The trajectory of gait development in mice

Abstract Objective Gait irregularities are prevalent in neurodevelopmental disorders (NDDs). However, there is a paucity of information on gait phenotypes in NDD experimental models. This is in part due to the lack of understanding of the normal developmental trajectory of gait maturation in the mouse. Materials and methods Using the DigiGait system, we have developed a quantitative, standardized, and reproducible assay of developmental gait metrics in commonly used mouse strains that can be added to the battery of mouse model phenotyping. With this assay, we characterized the trajectory of gait in the developing C57BL/6J and FVB/AntJ mouse lines. Results In both lines, a mature stride consisted of 40% swing and 60% stance in the forelimbs, which mirrors the mature human stride. In C57BL/6J mice, developmental trajectories were observed for stance width, paw overlap distance, braking and propulsion time, rate of stance loading, peak paw area, and metrics of intraindividual variability. In FVB/AntJ mice, developmental trajectories were observed for percent shared stance, paw overlap distance, rate of stance loading, and peak paw area, although in different directions than C57 mice. By accounting for the impact of body length on stride measurements, we demonstrate the importance of considering body length when interpreting gait metrics. Conclusion Overall, our results show that aspects of mouse gait development parallel a timeline of normal human gait development, such as the percent of stride that is stance phase and swing phase. This study may be used as a standard reference for developmental gait phenotyping of murine models, such as models of neurodevelopmental disease.


Supplementary Methods -Detailed Methods Description
gait parameters used for analysis ( Figure 1B). Gait was analyzed by quantifying components of the step cycle, or stride, broken into when a paw has contact with the ground, known as the stance phase, and when it is moving through the air, known as the swing phase ( Figure 1B). The stance phase is further broken down into the paw braking phase (heel strike to full stance) and propulsion phase (full stance to toe push off). All our trials comprised at least 12 strides based on previous work suggesting 9 strides or more are required for DigiGait data processing (Hampton et al., 2004) (C57, M=19.0, range: 12.0-24.5; FVB, M=18.9, range: 13.0-25.5). Selection of gait metrics for analysis. The DigiGait software analysis system outputs a comprehensive list of gait metrics. A concern with such a large list of variables is inflation of familywise error rate. To identify possible redundant variables, we generated scatterplot matrices. Pairs of variables that were visibly perfectly aligned were considered redundant and one variable within the pair was excluded from further analysis. The variables we excluded were stride duration, % brake of stride, % brake of stance, % propulsion of stride, % propulsion of stance, swing/stance ratio (Figures S1, S2), hindlimb shared stance time ( Figures S3, S4), and ataxia coefficient (Figures S5,S6) In addition to excluding variables based on redundancy, we also determined if variables required exclusion based on poor reliability. The post-video acquisition processing within the DigiGait software requires adjustment of filters to remove the snout and decrease noise from the digital paw prints, as well as manual corrections to errors within the paw contact area plots, introducing the possibility of inconsistencies across experimenters. We examined the inter-rater reliability of this processing between the measurements produced by two independent experimenters by calculating intraclass correlation coefficients (ICC) with their 95% confidence intervals using IBM SPSS Statistic software (v.25) based on absolute-agreement and two-way mixed-effects model. The data used for this were derived from an independent cohort of 10 FVB mice tested on P24, P27 and P30. For fore and hind limbs, respectively, 20/25 and 27/32 metrics showed excellent reliability (ICC ≥ .75) and another 2/25 and 2/32 showed good reliability (ICC = .60 -.74). The remaining metrics showed poor reliability (ICC ≤ .39; 3/25 and 3/32). ICCs are reported in Table 1. We excluded from further analysis any metrics with poor reliability: midline distance, axis distance, paw drag, and maximal rate of change of paw area contact during the propulsion phase. We also excluded tau propulsion despite good reliability because we felt this measure has not yet been defined or validated adequately in the literature and thus its usefulness is not clear.

Body Length Quantification
Animal body length was measured by importing videos from the Digigait software into Ethovision (Noldus Information Technology, The Netherlands, RRID:SCR_000441). The animal's body was automatically detected using the contrast of the darker body against a lighter background. A custom script provided by Noldus was used to calculate its length based on the coordinates of the nose, center point and tail base. A length measurement was calculated from those three coordinates for every frame of the video and then averaged into one score per animal for analysis. Thus, this accounted for any differences that occurred due to extensions or contractions of the body during a stride. To validate this method of body length measurement, the body lengths of a subset of FVB and C57 mice were also measured manually from the same videos at three different time points and then averaged. Intraclass correlation coefficients (ICC) with their 95% confidence intervals were calculated between the manual and automated measurements using IBM SPSS Statistics software (v.25, RRID:SCR_002865) based on absolute-agreement and two-way mixed-effects model. The ICCs indicated excellent reliability between the manual and automated measurements (FVB,ICC = .972,95% CI [.904,.992]; C57, ICC = .977, 95% CI [.955, .988]), providing confidence in and validation of our automated process for body length measurement.

