FR2819591A1 - DISCRIMINATION PROCESS WITH LOCATION AND / OR IDENTIFICATION OF SITUATIONS OF BIOLOGICAL DISTURBANCES BY SPECTROMETRY AND RECOGNITION OF SHAPE - Google Patents

Abstract

The invention concerns a discriminating method with location and identification of a situation of biological perturbation. Said method is characterised in that it consists in: subjecting a biological sample to a high-resolution analytical technique and in associating said technique with a validated statistical or classifying processing which takes into account all the vibrations observed, the set of vibrations referencing a specific biological perturbation situation. The invention is in particular applicable to characterisation of the physiological condition particular to a population, location and/or identification of a perturbation of a biological system by a substance with pharmacological activity or by a micro-organism, indirect discrimination of the use of anabolic agents in breeding and/or in sports and/or in the biomedical field and to constitute a repository or a databank.

Description

<Desc / Clms Page number 1>

L'invention a pour objet un procédé de discrimination avec repérage et/ou identification de situations de perturbations biologiques, qu'elles soient d'origine physiologique, génétique ou pathologique, auxquelles sont soumis des organismes vivants. Discrimination method with identification and / or identification of biological disturbance situations by spectrometry and shape recognition

The subject of the invention is a discrimination method with identification and / or identification of situations of biological disturbances, whether of physiological, genetic or pathological origin, to which living organisms are subjected.

 Physiological disturbance is defined as the state in which an organism is found under specific environmental conditions, it being understood that a disturbed physiological situation may also have genetic, environmental or pathological causes, as for example in the case in which type 1 or 2 diabetes.

 An example of physiological disturbance is illustrated by the application of anabolic treatments in cattle farming. Thus, the anabolic agents are administered in such a way as to cause the animal to increase muscle mass and, consequently, to decrease the proportion of fat. This use leads to a precise physiological situation, differentiable by morphological and anatomical criteria of the situation where anabolic agents are not used, criteria which are underpinned by metabolic differences.

 Anabolic expression, as used in the description and claims, unless otherwise indicated, includes a molecule or combination of molecules having the effect at the level of the body of substantially altering the overall metabolism, and thus the constitution of the body. body in terms of muscle tissue / adipose tissue ratio, as well as the tissue composition of lipids, carbohydrates and proteins. The modification of the general metabolism induced by the anabolic treatment is translated, at the level of the large-scale effects, in particular by the relative or absolute increase of the muscular mass, the melting of the fats, the modification of the behavior.

<Desc / Clms Page number 2>

 There are different families of anabolic steroids among which: steroids, ss-agonists, thyrostatic substances and peptides mimicking the action of growth hormone or stimulating its production.

 In the field of food safety, a great deal of research is being conducted to develop diagnostic methods (screening, recognition, authentication) for the use of anabolic steroids in animal husbandry, enabling veterinary control services to ensure quality. carcasses and animal products. Techniques of isolation and direct analysis have been implemented, such as immunological techniques, which are radioimmunological [Scippo et al., 1993; Scippo & Maghuin-Rogister, 1995] or immuno-enzymatic [Degand et al., 1989]. They allow a precise and very sensitive analysis of the incriminated substances. However, these immunological methods require having a priori knowledge of the substances to be controlled, in order to produce the antibodies specific for the desired anabolic agents. However, because of link affinities similar to those existing for the target molecule, they make it possible to diagnose a related molecule, but not necessarily a new family of anabolic agents. These techniques support a presumption of unlawful treatment.

[Daeseleire et al., 1992] ou le plasma [Van Ginkel et al., 1993]. Elle permet dans certains cas une analyse globale (multirésidus) des substances incriminées, avec de très bonnes sensibilités. Another widely used direct technique is the analysis by gas chromatography or liquid coupled to mass spectrometry. In contrast to immunochemical methods, this analytical method makes it possible to prove the presence of a prohibited molecule. It is the only method, to date, for identifying and confirming the presence of trace anabolic steroids in complex biological matrices, such as muscle [Hartwig et al., 1997], urine

[Daeseleire et al., 1992] or plasma [Van Ginkel et al., 1993]. It allows in some cases a global analysis (multiresidus) of the incriminated substances, with very good sensitivities.

<Desc / Clms Page number 3>

 Regarding the use of natural steroids (hormones naturally present in the body), a control method implements the calculation of the ratio of urinary concentrations between the desired molecule and one of its metabolites, such as testosterone ratio androstenedione [Gatti et al., 1993]. However, these analyzes can only be done with prior knowledge of said substance, which allows to set up a protocol adapted to the desired substance (s). Many have an enzymatic digestion stage that is controversial [Massé et al., 1989].

In the case of natural anabolic steroids, a variant also consists in measuring by mass spectrometry (GC-IRMS) the 13C / 12C isotope ratio for a target molecule, and thus determining the endogenous and exogenous quantities [Ferchaud et al., 1998] .

 These methods make it possible to directly search for an anabolic agent or its metabolites generated by the detoxification process (metabolization) by the body, from biological fluids such as urine or blood plasma, or in the tissues (muscle, liver , adipose tissue, kidney, etc.) or the integuments (hair). Detection by mass spectrometry allows a structural determination of the desired compound, whether xenobiotic or endogenous.

 In the case of the use of a new family of molecules with anabolic effect, there is in contrast no physicochemical knowledge of the anabolic in question, much less its metabolic pathways. Consequently, the search for the proof of the illicit use of such a molecule by all the analytical techniques targeted on already known molecules will be doomed to failure.

 The search for indirect arguments, based on the effects caused by anabolic treatment on the general metabolism whose result in terms of morphological effects is the increase of the muscular mass and the decrease of the

<Desc / Clms Page number 4>

 amount of fat, can move towards a monitoring of the metabolic disturbances generated.

 Some studies have reported the use of nuclear magnetic resonance (NMR) and multidimensional statistics for the diagnosis of anabolic steroids of the 3-agonist type used in breeding [Vogels et al., 1996], for the screening of substances with a therapeutic effect. or toxicological [Beckwith-Hall et al., 1998] and cancer diagnosis [Nadal et al. 1997; Somorj ai et al. 1995, Tate et al. 1996). In these different studies, these are one-dimensional NMR techniques (1H, 31p or 13C) for which the structural lines are superimposed. This leads to a bias in the integration of the NMR signals which is the preliminary step to the identification by the tools of the multidimensional statistics of the disturbance of the physiological situation.

