Application of multivariate statistics to the Steroidal Module of the Athlete Biological Passport: A proof of concept study

Metabolomic signatures in elite cyclists: differential characterization of a seeming normal endocrine status regarding three serum hormones

INTRODUCTION

Serum phenotyping of elite cyclists regarding cortisol, IGF1 and testosterone is a way to detect endocrine disruptions possibly explained by exercise overload, non-balanced diet or by doping. This latter disruption-driven approach is supported by fundamental physiology although without any evidence of any metabolic markers.

OBJECTIVES

Serum samples were distributed through Low, High or Normal endocrine classes according to hormone concentration. A ¹H NMR metabolomic study of 655 serum obtained in the context of the longitudinal medical follow-up of 253 subjects was performed to discriminate the three classes for every endocrine phenotype.

METHODS

An original processing algorithm was built which combined a partial-least squares-based orthogonal correction of metabolomic signals and a shrinkage discriminant analysis (SDA) to get satisfying classifications. An extended validation procedure was used to plan in larger size cohorts a minimal size to get a global prediction rate (GPR), i.e. the product of the three class prediction rates, higher than 99.9%.

RESULTS

Considering the 200 most SDA-informative variables, a sigmoidal fitting of the GPR gave estimates of a minimal sample size to 929, 2346 and 1408 for cortisol, IGF1 and testosterone, respectively. Analysis of outliers from cortisol and testosterone Normal classes outside the 97.5%-confidence limit of score prediction revealed possibly (i) an inadequate protein intake for outliers or (ii) an intake of dietary ergogenics, glycine or glutamine, which might explain the significant presence of heterogeneous metabolic profiles in a supposedly normal cyclists subgroup.

CONCLUSION

In a next validation metabolomics study of a so-sized cohort, anthropological, clinical and dietary metadata should be recorded in priority at the blood collection time to confirm these functional hypotheses.

Optimization and validation of a GC-MS quantitative method for the determination of an extended estrogenic profile in human urine: Variability intervals in a population of healthy women

An analytical method based on GC-MS was developed for the determination of a wide panel of urinary estrogens, together with their principal metabolites. Because of the low concentration of estrogens in urine, an efficient sample pre-treatment was optimized by a design of experiment (DoE) procedure to achieve satisfactory sensitivity. A second DoE was built for the optimization of the chromatographic run, with the purpose of reaching the most efficient separation of analytes with potentially interfering ions and similar chromatographic properties. The method was fully validated using a rigorous calibration strategy: from several replicate analyses of blank urine samples spiked with the analytes, calibration models were built with particular attention to the study of heteroscedasticity and quadraticity. Other validation parameters, including the limit of detection, intra-assay precision and accuracy, repeatability, selectivity, specificity, and carry-over, were obtained using the same set of data. Further experiments were performed to evaluate matrix effect and extraction recovery. Then the urinary estrogen profiles of 138 post-menopausal healthy women were determined. These profiles provide a representation of physiological concentration ranges, which, in forthcoming studies, will be matched on the base of multivariate statistics with the urinary estrogenic profile of women with breast or ovarian cancer.

Individual and cyclic estrogenic profile in women: Structure and variability of the data

