Experimental Assessment of Mouse Sociability Using an Automated Image Processing Approach

Summary

This protocol describes a method to quantify mouse sociability. Mice are videotaped as they move and interact in a special cage. Movie processing allows for the automated quantification of sociability with excellent accuracy and reliability.

Abstract

Mouse is the preferred model organism for testing drugs designed to increase sociability. We present a method to quantify mouse sociability in which the test mouse is placed in a standardized apparatus and relevant behaviors are assessed in three different sessions (called session I, II, and III).

The apparatus has three compartments (see Figure 1), the left and right compartments contain an inverted cup which can house a mouse (called "stimulus mouse").

In session I, the test mouse is placed in the cage and its mobility is characterized by the number of transitions made between compartments. In session II, a stimulus mouse is placed under one of the inverted cups and the sociability of the test mouse is quantified by the amounts of time it spends near the cup containing the enclosed stimulus mouse vs. the empty inverted cup. In session III, the inverted cups are removed and both mice interact freely. The sociability of the test mouse in session III is quantified by the number of social approaches it makes toward the stimulus mouse and by the number of times it avoids a social approach by the stimulus mouse.

The automated evaluation of the movie detects the nose of the test mouse, which allows the determination of all described sociability measures in session I and II (in session III, approaches are identified automatically but classified manually). To find the nose, the image of an empty cage is digitally subtracted from each frame of the movie and the resulting image is binarized to identify the mouse pixels. The mouse tail is automatically removed and the two most distant points of the remaining mouse are determined; these are close to nose and base of tail. By analyzing the motion of the mouse and using continuity arguments, the nose is identified.

Figure 1. Assessment of Sociability During 3 sessions. Session I (top): Acclimation of test mouse to the cage. Session II (middle): Test mouse moving freely in the cage while the stimulus mouse is enclosed in an inverted cup. Session III (bottom): Both test mouse and stimulus mouse are allowed to move freely and interact with each other.

Introduction

Impaired sociability is one of the primary symptom domains in a range of neurodevelopmental disorders, including autism spectrum disorders (ASDs)^1,2. It severely limits the patients' ability to develop relationships and greatly affects patients, families, and caregivers^2,3.

In an effort to better understand the mechanism underlying impaired sociability and develop treatments, several mouse models have been studied^1,4–9. One such strain is the BALB/c mouse, an inbred strain that exhibits strikingly low sociability, resembling features seen in patients with ASD5^8–12. For example, the BALB/c mouse displays a lower frequency of social approaches and a usually negative response to social approaches by a stimulus mouse (e.g., social avoidance) than the Swiss Webster comparator strain¹⁰.

To test whether a proposed treatment is effective in increasing sociability, objective measures are needed. A widely accepted framework is to analyze the sociability in a standardized apparatus that has three compartments (see Figure 1).

The left and right compartments each contain an inverted wire cup which houses a socially salient stimulus mouse. The test mouse is released into the apparatus for three intervals of 10 min (sessions I-III); session I assesses the locomotor activity of the test mouse, while sessions II and III assess different aspects of the test mouse's sociability.

In session I, the test mouse is allowed to acclimate to the apparatus. In this session, the number of transitions that the test mouse makes between the different compartments is measured. Importantly, sociability measures obtained in the standard three-compartment apparatus are dependent upon locomotor activity, and reduced locomotor activity can confound interpretation of sociability data.

In session II, the test mouse is moving freely and a stimulus mouse is placed under one of the inverted cups (counterbalanced). The stimulus mice are 4-week old outbred male ICR mice, age- and sex-matched to the test mice. The stimulus mouse is enclosed in an inverted cup in the side designated as the social compartment, and an empty inverted cup is placed in the side designated as the nonsocial compartment. In session II, the sociability of the test mouse is quantified by comparing the time it spends near the “social” inverted cup (less than 2 cm away) vs. the time it spends near the “nonsocial” inverted cup, and by the time it spends in the social compartment vs. the time it spends in the nonsocial compartment. Swiss Webster comparator mice spend significantly more time in the social compartment and near the social inverted cup compared to BALB/c mice that show no significant preference for the social inverted cup or the social compartment¹².