 The search for reliable means of discrimination with identification and / or identification of situations of biological disturbances led the inventors to take into account all the compounds present in a matrix or sample of biological origin, and to treat them according to a standardized protocol of spectroscopic and statistical analyzes.

 The invention thus proposes to identify the causal factor of the coordinated variations of the metabolism, by comparison of its signature with those recorded in a database.

 The studies carried out in this regard have shown that there are qualitative and quantitative constants specific to a physiological situation, characterized by a signature, and that it is possible to detect an associated variability.

 The term biological signature encompasses the notion of a characteristic and recognizable fingerprint of a metabolic situation described by means of a precise analytical technique.

 The method according to the invention for discrimination with identification and / or identification of a biological disturbance situation is characterized in that a biological sample is subjected to

<Desc / Clms Page number 5>

 highly resolutive analytical technique and that this technique is associated with a statistical treatment and / or classification taking into account all the variations observed with reference to a validated biological disturbance situation.

l'échantillon à étudier à au moins une mesure RMN multidimensionnelle, suivie d'une analyse quantitative des signaux, d'une intégration des signaux, et du traitement des données d'intégration. Les données obtenues sont alors caractéristiques des situations biologiques observées dans l'échantillon analysé et constituent une signature biologique. According to one embodiment of the invention,

the sample to be investigated to at least one multidimensional NMR measurement, followed by quantitative signal analysis, signal integration, and integration data processing. The data obtained are then characteristic of the biological situations observed in the sample analyzed and constitute a biological signature.

 NMR spectrometric analysis is, more particularly, two-dimensional NMR analysis. In this way we obtain correlations whose second-dimensional modulation gives rise to deconvolved variables. The 2D NMR used is homonuclear, for example H-H and / or heteronuclear, for example H-13C.

détection dite inverse H-C, de grande sensibilité puisque utilisant le proton'H, noyau d'hydrogène (99, 98% en abondance naturelle), comme noyau de détection au lieu du 13C (1,1% en abondance naturelle). Les corrélations entre lez et le 13C (taches de corrélation) sont obtenues par le biais de leur couplage mutuel. The protonic detection is obtained by means of a probe of

so-called reverse HC detection, of high sensitivity since using the proton'H, hydrogen nucleus (99, 98% in natural abundance), as detection nucleus instead of 13C (1.1% in natural abundance). Correlations between lez and 13C (correlation spots) are obtained through their mutual coupling.

1991]. Des impulsions de gradients de champs magnétiques sont envoyées au cours de l'expérience. Avec ces gradients, il est possible de sélectionner des cohérences H-C, ce qui permet une It is, in particular, an HMBC (Heteronuclear Multiple Bonding Connectivity) analysis which establishes heteronuclear correlations at long distances. It is possible to observe the couplings of carbon nuclei 13 located at two, three or even four chemical bonds of proton interval subject to detection. The pulse sequence used is, in particular, that of a HMBC-GAS (Gradient Accelerated Spectroscopy) [Hurd & John,

1991]. Pulses of magnetic field gradients are sent during the experiment. With these gradients, it is possible to select HC coherences, which allows a

<Desc / Clms Page number 6>

 very satisfactory suppression of the residual signals linked in particular to the resonance of the water and not containing information on the hydrocarbon skeleton situated in the median part of the proton dimension. This suppression is done in a reduced time of experience compared to the implementation of a HMBC without gradient.

 Moreover, the HMBC-GAS cards have the advantage of a great readability: they do not contain artefactual noise, in the form of vertical streaks located at the frequency of each signal and which are linked, in the case of a quadrature detection, an instrumental imperfection of the system of elimination of these signals in the experiments without gradient.

 The acquisition parameters, excitation frequency in the li dimension and in the 13C dimension, as well as the spectral windows in the two dimensions are determined specifically for each type of analysis. Each interferogram (FID) is multiplied by a square sinusoidal function. To increase the signal-to-noise ratio, an apodization (double zerofilling) is applied to each interferogram. In the case of HMBC-GAS, a Fourier transform of the signal in magnitude mode is then performed.

 The cards thus obtained are compared. The correlation spots are integrated and the data obtained are the subject of a multidimensional statistical analysis. The integration corresponds to the calculation of the integral of the NMR signal, represented in intensity per unit area, the smallest indivisible unit being the memory point. These memory points are characterized by a triplet: the two coordinates in the proton and carbon dimensions, as well as the intensity value of the signal for this point.

The integration of a given surface correlation spot then results from the sum of the intensity of all the memory points included in the surface in question, delimited around the correlation spot. From a statistical point of view, each task

<Desc / Clms Page number 7>

 Duly characterized and integrated correlation according to the same surface for all the maps becomes a precise variable, and each map established for an animal, a plant or any sample to be studied, is denominated individual.

 The exploitation of the data of the NMR cards obtained according to the invention makes it possible to have a reliable recognition technique of the physiological situations to which the samples at the origin of the variability have been submitted.

 According to another embodiment of the invention, mass spectrometry is used, additionally or alternatively, as a highly resolutive analytical technique.

 The variables obtained by the analytical technique used are subjected to a preliminary filtration step. The stages of discrimination of the groups of individuals and of classification of the so-called additional individuals are carried out on the reduced set of variables resulting from the filtration.

 The step of filtering the variables is advantageously performed by means of principal component analysis (PCA) and / or hierarchical classification and / or analysis of variance and / or partial correlation analysis, the analysis of variance and the analysis of the partial correlations that can be carried out on the whole set of variables or on the variables grouped in a cluster.

 The variables are filtered before the discriminant factor analysis (DFA) in order to eliminate the redundancy of information present in the initial system of variables, in particular, in very frequent cases where there are more variables than people.

 In the case of overinformation of the system, it follows a defect of invertibility (or pseudo-singularity) of the matrix to be analyzed, which makes it impossible to carry out the algorithm of AFD [Lebart, Morineau & Piron 1998; McLachlan, 1992].

The invention therefore seeks to reduce the number of redundant dimensions (therefore variables) without degenerating the information

<Desc / Clms Page number 8>

 total. Filtration by hierarchical classification of variables also makes it possible to generate groupings of variables that may be relevant from the point of view of the interpretation of these variables in terms of structure.

