All animals were DBA/2J (Strain #:000671, RRID: IMSR_JAX:00067) or C57BL/6 J (Strain #:000664, RRID: IMSR_JAX:000664) mice obtained from Charles River. Females (6–8 weeks old) used in behavior experiments were bred to males (8–10 weeks old) of their same strain. Female mice were group housed before and after breeding until approximately E17, when they were moved to single housing in a behavior box home cage prior to birth. All experiments were approved by the Animal Care and Use Committees of Johns Hopkins University. Animals were housed on a 12:12 light/dark cycle. Food and water were available ad libitum.

Behavior boxes were designed to enclose standard mouse cages (maximum of 5 mice per cage). The box was constructed with an aluminum frame (20 mm T-Slotted Rails, 80/20) and plastic walls (6 mm ABS Hair Panel, 80/20). Each box was equipped with a hinged door on the front, large enough to accommodate the cage and water bottle. Holes were drilled in the walls of the box to allow wires for sensors and actuators. Care was taken to make the boxes light tight and gaffer tape (25–50 mm black masking tape, Thor Labs) was used when necessary.

For our experiments, we included a number of sensors and actuators. Additional experimental components can easily be added as needed. We included (1) a ceiling mounted, downward facing, infrared (IR) video camera (Pi NoIR) to allow both daytime and nighttime video recording, (2) white and IR LEDs to illuminate the behavior box (Note: to match the Pi NoIR camera sensor, 940 nm or shorter wavelength IR LEDs are preferred for optimal nighttime illumination), (3) a temperature and humidity sensor, and (4) a circulating fan for climate stability. The IR camera has a flex cable connecting it directly to the Raspberry Pi computer. The white and IR LEDS as well as the fan were powered via a quad-channel relay allowing them to be independently turned on and off by the Raspberry Pi computer. The tops of the home cages were replaced with a clear acrylic panel to provide an open field of view and improve the quality of the video recording. Mirrors were placed on either side of each mouse cage to allow the ceiling-mounted camera to capture side view video as well. These strategies greatly improved the over-all video quality and allowed the precise scoring of complex maternal behaviors.

We used boxes of approximately 24” W × 24” D × 24” H, but strongly recommend setting up the mouse cage, water bottle, and mirrors to confirm width and length dimensions. Likewise, we recommend setting up the ceiling mounted IR camera to determine the video field-of-view before deciding on the height of the box.

All components including suppliers, price, and additional notes are provided in Materials and Equipment (Supplementary Table 1).

Each behavior box had a dedicated Raspberry Pi computer (Model 3B or newer, Raspberry Pi Foundation) to control the sensors and actuators within the box and to provide a web- and file- server to remotely control the experiment. Each Raspberry Pi computer included the computer itself, a 5 V/2A AC/DC power adapter, a 16–32 GB micro SD card for the operating system, a 64–256 GB USB thumb drive to locally store recorded files including video, and an ethernet cable. Each computer also had a ribbon cable to connect the camera to the computer (2 m Flex Cable, Adafruit). One dedicated Raspberry Pi computer was attached to the back side of each behavior box.

Each Raspberry Pi computer used the Raspberry Pi OS operating system (version Buster). This OS is based on the Debian operating system which provides stable and efficient operation, making it well suited for the demands of long behavioral experiments and running on low-cost hardware such as the Raspberry Pi.

All software is open-source and available on GitHub.¹ The VideoAnnotate software as well as data files and R code to perform analysis and generate plots is available on GitHub.² Both technical and end user documentation is provided including detailed software installation and troubleshooting for commonly encountered problems.³ Hardware schematics and wiring diagrams are available in the online documentation. An overview of how to use the PiE system is also available as a YouTube video.⁴ The PiE system has been in development since it was first used in 2017 (Cudmore et al., 2017). The Cudmore lab is actively developing the functionality and is dedicated to the maintenance of the PiE system.

