Mental health disorders, such as anxiety and depression, constitute one of the main causes of disease burden worldwide (Vos et al., 2015), and their prevalence in the United Kingdom is growing (Martín-Merino et al., 2009; Fineberg et al., 2013).

Behavioral disruption related to environmental or genetic changes are commonly evaluated through the use of animal models (Steimer, 2011; Rossignol and Frye, 2012), although often the methods and tests used are suboptimal, leading to mixed results and findings that may not translate well (Perel et al., 2007; Open Science Collaboration., 2015). Studies may use a single behavioral probe to represent complex behavioral traits, whereas behavioral outcomes are a culmination of a multifaceted system, are frequently subtle, and have aspects which can present in different ways (Ferreri et al., 2011). This could lead to subtle behavioral changes being missed, or anomalous data being given undue prominence. A targeted battery of behavioral tests can give insight to a greater number of behavioral traits and therefore give a more accurate representation of the specific behaviors being studied. However, a consequence of multiple testing is an increase in the probability of generating type I statistical errors and for this reason many studies restrict the number of tests performed (Shaffer, 1995).

The current study presents a novel method of utilizing a broader battery of tests to produce a simple score to represent each behavioral trait under investigation. Such unified rating scales are well established in the clinical assessment of symptom progression for diseases such as Parkinson’s or Huntington’s (Martínez−Martín et al., 1994; Huntington Study Group., 1996), whereby patients undergo a battery of assessment measures that are scored and classified into functional groups, such as cognition and motor function. These scores are used as a scale against which different aspects of disease progression or intervention can be monitored. Changes in mental health status may be incremental and not necessarily have statistical significance for any individual test measure, but together may lead to a general improvement in one or more behavioral traits or in general wellbeing. Here, a unified scoring model is applied to detect differences in anxiety- and sociability-related behaviors between two common mouse strains, with the aim of establishing a simplified, statistically sound method of comparing complex behaviors in preclinical models.

Despite changes in behavior being the primary focus of many studies of mood disorders, the background strain used is often determined by factors other than behavioral phenotype. Reasons for strain selection are not often discussed in published behavioral methods but understandably may be attributed to availability, ease of breeding, handling or genetic manipulation, desirable in vitro qualities, or simply being a vestige of previous studies probing non-behavioral phenotypes (Crawley et al., 1997). Many studies have shown significant strain differences in cognitive, social, emotional- and psychological-like behavioral characteristics (Liu and Gershenfeld, 2001; Võikar et al., 2001; Abramov et al., 2008; Johnson et al., 2010; Tuttle et al., 2017), highlighting the importance of using appropriate strains for the phenotypes behavioral studies are seeking to model. The prospects of detecting subtle changes in behavior, for example epigenetic manipulations or environmental influences on psychological-like traits, are decreased when experimental groups are compared to controls with a lower phenotypic baseline. For instance, where an increase of anxiety-like behaviors is hypothesized, manifestation is more likely to be detected in a non-anxious model, and/or populations which exhibit minimal variation in the targeted phenotype. Conversely, if a strain is inherently stressed, ceiling effects could mask detection of subtle increases in stress levels. Implementing the proposed scoring method can identify traits which support the selection of one strain over another for studies probing these behaviors.

In addition to strain differences, the present study utilizes the unified scores to compare behavioral traits of female and male mice. It is common for preclinical studies to use only male animals, despite many clinically diagnosed behavioral and emotional disturbances manifesting disparately between the sexes (Blanchard et al., 1995). Prevalence and scale of anxiety disorders and major depressive disorders, for example, is greater in females compared to males, whereas signs of autism spectrum disorder and attention deficit hyperactivity disorder are more prevalent in males (Romans et al., 2007; McLean et al., 2011; Willcutt, 2012; Loomes et al., 2017). This sexual dimorphism of behavioral disorder symptomatology underlines the necessity to include both sexes in any preclinical studies, indeed the National Institutes of Health policies stipulate that all preclinical work should include both sexes or a valid explanation as to why not.

The current study introduces a novel data-inclusive analysis of a comprehensive anxiety-related and social behavioral battery designed to produce a robust method of measuring subtle behavioral traits of two widely used background strains in behavioral research, the C57BL6/J and 129S2/SvHsd mouse strains. These particular behavioral traits and mice were selected to assess the validity of the unified scoring method presented here as there is a rich history of comparative studies against which the results can be validated (Hughes, 1989; Crawley et al., 1997; Montkowski et al., 1997; Belzung and Griebel, 2001; Liu and Gershenfeld, 2001; Võikar et al., 2001; Touma et al., 2004; Sankoorikal et al., 2006; Abramov et al., 2008; Yang et al., 2011; Lorenz et al., 2013). The objective is to provide a simple output for a complex system, which minimizes the risk of type I and type II statistical errors and increases reproducibility in preclinical behavioral neuroscience. In addition, this strain and sex comparison will provide a guide to strain and/or behavioral test selection aimed to help customize future experimental design to maximize useful output and consequently reduce animal use.

