Analysis of locomotor behavior in the German Mouse Clinic
Introduction
Recent advances in genome editing technology have revolutionized and accelerated the generation of mutant mouse lines for modeling human disease. This technological cataclysm has brought with it an upsurge in demand for comprehensive behavioral phenotyping that is reflected in the rising number of large-scale phenotyping centers and consortia (Infrafrontier Consortium, 2015, de Angelis et al., 2015). To meet the demand, new sophisticated phenotyping devices have been developed to provide an automatic and detailed characterization of mice. The goal of automation is not only to speed up data acquisition and to make it more objective and replicable, but also to reduce the burden on the animal through less handling. Ideally, a few automated systems would be sufficient to comprehensively phenotype a mouse model. The questions repeatedly arising in this context are how valid are individual tests; which test is the most informative and how many tests are really needed. It would also be beneficial to know if parameters of one test could predict the utility of another test for more detailed phenotypic characterization. Such knowledge would enhance the efficacy of phenotyping strategies, which have to be outlined in license applications for experiments on animals, thus saving time, animals and money.
Here we explore these questions with respect to locomotor phenotyping. Would it be sufficient, for example, to use the Catwalk (CW) from Noldus; an automated gait analysis system with video based paw tracking (Hamers et al., 2001, Batka et al., 2014)? Previously the assessment of gait parameters involved footprint analysis, where the paws were dipped into ink and the animal walked across a paper. Approximately 20 parameters were measured from these prints, however with the automated CW system, around 230 parameters are measured. With this dramatic increase in parameters/data points data analysis has become more complex. Traditionally only single parameters or a few selected ones were analyzed. But it is hard to formulate a clear picture of the phenotype when dealing with a large amount of parameters for one test. Thus it is necessary to decide which of the parameters give most information and is the best for detecting disease-relevant phenotypes. Alternatively, a compression of the data may be more appropriate to garner a comprehensive overview. As such reducing the number of parameters under consideration (aka dimensions) to a few components (i.e. artificial variables) by Principal Component Analysis (PCA) improves data handling without significant information loss (Budaev, 2010, Gonik et al., 2012, Vannoni et al., 2014, Ferguson et al., 2013, Ohl et al., 2003). The subsequent analysis then runs on the extracted components that substitute for the more numerous original parameters.
Running a PCA on behavioral data is not new; several groups have used it for analyzing data (Budaev, 2010, Gonik et al., 2012, Vannoni et al., 2014, Ferguson et al., 2013, Ohl et al., 2003). Yet running a PCA on a multidimensional data set (a high amount of animals and parameters, i.e. over 1000 cases and over 100 parameters, respectively) is not frequently done. With a few exceptions animal numbers are moderate (below 100 animals) and parameters analyzed small (below 50). To answer the above mentioned questions we apply a PCA to a multidimensional data set as collected by the CW, which, to our knowledge, has not been done so far. By including a fairly large sample size (over 1000 cases) we expect to produce reliable results, as factor instability is less likely to occur in a larger sample.
In this paper we focus on locomotor phenotypes, whereby the definition of a ‘locomotor phenotype’ is used very broadly. We include gait phenotypes (as analyzed by the CW), as well as activity (as in the Open Field (OF), home cage (HC) and SHIRPA) and motor ability phenotypes (as in the Grip Strength (GS), Rotarod (RR) and Vertical Pole (VP) test) into this term. Gait phenotypes describe how the animal moves, activity phenotypes describe how frequently the animal moves, and motor ability phenotypes describe in this case muscle strength, balance and coordination. To test for a disease-relevant locomotor phenotype generally one to several different tests are applied. Here we assess if the CW alone, with its large amount of parameters, can be sufficient to detect locomotor deficits and if a reduction in dimensions is appropriate to detect meaningful components. With historical data from our lab we ran a PCA to reduce the dimensions of the CW data. Furthermore, we characterized the resulting components and analyzed them with different statistical methods to evaluate the impact of different versions of the CW system (we have used two different versions in our lab), sex, body weight, age and genetics.
Likewise we ask if a combination of tests measuring locomotor behavior is necessary for comprehensive locomotor phenotyping, or if there are redundancies between tests. To answer these questions we included two data sets. In the first set we took historical data from our lab including parameters from the CW, OF and VP test. The second data set comprises publicly available data from the International Mouse Phenotyping Consortium (IMPC; (Brown and Moore, 2012)) website (https://www.mousephenotype.org) selecting different measures of motor performance (horizontal as well as vertical activity, muscle strength and coordination), which included parameters measured in the OF, SHIRPA, RR, GS and in the HC during indirect calorimetry at the German Mouse Clinic. With both data sets a PCA was run to see how the different parameters are distributed among the extracted components, thereby assessing possible redundancies between different tests.
Section snippets
Behavior
For all tests mice are transferred to the testing room at a minimum of 30 min prior to testing. Mutants are always concurrently tested with respective wild types and males prior to females. After each mouse the test apparatus is cleaned and disinfected. Different mutant lines were used in the behavioral assessments. For sample sizes by sex and genotype please refer to Table S1 in the Supporting Information. All experiments were approved by the government of Upper Bavaria, Germany.
