IntelliCage: the development and perspectives of a mouse- and user-friendly automated behavioral test system

Abstract

IntelliCage for mice is a rodent home-cage equipped with four corner structures harboring symmetrical double panels for operant conditioning at each of the two sides, either by reward (access to water) or by aversion (non-painful stimuli: air-puffs, LED lights). Corner visits, nose-pokes and actual licks at bottle-nipples are recorded individually using subcutaneously implanted transponders for RFID identification of up to 16 adult mice housed in the same home-cage. This allows for recording individual in-cage activity of mice and applying reward/punishment operant conditioning schemes in corners using workflows designed on a versatile graphic user interface. IntelliCage development had four roots: (i) dissatisfaction with standard approaches for analyzing mouse behavior, including standardization and reproducibility issues, (ii) response to handling and housing animal welfare issues, (iii) the increasing number of mouse models had produced a high work burden on classic manual behavioral phenotyping of single mice. and (iv), studies of transponder-chipped mice in outdoor settings revealed clear genetic behavioral differences in mouse models corresponding to those observed by classic testing in the laboratory. The latter observations were important for the development of home-cage testing in social groups, because they contradicted the traditional belief that animals must be tested under social isolation to prevent disturbance by other group members. The use of IntelliCages reduced indeed the amount of classic testing remarkably, while its flexibility was proved in a wide range of applications worldwide including transcontinental parallel testing. Essentially, two lines of testing emerged: sophisticated analysis of spontaneous behavior in the IntelliCage for screening of new genetic models, and hypothesis testing in many fields of behavioral neuroscience. Upcoming developments of the IntelliCage aim at improved stimulus presentation in the learning corners and videotracking of social interactions within the IntelliCage. Its main advantages are (i) that mice live in social context and are not stressfully handled for experiments, (ii) that studies are not restricted in time and can run in absence of humans, (iii) that it increases reproducibility of behavioral phenotyping worldwide, and (iv) that the industrial standardization of the cage permits retrospective data analysis with new statistical tools even after many years.

Keywords: animal welfare; automated behavioral analysis; comparative and evolutionary neuroscience; ethology and ecology; home-cage testing; marmoset (Callithrix jacchus); reproducibility; standardization.

Figure 1

From outdoor feeder boxes in Russia to a tool in the laboratory. The conceptual origin of IntelliCage were feeder boxes placed in the forest or in outdoor pens for recording and controlling the patrolling of wild and feralized mice (Lipp and Wolfer, 2013). (A) Set-up of feeder boxes to study natural learning in wild mice. (B) Closer view of a feeder box in the forest. Experiments in the forest failed because feeder boxes were partially destroyed by roaming bear cubs smelling the mouse food. (C) Outdoor pen (20 × 20 m) in the Russian field station Chisti Les containing eight feeder boxes and a central computer controlling the boxes. (D) Closer view of an automated feeder box recording entries of mice tagged with transponder chips. Food was only delivered upon a new visit. (E) First prototype of an IntelliCage operating on MS-DOS, constructed by Alexei Vyssotski and Giacomo Dell’Omo. (A,B) Courtesy of Patricia D’Adamo.

Figure 2

An eye-opening study comparing the spatial learning of trkB mutant mice in the lab with the behavior in a semi-naturalistic situation (A) trkB mutant mice were tested for water maze learning and showed a severe impairment, mostly visible in the homozygous mutants, while the heterozygous mice behaved like the controls. Modified after Minichiello et al. (1999). (B) In the same study, hippocampal slices had shown intermediate LTP values for the heterozygous animals. Modified after Minichiello et al. (1999). (C) Outdoor patrolling behavior of the same trkB mutant line in the Russian field station Chisti Les over 21 days. The mice had to patrol 8 boxes to obtain maximal food reward. Every third day, patrolling the loaded boxes was not necessary as food was placed inside the shelter, offering an opportunity for a one-day place reversal learning. Notably, the homozygous mutants ignored this opportunity, which was instead regularly exploited by the wildtype controls. Intriguingly, the heterozygous mutants felt in-between the groups, as would have been expected from the LTP data. Modified after Vyssotski et al. (2002a).

