Introduction

In mammals, growth hormone (GH) is a polypeptide of 191 amino acids released by the anterior pituitary gland [1]. It exerts its physiological effects on bone and tissue growth as well as in the immune response and reproductive functions [2, 3] by modulating carbohydrate, protein, and lipid metabolisms [4–6]. Initially administered to improve nitrogen balance in aged horses [7], equine GH (eGH) was later used in animal breeding and sports to promote foodstuff production, wound healing, and enhance physical performances.

Large-scale production of recombinant GH (rGH) using DNA technologies [8, 9] and its extensive use has led to the passing of new regulations prohibiting its use as growth promoter. These changes involved the development of analytical methods for short- and long-term detection of reGH abuse in animals. Recently, the challenge of rGH detection in biological matrices has been overcome by the development of successful confirmatory protocols based on LC-MS/MS measurement for rGH detection in animal blood [10–12]. In parallel, different screening tools have been developed. In this context, insulin-like growth factor-I (IGF-I) has been used as indirect biomarker of reGH administration since its concentration increases upon GH administration [13, 14]. This screening method, based on ELISA measurements in plasma samples, suffers from a lack of efficiency since it is strongly subjected to a threshold determination which is difficult to establish for inter- and intra-individual variation reasons (age, sex, circadian rhythm) [15]. Another approach consisting in bioassays, based on the monitoring of antibodies [16, 17] produced as a consequence of reGH administration, is very efficient for reGH long-term detection but carries practical testing limitations due to the delay necessary before inducing the immunological response. Therefore, the need for efficient screening tools is now of prime interest.

Metabolomics is an emerging field of “omics” research that focuses, in an untargeted mode, on large-scale and high-throughput measurement of small molecules (so-called metabolites) in biological matrices. Metabolomics, which is also known as metabolic profiling [18], only became possible as a result of recent technological breakthroughs in small molecule separation and identification. Nuclear magnetic resonance (NMR) and mass spectrometry (MS) are the two main analytical techniques employed in metabolomics to characterise physiological states or assess pharmacological and/or toxic effects of a drug [19]. Although NMR [20] has been traditionally used for these purposes, recent significant advances in MS have contributed to the expansion of its use. In particular, high MS resolution is now available with the development of new types of mass filtres such as the orbital trap (LTQ-HRMS) [21] allowing the detection of low-concentration compounds and with accurate mass determination for efficient subsequent structural elucidation. In terms of chromatographic separation, reversed-phase (RP) chromatography remains the most commonly used for metabolomic profiling by LC-MS; this technique allows a satisfying to fairly acceptable retention of multiple metabolites, with different physico-chemical properties, especially medium to relatively polar compounds [22]. New types of stationary phases such as HILIC [23] (hydrophilic interaction chromatography) have improved the analysis of polar compounds while monolithic columns [24] or fast-LC system have reduced the length of the analysis.

Until recently, most of the work in metabolomics had focused primarily on clinical or pharmaceutical applications such as drug discovery [25, 26], drug assessment [27, 28], clinical toxicology [29, 30], and clinical chemistry [31]. However, over the past few years, metabolomics has also emerged as a field of increasing interest in the context of doping control [32, 33]. Dumas et al. showed, for the first time, how metabolic perturbations induced by hormonal treatment can be evidenced through NMR characterisation of urine samples. Linear discriminant analysis (LDA) discriminated controls from treated animals and also correlated to quantitative physiological responses induced by anabolic steroids. This paper work was the first to demonstrate the suitability of metabolomic approaches as a powerful screening tool for anabolic steroid use in cattle with further potential application in sports and horseracing [32, 33].

In this context, the purpose of the present study was to set up, develop, and assess a new screening method based on global metabolomic profiling to screen for reGH abuse in horses. The idea is to measure the global biological effect consecutive to GH-treatment and not to focus anymore on a list of target biomarkers (IGFs, antibodies). The known effects of growth hormone on the hypothalamo-pituitary axis and as a consequence on carbohydrates (decrease of glucose concentration in blood), lipids (decomposition of fatty acids), and proteins (increase of lean fat) metabolisms suggests that such a strategy would allow pointing out the consequences of GH administration on the animal metabolism and be appropriated for the future development of a new screening method of growth hormone abuse.