In session III, test and stimulus mice are allowed to interact freely (inverted cups are removed). Measures of sociability, stereotypic behaviors and transitions between compartments are reliably obtained in the third 10 min session of free interaction between test and stimulus mice and analyzed. In this report, the sociability of the test mouse is quantified via two measures: 1) the number of social approaches it makes toward the stimulus mouse, where a social approach is defined as moving forward (nose first) towards the stimulus mouse so that the nose is less than 2 cm away from the stimulus mouse; 2) the number of avoidances of social approaches by the stimulus mouse. The test mouse is considered to be avoiding the approach if it turns or moves away from the stimulus mouse, or if it temporarily stops moving (“freezing behavior”) after the stimulus mouse has approached the test mouse. BALB/c mice show a decrease in number of social approaches made towards the stimulus mouse and an increase in the number of negative responses to social approaches by the stimulus mouse, compared to the Swiss Webster comparator strain¹⁰.

The evaluation of the behavioral movies has traditionally been done manually, i.e., by watching the movie carefully (possibly at reduced speed) and activating stopwatches for the time the mouse spends in a certain compartment or near one of the inverted cups¹⁰. This method requires watching of the movie repeatedly and is, therefore, very time-consuming. Also, the accuracy is limited by the rater’s reaction time for the stopwatch measurements and ability to perceive whether the mouse is within 2 cm of the inverted cup. In recent work, two compounds, (D-serine and D-cycloserine) have proven effective in restoring sociability in BALB/c mice^9,10,12. It is known that both components modulate the NMDA receptor, which is known to play a central role in social behavior^2,12. A large array of compounds is known to modulate NMDA receptor activity in a variety of well-defined ways, and the potential for a breakthrough in ASD treatment makes the expedited evaluation of these compounds a priority and a fast automatic method for evaluation highly desirable.

Described below is an automatic analysis procedure developed for the assessment of mouse sociability. It uses image processing methods on all the frames of the mouse movies to identify the position of the mouse or mice. In sessions I and II, it further detects nose position and uses it to compute sociability measures. In session III, it automatically detects all approaches, but manual classification (social approach, social avoidance, or other) is required.

Protocol

The protocol for the ease of explanation has been divided into 2 parts: 1) Experimental procedure and 2) Automated analysis. All animals were treated according to the Guide for the Care and Use of Laboratory Animals¹⁵, and all procedures were approved by the Institutional Animal Care and Use Committee.

1. Experimental Procedure

1. Set up the cage, which is a black non-reflective Plexiglas rectangular box (52 x 25 x 23 cm³) divided into three compartments: social, nonsocial, and a neutral middle compartment (Figure 1). Place inverted wire cups in each of the end compartments during sessions I and II (discussed below) and house the stimulus mouse during session II.
2. Set up the camera in such a manner that the cage fills the field of view. Since lighting can interfere with mouse behavior, maintain a luminance of no more than 3.5 lux inside the cage, coming from indirect light from an incandescent light bulb and measured on the cage floor in upward direction. Also use an infrared light source, which mice cannot sense but the camera can detect. Record all movies in 720 x 480 standard definition.
3. Record a single frame of an empty cage for later image processing (called the “reference frame”) and save as a grayscale image.
4. Session I: Introduce test mouse into the cage and wave hand over the cage, indicating the start of session (compare Automated Analysis). Record movie for 10 min. At the end of 10 min, put the mouse back into its original cage.
5. Session II: Put the stimulus mouse under one of the inverted cups (counterbalanced).
6. Use the “Social left” or “Social right” button to select whether the stimulus mouse is in the inverted cup on the left or in the inverted cup on the right.
   NOTE: The inverted cup with the stimulus mouse is called the “social cup”, whereas the other cup the “nonsocial cup”. The compartment in which the social cup is located is called the social compartment, the compartment with the nonsocial cup is the nonsocial compartment, and the middle compartment is the “neutral compartment”.
7. Introduce the test mouse into the cage. Wave hand over the cage to indicate the start of the session. Record movie for 10 min. Importantly, keep mice in the sociability apparatus before removing the inverted cups to begin Session III.
8. Session III: Remove the inverted cups, release both mice and allow the test and stimulus mice to interact freely with each other. Wave hand over the cage to indicate the start of the session. Record movie for 10 min. At the end of 10 min, put both mice back into their respective cages.

