Pfizer Inc; A.L. was an employee of Pfizer Health This study was supported by Pfizer. AB; Y.M. is now an employee of Fulgent Genetics; J.N.H. re- Samples from patients with ISS came from Pfizer-sponsored ceived research support from Pfizer Inc for this study and is study A6281280, with the following investigators in the on the scientific advisory board of Camp4 Therapeutics; S.V., United States: Moris Angulo; Yong Bao; Deborah Bowlby; B.W., L.A., and A.C. have nothing to declare. Dennis Brenner; Rosalind Brown; Mauri Carakushansky; Data Availability: The datasets generated during and/or Joseph Cernich; Jay Cohen; Debra Counts; Paul Desrosiers; analyzed during the current study are not publicly available Oscar Escobar; Myrto Frangos; Graeme Frank; Sherry but are available from the corresponding author on reason- L. Franklin; Mitchell Geffner; Michele Hutchison; Stephen able request. Kemp; Lynne Levitsky; Revi Mathew; Sunil Nayak; Robin Nemery; Ron Newfield; Richard Noto; Leslie Plotnick; Teresa References Quattrin; Robert Rapaport; Edward Reiter; David Repaske; 1. Guo MH, Shen Y, Walvoord EC, et al. Whole exome sequencing Samuel Richton; Judith Ross; Deborah Rotenstein; Ellen to identify genetic causes of short stature. Horm Res Paediatr. Sher; Lawrence Silverman; Michael Tansey; Svetlana Ten; 2014;82(1):44-52. Paul Thornton; Sharon Travers; Nancy Wright; Ping Zhou; 2. David A, Hwa V, Metherell LA, et al. Evidence for a con- William Zipf. tinuum of genetic, phenotypic, and biochemical abnormalities doi:10.1210/clinem/dgaa443 https://academic.oup.com/jcem  3213 in children with growth hormone insensitivity. Endocr Rev. 18. Wood AR, Esko T, Yang J, et al.; Electronic Medical Records 2011;32(4):472-497. and Genomics (eMEMERGEGE) Consortium; MIGen 3. Seaver LH, Irons M; American College of Medical Genetics Consortium; PAGEGE Consortium; LifeLines Cohort Study. (ACMG) Professional Practice and Guidelines Committee. Defining the role of common variation in the genomic and ACMG practice guideline: genetic evaluation of short stature. biological architecture of adult human height. Nat Genet. Genet Med. 2009;11(6):465-470. 2014;46(11):1173-1186. 4. Richmond E, Rogol AD. Current indications for growth 19. Blum WF, Ross JL, Zimmermann AG, et al. GH treatment to final hormone therapy for children and adolescents. Endocr Dev. height produces similar height gains in patients with SHOX defi- 2010;18:92-108. ciency and Turner syndrome: results of a multicenter trial. J Clin 5. Rosenbloom AL. Recombinant human insulin-like growth factor Endocrinol Metab. 2013;98(8):E1383-E1392. I (rhIGF-I) and rhIGF-I/rhIGF-binding-protein-3: new growth 20. Kanazawa H, Tanaka H, Inoue M, Yamanaka Y, Namba N, treatment options? J Pediatr. 2007;150(1):7-11. Seino Y. Efficacy of growth hormone therapy for patients with 6. Ranke MB, Lindberg A, Mullis PE, et al. Towards optimal treat- skeletal dysplasia. J Bone Miner Metab. 2003;21(5):307-310. ment with growth hormone in short children and adolescents: 21. Prader A, Largo RH, Molinari L, Issler C. Physical growth of evidence and theses. Horm Res Paediatr. 2013;79(2):51-67. Swiss children from birth to 20 years of age. First Zurich lon- Downloaded from https://academic.oup.com/jcem/article/105/10/3203/5870346 by guest on 04 May 2021 7. Ranke MB, Lindberg A. Predicting growth in response to growth gitudinal study of growth and development. Helv Paediatr Acta hormone treatment. Growth Horm IGF Res. 2009;19(1):1-11. Suppl. 1989;52:1-125. 8. Wassenaar MJ, Dekkers OM, Pereira AM, et al. Impact of the 22. Niklasson A, Albertsson-Wikland K. Continuous growth refer- exon 3-deleted growth hormone (GH) receptor polymorphism on ence from 24th week of gestation to 24 months by gender. BMC baseline height and the growth response to recombinant human Pediatr. 2008;8:8. GH therapy in GH-deficient (GHD) and non-GHD children 23. Cole TJ. Some questions about how growth standards are used. with short stature: a systematic review and meta-analysis. J Clin Horm Res. 1996;45(Suppl 2):18-23. Endocrinol Metab. 2009;94(10):3721-3730. 24. Goldstein JI, Crenshaw A, Carey J, et al.; Swedish Schizophrenia 9. Carrascosa A, Esteban C, Espadero R, et al.; Spanish SGA Study Consortium; ARRA Autism Sequencing Consortium. zCall: Group. The d3/fl-growth hormone (GH) receptor polymorphism a rare variant caller for array-based genotyping: gen- does not influence the effect of GH treatment (66 microg/kg per etics and population analysis. Bioinformatics. 2012;28(19): day) or the spontaneous growth in short non-GH-deficient small- 2543-2545. for-gestational-age children: results from a two-year controlled 25. Auton A, Brooks LD, Durbin RM, et al.; 1000 Genomes Project prospective study in 170 Spanish patients. J Clin Endocrinol Consortium. A global reference for human genetic variation. Metab. 2006;91(9):3281-3286. Nature. 2015;526(7571):68-74. 