The treadmill based scoring system presented here represents an inexpensive method of quantifying gait reliably in spinal cord injured paraparetic dogs with a wide range of disability. We demonstrated that neurologically normal, orthopedically sound and lame dogs of different sizes walk with a lateral gait (RF-LH-LF-RH) at a natural speed on the treadmill and we developed a scoring sheet that allows documentation of foot fall sequence from videotapes of dogs walking on the treadmill. Scores of both coordination and stepping frequency were developed and found to have high inter- and intraobserver reliability and to correlate closely with a previously developed ordinal scale of open field walking.
Our major requirements of a gait scoring system in spinal cord injured dogs were that it did not use expensive or specialized equipment, to allow multiple centres to acquire data for clinical trials; that it could be applied to dogs of varying size and conformation, over a wide range of disability; that it would not be affected by existing orthopaedic disease; that it did not require prior training; and that it provided discriminating continuous data with a high level of repeatability. Ordinal open field gait scoring systems are practical and have demonstrated utility and reliability [2, 7, 11, 12, 18], but do not generate continuous data and are limited in sensitivity. Such scoring systems are usually complemented by the addition of scoring techniques that assess different aspects of gait and provide continuous and highly discriminative data when used experimentally (e.g. [8, 19]). The spatiotemporal characteristics of gait are altered dramatically by spinal cord injury and can be quantified using kinematic analysis systems and PSWs. However, a major challenge posed by attempting to quantify this type of data is the inherent variability in many of the stride characteristics due to differences in size and conformation between breeds [14–16]. Accounting for covariates such as tibial length and velocity of walking can allay this problem and data generated in dogs with spinal cord disease using the PSW demonstrated that the fore limb stride length, swing and stance time shortens, and the hind limb swing time increases significantly with spinal cord injury . However, correlation to injury severity was lacking, likely because only independently ambulatory dogs could negotiate the walking surface. Moreover, the need to perform multivariate regression to factor in covariates complicates analysis in a clinical caseload of patients. Detailed kinematic analysis underlined the presence of inter-individual variability in many parameters, and indeed, the most consistent measures generated quantified the variability in both diagonal coupling time (time elapsed between fore limb and contralateral hind limb placement) and lateral paw placement [15, 16]. In normal dogs, these parameters do not vary, resulting in values close to zero. Changes in these parameters are sensitive markers of spinal cord injury severity [15, 16] and such data collection is objective but unfortunately costly. With the constraints of cost and availability, the use of kinematic data collection systems and PSWs was not viable for our purposes.
Given the need for a low cost system, we adapted strategies used in experimental models of spinal cord injury to count footsteps and evaluate step sequences. The two systems used most commonly in rodent work are foot print analysis accomplished by inking feet and evaluating the foot steps left on a paper surface , and catwalk analysis, a system in which the subject walks on a side-illuminated glass sheet that illuminates each paw as it is placed on the glass . The subject is videotaped from below and the videotapes analysed after the fact. These systems also allow collection of data such as base of support, stride length, and foot position [8, 9], data that we could not collect from our system, but data that is problematic due to the covariates encountered in dogs. We chose to collect our data when dogs were walking on a treadmill in order to allow us to control walking speed for individual dogs throughout testing [15, 16]. It is reported that people walk differently on a treadmill versus the open field , and so data on coordination in normal dogs walking on a treadmill was collected at the start of the study. The use of support for non-ambulatory dogs enables the investigator to quantify limb movements in even the most severely affected animals and has been shown not to affect gait patterns in normal dogs . By focusing simply on quantification of the numbers of steps taken and the sequence of footfalls, the issues associated with quantification of factors such as stride length were avoided.