2. Automated Analysis

NOTE: The commands in section 2 are issued via a graphical user interface (see Figure 2) that controls our software. Here we explain the step-by-step processing that our software allows; the software is also capable of performing the analysis in batch mode without user input.

1. For every movie, execute the following processing steps:
   1. Load movie (click Load button).
   2. Select cage configuration, which tells the analysis program the location of cage boundaries, compartment boundaries, and cup locations. Select one of four saved configurations or manually adjust all coordinates.
      NOTE: The configuration usually does not change from one experiment to the next and can therefore be reused.
   3. For each frame of the movie, perform the following processing steps (2.1.3.1-2.1.3.6).
      1. Convert the frame to grayscale by clicking the “Grayscale” button.
      2. Digitally subtract the grayscale reference frame (see step 1.3) by clicking the “Subtract” button. NOTE: The goal of this step is to make the pixel values of the frame very close to zero everywhere except for the mouse position.
      3. Binarize the frame by setting all pixels below a certain threshold level to black, all pixels above that level to white (click the “Binarize” button).
         Note: The idea is that since the mouse pixels generally have higher values than the remaining pixels, the mouse pixels will become white and all other pixels black. The optimal threshold level depends on the lighting condition and can be determined by assessing the mouse shape throughout a movie. If the threshold is chosen too low, the mouse appears too large, and pixels that do not belong to the mouse can be colored white. If the threshold is too high, the mouse becomes small and can disintegrate into disconnected pieces. In the recorded movies, the optimal threshold was always between 20 and 35 (on a scale from 0 to 255).
      4. Erode¹³ the binarized picture e times, then dilate¹³ 2e times, then erode again e times, by pushing the button “Erod/Dil”.
         Note: The purpose to this procedure is to remove the tail. For the resolution (720 x 480) that was used, e = 3 worked in all circumstances.
         Note: The software checks automatically for the presence of a hand in the picture (during the recording of movies, hands are waved through the picture to indicate the beginning of a session). If a hand is present, there will be unusually many white pixels in the binarized frame because the hand is much larger than a mouse. The condition for the presence of a hand is that there are more than 10,000 white pixels in the binarized frame. The software determines the beginnings of sessions I, II, and III, by detecting the range of frames in which hand movements occur.
2. In sessions I and II, continue as follows:
   1. Detect the largest connected component of the set of white pixels (Click “LComp”). Note: The purpose of this step is to eliminate white pixels that do not belong to the mouse, e.g., caused by the motion of the stimulus mouse in session II. The largest connected component is in almost all frames the mouse.
   2. Determine the two most distant white pixels (called the “ends” of the mouse) by clicking “Find Ends”.
      Note: The software establishes the continuity of the two ends, i.e., for each of the ends in the first frame to which of the ends in the subsequent frames they belong. The general idea is that since frames are recorded at high rate, the ends cannot travel far from one frame to the next, so that for the correct continuation of the ends from one frame to the next is that which minimizes the sum of the distances that the ends move. See the Discussion for more detail.
   3. Determine which end of the mouse is the nose.
      1. Determine the direction in which a mouse moves by observing how the center of gravity (COG) changes from frame to frame (the COG’s motion is the motion of the sum of all mouse pixels).
         Note: The COG of a set of pixels p₁, ... , p_n  with coordinates x₁, .. , x_n   is defined by:
         
