Subjects
All procedures were carried out in accordance with Guidelines for the Care and Use of Laboratory Animals, and approved by the common Ethical Committee of the National Veterinary School of Alfort, ANSES, and UPEC. Eight healthy golden retrievers and 20 GRMD dogs (all males) were included in the study. All animals came from the French GRMD colony. The healthy dogs were littermates of some of the GRMD dogs used. Six of the healthy dogs and 12 of the GRMD dogs had participated in a previous study [10]. All dogs were housed in the same facilities, and were genotyped as previously described [15]. Only the GRMD dogs that were still ambulatory after 9 months of age were included in this study.
Evaluation of gait quality using 3D accelerometry
As previously described [11], the 3-dimensional accelerometer recorder used was a Locometrix® gait analysis system, composed of three orthogonally positioned accelerometers, which can record accelerations along the dorso-ventral, cranio-caudal, and medio-lateral axes.
Immunosuppressed dogs
To assess the efficiency of our method, we analysed gait data acquired for four immunosuppressed dogs that had been published in a previous paper [12]. These dogs had been treated with high doses of oral prednisolone (2 mg/kg/d) and cyclosporine A (initial dose of 20 mg/kg/d) between 2 and 9 months of age.
Gait testing
Dogs were carried from the kennel to a 45-meter-long testing corridor located close to the laboratory facilities, as previously described [10]. The belt to which the accelerometric device was attached was fastened around the thorax of the dog, near the centre of gravity at rest. Each animal was tested twice per month from 2 months of age (when motor clinical signs appear in animals that survive the neonatal period) to 9.5 months of age, thus covering the period of growth and disease progression [4]. To enhance the discriminatory power of the method, values obtained for GRMD dogs that survived to 12 months were also included in the analysis. Young puppies were familiarized with the corridor and the belt before the test. The height at withers (HW) was measured at the end of each test. All tests were performed by the same experimenter (IB). In each test, the dog was encouraged to walk or run at its preferred gait, and its speed calculated over a distance of five meters, as previously described [11].
Data analysis
For quadrupedal gait analysis, acceleration curves were analysed using the software provided by the manufacturer of the recording device (Equimetrix®, Centaure Metrix, Evry, France). A 10-second sequence of steady-state locomotion, which was easily identifiable in the dorso-ventral acceleration curves, was analysed. The following variables, which have been previously described in detail [10, 11], were computed: stride frequency (SF, /s), stride regularity (Reg, dimensionless), total power of accelerations (TP, W/kg), relative components of the total power along the three axes (%) (calculated by dividing cranio-caudal (CCP), dorso-ventral (DVP), or medio-lateral power (MLP) by total power (TP)), and stride length (calculated by dividing the speed by SF), which was normalized to height at withers (SL/HW) in order to circumvent the effect of limb length on this variable. For the sake of consistency, only observations with a Reg value >70 were considered for analysis. Gait testing consisted of two consecutive round trips in the corridor. However, in contrast to healthy dogs, it was very difficult, if not impossible, for some GRMD dogs to complete the second round trip.
Discriminant analysis
The main objective of the present study was to assess the capacity of discriminant analysis (DA) to evaluate gait, detect functional alterations, and evaluate treatment benefits during the growth period in GRMD dogs. We used gait data obtained from healthy and GRMD dogs and performed DA using XLSTAT software (Addinsoft™). This Excel add-in extends the analytical functions of Excel and covers the key requirements for data analysis and statistics. DA is a commonly used multivariate data analysis method. The aim of this supervised method is to predict group memberships of a set of individuals based on multivariate data. In this scenario, the groups of individuals are assumed to be known a priori. DA reduces the dimensionality of the data at hand by computing synthetic variables, often called canonical variables or factors, the aim of which is to maximize inter-group variance while minimizing intra-group variance. DA yields new variables, which are linear combinations of the original variables. The maximum number of such variables is equal to the number of groups minus one and are usually noted F1 to F(n-1) where n is the number of groups. However, in practice, only the first few canonical variables are used for the purpose of discrimination, since the remaining canonical variables may be predominantly associated with noise present in the data. The graphical displays generated using the retained canonical variables are useful to depict the separation of groups. DA methods include linear DA (LDA) and quadratic DA (QDA) [16]. LDA is a parametric method that assumes Gaussian distributions with the same variance-covariance matrix within the various groups. In practice however, this is often not the case. Nonetheless, even in the case of a slight deviation from this requirement, LDA performs reasonably well [17], and was thus the method selected for the present study.