Several SCI experiments were performed using female C57BL/6J mice (Charles River Japan, Yokohama, Japan), including experiments to test potential therapeutics as well as impact force optimization experiments. The data used in this study was collected from approximately 800 measurements taken from 108 mice.
Mice were anesthetized using an intraperitoneal injection of a mixture of 0.3 mg/kg Medetomidine, 4 mg/kg Midazolam, and 5 mg/kg Butorphanol. Anesthesia was confirmed by pinching the hind paw prior to surgery. The spinal cord was exposed by laminectomy at the 11th thoracic vertebra. Contusion SCI was performed using the Infinite Horizons IH-0400 Impactor  (Precision Systems and Instrumentation LLC, Fairfax, VA, USA) using peak impact forces between 40 and 70 kdyne. The animal’s bladders were manually pressed every day to drain the urine and prevent urinary tract infections. After all motor function measurements were completed, mice were anesthetized again and euthanized by cervical dislocation.
Motor function was monitored using the Basso Mouse Scale (BMS) and Noldus CatWalk XT (Noldus Information Technology, Wageningen, The Netherlands) every week for 6 weeks post-injury.
BMS data was collected by observing a mouse in a 30 × 30 × 15 cm plastic cage for 5 min. Notes were taken describing several aspects of the animal’s gait, and BMS scores were calculated according to a flowchart .
CatWalk data was collected using a gain of 0.18, with a maximum compliant run time of 12.5 s. When possible, three compliant runs were recorded for each mouse, but poorly performing mice were often unable to produce runs faster than 12.5 s. In these cases, three non-compliant runs were collected. Runs were rejected if the animal turned around during the run. Post-injury data from each mouse was compared to a pre-injury baseline made of three pre-injury measurements to account for naturally occurring variation in motor function.
The Combined CatWalk Index (CCI) was developed by correlating 104 CatWalk parameters against observed BMS scores using linear regression, then combining all linear regression equations into a single weighted average (Fig. 1a). The R2 values are used as the weighting values, so that parameters with strong correlation have strong weights, while poorly correlating parameters have weak weights. Calculations were performed using both Microsoft Excel 2013 and LibreOffice Calc version 5, both programs produced identical results. Linear regression was performed on each parameter using the built-in SLOPE, INTERCEPT, and RSQ functions, which calculate the slope, Y intercept, and R2 values for a linear regression equation (Fig. 1c). Using these formulas is much simpler than preparing a separate scatterplot for each parameter and linear regression. BMS scores were used as the Y-axis data and CatWalk parameters were used as the X-axis data. The CCI coefficients (slope, intercept, and R2 values) for each parameter are listed in Table 1, in order from highest to lowest R2.
The CCI Score was calculated by combining all equations into a weighted average. “Adjusted CatWalk Parameters” were calculated using the original CatWalk data and CCI coefficients with this equation: R2 (Slope × CatWalk Parameter + Intercept) (Fig. 1d). Adjusted CatWalk Parameters were then summed and the sum was divided by 104, the number of CatWalk parameters used in this analysis. Mock ups of the spreadsheets used to calculate the CCI coefficients and CCI scores are shown in Fig. 1c, d, and an example spreadsheet containing the BMS and CatWalk data from impact force optimization experiments is available in Additional file 1.
CCI scores were calculated for every mouse at every time point based on the CatWalk data and the CCI coefficients in Table 1. The CCI scores were plotted against the corresponding BMS measurements and linear regression was used to determine how well the two scores correlated against each other, and an R2 value of 0.7093 was obtained (Fig. 1b). This R2 value is not perfect, but may be a reflection of the BMS system only being semi-quantitative. For example, if a mouse’s BMS score changes from 1 to 2 (a change from only showing partial ankle movement to full ankle movement without plantar placement), it is not the same as a change from 2 to 3 (the mouse shows plantar paw placement without weight support).
Table 1 shows that the CatWalk parameter that most closely correlates with BMS scores is the Step Sequence Regularity Index (SSRI), with an R2 of 0.7048, slightly lower than the R2 value for the correlation between BMS score and CCI score, but this difference is likely insignificant. The SSRI measures coordination by determining if the order of footprints falls into one of six regular patterns. Mice with poor motor function have poor coordination, and do not follow these regular patterns well, producing low SSRI scores. SSRI is often reported in studies that use the CatWalk system [4, 7,8,9,10,11], and the high correlation with BMS scores supports this practice. SSRI was chosen to represent CatWalk data when comparing CCI to CatWalk data in Fig. 2.
Data from the impact force optimization experiments was used to compare CCI scores to BMS scores and SSRI scores. Mean scores and sample standard deviations for each impact force were plotted for every time point (Fig. 2a–c). CCI and SSRI scores are presented as a percent of pre-injury baseline to account for naturally occurring variation between mice. BMS scores are presented directly, because all non-injured mice have a BMS score of 9, so there is no pre-injury variation. All three methods showed similar trends, with higher impact forces producing lower scores. All three methods show fairly large standard deviations, demonstrating the difficulty in producing consistent levels of injury with the contusion SCI model. This is in part due to the difficulty of controlling the Infinite Horizons impactor’s peak impact force. Actual impact forces were usually higher than the desired impact force, with substantial variation.
Average scores for each impact force were estimated by calculating the mean score across weeks 1–6 and were plotted against impact force (Fig. 2d–f). Linear regression was used to determine how well each method correlated with impact force. The CCI scores had a slightly higher R2 value (0.8854) than BMS (0.8636) or SSRI (0.8557). This indicates that the CCI score correlates well with the peak impact force in this contusion SCI model.
Coefficients of variation (CV) at each impact force and time point were determined by dividing the sample standard deviation by the mean. Each method’s CVs were averaged and compared using an unpaired, two-tailed T Test (Fig. 2g). The CCI method showed significantly smaller CV than either the BMS method or the SSRI method (P < 0.0001). This implies that the CCI method may be more precise than either other method.
One advantage of the human-observation based BMS method is that it produces a single score that can be easily be compared between mice, but suffers from the potential for human error and the requirement for training. The CatWalk system has the advantage of greater objectivity, but the large number of measurements can complicate several things, such as choosing parameters for publication or making comparisons between mice. For example, if one of set of mice has better coordination, but another set of mice has higher speed, which set of mice has better overall motor function? The Combined CatWalk Index appears to combine the advantages of both system by creating a single number based on objectively determined data. In addition, the CCI scores have slightly better correlation with BMS scores than any individual CatWalk parameter (Fig. 1b, Table 1), slightly better correlation with impact force than BMS scores or SSRI scores (Fig. 2d–f), and significantly smaller coefficients of variation than BMS scores or SSRI scores (Fig. 2g).
Although the CCI method requires a specialized instrument, the CatWalk system is fairly simple, and users can be quickly trained to measure mice. The BBB, BMS, and TMS systems require more extensive training, and steps must be taken to remove human bias or interrater variability. Additionally, the CCI method could potentially be modified and applied to any disease model that can be studied using the CatWalk system, such as chronic pain , arthritis, or vestibular disease . The main requirement is to have some semi-quantitative method to rank mice so that correlation between rank and CatWalk parameters can be determined.