Dogs in this study were presented to the Dept Veterinary Medicine for treatment of clinical SCI, or were owned by members of staff (normal dogs). The study was carried out in the UK under the jurisdiction of the Veterinary Surgeons Act (1966).
Normal dogs (n = 9)
These were of a variety of breeds and types, and owned by members of staff in the Department.
Spinal cord injured dogs (n = 20)
We restricted this investigation to dogs that had incurred thoracolumbar SCI, but exhibited perceptible stepping movements in the pelvic limbs, the clinical details are included in Table 1. Stepping activity was precisely defined by using data derived from digital gait data on step length; only dogs that exhibited pelvic limb steps of at least 40% of the length of the thoracic limbs were examined (in normal dogs pelvic limb steps vary between 80–109% of thoracic limb step length [data on file]).
SCI dogs were divided into two sub-groups:
a) 'Incomplete' SCI (n = 13)
These dogs had sustained thoracolumbar SCI but had not lost sensory function to the paws of the pelvic limbs. The term 'incomplete' is used here to facilitate comparison with human SCI patients and this group of dogs was categorised as having injuries equivalent to ASIA grade D . These dogs had sustained SCI sufficiently severe to prevent them from walking unaided but not to lose conscious pain perception in the pelvic limbs and had undergone conventional clinical treatment (decompressive spinal surgery or cage rest and physical therapy) and subsequently recovered the ability to step on the treadmill with support. These dogs were examined at a stage of recovery at which stepping occurred, although they were unable to walk without support of the hindquarters (and therefore exhibited comparable stepping competence to their counterparts that had 'complete' SCI – see below).
To permit comparisons of reliability of foot placement in the lateral plane at different stages of recovery after incomplete SCI we further sub-divided data on this cohort of dogs into two groups according to the extent of coordination between the thoracic and pelvic limbs. Thus, we compared the variability of foot placement in the y plane at the stage at which we could define intergirdle coordination as poor (i.e. they had a 'mean diagonal coupling interval' [see ] of greater than 0.1) with that at which they had recovered normal intergirdle limb coordination (i.e. mean diagonal coupling interval of less than 0.1).
b) 'Complete' SCI (n = 7)
These dogs had chronic SCI (onset was at least six months previously) with absent sensory function (including absent conscious pain perception) and absent voluntary movements in the pelvic limbs; these dogs therefore were equivalent to humans with ASIA grade A injuries. Despite the chronic severe SCI, all dogs in this category were able to make pelvic limb stepping movements when partially supported on the treadmill (this effect is sometimes referred to as 'spinal stepping' – see ) and they fulfilled the criteria of exhibiting pelvic limb step lengths of at least 40% of that of the thoracic limbs.
The methods of acquiring the digital gait data have been described previously . Briefly, dogs were held by a leash around the neck and encouraged to walk on the treadmill, until they found a comfortable speed (varying from 1–3 kmh-1) at which to walk. SCI dogs required support in the form of a support band placed under the abdomen to prevent falling during locomotion. Tension in the support band was adjusted so that the thoracolumbar part of the vertebral column was parallel to the treadmill belt surface and minor adjustments were made to maximise the intensity of stepping movements. Since ~60% of body weight is supported through the thoracic limbs of normal dogs and there is variation from instant to instant in the precise loading of the limbs during acceleration and deceleration  we estimate the load bearing of the band to be between ~10 and 35% of body weight during a step cycle, depending on the degree of lateral sway.
Four infrared motion capture cameras with a recording frequency of 100 Hz (Qualisys, Sweden) were positioned around a standard ex-gymnasium treadmill and calibrated to permit recording from the entire belt surface. The lateral plane was designated as the 'y-plane' in these recordings. 10 mm reflective markers were attached to the skin overlying specific anatomical landmarks: lateral fifth phalange, lateral humeral epicondyle, ulnar styloid process, greater trochanter of the femur, lateral femoral epicondyle, lateral malleolus of the tibia, and the interscapular region dorsal to vertebra C7.
During recording, the treadmill speed was set to the speed determined previously, at which the dog was walking consistently, and 60 seconds of motion was recorded. The treadmill speed was then increased and recordings repeated at a variety of speeds. In some normal animals we examined the effect of applying belt support, to act as a control for this means of assistance that was essential for dogs that had spinal cord injuries.
Processing of the recorded images was initially carried out using Qualisys Track Manager software (Qualisys QTM, Sweden). The 15 individual markers were identified and labelled to construct a 3D stick-diagram representation of the dog. Visual examination of lateral and forward movement was displayed in QTM graphical plots of y- (lateral) and x- (sagittal) plane position and used to exclude sections of data in which a dog was not walking consistently (recognised by abrupt irregularity in the plot, indicating acceleration or deceleration, or sudden movement to one side). Subsequent analysis of coordination was focussed on data obtained from the paws only (phalangeal markers). Positional data from y-plane plots of each paw were exported from QTM into Matlab (Release 14, Student version) and the distance between consecutive paw placements (i.e. from left to right and vice versa) of both limb girdles was calculated using a custom-designed script. For each paw, the instant of each consecutive paw placement was identified from the maximal stationary values in the sagittal (x-plane) data (corresponding to the maximal extent of paw protraction immediately prior to its placement on the treadmill surface); the y-plane coordinates of the paw at that instant were then extracted by the Matlab script and exported into an Excel file (and see Additional File 1).
Initially therefore we determined the base of support for the thoracic and pelvic limb girdles respectively. Because of the variability in conformation of our study group of dogs we anticipated the need to apply a formula to determine normal ratios for the base of weight support in the thoracic and pelvic limb girdles. However, more importantly, in this study we were primarily interested in the degree of variability in foot placement, rather than the mean position (since that would be restricted by the anatomy of the pelvic limb). We reasoned that animals that exhibited greater unreliability in placement would have an abnormally wide distribution of values compared to the mean. Therefore we calculated the coefficient of variation in foot placement by using the base of support data and dividing the standard deviation by the mean for the thoracic and pelvic limb pairs individually . We then compared this ratio between groups of dogs. Because we were interested in whether there was a difference in inconsistency of foot placement between the pelvic and thoracic limb girdles we then derived a second ratio – between the coefficient of variation of the pelvic and thoracic limb girdles. This parameter was then also compared amongst the groups.
Estimation of limb length
Measurement of 'limb length' in dogs is not straightforward, since different conformations of dog will stand with the joints at different angles, therefore we measured tibial length as a surrogate that would accurately represent the range in size of dog in each group. These measurements were made from the digital data acquired through the QTM software indicating the relative locations of the stifle (knee) and hock (ankle) joints. Tibial lengths for dogs in each group are listed in Table 2.
Data acquired in QTM were transferred as numerical data into Matlab. A custom-written script was used to extract the data points of interest and standard matrix addition or subtraction was used to calculate time intervals and position as described above. The resulting data was assembled in Excel spreadsheets and transferred into GraphPad Prism (Version 5.0 for Windows) for statistical analysis.
For each animal there were columns of data listing the distance between the intragirdle paw pairs at placement on the treadmill. From these we calculated the means, standard deviations and coefficients of variation for comparison between different groups. All groups of data (normal, incomplete and complete SCI) were initially compared using the Kruskal-Wallis test, followed by post hoc Dunn's tests where appropriate to determine differences between specific groups if significance was detected in the Kruskal-Wallis test. Where this occurred we have reported results of post hoc tests, full details are given in figure legends. Paired Student's t tests were used to compare data derived from normal animals at different speeds and walking with and without abdominal band support. The Mann-Whitney test was used to compare data from normal and lame dogs. For all tests, significance was assumed when p < 0.05.