 The filtration of variables is done in a composite way. It is necessary to use and cross several complementary methods in order to increase the informative nature of the selected or generated variables.

 In each cluster or cluster, which corresponds to a set of variables with very similar behavior, we try to synthesize the information as much as possible. For this, a variant of the method consists of the selection of a single variable by means of an analysis of variance (ANOVA). In a second variant, an analysis of the partial correlations is carried out. A third variant of the method consists in creating a synthetic variable from the initial variables, which takes into account the maximum amount of information (total inertia) contained in the cluster.

This is done through a PCA.

 The PCA represents individuals (animals, plants, microorganisms) in a space of smaller size, the individuals being initially present in a space with n dimensions, n denoting the number of initial variables. This analysis seeks to isolate the directions of maximum heterogeneity (or principal components) of the cloud of individuals and the cloud of variables, without formulating a priori on the group structure of individuals or of variables.

 AFD or another method of discrimination (classification tree or neuromimetic networks) is used to discriminate predefined treatment groups [Ripley, 1996; Venables & Ripley, 1999]. There is, at this level, a priori on the group structure. It is this which defines a priori the biological signature in the statistical sense of the term.

 The optimal discrimination of individuals divided into defined subgroups (species, varieties, treatment groups,

<Desc / Clms Page number 9>

physiological groups), these distinctions not being exhaustive or limiting, by a classification of individuals or by prior knowledge of these groups, is advantageously carried out by AFD or by neuromimetic analyzes.

Instead of separating individuals according to the heterogeneity of the cloud, by maximizing the distance between individuals, AFD seeks to maximize the distance between the centroids of the groups. In a second step, the AFD is used to recognize the signature of additional individuals to be tested, these having not been used to build the discriminating functions or axes, to assign them to the group of which they are the closest. The invention, from this point of view, generates a decision rule. The biological signature becomes equivalent to the set of discriminant functions (or axes) defined by linear combination of the chosen source variables.

 The biological signature can then be used for tracking purposes, on individuals who have not been used to construct the discriminant functions or the classification model. Concurrently, since the discriminant functions and the groups of variables are defined, it is possible to perform a physiological interpretation of the metabolic pathways affected by the perturbation in relation to its context, in the case where the variables used in the calculation of the discriminant functions are already known structurally.

 Then, the identification of unknown individuals is done by the comparative analysis of the results obtained by the implementation of analytical techniques and statistical treatment on these individuals and data from a reference system corresponding to the species or treatment to which the analyzed sample could be affected.

 This repository is evolutionary and obtained by accumulation of analysis results by operating as defined above, the samples being treated identically, in

<Desc / Clms Page number 10>

 conditions identical to those applied to the samples to be analyzed.

 The samples used according to the invention are of animal, plant or microbiological origin. They come most commonly from excretion or secretion products, tissues, cells or culture media. They may be in the form of total or partial lyophilisates, extracts or crushes prepared from biological matrices.

 Usable biological fluids include urine, blood plasma, seminal fluid, saliva, cerebrospinal fluid, lymph for animals, sap, leaf exfoliation and floral exfoliation for plants, microorganism culture media.

 The reliability of the process implies that the samples and the analyzes are carried out under the same conditions.

 The discrimination method of the invention is of great interest in many applications as mentioned below.

 The invention aims, in particular, the application of the method defined above as a method of investigating the phenotype to allow access to the genetic component of an individual. It can also reveal the impact on the metabolism of an environmental disturbance. For example, the determination of pollution indices.

 The invention also relates to the application of said method to the characterization of the physiological state specific to a population: the invention allows in particular the signature verification of various physiological parameters (age, sex, fattening state, lactation) or nutritional (nutritional balance, origin and / or qualification of food raw materials, presence of antinutritional factors or beneficial to health). The invention also allows the biomedical research of signatures and / or the diagnosis and / or the follow-up of a pathology (in particular those whose latency period can be long:

<Desc / Clms Page number 11>

 cancer, encephalopathies, neurodegenerations, autoimmune diseases, diabetes).

 According to yet another aspect, the invention aims at the application of said discrimination method to the identification and / or identification of disturbances of a biological system by a substance of pharmacological activity or by a microorganism: the invention makes it possible to monitoring the effects of a substance between treatment groups, that is to say the recognition via their biological signature of the use of anabolic agents, but also of therapeutic substances, as well as the effects of different pollutants or xenobiotics, same as the influence of infection by a pathogenic microorganism on a population.

 In the application, for example to the identification of anabolic agents in a biological sample taken from an individual, the simultaneous and unbiased study of all the variations of the metabolism according to the process of the invention makes it possible to account in a very fine variations in metabolism generated by the use of anabolic steroids.

 Thus, it is possible to release a biological signature from the physiological state of anabolic treatment, without directly dosing anabolic agents or their residues. It is possible in particular to indirectly differentiate the animals according to the anabolic treatment they have followed, but also according to the dose administered compared to untreated animals. It is also possible within a given group of individuals to explore the variability of sex and age-related metabolism of the animals examined.

 Thus, the invention allows an indirect and a priori observation of a phenotype, through the use of new non-invasive techniques, or if possible non-destructive for the body to obtain the sample studied.

 The invention makes it possible to obtain information on the entire metabolism of the animal, and thus to generate a biological signature of the physiological situation in which it occurs.

<Desc / Clms Page number 12>

 found, by comparison with the rest of the study population or a reference population.

 The invention also aims to use such data to form a reference system, and a database, for an animal species or another organism, developed from the characteristics of the profiles of the samples studied. Such means make it possible to verify the nature of the physiology of an unknown animal or of another living organism by its biological signature.

 Other features and advantages are given in the following examples for purposes of illustration. In these examples, reference is made to FIGS. 1 to 6, which represent respectively: FIG. 1: the HMBC-GAS map for the individual No. 22 (cull cow), FIG. 2: the result obtained from the filtration of variables, - figure 3: the multidimensional AFD analysis, - figure 4: the HMBC-GAS map for urine collected at slaughter on the individual no. 4463, implanted 4 times, - figure 5: the result of the filtration of variables, - figure 6: multidimensional analysis AFD.

