Characterizing the metabolome of children with growth hormone deficiency

Study design

A prospective clinical trial was conducted in the Endocrinology unit of Rambam Medical Center in Israel. The inclusion criteria for the study were age between 3 and 18 years old. The exclusion criteria included genetic syndromes or chromosomal abnormalitiesׂ such as Turner or Prader–Willi syndrome, systemic illness, and end-stage renal disease. Since the metabolomics profile may be influenced by the gut microbiome [1], children with antibiotics usage in the last three months were excluded from participation in the study. Children were referred to GHSTs following one of the following reasons: severe short stature (height 2.5 standard deviation score [SDS]), short stature relative to mid-parental height, or growth retardation, as specified in the consensus guidelines [15]. Prior to the test, all children underwent initial clinical and laboratory workup to exclude common underlying conditions for short stature such as celiac disease or thyroid function imbalance.

Medical records were reviewed for the following information: type of GHST, sex, chronological age, bone age, anthropometric measurements (height, weight), pubertal stage at the time of GHST, the use of sex hormone priming, GH levels during GHST, and basal IGF-I levels. Height was measured with a Seca 222 stadiometer and weight with a calibrated scale. BMI was calculated as weight (kg)/height (m²). Height, weight and BMI were expressed as SDS according to the recommendations of the Centers for Disease Control and Prevention (CDC) [16]. Prepubertal and pubertal stages were determined by physical examination. They are expressed as Tanner scores for genital status in boys and breast development in girls [17].

Testing protocol

Allocation to AST or GST was made by the referring pediatric endocrinologist. In general, our institution’s policy is to perform GST as the first provocative test, with the exception of children younger than 4 years of age, in which AST is performed first in order to minimize the chances of hypoglycemia events in this age group. A clonidine stimulation test is not routinely performed in our clinical center as it may lower blood pressure and requires monitoring.

There is known difficulty of diagnosing GHD during the immediate peripubertal period, as low GH levels in provocation tests frequently occur. However, no consensus on the use of priming with sex steroids before GH tests exists [15]. In our institution, girls>10 years and boys>11 years without signs of pubertal development are primed with sex hormones. Girls are given 2 mg β-estradiol orally daily for 3 days before testing, and boys receive a single intramuscular injection of 50 mg testosterone enanthate 10 days before the test.

All tests were performed following an overnight fast before collecting serum samples, and the serum samples were taken through a clinical standard procedure. Serum samples were obtained during the GHST and were immediately stored at −80 °C. GST is performed with an intramuscular injection of 30 μg/kg glucagon to a maximum of 1 mg. Blood samples for determination of GH levels are drawn at 0, 60, 90, 120, 150, and 180 min. AST is performed with an intravenous infusion of 0.5 gr/kg arginine over 30 min. Blood samples for determination of GH levels are taken at 0, 30, 60, 90, and 120 min. The serum GH concentrations are determined by a commercially available, solid-phase two-site chemiluminescent immunometric assay employing an Immulite 2000 automated analyzer (Siemens Healthcare Diagnostics). The GH assay is calibrated according to the WHO NIBSC 2nd IS 98/574. The analytical sensitivity of the method is 0.01 ng/mL. The consistency of the assay performance was checked by using Lyphochek Immunoassay control (Bio-Rad (Hercules, CA, USA). The GH total precision CV% was 6.35–5.21 % at GH levels of 3.27–13.31 ng/mL, respectively.

Outcome definition

In Israel, children with peak GH levels below 7.5 ng/mL during a GHST undergo a second test using a different pharmacological agent. A diagnosis of GH deficiency (GHD) is confirmed if the peak GH levels in the second test also remain below 7.5 ng/mL [18], 19]. Study participants were subdivided into two groups: children who were eventually diagnosed as having a GHD and controls. Notably, two participants who were initially referred for GHST due to an appropriate growth rate (defined as a decline of 0.3 SD from the child’s original growth curve ([18], 19]) had peak GH levels below 7.5 ng/mL (6.54 ng/mL and 5.7 ng/mL). However, their growth rate showed substantial improvement during clinical follow-ups, which was inconsistent with a diagnosis of GHD. Consequently, medical providers did not refer them for a second GHST. As these two children showed a suboptimal peak GH response during the first GHST but did not undergo a second confirmatory test, they were excluded from the analysis comparing children with GHD to controls.

Metabolite extraction

Serum samples were analyzed at the Weizmann institute of science. Extraction and analysis of lipids and polar/semipolar metabolites was performed as previously described in Malitsky et al. [20] and Zheng et al. [21] with some modifications: 100 µL of serum were extracted with 900 µL of a pre-cooled (−20 °C) homogenous methanol:methyl-tert-butyl-ether (MTBE) 1:3 (v/v) mixture, containing following internal standards: 0.1 μg*mL-1 of phosphatidylcholine (17:0/17:0) (Avanti), 0.4 μg*mL-1 of phosphatidylethanolamine (17:0/17:0, 0.15 nmol*mL-1 of ceramide/sphingoid internal standard mixture I (Avanti, LM6002), 0.0267 μg/mL d5-TG internal standard mixture I (Avanti, LM6000) and 0.1 μg*mL-1 palmitic acid-13C (Sigma, 605573).