We implemented custom software to control all aspects of a behavioral experiment. This software runs on a Raspberry Pi computer and once the hardware is configured can be installed with simple to use installation scripts. Implemented with a client/server architecture with a backend written in Python, a webserver (Flask), and a frontend web interface written in Javascript. We provide a point and click graphical user interface (GUI) that can run in any web-browser including on desktops, tablets, and smart-phones. All aspects of the software can be configured and controlled through a RESTful application-programming-interface (API) using Uniform Resource Locators (URLs) directly from a browser or with a programming language such as Python, R, or Matlab. Video acquisition to save files was implemented with the PiCamera Python package and video streaming was achieved using the UV4L software. Video files are initially saved in the h264 format and automatically (and optionally) converted to the mp4 video file format. A half-hour video at 1296 by 972 resolution with 30 frames per second creates a file just under 3 GB. With this file size, approximately 42.6 h of video can be stored on a 256 GB USB drive.

One important consideration when acquiring video with a Raspberry Pi camera is the possibility of dropped frames during a recording. For our maternal behavior video recordings, this was not an issue as the behaviors are relatively slow and dropped frames do not affect our analysis or results. If one seeks to use this system to record fast behaviors such as whisker movements or forepaw reaching, then dropped frames could potentially impact the analysis and results.

The Swarm software is designed to provide a web interface for the control of all aspects of any number of behavior boxes equipped with a Raspberry Pi running the PiE server software. All functionality of one behavior box with a PiE server is included in the Swarm software. The Swarm software can be run on any remote (analysis) computer including Raspberry Pi, macOS, and Windows operating systems. The Swarm software connects to individual PiE servers over the internet or a local area network (LAN) and itself provides an easy-to-use web interface.

The Swarm software will automatically (and optionally) copy recorded video files from the Raspberry Pi to a remote analysis machine for further analysis and safe storage. Thus, video can be recorded continuously for an arbitrary amount of time, regardless of the local storage limitations on a Raspberry Pi. In addition to full experimental configuration and control, the Swarm software allows experimenters to review recorded files and view real-time live video streams. With this design, fully remote control and review is possible, thus reducing potentially confounding experimenter interaction with animals and allowing experiments to be controlled from remote locations such as other labs, while at home, or even abroad.

Once installed on a local or remote computer, the Swarm software runs a web-server (Flask) and provides a point-and-click GUI in a browser (JavaScript). All communication with individual PiE servers is through their built-in RESTful API.

We have implemented desktop software, called VideoAnnotate, to perform behavioral analysis. Importantly, VideoAnnotate will work with any video, it is not dependent on video acquired with the PiE server software. At its core, VideoAnnotate provides video playback including play, pause, playback rate, and random access to any time within the file. As video is reviewed, the keyboard is used to indicate the start and stop of any number of different behavioral events. Point behavior event markers are also provided. All events within a video are presented in an intuitive GUI for review and curation. For example, clicking on an event will play the video corresponding to just that event.

A key design feature of the VideoAnnotate software is that it includes a randomization algorithm to reduce experimenter bias. This also allows random sampling of longer videos to reduce analysis time while not losing statistical power. In addition, this randomization is important because behavior often depends on the sequence of events such as ‘just placed in a cage’. By presenting the video in a random sequence of Chunks (see below) to the analyzer, we reduce bias as the analyzer does not know the sequence of events they are viewing. Once analyzed, all VideoAnnotate events are saved as a comma-separated-value (CSV) text file that can easily be opened and further analyzed with custom analysis scripts in almost any language.

As an example, we divided video files for our behavioral experiments into three Pieces of 10 min each, then 10 Chunks of video (10 s/Chunk) from each Piece were randomly selected, to total 30 Chunks of behavior video analyzed per 30-min recording. With this system, a 30-min behavioral video can be analyzed blind, while only having to score 5 min of video. These parameters are all user-configurable in the VideoAnnotate desktop application.