Example 1 Study of the physiological variability of a group of urine from cattle taken in slaughterhouses
These are urine from three categories of animals: cull cows, castrated males 3-4 years old, calves less than one year old.

This procedure comprises:
1-the preparation (lyophilization) of the urine,
2-the NMR analysis of the lyophilisates obtained,
3-the integration of the NMR analysis data,
4-the statistical analysis of the results.

 Each of the steps is detailed below.

<Desc / Clms Page number 13>

Lyophilization of Urines Each urinary sample is prepared as follows.

 The urine collected is stored at -30 C until analysis. A first analysis is designed to estimate the amount of urine needed to harvest 500 mg of dry matter.

About 10 ml of urine are placed in a 50 ml centrifugation tube, previously tared. The volume introduced is then weighed, before refreezing in the centrifuge tube. Freezing at a temperature of -80 ° C is preferable at a temperature of 30 ° C because it is faster (2 h) and provides a larger thermal flywheel when launching the next step.

The sample is then lyophilized until all the water in the sample has been sublimated. The lyophilizate is then weighed, which makes it possible to know the volume necessary to obtain 500 mg of lyophilizate. The freeze-drying operation is repeated on the same bottle. The lyophilizate thus harvested at the end of this second step is again weighed and then taken up in a volume of water to ensure a final concentration adjusted to 500 mg per ml. The sample is then frozen.

 Two lyophilizations are performed for each urinary sample. Each lyophilizate is independently measured by NMR.

NMR measurement
For each urine, two samples of constant concentration (500 mg / ml) are analyzed by 2D NMR. A series of experiments is recorded for each lyophilisate.

 An inverse detection probe is used for its high proton sensitivity. Are recorded: - a proton spectrum without elimination of the peak of water, - a proton spectrum with elimination of the peak of water by the so-called Watergate technique [Piotto et al., 1992J or by presaturation. In the case of a Watergate elimination, the attenuation reaches a factor of 105,

<Desc / Clms Page number 14>

une expérience HMBC donnant les corrélations hétéronucléaires ('H-13C) à longues distances (2JC-H et 3JC-H) [Bax & Summers, 1986], un proton Watergate de contrôle.

an HMBC experiment giving heteronuclear correlations ('H-13C) at long distances (2JC-H and 3JC-H) [Bax & Summers, 1986], a proton Watergate control.

 The pulse sequence used is that of an HMBC-GAS.

d'excitation basale dans la dimension li est de 500, 13 MHz. A cette fréquence basale est ajoutée une fréquence additionnelle 9031 Hz I 2Hz, fluctuant d'une mesure sur l'autre, car centrée sur le pic de l'eau. La fréquence d'excitation est de 125, 7728 MHz en dimension 13 C. La fenêtre spectrale en li est de 10,0982 ppm, soit 5050,50 Hz et de 222 ppm (soit 27921 Hz) en 13C. The acquisition parameters are as follows. Frequency

Basal excitation in the dimension li is 500, 13 MHz. At this basal frequency is added an additional frequency 9031 Hz I 2 Hz, fluctuating from one measurement to the other, because centered on the peak of the water. The excitation frequency is 125.7728 MHz in 13 C dimension. The spectral window in Li is 10.0982 ppm or 5050.50 Hz and 222 ppm (or 27921 Hz) in 13C.

 Each free precession signal (FID) is digitized in 2048 points. 256 increments (unit experiments of a 2D analysis) are recorded in 13C dimension. Each increment is the result of the accumulation of 16 interferograms (FID) under identical conditions, in order to increase the signal-to-noise ratio.

no 22 qui appartient au groupe des vaches de réforme. L'échelle dans la dimension horizontale correspond aux déplacements chimiques'H. Elle s'étend sur une fenêtre de 10,098 ppm autour de la résonance de l'eau. Le spectre projeté est celui du li Watergate. On note la présence du pic de l'eau fortement atténué au centre du spectre, ainsi que la forte redondance de signaux dans cette dimension, information qui est révélée par l'utilisation d'une technique multidimensionnelle. La seconde The data acquired in the time domain are multiplied by a window function of type sine squared, in order to increase the resolution of the signal, then they undergo a Fourier transformation in magnitude. Prior to obtaining the Fourier transform, a double zero-filling is performed in the carbon dimension, bringing the resolution from 256 to 512 points in this dimension. The acquisition time of this experiment is 2 hours, which brings the totality of the experiment to 2:40. An HMBC-GAS card is presented in Figure 1. This is the individual

# 22 which belongs to the group of cull cows. The scale in the horizontal dimension corresponds to chemical shifts'H. It extends over a window of 10.098 ppm around the resonance of water. The projected spectrum is that of the Watergate li. We note the presence of the peak of water strongly attenuated in the center of the spectrum, as well as the strong redundancy of signals in this dimension, information that is revealed by the use of a multidimensional technique. The second

<Desc / Clms Page number 15>

dimension, verticale, est la dimension 13c dont l'échelle s'étend sur 220 ppm.

dimension, vertical, is the dimension 13c whose scale extends to 220 ppm.

Integration
The information contained in the 2D NMR maps (HMBC) of each urine is translated into numerical data by calculating the volume of each correlation peak, using a computerized system. The latter applies a duly characterized pattern, defined once for all cards. Around each spot, sectors are delimited and the volume is calculated by adding the intensities of the points contained in the sector in question. Thus, each correlation task is digitized into a variable having for value the integral of the intensities on the given sector. The integration resulted in 266 variables. The data obtained is then subjected to multidimensional statistical processing.

Multidimensional statistical analysis of data obtained by HMBC-GAS
Filtration of variables
Variable filtering is used to define groupings of structurally and physiologically correlated variables. From these groupings, only the most informative variables are retained. Thus, of the 266 integrated variables, 94 were retained by the analysis of variance (F significant at the 5% threshold). They were then classified hierarchically according to their correlation profiles. Finally, a selection was made on the 30 candidate explanatory variables (significant at the 1% threshold) by the analysis of their partial correlations. At each step is selected the observed variable and are rejected all those that are linearly correlated to them. Figure 2 shows that the selected variables best describe all the explanatory variables with a minimum of information redundancy.

<Desc / Clms Page number 16>

The variables definitively retained are the following: varl83, var47, varl23, varl32, varl96, var180, var218, var175, var125, var58, var112, var266, var170, var133, var49, var142, var141. These variables mainly correspond to proton correlation spots between 1 and 4.3 ppm on the one hand, and between 6 and 7.5 ppm on the other hand.

