Three murine anxiety models: results from multiple inbred strain comparisons


  • L. C. Milner,

    Corresponding author
    1. Department of Behavioral Neuroscience, Portland Alcohol Research Center, Oregon Health & Science University and VA Medical Center, Portland, OR, USA
    Search for more papers by this author
  • J. C. Crabbe

    1. Department of Behavioral Neuroscience, Portland Alcohol Research Center, Oregon Health & Science University and VA Medical Center, Portland, OR, USA
    Search for more papers by this author

*Lauren C. Milner, Department of Behavioral Neuroscience, Oregon Health & Science University, Mail Code L470, Portland, OR 97239, USA. E-mail:


The literature surrounding rodent models of human anxiety disorders is discrepant concerning which models reflect anxiety-like behavior distinct from general activity and whether different models are measuring the same underlying constructs. This experiment compared the responses of 15 inbred mouse strains (129S1/SvlmJ, A/J, AKR/J, BALB/cByJ, C3H/HeJ, C57BL/6J, C57L/J, CBA/J, CE/J, DBA/2J, FVB/NJ, NZB/B1NJ, PL/J, SJL/J and SWR/J) in three anxiety-like behavioral tasks (light/dark test, elevated zero-maze and open field) to examine whether responses were phenotypically and/or genetically correlated across tasks. Significant strain differences were found for all variables examined. Principal components analyses showed that variables associated with both activity and anxiety-like behaviors loaded onto one factor, while urination and defecation loaded onto another factor. Our findings differ from previous research by suggesting that general activity and anxiety-related behaviors are linked, negatively correlated and cannot easily be dissociated in these assays. However, these findings may not necessarily generalize to other unconditioned anxiety-like behavioral tests.

Over the past 70 years, the use of rodent models has become an invaluable tool for anxiety disorder research. Many applications for these models exist, the most widespread being the examination of pharmacotherapeutic agents for anxiety disorders. Observing behavior in these tasks also allows researchers to investigate potential genetic, neurological and neurochemical contributions that underlie different aspects of anxiety (Henderson et al. 2004; Treit & Menard 1997; Turri et al. 2001).

Rodent assays of anxiety-like behavior can be subdivided into two categories: the first examines conditioned anxiety-like behavior using punished responding (Pollard & Howard 1979), while the second exploits rodents’ natural aversion to bright and/or exposed areas by measuring how subjects partition their behavior in these spaces (Hall 1934; Ramos & Mormède 1998). Tests that fall into the second category include the open field (OF), elevated plus-maze (EPM) and the light/dark (LD) test. These tests require no conditioning, do not cause physical discomfort and are considered to have ecological face validity (Rodgers 1997).

Despite the progress in preclinical anxiety research, certain questions remain unanswered. Some anxiety-related behavioral tests appear to be highly variable (Carobrez & Bertoglio 2005; Hogg 1996). Researchers have also found that anxiety-like behavioral measurements are not always highly correlated across tasks, suggesting that different anxiety-like behavioral tests do not necessarily address the same underlying construct. Griebel et al. (2000) tested the behavior of various mouse strains in the LD and EPM tests and found that basal anxiety-related behavior in the LD test was not strongly associated with basal anxiety-like behavior in the EPM and that diazepam’s anxiolytic effects in one assay were not necessarily observed in the other. These discrepancies can be explained by the idea that different tests reflect different types of anxiety (Belzung & Griebel 2001). However, other differences such as sensitivity to light (DeFries et al. 1966) or locomotor behavior (Henderson 1986) could be responsible.

Even if a single underlying ‘anxiety’ condition exists in rodents, inbred strain variability in nonanxious behaviors (e.g. locomotor activity) may confound behavioral interpretations. Most studies examining anxiety-like behavior have been limited to a few inbred strains (Avgustinovich et al. 2000; Võikar et al. 2001), although some have examined more (Griebel et al. 2000; Trullas & Skolnick 1993). The use of a larger number of genotypes helps to account for some of these confounding behaviors by increasing the likelihood that traits are widely distributed across the strains studied, reducing the likelihood of seeing spurious associations and aiding the interpretation of anxiety-like behavior.

The current experiments tested a substantial number of inbred mouse strains for behaviors in the LD test, elevated zero-maze (EZM) and OF. By correlating strain mean responses for anxiety-like behavior measures, we explored commonalities of genetic contributions across these tests to see whether they suggest a single anxiety-related behavioral state. We also examined whether genetic influences on anxiety-like behavior could be distinguished from those on locomotor activity. Finally, using principal components analyses, we asked whether activity and anxious-like behavior indices could be effectively disambiguated across tests.

