Biochemical Signatures in Growth Hormone Deciency: a Pilot Study Using Metabolomics Approaches by Direct Infusion-mass Spectrometry

Purpose: The discovery of biomarkers to detect growth hormone deciency (GHD) and monitor growth hormone replacement therapy (GHRT) remains challenging. Among “omics” technologies used for screening biomarkers, metabolomics stands out as a powerful tool for large-scale identication and quantication of small molecules present in biological matrixes. Metabolomic proles allow us to infer the phenotypic state; therefore, metabolomics is a great ally in investigating biomarkers and understanding biological processes. Methods: In this study, global metabolomics (N = 39; range scan 50-600 m/z) and lipidomics (N = 36; range scan 50-1500 m/z) approaches were performed by high-resolution direct infusion-mass spectrometry. Partial least-square discriminant analysis (PLS-DA) models were used for data-driven diagnosis of GHD and evaluation of GHRT (VIP score >1.5). Cross-validation, permutation test, and ROC curves conrmed the predictive accuracy of PLS-DA models. The features were annotated using accurate mass measurements matched against the metabolomics database. Results: Data analysis revealed changes in the class of proteinogenic/glycogenic acids, carnitines, n-acyl-amines, unsaturated fatty acids, and sulfur amino acids, and pathway analyzes revealed changes in glycerophospholipid metabolism. Regardless of GH therapy, GHD individuals remain with changes in lipids and amino acids compared to healthy control. Conclusion: GHRT inuences the metabolism status of GHD patients in order to compensate for dysregulations caused by GHD. The data has corroborated the action of GHRT and indicated new potential biomarkers for treatment follow-up. pathophysiology, tests designed to measure levels of the factor lack precision, specicity, and robustness. current paper presents a cross-sectional study design that assessed adult patients with congenital growth hormone deciency (GHD) treated with GHRT compared with healthy adults. Inclusion criteria as follows: be in adult age; for the patient’s was preliminary proper compliance prescribed Exclusion criteria active smokers; diagnosis poor samples collected in endocrinology were used for the mass spectra acquisition: ion source voltage of 3.5 kV (both ionization modes); HESI source temperature of 60 ºC; the capillary temperature of 250 ºC; and multiple m/z range acquisitions. Mass spectrometric data were acquired by direct infusion at a ow rate of 5 µL min − 1 . The mass spectrum scanning range used in metabolomics ranged from 50 to 600 m/z, and lipidomics ranged from 50 to 1500 m/z. The samples were analyzed randomly to avoid bias.