Supplementary Figures
Supplementary Figure 1. Scatter matrices of spatiotemporal metrics measured in both fore and hindlimbs of C57 mice at P21, P24, P27 and P30. One variable from a pair that was considered close to or at perfect alignment was excluded from further analysis. Figure 4. Scatter matrices of spatiotemporal metrics measured in only the hindlimbs of FVB mice at P21, P24, P27 and P30. One variable from a pair that was considered close to or at perfect alignment was excluded from further analysis. Figure 5. Scatter matrices of postural and variability metrics measured in both fore and hindlimbs of C57 mice at P21, P24, P27 and P30. One variable from a pair that was considered close to or at perfect alignment was excluded from further analysis. Figure 6. Scatter matrices of Scatter matrices of postural and variability metrics measured in both fore and hindlimbs of FVB mice at P21, P24, P27 and P30. One variable from a pair that was considered close to or at perfect alignment was excluded from further analysis. Step Angle, (B) step angle coefficient of variance (CV), and (C) peak paw area CV appeared to significantly change with age in C57 mice. However, after adjusting for differences in body length from P21 to P30, these variables no longer significantly changed with age. Data are means ± SEM and covariate adjusted means ± SEM. Figure 8. Gait metrics that only reflected change in body length in FVB mice. Raw data means and covariate adjusted means are presented for all variables. (A) Propulsion duration, (B) absolute paw angle, (C) paw angle coefficient of variance (CV), and (D) peak paw area CV appeared to significantly change with age in FVB mice. However, after adjusting for differences in body length from P21 to P30, these variables no longer significantly changed with age. Data are means ± SEM and covariate adjusted means ± SEM.

Supplementary
Supplementary Figure 9. The directional trajectory of hindlimb stance duration in C57 from P21 to P30 is flipped when adjusted for body length changes. Hindlimb stance duration raw means and covariate adjusted means are presented for C57 mice. Before adjusting for body length, stance duration appeared to significantly increase. However, after adjusting for differences in body length from P21 to P30, stance duration significantly decreased from P21 to P30. Data are means ± SEM and covariate adjusted means ± SEM.
Supplementary Figure 10. Scatter matrices of Scatter matrices of postural and variability metrics measured in both fore and hindlimbs of FVB mice at P21, P24, P27 and P30. (A) C57 fore-and hindpaw lengths increase with age, as measured at full stance. Data are means ± SEM. (B) Scatterplots of the strong positive relationship between body length and fore-and hindpaw lengths from P21 to P30 in C57 mice. All data represented. (C) FVB fore-and hind paw lengths increase with age, as measured at full stance. Data are means ± SEM. (D) Scatterplots of the strong positive relationship between body length and fore-and hindpaw lengths from P21 to P30 in FVB mice. All data represented. Figure 11. Absolute paw angle and paw able coefficient of variation for C57 forelimbs.

Supplementary
(A) The postural metric absolute paw angle significantly changed across P21 to P30 for the forelimbs in C57 mice, but meaning of the change across time is unclear. (B) The variability of this metric, paw angle CV, significantly decreased from P21 to P30.

Supplementary Tables
Supplementary Table 1

Supplementary Discussion
Strain Comparisons: While different strains of mice often show differential performance in in many behavioral assays (Eisener-Dorman et al., 2011;Keum et al., 2016;Liu et al., 2011;Martin et al., 2014;S. S. Moy et al., 2004;Sheryl S. Moy et al., 2008), our study was not designed to directly compare the two strains tested so our ability interpret any differences is limited. The differences observed may arise from size discrepancies, as FVB mice are consistently larger, or experimental parameters, such as different sample sizes. It is also possible that motor development in the FVB mouse occurs earlier than in the C57 mouse, as suggested by the fact that FVB mice were able to run successfully on the DigiGait at a younger age (P17) than C57 mice (P21, pilot data not shown). Future direct comparison studies are needed to validate strain differences suggested by the present study.
In any case, our results provide a reference dataset of gait maturation for two commonly used mouse strains, often employed as the genetic backgrounds for models of disease.
Body Length: We also recognized that the rapid growth in body size across the age range examined could confound our gait data, and indeed we found that more than a third of all metrics were significantly influenced by body length. Decreased stride frequency, or cadence, and increased stride length have been suggested as markers of gait maturity in humans, however they are mainly driven by limb lengthening and thus likely do not reflect underlying motor circuit maturation -a more relevant phenotype for brain disorders like NDD. In both strains examined, we initially observed a comparable pattern of decreased stride frequency and increased stride length. However, body length significantly impacted both of these variables, and controlling for this influence revealed that neither stride frequency nor length changed with age from P21 to P30. Thus, we presented the data both before and after controlling for the impact of body size on each gait parameter to highlight the possibility of erroneous interpretations when body size is not considered, and to help define those features that could reflect true differences in CNS circuits rather than simply changes in limb length.