Materials and methods

Animals and husbandry

Four male and four female mice (= 120) from 15 inbred strains (129S1/SvlmJ, A/J, AKR/J, BALB/cByJ, C3H/HeJ, C57BL/6J, C57L/J, CBA/J, CE/J, DBA/2J, FVB/NJ, NZB/B1NJ, PL/J, SJL/J and SWR/J) were purchased from The Jackson Laboratory (Bar Harbor, ME, USA) at 6 weeks of age and were allowed to adapt to housing conditions for 2 weeks prior to start of testing. These strains were selected because of their correspondence with those studied by Trullas and Skolnick (1993), their priority listing on the Mouse Phenome Database ( and because we have extensive data for a number of behavioral responses for many of these strains. Animals were housed at the Portland VA Medical Center in the Veterinary Medical Unit. All subjects were housed in the same room and on the same rack, ordered from top to bottom by their experiment run order (pseudorandomly designed), with two mice per shoebox-sized standard plastic cage (28 × 17 × 11.5 cm). Our animal facility uses cages of either polysulfone or polycarbonate indiscriminately. Food (Purina 5001) and water were available ad libitum. Cages were lined with Bedicob® bedding and changed twice weekly. Both the housing and testing room temperatures were maintained at 22 ± 1°C, and all animals were kept on a 12:12-h light:dark schedule with lights on at 0600 h. All housing and testing were performed according to the US Department of Agriculture and the US Public Health Service guidelines and was approved by the Institutional Animal Care and Use Committee.


Two copies of each apparatus were available for testing, so two subjects were run simultaneously in each task. However, the apparatuses were positioned so that the mice could not see each other.

LD task

The apparatus used for this study was constructed according to Crawley and Goodwin (1980) and comprises two connected compartments. The light compartment is a 28 × 28 × 30 cm clear acrylic box that shares one wall with the dark compartment, which is 28 × 17 × 30 cm, composed of dark acrylic and has a lid to ensure that little light enters. An opening in the shared wall allows subjects to move between the two compartments.

Elevated zero-maze

The EZM was chosen over the more traditionally used EPM because of the lack of a center platform in the EZM. The center area in an EPM can be a complicating factor when interpreting activity and anxiety-like responses (Cook et al. 2001; Shepherd et al. 1994). The EZM was constructed by Flair Plastics (Portland, OR, USA) and comprises four equally proportioned arms forming a circle with an external diameter of 45 cm. The floor of the alleys is 5.5-cm wide and composed of black acrylic. The walls of the two ‘closed’ arms are on opposite segments of the maze and are made of clear acrylic 11 cm in height. The ‘open’ arms are also on opposite segments of the maze and have a lip, composed of clear acrylic and 3 mm in height, to prevent the subjects from falling from the apparatus, which is elevated 45 cm above the floor.

Open field

The round OF apparatus used in this study was constructed by Lehigh Valley (Bethlehem, PA, USA) and is 61 cm in diameter with walls that are 26 cm in height. The floor and walls of the apparatus are made of stainless steel.


In an attempt to limit circadian effects, all testing was carried out between 0700 and 1300 h. However, because of this temporal restriction and the duration of each exposure, we tested our subjects in four groups over the course of 4 days (Friday, Saturday, Sunday and Monday) for each apparatus. Although strain differences may be affected by prior testing (Crabbe et al. in press; McIlwain et al. 2001), cost and practical considerations required us to test all mice in all three tasks. We elected to test mice first in the LD test and then in the EZM and the OF, with 1 week between each test. Similar intervals have been used by others and us for testing multiple responses in surveys (Crawley & Paylor 1997; Rhodes et al. 2007; Wall et al. 2004). The experiment was designed so that one cage of mice from each strain would be tested on each day. Because the animal facility is considerably less busy on weekends, we also wanted each sex to be tested both on weekends and on weekdays in order to offset the influence of any environmental factors that might differ between the two. Thus, half the males of each strain were tested on Fridays or Sundays, and half the females were tested on Saturdays or Mondays. The opposite sex/day pairing was used for the other half of the subjects, and once randomly assigned, test day was held constant for all three tests. Each of the strains was randomly assigned to a test order for each day-group, and that order was also held constant for all three tasks. For all tests, the animals were moved to the test room and allowed to adapt to the testing room conditions for 1 h before testing started, and the apparatus was wiped with 10% isopropyl alcohol between each animal. Each subject’s activity in all tests was captured separately by a video camera (Sony Handycam NP-F330) mounted 78.7 cm (LD and EZM) or 155 cm (OF) above the apparatus. Videotape records were subsequently scored from a monitor through a transparency on which areas of the apparatus were demarcated by lines as described later. Scoring from these tapes began as soon as the subject was placed in the apparatus. Line crossings were counted when all four paws crossed. A nose poke across a line was counted when the whole head crossed the line. A rear was counted when the animal lifted both forepaws from the floor, with or without touching the wall. All videotaped data are available upon request.