Introduction
The pituitary gland sometimes referred to as the body's "master gland," controls the production of several hormones that are essential for human development, such as ACTH (adrenocorticotrophic hormone), responsible for the production of cortisol; LH (luteinizing hormone), linked to ovulation or testosterone production; PRL (prolactin), related with breast milk production after childbirth; and GH (growth hormone), responsible for the human growth, including muscular and bone effects [1]. When the pituitary gland fails to produce one or more hormones, it causes congenital hypopituitarism, estimated to affect one in 3000-4000 newborns. Among hypopituitarism disorders, one of the most relevant is the disturbance in GH production, causing severe disruption in infant development [2]. GH plays a fundamental role in postnatal growth, with several biological functions, including intermediate lipid metabolism and homeostasis [3,4]. Consequently, growth hormone de ciency (GHD) is often accompanied by obesity, hyperlipidemia, abnormalities in cardiac structure, and bone mass decrease [5]. Most of the actions of GH are mediated by IGF-1 (insulin-like growth factor 1); therefore, GHD may occur not only by GH de ciency itself but also due to abnormalities in the GH-IGF-1 axis [6].
In adulthood, the GHD condition is a well-known clinical condition caused by genetic factors or acquired injury and leads to abnormal body composition, poor quality of life, dyslipidemia, increased cardiovascular risk, and mortality [7]. These symptoms can be mitigated with growth hormone replacement therapy (GHRT); however, adverse effects are also reported (e.g., insulin resistance, hypertension, uid retention). Although been a well-de ne condition, the lack of uniformity of the disease and analytical protocols remains a challenge for the precise diagnosis of GHD [4,8].
For GHD diagnosis, it is necessary to evaluate extensive biochemical and imaging exams, including quanti cation of IGF1 and insulin-like growth factor binding protein-3 (IGFB3) to assess GH function indirectly. However, only IGF1 or IGFB3 levels alone are neither sensitive nor speci c enough to diagnose hypofunction, and it usually is necessary to expose the patients to more stressful and particular tests that induce maximum viable GH secretion by hypophysis (e.g., arginine, clonidine, and hypoglycemia tests) [9][10][11]. Even though IGF-1 is still considered a biological marker for GHD pathophysiology, tests designed to measure levels of the factor lack precision, speci city, and robustness.
Metabolomics e ciently creates explanatory and individualized physiological panoramas that re ect the phenotypic state [12,13].
Notwithstanding, only a few papers reported the application of omics for GHD studies [14][15][16][17][18]. In a recent review published by this group, San-Martin and Cols (2020) compiled the studies regarding GHD and considered metabolomics an essential and underused tool for the diagnosis of GHD due to differences in the lipid metabolism between the evaluated cohorts (e.g., fatty acids, β-oxidation, glycerolipids) [19]. In metabolomic studies, liquid chromatography coupled to mass spectrometry (LC-MS) is consecrated; however, with better resolution mass spectrometers, it is possible to enhance a broad metabolic coverage by directly injecting samples into the ionization source [13,[20][21][22]. Moreover, metabolomics can be applied not only for biomarker screening but also for a better understanding of GHD and GHRT, contributing to the growth of individualized medicine, the concept when therapies are provided according to the needs of individuals on a case-by-case basis [19].
This pilot study applied metabolomic and lipidomic pro ling by direct infusion mass spectrometry (DIMS) to congenital GHD patients with and without GHRT. We evaluated the metabolism effects of the disease and its treatment with growth hormone replacement therapy, showing signi cant differences between the groups regarding proteinogenic/glycogenic amino acids, carnitines, n-acyl-amines, unsaturated fatty acids, and sulfur amino acids, as the signi cant changes in glycerophospholipid metabolism.

Study design and settings
The current paper presents a cross-sectional study design that assessed adult patients with congenital growth hormone de ciency (GHD) treated or not with GHRT compared with healthy adults. Inclusion criteria were as follows: be in adult age; for the patient's group was necessary a preliminary diagnosis of GHD and be in proper compliance with prescribed treatment (at least in the past six months). Exclusion criteria were: active smokers; a diagnosis of diabetes mellitus or arterial hypertension; patients with poor compliance (Figure 1).
The samples were collected in the hypopituitarism outpatient clinic developmental endocrinology unit of the University of São Paulo Medical School (FMUSP). In the clinic, 318 patients were initially diagnosed with congenital hypopituitarism, among which 217 are under active monitoring, 83 patients in childhood and transition, and 134 in adulthood. The endocrinology unit of FMUSP is a national reference for GHD patients, an outcome of several years of intense care for hypopituitarism patients. Considering the rarity of the condition, the clinic holds a signi cant number of GHD patients in the country. After assessing inclusion and exclusion criteria, 39 patients were included in this pilot study ( Figure 1). Amongst 39 GHD patients, 25 were non GHRT (GHDoff), while 14 were undergoing GHRT (GHDon), and 14 individuals were selected as healthy controls (Figure 1). This paper's primary goal consists of exploring GHD through an exploratory metabolic analysis; therefore, three comparative data on metabolomics and lipidomics were performed (i.e., GHDon versus GHDoff, GHDon versus Healthy Control, GHDoff versus Healthy Control). Ethical committee approval can be accessed at CAAE no. 98372718.7.0000.0068 (University of São Paulo) and CAAE no. 98372718.7.3001.5505 (Federal University of São Paulo) authorization. All patients signed a term of consent for participation in the study.

