Diagnosis of Childhood and Adolescent Growth Hormone Deficiency Using Transcriptomic Data

Abstract

Background: We have shown that gene expression (GE) data have promise as a novel tool to aid in the diagnosis of childhood growth hormone deficiency (GHD)1. Our previous study compared GE data in children with GHD to healthy control children of normal stature. The aim of this study was to assess the utility of GE data in the diagnosis of GHD in childhood and adolescence using non-GHD short stature children as a control group. Methods: GE data were obtained from patients undergoing growth hormone stimulation testing via a sample of blood taken at the start of the test. Arginine and glucagon stimulation tests with a cut-off for peak GH of <7mcg/L (IDS iSYS assay) were used for the diagnosis of GHD. GE was assessed in peripheral blood mononuclear cells via RNA-seq using the Illumina HiSeq 4000 platform. Data were taken for the 271 genes whose expression was utilised in our previous study. The synthetic minority oversampling technique was used to balance the dataset and a random forest algorithm applied to predict GHD status. Boruta was used to assess which of the genes were contributing to the predictive capacity. Results: Twenty-four patients were recruited to the study, with eight subsequently diagnosed with GHD. Of the eight patients diagnosed with GHD, three had two stimulation tests and five had one stimulation test with anterior pituitary hypoplasia (in addition one patient had an arachnoid cyst and another a thin stalk). Median (range) peak GH was 2.5 (0.1 - 5) mcg/L in the GHD group and 11.0 (7.4 - 31) mcg/L in the non-GHD group.There were no significant differences in gender, age, auxology (height SDS, weight SDS, BMI SDS) or biochemistry (IGF-I or IGFBP-3 SDS) between the GHD and non-GHD subjects. 82 of the 271 genes used in our previous study were above the threshold of detection for RNA-seq in this study. A random forest algorithm using these 82 genes gave an AUC of 0.97 (95% CI 0.93 - 1.0) for the diagnosis of GHD. Boruta was able to identify 65/82 genes with predictive capacity greater than permuted data within the dataset. Using a gene ontology approach the top fifty biological processes generated 16 clusters by affinity propagation which included regulation of TORC1 signalling and inositol phosphate metabolism. Conclusion: This study demonstrates highly accurate diagnosis of childhood GHD using a combination of GE data and random forest analysis and validates the findings of our original study. 1Murray etal (2018) JCI Insight 3(7): e93247