LD test

Lighting conditions for this task were set at 250 lux. At the beginning of each run, the mouse was taken from its home cage, placed next to the wall in the light compartment furthest from the dark compartment and the experimenter moved out of the subject’s line of vision. The mouse was allowed to explore this apparatus for 10 min. For scoring, a transparency was overlaid on the light compartment (the only one visible) image on the video monitor that divided this compartment into four equal sections. Variables scored included number of boli and urinations in both light and dark compartments, number of transitions between light and dark compartments, line crosses in the light compartment, number of rears in the light compartment, number of nose pokes from dark side of compartment into light side of compartment, time spent in light compartment and latency to leave the light compartment. Additionally, voluntary reentry time (total time spent in light side of compartment minus initial latency to leave light compartment; Kliethermes et al. 2003) reflects the amount of time spent in the light compartment following initial entry into the dark compartment. This variable was created to address the potential confound between total time in the light compartment and latency to leave the light compartment as the latter may actually reflect anxiety-like behaviors (i.e. ‘freezing’ behavior).

Elevated zero-maze

For this task, lighting conditions were set at 15 lux. At the beginning of each run, the subject was removed from its home cage and placed in a closed arm facing an open arm, the experimenter moved out of sight and the mouse was allowed to explore the apparatus for 5 min. For scoring, the transparency had 12 lines, effectively splitting the total circular alley into 12 equal segments. Variables included number of fecal boli/urinations, time spent in open arms, number of line crosses in the open arms, number of line crosses in the closed arms, number of rears, latency to enter the open arm, number of stretched-attend postures, a common measure of risk assessment that consists of a subject extending its torso into an open or novel area and then immediately retracting its torso (Weiss et al. 1998), and head dips over the sides of the open arms. The four lines at the interfaces between open and closed arms were therefore used for assessments of stretched-attend postures but were not counted in the open or closed arm number of crosses. In order to remain consistent with previous literature (for review, see Hogg 1996), we have reported time spent in open arms as a percent of the total time in the apparatus. We have also reported percent of line crosses in the open arms (out of a total number of line crosses in the apparatus) because ‘percent of open arm entrances’, an anxiety-like behavior measure commonly reported in EPM literature, is not interpretable in EZM data.

Open field

For this task, lighting conditions were set at 350 lux. At the start of the run, each subject was removed directly from its home cage, placed in the center of the OF and allowed to explore the apparatus for 15 min while the experimenter was hidden. For scoring, the transparency comprised a large circle with radial lines drawn to divide the outer track of the entire OF into 16 equal segments. A smaller circle (capturing the center circle, diameter = 45.7 cm) inside the larger circle had four radial lines that divided the center of the OF into four equal segments, each of which was 4× as large as the size of an outer segment. Variables scored were fecal boli/urinations in apparatus, total time in center of apparatus, number of line crosses in the periphery of the OF, number of line crosses from the periphery to the center of the OF, number of line crosses in the center of the OF and number of rears in apparatus. Rearings in the center and periphery of the OF were counted separately, but we did not note whether mice were touching the wall (‘leaning’) during a rear in the periphery. However, we recognize the potential significance of this measurement in the OF test (Clément et al. 1995), and all videotapes of this task are available to anyone interested in exploring this issue.

Statistical analyses

Dependent variables were analyzed using Systat® with analyses of variance (anovas). Main effects and interactions from anovas were followed up using Tukey’s post hoc analyses, and the threshold for significance was set at < 0.01. Estimates of genetic effect size are taken from adjusted r2 values (sum squares between subjects/sum squares total) automatically calculated when analyzing one-way anovas by strain. Occasional outliers for certain variables were defined as being greater or less than three standard deviations from the mean for the group and were removed from all analyses. Additionally, the percent open-arm line crosses in the EZM and percent line crosses in the center of the OF were included in the anova analyses but were removed from the correlational and principal components analyses (PCA) as they are computed from other variables and are therefore redundant.

Pearson’s r was used for correlational analyses. Strain mean correlations were estimated from the correlation of strain mean values for a pair of variables. As discussed in detail elsewhere (Crabbe et al. 2005), we were more interested in the pattern of strain mean correlations than in the statistical significance of each given correlation. Because the low n (15) for strain mean correlations limits power to detect associations, we occasionally discuss nonsignificant correlations. All correlations were analyzed by examining scatter plots to detect outlier strains, which can easily distort underlying patterns of genetic association.

We also computed phenotypic correlations across the entire data set, ignoring strain because each strain was represented by approximately the same number of mice (eight). The phenotypic covariance matrix was used for PCA for individual tasks using a varimax rotation, from which components with eigenvalues >1 are reported. A separate PCA analysis performed across all tasks reported numerous factors (eigenvalues >1) but showed no pattern informative for the current study. A scree plot analysis showed a natural break at four factors, with factors having eigenvalues >2 contributing most of the total variance to this model. A PCA analysis was then run fitting the data to four factors and these values are reported in Table 2.