Male and female C57BL6/J and 129S2/SvHsd mice were tested on a battery of anxiety- and sociability-related tasks and scored on cohort-normalized scales between 0 and 1, whereby a score of 0 represents the lowest measured anxiety/sociability signs and a score of 1 represents the highest.

Principal component analyses of outcome measures for each multi-outcome test were used to validate their assignment as contributing either positively (higher score, greater anxiety/sociability) or negatively (higher score, low anxiety/sociability) to the behavioral trait scores being probed, Figure 2. For the elevated zero maze, PC1 described the ‘latency’ measure as contributing contrariwise to the variance compared to the ‘number of crosses’ and time spent in the open’ measures (+ve, −ve, and −ve respectively, Figure 2A), i.e., a large score in the latency measure would indicate a higher anxiety score, whereas a large score in the other measures would indicate a lower anxiety score. PC1 was also able to demonstrate the opposing contributions of the ‘latency’ measure and ‘number of crosses’ and ‘time to cross’ measures of the light/dark box test to the anxiety score, Figure 2B. The sociability outcome measures of the 3-chamber test were grouped by PC1 into ‘time spent in the empty chamber,’ ‘crosses into the empty chamber’ and ‘latency to enter the occupied chamber’ versus ‘crosses to,’ ‘time spent’ and ‘time per visit’ in the occupied chamber, Figure 2C. For the direct social interaction test, PC1 did not reflect the directional influence of sociability-related factors, indicating that this represents an extraneous factor to sociability, Figure 2D, with an environmental exploration component providing a more feasible explanation. Since it did not appear to be relevant to measures of sociability, the ‘exploration’ measure was removed and the PCA repeated, Figure 2E. PC2 reflected the directional influence of sociability-related factors, with the time spent interacting with the host (‘sniffing,’ ‘following,’ ‘being followed,’ ‘attacking’ and ‘being attacked’) (socially interactive factors) opposing the time spent self-grooming or immobile (socially anxious factors). For the social propinquity test, PC1 was descriptive of the directional influence of sociability-related factors, with the duration the tube was vacant and the latency to share the tube (socially anxious factors) opposing the time spent sharing the tube (socially interactive factor). The allocation of outcome measures to positive or negative score contributions is summarized in the ‘Influence’ column of Table 1.

129 mice demonstrated a higher anxiety score compared to the BL6 mice, with no difference between males (STRAIN: F₁,₆₉ = 6.91, p = 0.011, Figure 3A). 129 mice also exhibited a greater level of social interaction compared to the BL6 mice (STRAIN: F₁,₆₉ = 10.90, p = 0.002, Figure 3B). No difference in the integrated anxiety or sociability scores between sexes was identified [SEX(Anxiety): F₁,₆₉ = 1.71, p = 0.196 and SEX(Sociability): F₁,₆₉ = 0.30, p = 0.583 respectively].

PCA of individual animal scores based on all behavioral tests conducted demonstrated significant clustering of mice in PC1 into strain groups, although separate clustering of sexes was not detected (STRAIN: F₁,₆₉ = 28.45, p < 0.001 and SEX: F₁,₆₉ = 0.09, p = 0.766 respectively, Figures 3C,D).

Behavioral traits in animal models are often subtle and complex, and detecting replicable disturbances can be challenging. The multifaceted nature of any behavior necessitates the use of multiple tests to obtain a truly reflective assessment (Belzung and Griebel, 2001). Anxiety-related behaviors, for example, are known to arise from several different neuronal systems (Ferreri et al., 2011), may present in slightly different ways or intensities (Montkowski et al., 1997) and certain behavioral tests may not be sensitive enough to detect subtle changes. The data presented here, from three different tests of anxiety and stress (the elevated zero maze, light/dark box and FCM analysis), show how disparate outcomes can be generated depending on the test used. Many studies will often probe behaviors using just one test. However, as we show here, one test may not be adequate to detect true effects. A major problem with conducting multiple tests is that the probability of generating type I errors increases, and while this can be corrected for by using more rigorous statistical thresholds, it is not ideal (Shaffer, 1995). The current study presents a novel method for utilizing a broad range of behavioral probes and consolidating all outcome measures to provide a simple and comparable score for each type of behavioral trait being investigated. The combined score incorporates equal input from each test performed whilst minimizing the probability of reporting chance results as significant effects or missing subtle behavioral disturbances.