The concentration of estrogens in the body fluids of women is highly variable, due to the menstrual cycle, circadian oscillations, and other physiological and pathological causes. To date, only the cyclic fluctuations of the principal estrogens (estradiol and estrone) have been studied, with limited outcome of general significance. Aim of the present study was to examine in detail the cyclic variability of a wide estrogens' panel and to interpret it by multivariate statistics. Four estrogens (17α-estradiol, 17β-estradiol, estrone, estriol) and eleven of their metabolites (4-methoxyestrone, 2-methoxyestrone, 16α-hydroxyestrone, 4-hydroxyestrone, 2-hydroxyestrone, 4-methoxyestradiol, 2-methoxyestradiol, 4-hydroxyestradiol, 2-hydroxyestradiol, estriol, 16-epiestriol, and 17-epiestriol) were determined in urine by a gas chromatography - mass spectrometry method, which was developed by design of experiments and fully validated according to ISO 17025 requirements. Then, urine samples collected every morning for a complete menstrual cycle from 9 female volunteers aged 24-35 years (1 parous) were analysed. The resulting three-dimensional data (subjects × days × estrogens) were interpreted using several statistical tools. Parallel Factor Analysis compared the estrogen profiles in order to explore the cyclic and inter-individual variability of each analyte. Principal Component Analysis (PCA) provided clear separation of the sampling days along the cycle, allowing discrimination among the luteal, ovulation, and follicular phases. The scores obtained from PCA were used to build a Linear Discriminant Analysis classification model which enhanced the recognition of the three cycle's phases, yielding an overall classification non-error rate equal to 90%. These statistical models may find prospective application in fertility studies and the investigation of endocrinology disorders and other hormone-dependent diseases.

Multivariate interpretation of the urinary steroid profile and training-induced modifications. The case study of a Marathon runner

The steroidal module of the athlete biological passport (ABP) introduced by the World Anti-Doping Agency (WADA) in 2014 includes six endogenous androgenic steroids and five of their concentration ratios, monitored in urine samples collected repeatedly from the same athlete, whose values are interpreted by a Bayesian model on the basis of intra-individual variability. The same steroid profile, plus dihydrotestosterone (DHT) and DHEA, was determined in 198 urine samples collected from an amateur marathon runner monitored over three months preceding an international competition. Two to three samples were collected each day and subsequently analyzed by a fully validated gas chromatography-mass spectrometry protocol. The objective of the study was to identify the potential effects of physical activity at different intensity levels on the physiological steroid profile of the athlete. The results were interpreted using principal component analysis and Hotelling's T² vs Q residuals plots, and were compared with a profile model based on the samples collected after rest. The urine samples collected after activity of moderate or high intensity, in terms of cardiac frequency and/or distance run, proved to modify the basal steroid profile, with particular enhancement of testosterone, epitestosterone, and 5α-androstane-3α,17β-diol. In contrast, all steroid concentration ratios were apparently not modified by intense exercise. The alteration of steroid profiles seemingly lasted for few hours, as most of the samples collected 6 or more hours after training showed profiles compatible with the "after rest" model. These observations issue a warning about the ABP results obtained immediately post-competition.

Correlation between chronological and physiological age of males from their multivariate urinary endogenous steroid profile and prostatic carcinoma-induced deviation

The biosynthesis of endogenous androgenic anabolic steroids (EAAS) in males varies with age. Knowledge of the general urinary EAAS profile's dependence from aging - not reported up to now - may represents a prerequisite for its exploitation in the screening and diagnostic support for several pathologies. Extended urinary EAAS profiles were obtained from healthy and pathological individuals, using a GC-MS method which was fully validated by a stepwise, analyst-independent scheme. Seventeen EAAS and five of their concentration ratios were determined and investigated using multivariate statistical methods. A regression model based on Kernel partial least squares algorithm was built to correlate the chronological age of healthy male individuals with their "physiological age" as determined from their urinary EAAS profile. Strong correlation (R² = 0.75; slope = 0.747) and good prediction ability of the real chronological age was inferred from EAAS data. In contrast, patients with recent diagnosis (not pharmacologically treated) of prostatic carcinoma (PCa) exhibited a comprehensive EAAS profile with strong negative deviation from the model, corresponding a younger predicted age. This result is possibly related to the activation of anomalous steroid biosynthesis induced from PCa. Over a restricted 60-80 years-old population, PLS-discriminant analysis (DA) was used to distinguish healthy subjects from patients with untreated PCa. PLS-DA yielded excellent discrimination (sensitivity and specificity >90%) between healthy and pathological individuals. This proof-of-concept study provides a preliminary evaluation of multivariate DA on wide EAAS profiles as a screening method to distinguish PCa from non-pathological conditions, overcoming the potentially interfering effect of ageing.