Metabolite and lipid extraction (plasma)
Ice-immersed blood samples were vortexed for 2 min in ice, and a 100 µL-aliquot of the plasma was pipetted into an Eppendorf® tube for processing; 300 µL of a methanol/ethanol (1:1 v/v) solution was added to ensure protein precipitation in each tube. Samples were vortexed for 1 min and incubated for 5 min in an ice bath. Finally, the samples were vortexed for a few seconds and centrifuged at 13,000 rpm (15.7 rcf) for 20 min at 4 ºC. The supernatants (hydromethanolic extracts) were stored at -80 ºC and used for global metabolomics [23].
For the lipidomics approach, 30 µL of plasma was transferred to an Eppendorf® tube, and 225 µL of methanol was added in each tube, followed by vortexing for 10 s. Posteriorly, 445 µL of chloroform and 185 µL of water were added and stirred for one hour. After separating the organic phase (lipids) from the aqueous phase (primary metabolites), the phase with lipids was dried and stored at -80 ºC. On the day of the analysis, the lipidic extracts were resolubilized in ethyl acetate.
As internal standard (IS), a solution containing 15 N and 13 C labeled tryptophan in water with a nal concentration of 40 µM was prepared, and 20 µL was spiked into each sample. Before direct infusion to the mass spectrometer (DIMS), the samples were ltered using a 0.22 µm nylon syringe lter.

Quality control samples (QC)
Quality control samples (QC) were prepared from an aliquot of 20 µL of each sample and mixed thoroughly. QCs samples were later processed as a regular sample, following the protocol in section 2.2. One QC analysis was carried out throughout the experimental design every ve analyses of the random samples. QC samples were displayed on a PCA plot to assess the stability and performance of the analytical system.
The following conditions were used for the mass spectra acquisition: ion source voltage of 3.5 kV (both ionization modes); HESI source temperature of 60 ºC; the capillary temperature of 250 ºC; and multiple m/z range acquisitions. Mass spectrometric data were acquired by direct infusion at a ow rate of 5 µL min − 1 . The mass spectrum scanning range used in metabolomics ranged from 50 to 600 m/z, and lipidomics ranged from 50 to 1500 m/z. The samples were analyzed randomly to avoid bias.

Data pre-processing and alignment by m/z ratio
The .RAW les from the MS analysis were converted to .csv using the Xcalibur® software (version 3.0.63, Thermo Fisher Scienti c). To better understand the pre-processing and data alignment of DIMS analysis, results must consist of a table containing two columns: m/z and intensity, with one m/z ion and its respective intensity in each row. Different samples contain different ions and generate distinct m/z columns between samples. For multivariate statistical analysis, one column must remain constant in all samples; in LC-MS data, for example, time is the column used as constant. On DIMS analysis, there is no separation, so time cannot be used as constant. Therefore, it is necessary to align the m/z columns from all samples.
In this work, we developed a user-friendly software called MassAligner, that creates a column containing all the acquired m/z in all samples but removing redundancies. After that step, it searches in the original table from each sample the intensity of each m/z in the column. Once an m/z is not present in that particular sample, the software input zero as intensity, and thereof, the nal result is a worksheet within all the samples, with the same m/z column (i.e., a constant column). MassAligner allows round the decimal cases in the m/z column since high-resolution m/z data requires high computational power.
After alignment, the data was processed using the MetaboAnalyst 5.0 platform (https://www.metaboanalyst.ca/) to normalize and prepare the data for multivariate analysis. This study applied normalization, followed by log transformation and Pareto scaling to transform the data matrix into a Gaussian-type distribution. The MassAligner software and raw data are available in a cloud-based repository (http://dx.doi.org/10.17632/dd3p5tfcd7.1 and http://dx.doi.org/10.17632/8z736ssxdj.1, respectively).

Statistical analyses
Principal component analysis (PCA) was performed to verify the data grouping, trends, variables selection, and/or outliers from the group. For group classi cation, Partial least squares discriminant analysis (PLS-DA) was carried out on the data. The predictive ability and statistical signi cance of the PLS-DA model were accessed through cross-validation applying a permutation test with 1000 interactions, and ROC curves for features with Variable importance for the projection (VIP) score >1.5 were performed for accuracy veri cation. Only 36 of the 39 samples were used for data analysis in the lipidomics approach due to experimental problems in three of them. In total, 12 PLS-DA and 4 PCA analyses were performed and are available as supplementary material together with the permutation tests and ROC curves.

Putative annotation, enrichment, and pathways analysis
METLIN, Human Metabolome Database (HMDB), and LIPIDMAPS databases were employed for the putative annotation. A mass error of 5 ppm was set as the maximum tolerable variation from the measured m/z. Chemical modi cations (adducts) in both ionization modes were considered. Enrichment and pathways analysis was performed in MetaboAnalyst (version 5.0).