Characterization of the ten principal components revealed the possibility to focus on a reduced parameter-set in future analysis
We firstly ran a PCA only on the CW data set to reduce dimensions and to analyze the impact of sex, body weight, age and genotype. Historical data of 1499 cases was used for the PCA analyzing CW data. As there was a change of the CW version (from 7.1 to XT) in our institute we took data from both, analyzing only parameters that were measured by the two. After the omission of one parameter due to low correlations in the correlation matrix (“StepSequence_Rb”), we analyzed 112 parameters. The
Discussion
In our study we performed PCAs on different data sets related to locomotor phenotypes to refine future data analysis and to investigate possible redundancies between different tests. Knowledge of the latter better informs future decisions on phenotyping strategies: if tests were redundant, one could focus on the simplest one. Our results suggest that for future CW analysis a primary focus can be on FP and HP measures, and not all individual paws, as well as one of the interlimb coordination
Conclusion
In conclusion, we have shown that a large amount of data points generated by the CW system can be meaningfully compacted via PCA to a few components. Here we extracted 10 components explaining 80% of the total variance in the data. The loadings of parameters onto components suggest the possibility to focus on FP and HP parameters neglecting the parameters for every single paw, as well as analyzing only one measure of coordination, i.e. couplings or phase dispersion, as long as the experimental
Acknowledgments
This work was supported by the German Science Foundation Collaborative Research Centre (CRC) 870 to L.E., by funds from the Bayerisches Staatsministerium für Bildung und Kultus, Wissenschaft und Kunst within Bavarian Research Network “Human Induced Pluripotent Stem Cells” (ForIPS) to F.G., the German Federal Ministry of Education and Research (BMBF) through the Integrated Network MitoPD (Mitochondrial endophenotypes of Morbus Parkinson), under the auspices of the e:Med Programme (grant 031A430E
References (32)
- et al.
Sex bias in neuroscience and biomedical research
Neurosci. Biobehav. Rev.
(2011) - et al.
Detection of subtle neurological alterations by the Catwalk XT gait analysis system
J. Neuroeng. Rehabil.
(2014) - et al.
Sex differences in Parkinson’s disease
Front. Neuroendocrinol.
(2014) - et al.
Assessment of locomotor activity, acoustic and tactile startle, and prepulse inhibition of startle in inbred mouse strains and F1 hybrids: implications of genetic background for single gene and quantitative trait loci analyses
Neuroscience
(1997) - et al.
Neural control of sexually dimorphic behaviors
Curr. Opin. Neurobiol.
(2013) - et al.
Spontaneous behavior in the social homecage discriminates strains, lesions and mutations in mice
J. Neurosci. Methods
(2014) - et al.
The need for speed in rodent locomotion analyses
Anatomical Record (Hoboken, NJ: 2007)
(2014) - et al.
Open-field behavior of house mice selectively bred for high voluntary wheel-running
Behav. Genet.
(2001) - et al.
Tests to assess motor phenotype in mice: a user’s guide
Nat. Rev. Neurosci.
(2009) - et al.
Towards an encyclopaedia of mammalian gene function: the International Mouse Phenotyping Consortium
Dis. Models Mech.
(2012)
Using principal components and factor analysis in animal behaviour research: caveats and guidelines
Ethology
Sex modifies the relationship between age and gait: a population-based study of older adults
J. Gerontol. Ser. A: Biol. Sci. Med. Sci.
Principal component analysis of the effects of environmental enrichment and (−)-epigallocatechin-3-gallate on age-associated learning deficits in a mouse model of down syndrome
Front. Behav. Neurosci.
Behavioral phenotypes of inbred mouse strains: implications and recommendations for molecular studies
Psychopharmacology
Analysis of mammalian gene function through broad based phenotypic screens across a consortium of mouse clinics
Nat. Genet.
Open-field behavior in mice: selection response and situational generality
Behav. Genet.
Cited by (10)
Systematic data analysis and data mining in CatWalk gait analysis by heat mapping exemplified in rodent models for neurodegenerative diseases
2019, Journal of Neuroscience MethodsCitation Excerpt :Heat maps have been applied in biology (Weinstein, 2008), behavioral science (Anderson and Perona, 2014; Kas et al., 2011), social statistic (Loua, 1873), accounting, graphic design, computer engineering, and others (Wilkinson and Friendly, 2009). In the field of gait analysis, heat maps have been applied to describe principal components resulting from principal component analysis (PCA) algorithm (Courtine et al., 2009; Heikkinen et al., 2015a, 2015b; Mondello et al., 2015; Nurmi et al., 2015; Sweeney et al., 2015; Zainana et al., 2015; Zimprich et al., 2018) and to report correlation coefficients (Machado et al., 2015). The systematic IDA for CatWalk gait parameters presented here provides a comprehensive map of between-group differences, and is especially helpful for exploratory data mining.
Preface: Special issue on measuring behaviour 2016
2018, Journal of Neuroscience MethodsMultidimensional analysis of behavior predicts genotype with high accuracy in a mouse model of Angelman syndrome
2022, Translational PsychiatryParkinson's disease motor symptoms rescue by CRISPRa-reprogramming astrocytes into GABAergic neurons
2022, EMBO Molecular MedicineA bioisostere of Dimebon/Latrepirdine delays the onset and slows the progression of pathology in FUS transgenic mice
2021, CNS Neuroscience and Therapeutics