Table 2.  PCA for individual and combined anxiety-like behavior tasks
TaskVariableFactor 1Factor 2Factor 3Factor 4
  • a

    Reported eigenvalues >1.

  • b

    Reported eigenvalues >2.

  • Italics indicate variable loading equally onto two factors.

LDaBoli −0.72 
Urinations −0.86 
Line crosses0.94 
Nose pokes −0.76 
Total time in light 0.77 
Latency to leave light 0.58 
Voluntary reentry time0.75 
EZM*Boli −0.79 
Urinations −0.85 
% Time in open arms0.91 
Open-arm line crosses0.90 
Closed-arm line crosses0.84 
Head dips0.76 
Latency to enter open arms−0.76 
Stretched attends−0.40 
OF*Boli 0.76 
Urinations 0.81 
Time in center0.76 
Line crosses in periphery0.68 
Line crosses from periphery to center0.91 
Line crosses in center0.92 
Rears in periphery0.71 
Rears in center0.60 
All TasksbLD boli 0.56 
LD urinations 0.68 
LD transitions0.58 0.54
LD line crosses0.66 
LD rears0.67 
LD nose pokes −0.70 
LD total time in light 0.85 
LD latency to leave light 0.48 
LD voluntary reentry time0.61 
EZM boli 0.64 
EZM urinations 0.66 
EZM % time in open arms0.87 
EZM open-arm line crosses0.83 
EZM closed-arm line crosses0.77 
EZM rears0.60 
EZM head dips0.70 
EZM latency to enter open arms−0.74 
EZM stretched attends−0.41 
OF boli 0.71 
OF urinations 0.54 
OF time in center 0.86
OF line crosses in periphery0.49 0.46
OF line crosses from periphery to center 0.77
OF line crosses in center 0.88
OF rears in periphery0.67 
OF rears in center 0.58

Principal components analyses were also performed on strain mean values for each variable in all tasks. For this analysis (eigenvalue >1), the scree plot showed a natural break after four factors, so the PCA analysis was repeated by fitting the analysis to four factors. The results are reported in Table S2.

Although the number of ‘subjects’ (strain mean) in this analysis was too low compared with the number of variables to support PCA on the strain mean covariance matrix, we ignored this requirement to see whether the genetic components appeared to resemble the phenotypically derived components.


Data were collected from all mice within each strain (= 8) for both the LD test and the EZM. However, one AKR/J female mouse died before being tested in the OF.

Each variable was initially analyzed using a three-way anova in order to determine potential sex and day effects, in addition to strain differences, with the goal of simplifying the anovas as much as possible. Day effects were analyzed by combining data from weekends and comparing them with data from weekdays. A main effect of sex was noted for number of boli in the LD test (F1,57 = 11.1; < 0.01), with the males depositing more boli than the females. A main effect of day (F1,60 = 10.7; < 0.01) and sex × day interaction (F1,60 = 8.3; < 0.01) were noted for number of urinations in the EZM, with urine counts higher for the weekend than the weekday groups and the male weekend group urinating more than any other group. The number of stretched attends in the EZM also showed a main effect of day (F1,58 = 8.0; < 0.01), with the weekend group showing more stretched attends than the weekday group. Last, a strain × sex interaction (F14,59 = 17.1; < 0.01) was noted for number of boli counted in the OF, with DBA/2J females depositing more boli than DBA/2J males and CE/J females depositing fewer boli than that of CE/J males. Given that these were the only significant sex and day effects/interactions observed in a substantial number of analyses, we report data collapsed across day and sex. However, it is important to note that our small sample size for each same-sex, same-strain group (= 4) likely provided inadequate power to detect significant sex differences (or, especially, strain × sex interaction). Future studies with increased numbers may show sex differences not detected in the current study. All original data will be posted to the Mouse Phenome Database and are available from the authors upon request (the Mouse Phenome Database will post data grouped by sex).

