Targeted Metabolomics Approach To Detect the ... - ACS Publications

Subscriber access provided by The University of Melbourne Libraries

Article

Targeted Metabolomics Approach to Detect the Misuse of Steroidal Aromatase Inhibitors in Equine Sports by Biomarkers Profiling George Ho-Man Chan, Emmie Ngai Man Ho, David Kwan Kon Leung, Kin-Sing Wong, and Terence See Ming Wan Anal. Chem., Just Accepted Manuscript • DOI: 10.1021/acs.analchem.5b03165 • Publication Date (Web): 03 Dec 2015 Downloaded from http://pubs.acs.org on December 12, 2015

Just Accepted “Just Accepted” manuscripts have been peer-reviewed and accepted for publication. They are posted online prior to technical editing, formatting for publication and author proofing. The American Chemical Society provides “Just Accepted” as a free service to the research community to expedite the dissemination of scientific material as soon as possible after acceptance. “Just Accepted” manuscripts appear in full in PDF format accompanied by an HTML abstract. “Just Accepted” manuscripts have been fully peer reviewed, but should not be considered the official version of record. They are accessible to all readers and citable by the Digital Object Identifier (DOI®). “Just Accepted” is an optional service offered to authors. Therefore, the “Just Accepted” Web site may not include all articles that will be published in the journal. After a manuscript is technically edited and formatted, it will be removed from the “Just Accepted” Web site and published as an ASAP article. Note that technical editing may introduce minor changes to the manuscript text and/or graphics which could affect content, and all legal disclaimers and ethical guidelines that apply to the journal pertain. ACS cannot be held responsible for errors or consequences arising from the use of information contained in these “Just Accepted” manuscripts.

Analytical Chemistry is published by the American Chemical Society. 1155 Sixteenth Street N.W., Washington, DC 20036 Published by American Chemical Society. Copyright © American Chemical Society. However, no copyright claim is made to original U.S. Government works, or works produced by employees of any Commonwealth realm Crown government in the course of their duties.

Page 1 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Analytical Chemistry

Targeted Metabolomics Approach to Detect the Misuse of Steroidal Aromatase Inhibitors in Equine Sports by Biomarkers Profiling

George Ho Man Chan,* Emmie Ngai Man Ho, David Kwan Kon Leung, Kin Sing Wong and Terence See Ming Wan* Racing Laboratory, The Hong Kong Jockey Club, Sha Tin Racecourse, Sha Tin, N.T., Hong Kong, China

Corresponding authors: George Ho Man Chan Tel: +852-2966-6681; Fax: +852-2601-6564; E-mail: [email protected] Terence See Ming Wan Tel: +852-2966-6297; Fax: +852-2601-6564; E-mail: [email protected]

ACS Paragon Plus Environment


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 2 of 30

ABSTRACT The use of anabolic androgenic steroids (AAS) is prohibited in both human and equine sports. Conventional approach in doping control testing for AAS (as well as other prohibited substances) is accomplished by the direct detection of target AAS or their characteristic metabolites in biological samples using hyphenated techniques such as gas-chromatography or liquid-chromatography coupled with mass-spectrometry. Such approach, however, falls short when dealing with unknown designer steroids where reference materials and their pharmacokinetics are not available. In addition, AASs with fast elimination times renders the direct detection approach ineffective as the detection window is short. Targeted metabolomics approach is a plausible alternative to the conventional direct detection approach for controlling the misuse of AAS in sports. Since the administration of AAS of the same class may trigger similar physiological responses or effects in the body, it may be possible to detect such administrations by monitoring changes in the endogenous steroidal expression profile. This study attempts to evaluate the viability of using the targeted metabolomic approach to detect the administration of steroidal aromatase inhibitors, namely androst-4-ene-3,6,17trione (6-OXO) and androsta-1,4,6-triene-3,17-dione (ATD), in horses. Total (free and conjugated) urinary concentrations of 31 endogenous steroids were determined by gas chromatography-tandem mass spectrometry (GC-MS/MS) for a group of 2 resting and 2 in-training thoroughbred geldings having been treated with either 6-OXO or ATD. Similar data were also obtained from a control (untreated) group of in-training thoroughbred geldings (n = 28).

Statistical processing and

chemometric procedures using principle component analysis (PCA) and orthogonal projection to latent structures-discriminant analysis (OPLS-DA) have highlighted 7 potential biomarkers that could be used to differentiate urine samples obtained from the control and the treated groups. Based on this targeted metabolomic approach, the administration of 6-OXO and ATD could be detected much longer as compared with the conventional direct detection approach.