Anthropometric and clinical data
Clinical and anthropometric parameters from the patients are presented in Table 1. It is interesting to highlight that among all GHD patients, 12 have isolated GHD diagnosis, i.e., only GHD hormone is unregulated. In contrast, 27 patients have combinate GHD diagnosis, i.e., GHD and other(s) hormones are unregulated. Therefore, due to similar hormonal conditions, some patients utilize levothyroxine, sex steroids, and glucocorticoids. Although the difference in patients under GHD therapy on and off is not statistically signi cant, the therapy may contribute to the alterations between the GHD and control groups.
Although higher triglycerides and lower HDL levels were observed in the GHDoff group, there was no signi cant statistical difference between the groups in any classic clinical lipid pro le parameter (Table 1). It is important to note that three patients in the GHDoff group were under statin therapy for control of cholesterol levels; nonetheless, once these samples were excluded, the median and statistical signi cance of the clinical classic lipidic pro le between the groups had no alteration; therefore, the three samples were used for the omics studies.
From the anthropometric data, one may notice a signi cant difference in Body Mass Index (BMI) between GHDoff vs. GHDon, with a higher percentage of eutrophic individuals between control patients and GHDon group; these body characteristics are closely related to the role of GH in the body composition. There is a signi cant difference in IGF-SD between GHD and controls patient, but no difference between GHDon vs. healthy controls, indicating that the therapy could correct the IGF-1 to normal levels.

Statistical analysis
Initially, PCA analysis was applied to the data to search for grouping tendencies and analytical variation; however, no clear group separation could be observed in PCA analysis for metabolomics and lipidomics data. The QCs clustering can establish the robustness of the analytical method in metabolomics approach analysis ( Figure 2A). The other PCA gures for analysis of lipidomic and metabolomic approaches are available in the supplementary material in Figure S1.
Considering that the PCA model could not clearly show the grouping of the samples, PLS-DA, a supervised multivariate analysis, was employed in the data. Figures 2B and 2C report the PLS-DA model and the coe cients that infer the tting and predictive ability of each PLS-DA. Although the R 2 and Q 2 metrics indicate the reliability of the generated model, metabolomics and lipidomics PLS-DA analysis were validated by permutation test and area under the curve (AUC) of the ROC curve for assurance of statistical signi cance and predictive performance of the model. The PLS-DA shown in Figure 2C shows the GHDon vs. GHDoff (+ mode) group comparison for the metabolomic approach; Figure 2D and 2E represent the permutation test and roc curve used in the model validation, respectively. In the supplementary material, all the PLS-DA comparison models are presented in Figure S2 for the metabolomics approach, followed by Figure S3 with the ROC curves for all the PLS-DA models and the permutation tests in Figure S4. For the lipidomics approach, the PLS-DA models are presented in Figure S5, and the ROC curves and permutation tests in Figure  S6 and S7.
For multivariate exploratory analysis based on the ROC curve, only features with VIP > 1.5 were selected. ROC curves for the comparisons of GHDon vs. GHDoff, GHDon vs. healthy control, and GHDoff vs. healthy control are reported in Figure 3.
After the reliability check of the PLS-DA model, the annotation of the features was performed based on the high-resolution m/z data, based on the set of variables that distinguished the groups and allowed the veri cation of the similarities in the annotated species, regardless of the GHRT and "omics" approach ( Figure 3). The area under the curve (AUC), the 95% con dence interval, and the p-value for each variable putatively annotated from the univariate ROC curve is reported in Table S1.

Putative annotation, enrichment, and pathways analysis
Metabolomics standards initiative (MSI) [24] guidelines and International lipid classi cation and nomenclature committee (ILCNC) [25] guidelines were adhered to for the annotation of signi cant metabolites (Table 2). Table 2 reports the database correspondences based on the precise mass within a low mass tolerance (5 ppm) corresponding to a single chemical formula; identi cation numbers (ID) of each annotated feature are available in Table S1 and Table S2.
Metabolomics and lipidomics approach of the GHD group revealed features inherent in the changes caused by GHD in adults and the action of GHTR (Table 2). Twelve PLS-DA models were constructed using two-class input (i.e., GHDon vs. GHDoff; GHDon vs. healthy; GHDoff vs. healthy, both with negative and positive ion mode). The enrichment analyses showed the main chemical classes of annotated features in metabolomics and lipidomics approaches ( Figure 4A and 4B). Pathway analyzes showed metabolic changes linked to GHD ( Figure 4A and Figure 4B). For enrichment and path analysis, only the features of comparisons the GHDon vs. control healthy and GHDoff vs. healthy control were used to build them. Figure 4C presents the main classes of altered metabolites and their path according to the therapeutic conditions (i.e., with GHTR and without GHRT).