Lactation day 0 (LD0) was demarcated as the first day live pups were seen in the cage by 12 PM. Behavior experiments were conducted on Lactation Days 1, 3, and 5 (LD1, LD3, LD5) during the first hour of the light cycle from 09:00–10:00 AM. Immediately prior to the Maternal Behavior Test, the mouse dam was removed from the cage and placed in a clean separation chamber for 15 min. The cage was then cleaned, and food and water removed for the assay. Just before the start of the assay, the pups were returned to the cage and placed in the three corners opposite a new nestlet. The video recording was begun at the dam’s moment of return to the cage, which commenced a 30-min video capture period.

The 30-min Maternal Behavior Test video was scored blind, consistent with standard maternal behavior testing (Numan and Callahan, 1980; Matthews-Felton et al., 1995; Felton et al., 1998; Nephew and Bridges, 2011; Wu et al., 2014). Nine parameters were scored as either present or absent in each 10-s randomly presented video Chunks as detailed in the video annotation software section. These were: all pups retrieved and grouped, all pups nested, pup retrieval, pup interaction (sniffing, anogenital licking), crouching over pups, resting with pups, nest building, solo activity, solo rest. Three additional parameters were scored by viewing the entire 30-min Maternal Behavior Test as a whole: delay to initiate first pup retrieval, delay to complete pup retrieval, and nest score at 30 min. Nests were scored 0 to 4 as follows: 0, no nest attempted; 1, poor nest, not all of the nesting material was used, lacks structure; 2: fair nest, all the nesting material was used but the nest lacks structured walls; 3, good nest, all nesting material was used and the nest has low walls; 4, excellent nest, all the nesting material was used and the nest has high structured walls (Numan and Callahan, 1980).

All behavior data was analyzed using the Wilcoxon Rank Sum Test, a non-parametric test of distribution differences in continuous data between two groups.

This software runs on an individual Raspberry Pi computer and provides both a web- and file- server. Throughout the development of this system, we performed extensive testing and modifications to ensure it performs reliably and that it captures the general needs of a remotely controlled behavioral experiment. The provided web interface allows all aspects of the behavioral data acquisition to be remotely controlled (Figure 3A). This includes starting and stopping of video recording as well as live streaming of the video feed. This also includes manual control of the daytime (white) and nighttime (IR) LEDs as well as the fans installed in the box for environmental control. To allow semi-autonomous data acquisition, the web interface provides control of the parameters for video acquisition and environmental control (Figure 3B). This includes parameters for video acquisition such as video recording duration and the number of repeats, as well as video acquisition resolution and frame-rate. In addition, environmental variables can be configured in this web interface including the timing of light/dark period transitions.

As video and sensor data is acquired, it is saved locally on the dedicated Raspberry Pi computer attached to each behavior box. The PiE server software provides additional web interfaces to monitor this information and download it to a remote analysis computer. This includes a web interface to download recorded video files and textual logs of sensor data such as temperature and humidity (Figure 4A). Finally, the web interface provides plots and tabular logs of sensor data such as temperature and humidity (Figure 4B). This is critically important to monitor a behavior box remotely and can be used during offline behavioral analysis to determine if a disruption in these parameters occurred.

The PiE server software implements a web-based API to provide access to all its functions. This is called a RESTful API and basically provides a set of URLs to get and set all data and features of the provided web GUI. This includes an API to, for example, start and stop video recording, control actuators such as LEDs and fans, and to download acquired data. With this API, others can extend the capabilities of the PiE server software by creating customized scripts to implement novel behavioral data acquisition and control of the sensors and actuators included in the behavior box.

Once an individual behavior box is configured and controlled with the PiE server software, a key feature of our pipeline is to extend this functionality to any number of behavioral boxes. This is achieved with the Swarm software that allows full control of any number of PiE servers to be remotely controlled (Figure 5A). The Swarm software provides the same control of individual PiE servers across any number of behavioral boxes. It allows for the video recording to be started and stopped, and for daytime and nighttime lights to be set to auto or manual. The Swarm software also provides a surveillance interface where the video recording and live feed of each behavior box can be monitored (Figure 5B).