2 ACS Paragon Plus Environment

Page 3 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


INTRODUCTION The efficacy of conventional drug testing, which aims to identify the presence of individual banned substances or their unique metabolites in biological fluids, is diminishing quickly over time. Despite advances in analytical sciences, doping with proteins, peptides and other emerging products of biotechnology is most difficult if not impossible to identify by conventional drug testing. These substances are becoming cheaper to obtain and growing rapidly in number. Their covert use, particularly the anabolic or growth enhancing agents with lasting effects during the long period of training, poses not only a significant threat to the integrity of human and animal sports, but also a heavy burden for laboratories to keep performing far more and far better drug testing. Consequently, additional resources are constantly needed. With the diminishing effectiveness of conventional drug testing, which is increasingly more expensive, in order to control current and emerging threats, many anti-doping efforts worldwide have been evaluating indirect methods that are based on the identification and monitoring of biomarkers for both human and equine athletes. In human doping control of AAS, Boccard et al. reported an untargeted steroidomic approach to analyse urine samples from a human clinical trial for the discovery of biomarkers to correlate with testosterone undecanoate oral intake.1 Van Renterghem et al. reported four urinary steroid ratios in addition to the well-known testosterone/epitestosterone ratio to support positive findings of testosterone abuse in human.2

These researchers had also identified novel biomarkers for

dihydrotestosterone and dehydroepiandrosterone administration in human athletes based on comprehensive steroid profiling.3 In equine, Kaabia and coworkers have successfully established statistical models and weighted equation to discriminate untreated male horses from those administered with nandrolone based on their urinary and plasma endogenous steroid profiles.4 The detection window for the administration with this endogenous steroid could be significantly improved to up to three months.

The study provided insight to the industry on adopting the

metabolomics approach to control the misuse of anabolic steroids in horse with a prolonged 3 ACS Paragon Plus Environment


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

detection and also demonstrated the potential to regulate the misuse of endogenous prohibited substances. The approach of using biomarkers profiling for doping control has already been adopted in human sports. Since December 2009, WADA has implemented the athlete biological passport (ABP) for anti-doping purpose. Initially, the ABP included only a haematological module to detect blood doping (e.g., the use of erythropoiesis-stimulating agents, blood transfusion or gene manipulation) based on the haematological variables in blood. More recently, a steroidomic module was introduced in January 2014 to detect exogenous administration of endogenous AAS and other anabolic agents. It is reckoned that administration of endogenous AAS can alter one or more of the seven targeted markers or ratio of the urinary steroid profile, namely testosterone, epitestosterone, androsterone, etiocholanolone, 5α-androstane-3α,17β-diol, 5β-androstane-3α,17β-diol and the ratio of testosterone to epitestosterone (T/E). Longitudinal monitoring of these markers may provide indication on the abuse of the endogenous AAS in human athletes. On the other hand, although doping control by monitoring biomarkers is currently not prevalent in the horseracing industry, it would undoubtedly be the future trend of development. Aromatase inhibitor is an estrogen blocker that can effectively block estradiol biosynthesis by inhibiting the pathway of aromatisation. Due to the sex-steroid negative feedback, secretion of endogenous hypothalamic gonadotrophin-releasing hormone (GnRH) and luteinizing hormone (LH) would be enhanced, which would consequently stimulate testosterone biosynthesis and secretion. Therefore, the abuse of aromatase inhibitors is regarded as a means of indirect androgen doping.5 Albeit the absence of gonads, trace levels of testosterone, reportedly produced by the adrenal cortex, are often found in gelding urine.6,

7

For this reason, the abuse of testosterone in geldings is

controlled by internationally agreed thresholds.8-10 It is envisaged that aromatase inhibitor should also have some effect on geldings, despite less pronounced than in entire male horses. In the horseracing industry, aromatase inhibitors including steroidal ones have recently been listed in Article 6E of the International Agreement on Breeding, Racing and Wagering published by the 4 ACS Paragon Plus Environment

Page 4 of 30

Page 5 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


International Federation of Horseracing Authorities as one of the classes of hormones and metabolic modulators that are banned at all times in the career of a racehorse. The authors’ laboratory had previously conducted in vitro and in vivo metabolic studies of two steroidal aromatase inhibitors, namely androst-4-ene-3,6,17-trione (6-OXO)11 and androsta-1,4,6-triene-3,17-dione (ATD).12 Some useful metabolites with longer detection times than their respective parent steroids were identified and proposed as suitable screening targets. As mentioned earlier, this direct detection approach is only effective in detecting the intended targets and cannot be applied to the detection of other aromatase inhibitors. This study attempts to evaluate if a targeted metabolomics approach could establish a statistical model to be used as a criterion for the indirect detection of the class of steroidal aromatase inhibitors in general.

Urinary concentrations of 31 endogenous steroids were quantified by gas

chromatography-tandem mass spectrometry (GC-MS/MS) in samples taken from thoroughbred geldings in-training (n=28) and thoroughbred geldings having been treated with 6-OXO (n = 2) and ATD (n = 2). Seven of the 31 endogenous steroids were identified as potential biomarkers based on the unsupervised Principal Component Analysis (PCA) and the supervised Orthogonal Projection to Latent Structures-Discriminant Analysis (OPLS-DA) models. A weighted equation employing the urinary concentrations of the 7 potential biomarkers was established to serve as a criterion to identify potential positive samples. Based on this targeted metabolomic approach, the administration of 6OXO could be detected about 2.1 times longer (up to 95 hours) than that based on the direct detection of the 6-OXO metabolite (3,17-dihydroxyandrostan-6-one) reported previously.11 Similarly, the detection time for ATD administration was about 2.5 times longer (up to 195 hours) compared to the reported direct detection of ATD metabolites (androsta-1,4,6-trien-17-ol-3-one and 4,6-androstadien-17β-ol-3-one).12



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

EXPERIMENTAL SECTION Sample preparation, extraction and derivatisation Urine pretreatment

