The authors engineered a compact, programmable operant device that integrates a touchscreen, retractable levers, and a water-reward nozzle into individually housed mouse home cages. The manuscript describes two hardware variants: version 1 couples an external laptop with an IR-equipped touchscreen sensor for nose-poke detection, while version 2 replaces the laptop with a Raspberry Pi 4 and uses IR camera–based image recognition to detect mouse actions. Optional modules—an operation monitor, lever modules, and an AirBar IR sensor—are available to tailor the set-up to different task demands. The project is fully open-source; build instructions, software, and raw data are accessible on GitHub and Zenodo as reported by the authors. The authors provided a per-unit cost range (US $380–660), noting assembly using off-the-shelf parts and 3D-printed components.
Habituation began with daytime transfer into the chamber and a scheduled water-withholding period until 9:00 p.m., producing an initial 9–12 hour water restriction before ad libitum access overnight. From day two, sessions ran daily with a midnight start signaled by a green LED and a brief water delivery. Sessions continued until animals reached a predetermined number of water accesses or until the programmed session duration elapsed. The manuscript states cumulative water-access duration was monitored throughout sessions to ensure opportunity for drinking; however, the actual water volume consumed per mouse was not directly measured. The authors monitored body weight as part of welfare checks and compared paired pre- and post-testing weights in a subset of six wild-type mice, but the source did not provide detailed numerical outcomes for those comparisons in the text provided here.
The delayed non-matching-to-position (DNMTP) protocol used a three-aperture mask aligned to illuminated rectangles on an LCD. Training progressed through three blocks: VD1 (initial visual detection with stimulus rewards when mice failed to respond), VD2 (visual discrimination with punishment for incorrect responses), and a DNMTP pre-acquisition stage to teach rule structure before the full DNMTP task. In DNMTP trials the sample panel was presented, followed by a delay phase and then a choice phase in which both lateral panels were illuminated; mice had to touch the panel opposite the sample location to receive a 2-second water reward. Acquisition criteria were defined (for example, reaching a set success rate across consecutive days), after which delay durations were increased. To avoid fixed timing adaptation, two delay conditions were randomized trial-by-trial using a common seed across animals. The authors reported that most wild-type mice reached criteria within one to two weeks.
The authors also handled prolonged actual delays—occasionally lengthened by behaviors such as eating or sleeping—by performing sensitivity analyses restricted to trials with actual delay duration below a prespecified threshold (programmed delay × 1.3 + 10 s). Those delay-restricted analyses were treated as sensitivity checks rather than independent confirmatory tests.
The two-choice spatial discrimination (TCSD) task used a five-aperture mask and began with VD1/VD2 visual-discrimination training analogous to the DNMTP pretraining. In the TCSD block, two panels were illuminated simultaneously and the correct side was assigned randomly at trial start. Correct responses yielded 2 seconds of water access; incorrect choices triggered 5 seconds of chamber-light punishment. The correct-side assignment was reversed when the subject achieved more than seven correct responses in the most recent eight trials, enabling a measure of behavioral flexibility. Daily trials were limited by either a total-correct threshold (150 correct responses) or a 6-hour session cap.
The authors applied both DNMTP and TCSD to App NL–G–F knock-in mice, which carry three familial Alzheimer’s disease mutations in the APP gene without overexpression artifacts. Using Operant House, the investigators reported detection of working-memory deficits in these knock-in mice as early as 5 months of age. At the same age they observed subtle, spatially heterogeneous synaptic alterations in the prefrontal cortex, while conventional electrophysiological measures were described as largely preserved. The source text links these behavioral and anatomical observations but does not provide detailed numerical statistics within the section presented here.
The platform tests individual mice in their home cages, enabling training during the animals’ active phase and reducing experimenter transfers that can induce stress. Because each device accommodates a single animal, individual housing is an intrinsic design feature. To mitigate welfare concerns the system supports continuous monitoring via live camera feeds and recorded videos, and body-weight monitoring was implemented. The Operant House is positioned as a scalable, low-cost alternative to commercial systems, suitable for parallel automated testing across multiple individually housed subjects.
The authors explicitly acknowledge that individual housing is a limitation requiring careful animal-welfare oversight. The manuscript notes that actual water volume consumed per animal was not measured directly. Paired body-weight comparisons were performed in a small wild-type subset, but the present text does not report the numerical test results. Detailed parameter values for some experimental elements—such as exact per-trial reward volumes, or full statistical outputs from the App NL–G–F behavioral comparisons—are not included in the source excerpt and therefore are not reproduced here. The open repositories linked in the paper (GitHub and Zenodo) are cited as the locations for complete protocols and raw data, which readers should consult for additional methodological specifics and full datasets.