Materials and methods

Chemicals

HPLC grade acetonitrile (MeCN) and acetic acid were purchased from Carlo Erba Reactifs (SDS, Peypin, France). Deionised water was produced using an ultra-pure water system (Milli-Q, Millipore Corporation, USA).

Animal experiment

Experimentation was led in agreement with animal welfare rules at the administration and sampling Centre of Fédération Nationale des Courses Hippiques (FNCF, Coye-la-Forêt, France) on one stallion (horse # A) and one gelding (horse # B). A pharmaceutical preparation of recombinant equine growth hormone (EquinGen-5®, BresaGen, Australia) was administered daily to both horses at a concentration of 18 µg kg−1 during the first week and 25 µg kg−1 during the second week. Two additional injections were made on the third week at a concentration of 30 µg kg−1. Urine samples were collected daily during the week preceding reGH administration (from D−7 to D−1), and during the three first weeks after the first injection (Days D+1 to D+20) of the experiment. Samples were then collected every 2 days until D+30 (Days D+22 to D+30). Samples were stored at −20 °C until analysis. ReGH doses were chosen according to the recommendations of the manufacturer. In order to take into account the restricted number of animals, each urine sample was injected three times in the same sequence. All samples were analysed on two randomised sequences of 48 h.

Sample preparation

Two different protocols (solid-phase extraction and filtration followed a freeze-drying step) were used and compared as sample preparation for the metabolomic study.

In the first protocol, the adsorbent was activated and conditioned first with 2 mL of methanol and then with 2 mL of distilled water. Urine sample (1 mL) was loaded onto an Oasis HLB cartridge (60 mg, Waters, Saint-Quentin en Yvelines, France), washed with distilled water (1 mL) and extracted with 4 mL of methanol. The eluent was evaporated under nitrogen and dissolved with 1 mL of distilled water.

In the second protocol, one millilitre of each urine sample was filtered through a 10-kDa membrane (Microsep 10 K Omega, VWR, Fontenay sous bois, France) by centrifugation at 4,000 rpm at 6 °C for 30 min. Filtered samples were then lyophilized for 48 h. The dry mass of each sample was then measured and all samples were reconstituted in deionised water at the same concentration level (i.e. 30 mg mL−1) for normalisation purpose.

All samples were frozen until LC-ESI-HRMS analysis.

LC-ESI-HRMS measurements

Reversed-phase liquid chromatography (RP-LC)

LC separation was performed with a Surveyor quaternary pump (ThermoElectron, Bremen, Germany) on an Uptisphere C18 column (3 µm, 2.1 mm × 150 mm) from Interchim (Montluçon, France) fitted with the corresponding guard-column (2 µm, 10 mm × 2 mm). The injected volume was set at 5 µL. The mobile phase consisted of two solvents: 0.1% acetic acid in water (A) and 0.1% acetic acid in acetonitrile (B). A “generic” elution gradient was used starting from 100:0 to 0:100 (A: B; v/v) over 30 min. The mobile phase was kept at isocratic conditions (100 % B) for 10 min and returned at 100% A in 20 min. Total run time was 60 min. The flow rate was set at 200 µl min−1.

Mass spectrometry

High-resolution MS fingerprints were achieved on a linear ion-trap coupled to an orbital trap instrument (LTQ-OrbitrapTM, Thermo, Bremen, Germany) after positive electrospray ionisation (ESI+). Capillary and cone voltages were set at 5 kV and 30 V, respectively. The other spectrometer parameters were set as follows: capillary temperature, 350 °C; gas cone flow, 40 u.a.; gas auxiliary flow, 6 u.a. MS resolution was fixed at 30, 000 FWHM and data were acquired in the full scan mode from m/z 50 to 800.

Data processing

MS data acquired in the instrument-specific data format (raw data) were converted to NetCDF files for data processing using a conversion software programme (file converter programme available from ThermoFisher Scientific Xcalibur software). The total data set was divided in two folders: the first one corresponding to urine collected before reGH administration (non-treated animal), the second one corresponding to urine collected from the first reGH injection until the end of experiment (GH-treated animal).

The data processing step was achieved using xcms software implemented with the freely available R statistical language [34] (v 2.6). The process was based on four main steps: peak picking, peak grouping, retention time correction followed by a second peak grouping. The final report generated by the software is a table where all MS signals constituting the metabolic fingerprints, i.e. ions characterised by their accurate m/z and retention time values, are ranked according to a p value (statistical significance associated to this difference of abundance) and a ratio (difference of abundance observed between treated and non-treated animals).