Urine (about 5 mL) was centrifuged at ~1500 g for 10 minutes and the

supernatant (4 mL) was transferred to a tube containing ammonium sulfate (0.5 g). d3-Testosterone sulfate (equivalent to 240 ng of free d3-testosterone), d3-nandrolone (200 ng) and d3-androstanediol (200 ng) were added as internal standards, and the mixture was vortexed until the ammonium sulfate had dissolved. Following centrifugation (~1500 g for 10 minutes), 3 mL of the supernatant was loaded onto a Sep-Pak C18 cartridge, which had been pre-conditioned with methanol (5 mL) followed by deionised water (5 mL × 2). The loaded cartridge was rinsed with deionised water (6 mL × 2) and n-hexane (5 mL), and then eluted with methanol (3 mL). The eluate was evaporated to dryness at 60 °C under nitrogen and reconstituted in 2 mL of phosphate buffer (0.1 M, pH 6.0). Enzyme Hydrolysis

The pH of the reconstituted urine SPE extract was adjusted to 6.4, vortexed,

and incubated with β-glucuronidase from E. Coli (30 µL, 140 U/mL) for 1 hour at 55 °C. The enzyme-hydrolysed extract was cooled to ambient temperature and then loaded onto a Nexus cartridge, which was rinsed with deionised water (3 mL) and n-hexane (3 mL). The cartridge was then eluted with chloroform (2 mL) and 5 % methanol in ethyl acetate (v/v, 3 mL) successively. The combined eluate was evaporated to dryness at 60 °C under nitrogen. Methanolysis Anhydrous methanolic hydrogen chloride (1 M, 0.5 mL) was added to the dried enzyme-hydrolysed residue and the mixture was heated at 65 °C for 15 minutes. After cooling to ambient temperature, diisopropyl ether (3 mL) was added and the resulting mixture was transferred to a 15-mL graduated centrifuge tube containing 2 mL of aqueous base (1 M NaOH and 0.15 M NaCl) for neutralisation and liquid-liquid extraction. The mixture was centrifuged at ~1500 g for 0.5 minute. The organic layer was filtered through a cotton-wool-packed Pasteur pipette into a 5-mL Reacti-vial and then evaporated at 60 °C under nitrogen. 6 ACS Paragon Plus Environment

Page 6 of 30

Page 7 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


PFPA derivatisation Dried acetonitrile (100 µL) and pentafluoropropionyl anhydride (PFPA) (30 µL) were added to the residue. The mixture was incubated at 60 °C for 15 minutes, then evaporated to dryness at 60 °C under nitrogen. The residue was finally reconstituted with 30 µL of n-heptane and transferred to a conical insert in a Chrompack autosampler vial for GC-MS/MS analysis.

Instrumentation and GC-MS/MS conditions Solid-phase extraction was carried out using a RapidTrace SPE workstation (Zymark Corporation, Hopkinton, MA, USA). GC-MS/MS analyses were performed on a ThermoScientific TRACE 1300 Series gas chromatograph coupled with a Thermo Scientific TSQ 8000 Evo Triple Quadrupole mass spectrometer (Thermo Fisher Scientific, San Jose, CA, USA) equipped with a ThermoScientific TriPlus RSH liquid autosampler. Separation was performed on a DB-5MS (~ 30 m × 0.25 mm, 0.25 µm film thickness) column (J&W Scientific, Folsom, CA, USA). The oven temperature was set initially at 110 °C and held for 1 minute, increased to 150 °C at 60 °C/min and then increased to 320 °C at 15 °C/min, and finally held at 320 °C for 4 minutes. A constant helium flow at 1.2 mL/min was used for all analyses. Sample (1 µL) was injected at 260 °C in splitless mode. All GCMS/MS analyses were performed in the Electron Ionisation mode at 70 eV. The transfer line and ion source temperatures were set at 320 °C and 280 °C respectively. Mass spectral data were acquired in selected reaction monitoring (SRM) mode, with emission current set at 25 µA and scan time at 300 msec, while the Q1 resolution was set at 0.7 FWHM. The precursor and product ions monitored and the respective collision energies employed for the target steroids are specified in Table S1 in the Supporting Information. TraceFinder (version 3.2.368.22, Thermo Fisher Scientific) software was used for method setup and data processing.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Quantification of target compounds The treated samples (post-administration) and untreated control samples (pre-administration and prerace) were randomised and proportionally divided into 5 batches for analysis. For each analytical batch, calibrators were prepared in XAD-treated urine at 0, 5, 10, 20, 30 and 50 ng/mL in duplicate. The average peak area ratios of each target steroid to its corresponding internal standard (as stated in Table S1) were plotted against the concentrations in the calibrators and fitted to obtain a linear regression calibration curve. The concentrations of the target steroids in the test samples were interpolated from the calibration curves.

Samples with steroid concentrations exceeding the

calibration range were diluted accordingly with deionised water and reanalysed. Quality control samples (at 10 ng/mL for each target steroid) in duplicate were processed in parallel with the test samples.