Discriminant factor analysis
Discriminant factor analysis was performed on the matrix of reduced dimensions (18 variables retained), concerning a measurement for each lyophilization of a urine sample. The variability of the three groups is fully summarized in the first two factors. Figure 3 shows the projection on the plane defined by factorial axes 1 and 2 of individuals belonging to the following three groups: group 1: cull cows (F),
D group 2: castrated males (3-4 yr olds) (Mc), group 3: calves less than one year old (V).

 The discrimination is significant at the 10-4 level (n = 54). We see that the three groups are perfectly separated.

 This analysis highlights the possibility of discriminating animals according to their sex and age, that is to say according to phenotypic characters.

Validation of the decision rule
Once generated the rule of discrimination, the estimate of the errors associated with this rule provides a validation tool of this one.

 Cross-validation makes it possible to understand the robustness of the discrimination method, without resorting to a set of individuals to be tested, and to clarify the impact of a limited number of measures on the model [MacLachlan, 1992]. An individual is successively removed from the total statistical sample. For n individuals, n-1 samples are generated. On each

<Desc / Clms Page number 17>

 sample, an AFD is performed, and then the removed individual is screened. Its coordinates are calculated from the discriminating equations of the model obtained with n-1 individuals. The individual is then assigned to the group he is closest to, according to the prediction rule determined during the calibration of the AFD. This attribution is found in the confusion matrix. A misclassification rate relative to the original group is then calculated [Lebart, Morineau, Piron, 1998].

 The assessment of these rates validates the technique and assigns a confidence interval.

Confusion matrix

<Tb>
<tb> Predictions <SEP> Appartenances
<tb> (numbers <SEP> and
<tb> Initials <SEP> Predictions
<tb> percentages)
<tb> Groups <SEP> Total <SEP> F <SEP> Mc <SEP> v
<tb> F <SEP> 32 <SEP> 28 <SEP> 1 <SEP> 3
<tb>% <SEP> 100 <SEP> 87.5 <SEP> 3.13 <SEP> 9.38
<tb> Mc <SEP> 16 <SEP> 0 <SEP> 16 <SEP> 0
<tb>% <SEP> 100 <SEP> 0 <SEP> 100 <SEP> 0
<tb> V <SEP> 6204
<tb>% <SEP> 100 <SEP> 33, <SEP> 33066, <SEP> 67
<tb> Total <SEP> 54 <SEP> 30 <SEP> 17 <SEP> 7
<tb>% <SEP> 100 <SEP> 55.56 <SEP> 31.48 <SEP> 12.96
<Tb>
F: cull cows, Mc: steers aged 3-4, V: calves less than one year old
The individuals belonging to the groups of the first column are attributed according to each line to the different groups.

Estimation of classification error rates according to membership groups

<Tb>
<tb> Groups <SEP> F <SEP> Me <SEP> V <SEP> Error
<tb> total
<tb><SEP>SEP> Rate of <SEP> of <SEP> 12.50 <SEP> 0 <SEP> 33.33 <SEP> 15.28
<tb> ranking <SEP> (%)
<Tb>

<Desc / Clms Page number 18>

F: cull cows, Mc: steers aged 3-4, V: calves less than one year old
This error rate is relatively low, given the size of the sample, and its large variability. It stems mainly from the misallocation of 2 calves out of a total of 6, which is low compared to the numbers of the other groups. In addition, 3 cull cows were classified in the group of calves less than one year old. A good level of discrimination can be achieved by increasing the number of animals analyzed.

Example 2: Discrimination of cattle urine according to the anabolic treatment
The procedure applied to the urine of cattle implanted in 1999 with the trenbolone acetate-estradiol-17ss (Revalor-S) hormone combination is reported according to several treatment groups. These are urine from four categories: non-implanted castrated males called controls, treated with an implant at the beginning of treatment, an implant at the beginning plus a implant at the implantation period, or after 45 days, and finally 4 implants at the beginning of treatment. Two urine samples were taken at 10 and 90 days after implantation. This procedure comprises:
1-lyophilization of the urine, 2-NMR analysis of the lyophilisates obtained,
3-the integration of the NMR analysis data,
4-the statistical analysis of the results.

Sample preparation
The sample preparation protocol is the same as in Example 1.

NMR analysis of lyophilizates
For each urine, two samples of constant concentration (500 mg / ml) are analyzed by 2D NMR. A series

<Desc / Clms Page number 19>

d'expériences est enregistrée pour chaque lyophilisat. Une sonde de détection inverse'H-13 C est utilisée pour sa grande sensibilité proton. Sont enregistrés : - un spectre proton sans élimination du pic de l'eau, - un spectre proton avec élimination du pic de l'eau par la technique dite Watergate [Piotto et al., 1992] ou par présaturation. Dans le cas d'une élimination de type Watergate, l'atténuation atteint un facteur 105, une expérience HMBC donnant les corrélations hétéronucléaires ('H-13C) à longues distances 2JC-H et 3JC-n, [Bax & Summers, 1986], - un spectre proton de contrôle avec élimination du pic de l'eau par la technique dite Watergate.

of experiments is recorded for each lyophilisate. An H-13 C reverse detection probe is used for its high proton sensitivity. Are recorded: - a proton spectrum without elimination of the peak of water, - a proton spectrum with elimination of the peak of water by the so-called Watergate technique [Piotto et al., 1992] or by presaturation. In the case of Watergate removal, the attenuation reaches a factor of 105, an HMBC experiment giving heteronuclear correlations ('H-13C) at long distances 2JC-H and 3JC-n, [Bax & Summers, 1986] a proton spectrum of control with elimination of the peak of water by the so-called Watergate technique.

 The pulse sequence used is that of an HMBC-GAS.

The acquisition parameters are the same as those described in Example 1. The acquisition time of this experiment is 2 hours, bringing the totality of the experiment to 2:40. Figure 4 presents an HMBC -GAS for the individual no 4463. This is a calf treated with 4 implants of the Revalor-S type. The measurement was made on the slaughter levy.

Integration of NMR analysis data
The information contained in the 2D NMR maps (HMBC) of each urine is translated into numerical data by calculating the volume of each correlation peak, using a computerized system as described in Example 1.