Figure 3

Modern IntelliCage since 2006. (A) Complete view of the system integrated into a commercially available polycarbonate rat cage (20.5 cm high × 58 × 40 cm at top, 55 × 37.5 cm at bottom, Tecniplast 2000, Buguggiate, Italy). The entire cover plate with the corners can be lifted for cleaning or exchanging the cage body. The electronic control unit integrates light and temperature sensors. It connects with up to 8 IntelliCages running the same or different programs. (A) Combination of 4 standard Tecniplast mouse-houses permits preferential huddling of mice. (B) Inside view of the conditioning corner faced by the mouse when advancing through the ring antenna. Walls, nose-poke-holes and grids are made from stainless steel. (C) Outside view of the conditioning corner. The sliding doors are moved by means of a cogwheel-operated mechanism preventing squeezing of the mouse nose. Part of the operating circuitry is integrated in the blue plastic container.

Figure 4

Designing simple and complex tasks in the designer program with a graphic user interface (GUI). (A) Graphic design for spatial learning. This requires a simple sequence: a specific corner is assigned to one or several animals. Upon identification of an assigned individual, a timer is activated and the door leading to the drinking nipple opens. The door closes after a defined period or after the mouse has left the corner. (B) Graphic design for discount delay-conditioning. This procedure measures how well mice can solve a conflict between easy access to plain water and the need to wait a defined time for obtaining a sucrose/saccharine reward. Upon entering a corner, the mouse is identified, two timers are activated according to the learning progress of the mouse, and an LED signal is activated to mark the beginning of the procedure. After having made a nose-poke choice towards one of the bottles, the system will deny access to the sweetened bottle if the nose-poke is too early. The recording of the animal’s actions indicates its ability to inhibit learned local movements, yet also a sense for time at short-term scales. (C) Data example of simple spatial programming: MHB-Cre:DTA mice carrying a mutation causing postnatal ablation of medial habenular cells are impaired in their ability of spatial reversal learning, however combined with other behavioral deficits (Kobayashi et al., 2013). (D) Strain comparison using discount-delay conditioning. C57BL/6 and DBA/2 mice typically differ in their ability of controlling behavior under conflicting situations (Wolfer et al., 2012). Saccharine preference was established rapidly in both strains when there was no imposed delay. Upon increasing waiting times, DBA/2 mice quickly switched to drink plain water, while C57BL/6 mice maintained a preference for saccharine, also with increasing waiting times, but eventually switched to the plain water solution. Presenting immediate reward re-established the saccharine preference in both strains. Example set up by Elisabetta Vannoni.

Figure 5

Ongoing information provided on-screen by the controller. (A) The default controller screen just shows the activity state of sensors and actors. Yet, the menu provides numerous opportunities to call the actual state of the data in both alphanumerical and graphic form. The graphs can be selected for single animals, subgroups, or all mice in the cage. (B) Quick monitoring of corner preferences by individual mice. (C) Continuous monitoring of behaviors considered as errors or success permits to recognize developing trends resulting from treatments. The screen shows the mean cumulative error rate in reversal learning as observed in a group of mice with hippocampal lesions. (D) Individual learning or preference curves can also be plotted, e.g., for saccharine preference. Note that the final mean score of the animals in the cage is around 900, because some of the mice ignored or even avoided saccharine. Also note that every experiment can be graphically replayed (from archive files), for individuals or for treatment groups, by using selectable time windows from seconds to weeks, thus recognizing the development of odd behavior patterns of treatment groups or strangely behaving animals.

Figure 6

Automated statistical analysis of IntelliCage data by FlowR. (A) Graphic interface for creating a workflow connecting various R scripts for simple or complex statistics. The program reads in archive files from IntelliCage experiments, leaving the original data intact. (B) The extracted data are read-inand analyzed by pre-assembled R-routines including publication-ready graphic displays and statistical analysis in PDF format. Shown here are simple bar graphs, and 3D multidimensional scaling and principal component analysis. The analysis requires a minimum of computer experience and knowledge in R or other statistics programs. (C) Chronometric analysis including simple activity plots, cosinor analysis and vector rose plots of acrophases for rapid comparison of groups. Picture provided by courtesy of XBehavior.

