Targeted Searches of the Electronic Health Record and Genomics Identify an Etiology in Three Patients with Short Stature and High IGF-I Levels

Concerns about short stature are one of the most common reasons for referral to a pediatric endocrinologist. Human height is influenced by a multitude of biological processes, and molecular variants in a wide array of genes can result in short stature [1]. Consequently, despite adequate clinical and laboratory evaluation, a pathological diagnosis is not identified in the vast majority of children evaluated for short stature [2]. While many of these children have variants of normal physiology, a number have undiagnosed genetic growth disorders. Recent advances in genomic technologies have revolutionized our ability to discover novel or rare genetic etiologies of growth failure. However, it is often quite difficult to determine which of the many patients seen for short stature may have an undiagnosed genetic cause.

The widespread adoption of electronic health records (EHRs) in the US has created a rich asset to facilitate clinical research. EHRs are commonly used as the main data source for safety surveillance, observational, epidemiological, and prospective cohort studies [3, 4]. Most recently, with the increased popularity of EHR-linked biorepositories, EHRs have become a valued tool for the augmentation of clinical characterization to drive discovery of phenotype-genotype associations [5, 6]. EHRs improve our ability to identify cohorts of patients with a specific phenotype of interest by using a combination of diagnosis codes, pharmacy data, narrative text, laboratory values and other clinical features including anthropometric data [5].

The objective of our study was to assess the feasibility of using dense phenotypic information from the EHR to systematically identify a distinct cohort of patients with short stature who have a high probability of harboring a monogenic etiology for their short stature. As an initial proof of principle, we chose to focus on patients with elevated insulin-like growth factor I (IGF-I) levels, a rare finding in patients with short stature. High IGF-I levels are most commonly seen in patients with heterozygous mutations of the gene encoding the receptor for IGF-I (IGF1R). Affected individuals typically present with distinct phenotypic features including prenatal and postnatal growth retardation, microcephaly, and often with serum IGF-I concentrations above the normal ranges [7‒9]. However, despite multiple reports in the literature, IGF-I resistance due to IGF1R mutations is underdiagnosed in clinical practice, perhaps because of its wide phenotypic spectrum [9‒11] as well as laboratory evaluation that can be misinterpreted as reassuring. Our group previously conducted large-scale candidate gene sequencing in a cohort of patients with idiopathic short stature and identified a frameshift mutation in IGF1R in a patient with clinical features suggestive of a pathological IGF1R deficiency state [12]. Similarly, Storr et al. [13] reported a case of a novel pathogenic IGF1R mutation identified in a cohort of patients with IGF-I resistance. Most recently, Yang et al.[11] reported four novel variants in IGF1R in a cohort of patients with prenatal and postnatal growth failure. In addition, numerous other conditions might resemble the phenotype of IGF-I resistance [14].

In the present study, we sought to provide a proof of principle that searches of the EHR could identify a discrete clinical subgroup of patients with short stature who were at high risk of having a particular type of undiagnosed genetic disorder. We executed this algorithm at three leading pediatric hospitals and discuss the challenges in its implementation.

We performed a targeted bioinformatics search of the EHR at three large academic pediatric endocrine departments (Boston Children’s Hospital (BCH), Children’s Hospital of Philadelphia (CHOP), and Cincinnati Children’s Hospital Medical Center (CCHMC)). CHOP and CCHMC use the EPIC EHR system (Epic, Verona, WI, USA) while BCH uses the Cerner system (Cerner, North Kansas City, MO, USA). These are two of the most widely used EHR systems in the United States. The research study protocol was approved by the institutional review board of each institution. Only subjects seen since 2012 were included as they were the most likely to be available for prospective recruitment. Inclusion criteria were height below –2 SD (adjusted for age and sex, based on CDC reference tables [15]) and an IGF-I level above the 90th percentile (assay specific and adjusted for age and sex [16]). An elevated IGF-I level was required only at a single time point in order to meet screening criteria to maximize capture rates. At all sites, the bioinformatics algorithm excluded patients who had growth hormone (GH) or IGF-I listed as an active medication at the time of IGF-I measurement. This was done to exclude patients in whom the IGF-I elevation was likely a result of treatment.