 Integration has generated 313 variables. The data obtained is then subjected to multidimensional statistical processing.

<Desc / Clms Page number 20>

Multidimensional statistical analysis of data obtained by HMBC-GAS
Filtration of variables
Variable filtering is used to define groupings of correlated variables because they are structurally or physiologically related. From these groupings, only the most informative variables are retained. The analysis of variance revealed 106 explanatory variables (F significant at the threshold of 5%). The hierarchical classification was performed on the correlation profile (r2) between the candidate explanatory variables. The iterative selection on the analysis of the partial correlations (p2) of these candidate variables makes it possible to reject those which are redundant with the selected variables. This classification has been cross-checked with the partial correlation analysis.

Finally, 10 groupings of variables were selected. Only one variable partially correlated to the class variable is retained per group. The results of this filtration are presented in figure 5. This figure shows the interest of a drastic selection of the variables: out of 313 integrated NMR correlations, 10 have been selected so that they are all explanatory and present a minimum redundancy of information.

cluster et qui ont le p2 le plus élevé du cluster, avec un seuil de rejet de 104. Les variables sont les suivantes : var28, varl90, var63, varl87, var51, var49, varl27, varl83, var231, var215. Ces variables correspondent principalement à des taches de corrélation de protons se situant entre 1 et 4, 3 ppm, d'une part, et entre 6 et 7,5 ppm d'autre part. The selected variables are those that characterize a

cluster and which have the highest p2 of the cluster, with a rejection threshold of 104. The variables are: var28, varl90, var63, varl87, var51, var49, varl27, varl83, var231, var215. These variables mainly correspond to proton correlation spots between 1 and 4, 3 ppm, on the one hand, and between 6 and 7.5 ppm, on the other hand.

Discriminant factor analysis
Discriminant factor analysis was performed on the reduced size matrix (10 variables) for the 90 day sampling. The variability of the four groups is summarized in the

<Desc / Clms Page number 21>

trois facteurs discriminants. On rapporte sur la figure 6 les résultats de l'AFD sur le plan décrit par les deux premiers axes factoriels. Les groupes projetés sont les suivants : groupe 1 : témoins non implantés (T), groupe 2 : traitement avec 1 implant de stéroïdes (1), groupe 3 : traitement avec 2 implants de stéroïdes (2), groupe 4 : traitement avec 4 implants de stéroïdes (4).

three discriminating factors. Figure 6 shows the results of AFD on the plane described by the first two factorial axes. The proposed groups are: group 1: non-implanted controls (T), group 2: treatment with 1 steroid implant (1), group 3: treatment with 2 steroid implants (2), group 4: treatment with 4 implants steroids (4).

 Discrimination is significant at the 10-4 level (n = 30).

 These results allow excellent graphic discrimination which is also very highly significant. It is thus possible to discriminate the control animals from animals treated qualitatively on the first axis. This separation represents 69% of the total variability. The urinary biological signature of the control animals is very different from that of the implanted animals. On the other hand, the second axis discriminates the animals treated according to the implantation dose. This axis represents 24% of the total variance. Discrimination along this axis reveals the intensity of the treatment applied. This analysis highlights the possibility of discriminating the implanted animals (steroid-treated) of the controls, not only qualitatively but quantitatively, and this, from the tenth day of treatment (not shown), during the establishment of the physiological disturbances leading to to protein accretion and lipid melt.

Validation of the decision rule
The same validation test rules as those presented in Example 1 were used to attest to the robustness of the discrimination method established on the test set.

<Desc / Clms Page number 22>

Confusion matrix

<Tb>
<tb> Predictions <SEP> Appartenances
<tb> (numbers <SEP> and <SEP> Initial <SEP> Predictions
<tb> percentages) <SEP> s
<tb> Groups <SEP> Total <SEP> T <SEP> 1 <SEP> 2 <SEP> 4
<tb> T <SEP> 8 <SEP> 8 <SEP> 0 <SEP> 0 <SEP> 0
<tb>% <SEP> 0
<tb> 1 <SEP> 4 <SEP> 0 <SEP> 3 <SEP> 1 <SEP> 0
<tb>% <SEP> 0
<tb> 2 <SEP> 0
<tb>% <SEP> 100 <SEP> 0 <SEP> 10 <SEP> 90 <SEP> 0
<tb> 4 <SEP> 80008
<tb>% <SEP> 100 <SEP> 0 <SEP> 0 <SEP> 0 <SEP> 100
<tb> Total <SEP> 30 <SEP> 8 <SEP> 4 <SEP> 10 <SEP> 8
<tb>% <SEP> 100 <SEP> 26.67 <SEP> 13, <SEP> 33 <SEP> 33.33 <SEP> 26.67
<Tb>

T: controls, 1: treatment with 1 implant, 2: treatment with 2 implants successively, 4: treatment with 4 implants.

 The individuals belonging to the groups of the first column are attributed according to each line to the different groups.

Estimation of classification error rates according to membership groups

<Tb>
<tb> Groups <SEP> Error
<tb> total
<tb><SEP>SEP> error rate of <SEP> 0 <SEP> 25, <SEP> 0 <SEP> 10, <SEP> 0 <SEP> 0 <SEP> 8, <SEP> 75
<tb> ranking <SEP> (%)
<Tb>
T: controls, 1: treatment with 1 implant, 2: treatment with 2 implants successively, 4: treatment with 4 implants.

 The low error rate results from the fact that the sample is well described by the variables retained during the learning phase (establishment of the discrimination rule), which makes it possible to attribute the withdrawn individual with a risk of relatively small error. If we increase the number of introduced variables, the discrimination is not more precise. The rule of discrimination thus calibrated makes it possible to recognize a fraudulent situation of anabolic treatment on this type of animals.

<Desc / Clms Page number 23>

Legends of the figures
Example 1
Figure 1: HMBC-GAS map for individual # 22 (cull cow). The scale in the horizontal dimension corresponds to the chemical shifts lH. It extends over a window of 10.098 ppm around the resonance of water. The projected spectrum is Watergate, for guidance. Note the presence of the strongly attenuated water peak in the center of the spectrum, as well as the strong redundancy of signals in this dimension, information that is revealed by the use of a multidimensional technique. The second dimension, vertical, is dimension 13C, whose scale extends over 222 ppm.