LD test

Significant strain differences were found for all variables examined, with genetic effect sizes ranging from 0.26 to 0.73 (Table 1). Results are shown in Figure S1. Post hoc analyses performed following anovas for each variable showed the following: A/J, SJL/J and BALB/cByJ strains were high for number of boli produced, while the FVB/NJ, CE/J, C57BL/6J and C57L/J strains were the lowest for this variable. This strain pattern differed somewhat from that of urinations produced, in which the AKR/J, BALB/cByJ, PL/J and SJL/J strains were high and the C57BL/6J, A/J and FVB/NJ strains were low. For number of transitions between the light and dark compartments, the FVB/NJ, C57L/J and CE/J strains were high and A/J and BALB/cByJ strains were low. The FVB/NJ, PL/J, CE/J, SWR/J and C57L/J strains made the most line crosses in the light compartment, while the AKR/J, BALB/cByJ, DBA/2J and A/J strains made the least. This pattern was similar for number of rears in the light compartment, except that the PL/J strain did not score high on the rearing variable. The C57L/J, 129S1/SvlmJ and BALB/cByJ strains showed the highest number of nose pokes, while the A/J strain was the lowest for this variable. For total time spent in the light compartment, the A/J strain was higher than all other strains, but the A/J strain also showed a much longer latency to leave the light compartment than all other strains. The voluntary reentry variable accounts for this potential confound (see Materials and methods for details), and using this measure, the A/J and BALB/cByJ strains were among the lowest, while the C3H/HeJ, PL/J, SWR/J and FVB/NJ were among the highest strains for time spent in the light compartment following initial latency to leave the light compartment.

Table 1.  Results from anovas for main effects of strain in three tests of anxiety-like behavior*
TestVariableAbbreviationsF valueGenetic effect size
  1. *All F with 13–14 and 55–60 degrees of freedom. For ≥ 3.8, ≤ 0.001; for ≥ 2.4, ≤ 0.01.

Line crosses in lightLDlc6.80.49
Nose pokesLDnp5.70.44
Total time in lightLDttl11.60.62
Latency to leave lightLDlll7.60.50
Voluntary reentry timeLDvr19.40.73
% Time in open armsEZMptto9.70.57
Open-arm line crossesEZMoalc9.00.55
Closed-arm line crossesEZMcalc8.70.54
% Open-arm line crossesEZMpoalc7.60.51
Head dipsEZMhd8.70.54
Latency to enter open armsEZMleo6.20.45
Stretched attendsEZMsa3.80.34
Time in centerOFtct11.50.62
Line crosses in peripheryOFlcp15.60.69
Line crosses from periphery to centerOFlcpc27.00.77
Line crosses in centerOFlcc22.60.75
% Line crosses in centerOFplcc10.10.58
Rears in peripheryOFrp39.60.85
Rears in centerOFrc4.60.39

Elevated zero-maze

Significant strain differences were found for all variables studied, with genetic effect sizes ranging from 0.30 to 0.65 (Table 1). Results are shown in Figure S2. Number of boli produced in this task showed that the SJL/J, BALB/cByJ, CBA/J and DBA/2J strains produced the most fecal boli, while the FVB/NJ and CE/J strains did not produce any boli. This pattern was different for number of urinations in that the BALB/cByJ, SJL/J and NZB/B1NJ strains exhibited a high number of urinations, while the A/J strain exhibited a low number and the FVB/NJ, C57BL/6J and C3H/HeJ strains did not urinate during this task. Percent time spent in the open arms showed the SWR/J, CE/J and 129S1/SvlmJ strains as exhibiting the highest percentage and the A/J and BALB/cByJ strains exhibited the lowest percentage. The SWR/J and CE/J strains were high for open-arm line crosses, while BALB/cByJ strain was low for this variable, and the A/J strain showed no open-arm line crosses. Closed-arm line crosses showed a similar pattern to open-arm line crosses, although the CE/J, C57L/J and FVB/NJ strains were all high for this variable. Percent open-arm line crosses (out of total line crosses) showed a similar pattern to open-arm line crosses, with the exception of the 129S1/SvlmJ strain, which was high. The FVB/NJ, C57L/J, SWR/J and CE/J strains exhibited more rears than other strains tested. For both number of stretched attends and latency to enter the open-arm variables, the BALB/cByJ and A/J strains were the highest, while the SWR/J and FVB/NJ strains were among the lowest. Last, the SWR/J strain exhibited substantially more head dips than all other strains tested.

Open field

Significant strain differences were found for all variables, with genetic effect sizes ranging from 0.35 to 0.85 (Table 1 and Figure S3). A/J, BALB/cByJ and SWR/J strains were the highest for number of boli, whereas the C57BL/6J, FVB/NJ and CE/J strains were the lowest for this variable. Urination counts followed a different pattern, with the AKR/J and NZB/B1NJ strains urinating most frequently and the C57BL/6J and FVB/NJ strains not urinating in this task. For time spent in the center of the OF, the C57L/J, NZB/B1NJ, CE/J and FVB/NJ strains spent the greatest amount of time in the center, while the BALBc/ByJ, AKR/J, and 129S1/SvlmJ strains spent the least amount of time in the center. The CE/J, C57L/J and FVB/NJ mice were the highest for line crosses in the periphery, center and periphery to center in the OF, whereas the BALB/cByJ, A/J and CBA/J strains were the lowest for these variables. For percent of lines crossed in the center of the OF, the C57L/J strain was high, while the A/J and BALB/cByJ strains were low for this variable. The C57L/J, FVB/NJ, SWR/J and CE/J strains also exhibited the highest number of rears in either the periphery or the center areas of the OF, while the CBA/J, BALB/cByJ, 129S1/SvlmJ and A/J strains reared either very little or not at all in this task.