Figure 7

Early validation studies of IntelliCages in 2003 and 2004. (A) Estimating robustness of expected differences in a largely uncontrollable environment in a Russian field station. Two old-type IntelliCage were used to run a pilot study with cranially irradiated mice, but the information provided by the radiologists was lost due to the untimely death of Nada Ben Abdallah who was running the study. However, an IntelliCage data archive file could be recovered (by Pascal Zinn) and permitted to run a data analysis using the stored information only. There were clearly some differences between treatment groups that cannot be interpreted, however. On the other hand, the data demonstrate that IntelliCages can reveal significant behavioral differences between treatment groups even in noisy environments. The same cages were also used in that year to study differences between wild voles and mice (Galsworthy et al., 2005). (B) IntelliCages revealing extremely subtle transgenerational effects. Two IntelliCages housed 20 female DBA mice, 9 controls and 11 animals whose grand-grand-fathers (3 generations ago) had received postnatal thyroxine injections that changed brain and body features that were transmitted paternally (yet variably) over 3 generations of dam-raised DBA/2 mice. For details see Vyssotski et al. (2002b) and Vyssotski (2011). The observed behavior was how frequently the mice were visiting corners where they received air-puffs, which was rarely observed in other studies. The cages were situated in a non-climatized laboratory. Given the heat of summer 2003, we suspect that some mice were actively seeking the air-blows, which in this context provided a rewarding cooling. Modified after Lipp (2005) and Lipp et al. (2005). (C) Pooled presentation of non-systematic IntelliCage tests with knockout mice provided by collaborators and not being used in conventional tests, including a few mice with hippocampal lesions available for pilot studies. CREB/CREM double mutants and mice with knockout of the mineralocorticoid receptor were provided by Peter Gass and Thomas Lemberger in Heidelberg. Data were presented repeatedly by Lipp (2006) and Wolfer et al. (2012).

Figure 8

Influential studies giving rise to different directions of IntelliCage use. (A) The central amygdala (CE) shows activation of neurons as indicated by c-Fos-Expression when mice in an IntelliCage had to learn to visit a corner to obtain sweet reward. Their companions in the same cage, that had free access to sucrose solution in all corners, did not show activation of the central amygdala, indicating that the c-Fos activation was not due to a gustatory sensation. In a second cage, one group of mice had access to plain water in all corners, while their companions received air-puffs after having visited their preferred corner (as identified during the adaptation period). This study showed that the IntelliCage can provide unique testing procedures for dissecting the involvement of neuronal structures in motivationally different tasks during similar learning requirements tasks. Figure redrawn after Knapska et al. (2006); see also there for methods. CEm, central nucleus (amygdalae) medial part; Ld, lateral nucleus (amygdalae) dorsal part; P-Av, place avoidance task; P-Pref, rewarded place preference learning. (B) Mutations of the gene Arhgef6 in humans are known for causing intellectual disability. The corresponding mouse model underwent a series of behavioral tests including IntelliCage tests. In a simple place learning test, mutants were more active but learned the simple task as the wildtypes. Complicating the task by introducing left/right differences in the corners was associated with increased activity of the mutants, associated with higher error rates. Water maze learning showed modest differences, but the radial maze did not. Figure redrawn from Ramakers et al. (2012). (C) Two paradigms of behavioral flexibility based on learning a switching routine for obtaining water. The initial task was devised by Endo et al. (2011) and included learning a shuttling routine between diagonally opposite corners. After several sessions (usually days), the position of the active corners are switched and the mice must relearn the new positions, thus providing a measure for spatial reversal learning. The error rates after a new reversal are initially high, but gradually decline after every reversal, providing a measure for rule learning. As this protocol is time consuming, new versions were developed by one of us (Toshihiro Endo), based on a self-paced reversal (SPRT), usually after a mouse has reached a criterion between 30% correct responses. This less tedious (automated) procedure is particularly suitable for older animals and different wild species hard to test in common behavioral laboratories due to handling difficulties. For example, wood mice (Apodemus sylvaticus) learn this procedure easily as compared to bank voles (Clethrionomys glareolus). The power of this IntelliCage approach is that higher cognitive abilities of rodents can be assessed subtly and without stress. Graphs were modified after Endo et al. (2011) and Jörimann et al. (2023).

Figure 9

Use of IntelliCage systems in behavioral research. (A) Cumulative plot of papers dealing practically with or describing IntelliCages since 2005. The year 2023 includes publications at 15 October 2023, including some reviews and discussion papers. Searching criteria in Google Scholar (screening the entire paper) were: presence of the keyword “IntelliCage” together with (i) Primary journal articles that use “IntelliCage” as part of the methodology, (ii) Review papers/textbook chapters only if they focus on rodent behavior, (iii) Preprints (bioRxiv), (iv) Articles in languages other than English. Anything else is not included, for example conference abstracts, theses, articles that only mention “IntelliCage” without actual use or specific focus on it, etc. A complete list of papers, ordered alphabetically or chronologically, can be found in Supplementary References. (B) Proportions of IntelliCage papers classified according to scientific fields. For a description of the main disease classifications, see Table 1.

