| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
INNOVATIVE METHODOLOGY
MOUSE PHENOME PROJECT
1Department of Zoology, George S. Wise Faculty of Life Sciences, and 2Department of Statistics and Operations Research, The Sackler Faculty of Exact Sciences, Tel Aviv University, Tel Aviv, Israel 69978; and 3Maryland Psychiatric Research Center, Department of Psychiatry, School of Medicine, University of Maryland, Baltimore, Maryland 21228
Submitted 10 February 2004 ; accepted in final form 15 February 2004
 |
ABSTRACT |
|---|
 TOP ABSTRACT METHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
behavioral phenotyping; thigmotaxis; ethological measures; Mouse Phenome Project; C57BL/6J; DBA/2J
Behavior in the Open Field Test (OFT) is a widely used test of anxiety as well as of exploration and locomotor activity (1, 39) and is one of the most commonly used behavioral tests in genetically engineered mice research (5). As a test of spontaneous (unconditioned) behavior, it allows the animal to exhibit a wide range of behaviors and therefore is highly suitable for the study of complex phenomena such as anxiety. Typically, however, only two spatial measures are collected in this test: the total distance traveled (considered as a measure of general activity) and some measure of the animal's tendency to avoid the arena center. The latter measure is considered anxiety related, based on the assumption that the arena center is more threatening for rodents than its periphery and based on the increase in center occupation seen after administration of anxiolytic drugs (33, 36). These simple measures can be recorded automatically, allowing high-throughput analysis, but they far from exhaust the behavioral richness exhibited by animals in the OFT. More comprehensive analyses of tests of unconditioned behavior are needed (30, 31, 38) that would provide detailed and articulated information about different aspects of the animal's behavior, thus allowing for the reliable interpretation of genetic and pharmacological results. One way to meet this need is to use test batteries, which would include many different behavioral tests (5, 14, 32). Another (not incompatible) approach is to extract more information out of a single test. This is generally done by adding "ethological" measures, collected by human observers (e.g., the frequency of grooming, rearing, stretch attends, and so forth; Refs. 4, 6, 7, 11, but see Ref. 29 for an exception). This type of analysis adds a considerable amount of information but is relatively low throughput, involves subjective judgment, and uses ad hoc criteria.
A complementary high-throughput approach is offered via a software-supported strategy called Software for the Exploration of Exploration (SEE; Refs. 15, 16, 23), which is aimed at the investigation of rodent OFT behavior. SEE includes statistical algorithms that allow a reliable estimation of the animal's locations and momentary speeds at a high spatial and temporal resolution and the segmentation of the continuous x,y-location time series into a string of intrinsically defined, ethologically meaningful units. This allows the automatic calculation of multiple behavioral measures (end points) based on the quantification of the various features of these units. SEE end points have been shown to discriminate between inbred mouse strains in a replicable manner across laboratories (22).
Here we present a further development of SEE, which provides a detailed description of mouse behavior near the wall and in the center of the arena. This behavior is commonly quantified by several measures: the time spent in the center (or its proportion out of the entire session), the distance traveled in the center (or its proportion out of the total distance traveled), the number of entries into the center, and the latency to enter the center. Most studies use only one of these measures, but some use several. Sometimes one or more of these measures are quantified separately for a certain time period at the beginning of the session (e.g., first 5 min), based on the assumption that the mouse's anxiety level changes with time.
Obviously, any measure concerned with avoiding the center requires a definition of "being in the center." The arena center is defined arbitrarily as the area at a distance of more than x cm from the wall, where x is a "cutoff" number ranging between 5 and 20 cm in different studies. The value of the cutoff may be based on ethological considerations (e.g., a distance that enables a mouse to keep its vibrissae in contact with the wall; Ref. 35) but is often determined by technical factors, such as the dimensions of the squares composing a photocell chamber. If behavior in the OFT is stochastic and if the transition between behavior near the wall and in the center is continuous, then the precise definition of the cutoff value (e.g., 5, 10, or 15 cm from the wall) between the two is not of crucial importance, and an arbitrary choice would be justified. This, however, does not seem to be the case. Observation of mouse behavior in the OFT reveals a clear structure: mice typically either run parallel to the wall in close proximity to it or make forays into the center. Furthermore, these forays are of different types: small arc-shaped forays performed near the wall or long forays that cross the arena. A structured behavior demands intrinsic definition of cutoff values, since arbitrarily defined cutoff values might either cut across a continuous structure or indiscriminately lump together distinct structures. Because the behavioral structure is, moreover, likely to differ across preparations (e.g., inbred strains or drug-treated animals), the use of arbitrary criteria might blur important differences between them. Analysis based on intrinsic criteria should therefore considerably enhance the discriminative power of the OFT, and applying such analysis to the behavior near the wall and in the center should increase the discriminative ability of its anxiety-related aspects. In addition, measures based on intrinsic properties of behavior have been shown to provide results that are replicable across laboratories (22), a major concern in present-day research (12).