Strain mean correlations

The strain mean correlation matrix is given in Table S1. Many correlations attained statistical significance, and the pattern of high association did not clearly cluster within those variables usually taken as indices of anxiety-related behavior but also included indices of activity. Furthermore, they were quite high across the three different tests as well as within each test. To clarify these relationships, we depict three groups of correlated variables as scatter plots. Figure 1 shows one principal customary measure of activity and one anxiety-like behavior from each apparatus. These six variables are generally quite strongly correlated. Within each test, the correlations of the activity and anxiety-related behavior measures were 0.87 (LD), 0.82 (EZM) and 0.72 (OF). In each case, high activity was positively correlated with variables taken to indicate low anxiety. Other strain mean correlations are discussed later.

Figure 1.

Strain mean correlations between one principal activity and anxiety-like behavior measure in each task. Correlation coefficient (r) for each pair is indicated within inset in bold. Critical values: ≥ 0.52 (≤ 0.05); ≥ 0.64 (≤ 0.01). ‘OF line crosses’represent total number of lines crossed (center, periphery and center to periphery). In the LD test, number of transitions is the usual index of anxiety-like behavior. Ellipses represent Gaussian confidence intervals.

Principal components analyses on phenotypic data

Results are given in Table 2. We first performed PCAs for the variables within each task separately. For the LD test, this yielded three factors, which together accounted for 73.8% of the variation (factor 1: 39.0%, factor 2: 19.8% and factor 3: 15.0% of variation) in this model. The variables loading strongly and positively onto the first factor were transitions, line crosses in the light compartment, rears and voluntary reentry time. Total time in the light compartment and latency to leave the light compartment loaded strongly and positively onto the second factor, while nose pokes loaded strongly and negatively. Urine and boli loaded strongly and negatively onto the third factor.

A PCA on the EZM data generated two factors that together accounted for 64.5% of the variation in this model (factor 1: 45.6% and factor 2: 18.9% of variation). Percent time spent in the open arms, open-arm line crosses, closed-arm line crosses, rears and head dips loaded positively onto the first component, while latency to enter the open arm loaded negatively onto this factor. Boli and urine counts loaded strongly and negatively onto the second factor, while stretched attends did not load strongly onto either factor.

For the OF, the PCA analysis produced two factors that together accounted for 64.8% of the variation in this model (factor 1: 46.3% and factor 2: 18.5% of variation). Time in the center, line crosses in the periphery, periphery to center of the OF, and center and rears in both the periphery and center loaded positively onto factor 1, while boli and urine counts loaded positively onto factor 2.

Thus, the individual PCA for the LD test discriminated activity from location choice (anxiety-like behavior) to some degree and both variables from the boli/urination variables. However, for the EZM and OF tasks, the PCAs tended to dissociate a combined activity/anxiety factor and a boli/urination factor.

Next, a PCA was performed using all the variables from all three tests. This yielded four factors with eigenvalues >2 that together accounted for 62.0% of the variance. Variables that loaded strongly on factor 1, which accounted for 25.6% of the variance, were as follows: rears and line crosses in the light side of the LD test and voluntary reentry time in the light compartment of the LD test, percent total time spent in the open arms, open-arm line crosses and closed-arm line crosses in the EZM, as well as rears, head dips and latency to enter the open arm (loading negatively) of the EZM, and rears in the periphery of the OF. Variables that loaded strongly on factor 2, which accounted for 9.2% of the variance, were total time spent in the light compartment, latency to leave the light compartment and nose pokes between compartments (loading negatively) in the LD test. The variables that loaded strongly on factor 3, which accounted for 11.1% of the variance, were boli and urination counts for all three tasks. Last, total time spent in the center, periphery to center and centerline crosses and rears in the center of the OF loaded strongly onto factor 4, which accounted for 16.0% of the variance in this data set. Number of transitions in the LD test and line crosses in the periphery of the OF loaded similarly onto factors 1 and 4 and EZM stretched attends did not load highly onto any factor, although this factor loaded moderately (negatively) onto factor 1. In general, these results were quite similar to the results from each task considered separately.

Principal components analyses on strain mean data

Because the results from the strain mean correlation matrix and the phenotypic PCAs reported earlier suggest that activity and anxiety-related variables tend to be strongly associated, we examined the data with a PCA using strain mean values for each variable. The results were somewhat consistent with the overall phenotypic PCA reported above (data shown in Table S2), but also showed a number of differences. The strain mean PCA analysis is discussed in Table S2. Figure 2 shows scatter plots of the strain mean associations designed to show the underlying association between boli/urination vs. variables indicative of activity, with high activity correlated with low boli/urination. A similar association between defecation/urination and anxiety-related variables was observed, with higher scores on anxiety-like behavior variables (indicating a low anxiety-related behavioral state) correlated with low boli/urination.