The tubes were vortexed at 500 rpm for 1 min, and then sonicated for 30 min in an ice-cold sonication bath (taken for a brief vortex every 10 min). Then, double deionized water (DDW): methanol (3:1, v/v) solution (450 µL) containing internal following standards: C13 and N15 labeled amino acids standard mix (Sigma) was added to the plates with extracted samples, followed by centrifugation. The upper, organic phase (450 µL), was transferred into a 2 mL Eppendorf tube. The polar phase was re-extracted as described above, with 200 µL of MTBE. The 150 µL of the upper organic phase were combined with the first one, and dried in a speedvac, and then stored at −80 °C until analysis. For analysis, the dried lipid extracts were re-suspended in 250 μL mobile phase B (see below) and centrifuged at 3,500 rpm at 4 °C for 30 min, and 200 µL were transferred into a new plate for injection. Lower, polar phase (550 µL) used for polar metabolite analysis, was lyophilized and resuspended in 200 µL methanol: DDW (50:50), centrifuged at 3,500 rpm at 4 °C for 30 min and 150 µL were transferred into new plate, centrifuged in the same way, and 120 µL were transferred into new plate for injection.

LC-MS for lipidomics analysis

Lipid extracts were analyzed using a Waters ACQUITY I class UPLC system coupled to a mass spectrometer (Thermo Exactive Plus Orbitrap) which was operated in switching positive and negative ionization mode. The analysis was performed using Acquity UPLC System combined with chromatographic conditions as described in Malitsky et al. (2016) with small alterations. Briefly, the chromatographic separation was performed on an ACQUITY UPLC BEH C8 column (2.1 × 100 mm, i.d., 1.7 μm) (Waters Corp., MA, USA). The mobile phase A consisted of DDW: Acetonitrile: Isopropanol 46:38:16 (v/v/v) with 1 % 1 M NH4Ac, 0.1 % acetic acid. Mobile phase B composition is DDW: Acetonitrile: Isopropanol 1:69:30 (v/v/v) with 1 % 1 M NH4Ac, 0.1 % acetic acid. The column was maintained at 40 °C and the flow rate of the mobile phase was 0.4 mL*min-1. Mobile phase A was run for 1 min at 100 %, then it was gradually reduced to 25 % at 12 min, following a decrease to 0 % at 16 min. Then, mobile phase B was run at 100 % till 21 min, and mobile phase A was set to 100 % at 21.5 min. Finally, the column was equilibrated at 100 % A till 25 min.

Lipid identification and quantification

Orbitrap data was analyzed using LipidSearch™ software (Thermo Fisher Scientific). The validation of the putative identification of lipids was performed by comparing a library which contains lipids produced by various organisms and on the correlation between retention time (RT) and carbon chain length and degree of unsaturation. Relative levels of lipids were normalized to the internal standards and the amount of plasma used for analysis.

LC-MS polar metabolite

Metabolic profiling of polar phase was done as described at Zheng et al. [21] with minor modifications described below. Briefly, analysis was performed using Acquity I class UPLC System combined with mass spectrometer Q Exactive Plus Orbitrap™ (Thermo Fisher Scientific) operated in a negative ionization mode. The LC separation was done using the SeQuant Zic-pHilic (150 × 2.1 mm) with the SeQuant guard column (20 × 2.1 mm) (Merck). The Mobile phase B: acetonitrile and Mobile phase A: 20 mM ammonium carbonate with 0.1 % ammonia hydroxide in water: acetonitrile (80:20, v/v). The flow rate was kept at 200 μL* min−1 and gradient was as follow: 0–2min 75 % of B, 14 min 25 % of B, 18 min 25 % of B, 19 min 75 % of B, for 4 min, 23 min 75 % of B. Polar metabolites data analysis was done using TraceFinder (Thermo Fisher Scientific), when detected compounds were identified by accurate mass, retention time, isotope pattern, fragments and verified using in-house-generated mass spectra library.

Statistical analyses

Summary statistics were computed in Python [22]. To study the metabolomics profile of children with GHD, we first compared the metabolomics profile of cases and controls using the Mann-Whitney U test. Then, for each metabolite, we constructed a logistic regression model adjusting for age, sex, tanner stage and type of GHST. We further clustered the metabolites into 50 clusters (lipids and polar separately) and repeated the analyses using the average of metabolites in each cluster as representative.

To investigate the impact of varying peak GH thresholds on metabolomic profiles we conducted a principal component analysis (PCA). This analysis, which included all children in the study, compared cases and controls based on their peak GH levels during GH stimulation tests (GHST) and different threshold criteria. Statistical significance was set at p<0.05. False Discovery Rate (FDR) was employed at the rate of 0.1.