10. Dos Santos C, Essioux L, Teinturier C, Tauber M, Goffin V, 26. Marchini J, Howie B, Myers S, McVean G, Donnelly P. A new Bougnères P. A common polymorphism of the growth hormone multipoint method for genome-wide association studies by im- receptor is associated with increased responsiveness to growth putation of genotypes. Nat Genet. 2007;39(7):906-913. hormone. Nat Genet. 2004;36(7):720-724. 27. Howie BN, Donnelly P, Marchini J. A flexible and accurate geno- 11. Binder G, Baur F, Schweizer R, Ranke MB. The d3-growth type imputation method for the next generation of genome-wide hormone (GH) receptor polymorphism is associated with in- association studies. PLoS Genet. 2009;5(6):e1000529. creased responsiveness to GH in Turner syndrome and short 28. Purcell S, Neale B, Todd-Brown K, et al. PLINK: a tool set for small-for-gestational-age children. J Clin Endocrinol Metab. whole-genome association and population-based linkage ana- 2006;91(2):659-664. lyses. Am J Hum Genet. 2007;81(3):559-575. 12. Binder G, Trebar B, Baur F, Schweizer R, Ranke MB. 29. Price AL, Patterson NJ, Plenge RM, Weinblatt ME, Shadick NA, Homozygosity of the d3-growth hormone receptor poly- Reich D. Principal components analysis corrects for strati- morphism is associated with a high total effect of GH on growth fication in genome-wide association studies. Nat Genet. and a low BMI in girls with Turner syndrome. Clin Endocrinol 2006;38(8):904-909. (Oxf). 2008;68(4):567-572. 30. Willer CJ, Li Y, Abecasis GR. METAL: fast and efficient meta- 13. Costalonga EF, Antonini SR, Guerra-Junior G, Mendonca BB, analysis of genomewide association scans. Bioinformatics. Arnhold IJ, Jorge AA. The -202 A allele of insulin-like growth 2010;26(17):2190-2191. factor binding protein-3 (IGFBP3) promoter polymorphism is 31. Huyghe JR, Jackson AU, Fogarty MP, et al. Exome array ana- associated with higher IGFBP-3 serum levels and better growth lysis identifies new loci and low-frequency variants influencing response to growth hormone treatment in patients with se- insulin processing and secretion. Nat Genet. 2013;45(2): vere growth hormone deficiency. J Clin Endocrinol Metab. 197-201. 2009;94(2):588-595. 32. Wu MC, Kraft P, Epstein MP, et al. Powerful SNP-set analysis for 14. van der Kaay DC, Hendriks AE, Ester WA, et al. Genetic and case-control genome-wide association studies. Am J Hum Genet. epigenetic variability in the gene for IGFBP-3 (IGFBP3): cor- 2010;86(6):929-942. relation with serum IGFBP-3 levels and growth in short chil- 33. Feng S, Liu D, Zhan X, Wing MK, Abecasis GR. RAREMETAL: dren born small for gestational age. Growth Horm IGF Res. fast and powerful meta-analysis for rare variants. Bioinformatics. 2009;19(3):198-205. 2014;30(19):2828-2829. 15. Clayton P, Chatelain P, Tatò L, et al. A pharmacogenomic ap- 34. Purcell SM, Wray NR, Stone JL, et al.; International proach to the treatment of children with GH deficiency or Turner Schizophrenia Consortium. Common polygenic variation con- syndrome. Eur J Endocrinol. 2013;169(3):277-289. tributes to risk of schizophrenia and bipolar disorder. Nature. 16. Stevens A, Murray P, Wojcik J, et al.; PREDICT Investigator 2009;460(7256):748-752. Group. Validating genetic markers of response to recombinant 35. Campbell CD, Kirby A, Nemesh J, Daly MJ, Hirschhorn JN. human growth hormone in children with growth hormone defi- A survey of allelic imbalance in F1 mice. Genome Res. ciency and Turner syndrome: the PREDICT validation study. Eur 2008;18(4):555-563. J Endocrinol. 2016;175(6):633-643. 36. Wajnrajch M, Meng Y, Dauber A, et al. Supplemental Table 1. 17. Stevens A, Clayton P, Tatò L, et al. Pharmacogenomics of insulin- Figshare, 2020. Posted June 14, 2020. doi:10.6084/ like growth factor-I generation during GH treatment in children m9.figshare.12479090.v1. with GH deficiency or Turner syndrome. Pharmacogenomics J. 37. Lettre G, Butler JL, Ardlie KG, Hirschhorn JN. Common 2014;14(1):54-62. genetic variation in eight genes of the GH/IGF1 axis does not 3214  Dauber et al    GWAS of Growth Hormone Responsiveness J Clin Endocrinol Metab, October 2020, 105(10):3203–3214 contribute to adult height variation. Hum Genet. 2007;122(2): B4GALT7 deficiency and review of the literature. Am J Med 129-139. Genet A. 2013;161A(10):2519-2527. 38. D’Aloisio AA, Schroeder JC, North KE, et al. IGF-I and IGFBP-3 40. Stevens A, De Leonibus C, Whatmore A, et al. Pharmacogenomics polymorphisms in relation to circulating levels among African related to growth disorders. Horm Res Paediatr. American and Caucasian women. Cancer Epidemiol Biomarkers 2013;80(6):477-490. Prev. 2009;18(3):954-966. 41. Ouni M, Belot MP, Castell AL, Fradin D, Bougnères P. The P2 39. Guo MH, Stoler J, Lui J, et al. Redefining the progeroid form promoter of the IGF1 gene is a major epigenetic locus for GH of Ehlers-Danlos syndrome: report of the fourth patient with responsiveness. Pharmacogenomics J. 2016;16(1):102-106. Downloaded from https://academic.oup.com/jcem/article/105/10/3203/5870346 by guest on 04 May 2021