In this work, we present a new SEE algorithm that automatically separates the path of the mouse into intrinsically defined patterns of movement near the wall and in the center. We also demonstrate its use in the design of new, replicable end points. This is done with data from two inbred strains, C57BL/6J (B6) and DBA/2J (D2), tested across three laboratories. These two strains are widely used in behavioral, genetic, and pharmacological research. Studies that have used the accepted OFT measures of distance traveled and center occupancy usually show B6 to be a more active strain than D2 and to occupy the center more (reviewed in Ref. 13).
After developing new end points of wall vs. center behavior and demonstrating their discriminative power and replicability, we use an independent set of data to assess the validity of these results.
|
METHODS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
ilan99/see/help) and have already been used in previous studies (22, 24). Data from three of the five experiments available in the database comparing B6 and D2 (one from each laboratory) were used. These experiments had the least differences in experimental conditions between laboratories (the differences are summarized in Table 1).
|
The experimental and housing protocols were identical for both studies and are described in detail elsewhere (22). Here we repeat the main points.
Animals
B6 and D2 male mice 9–14 wk old were shipped from Jackson Laboratories.
Housing
Animals were maintained in a 12:12-h reversed-light cycle (lights on from 8:00 PM to 8:00 AM) and were housed two to four per cage under standard conditions of 22°C room temperature and water and food ad libitum. The animals were housed in their room for at least 2 wk before the start of the experiment. All animals were maintained in facilities fully accredited by the American Association for the Accreditation of Laboratory Animal Care (AAALAC, MPRC, and NIDA) or by National Institutes of Health Animal Welfare Assurance Number A5010-01 (TAU). All studies were conducted in accordance with the National Research Council's Guide for the Care and Use of Laboratory Animals.
Experimental Procedure
The arena was a large (210- to 250-cm diameter, see Table 1), circular area with a nonporous gray floor and a 50-cm-high primer gray-painted continuous wall. Several landmarks of various shapes and sizes were attached to different locations on the arena wall and on the walls of the room where the arena was located. The arena was illuminated with two 40-W neon bulbs on the ceiling above the center of the arena.
The experiments were conducted during the dark part of the cycle, 1–2 h after its onset. Each experimental animal was brought from its housing room to the arena in a small opaque box and placed within the arena (in a standardized location, near the wall) while still in the box. After 20 s, the box was lifted, and a 30-min session began. The animal's movement was tracked with the Noldus EthoVision automated tracking system (34).
Data Analysis
The raw data obtained from the tracking system were smoothed with the use of a specialized algorithm implemented in the stand-alone program "SEE Path Smoother" (20). This procedure produces reliable estimates of momentary speeds during motion (momentary speeds during arrests were defined as zero).
As was previously shown, rodent locomotor behavior consists of two distinct modes of motion: progression segments and lingering episodes (15, 19). During progression segments, the animals traverse relatively large distances, attaining relatively high speeds. During lingering episodes, the animals stop and perform scanning movements, while staying in a circumscribed neighborhood. Segmentation of the smoothed path into progression segments and lingering episodes was done with the expectation maximum (EM) algorithm (17), using a two-Gaussian mixture model. Stand-alone user-friendly software for smoothing (SEE Path Smoother) and for segmentation (SEE Path Segmentor) can be downloaded at http://www.tau.ac.il/ilan99/see/help.
We developed the new algorithms and end points described in RESULTS using the Mathematica-based program SEE Package (16) and two recently developed extension programs, "SEE Experiment Explorer" and "SEE Endpoint Manager" (21).