In-house drug administration and control samples Control (untreated) urine samples were collected from twenty-four healthy castrated male (gelding) thoroughbreds (imported from different countries in both hemispheres and aged between 3 and 9 years old) over a period from December 2013 to April 2014. These horses were in-training horses kept at the Hong Kong Jockey Club stables with 24-hours surveillance by closed-circuit television with access control to authorized person only. All medication and treatment were provided by inhouse veterinarians and recorded. Sampling was conducted under pre-race conditions on the morning of a raceday. The protocols for the administration of androst-4-ene-3,6,17-trione (6-OXO) and androsta-1,4,6-triene-3,17-dione (ATD) to geldings have already been published by the authors’ laboratory.11, 12 Details were also described in the Supporting Information.


Page 8 of 30

Page 9 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Method validation Inter-day accuracy and precision were evaluated using spiked urine samples at 3 concentrations (5, 20 and 40 ng/mL).

The inter-day accuracy and precision of the method were determined by

analysing the spiked urine samples in five replicates on three different days. Extraction recoveries were determined by preparing target steroids spiked at 20 ng/mL in blank urine samples (n=4) before and after sample extraction. Internal standards were each added after extraction to all samples for the extraction recovery study. The peak area ratios of each recovered target to the internal standard obtained from the samples spiked with steroids before extraction were compared with those obtained from the corresponding blank extracts spiked with the same amount of steroids after extraction (assuming 100 % recovery). The limits of detection (LoD) and limits of quantification (LoQ) were estimated by replicate analyses of control urine samples (n = 6) spiked with target steroids at 5 ng/mL.

Statistical analysis Statistical analysis on the unsupervised Principal Component Analysis (PCA) and supervised Orthogonal Projection to Latent Structures-Discriminant analysis (OPLS-DA) models were performed using SIMCA (version 14, Umetrics, Sweden). Cross-validation (CV) was performed to test the predictive ability of the model with the default settings in the software. One-seventh of the samples were left out from the mathematical model in each round of the CV test. All data were log transformed and Pareto scaled to minimise heteroscedasticity and to adjust for fold differences between metabolites.13 The model validity was verified using permutation tests and CV-ANOVA by SIMCA, and also by comparing the goodnesses of fit (R2 and Q2).14, 15 The rank of each metabolite was calculated from the Variable Influence on Projection (VIP) score in the OPLS-DA. In addition,



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

a receiver operator characteristic (ROC) analysis was conducted to evaluate the ability of the model to classify individual samples into the treated and control group.

RESULTS AND DISCUSSION Method Validation Table S2 in the Supporting Information summarises the results of method validations (abbreviations for the analytes are shown in Table S1). Apart from 5α-DHT at 40 ng/mL and 20 ng/mL, all target steroids spiked at the three levels exhibited accuracy within ± 20 %. As for the precision, all target steroids, except 6-T and 6-ADD at 40 ng/mL, were less than the benchmark for precision (22 %) proposed by Thompson based on the “corrected” Horwitz equation, σR = 0.22C for C < 120 ppb, where σR = standard deviation and C = concentration of analyte.16 The extraction recoveries for the majority of the target steroids were over 40 %. The relatively low recoveries for the estrogens (E1, E2 and E2α) were likely due to loss during the base-wash step with NaOH/NaCl, as these phenolic estrogens are mildly acidic. Albeit the relatively low recoveries, method accuracies (within ± 20 %) for most of the targets suggested that the method was fit for its intended purpose in quantification. The recovery issue was hence not reckoned having significant impact to the developed method. Linear calibration curves for the range of 0 to 50 ng/mL were established for each target steroid, with correlation coefficient (R) greater than 0.99 in all cases. The limits of detection (LoD) and the limits of quantification (LoQ) were taken to be respectively 3 and 10 times the standard deviation at 5 ng/mL. The LoD of target steroids were found to range from 0.3 ng/mL to 1.9 ng/mL, whereas the LoQ were from 0.9 ng/mL to 6.2 ng/mL. The respective LoD and LoQ of each target steroid were listed in Table S3 in the Supporting Information.


Page 10 of 30

Page 11 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Steroid Profiles in Urine Samples from Treated and Control Horses The steroid profiles (consisted of 31 steroids) for a total of 64 urine samples, including 28 from the control group (24 pre-race plus 4 pre-administration urine samples) and 36 treated samples (postadministration urine samples from 4 treated geldings), were evaluated in this study, resulting in a data matrix composed of 1984 observations. Two-third of the urine profiles were randomly allocated to the training set for prediction of class membership (treated or control) and statistical modelling, while the remaining one-third were allocated to the validation set to test for the applicability of the established statistical model. In this study, target concentrations below LoQ were regarded as missing values in the profile, and would be substituted with a value equivalent to LoQ/2 for statistical modelling.17 Figure 1 shows the unsupervised principal component analysis (PCA) score plot of the extracted data as correlated to 21 of the target steroids. The other 10 target steroids, although profiled, showed zero variance over all 64 urine samples and were therefore excluded. A plausible metabolic pathway of the 21 target steroids is shown in Figure 2. The corresponding PCA loading plot can be found in the Supporting Information (Figure S1).

The PCA model contained four components with the

performance of R2X = 0.922 and Q2 = 0.674. Three components with eigenvalue larger than one accounted for nearly 90 % of the variations. In addition, the dynamic change of the steroid profile in correlation with the pre- and post-administration time was inferred by the trajectory in the score plot (grey arrow, Figure 1). It was observed that all pre-administration samples resided in the same cluster as the other control samples in the PCA score plot, indicating the similarity of their steroid profiles at basal situation (blue circle, Figure 1). Notwithstanding, the steroid profiles changed remarkably on the day of administration within 24 hours (red circle, Figure 1) and up to 71 hours post-administration (yellow circle, Figure 1), and these samples were accordingly deviated from the untreated group in the score plot.