 Figure 2: Filtration of variables by hierarchical classification from individuals describing biological variability in cattle rearing. The hierarchical classification makes it possible to classify according to their correlations the candidate variables. The variables whose name is indicated were retained at the end of the filtration. The aggregation index scale indicated on the dendrogram corresponds to the correlation coefficient of the clusters (groupings or singleton). The strong correlations at the bottom of the dendrogram are essentially structural.

The physiological correlations are above.

It appears that the selected variables are uncorrelated and have minimal information redundancy.

 The variables in the classification are significantly explanatory for a 5% probability of rejection. The variables defined as candidates are explanatory of the group structure because their F is significant (reject probability threshold set at 1%). Are then selected iteratively those whose p2 is maximum. Those that are linearly correlated to the selected variable are rejected. Thus, the selected variables are explanatory of the structure in groups and have very little correlation with each other. 18 variables were selected.

<Desc / Clms Page number 24>

 Figure 3: Discriminant Factor Analysis on the biological variability of bovine animals slaughtered. The first discriminant axis allows discrimination between females and calves (on the left), on the one hand, and castrated males (on the right) on the other hand. This discriminating axis explains 90% of the total variability of the game. The second axis explains 10% of the total variability. The adults are at the bottom, the calves at the top, which corresponds to a discrimination according to the maturity of the animal.

 The significance of the discrimination is calculated by the Wilks test, which gives the probability that the observed discrimination is due to chance. In this case, it is 0.0001.

Group captions:
F: cull cows, Mc: castrated males 3-4 years old,
V: calves under one year old.

Figure 4 : Carte HMBC-GAS pour l'individu 4463 traité par 4 implants de Revalor-S. La mesure est effectuée sur l'urine collectée à l'abattage. L'animal étudié a été implanté avec une quadruple dose pendant 90 jours. L'échelle dans la dimension horizontale correspond aux déplacements chimiques'H. Elle s'étend sur une fenêtre de 10,098 ppm autour de la résonance de l'eau. Le spectre projeté est celui du li Watergate, à titre indicatif. Noter la présence du pic de l'eau fortement atténué au centre du spectre, ainsi que la forte redondance de signaux dans cette dimension, information qui est révélée par l'utilisation d'une technique multidimensionnelle. La seconde dimension, verticale, est la dimension 13C, dont l'échelle s'étend sur 222 ppm. Example 2

Figure 4: HMBC-GAS card for the 4463 individual treated with 4 Revalor-S implants. The measurement is performed on urine collected at slaughter. The animal studied was implanted with a quadruple dose for 90 days. The scale in the horizontal dimension corresponds to chemical shifts'H. It extends over a window of 10.098 ppm around the resonance of water. The projected spectrum is Watergate, for guidance. Note the presence of the strongly attenuated water peak in the center of the spectrum, as well as the strong redundancy of signals in this dimension, information that is revealed by the use of a multidimensional technique. The second dimension, vertical, is the 13C dimension, whose scale extends to 222 ppm.

<Desc / Clms Page number 25>

 Figure 5: Filtration of variables by hierarchical classification of the candidate variables, cross-checked with an analysis of the partial correlations on the game steroidal treatments. The hierarchical classification makes it possible to classify according to their correlations the candidate variables. The variables whose name is indicated were retained at the end of the filtration. The aggregation index scale indicated on the dendrogram corresponds to the correlation coefficient of the clusters (groupings or singleton). The strong correlations at the bottom of the dendrogram are essentially structural.

The physiological correlations are above.

It appears that the selected variables are uncorrelated and have minimal information redundancy.

 The variables defined as candidates are explanatory of the structure in groups because their value of F is significant (threshold of probability of rejection 5%). Are then selected iteratively those whose p2 is maximum. Those that are linearly correlated to the selected variable are rejected. Thus, the selected variables are explanatory of the group structure and have very little correlation with each other. 10 variables were selected.

discrimination selon la dose implantée, c'est-à-dire selon le gradient d, intensité du traitement anabolisant. Le troisième axe (non représenté) explique les derniers 7% de variabilité. La significativité de la discrimination est calculée par le test À de Wilks, qui donne la probabilité pour que la discrimination observée soit due au hasard. Dans ce cas, elle est de 0,0001. Figure 6: Biological signature of anabolic treatments by Discriminant Factor Analysis. The first discriminant axis allows discrimination between control animals (on the left), on the one hand, and animals treated with anabolic (Revalor-S) for 90 days (on the right) on the other hand. This discriminating axis explains 69% of the total variability of the game. The second axis explains 24% of the total variability, and corresponds to a

discrimination according to the implanted dose, that is to say according to the gradient of intensity of the anabolic treatment. The third axis (not shown) explains the last 7% of variability. The significance of the discrimination is calculated by the Wilks test, which gives the probability that the observed discrimination is due to chance. In this case, it is 0.0001.

<Desc / Clms Page number 26>

Legend of groups:
T: non-implanted control animals,
1: animals having undergone anabolic treatment with 1 steroid implant,
2: animals having undergone anabolic treatment with 2 steroid implants,
4: animals having undergone anabolic treatment with 4 implants of steroids.

Bibliographical references
Beckwith-Hall BM, Nicholson JK, Nichols AW, Foxall PJD, Lindon JC, Connor SC, Abdi M, Connelly J & Holmes E. Nuclear magnetic resonance spectroscopy and the main component of the analysis of biochemical effects of three model hepatotoxins. Chem. Res. Toxicol. (1998) 11, 260-272.

 Daeseleire EAI, From Guesquière A & van Peteghem CH.

Multiresidue analysis of anabolic agents in muscle tissue and urine of cattle by GC-MS. J Chromatogr Sci (1992) 30, 409-414.

 Degand G, Schmitz P & Maghuin-Rogister G. Enzyme immunoassay screening procedure for the synthetic anabolic estrogens and androgens, diethylstiltestrol, nortestosterone, methyltestosterone and trenbolone in bovine. J Chromatogr (1989) 489, 235-243.

 Ferchaud V, Bizec B, Monteau F & André F. Determination of the exogenous character of testosterone in bovine urine by gaschromatography-combustion-isotope-ratio mass spectrometry. Analyst (1998) 123, 2617-2620.