Determination of IGF-I percentiles was site-specific. At CCHMC, serum IGF-I was determined by a chemiluminescent immunometric assay (IDS-iSYS; Immunodiagnostic Systems Ltd., Boldon, UK). A single IGF-I assay was used for all the samples during this period. IGF-I percentiles were calculated using the assay’s manufacturer datasheets [16]. At CHOP, IGF-I levels were measured by various commercial laboratories, per clinical routine (patient and insurance preference), and some of the laboratories utilized different IGF-I assays during the study period. Representative samples were identified for each of the IGF-I assays for each laboratory using the EHR. The reference ranges listed in each assay were recorded manually for the pubertal stage and/or the age of the patient. Mean and standard deviations were generated from these, and an algorithm was created to query the EHR and generate percentiles for each test according to pubertal stage and/or age. Some assays reported Z scores directly and, where possible, these were extracted and converted to percentiles. In order to validate this approach, we counted the number of samples with IGF-I concentrations above the 90th percentile for each assay to ensure that no assay was overrepresented on initial screen. At BCH, a single IGF-I assay (Immulite, Siemens Medical Solutions, Tarrytown, NY, USA) was used except for a 6-month period when samples were sent to ARUP. IGF-I percentiles were calculated for these 2 assays using the assay’s manufacturer datasheets.

Charts of the patients identified by the bioinformatics screening algorithm were manually reviewed by one of the investigators at each of the participating institutions. Patients with known underlying genetic conditions or other chronic conditions (not a structural birth defect) explaining their short stature, and patients with precocious puberty were excluded from the study (Table 1). Patients who subsequently achieved a normal height or who had a repeat IGF-I level that was not elevated without treatment were excluded as the intention of the study was to only examine patients with a likely genetic etiology for their clinical presentation. All eligible patients were approached by mail or telephone and offered the opportunity to participate in the study. Written informed consent and assent were obtained from all participants and/or their parents/guardians, as age appropriate. The eligible patients and their immediate relatives were invited to attend a single research study visit at each of the participating institutions for standardized phenotyping and sample acquisition (blood, saliva, or both).

Whole exome sequencing was performed either at CCHMC or at the Broad Institute on DNA extracted from whole blood or saliva samples from 9 patients and their immediate family members. Details of the exome sequencing methods have been previously described [17]. For each patient, we performed a number of analyses. For all subjects, we searched for nonsynonymous, homozygous, or compound heterozygous recessive variants with a minor allele frequency below 0.01 in the 1,000 Genomes and gnomAD databases (http://browser.1000genomes.org/, and http://gnomad.broadinstitute.org/). Missense variants were only included if they were predicted to be pathogenic by a minimum of 2 of the following in silico prediction programs: Polyphen2, MutationTaster, and SIFT. If DNA was available from both parents, we searched for de novo variants in the proband which were required to be absent from the aforementioned databases (i.e., novel variants). In male subjects, we included a search for potential X-linked variants which were also required to be novel. Finally, if a parent was also affected with short stature, we looked for novel heterozygous variants inherited from the affected parent. In addition to exome sequencing, the subjects had previously undergone clinical testing for copy number variants, as assessed by chromosomal microarray analysis, and were all negative for significant copy number variants.

Each of the identified IGF1R missense variants was regenerated with the QuikChange® II site-directed mutagenesis kit (Agilent Technologies Santa Clara, CA, USA) using IGF1R cDNA in the pcDNA3.1/AMP expression plasmid, as previously described [18, 19]. The primers used for generating the variants were: Val1013Phe (V1013F) 5′-GGGGTCGTTTGGGATGTTCTATGAAGGAGT­TGC-3′ (sense), 5′-GCAACTCCTTCATAGAACATCCCAAA­CGACCCC-3′ (antisense); Thr28del: 5′-GCTCTGGCCGAGT­GGAGAAATCT-3′ (sense), 5′-ATTTCTCCACTCGGCCAGA­GC-3′ (antisense). The resultant cDNA constructs carrying the point mutations were verified by Sanger DNA sequencing.