The profile migrated to the intermediate phase, eventually

returning to the basal value after 4–9 days post-administration (purple circle, Figure 1), and 11 ACS Paragon Plus Environment


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

consequently converged to the untreated cluster in the score plot. The dynamic trajectory has projected the comprehensive pharmacokinetics of the 21 target steroids in response to the administration. With the preliminary evaluation that class membership can be distinguished by the principal components, a model of supervised orthogonal projection to latent structures-discriminant analysis (OPLS-DA) was then built to differentiate steroid profiles between the treated and control populations without considering the time profile. The OPLS-DA score plot (Figure 3A) with the performance of R2X = 0.909, R2Y = 0.75 and Q2 = 0.637 has successfully separated a majority of the treated urine samples from the control samples. Four samples were noticed to be fallen out of the Hotelling ellipse however not classified as outliers. These samples were known to be early postadministration samples with the target steroids significantly up-regulated in response to the administration of aromatase inhibitors and produced a highly different steroid profile as compared with the control samples. Upon fast elimination of these target steroids, the steroid profiles of the later post-administration samples had only moderate changes as compared with the controls. Due to the obvious difference in steroid profiles, the early post-administration samples were therefore fallen out of the Hotelling ellipse while those collected in later time points, however still expressed, were grouped inside the right-half of the Hotelling ellipse. On the other hand, some of the treated samples which fell under the cluster of control samples in the OPLS-DA score plot (as indicated by the red arrow in Figure 3A) were revealed to be either samples collected very close to the time of administration (e.g., 10 minutes post-administration) when the steroid profile may not have altered in response to the oral uptake of the steroidal aromatase inhibitor, or samples collected during the terminal phase of elimination when the steroidal profile may have already returned to the basal value. The goodness of fit and predictive ability of the model was validated by the 100Y-permutated model as shown in Figure S2 in the Supporting Information. Since the Q2 (blue) regression line had an intercept of less than 0.05, and all 100 permuted R2 values (green) on the left were lower than the 12 ACS Paragon Plus Environment

Page 12 of 30

Page 13 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


original point of the R2 value on the right, indicating that the original model was not over-fitted, the validity of the OPLS-DA model was confirmed.14, 18 In addition, the result of the CV-ANOVA variant suggested that the model is highly significant, with a p-value of 0.00011.15 The loading plot (Figure 3B) of the OPLS-DA model illustrated that most of the target steroids showed a tendency towards the dummy variable of the treated group, meaning that the separation were mainly resulted from the expression changes in concentration of these 21 target steroids in the treated samples. The receiver operating characteristics (ROC) analysis was also conducted to evaluate the discriminative power, sensitivity and specificity of the model (Figure 4). The area under curve (AUC) obtained by the analysis was 0.946, which means the probability that the classifier will rank a randomly chosen positive instance is higher than a randomly chosen negative instance, i.e., the model has a good predictive ability towards discriminating between treated and control samples.19, 20

Selection of Biomarkers for Screening of Aromatase Inhibitor Administration Statistically-significant biomarkers capable of differentiating between treated and control urine samples were selected from an analysis of the S-plot and Variable Importance Plot (VIP). Target steroids with VIP score larger than one were considered critical variables in differentiating between the two classes of samples.14 Figure 5A shows the VIP plot of the OPLS-DA model. Seven target steroids, namely 5α-androstane-3β,17α-diol (αβα), 4,6-androstadien-17β-ol-3-one (6-T), 5αandrostane-3β,17β-diol (αββ), androsterone (A), testosterone (T), 5-androstene-3β,17β-diol (AED) and 4,6-androstadien-3,17-dione (6-ADD), were evaluated as significant mutual biomarkers correlated with the administration of steroidal aromatase inhibitors. These seven targets were also outstanding in the S-plot, with high correlation (reliability, y-axis) and high covariance (contribution, x-axis) as compared to the other target steroids (Figure 5B).21 These identified biomarkers, as highlighted in red in Figure 2, are all precursors or metabolites of testosterone and androstenedione,



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

the two androgens that can be converted to estradiol and estrone by aromatase. Figure S3 in the Supporting Information shows the urinary expression profile of the seven identified biomarkers. The concentrations of these anabolic and androgenic biomarkers were observed to be significantly elevated by 2 to 4 order of magnitudes after treatment. This phenomenon correlated with the effect of an aromatase inhibitor in the mammalian body system.5 After administering a steroidal aromatase inhibitor, biosynthesis of estrogens is hindered, while the production of endogenous androgens is induced due to the sex-hormone feedback. As a result, treatment with an aromatase inhibitor is a type of indirect androgen doping. The expression profiles addressed the significance of these seven biomarkers in classifying control and treated samples, and also supported the findings of the S-plot and VIP analyses. While it is expected that the concentrations of estrogens (i.e. E1, E2 and E2α) should theoretically be reduced in response to the administration, down-regulation trends were not observed due to their low endogenous levels in geldings, which were all below the LoQ for the preadministration urine samples (data not shown). This implies that the quantification method used was not sensitive enough to reveal any down regulation of these estrogens and hence their insignificance in the S-plot (Figure 5B). Based on the above statistical results, the contribution of each endogenous steroid biomarker in the OPLS-DA model was determined and a weighted equation for screening purpose was established as follow: Y = 1.86 × log [αβα] + 0.21 × log [6-ADD] + 0.07 × log [6-T] + 0.06 × log [AED] – 0.05 × log [A] – 0.22 × log [T] – 0.56 × log [αββ] where Y is the screening criterion; the coefficient for each biomarker is extracted from SIMCA, which re-expresses the influence of the predictor variables (i.e., concentration of the target steroids) on the response variable (i.e., administration of steroidal aromatase inhibitor) of the OPLS-DA model; and the concentration of each biomarker is applied unless it is smaller than its corresponding