Data analysis

Statistica (v 7.1, www.statsoft.com) and SIMCA-P+ (v. 11.5, Umetrics, Sweden) software were used for multivariate data analysis. Orthogonal partial least squares (OPLS) were applied to build descriptive and predictive models. The various MS signals constituting the fingerprints were considered as independent variables while days of the experiment were assigned to the dependent variable. Models built with an OPLS attempt to explain a Y variable which in the present study represents the status of the animal (i.e. GH-treated or Control) from the X matrix of all the ions constituting the fingerprint. The two main components generated by the analysis were as follows: axis 1 is a linear combination of ions selected after data processing; it aims at better explaining the discrimination between GH-treated and non-treated animals (i.e. Y). Axis 2, orthogonal to axis 1, is a linear combination of the ions detected during data processing which aims at restituting the residual variability.

Results and discussion

Metabolomic studies require development of analytic strategies beginning with the choice of sample preparation techniques such as solid-phase extraction or liquid–liquid extraction [35, 36]. In this study, a minimal and selective preparation technique, giving access to a large range of compounds, was developed. In this context, two sample preparation techniques were used and compared. First, solid-phase extraction was performed to clean-up the urine matrix from mucus and salts. A simple filtration step on 10-kD membrane was used in order to remove high-molecular-weight molecules from the urine samples. The freeze-drying step was added to normalise the dry mass of each urine sample. Concerning the liquid chromatography and mass spectrometry parameters, standard conditions were used to separate, analyse and ionise a maximum of molecules. Positive electrospray mode was preferentially chosen because of its low ionisation energy.

Assessment of the analytical variability

Assessment of the repeatability of both sample preparation and MS measurement is a crucial step in conventional analytical chemistry, and even more in the field of metabolomics. Given that all data processing and subsequent statistical analysis are based on the comparison of MS signal abundances measured in each analysed sample, it is absolutely necessary to ensure that MS signal variations are not related to MS measurement and/or sample preparation. Otherwise, no confidence level could be granted to a difference which will be observed between two or several analysed samples. To assess the repeatability of MS measurement and sample preparation steps, a number of samples collected during the experiment have been prepared and analysed several times. Sample preparation repeatability was evaluated by extracting the same urine samples five times while MS fingerprinting process repeatability was assessed through five repeated analysis of the same urine extract. Relative standard deviations (RSD) were calculated for all ions detected after the processing step.

It appeared that the filtration followed by a freeze-drying step as sample preparation protocol did not induce any analytical variability since more than 85% of the ions constituting the metabolomic fingerprints exhibited relative standard deviations lower than 25% with regard to their observed intensities. Reproducibility of this protocol was therefore considered as very acceptable, especially when compared to that of the SPE-based protocol from which only 50% of the ions showed acceptable RSD and even if 50% RSD might be acceptable, depending on the finally aimed significance of regulation factors. According to this result, the filtration and freeze-drying steps were preferred, and were chosen as clean-up protocol in order to prepare all urine samples involved in the present study.

Results, as presented in Fig. 1, show that most RSDs of the ions constituting the fingerprints vary from 0 to 30 % with very few outliers for both MS measurement and sample preparation repeatability. Consequently, the analytical variability associated to the present developed method was found satisfactory since (a) the observed values are in accordance with variability commonly accepted for conventional targeted measurement and (b) some normalisation solutions may be used to further decrease these values if needed.

Fig. 1
figure 1

Assessment of MS measurement (a) and sample preparation (b) repetabilities. Five aliquots of horse B (Gelding) urine collected at D-2 were prepared and analysed simultaneously for sample preparation assessment and one aliquot of the same urine was prepared and injected five times for MS measurements evaluation. Analyses were performed on LC-ESI-LTQ-Orbitrap in positive mode following by a processing step of MS fingerprints by xcms software. Relative standard deviations were then calculated on ions sorted out by xcms in order to evaluate MS measurements (a) and sample preparation (b) repeatabilities

Full size image

Effect of reGH administration on global horse metabolism

Considering the acceptable repeatability of the fingerprinting strategy defined in this study, the potential effects of reGH on the metabolome of animals was further investigated. From the LC-HRMS analysis, data were processed by xcms software in order to compare the urine fingerprints collected before reGH (D−7 to D−1) treatment with those collected after reGH administration (D+3, D+7, D+10 and D+ 25). From this analysis, a total of 2,000 ions were pointed out after processing by xcms.

