Gene expression signatures predict response to therapy with growth hormone

Adam Stevens,Cheri Deal,Mohamad Maghnie,Julia Skorodok,Pierre Chatelain,Geoffrey Ambler ,Philip Murray ,Diego Yeste,Eirik Vangsøy-Hansen ,Stefano Zucchini,Ekaterina Koledova,Alicia Belgorosky,Terence Garner,Régis Coutant ,Juan-Pedro Lopez Siguero ,Lars Hagenäs ,Gerhard Binder,Klaus Kapelari,Chiara De Leonibus,Elena Bashnina,Jovanna Dahlgren,Peter Clayton

doi:10.1038/s41397-021-00237-5

Abstract

Recombinant human growth hormone (r-hGH) is used as a therapeutic agent for disorders of growth including growth hormone deficiency (GHD) and Turner syndrome (TS). Treatment is costly and current methods to model response are inexact. GHD (n = 71) and TS patients (n = 43) were recruited to study response to r-hGH over 5 years. Analysis was performed using 1219 genetic markers and baseline (pre-treatment) blood transcriptome. Random forest was used to determine predictive value of transcriptomic data associated with growth response. No genetic marker passed the stringency criteria for prediction. However, we identified an identical set of genes in both GHD and TS whose expression could be used to classify therapeutic response to r-hGH with a high accuracy (AUC > 0.9). Combining transcriptomic markers with clinical phenotype was shown to significantly reduce predictive error. This work could be translated into a single genomic test linked to a prediction algorithm to improve clinical management. Trial registration numbers: NCT00256126 and NCT00699855.

Highlights

These authors contributed : Adam Stevens, Philip Murray
We first demonstrated that a fundamental relationship existed between the baseline blood transcriptome and response to Recombinant human growth hormone (r-hGH) over the 5 years of the study (GHD n = 50, Turner syndrome (TS) n = 22) using Discriminant Analysis of Principal Components (DAPC) on the unsupervised baseline transcriptome (GHD = 8875, TS = 8455 gene probe sets)
We examined classification of growth response using random forest (RF) with oversampling by Synthetic Minority Oversampling Technique (SMOTE) to correct for uneven class size (GHD, Table S3A and TS, Table S3B)

Summary

Objectives

This study aimed to identify for the first time the genomic associations that classify response to r-hGH therapy from 1 year up to 5 years of treatment with r-hGH in children with TS and GHD

Methods

Results

Conclusion