         Mice usually move forward (i.e., towards their nose). In a section of several hundred frames, there is always very high confidence in the head/tail detection.
         Note: The software automatically determines towards which of the detected ends the mouse is moving; this end is identified as the nose and marked with a red circle in the movie. When the detected nose is marked with a red circle in every frame, the user can check the quality of the detection.
3. In session III, continue as follows:
   1. Detect two largest components of white pixels.
      1. If both are of acceptable size for a mouse and their centers of gravity (COGs) are compatible with the mouse COGs of the previous frame, consider these two components the two mice.
      2. Consider mouse sizes acceptable if they have not changed by more than 20% since the previous frame, and consider COGs of two consecutive frames as compatible if their difference is not larger than the distance a mouse can travel in the time between frames (this distance follows from the maximum mouse speed, which can be determined as part of the preparations of an experiment series by analyzing the motion of several individual mice (as in sessions I and II).
         Note: If the two largest components are not acceptable sizes for a mouse, or the COGs are too different from those in the previous frame, the software automatically increases the binarization threshold until the two largest components are of acceptable size and their COGs are compatible with those of the previous frame.
   2. Click the test mouse in the first frame. Use the analysis in 2.3.1 to identify both test and stimulus mouse in every frame of session III.
   3. Identify the frames in session III in which the one mouse socially approaches the other. Note: The software implements this step by automatically checking that the distance between the mice is less than the typical interaction distance of mice (we use 2 cm).
   4. When the software plays a short segment of the movie with the detected approach to the user, click a button to indicate whether this approach qualifies as a social approach by the test mouse, social avoidance by the test mouse, or neither.
      Note: Step 2.3.4 can be automatized if it is known which mouse is the test mouse. Since this information gets lost when the mice get too close to each other (or climb on top of each other), and since we have not yet found a way to label the mice in a way that can be detected reliably by our software AND does not affect their behavior, we have reverted to manual classification for now.
4. To evaluate processed movies, load movie using Load button.
   1. Click on Start Session 1 to display the first frame of session 1 (or Start Session 2 to display the first frame of session 2).
   2. Click on Play/Stop to play and stop movies.
      Note: The processed movies will display mouse with head and tail marked with differently colored circles. All sociability measures are displayed on the screen and updated as the movie plays.
   3. Export sociability data in Excel format by clicking “Export”.
   4. Click on “Compile Data” to compile all the data of all movies in a folder into a single excel file.

Results

Figure 3 shows a color frame of a behavioral movie with an empty cage. The mouse has not been introduced into the cage yet. The position of the cups, (social and nonsocial) and the compartment boundaries are superimposed. The color frame of empty cage is converted to 8 bit grayscale as shown in Figure 4 and used as a reference frame. The reference frame is subtracted from every other frame of the captured movie.

Figure 5 shows one of the frame...

Discussion

By combining videotaping mice and automatic movie analysis, we have created an affordable, reliable high-throughput screening technique.

The accuracy of the automated analysis was compared with the results of manual analysis for more than 100 movies. In the overwhelming majority of frames (> 99%) of frames, the nose of the mouse is identified with good precision. Most of the (few) misdetections have no effect on the sociability measures, so that around 80% of the movies do not require any c...

Materials

List of materials used in this article

Name	Company	Catalog Number	Comments
Plexiglas cage	Norva Plastics, Norfolk, VA	custom made	Dimensions and layout described in manuscript, can be adjusted according to needs. Use non-reflective plexiglas for to facilitate image processing
Wire cups	Kitchen plus	315	Use one to house stimulus mouse, one empty
Video camera	SONY	HDR-PJ790	Can be replaced by any camera with Comparable specifications
OpenCV (Image processing library)	Willow Garage	N/A	Any modern image processing library can be used.