Page 14 of 30

Page 15 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


LoQ, in which case it is substituted with the value of LoQ/2. This equation was subsequently applied to all control samples in the training set (n=19) in order to establish a threshold to identify the possible administration of an aromatase inhibitor.

The criterion threshold (Ythreshold) was

established by taking the average value of Y for the control samples plus 3 standard deviations (mean+3SD).4 This Ythreshold was determined to be -0.27. In principle, any sample with Ysample > Ythreshold is considered suspicious for having had an exposure to a steroidal aromatase inhibitor. The validity and applicability of the Ythreshold in the screening of aromatase inhibitor administration was further investigated by applying the weighted equation to the treated and control samples in the validation set. Figure 6 illustrates the prediction ability of the screening criterion Y. All pre-race urine samples (Areas A) in the training and validation sets were having Ysample < Ythreshold and were therefore correctly identified as negative samples (true negative).

Pre-administration samples (Areas B)

collected from horses before an oral administration of 6-OXO or ATD were expected to have a small value of Y since the physiological conditions of the horses should in general be similar to those providing the pre-race samples. However, one of the four pre-administration samples was different from the rest of the control samples with Y >> Ythreshold. Based on the model, this sample was suspected to have been collected from a potentially doped horse and should be investigated further. However, this could also be a false positive. Areas C show the prediction ability of our model towards 6-OXO or ATD administration. Two samples in the training set and two in the validation set were missed by the screening (i.e., false negatives). However, as noted earlier, most of these samples had been collected either at a very short time after administration or at a much later time point when the kinetic profiles of the steroids might have returned to basal values. Based on the ROC curve in Figure 4, it was revealed that the best cut-off that maximised sensitivity and specificity of the model was at 0.89 (sensitivity, true positive rate TPR) and at 0.96 (specificity, 1 - false negative rate FPR), as determined by the smallest value of distance (d) from the top-left 15 ACS Paragon Plus Environment


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Page 16 of 30

corner (0, 1) where ݀ = ඥ(‫)ܴܲܨ‬ଶ + (1 − ܴܶܲ)ଶ . The optimal cut-off indicated that when Y > Ythreshold, there was 89 % that the model would correctly identify a positive sample as positive (i.e. true positive), and on the contrary when Y < Ythreshold, there was 96 % that the model would correctly identify a negative sample as negative (i.e. true negative). From our experimental results in this study on the prediction capability of the established screening criterion towards the detection of aromatase inhibitor administration as shown in Figure 6, the true positive rate (TPR) of our model, equivalent to true positive (TP) / [true positive (TP) + false negative (FN)], was 32/36 or about 89 %; while the false positive rate (FPR), equivalent to false positive (FP) / [false positive (FP) + true negative (TN)], was 1/28 or about 4 %. These results demonstrated that our seven-biomarkers screening criterion model was highly correlated with the optimal value, and supported that the model has achieved a maximum sensitivity and specificity in discriminating between control samples and aromatase inhibitor treated samples. The detection of 6-OXO administration has been reported previously to be best performed by monitoring its metabolite 3,17-dihydroxyandrostan-6-one, with a detection time of up to 46 hours.11 For ATD administration, a detection time of up to 77 hours was observed by monitoring its metabolites androsta-1,4,6-trien-17-ol-3-one and 6-T.12 Using the proposed steroidomics model, the detection time could be extended about 2.1 times (up to 95 hours) for 6-OXO administration and about 2.5 times (up to 195 hours) for ATD administration. In order to test the robustness of the developed model, attempt was made to incorporate some postrace urine samples for evaluation. The result was however not very useful. Apparently, urine samples collected immediately after strenuous exercise (such as racing) could have a significantlyaltered steroid profile. The post-race urine samples could not converge with either the control or treated samples but differentiated themselves statistically into a third class.


Nevertheless, the

Page 17 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


developed model should still be applicable to in-training (out-of-competition) or pre-race urine samples. Theoretically, the sensitivity of a biomarkers profiling approach should be higher than traditional methods to detect a single target analyte. Since a profile assay is composed of multiple biomarkers, as long as one of the markers alters in expression, the presentation of the entire profile becomes different compared with the control samples. The sensitivity of the metabolomics profiling approach is therefore magnified in proportion to the number of diagnostic biomarkers used. In this study, we have demonstrated that metabolomics analysis by biomarker profiling is a viable approach to detect the administration of 6-OXO and ATD. It is however still too early to say if the established OPLSDA model and the resulting weighted equation could serve as a generic screening tool for the administration of other aromatase inhibitors. The fact that post-race urine samples did not fit into the model would suggest that the problem is far more complicated as there are many more variables in play (such as diet, exercise, health, environmental conditions, etc.) that could affect the steroid profile of a horse. The authors’ laboratory is planning to follow up on this model by conducting more administration trials with other aromatase inhibitors, as well as testing the developed model for the detection of other AAS.