Descriptions of quadrupedal gait date back to the 1800s when Muybridge  used photography to document gait sequences in a range of species walking in the open field. At that time he described a canine walking gait as RF, LH, LF, RH (now called the lateral gait) and included documentation of a pacing gait (in which both fore and hind limbs on the same side are placed on the ground simultaneously) as dogs increased speed to transition to a trot. Beyond this early work, detailed literature on the frequency with which particular stepping sequences are used in walking dogs on and off the treadmill is limited . Footfall sequences have been described in detail in rats walking in the open field and a total of six different sequences, composed of two mirror versions of three basic patterns: cruciate, alternating and rotary, were documented . Of these patterns, 80% of female wistar rats used the ‘alternating b’ pattern that corresponds to the lateral pattern in dogs . Our data revealed that a large cohort dogs of different sizes, and including lame dogs, adopted the lateral pattern [17, 22] when walking at different speeds on a treadmill. While some of these dogs did occasionally adopt a pacing gait when walking at faster speeds in the open field (unpublished data), this was not recorded while they were walking on the treadmill. A small number of dogs had brief periods of more random stepping sequences when walking at the slow speed, in particular the lame dogs, likely because of attempts to shift weight off the lame limb(s), but all walked with the lateral pattern when the speed of the treadmill was increased. We therefore adopted the lateral pattern as our definition of normal foot sequences when scoring videotapes of gait.
We were interested in both the number of hind limb steps that are taken and the fore limb hind limb coordination because it is proposed that animals can recover walking ability through increased activation of the local central pattern generators and through recovery of ascending and descending input via the long tracts of the spinal cord . The sequence of foot falls in quadrupedal mammals is controlled by both local ascending and descending propriospinal pathways that influence the thoracic and pelvic limb central pattern generators, as well as by supraspinal input [25, 26]. The coordination of fore and hind limbs is therefore affected by spinal cord injuries [15, 27]. It has been shown that accurate quantification of coordination between fore and hind limbs provides a sensitive measure of spinal cord injury in rats  that complements ordinal scales such as the BBB scale. In addition, the recovery, or lack thereof, of coordination in the presence of recovery of stepping ability potentially provides information on the mechanism of motor recovery.
The parameters that we scored were the ability to take a hind limb step, and the percentage of hind limb steps that occurred within a normal stepping cycle with the rationale of providing a simple measure of stepping frequency and strength (the stepping score) and of ascending and descending input (the coordination score). An alternative method of computing coordination known as the regularity index (RI) has been reported in rats  and can be calculated readily from our data by multiplying the number of footsteps taken in a normal stepping sequence by 4 to give the total number of coordinated steps and dividing by the total number of steps taken . The stepping score can be used to assess motor function by quantifying the number of weight-bearing steps, related to frequency and strength of stepping, and by quantifying the number of steps taken when the dog is walked with support. The extremely high correlation between the stepping score and the OFS supports the view that the stepping score quantifies degree of successful ambulation. Recovery of coordination is less complete than stepping, as reflected in the lower coordination scores. The coordination score correlated less closely (although still significantly) with the OFS than the stepping score, reflecting the fact that the OFS, unlike the BBB scale used in rodent spinal cord injury research , does not focus on quantifying coordination between fore and hind limbs. It is interesting to note that recovery of coordination was quite variable in the ambulatory dogs, for example, dogs 17 and 18 were both stepping with their hind limbs as frequently as their forelimbs, but in dog 17 only 60% of the hind limb steps were coordinated in a normal stepping cycle, compared with 86% in dog 18.
The definition of a footstep used was refined during the development of the scoring system to enhance scoring consistency and to capture all stepping movements. The observers in this study had very different backgrounds and included a veterinary student with no prior training in gait analysis, and three veterinarians one of whom had never worked in the field of spinal cord injury and two who were experienced in this field. The high repeatability of the data between these individuals and upon repeat scoring by one individual supports the utility of this scoring system. However, wider use of the system will allow more rigorous assessment of the extent of training needed to maintain consistent results in a wider population of observers. The major drawback of this scoring system is that, while videotapes of dogs can be obtained very efficiently, it is time intensive to score the videotapes. Time taken to score each videotape was not recorded by the observers in this study, but typically, it takes approximately 5 minutes to score a neurologically normal dog or a completely paraplegic dog, while paraparetic dogs could take as long as 30 minutes to score due to the need to watch the videotapes at reduced speed to accurately identify each foot step.