Principal component analysis of the test outcome measures was able to effectively categorize their contribution to the traits being probed, thus justifying the positive/negative allocation of behaviors to each test score. The resulting unified scores were able to detect a difference in the degree of anxiety-related behavior that female 129s exhibit compared to female BL6 mice, a result which was reflected significantly in some but not all behavioral tests or outcome measures. This is congruent with the conclusion of several previous strain comparison studies, albeit with different tests or test combinations (Rogers et al., 1999; Võikar et al., 2001; Abramov et al., 2008; Harms et al., 2008). Interestingly, contrary to these previous reports, the unified score for anxiety did not detect a difference between the male mice. In fact, it was only the light/dark box test ‘time in light’ measure that demonstrated a significant increase in anxiety-like behavior in the male 129s – a test that is frequently used as the sole measure for reporting this behavior. The data suggests that anxiety-related behaviors in the males may be affected by a more restricted range of aspects of anxiety, whereas the female mice may present a more generalized anxious state. This result may reflect the dimorphic manifestation of anxiety-related symptoms in people, where women are twice as likely to be diagnosed with anxiety related conditions than men (Martín-Merino et al., 2009). Female mice demonstrated a higher baseline concentration of corticosterone metabolites in their feces compared to the males. Whilst high levels of this stress hormone may suggest female mice are generally more stressed than the males, it was observed that following exposure to predator odor all groups showed a similar proportional increase in FCMs. Other than ‘amount’ of stress per se, sex differences in the FCM baseline measures could be explained by differences in metabolism or excretion (Touma et al., 2003, 2004), or due to disparities in the proportion of metabolite types being produced, since only one marker was used in the current study.

The unified scores for sociability detected a greater tendency for interaction in both sexes of the 129s compared to the BL6 mice, a result which was less clearly discerned from the individual tests. This trait is consistent with direct social interaction findings presented by Hughes (1989) and Harms et al. (2008) who observed that 129/Sv mice spent more time engaged in social interaction than other strains. Importantly, PCA of the individual test scores revealed significant clustering in PC1 of the experimental groups, indicating that the effect of ‘group’ may be the primary factor affecting the greatest variance within the data set and thus supporting the effectiveness of the normalized unified scoring system.

The data presented demonstrate a proof of concept for the use of unified scores to present a clear behavioral phenotype using data from multiple tests – the results of which are validated by a broad range of previously published studies probing aspects of these behaviors in these strains (Hughes, 1989; Crawley et al., 1997; Montkowski et al., 1997; Belzung and Griebel, 2001; Liu and Gershenfeld, 2001; Võikar et al., 2001; Touma et al., 2004; Sankoorikal et al., 2006; Abramov et al., 2008; Yang et al., 2011; Lorenz et al., 2013), however, further validation of the sensitivity of the unified scoring method through pharmacological studies would be beneficial. These studies could also be used to determine the optimum number of different tests to include within a test battery for each particular behavioral trait. The current study used at least 3 different tests to probe each behavioral trait, which resulted in a clear result backed up by previous studies, however, it would be useful to investigate when the benefits of including more tests to increase the integrity of the data is outweighed by the amount of work and logistical practicalities involved.

The underlying method presented here for anxiety-related and social-related tests has potential to be applied to other aspects of behavior by amalgamating related output measures probing behavior outcomes such as depression, cognition, or motor deficiencies. For example, combining measures of the tail suspension test, forced swim test and hedonistic response could provide a more accurate indication of the depressive-like state of an individual than either one on its own, and could therefore help to improve the reporting of behavioral results and subsequently reproducibility, for which there is growing call (Landis et al., 2012). Furthermore, since all data is normalized to a score between 0 and 1 based on comparisons within a dataset, it allows the unified scores to be compared between experiments and labs by reducing the influence of environmental/temporal variations on basic outcome measures. Importantly, the lack of sex effects between the female and male scores demonstrate a counterargument to the common presumption that female mice are less suited to behavioral testing. The normalization of data into unified comparative scores could further support the use of both sexes for investigation of sexual dimorphic patterns of behavior.

The unified behavioral score can incorporate a broad range of behavioral probes and present a simple comparable score reflective of generalized behavioral traits. The method is sensitive enough to detect subtle differences in complex behaviors that may be easily missed by individual tests. This novel approach aimed at presenting and comparing behavioral traits will be a useful tool for exploring the effects of genetic manipulations, disease, adversity or interventions in animal models of psychiatric disorders.

The raw, transformed and unified score datasets for this study can be found in the FigShare Repository,

The animal study was reviewed and approved by the local Cardiff University Ethics Review Committee.

DH, RJ, and AI contributed to the concept and design of the study. DH, HC, HT, and RB-S conducted the behavioral tests. RP and CT corticosterone metabolite analyses. DH wrote the draft manuscript. All authors contributed to the manuscript revision, read and approved the submitted version. RJ acquired the funding.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.