Statistical Methods
Calculation of radial distances and radial speeds. The wall/center separation procedure requires the calculation of the radial distance (the distance from the wall) and the momentary radial speed (the component of the velocity in the direction of the center) for all time points. The computation of these two quantities is based on the arena radius. In a perfectly circular arena, the radius is, of course, a constant. However, in real life, an arena is seldom a perfect circle, often having some dents and irregularities in its shape. Therefore, calculating the radial distance at any time point with a constant radius introduces an error into the computations, and the effect of this error is amplified when the radial velocity (the first derivative of the radial distance) is computed. It was found that even very small dents in the arena shape can cause relatively large errors in the calculation of radial speeds.
To overcome this difficulty, the radial distances and velocities should be calculated with the actual shape of the arena. Because Ethovision (and other tracking systems) does not provide the spatial coordinates of the arena wall directly, an algorithm was developed for the estimation of the arena shape from the movement of the mouse in it. The algorithm, called "Arena Builder," utilizes the fact that most mice spend a large portion of the session running very close to the arena wall (actually touching it). Therefore, their locations can be used to produce an estimate of the arena shape. This estimated arena shape is in turn used to compute the radial distances and the radial speeds at all time points, without the need to use a constant arena radius. The arena shape is calculated separately for each session, thus correcting for small changes in the arena location and small shifts in the position of the camera that might occur between sessions. A detailed description of the algorithm is available online at http://www.tau.ac.il/ilan99/see/help.
Comparing end point results between strains and across laboratories. To assess the discrimination between strains and the replicability across laboratories of end points generated by the wall/center separation procedure (see RESULTS), we used the linear mixed-effects ANOVA model (25, 27). In this model, the strain is considered as a fixed factor, whereas the effect of laboratory is considered as random. This means that we think of the laboratory effect as being drawn from the population of all possible laboratories' effects. The interaction between strain and laboratory is also considered random. Thus a significant strain difference yielded by the mixed-effects ANOVA model can be regarded as replicable across laboratories. This approach is more conservative than the widely used linear fixed-effects ANOVA model: if a difference between two strains was found to be significant under the mixed model, it will also be significant under the fixed-effects model, but the opposite is not necessarily true. Transformations (listed in Table 2) were used for the analysis of each end point to correct toward approximately normal distributions and to stabilize group variances (which tend to increase with the increase in group means).
|
Correlation between end points. The Pearson correlations between the end points, on the transformed and within-strain normalized data, were computed.
Validation of the new end points using a second data set. Two aspects of the newly developed end points were examined with a second, independent data set. 1) To assess the stability of the strain differences, we tested the hypothesis that the difference between the two strains, in each end point, is the same as in the first data set. Note that, even within the mixed-model framework, such a difference of differences can be simply tested by using a variation of the two-sided t-test because the laboratory and interaction effects drop out. The false discovery rate (FDR) controlling Benjamin Hochberg (BH) procedure (3) was used to control for multiple comparisons. 2) The first data set was used for the development and selection of potential end points as well as for testing the strain differences. This may hamper the discriminatory value of the subset of end points that were found to be significant. For that reason, we retested end points that were found to be significant in the first data set for their ability to distinguish between strains and across laboratories in the second data set (using the mixed-model and FDR control).
|
RESULTS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
The wall/center separation procedure is an algorithm for separating the path of the mouse into intrinsically defined components of movement in reference to the wall and for defining discrete units of wall/center behavior. It consists of two stages. In the first, movement along the wall is distinguished from movement in the center. In the second, movement in the center is further partitioned into several components.
First stage: distinguishing between movement along the wall and in the center. This stage is based on the observed tendency of mice to run along the arena wall. While a mouse is running along the wall, both its radial speed and its distance from the wall are small. Therefore, to isolate the component of movement along the wall, we classify data points according to both their radial speed and their distance from the wall. Having performed this classification, we divide each class into the following discrete units: segments of progression along the wall, segments of progression in the center, and incursions into the center (forays into the center, which may be composed of several progression segments). Below is a detailed description of this stage of the wall/center separation procedure (also illustrated in Fig. 1).
|
2) With the use of the EM algorithm, a two-Gaussian mixture model is fitted to the frequency distribution of the momentary radial speeds (using square root transformation) of data points belonging to progression segments. The intersection point between the two fitted Gaussians serves as the first criterion (or cutoff value) according to which the data points are classified into two groups. As can be seen in Fig. 1B, the frequency distributions contain a prominent component of low radial speeds, which is captured by the first Gaussian fitted by the EM algorithm. Figure 1C shows that this component contains data points in which the mouse moves in parallel to the wall; the component that is captured by the second fitted Gaussian contains data points in which the mouse either approaches or moves away from the wall.