Figure 10

Male–female differences and inter-laboratory comparisons. (A) Significantly less errors in spatial learning of aged females (n = 15) as compared to males (n = 14), *p < 0.05, means and S.E.M. After 7 days of habituation to the IntelliCage environment with free access to all corners for drinking, water delivery occurred for 4 days only in the corner opposite to the one preferred during the adaptation period. (B) Reversal phase: during days 12–14, all mice underwent a 3-day spatial learning reversal where the only corner available for drinking was the preferred one during the adaptation phase (Berry et al., 2012). (C) Grand average of 288 mice of different strains (C57BL/6, DBA/2, F1 B6xD2) as observed for nocturnal activity in four different laboratories across Europe, indicating a highly significant strain effect (Krackow et al., 2010). (D) Laboratory-specific results show clear differences among the single strains at Karolinska Institutet (NKAR) in Stockholm, Evotec in Hamburg, Istituto Superiore di Sanità (ISS) in Rome; and University of Zürich (UNIZH), but the laboratory-specific comparisons revealed in all cases comparable strain differences. For details of the statistical design, see Krackow et al. (2010).

Figure 11

IntelliCage add-ons that can be controlled by the IntelliCage system. (A) Overview of all add-ons constructed under the FP6 programs “Intellimaze” and “Noveltune.” Picture provided by the University of Zürich. (B) Animal gate permitting or denying access to different test systems. The gate contains three doors regulating to-and-from traffic to external devices, supported by air-puffs driving away dawdling mice. The first compartment contains the RFID reader. For a figure showing an outside home cage connected to the IntelliCage for auditory testing, see Kahnau et al. (2023a,. The add-on most successfully used up-to-now is the audiobox.

Figure 12

RFID floor plate (Phenovance) fitting exactly under the IntelliCage, comprising 50 × 50 mm antennas capable of recording proximity and trajectories of moving mice even when multiple mice are on the same antenna. Although the RFID floor plate requires a different type of transponder from that used in IntelliCage, these two types of transponders do not interfere with each other. Picture courtesy by Phenovance.

Figure 13

IntelliCage for rats. (A) Larger housing space required adaption of the IntelliCage system, and (B) adaptation of the corner system. The software of the mouse IntelliCage runs also with the rat system. Picture provided by TSE-Systems International.

Figure 14

IntelliCage system for marmosets (Callithrix jacchus) as constructed by Seico Benner and Toshihiro Endo. (A) Modified rat corner. (B) Animal-friendly marmoset housing at the Yamasue Laboratory, Department of Psychiatry, Hamamatsu University School of Medicine (Hamamatsu, Japan). The position of a rat corner is marked by a red circle. Other corners can be placed everywhere, provided cabling and backside of the corner are protected. Mounting at the wall is the easiest solution. (C) Marmoset entering the tubular RFID identifier, which blocks access to others while one subject is working inside. (D) Inside view of the conditioning corner corresponding to the arrangement in Figure 4G. (E) Outside view of a corner, with the operating gate blocking access to the nipples. (F) Screenshot of the standard IntelliCage controller during training of monkeys showing activated signals for presence and nose-poke. (G) Set-up of a simple fixed-ratio (FR) discrimination learning task during which the monkey has to nose-poke or to touch the light barrier in front of the closed barrier several times to open the barrier. The position of the rewarded site is signaled by LEDs. The programming was done with the designer program. (H) Proof-of-principle pilot study demonstrating differential learning by 8 marmosets under fixed ratios (FR1–FR4, the latter indicating that 4 pokes/touches are required for gate opening). Data published by Benner (2022). Videos showing the various actions are provided in Supplementary Videos S1, S2.

Figure 15

Comparison of conventional (linear) multivariate statistics and machine learning. (A) Standard canonical discrimination analysis of mouse learning in the IntelliCage showing decent separation of hippocampally lesioned mice (HIPP) from controls in the same cage, but considerable overlap between controls and prefrontal lesions (PFC). Data from Voikar et al. (2018). (B) Random forest analysis: same data analyzed by the random forest algorithm, which eventually could separate the three groups much better. Data presented by Krackow and Lipp (2023).