A key feature of the Swarm software is its ability to copy all raw data from any number of PiE servers to a remote analysis machine. This can be done while other experiments are running on the PiE servers and has an alarm-clock feature such that files can be automatically transferred at a given time or times per day. This is a requirement of our behavioral pipeline as just using 8 behavior boxes, we found manually collecting all the raw video data was tedious and error prone.

For the analysis of the raw video recorded and organized by the PiE server and Swarm software, we provide a desktop GUI for video annotation and scoring (Figure 6A). Unlike the PiE server software, this is designed to run on an analysis computer such as a macOS or Microsoft Windows operating system. At its core, this software interface provides all the expected video player capabilities including playing and pausing a video and random access to any frame of the video.

The video annotation software is designed for the manual scoring of events within a recording. As the video is played, keyboard shortcuts allow the user to mark the beginning and ending of different behavioral events. This interface also allows the user to mark single time point behavioral events. All events are tallied and displayed in a GUI resembling a video editor interface.

Because behavioral experiment analysis has the potential for experimenter bias, the video annotation software includes a randomization algorithm (Figure 6B). For example, if the user has a 30-min recording, this recording can be split into a desired number of Pieces (for example, 3 Pieces with a Piece duration of 10 min). VideoAnnotate then randomly selects an equal number of Chunks from within each Piece (for example, 10 Chunks per Piece with a Chunk length of 10 s). This allows for a total of 30 Chunks, with 10 Chunks from the first, middle, and last 10-min of the 30-min assay. The Piece functionality is important for assays such as Maternal Behavior Testing, where certain behaviors, like pup retrieval or having all pups nested, are not equally likely to occur at the beginning or end of an assay. The three Pieces prevent the random sampling from randomly introducing a bias by selecting many Chunks toward the beginning or end of the assay. Though for assays where the behaviors scored are equally likely to occur at any point in the assay, Piece number should be set to “1” to nullify the Piece function. The user then proceeds to scoring, where VideoAnnotate presents each Chunk (pooled across all animals and experimental days the user wishes to score together), in a random order. With this system, videos of 30-min duration can be scored in as little as 5 min. Finally, experimenter bias is reduced because the scoring is done in a random order and the scorer is blind to the experimental day, individual animal, and experimental group, as well as to the unique context of each Chunk within its total recording.

All behavioral events are saved as CSV text files which facilitate additional analysis and plotting by end users in any number of scripting languages. Importantly, the video annotation software does not require video to be acquired with the PiE server, it is designed to open, display, and analyze several common video formats including mp4.

To demonstrate the flexibility of the PiE system, we collected and analyzed home cage data using maternal behavior tests, a complex social behavior requiring simultaneous scoring of many behaviors (for a review of mouse maternal behavior, see Weber and Olsson, 2008).

Consistent with other work examining innate differences in maternal behavior in the DBA/2 J and C57BL/6 inbred mouse strains, our data show significant differences between mouse dams of the two strains across a number of behaviors assayed (Figure 7). These measures included the delay to initiate pup retrieval (Figure 7A), scored video Chunks in which all pups were gathered in the nest (Figure 7B), and the nest score at the end of the assay (Figure 7C). These tests illustrate the different uses of the video annotation software including the analysis of complete video files (Figure 7A), analysis of randomly selected Chunks of video (Figure 7B), and analysis of the end of each assay (Figure 7C).

Across all three behavioral measures, DBA/2J dams show reduced maternal responsiveness as compared to C57BL/6 dams. DBA/2J dams take longer to initiate pup retrieval, they have all of their pups retrieved and nested in fewer of the video Chunks scored, and they build weaker nests than C57BL/6 dams. These findings are consistent with past findings comparing DBA/2J and C57BL/6 mouse strains (Carlier et al., 1982; Cohen-Salmon et al., 1985; Brown et al., 1999; Shoji and Kato, 2006).