3) Most of the movement in parallel to the wall is performed along a narrow strip near the wall (as is obvious from Fig. 1C, top), but some of it is performed at a distance. To isolate the movement along the wall, all data points in which the mouse moves in parallel to the wall (i.e., all data points that belong to the first component in point 2 above) are further separated into two groups according to their distance from the wall, again using the EM algorithm with a two-Gaussian mixture model. The intersection point between the two fitted Gaussians is used as a second cutoff value. The frequency distributions of the distances from the wall show a salient component of small distances, which is captured by the first Gaussian (Fig. 1D), containing data points in which the mouse moves along the wall (i.e., in parallel to the wall and close to it).
Definition of units of behavior. DEFINITION I. A wall segment is defined as any series of successive data points belonging to the wall class. Similarly, a center segment is defined as any series of successive data points belonging to the center class (Fig. 2A). A wall or a center segment is bounded by either a lingering episode or a transition between wall and center motion. Note that a progression segment may include more than one wall and/or center segment (as when a mouse progresses along the wall, leaves it, and returns to it again, all without stopping).
|
There are two exceptions for definition II. Exception IIa is described as follows. It sometimes happens that a series of center segments contains more than one entry into the center, i.e., when a mouse enters the center, returns to the wall, performs a stop there (but does not run along the wall), and then enters the center again. To account for these rare cases, we modify the incursion definition to a series of successive center segments that begins and ends either with a wall segment or with a lingering episode performed near the wall, i.e., a lingering episode of which at least one data point is within the range of the "ring" formed by movement near the wall (i.e., the blue part of the path plots in Fig. 1E). To avoid artifacts caused by outliers, this range is defined as the 98th quantile of the distances from wall of all data points belonging to wall segments.
Exception IIb is described as follows. It was found that a small fraction of incursions (as defined above) is performed entirely within the near-wall domain (i.e., their maximal distance from the wall is smaller than the 98th quantile of the distance from wall of wall segments). These are small movements directed toward the center (therefore having high radial speeds), which do not proceed beyond the wall "ring." Despite their high radial speeds, these minute incursions are reclassified as movement near the wall because they are performed in the same distance from the wall as that of most wall segments. Data points that belong to these incursions are pooled with those that belong to wall segments, and new wall segments (i.e., series of successive data points) are defined from the pooled data.
DEFINITION III. Lingering episodes are classified into two groups: 1) center lingering comprises all lingering episodes that are enclosed within incursions, and 2) wall lingering comprises all the remaining lingering episodes.
Second stage: classifying incursions into several components. Even a cursory observation of the center component of any mouse's path (Fig. 1E and Fig. 2B) indicates that incursions are not a homogeneous group: some incursions are performed in relative proximity to the wall, others reach further into the center, and still others even cross the center. These different types of incursions can be characterized by the incursions' maximal distances from the wall. However, because the number of incursions in some of the sessions is too small to get a reliable classification, we use the maximal distances of center segments instead (due to there being more center segments than incursions in a session because some incursions are composed of several center segments). With the use of the EM algorithm, a Gaussian mixture model is therefore fitted1to the frequency distribution of the maximal distances from wall of center segments (after being log transformed). Figure 3A shows the results for a single B6 session, where three Gaussians were fitted by the EM algorithm. The intersection points between the fitted Gaussians are used as cutoff values between separate groups of center segments. These groups are in turn used to classify incursions into types. For example, in the session shown in Fig. 3, all incursions that contain only center segments from group 1 constitute a single type: near-wall incursions (Fig. 3B, label 1). Furthermore, all incursions that contain at least one center segment from group 2 and no center segments of group 3 constitute a second type: intermediate incursions (Fig. 3B, label 2). Finally, all incursions that contain at least one center segment from group 3 constitute the third type: arena-crossing incursions (Fig. 3B, label 3).
|
Using the Wall/Center Separation Procedure for Behavioral Phenotyping
The newly defined patterns can now be used to design new end points. One way to do this is to calculate existing end points separately for movement along the wall and in the center. Another way is to characterize features that are relevant only in the newly generated behavior patterns. Using both of these approaches allows the generation of a multitude of new end points of behavior near the wall and in the center. Below, we present several examples of such end points (their values are summarized in Table 2).