HEK293 cells (ATCC; LGC Standards, Wesel, Germany) were cultured in Dulbecco’s modified Eagle’s medium (DMEM) supplemented with 10% fetal bovine serum, 100 U mL^–1 penicillin, and 100 μg mL^–1 streptomycin at 37°C in 5% CO₂. Cells were seeded in 6-well dishes at a density of 30–50,000 cells/well and transfected when approximately 70% confluent. Cells were transiently transfected with pcDNA3.1 vector or plasmids encoding WT or variant IGF1R for Thr28del and V1013F, using PolyJet In Vitro DNA Transfection Reagent (SignaGen; Rockville, MD, USA) following the manufacturer’s instructions. Transfected HEK293 cells were placed in serum starvation media (DMEM supplemented with 0.1% bovine serum albumin (BSA) and penicillin/streptomycin) for 8–24 h followed by fresh starvation media for 90 min. After 90 min, cells were treated with 100 ng of recombinant human IGF-I (Life Technologies, Frederick, MD, USA) in DMEM or a negative treatment of DMEM with phosphate-buffered saline (PBS) for 10 min, washed in cold PBS, and cells solubilized in lysis buffer (1× PBS, 1% Nonidet P-40, 0.5% sodium deoxycholate, 0.1% SDS, 10 mg mL^–1 phenylmethylsulfonyl fluoride, 100 nM sodium orthovanadate) and protease inhibitor mixture (complete Mini-EDTA-free; Sigma, St. Louis, MO, USA). Cell debris was removed by centrifugation, and final whole-cell lysates were stored at –20°C.

We performed the BioRad DC protein assay according to the manufacturer’s protocol (Biorad, Hercules, CA, USA). For immunoblot analysis, 30 μg of whole-cell lysates were separated on 4–20% SDS-polyacrylamide gels, transferred to 0.45-μm nitrocellulose membranes. Membranes were blocked in 5% BSA prior to primary antibody exposure (overnight at 4°C). The primary antibodies were: rabbit monoclonal IgG against IGF1R β-subunit (D23H3) (dilution 1:1,000), rabbit monoclonal IgG against phospho-IGF1R β-subunit (Tyr 1135/1136) (dilution 1:1,000), and mouse monoclonal IgG against alpha-tubulin (DM1A) from Cell Signaling Technologies (Beverly, MA, USA). Secondary antibodies (anti-mouse IgG and anti-rabbit IgG) conjugated with horseradish peroxidase were obtained from BioRad. Detection of immune-labeled proteins was performed using a commercial chemiluminescent assay (ECL prime; Amersham, UK). Visualization and quantitative measurements were performed on the BioRad ChemiDoc Touch Imaging System, and densitometry was performed using Image J software. All immunoblot data shown are representative of at least three independent experiments. Unpaired T tests were used to compare densitometric analyses between wild-type and variant IGF1R.

An overview of eligibility assessment and recruitment is provided in Table 1. In total, 234 patients were identified by the algorithm across all three institutions with 39 being deemed eligible after manual review (17% eligibility rate). Of those, 9 were successfully recruited (23% recruitment rate). Interestingly, 89 of a total of 234 patients (38%) flagged by the algorithm had a preidentified genetic condition including 65 patients with Duchenne muscular dystrophy. Additional details of the excluded patients are available in Table 1. The 9 recruited subjects ranged in age from 2 to 15 years and had a mean height SDS at presentation of –2.7 SD (range –2.2 to –4.2 SD) (Table 2). The majority of subjects had additional syndromic features, although some were classified as idiopathic short stature. Additional clinical details are provided in Table 2.

Exome analysis was performed in all 9 patients. DNA was available for both parents in 6 of the families, and paternal samples were missing in 3 families (patients 1, 6, and 9). Prioritization of exome sequencing variants was performed as per the methods stated above. Details of all variants identified in the analysis can be found in the online supplementary Table and Excel File (see www.karger.com/doi/10.1159/000504884 for all online suppl. material). We identified a likely genetic etiology in 3 of the 9 cases. Two of the patients had novel, likely pathogenic variants in IGF1R including a novel missense variant (p.Val1013Phe) and a maternally inherited single amino acid deletion (p.Thr28del) (Fig. 1a). Functional validation studies supported the pathogenicity of these 2 variants (see below). In addition, patient 5 was found to have a novel, de novo missense variant in CHD2 (p.Val540Phe) which is the likely etiology of his short stature and intellectual disability. Details of these 3 cases are presented below.

Patient 6 is a 7.5-year-old African-American male with short stature (SDS –3) and attention deficit hyperactivity disorder (ADHD) born at 40 weeks of gestation with a birth weight of 2.69 kg (SDS –2.0). His endocrine evaluation at the age of 7.5 years showed elevated serum IGF-I (261 ng/mL, +2.8 SDS) and serum insulin-like growth factor-binding protein-3 (IGFBP-3) of 3 mg/L (normal range 2.1–4.2). On physical exam, he was prepubertal, nondysmorphic, and had proportional body segments. His head circumference was –1.5 SD. Family history revealed that his father had short stature with a height of 160 cm (SDS –2.3), whereas his mother had a normal height of 155 cm (SDS –1.3). The patient, his unaffected mother, and his aunt underwent whole exome sequencing. DNA from his father was not available. Based on the available clinical information, we prioritized heterozygous nonsynonymous variants only present in the patient which could theoretically be de novo or inherited from the affected father. This identified a novel missense variant (c.3037G>T, p.Val1013Phe) in IGF1Rwhich was predicted to be damaging by Polyphen2, MutationTaster, and SIFT. The missense variant was validated by Sanger sequencing.