Discussion
The IGF-1 as a biomarker did not show differences between GHDon and healthy control (Table 1); however, the metabolomics/lipidomics could predict signi cantly altered sets of small molecules. Consequently, the use of IGF-1 as a biomarker is again considered reductionist and not very sensitive during adulthood. The low speci city of biomarkers for GHD and the evident change in lipid metabolism led to the application of nontargeted metabolomics/lipidomics in the search for new sources of biomarkers linked to the phenotypic state of GHD (i.e., downstream readouts of cellular signaling, transcriptomic and proteomic changes).
Metabolomics and lipidomics analysis by DIMS showed good coverage of the GHD metabolome; notwithstanding, it is important to highlight those similar results can be achieved computational algorithms in treating DIMS data, then such platforms require previous knowledge in computing programming. For this reason, we developed MassAligner software to be user-friendly, regarding previous programming knowledge.
DIMS may be a more practical alternative in large-scale metabolite studies due to lower time consumption in data acquisition, and MassAligner Software guaranteed the accessible Metaboanalyst® platform.
The PLS-DA model proved to be e cient in the data drive prediction of GHD through the metric values of the cross-validation; prediction coe cients (Q 2 ) are in agreement with those suggested in metabolomic studies (i.e., Q2> 0.6) ( Figure 2) [26]. Permutation test and ROC curve corroborate the observed statistical difference between the sample means and predictive performance of PLS-DA models [27].
Other studies have already demonstrated the potential of metabolomics to elucidate biochemical pathways and unknown metabolites [28,29].
Among the Omics sciences applied in GHD, metabolomics is scarcer than the large-scale molecular study techniques used in molecular biology (i.e., transcriptomics) [17,18]. A recent study using single-cell transcriptomics has provided insights into the transcriptional landscape of human pituitary development, which will allow data integration for the development of multi-omics GHD studies [18].

Metabolomics approach
Although the metabolomics sample preparation protocol favors polar metabolites, the glycerophospholipid class produced signi cant features that classify GHD ( Figure 4A and 4B). It is interesting to highlight that the participation of glycerophospholipids can be established from the structural composition of cells, mediation of cellular responses, and precursors of biological effectors [30][31][32]. Thus, the glycerophospholipid alterations in the GHD population were considered: 1) subsequent action of the change in fatty acids; 2) high oxidative stress coupled with the imbalance of reactive oxygen species can destabilize membranes glycerophospholipids; 3) sources of building blocks for biological effectors.
Amino acids also stood out as signi cant metabolites to differentiate GHD individuals from healthy controls, which directly re ects the ability of metabolomics and its branches to create panoramas of the studied phenotype. The pathways analysis indicated a signi cant change in the metabolism of the following amino acids: glutamine/glutamate, taurine/hypotaurine, cysteine/methionine, arginine/proline, alanine/serine/glutamate ( Figure 4B). Alteration of amino acids metabolism are re ective of the metabolic actions of homeostasis during the lack of GH as a biological effector so that it can lead to hypoglycemia (i.e., GH acting as an insulin counter-controller); consequently, these proteinogenic and glycogenic amino acids are redirected to catabolism (e.g., glutamine, arginine, proline, alanine). Amino acids can be obtained from proteolysis of muscle tissues and follow as an energy substrate in GHD [33]; hence, glycolysis/gluconeogenesis and the TCA cycle are intermediary pathways that suffer from substrate alternation.
The change in the pathways of sulfuric amino acids (e.g., taurine/hypotaurine, cysteine /methionine) makes up the need to use the thiol group available in these amino acids for various molecules with biological actions (e.g., Glutathione) [34]. From an endocrinological point of view, this condition may be related to the fact that several biomolecules require the help of sulfotransferase enzymes to perform their metabolic functions effectively (e.g., steroids, bile salts, peptides) [35]. With proteomic analysis and omics data integration, it will be possible to determine which enzymes are involved with oxidative stress in GHD conditions.
Protein synthesis uses arginine in the anabolic process, and glutamine is an essential precursor for arginine biosynthesis [36,37]. The annotation of glutamic acid, cysteinyl-proline peptide, arginine, capreomycidine directly impacted the arginine metabolism in GHDon and GHDoff conditions (Table 2)