Examples of End Points That Are Calculated Separately for Movement Near the Wall and in the Center
Median path curvature during progression. The path curvature is defined as the change in direction between data points located at a fixed distance from each other (5 cm in the present case). This is an improved version, developed by Ehud Fonio (TAU), of the "radius of turn" end point (22). The median curvature was calculated for the pooled data points from all progression segments (Fig. 4A) and separately for pooled data points from wall segments only (Fig. 4B) and from center segments only (Fig. 4C). The results were compared for B6 and D2 mice in the three laboratories.
|
Median duration of lingering episodes. The median duration was first calculated for all lingering episodes (Fig. 5A) and then separately for wall (Fig. 5B) and center (Fig. 5C) episodes. Figure 5A shows that the median duration of lingering episodes is higher in D2 than in B6 mice. Figure 5, B and C, shows, however, that the strain difference holds only for lingering near the wall. The median lingering duration in the center is similar in both strains.
|
Number of incursions. Figure 6A shows that the number of incursions performed during a session is higher in B6 than in D2 mice in the three laboratories, but this strain difference is not statistically significant.
|
The median of the maximal distance from wall of incursions. The median of the maximal distance from wall of incursions indicates how far from the wall the mouse tends to venture during its center activity. The results for this end point are shown in Fig. 6C. The median maximal distance from wall in incursions is significantly higher in B6 than in D2 mice.
The proportion of incursions beginning with a lingering episode. Sometimes an incursion is directly preceded by a stop (lingering episode); at other times, a mouse proceeds from progressing along the wall to the center without stopping. The proportion of incursions that begin with a stop, out of the total number of incursions, is a measure for the tendency to stop before entering the center. This proportion is significantly higher in D2 than in B6 mice (Fig. 6D).
Average segment length in incursions. The average segment length in an incursion is calculated as the length of the incursion divided by the number of center segments included in it. The median of this measure over all incursions was calculated for each mouse. This end point can be regarded as a measure of the tendency to stop during progression in the center. Figure 7A shows that, in incursions, the median of the average segment length is significantly higher in B6 than in D2 mice. Another end point previously measured with the same data, the median length of progression segments (i.e., with no distinction between wall and center behavior), which may be regarded as the overall tendency to stop, showed no difference between these two strains (22). D2 mice thus have a higher tendency to stop than B6 mice only when progressing in the center.
|
1, meaning that B6 mice reach similar mean speeds when moving away as when approaching the wall. In contrast, the value in D2 mice was <1, meaning that D2 mice reach lower mean speeds when moving away from the wall than when approaching it. End Points Calculated Separately for Different Incursion Types
The end points presented above were calculated for the entire movement in the center, ignoring its subclassification into incursion types. Any end point characterizing center activity, however, can also be calculated separately for the data in each incursion type. This may potentially accentuate differences between strains but is relatively less useful for strains that are not very active in the center and have a relatively small number of incursions, such as D2 mice. This is because in such strains the number of incursions belonging to each type is often too small for meaningful statistical analysis. Because this was the case with the D2 mice in the present study, there were only two end points that could be calculated separately for each incursion type: 1) the median curvature (because this measure is not calculated per incursion but for all the pooled data points belonging to incursions) and 2) the number of incursions per session.
Median curvature. Figure 8A shows the results for median curvature during movement in near-wall, intermediate, and arena-crossing incursions. In the first two incursion types, there is a significant strain difference in the same direction as the difference in curvature calculated for overall progression in the center (see Fig. 4C), the path of D2 mice being more curved than that of B6 mice. The strain difference disappears, however, during arena-crossing incursions, where both strains have similar (and relatively low) path curvature.
|
Correlations Between End Points
The median of all correlations between all the new end points was about zero (0.007); the lower quartile was –0.21 and the upper quartile was 0.23; the lower and upper deciles of the correlations were –0.34 and 0.44, respectively. In addition, 59% of the correlations were statistically indistinguishable from 0. The pairs of end points with the most extreme correlation (above 0.8 in absolute value) were as follows: number of incursions and number of near-wall incursions (correlation = 0.86), median curvature in the center and median curvature in arena-crossing incursions (–0.89), and median curvature in the center and median curvature in intermediate incursions (–0.86).