Figure 2.

Strain mean correlations of boli and urination counts with activity and anxiety-related behavior measures in each task. Correlation coefficients (r) indicated within inset in bold. Because boli and urination counts were highly correlated among tasks, an average for all three tasks was used in the correlational analyses. (a) Correlations between activity variables and boli/urine variables; (b) Correlations between anxiety-related behavior variables and boli/urine variables. ‘OF line crosses’represent total number of lines crossed (center, periphery and center to periphery). Ellipses represent Gaussian confidence intervals.


In agreement with earlier studies, our results illustrate the existence of strain differences for these three tasks (Avgustinovich et al. 2000; Bouwknecht & Paylor, 2002; van Gaalen & Steckler 2000; Griebel et al. 2000; Kopp et al. 1999; Wahlsten et al. 2006). Our estimates of genetic effect size were also substantial (0.26–0.85) for both activity measures (number of line crosses in the LD test, number of closed line crosses in the EZM and number of line crosses in the OF) and for anxiety-related behavior indices (number of LD transitions, percent of time in open arms of EZM and time in the center of the OF). Another similarity between our study and earlier work is the negative correlation observed between boli and activity measures (DeFries et al. 1978; Hall, 1934).

Our data are also in general agreement with studies examining the relative ranks of strains tested for similar measures. Results from selected experiments are compared in Table S3. Although a variety of inbred strains were used for these studies, some commonalities emerged. For example, in most studies, the C57BL/6J strain made more LD transitions than those of the other strains but ranked lower in EPM percent total open arm time. Additionally, our study supports previous research suggesting that both the A/J and the BALB/cByJ strains exhibit a low number of LD transitions and is consistent with previous reports of OF activity in the A/J and BALB strains, which show lower ambulation than other strains tested.

In contrast, the results of our all-task PCA were not in complete agreement with previous research. In our analysis, none of the factors could be characterized unambiguously as reflecting ‘activity’ or ‘anxiety’. The first and fourth factors loaded a number of general locomotor activity measures as well as anxiety-like behavior measures, indicating that activity and anxiety-like behaviors cannot be separated effectively using these specific tasks. The interpretation of our analysis differs from the PCA performed by Trullas and Skolnick (1993), who dissociated anxiety-like behavior from general activity in the EPM and OF tests. Carola et al. (2002) performed a study using two inbred strains (BALB/c and C57BL/6) in two anxiety-related behavioral tasks (EPM and OF). This study correlated factors obtained by performing separate PCAs on OF and EPM and found that OF factor 1, ascribed to general locomotor activity, correlated more strongly with EPM factor 2, defined as ‘anxiety-like activity’, than with EPM factor 1, which was believed to reflect general locomotor activity. Belzung and LePape (1994) encountered similar issues when testing mice created from the breeding of two F1 hybrid crosses (C57BL/6J × DBA/2J and C3H × BALB/c+ccrosses). Two distinct factors emerged in their PCA: an unfavorable reaction to novelty in the EPM and novel object tasks and general activity. However, this analysis showed LD test behavior as related to both factors, which makes it difficult to conclude that these tests are unequivocally measuring unique factors for anxiety-like behavior and activity. Each study concluded that anxiety-like behavior in these models could be dissociated from general locomotor activity, a conclusion that our study does not strongly support.

Our results for PCAs on variables from individual tasks, which were in general agreement with our all-task PCA, were also somewhat discrepant from previous literature. Carola et al. (2004), using C57BL/6 and BALB/c mice, performed PCA analyses on EPM and OF behavioral data and found that anxiety-like variables loaded onto different factors than activity variables. The dissociation between activity and anxiety-like behaviors was also found for a number of studies examining rodent behavior in the EPM, with researchers reporting general activity variables loading onto a separate factor from anxiety-like behaviors (Aguilar et al., 2002; Doremus et al. 2006; Wall & Messier 2000). One variable that remained consistent across tasks in the current study was boli. Hall (1934) first described the relationship between boli and urine counts and ‘emotionality’ in rats, with rats producing more boli/urine during periods of high ‘emotionality’ (i.e. when placed in a novel or stressful environment). This relationship has been repeatedly documented in rodents (Castanon & Mormède 1994; Carola et al. 2004; Gentsch et al. 1981; Henderson et al. 2004) and does not appear to be influenced by locomotor activity levels. However, caution must be used when using defecation as an indicator of anxiety-like behavior because strain differences in metabolism or intestinal function could potentially affect the expression of this variable.