Lipidomics approach
As the metabolomics results suggested the importance of lipids in GHD patients, a lipidomics approach was carried out in the samples. The results from the new approach corroborated with the previous study, indicating a signi cant change in the glycerophospholipid class ( Figure 3A and 4A). It was observed that the enrichment analysis revealed important species of lipids that act as biological effectors during the bioenergetic metabolism of GHD, such as acyl-carnitines, glycerolipids, unsaturated fatty acids, and n-acyl-amines. These classes of metabolites proved to be signi cant regardless of therapeutic approaches in individuals with GHD ( Figure 4A). In parallel, metabolomics pointed to similar lipid species, although only the glycerophospholipids were signi cant in the metabolomics pathway analysis ( Figure 4B).
Another metabolic difference in GHD patients revealed by the lipidomics approach was the change of fatty acyls metabolism, noted by the putative annotation of lipids such as Palmitic acid; Docosanoic acid; Myristic acid; 8-Hydroperoxylinoleic acid. It must be noted that Fatty Acyls correspond to a "super" lipid class formed by the loss of hydroxyl from the carboxy group of a fatty acid. The synthesis and oxidation of fatty acids are well de ned by the enzymes ACC1 and ACC2 that regulate a set of reactions in the cytosol and mitochondria, respectively [39]. The condition of GHD prevents insulin counter-regulation; therefore, there will be greater intensity in fatty acid elongation, which explains one of the parameters that unbalance the metabolism of fatty acids [33,40].
Moreover, acyl-carnitines such as 3-hydroxyoctanoyl carnitine; Linoleoyl carnitine were up-regulated in the GHDoff individuals (Table 2). These unbalanced lipids are participants in biosynthesis and lipid oxidation metabolism, indicating that the absence of GHRT in adulthood affects bioenergetic metabolism [40].
Lipolysis is directly linked to GH stimuli; however, the absence of GH does not correctly stimulate the use of lipid reserves found in adipocytes to obtain energy and/or intermediate metabolites [41]. At the same time, the proteolysis of muscles becomes a source of energy, which is signi cant in GHD [42,43]. Treep and Cols (2008) showed that GH therapy stimulates oxidation of lipids originating from skeletal muscle in GHD patients [44]; therefore, GH is an important parameter that regulates the use of these lipid reserves e ciently, and metabolomic/lipidomic pro le could guarantee a better understanding of GHRT.
Phosphatidylcholine, phosphatidylserine, phosphatidylglycerol, and phosphatidic acids are up-regulated features in GHD individuals regardless of GHRT (Table 2). Differences between patients on GHRT and healthy controls also remained signi cant in the work of Abd Rahman (2013).
GHD condition has been shown to limit lipids as an energy source and could subsequently affect other biochemical pathways that use lipids, such as in ammation, cognition, metabolic homeostasis, and transport [45,46]. The imbalance in lipid diversity can affect essential unsaturated fatty acids that are precursors or intermediates of biological effectors (i.e., prostaglandins, glycerophospholipids, and post-translation modi cation (PTM) of proteins with acyl groups regulates various biological processes [47,48].

Conclusion
The di culty in nding robust biomarkers for GHD and monitoring GHRT action in adults remains a challenge. The search to understand metabolome in GHD patients aimed to clarify the leading products and/or intermediates in the metabolism of GH de ciency, contributing to elucidating the GH axis. Several parameters in uence changes in the lipid pro le; however, further study using metabolomics/lipidomics and data integration with the other omics will be essential to create individualized and explanatory panels on the pathophysiological condition of GHD. Metabolomic and lipidomic approaches showed consistent changes in the GHD metabolome and correlated them with the treatment approach.
Notwithstanding, other targeted studies with a more signi cant number of individuals must be carried out to validate the unregulated lipids and/or amino acids as GHD biomarkers.

Declarations
Con ict of Interest: The authors declare that they have no con ict of interest with this paper. All patients provided written informed consent for participation in this study. Only after the written consent, their blood was sampled.