Towards large scale automated cage monitoring – Diurnal rhythm and impact of interventions on in-cage activity of C57BL/6J mice recorded 24/7 with a non-disrupting capacitive-based technique

Background and aims Automated recording of laboratory animal’s home cage behavior is receiving increasing attention since such non-intruding surveillance will aid in the unbiased understanding of animal cage behavior potentially improving animal experimental reproducibility. Material and methods Here we investigate activity of group held female C57BL/6J mice (mus musculus) housed in standard Individually Ventilated Cages across three test-sites: Consiglio Nazionale delle Ricerche (CNR, Rome, Italy), The Jackson Laboratory (JAX, Bar Harbor, USA) and Karolinska Insititutet (KI, Stockholm, Sweden). Additionally, comparison of female and male C57BL/6J mice was done at KI. Activity was recorded using a capacitive-based sensor placed non-intrusively on the cage rack under the home cage collecting activity data every 250 msec, 24/7. The data collection was analyzed using non-parametric analysis of variance for longitudinal data comparing sites, weekdays and sex. Results The system detected an increase in activity preceding and peaking around lights-on followed by a decrease to a rest pattern. At lights off, activity increased substantially displaying a distinct temporal variation across this period. We also documented impact on mouse activity that standard animal handling procedures have, e.g. cage-changes, and show that such procedures are stressors impacting in-cage activity. These key observations replicated across the three test-sites, however, it is also clear that, apparently minor local environmental differences generate significant behavioral variances between the sites and within sites across weeks. Comparison of gender revealed differences in activity in the response to cage-change lasting for days in male but not female mice; and apparently also impacting the response to other events such as lights-on in males. Females but not males showed a larger tendency for week-to-week variance in activity possibly reflecting estrous cycling. Conclusions These data demonstrate that home cage monitoring is scalable and run in real time, providing complementary information for animal welfare measures, experimental design and phenotype characterization.


S1 Fig Response to lights on
The response to lights on is identified as follows (see Fig. S1): • Smooth the minute-based activity time series (averaged across all 12 electrodes) with a low pass filter (moving average of 30 minutes) • Find the peak of the time series (within +/-3 hours from lights on) • Find contiguous blocks of minutes whose activity is larger than half of the peak • If there are less than 5 minutes between two blocks, consider them as a single block • Pick the block which contains the maximum value (block 2 in the example in Fig. S1) • Set the response duration as the distance between the extrema of the identified block (points C and D in the example in Heat maps showing average global activity of four cages with male C57B/6J mice, kept 5 to a cage, during 4 consecutive weeks (day 1-28); conversion of activity to color according to scale to the right. The basic pattern of day and night time activity levels are the same as for the cages with weekly cage-change (cf. S2 Figand Fig 2). Following the cage-change day the day and night activity pattern reach a level that is essentially maintained as a basal undisrupted activity patter until next cage-change. Cage-change day 2 and day 16 have been indicated with an asterisk and white vertical line indicates transition to night time while left and right border of the heath map correspond to day break.

Digital ventilated cage TM (DVC TM )
Digital ventilated cage (DVC TM ) is a commercially available system designed to collect information from individual ventilated cages (IVC) directly at rack level. DVC TM system builds up on the top of a standard IVC rack by resorting to a set of sensors, which are able monitor cage conditions as well as animal activity. All sensors are placed outside the cage and do not require any human intervention to collect data, nor impacting conventional IVC cage operations. The key DVC TM sensing component is an electronic board installed below each IVC cage and mechanically connected to the rack. This electronic board is composed of 12 planar electrodes connected to an integrated circuit (proximity sensor) that continuously measures their electrical capacitance. Since electrical capacitance is generally influenced by the matter present in each electrode's surrounding, the capacitance measured by the proximity sensor depends on the presence of, e.g., bedding material, water, animals, etc. By properly tracking and analyzing capacitance variation over time, it is possible to detect animal activity as well as bedding humidity condition (due to e.g., latrine). The proximity sensor measures the electrical capacitance of each of the 12 electrodes 4 times per second (i.e., every 0.25 seconds), thus generating a total of 48 samples per second per board (cage). In addition to the 12-electrodes board, the DVC TM system also embeds IR sensors at each cage position (connected to the rack laterally with respect to the cage) to monitor the presence of food and water.
All electronics components of DVC TM are connected via wires to a dedicated computer, referred to as DVC TM master, which provides both power and data connection. The DVC TM master collects raw data coming from all cage positions, with the option of both transferring them directly to a web-based software application or to a dedicated storage device for later data processing purposes.
It is worth to mention that, DVC TM is a big-data oriented system that is designed to collect, store and process data from thousands of cages simultaneously (i.e., high throughput). In addition, DVC TM system requires neither manual intervention to set up the sensing infrastructure, as this is inherently connected to the rack and always on, nor maintenance or removal of components as these have been designed to be safely sanitized and/or inserted into autoclaves while still connected to the rack.

Computation of the activity metric
For the scope of this study, it was decided to measure animal activity in terms of perturbation of electrode capacitance over time. As mentioned above, the rational is that, when an animal moves closer to an electrode or away, the measured capacitance varies accordingly, whereas if no movements occur, no capacitance changes (or very small due to noise) are recorded. The DVC TM animal activity algorithm used throughout the paper translates these observations into metrics as follows. For any given DVC TM board corresponding to a TMspecific cage position, let ( ) be the measure of the th electrode at discrete time , with ∈ {1,2, … ,12} and ∈ {0,1,2, … }. Recall that, a sample is collected every 0.25 secs. Capacitance perturbations are then evaluated via one-step finite difference ( ) = ( ) − ( − 1), with > 0. The activity indication at the th electrode at time is determined by comparing the finite difference ( ) against a fixed threshold as follows ( ) = 1(| ( )| ≥ ), where 1( ) is an indicator function for event , 1( ) = 1 if event is true and 1( ) = 0 if event is false. The threshold is conveniently selected based on noise magnitude considerations and it is assumed fixed and equal to = 2 throughout the whole paper. Depending on the needs, a summarizing activity metric can be built up upon ( ) by summing up activations within a given time interval, and possibly by averaging across a set of electrodes. More specifically, let 1 and 2 be the starting and ending time of the required time interval, and let be a set of electrodes of interest, then the average activity metric is calculated as: ( 1 , 2 ) = 1 ( 2 − 1 )| | � � ( ) 2 = 1 ∈ . Note that, activity metric ( 1 , 2 ) is mainly used in this paper to measure activity within 1 minute intervals for each minute of the day ( 1 , 2 are selected to cover 1 minute), or for lights ON/lights OFF periods (where