Our individual PCAs support the idea that high activity is correlated with low anxiety-like behavior. Notable exceptions are the time spent in the light compartment and latency to leave the light compartment of the LD test, both of which loaded onto a different factor than did activity and anxiety-like behaviors. This agrees with Guillot and Chapouthier (1996) who found that number of transitions and time spent in the light side loaded onto two different factors in this task, suggesting to the authors that these two variables are influenced by processes that are under the control of different genetic factors.

While analyses using individual phenotypic data reflect genetic and environmental influences (or an interaction between the two), analyses using inbred strain mean allow researchers to reduce environmental influence and examine common genetic factors underlying reported behaviors (Hegmann & Possidente 1981). With this in mind, we repeated our PCA analysis using strain mean values. Although this PCA analyses differed somewhat from our phenotypic analyses (see supplementary materials for details), none of these analyses could effectively dissociate anxiety-like behavior variables and activity variables in these tasks. It should again be noted that the use of strain means violates the principles of PCA analysis as many more ‘subjects’ (strains) would be required for adequate statistical power.

One issue that may have affected the results of our study was the inclusion of the A/J inbred mouse strain, whose behavior in these types of tests can vary from that of other strains because of very low locomotor activity. To explore this issue, all statistical analyses were performed again after excluding this strain. Exclusion of the A/J did not significantly affect the results of these analyses, so the A/J strain has been included in the reported results.

Another issue that may have affected the results of our study is the discovery that laboratory environment can have a substantial impact on results of anxiety-related behavioral tests. Crabbe et al. (1999) found that, despite considerable efforts to standardize all housing and testing conditions, strain differences in variables associated with both activity and anxiety-like behavior in an OF and an EPM differed between laboratories. Furthermore, a recent study conducted by Izídio et al. (2005) tested SHR and LEW rat strains in an OF, EPM and LD test to examine the effects of a number of environmental variables and found that both position of home cage on the rack and arousal level directly before testing can significantly affect anxiety-like behavior indices in all three tests. Other studies have shown that environmental factors can affect genetic ‘signal’, and have had some success in identifying these sources of influence (Chesler et al. 2002; Valdar et al. 2006). Although our current study attempted to control for these types of factors, it is possible that our laboratory environment may have influenced the behavioral results seen.

Although the EZM has been established as a valid alternative to the EPM test (Shepherd et al. 1994), our use of the EZM may also have affected the results because of its lack of a central platform. Subjects in an EPM can spend a large proportion of their test time in the central platform (Lee & Rodgers 1990; Rodgers et al. 1992) and may enter the closed arms without entering an open arm. Thus, the lack of such an area in this study may have forced the subjects into a different pattern of responding, yielding results different from what we might have seen in an EPM.

Lastly, as previously noted, all subjects were tested in an LD box in which the dark compartment is completely enclosed. Therefore, our measurement of locomotor activity (line crosses in the light compartment) may not be readily comparable to activity measures in other apparatuses. Specifically, the light compartment is typically thought to be ‘anxiety’ provoking, and line crosses in this compartment may not reflect locomotion entirely unaffected by an ‘anxiety’ state in rodents. This may have inflated the correlation between our locomotor activity measurement (line crosses in the light compartment) and anxiety-like behavioral measurement (transitions) in this task.

In summary, the results of the present study support previous research suggesting that significant strain differences exist for variables thought to measure anxiety-like behavior in mice. However, because the indices of anxiety-related behavior and activity in these tasks are so highly correlated, both within and between each of the tasks studied, this suggests that anxiety-like behavior measured in inbred mice must be interpreted with caution. Many studies have found that these types of anxiety tasks are only consistently predictive for benzodiazepine therapies, as opposed to other effective drugs such as those that affect the 5-HT system (Handley & McBlane 1993). Thus, the debate surrounding the ability of each of these tasks truly to capture human anxiety in an animal model is far from over. Ultimately, our experiments cannot completely answer the question whether these three tests measure the same underlying anxiety-like condition or whether our findings can be generalized to other unconditioned anxiety-like behavior tasks, such as the EPM. Furthermore, we did not extend our analysis to any tasks studying conditioned anxiety-like behavior. The use of inbred strain panels allows the cumulation of data across experiments with these stable genotypes, and a collection of conditional anxiety-like behavioral data in these murine strains would be a valuable addition to the literature.


The authors thank Dr John Belknap for statistical advice, Dr Christopher Kliethermes for advice on experimental design and interpretation and Christina Cotnam for technical assistance. The authors also thank the anonymous reviewers and editors of this journal for their helpful suggestions. These studies were supported by National Institute on Alcohol Abuse and Alcoholism grants AA07468, AA10760, AA13519, T32 AA015822 and a grant from the US Department of Veterans’ Affairs.