Automated behavioral analysis of limbs’ activity in the forced swim test

Abstract

The forced swim test in rodents has been widely used to evaluate potential effectiveness of antidepressant medications since it was described in 1977 by Porsolt. In this test, a rodent is placed in a water container, and its immobility time is measured. The immobility time indicates the level of inactivity, interpreted as ‘hopelessness’, and has been shown to decrease when the rodent is treated with antidepressant medications. The simple measure of immobility time does not take into account intermediate behaviors during testing (ranging from total immobility to extensive ‘struggling’ behavior) and does not show normal Gaussian distribution in tested groups of rats.

We have previously developed a software allowing an observer to assign scores to the full range of intermediate behaviors by continuously reporting the motion of the limbs using a joystick. Based on the joystick score, we have now developed an automatic tool that uses computer vision algorithm (CVA) to analyze specifically the motion of the limbs and generate an objective, reproducible and automated score.

In the current study we have analyzed data obtained during the swim test using the traditional immobility time, the joystick analysis and the new CVA method to test the distribution of scores in a group of 20 rats. In addition, we tested the effects of various medications using these different scoring methods. The CVA method has been validated and positively correlated with the joystick score. Data obtained using the CVA method is objective, reproducible, and significantly reduces the time required for human analysis.

Introduction

For several decades, research communities and pharmaceutical industries have used the forced swim test to evaluate potential new antidepressant medications (Borsini et al., 1991, Dhir and Kulkarni, 2008, Porsolt et al., 1977a, Wong et al., 2000). The paradigm consists of a pretest session, in which a rat is placed into a cylinder filled with water for 15 min and a test session 24 h later, in which the rat is placed again in the same tank for 5 min. The test is based on the observation that rats, following initial escape-oriented movements, develop an immobile floating posture in the water cylinder. When they are placed again in the testing apparatus 24 h later, they resume this posture quickly. This posture was interpreted by Porsolt et al. (1977b) as reflective of the animal's state of despair, elicited by its perception of the hopelessness of the situation learned during the first session. The total amount of time in which the animal demonstrates this behavior is therefore measured and termed ‘immobility time’.

Acute administration of candidate compounds between the two exposures to the swimming tank may reduce or prevent the development of such immobility (Porsolt et al., 1978). Tricyclic antidepressants, monoamine oxidase inhibitors, and atypical antidepressants such as mianserine and iprindole, are all effective in this paradigm (Borsini and Meli, 1988). However, there are several shortcomings with the traditional type of measurement and several compounds have been identified as generating false positive (Borsini and Meli, 1988, Delini-Stula et al., 1988, Panconi et al., 1993) or negative (Lucki, 1997) results. In his original paper, Porsolt described the immobility posture of the rat “only [as] those movements necessary to keep its head above the water” (Porsolt et al., 1977b). The amount of time that the animal spends immobile can indicate, according to Porsolt, a state of despair in which the rat has learned that escape is impossible. In order to remain afloat, the animal makes certain, slight swimming movements that are less relevant to its escape behavior, than active swimming using its limbs. The observer in the traditional Porsolt test does not account at all for the time the animal spends during the active behavior. In order to monitor and measure other behaviors during the swim test (including ‘intermediate’ behaviors that cannot be defined as clear struggling or clear immobility), a more flexible measure is needed.

Some efforts to overcome these limitations and to enhance the sensitivity of the traditional Porsolt paradigm have been previously reported. Some methods try to give a broader definition to the active part of the behavior. For example, Detke and Lucki (1996) and Cryan et al. (2005) measured different types of behaviors, divided into swimming, climbing and immobility. They define the behavior for each interval of 5 s during the swim test, as one of the above-mentioned three behaviors according to which was more prominent in the 5 s period. This method was shown to allow detection of antidepressant effects of medications that were not effective in the traditional scoring method.

Other methods attempted to make the measurement more physical and objective. Shimazoe et al. (1987) used tremor sensors surrounding the cylinder to record water vibrations while rats were swimming in it. De Pablo et al. (1989) measured the variation in the frequency of the natural electromagnetic field of water induced by movements of rats. Hedou et al. (2001) measured the distance that the animal moved using a special tracking system and software (Ethovision by Noldus). The advantage of such automated methods is that they are not biased by subjective or inexperienced observers. However, the first two methods can measure only a defined area of the water tank, and the third method measures only the total activity or distance that the animal moved. There are situations in which the animal makes vigorous movements with its paws within one area of the swimming tank for long periods without significant changes in the position of the entire body, or with relatively small effects on water vibrations in other parts of the tank. Such periods are not evaluated properly using the above-mentioned approaches. In addition, all of these methods require complex, expensive, and dedicated equipment.

In our recent study (Gersner et al., 2005), we proposed the joystick analysis, which allows measurements of the full range of the intermediate behaviors (from total immobility to an extensive struggling behavior), and was validated against the traditional Porsolt paradigm. The ability to generate any number between 0 and 100 creates a continuous score rather than a binary score of the activity and avoids labeling the behavior into three types as in the method suggested by Detke and Lucki (1996). Unlike the immobility time measures, which did not show a normal Gaussian distribution in tested groups of rats, the joystick score generated a normal Gaussian distribution, which is of a statistical value. Although the joystick score is valid, dependable and has been successfully used in recent works (Lewitus et al., 2008, Toth et al., 2008), it has disadvantages: it is time-consuming, and like all other measures that are not automated, it involves subjective decisions of the raters and therefore might be less reproducible and less accurate.

To overcome these problems and still measure accurately specific activity of the limbs only (and not the whole body due to the caveats mentioned above), we have established a novel automated method, which is based on the sensitive joystick score of limbs’ activity. Due to its length, measurement of the tail movements would be the dominant factor of activity and limbs movements would become negligible in the final score. Moreover, tail movements is mostly required for animal's balance and the aim of our measurements is to monitor activity that is interpreted as animal's attempt and ‘motivation’ to struggle and escape from an unpleasant situation. We have therefore considered the tail movements largely as unspecific behavior. Considering these goals, we cooperated with ProTrack, a vision technology company, to create an automated measure based on a computer vision algorithm. Given a video file shot from below the water tank we developed a software based on the following specifications: (1) score high frequency movements of the front and hind limbs; (2) disregard whole-body turns and movements of the tail; (3) analyze videos taped from a position below the water tank; (4) ignore noise factors caused by water movement, differences in illumination and animal feces in the water tank, so that these would not contribute to the final score.

Methods from computer vision using proprietary algorithms of Protrack Ltd. were used to score the animal movement, thus achieving an objective and reproducible score that required only the video file to be analyzed.