A first possible way to assess the effects of reGH consists in plotting the p values of ions detected during the processing step (Fig. 2) from the more to the less significant. This figure allows a rapid and visual global overview of the metabolome perturbation in a time-related (kinetic) way. The results show that the different curves, corresponding to different days of urine collection, are not superimposed and change according to time and thereby to treatment length. The discontinued line in Fig. 2 represents a statistically significant p value limit (p ≤ 0.05). Classically, ions with p values below 0.05 are considered as ions exhibiting a significant variation of abundance between groups to be compared (in this study, non-treated and treated animals). The higher the number of ions below this limit, the more the metabolome of the animals is modified. From this point of view, the different curves seem to indicate that the metabolome of the animals is globally and significantly modified after reGH treatment. In the present study, the metabolome is the most affected 25 days after the first administration which seems to reveal a long-term effect of equine growth hormone. This result is consistent with previous experiments [37, 38] but it should however be underlined that metabolome changes could also be induced by other external factors such as wounds, inflammation, stress, etc … which may participate to these metabolome changes even if rearing conditions (age, feed, exercise) were controlled during the experiment.

Fig. 2
figure 2

Evaluation of the impact of reGH administration on global metabolome of horses A and B. MS fingerprints of urine collected at D−7 to D−1, D+3, D+7, D+10 and D+25, obtained by LC-ESI-LTQ-Orbitrap, were processed by xcms software. The p value of all ions detected during this step was plotted on a graphic according to their rank.

Full size image

Finally, this first set of results demonstrated that the administration of reGH globally induces significant modifications of urine metabolome and that these modifications are accessible and revealed by the present analytical MS fingerprinting strategy.

Multivariate statistical analysis to build up a descriptive and predictive model

An OPLS analysis was performed on the MS intensities of the 2,000 ions pointed out by the xcms software and resulting from the comparison of the urine fingerprints recorded before and after reGH administration (Fig. 3). Two components R 2(Y) and Q 2(Y) were extracted of the model to assess the percentage of Y variance which can be explained and predicted by the model, respectively. Hotelling’s T2 range (represented by an ellipse on Fig. 3) was represented to evaluate the number of outliers among the sample at a confidence level of 95%.

Fig. 3
figure 3

Evaluation of the impact of reGH administration to horses on their urinary MS metabolomic fingerprints by using a supervised multivariate analysis. Data provided from urines collected at D−7 to D−1, D+3, D+5, D+7, D+10, D+15, D+19, D+25, were separated specifying control (0) and GH-treated (1) groups. The mass spectra were processed using xcms software for background suppression, peak matching, peak alignment. The MS abundances obtained for each variable (ions detected) were then analysed by Orthogonal Partial Least Square (OPLS) by means of SIMCA-P software

Full size image

The efficient discrimination of urine extracts collected before (0) and after the treatment (1), presented on Fig. 3, is correlated to a high value of R 2(Y) = 0.85. Moreover, the encouraging value of Q 2(Y) = 0.7 suggests that the model could be used for the prediction of future data. Hotelling’s T2 range indicates the absence of outliers which can give additional confidence to the results. The prediction ability of the previous OPLS model was assessed by a new OPLS built with 2/3 of the previous dataset (Fig. 4). The other one third of the dataset was considered as a validation test and was incremented in the model for the prediction. The results show the possibility to build a reliable descriptive model (R 2(Y) = 0.9 Q 2(Y) = 0.8) with two thirds of data and to use this model for prediction of the remaining one third of data.