Patient 9 is an 8-year 11-month-old Caucasian female referred to endocrinology for short stature (SDS –3.1). Her birth weight was 2.438 kg (–1.8 SD). On examination, the patient presented with microcephaly with a head circumference of –3.6 SD. At the age of 8 years 5 months, she had an IGF-I level of 310 ng/mL (90–95th percentile) and IGFBP-3 of 5.2 mg/L (normal range 1.6–6.5). Her mother had a normal height just below the 5th percentile (152 cm, SDS –1.7). The patient, her mother, and sister underwent WES. DNA from her father was not available. Given the available clinical information and history of mother’s height below the 5th percentile, we prioritized nonsynonymous, heterozygous variants inherited from the mother but not present in the sister whose height was at the 46th percentile. A novel maternally inherited single amino acid deletion (c.80_82del, p.Thr28del) in IGF1R was identified and predicted to be damaging. The variant was validated by Sanger sequencing.

Patient 5 is a 15-year-old Caucasian male with short stature (SDS –2.6), intellectual disability, ADHD, and autism who was born at 40 weeks’ gestation with a birth weight of 3.062 kg (SDS –1.1) and length of 47.6 cm (SDS –1.21). The patient had a history of microcephaly but his head circumference at birth was not available. In early infancy, he was noted to have mild developmental delay and feeding difficulties. His physical exam revealed multiple dysmorphic features including prominent curved eyebrows, prominent nasal tip, short wide philtrum, full lips, and down turned corners of mouth. He was diagnosed with ADHD and autism during childhood and underwent formal neuropsychological evaluation which revealed an IQ of 72. Parents denied history of seizures. An endocrine evaluation revealed a serum IGF-I of 535.1 ng/mL (90–95th percentile) and IGFBP-3 of 7.7 mg/L (normal range 2.6–6.3). His pubertal development advanced normally. There was no family history of short stature; his mother’s height was 155 cm (SDS –1.3) and his father’s 175 cm (SDS –0.2). Exome sequencing identified a novel de novo missense variant (c.1618G>T, p.Val540Phe) in CHD2 which was predicted to be damaging by Polyphen2, MutationTaster, and SIFT. The variant was validated by Sanger sequencing. CHD2 is involved in chromatin remodeling, and denovo mutations in CHD2 have been previously shown to cause a wide range of phenotypes including autism, developmental delay, and early-onset seizures [20, 21]. Multiple patients with mutations in CHD2 have been reported to have short stature [22], and a mouse model of CHD2 deficiency demonstrated growth retardation [23]. The patient’s variant is found in the critical SNF2-related helicase domain, a site of previously identified pathogenic missense mutations [20].

To investigate the pathogenicity of the novel missense variants in IGF1Ridentified in our patients, plasmids containing wild-type and mutant IGF1R were transfected into HEK293 cells. Immunoblot analyses indicated that both the p.V1013F and p.Thr28del variants caused significant decreases in IGF1R expression with concomitant poor responses to IGF-I stimulation (Fig. 2), thus supporting the pathogenic impact of the variants. It is of note that pIGF1R was detected with increasing loading of p.Thr28del cell lysates, but not in response to IGF-I (online suppl. Fig. 1).

Short stature can be caused by numerous genetic etiologies. Often, it can be difficult to ascertain the specific genetic etiology based on clinical grounds alone. In one study, detailed clinical and biochemical evaluation by an endocrinologist and a geneticist yielded an etiology in approximately 14% of patients referred for evaluation of growth retardation or short stature [24]. In the same study, an additional 16.5% of cases were molecularly diagnosed by exome sequencing in a random subset of 200 patients with short stature of no clear etiology [24]. This supports the power of larger-scale genetic sequencing to identify causes of short stature that are missed on clinical grounds alone as many patients present with subtle phenotypic or biochemical abnormalities. However, larger-scale genetic sequencing is too expensive to feasibly perform as a screen for patients presenting with short stature. The purpose of the current study was to leverage an EHR-based algorithm to narrow the scope and focus large-scale genetic screening to identify genetic etiologies in patients with a rare form of short stature. We believe that this could prove a much more efficient approach to identifying patients with rare genetic growth disorders. As a proof of principle, we focused on a distinctive, rare biochemical phenotype, namely patients with elevated IGF-I levels in the setting of short stature. Since IGF-I is a key mediator of GH action, it is unusual for patients with short stature to present with elevated IGF-I levels. After manual review, we successfully identified a total of 39 patients fitting our criteria, but despite approaching all eligible patients at three large academic centers, only 9 patients were willing to participate in our study. This further underscores the importance of multicenter collaborations for the investigation of rare phenotypic presentations.