 Gatti GC, Zoboli G, Zuchi M Ramazza V & Cantoni C. Urinary excretion of androstenedione in cattle as an indicator of possible illegal use in livestock breeding. Anal Chim Acta (1993) 275, 105-112.

 Hartwig M, Hartman S & Steinhart H. Physiological quantities of naturally occurring steroid hormones (androgens & progestogens), precursors and metabolites in beef of differing sexual origin. Z Lebensm Unters Forsch A (1997) 205,5-10.

<Desc / Clms Page number 27>

 Hurd RE and John BK. Gradient-Enhanced Proton-Detected Heteronuclear Multiple Quantum Coherence Spectroscopy. Journal of Magnetic Resonance (1991) 91: 648-653.

 Lebart L, Morineau A & Piron M. Exploratory Statistics lUUltidimensional. 2 Edition, 1998, Dunod, Paris.

 Mac Lachlan GJ. Discriminant Analysis and Statistical Pattern Recognition, 1992, John Wiley & Sons, NY.

 Massé R, Ayotte C. & R. Dugal Studies on anabolic steroids. I integrated methodological approach to the gas chromatography-mass spectrometric analysis of anabolic steroid metabolites in urine. J Chromatogr (1989) 489 (1), 23-50.

 Nadal L, Leray G, Desbarats C, Darcel F, Bansard JY, Bondon A & Some JD. Proton and phosphorus magnetic nuclear resonance spectroscopy of human brain tumors with automated data classification: a preliminary study. Cell. Mol.

Biol (1997) 43, 659-673.

 Piotto M, Saudek V. & Sklenar V. Gradient-tailored excitation for single quantum NMR in aqueous solutions. Biolol NMR (1992) 2 (6), 661-665.

 Ripley BD. Pattern Recognition and Neural Networks, 1996, Cambridge University Press, Cambridge.

 Scippo ML & Maghuin-Rogister G. Methods of detection and surveillance of natural sex steroid hormones, in Scientific conference on growth promotion in meat production, Europan commission, directorate general VI Agriculture, Brussels, Belgium, 29/11 to 1/12/1995 , 541-566.

 Scippo ML, Gaspar P, Degand G, Brose F and Maghuin-Rogister G.

Control of the illegal administration of natural steroids in urine and tissues of veal calves and in plasma of bulls.

Analytic Chimica Acta (1993) 275, 57-74.

 Somorjai RL, Nikulin AE, Pizzi N, Jackson D, Scarth G, Dolenko B, Gordon H, Russell P, Lean CL, Delbridge L, Mountford CE & Smith ICP. Computerized consensus diagnosis: A classification strategy for the robust analysis of MR spectra. I Application to

<Desc / Clms Page number 28>

 1H spectra of thyroid neoplasms. Magnetic Resonance in Medicine (1995) 33, 257-263.

 Tate AR, Watson D, Eglen S, TN Arvanitis, Thomas EL & Bell JD. Automated feature extraction for the classification of human in vivid 13C spectra using statistical pattern recognition and wavelets. Magnetic Resonance in Medicine (1996) 35,834-840.

 Van Ginkel LA, Stephany R, Spaan A & Sterk SS. Bovine Blood Analysis for Natural Hormones: An Overview of Analytical Strategies. Analytic Chimica Acta (1993) 275, 75-80.

 Venables WN & Ripley BD. Modern Applied Statistics with SPLUS. 3rd Edition, 1999, Springer-Verlag, NY.

 JWTE Vogels, CJM Arts, AC Tas, Van Baak MJ & van der Greef J.

Application of proton nuclear magnetic resonance spectroscopy and multivariate analysis as an indirect screening method for monitoring the use of growth promoters. in EuroResidue III, 6-8 May 1996, Veldhoven, The Netherlands, 968-972.

Claims (16)

1. Discrimination method with identification and / or identification of a biological disturbance situation, characterized in that a biological sample is subjected to a highly resolutive analytical technique and this technique is associated with a valid statistical or classification treatment which takes into account all the variations observed, the set of variations referencing a given biological disturbance situation.  2. Method according to claim 1, characterized in that

 subjects the biological sample to be studied to at least one multidimensional NMR measurement, followed by quantitative signal analysis, signal interaction, and statistical processing of the integration data.  3. Method according to claim 2, characterized in that the NMR analysis is a homonuclear analysis H-H and / or heteronuclear C two-dimensional.  4. Method according to claim 3, characterized in that it is an HMBC analysis and more particularly HMBC-GAS.  5. Method according to claim 3 or 4, characterized in that the acquisition parameters, excitation frequency in the dimension li and in the dimension 13C, as well as the spectral windows in the two dimensions are determined specifically for each type of analysis, and each interferogram (FID) is multiplied by a square sinusoidal function.  6. Method according to one of the preceding claims, characterized in that it performs a filtration of variables prior to the step of discriminating groups of individuals to validate the biological disturbance and at the classification stage d unknown individuals using the previously established rule of discrimination or classification. <Desc / Clms Page number 30>  7. Method according to claim 6, characterized in that the variable filtration is performed by a hierarchical classification coupled to an ACP and / or a hierarchical classification coupled to an analysis of variance and / or a partial correlation analysis, the analysis variance analysis and partial correlation analysis that can be performed on the entire set of variables or variables grouped in a cluster.  8. Method according to claim 1, characterized in that mass spectrometry as a highly resolutive analytical technique can be used complementarily or alternatively.  9. Method according to one of the preceding claims, characterized in that the biological samples are of animal, plant or microbiological origin.  10. Method according to any one of claims 1 to 9, characterized in that the samples come from biological fluids, tissues, cells, culture media.  11. The method of claim 10, characterized in that it is total lyophilisates or partial extracts or broyas.  12. Application of the method according to any one of claims 1 to 11, phenotyping of living organisms, to access the genetic component of an individual and / or reveal an environmental component.  13. Application of the method according to any one of claims 1 to 11, the characterization of the physiological state specific to a population.  14. Application of the method according to any one of claims 1 to 11, the identification and / or identification of a disturbance of a biological system by a substance with pharmacological activity or a microorganism.  15. Application according to claim 14 to indirect discrimination of the use of anabolic agents in breeding and / or in the sports field and / or in the biomedical field. <Desc / Clms Page number 31>  16. Application of the method according to any one of claims 1 to 11, to constitute a repository or a database.