Despite the complexity, the approach of using steroidomics or

metabolomics as an indirect tool for detecting exposure to a class of AAS or prohibited substances with similar pharmacological effects is appealing. Assuming a robust model can be developed for a certain class of prohibited substances, it will be an excellent method to detect new or designer drugs with the same pharmacological effects. Having said this, it is envisaged that the proposed indirect metabolomic approach in proving exposure to prohibited substances will face many technical and regulatory challenges ahead. For examples, the potential technical complexity in applying metabolomic approach in screening and referee analysis (i.e. A and B sample analyses) involving different laboratories and the potential challenges to the validity (or confidence level) of the test results for the purpose of sanction would be just some of the few major challenges among many. It 17 ACS Paragon Plus Environment


1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

will require great effort from the horseracing community and authorities to come up with relevant guidelines and rules to address these issues. Due to the potential variability among racehorses, it would seem appropriate for racing authorities to consider adopting WADA’s ABP model of longitudinal profiling as an alternative to doping control testing as this would eliminate the argument of inter-horse variations and would be legally more defensible.

CONCLUSION This paper describes a metabolomics approach to establish a model to screen for potential misuse of steroidal aromatase inhibitors in geldings under in-training (out-of-competition) or pre-race conditions. Free and conjugated endogenous steroids in urine were quantified by a validated GCMS/MS method for profiling purpose. Seven out of the thirty-one target steroids were highlighted by statistical modelling as significant mutual biomarkers to discriminate urine samples collected from horses administered with steroidal aromatase inhibitors from the control group. Based on the OPLS-DA model and the resulting weighted equation, the administration of 6-OXO and ATD could be detected for up to 4 and 9 days respectively. The detection times were about 2.1 times longer for 6-OXO and 2.5 times longer for ATD as compared with the direct detection of unique metabolites reported by the authors’ laboratory. The results demonstrated that the metabolomic approach of profiling diagnostic biomarkers is a viable tool to detect the misuse of different drugs of the same class in horse sports. In addition, to the best of our knowledge, this is the first report on establishing a time-dependent endogenous steroidal profile correlated with the administration of a class of drugs rather than an individual drug or prohibited substance. With the temporal trajectory, longitudinal monitoring of biomarkers in response to the administration could also facilitate the in-depth investigation of abnormality, thus revealing the excretion kinetics. As a proof-of-concept, this study also demonstrated the feasibility to detect illegal administration of different drugs by a single metabolomics model with good predictability towards compliance status. Ideally, each established 18 ACS Paragon Plus Environment

Page 18 of 30

Page 19 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


model could be applied generically to detect drugs or substances having similar actions or effects. However, as the steroid profile of a horse can be affected by many variables (such as diet, exercise, heath, environmental conditions, etc.), more work will need to be done to validate if the proposed weighted equation can be used as a generic tool to screen for the administration with the class of steroidal aromatase inhibitors in out-of-competition or pre-race urine samples.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

FIGURES Figure 1 Three dimensional PCA score plot with respect to the treatment classification and postadministration time profile. Control (untreated) urine samples, which had been collected in pre-race and pre-administration environment, were well separated from the steroidal aromatase inhibitor postadministration samples.


Page 20 of 30

Page 21 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 2 Plausible metabolic pathway for the target steroids in related to the administration of steroidal aromatase inhibitor. Seven target steroids that were evaluated to be potential biomarkers for discriminating between control and treated samples were highlighted in red.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Figure 3

OPLS-DA (A) score plot and (B) loading plot constructed with the basis of the

contribution from the 21 target steroids. The two plots illustrated that the separation of the control and treated groups were mainly based on the elevation in expression of some target steroids in the treated samples. Some of the treated samples fell under the cluster of control samples (as indicated by the red arrow).


Page 22 of 30

Page 23 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 4 Receiver operator characteristic (ROC) curve representing the true positive rate TPR (sensitivity) against the false positive rate FPR (1 - specificity) according to the 21 target steroids that established the statistical model. The optimal cut off point (FPR, TPR) was determined to be (0.04, 0.89) as indicated by the red arrow, representing 89 % of the model can correctly identify a positive sample as positive while 96 % of the model can correctly identify a negative sample as negative.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

Figure 5 (A) VIP score plot and (B) S-plot in response to the OPLS-DA model constructed. Target steroids with VIP score larger than 1 (pointed by red arrows) were considered to be potential biomarkers in discriminating control and treated samples. The finding were further supported by the S-plot projection, were 7 of the target steroids with VIP score > 1 exhibiting up-regulated expression change in treated urine samples (bracketed with red rectangle).