Validation of Wall/Center End Points
The results presented so far were calculated on the same data set that was used for the development of the new end points. Having access to a new, publicly available data set contributed by our group to the Mouse Phenome Database (28) and also having access to the raw data of these strains (which are also publicly available at http://www.tau.ac.il/ilan99/see/help) allow us to validate the results derived on the first data set.
We start by assessing the stability of the strain differences on the 16 newly developed end points. Table 3 summarizes the results (means ± SE) for the two strains in the second data set. In addition, it displays the strain differences in both data sets and the P values for the comparison between them. The two P values of <0.05 do not remain statistically significant after we controlled for multiplicity using the FDR criterion. Thus there is no evidence for lack of stability in the strain differences across the two data sets in all 16 end points.
|
|
DISCUSSION |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
In this work, we present a SEE algorithm for the identification of the intrinsic structure of mouse behavior in reference to the wall of the arena. It reveals the existence of four distinct patterns: one of movement along the wall and three of movement in the center. Within these patterns, discrete units are defined: wall and center segments and wall and center lingering. Center segments and center lingering are concatenated into more complex units termed incursions. With the use of this structural basis, new end points characterizing the behavior near the wall and in the center can be designed and tested for discrimination and replicability. Several such end points were calculated on data from two inbred strains, B6 and D2, tested in three laboratories to demonstrate the phenotyping advantages offered by an SEE-based approach to wall/center behavior. One advantage is the ability to compute various parameters separately for different patterns and possibly reveal strain differences that are masked when the same parameters are computed indiscriminately for the entire data. This is demonstrated by the results for path curvature and for lingering duration shown in Figs. 4 and 5. When path curvature is computed separately for progression near the wall and in the center, a previously masked phenomenon is uncovered: D2 mice have higher path curvature than B6 mice during progression in the center (Fig. 4C). Consequently, path curvature results during overall progression (showing B6 to have higher path curvature than D2 mice, Fig. 4A) actually reflect only the strain difference near the wall. The opposite-direction strain difference in the center is completely obscured and serves only to somewhat reduce the end points' discriminatory powers (hence the higher statistical significance for path curvature near the wall than for the overall path curvature). Similarly, median lingering duration results (Fig. 5A) reflect a "mixture" of two phenomena: D2 make longer stops than B6 mice near the wall, and both strains have similar lingering duration in the center (Fig. 5, B and C). Using the wall/center separation procedure, we identify both phenomena and express these as two distinct end points. A yet finer discrimination is obtained when the behavior in different patterns of center movement is examined separately: path curvature of B6 mice is lower than that of D2 mice during near-wall and intermediate incursions but not during arena-crossing incursions (Fig. 8A).
Another fruitful phenotyping approach is demonstrated by the end points shown in Figs. 6 and 7, which describe different features of incursions. In addition to analyzing the entire data during activity in the center as a whole, we can examine separate forays into the center and compute their properties, further increasing the articulation of the behavioral phenotype. The only measure of this kind available in the present research is the number of entries into the (arbitrarily defined) center. Our technique, in addition to providing an intrinsic definition of center entries (incursions), allows the description of many characteristics besides their number. The monitored characteristics of incursions range from rather simple ones, such as length (Fig. 6B), to more complex ones, such as outward/inward speed ratio (Fig. 7B). An incursion, moreover, is not an indivisible entity; rather, it consists of discrete subunits, center segments and center lingering, thus further increasing the variety of possible end points (e.g., average segment length in incursions, Fig. 7A). Here too, the end points can be computed separately for different incursion types (Fig. 8B), and it is thus revealed that the strain difference in the number of near-wall incursions is strongly affected by laboratory, whereas the number of intermediate and arena-crossing incursions is not much affected by laboratory; thus there are replicable strain differences.
The fact that all of the 16 end points were validated with a separate data set, collected nearly 1 yr later across three laboratories, attests to the robustness of the end points and the usefulness of this SEE-based methodology for behavior phenotyping.
All of the 16 new end points can be applied to various strains and preparations with good prospects of providing stable results. Ten of these reliably discriminate between B6 and D2 mice.