Ethical considerations

The study complied with the World Medical Association Declaration of Helsinki regarding ethical conduct of research involving human subjects and/or animals. Study protocol was reviewed and approved by the Human Research Ethics Committee of Rambam medical center Institutional Review Board (IRB) and Weizmann Institute of Science IRB.

Comparison between the metabolite profiles of children with GH deficiency to controls

We first analysed the differences in the metabolomics profile of children diagnosed with GHD compared to controls. Of the 68 children participated in the study, 25 (36.8 %) were clinically diagnosed with GHD, and 41 (60.3 %) served as controls. Two children (2.9 %), were excluded from this analysis because although their initial GH stimulation test was abnormal, they did not undergo the necessary confirmatory test due to maintaining a normal growth rate (see Methods). The characteristics of the children included in the analyses are presented in Table 1. No statistical differences were found in the age, sex, tanner stage, and BMI percentile of the children. As expected, as children with normal peak GH on GHST are not referred to a second test, more children who had GHD were recruited in the second GHST test compared to the control group (p<0.05). Of the 25 children diagnosed as having a GHD, MRI results were available for 19 children. In 15 children (78.9 %) MRI was normal. Abnormal MRI findings appearing each in one child included a very small cyst on the right size of the hypophysis, a relatively small hypophysis size and a relatively weak signal from the neurohypophysis. Two children had ectopic neurohypophysis.

Overall, 7 polars metabolites (Table S1) and 50 lipids (Table S2) were significantly different between groups (p<0.05; two sided Mann-Whitney U test). The majority of significantly different lipid metabolites, including various phosphatidylcholines, phosphatidylserines, lysophosphatidylcholines, sphingomyelins, ceramides, cholesteryl esters, and triacylglycerols, were elevated in the control group. Notably, only three metabolites – acetylhexosyl sitosteryl ester (18:2), plasmalogen phosphatidylcholine (18:0e/16:0), and phosphatidylethanolamine (20:0/18:2) – were higher in the GHD group. Polar metabolites with a higher level in children with GHD included Heptanoic acid and N-Octanoylglycine. However, after implying FDR correction, only one lipid metabolite, which was higher in controls, phosphatidylserine (PS) (40:3), remained significant. To adjust for potential confounders, we next constructed a logistic regression model for each metabolite adjusting for age, sex, tanner stage and type of GHST. In this analysis, no metabolite achieved the significance threshold (p<0.05). Since many of the metabolites are correlated to each other, we next separately cluster the lipid and polar metabolites into 50 clusters each, and repeat the analyses using the average of metabolites in each cluster as representative. In an adjusted logistic regression model for each metabolite cluster, two lipid clusters were significantly different between groups.

The influence of different peak GH thresholds on metabolomic profiles

In Israel, a peak serum GH level of below 7.5 ng/mL is used to diagnose GH deficiency [19]. However, other countries utilize a different threshold and it has been previously suggested that threshold definitions are arbitrary and not precise, leading to misdiagnosis of numerous children and adolescents [25]. To examine differences in metabolomic profiles among all children in the study, we determined the peak GH level during GH stimulation testing for each participant. Based on this, we classified them as cases or controls according to three different diagnostic thresholds for peak serum GH levels: (A) 5 ng/mL or lower (9 children were compared to 59 controls) (B) 7.5 ng/mL or lower (27 children were compared to 41 controls) and (C) 10 ng/mL or lower (37 children were compared to 31 controls). We then performed dimensionality reduction analyses of polar and lipid metabolites for each scenario. Notably in the threshold of 7.5 ng/mL, the threshold used for GHD definition in Israel, and 5 ng/mL, PCA revealed statistically significant separation between groups by lipids (p<0.05) (Figure 1, Full results of PCA analyses are presented in Figure S2).

Figure 1:

Principal component analysis (PCA) of the lipid metabolomics profile of children with peak growth hormone of (A) 5 ng/mL or lower and (B) 7.5 ng/mL or lower during GHST vs. controls. Shown are the principal components which revealed statistically significant separation between groups alongside their ratio of the variance explained. PCA was applied over the log10 transformed data. Full results of PCA analyses are presented in Figure S2.

To further analyze the difference in the metabolomics profile of children who would have been diagnosed with GHD according to different peak GH thresholds in the GHST vs. controls, we calculated the mean distance in lipids and polar metabolomics by peak GH threshold (Figure 2). This analysis demonstrates a greater separation through metabolomic analysis with stricter diagnostic thresholds for GHD.

Figure 2:

Mean euclidean distance (y-axis) between the metabolomics profiles of study participants separated based on lipids (A) and polar (B) metabolites. The distances are computed between children who would have been diagnosed with GHD according to different GH peaks on GHST vs. controls (x-axis). The peak GH threshold of 7.5 ng/mL used in our practice is marked in red. Solid blue line shows the mean while the shaded area describes the standard error of the mean. Distances were computed over the first five PCs of the metabolomics data. PCA was applied over the log10 transformed data..