Page 24 of 30

Page 25 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


Figure 6 Prediction ability of the screening criterion Y. Samples with Y larger than threshold Ythreshold were considered suspicious for a prior administration of steroidal aromatase inhibitor. The samples had been collected under pre-race (Areas A), pre-administration (Areas B), and postadministration (Areas C) environment respectively. The number on top of the symbol indicated the post-administration day, e.g. 1 refers to day 1 (within 24 hours), 9 refers to day 9, etc. For the establishment of the model, two-third of the samples were allocated to the training set for building the statistical model, while the remaining were allocated to the validation set for ratifying the model’s validity.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

REFERENCES (1)

Boccard, J.; Badoud, F.; Grata, E.; Ouertani, S.; Hanafi, M.; Mazerolles, G.; Lanteri, P.; Veuthey, J. L.; Saugy, M.; Rudaz, S. Forensic Sci Int 2011, 213, 85-94.

(2)

Van Renterghem, P.; Van Eenoo, P.; Sottas, P. E.; Saugy, M.; Delbeke, F. Clin Endocrinol

(Oxf) 2011, 75, 134-140. (3)

Van Renterghem, P.; Van Eenoo, P.; Sottas, P. E.; Saugy, M.; Delbeke, F. Drug Test Anal 2010, 2, 582-588.

(4)

Kaabia, Z.; Dervilly-Pinel, G.; Popot, M. A.; Bailly-Chouriberry, L.; Plou, P.; Bonnaire, Y.; Le Bizec, B. Drug Test Anal 2014, 6, 376-388.

(5)

Handelsman, D. J. Br J Pharmacol 2008, 154, 598-605.

(6)

Silberzahn, P.; Rashed, F.; Zwain, I.; Leymarie, P. Steroids 1984, 43, 147-152.

(7)

Cox, J. E.; Williams, J. H.; Rowe, P. H.; Smith, J. A. Equine Vet J 1973, 5, 85-90.

(8)

Ho, E. N. M.; Kwok, W. H.; Leung, D. K. K.; Riggs, C. M.; Sidlow, G.; Stewart, B. D.; Wong, A. S. Y.; Wan, T. S. M. Drug Test Anal 2015, 7, 414-419.

(9)

Tang, P. W.; Ip, W. C.; Law, W. C.; Wan, T. S. M.; Crone, D. L. Proceedings of the 12th

International Conference of Racing Analysts and Veterinarians, Vancouver, Canada 1998; 107-112. (10)

Tang, P. W.; Law, W. C.; Crone, D. L. Proceedings of the 11th International Conference of

Racing Analysts and Veterinarians, Queensland, Australia 1996; 68-72. (11)

Leung, G. N. W.; Tang, F. P. W.; Wan, T. S. M.; Wong, C. H. F.; Lam, K. K.; Stewart, B. D.

Biomed Chromatogr 2010, 24, 744-751. (12)

Kwok, W. H.; Leung, G. N. W.; Wan, T. S. M.; Curl, P.; Schiff, P. J. J Steroid Biochem Mol

Biol 2015, 152, 142-154. (13)

van den Berg, R. A.; Hoefsloot, H. C.; Westerhuis, J. A.; Smilde, A. K.; van der Werf, M. J.

BMC Genomics 2006, 7, 142. (14)

Zhang, T.; Wu, X.; Ke, C.; Yin, M.; Li, Z.; Fan, L.; Zhang, W.; Zhang, H.; Zhao, F.; Zhou, X.; Lou, G.; Li, K. J Proteome Res 2013, 12, 505-512.

(15)

Eriksson, L.; Trygg, J.; Wold, S. Journal of Chemometrics 2008, 22, 594-600.

(16)

Horwitz, W.; Albert, R. J AOAC Int 2006, 89, 1095-1109.

(17)

Fiehn, O.; Kind, T. In Metabolomics; Weckwerth, W., Ed.; Humana Press, 2007; Vol. 358, pp 3-17.

(18)

Wiklund, S.; Nilsson, D.; Eriksson, L.; Sjöström, M.; Wold, S.; Faber, K. Journal of

Chemometrics 2007, 21, 427-439. 26 ACS Paragon Plus Environment

Page 26 of 30

Page 27 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


(19)

Linden, A. J Eval Clin Pract 2006, 12, 132-139.

(20)

Xia, J.; Broadhurst, D. I.; Wilson, M.; Wishart, D. S. Metabolomics 2013, 9, 280-299.

(21)

Wiklund, S.; Johansson, E.; Sjostrom, L.; Mellerowicz, E. J.; Edlund, U.; Shockcor, J. P.; Gottfries, J.; Moritz, T.; Trygg, J. Anal Chem 2008, 80, 115-122.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

ACKNOWLEDGEMENT The authors would like to thank Ms. Rebecca Cheng and Mr. Raven Kan for their technical assistance.


Page 28 of 30

Page 29 of 30

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60


SUPPORTING INFORMATION Materials; In-house drug administration protocols; Deconjugation study; GC-MS/MS experimental parameters of target steroids and their corresponding internal standards; Method Validation data; The PCA loading plot of the treated urine samples against control urine samples; OPLS-DA permutation plot; and the urinary elimination profile of the seven potential biomarkers in four treated horses. This material is available free of charge via the Internet at http://pubs.acs.org.



1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

For TOC only


Page 30 of 30

Targeted Metabolomics Approach To Detect the ... - ACS Publications

Recommend Documents