The 9 patients in our cohort had a range of phenotypic features including many with dysmorphic features or developmental delay. We identified a genetic etiology in each of 3 patients, two of whom had novel missense variants in IGF1R. Notably, these two patients were classified as having idiopathic short stature and did not have additional syndromic features. Both individuals inherited the variants from one of their parents, one of whom had short stature and one of whom had a height in the low normal range (–1.7 SD). Patients who carry pathogenic mutations in IGF1R present with a wide range of heights and it is not unexpected to have a parent whose height is in the low-normal range [9]. IGF1R is a lead candidate gene for patients with short stature and elevated IGF-I levels, but neither of these patients had clinical testing for mutations in IGF1R. One of the patients had a frankly elevated IGF-I level, while serum IGF-I was at the upper end of the normal range for the other 2 patients. The importance of these diagnoses is that GH therapy can increase height in patients with IGF1R mutations, although the IGF-I level may need to be titrated to a higher level [9, 25]. The third patient had a de novo missense variant in CHD2, a gene involved in chromatin remodeling. Chromatin dynamics have been identified as playing an important role in overall height biology [26], and defects in this specific gene cause a syndrome marked by developmental delays, autism, early-onset seizures, and often short stature [22]. Of note, our patient did not have seizures, one of the more common manifestations of mutation in CHD2, which may be why this gene was not considered as part of the clinical evaluation. It remains mechanistically unclear how CHD2 perturbs the GH/IGF-I axis. We do not know the reason for the elevation of IGF-I levels in the patients without a molecular diagnosis. It is possible that these patients have mutations outside of the exome, such as in gene regulatory regions or possibly epigenetic changes, which affect the GH/IGF-I axis. Additional research is needed to determine potential etiologies in these patients.

The novel IGF1R variants, p.Thr28del and p.V1013F, identified in patient 6 and patient 9 respectively, were likely causal of impaired growth. Both were poorly expressed in reconstitution studies. The p.Thr28del variant, within the secretory signal peptide (amino acid residues 1–30), is close to the proteolytic cleavage site and is the first pathological mutation discovered that involves the signal peptide of IGF1R. Appropriate removal of secretory signal peptides is essential for exporting proteins destined for insertion into the plasma membrane or the extracellular milieu. Interestingly, an in silico signal peptide program (http://www.cbs.dtu.dk/services/SignalP/) indicates that 5 critical residues, ²⁸ThrSerGlyGluIle³² (Fig. 1b), are important for normal cleavage between Gly30 and Glu31 (predicted high probability of 0.9354 out of 1.0), but loss of residue Thr28 shifts the 5 critical residues to ²⁷ProSerGlyGluIle, which destroys the cleavage site (predicted probability of cleavage dropped to 0.2446). Hence, the poorly immune-detected IGF1R p.Thr28del suggests that the aberrant IGF1R variant may be targeted for degradation pathways. Previously, several inactivating mutations have been reported within the intracellular tyrosine kinase domain including missense variants p.E1080K [27] and p.G1155A [28]. The IGF1R p.V1013F identified in this report lies within the nucleotide binding region of ATP (residues 1,005–1,013), a motif that binds nucleotide phosphates. The missense p.V1013F not only results in significantly reduced protein expression, but the tyrosine kinase activities were also significantly reduced, comparable to the previously described IGF1R p.E1080K mutation [26]. Altogether, our studies support IGF1R haploinsufficiency in patients 6 and 9 as the likely established cause of short stature consistent with previously reported cases.