Beyond Center Avoidance
All presently accepted measures of wall/center behavior assess the degree to which an animal avoids the presumably threatening arena center. Several of the new end points presented in this work are, like the presently accepted measures, also measures of center avoidance. In addition, our analysis includes end points that characterize the activity in the center, the activity near the wall, and the ways in which they differ. The number of incursions, median incursion length, and the median of an incursion's maximal distance from the wall can be considered as a measure of center avoidance. The results with these end points corroborate the results of previous OFT studies (reviewed in Ref. 13), which show B6 as a strain more prone to occupy the center than D2: B6 mice make more incursions into the center than D2 mice (Figs. 6A and 8B), incursions of B6 mice are longer on average than those of D2 (Fig. 6B), and B6 mice get farther into the center during incursions (Fig. 6C). However, our results show that, in addition to occupying the center more than D2, B6 mice move in straighter lines while in the center (Fig. 4C), are less prone to stop before embarking on a trip to the center (lower proportion of incursions beginning with a lingering episode, Fig. 6D), tend to run longer distances in the center without stopping (higher average segment length in incursions, Fig. 7A), and have similar speeds while moving away or toward the wall during center activity (whereas D2 mice move faster toward the wall than away from it, Fig. 7B). The two strains also differ in some properties of their activity near the wall: the path of B6 mice is more curved than that of D2 mice (Fig. 4B), and B6 mice make shorter stops while near the wall (lower median lingering duration, Fig. 5B).
Center avoidance is considered as indicative of anxiety, an interpretation that is backed by pharmacological and genetic studies (10, 18, 33). The interpretation of some of the new measures that characterize the activity in the center seems straightforward. Thus it is likely that an anxious mouse will hesitate more before going into the (aversive) arena center than a nonanxious mouse and therefore will have a larger proportion of incursions beginning with a lingering episode. Likewise, an anxious mouse may be expected to progress more slowly toward the center than away from it (thus having an outbound/inbound speed ratio, which is <1). Other measures, however, are less easy to interpret. For example, when in the center, would an anxious mouse run in a curved or a straight path? Would it tend to stop more or less than a nonanxious mouse? Would it make longer or shorter stops? At present, we are not able to propose an interpretation as to the face validity of these end points as measures of anxiety, and, of course, it is necessary to measure all of these new end points in pharmacological and genetic models of anxiety to discover whether any of them has predictive validity. However, by expanding the behavioral repertoire to include measures that describe behavior in the center in positive terms rather than only in terms of center avoidance, we aim at enhancing the articulateness and comprehensiveness of phenotypes of anxious (vs. nonanxious) mice. Applying the analysis to additional strains, and to various pharmacological and genetic preparations, should provide an answer to the question of which of the new measures are related to anxiety and in what way.
Potential Contribution of the Wall/Center Separation Procedure to the Study of Anxiety
Anxiety is a psychological construct that is thought to have complex behavioral manifestations and a complex genetic basis. The development of advanced tools for behavioral analysis can assist in addressing the complexity of this behavior with a degree of precision sensitive to genetic and pharmacological manipulations; these tools could increase the articulation and richness of behavioral phenotypes. The wall/center separation procedure generates such a phenotype for a classical anxiety-related test: behavior in reference to the arena wall in the OFT.
To obtain a larger number of relatively independent end points characterizing anxiety, behavior geneticists use measures from several tests (14, 38). This methodology has the potential advantage of highlighting properties of anxiety that are stable across situations. It is also possible, however, that the same level of presumed anxiety may be expressed in opposite directions in different stimulus situations. For example, although a dark compartment in the light/dark test may attract the animal to stay away from a lit open space, a bare wall in the OFT may prompt it to rush into the center. By providing many end points that characterize anxiety in the same stimulus situation, SEE has the potential advantage of characterizing the multifaceted nature of anxiety in the very same situation. The two methodologies are, therefore, complementary.
|
GRANTS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
|
FOOTNOTES |
|---|
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
1 Unlike the first stage of the procedure, in which a two-Gaussian mixture model is used, here the number of components in the Gaussian mixture model is chosen automatically using the likelihood ratio test. 
|
REFERENCES |
|---|
|
TOPABSTRACTMETHODSRESULTSDISCUSSIONGRANTSACKNOWLEDGMENTSREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  G. I. Elmer and N. Kafkafi Drug Discovery in Psychiatric Illness: Mining for Gold Schizophr Bull, March 18, 2009; (2009) sbn194v1. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 N. Kafkafi, Y. Benjamini, A. Sakov, G. I. Elmer, and I. Golani Genotype-environment interactions in mouse behavior: A way out of the problem PNAS, March 22, 2005; 102(12): 4619 - 4624. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
| Visit Other APS Journals Online |