Fig. 4
figure 4

Assessment of the prediction ability of the OPLS model build in the Fig. 3. Two thirds of data was used for the building of predictive model. One-third data was used for the prediction. Data provided from urines collected at D−7 to D−1, D+3, D+5, D+7, D+10, D+15, D+19, D+25 were separated specifying control (0) and GH-treated (1) groups for the descriptive model and control (2) and GH-treated (3) groups for the predictive model

Full size image

Selection of potential discriminative ions

The OPLS performed previously and presented on Figs. 3 and 4 allowed the building of a tool able to discriminate treated from control samples and to predict new samples. In the spirit of developing a future detection method to screen for growth hormone administration, based on monitoring new metabolites, a focus on the ions which intervene the most in the discrimination of non-treated and GH-treated animals must be performed. S-plot representation was carried out to select the variables which are closely correlated to axis of Fig. 3 and thus highlighting potential biomarkers of reGH administration. The coordinates that are plotted in the S-plot from the predictive component are p1 vs. p(corr)1, representing the magnitude (modelled covariation) and reliability (modelled correlation), respectively (Fig. 5). This approach enabled to highlight the presence of twenty ions strongly correlated to the axis (Table 1). To confirm the ability of these ions to discriminate treated from non-treated animals, an OPLS was performed on these 20 ions (Fig. 6) which showed a very efficient discrimination power (R 2(Y) = 0.85) of samples collected before (0) and after treatment (1) demonstrating a high discriminating ability of the twenty statistically relevant ions. Moreover, the examination of their ion chromatograms confirms the good chromatographic resolution of these ions as well as their abundance differences observed between control and GH-treated animals (Fig. 7) and confirms their interest as potential biomarkers, for which specific monitoring methods will, in the future, have to be developed according to their individual chemistry. All these results indeed take place in the context of the development of new screening strategies applied to the control of growth promoter use for which similar studies have been recently carried out [39, 40].

Fig. 5
figure 5

S-plot representation of variables (ions detected) intervening in the formation of axis of OPLS presented in Fig. 3 to visualise the ions which participate the most in the discrimination of control and GH-treated animals

Full size image
Table 1 List of ions that were the most correlated to the discriminant axis and that were selected as potential biomarkers
Full size table
Fig. 6
figure 6

Orthogonal partial least square performed on the MS abundance of 20 ions selected for their high contribution in the formation of axis of OPLS presented in Fig. 3. After the processing of MS fingerprints of urines collected at D−7 to D−1, D+3, D+5, D+7, D+10, D+15, D+19 and D+ 25 by xcms software, an OPLS (Fig. 5) was performed on all ions detected. Only the 20 ions the most correlated with axis of this OPLS were selected in order to evaluate their role in the discrimination of control animals (0) and treated animals (1)

Full size image
Fig. 7
figure 7

Extracted ion chromatograms (XIC) of two ions selected among the twenty ions chosen for the OPLS analysis presented in Fig. 6. These ions are present with different MS abundances in the urine collected at D−7 to D−1, D+3, D+5, D+7, D+10, D+15, D+19 and D+ 25

Full size image

Towards the development of a new screening method

The strategy presented and assessed in the present article was set up on the basis of biological samples collected on two horses. Although the results show changes in horse metabolism as a consequence of reGH administration, the metabolomic models presented in this paper cannot be used to date as it is and as a new tool for reGH screening. Indeed, a higher number of samples collected on a large number of animals should be analysed to take into account the inter-animals variability and thus lead to the building of more robust metabolomic models. However, this method has the merit of demonstrating the potential of metabolomic studies in the future development of new screening tools applied to anti-doping

Conclusions

The metabolomic approach developed in the present study was based on the comparison of horse urine fingerprints obtained by LC-ESI-HRMS. This study required the preliminary assessment of repeatability for both sample preparation and mass spectrometry measurements to evaluate the analytical variability associated to the method. The results showed that the induced variability was satisfactory and therefore not a limiting factor in the proposed strategy. Considering this acceptable repeatability, the protocol was applied to samples collected on reGH-treated animals in order to investigate its ability to highlight metabolome’s modifications upon growth hormone administration. The analysis of recorded fingerprints showed that it was possible to give evidence for modifications in the metabolome of horses after reGH administration. Moreover, it was found that the metabolome seemed to be the most affected at the end of treatment which may indicate a long-term effect of reGH, which is coherent with known effects of the growth hormone. Finally, some 20 ions responsible for the discrimination between treated and non-treated animals were pointed out as potential biomarkers of such a hormonal treatment. These preliminary results demonstrate the relevance of such an untargeted approach in the context of growth hormone abuse screening. The strategy is now being applied to a large set of samples collected on heterogeneous populations. This study will help to know the influence of the inter-individual variability on the establishment of metabolomic models and whether this new technique can be transposed and compatible in a routine laboratory.