This study highlights some of the key difficulties in implementing an EHR-based search for rare phenotypes. First, our algorithm was limited by its ability to only search for data elements that are collected as discrete data fields in the EHR. This allowed us to identify patients based on their height Z scores and IGF-I levels. However, manual review was required and resulted in exclusion of the majority (83%) of screened cases. The major reasons for exclusion were the presence of another chronic condition, IGF-I levels appropriate for pubertal stage, or lack of persistence in IGF-I elevation. Interestingly, 65 out of the 234 (28%) patients flagged by the algorithm had Duchenne muscular dystrophy. Previous reports examining the etiology of short stature in these patients have found IGF-I levels within the normal range [29]. However, the elevated IGF-I levels may be due to concomitant treatment with high-dose steroids which can lead to IGF-I resistance and elevation in circulating IGF-I levels [30]. The large number of patients with this condition is likely due to a specific multidisciplinary clinic at CCHMC for these patients. Diagnostic coding is not particularly reliable in the EHR and thus simple selection of patients with diagnostic codes for short stature or growth failure would not exclude patients with underlying chronic condition. It is possible to create a list of diagnoses to exclude which may be a more reliable approach and would decrease the numbers of charts needed for manual review, but that list would never be comprehensive enough to fully eliminate the need for manual review. Pubertal status is not routinely recorded in our current EHR systems as a discrete data element but is rather often found as text within the clinic note. To further streamline the EHR search, a discrete data field for pubertal status could be created within note templates for growth evaluations, or alternatively, searches that employ natural language processing could “read” the notes and incorporate the pubertal information into the search. Finally, for simplicity of algorithm coding, we did not include a rule requiring serial IGF-I levels to be persistently elevated, but this could be easily incorporated into a future version of the algorithm. In sum, our study provides proof of principle that an EHR algorithm can identify a relevant patient subpopulation among the many patients presenting with short stature, but significant improvements in search strategy are required before this can be fully automated into routine clinical practice. Despite these limitations and the need for manual review, our approach was able to relatively quickly identify a high yield group of individuals from the thousands of children evaluated for short stature at our three large medical centers.

In conclusion, our study demonstrates the feasibility of identifying rare phenotypic subgroups based on discrete data elements in the EHR. With further improvements in the algorithm, it should now be possible to proactively flag these patients as being at high risk for having an underlying genetic etiology. In our cohort, after significant manual review, we found a genetic etiology in 3 of 9 patients with short stature and high IGF-I levels, and evaluation of the IGF1R gene alone would have identified a molecular cause in 2 of the 9 patients. As these are rare genetic etiologies, practicing clinicians may not be aware of the likelihood of identifying a specific genetic etiology or realize that such an etiology exists. Decision support through the EHR can alert clinicians to the possibility of rare disorders whose diagnosis could affect therapeutic decisions. Our study was not designed to examine the yield of an algorithm-based approach versus comprehensive sequencing as an initial approach in the diagnosis of short stature as some have advocated. Further research will be needed to compare these approaches, but we first sought to provide proof of principle that an algorithm-based approach could prove useful in the evaluation of rare subgroups of patients with short stature. Additional research is needed to understand the applicability of this approach to a wider range of phenotypic subgroups, but our model of multi-institutional collaboration will be critical to the success of these endeavors.

We would like to thank Casey Gorman and Anna Bartels for assistance with subject recruitment. Research reported in this publication was supported by the Eunice Kennedy Shriver National Institute of Child Health and Human Development of the National Institutes of Health under Award Number R01HD093622. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Additional funding support was provided by the Genomics Research and Innovation Network.

All studies were conducted in compliance with the Declaration of Helsinki. Institutional review board approval was obtained for all studies, and written informed consent was obtained from subjects or their guardians as appropriate.

A.G. serves on the Steering Committee of the Pfizer International Growth Study Database and is a Consultant for Sandoz Inc. A.D. has received research support from Novo Nordisk and Ipsen and has served as a consultant for Ascendis, OPKO Biologics, Pfizer, Ipsen, Sandoz, and Novo Nordisk. J.N.H. is a member of the scientific advisory board for Camp4 Therapeutics.

Research reported in this publication was supported by the Eunice Kennedy Shriver National Institute of Child Health and Human Development of the National Institutes of Health under Award Number R01HD093622. Additional funding support was provided by the Genomics Research and Innovation Network. None of the funding agencies played any role in the study design, conduct, analysis, or writing of the manuscript.

C.C.S. and A.D. designed the study, performed data analysis, and wrote the manuscript. C.P.H., L.T., A.F., J.N.H., and A.G. recruited patients, phenotyped patients, and participated in data anlaysis. G.L. and D.C. performed the bioinformatics analysis. M.A., A.D., and V.H. designed and performed the in vitro studies. All authors reviewed and edited the manuscript.

Members of the Genomics Research and Innovation Network are listed in the Supplementary Appendix