Automated monitoring of mouse feeding and body weight for continuous health assessment

Introduction

Accurate health monitoring of research animals is necessary to provide adequate care and avoid unnecessary suffering. In a recent survey of mice at a large research institution, “spontaneous mortality with unknown cause” was more common than all other adult morbidity types combined. ¹ Simply put, most mouse illnesses were not detected until it was too late.

Common approaches to routine health surveillance include visual observation, body condition scoring, ² and targeted assessment of experiment-specific outcomes.^3–5 These labour-intensive methods become less practical as population size increases and have other important limitations. First, many diseases manifest in ways that are difficult for human observers to detect. The classic response to infectious disease – mirrored in several other disease states – is a suite of behavioural changes collectively known as “sickness behaviour.”^6–9 Lethargy, reduced feeding and reduced grooming are virtually impossible to detect in brief health checks, particularly when conducted during the day (light phase), when mice are least active and often huddled together. ¹ An additional concern is the oft-repeated (although not empirically proven) claim that mice hide signs of pain or disease to avoid being targeted by predators (e.g. humans) or aggressive conspecifics.^1,3,9,10 Human observers often will not notice that mice are sick until their illness has conspicuous physical consequences, such as reduced body weight, low body condition score, or an unkempt coat. These subjective assessments also have the potential for inconsistency between observers.^5,11

Second, handling mice for visual inspection, palpation or weighing ² is typically aversive and provokes a physiological stress response,^12,13 perhaps doubly so for animals who are sick or experiencing pain or distress. ¹⁴ Handling may also negatively affect health and experimental outcomes, for example by dampening mice’s subsequent immune responses to a challenge ¹⁵ or otherwise altering results of behavioural assays;^16–18 this can be avoided by non-aversive handling techniques. ¹⁹

Therefore, routine health assessment should ideally rely upon non-stressful measurement – including at times when healthy mice are active – of parameters that are not prone to human error and are difficult for animals to hide. Various automated techniques allow continuous, objective, remote, handling-free observation of mice in their home cages, while they are unaware of being monitored.^8,20–23

Automated home cage health assessment can detect condition-specific symptoms and changes consistent with sickness behaviour in various veterinary and experimental contexts. In neurodegenerative disease models, this approach distinguished sick from control mice weeks before standard assessments did, finding differences in rest, locomotion, grooming and their circadian rhythms, as well as in motor learning and performance.^8,24–26 Automated monitoring revealed short-term decreases in locomotion and drinker visits in mice following traumatic brain injury ²⁷ and decreased locomotion and social connectedness in commensal mice with experimentally induced sickness behaviour. ²⁸ Automated video assessment readily detects reduced activity in post-surgical mice, where human observers instead rely on more obvious pain-specific behaviour (e.g. twitching or licking the affected area). ⁴

Here, we use automated home cage monitoring to detect changes in body weight and food intake, key signs of sickness behaviour that mice cannot voluntarily hide or alter and that can indicate deteriorating overall health. Sustained weight loss was predictive of imminent death in a rat cancer model, ²⁹ and decreasing body weight and food intake are integral components of many humane endpoints for laboratory rodents.^11,30,31 To our knowledge, automated body weight tracking has not previously been used for rodent health surveillance.

We developed an apparatus for automated, objective, accurate, continuous and non-invasive health monitoring of individual mice in group-housing. We hypothesized that a combination of body weight, feed intake and visual appearance could be used to monitor health. We predicted that automated assessment would show construct validity by successfully distinguishing mice with experimentally induced sickness from healthy controls. We predicted it would detect illness sooner than human observers conducting twice-daily assessments. Additionally, we opportunistically examined circadian fluctuations in feed intake and body weight.

Animals, material and methods

Experimental trials

We tested whether the automated monitoring apparatus could detect the effects of an LPS challenge. LPS is a component of the outer membrane of many pathogenic bacteria that reliably elicits sickness behaviour with systemic administration in mice.^7,34 We administered 1 mg/kg Escherichia coli 0111:B4 LPS (Sigma Aldrich St. Louis, MO, USA) in saline vehicle intraperitoneally to experimental animals after determining, in the pilot, that this sub-lethal dose would induce short bouts of illness.

The main experiment consisted of five trials in which a same-sex pair of mice lived in the apparatus and were monitored continuously. Each pair was allowed at least 2 days in the apparatus for acclimation and baseline data collection. Following this, at 1600 hours we injected one randomly assigned mouse per pair with LPS and the other with saline. We kept the pair in the apparatus for at least 1 full day following injection and then euthanized them using rising CO₂. We kept some pairs in the apparatus for longer periods before or after injection to collect further data (not reported here). As a humane endpoint, we euthanized LPS-injected mice immediately if they did not resume feeding after 24 hours or if recommended by veterinarians. We thoroughly cleaned and disinfected the apparatus between trials.

A human experimenter (an experienced mouse researcher) visually assessed each mouse twice a day (about 0900 and 1500 hours) through the cage walls and lid, starting in the afternoon 2 days before injection. The experimenter gave mice a body condition score (BCS) ² and a specific illness behaviour score (IBS) compiled from other assessments approved by Purdue’s IACUC^5,14,35 (Table 1). This experimenter identified mice solely by ear notch and was blind to treatment.

OPEN IN VIEWER

Table 1.

Illness Behaviour Score. Criteria were compiled from health assessment plans used in mouse and rat experiments monitoring illness and abdominal pain (LPS-induced and otherwise). Scores were summed across five criteria, with lower scores signifying better health.

Category	Score	Criteria
Movement / demeanor	0	Movement (change of spatial location) within 15 seconds of observation and/or interaction with cage-mate
	1	Movement within 30 seconds and/or interaction with cage-mate
	2	No movement nor interaction with cage-mate within 30 seconds
Eyes	0	Shiny, neither bulging nor sunken
	1	Possibly shiny, eyes beginning to bulge or sink
	2	Dull appearance, eyes are noticeably bulging or sinking
Coat (appearance)	0	Shiny, smooth, even coat
	1	Coat is rumpled in areas (not smooth), little shine
	2	Coat appears bristly, no shine, possible clumps / patches missing
Posture	0	No hunching / full body extension occurs (such as stretching upward to smell through lid)
	1	Some hunching (when motionless) / little full body extension
	2	Constant or near-constant hunching / no full body extension
Discharges	0	No discharge around eyes or nose
	1	Discharge noticeable for at least one of eyes or nose
	2	Discharge noticeable for both eyes and nose

Results

During the 24 hours preceding LPS or sham injections, mice obtained an average of (+/− SD) 169.7 +/− 63.7 (range 90–283) pellets in 811.6 +/− 567.4 (range 381–2274) separate visits to the feeder, 227.5 +/− 103.7 (range 121–404) of which yielded stable body weight readings. The longest time without a stable reading for any individual mouse was just over 4.5 hours, during a period of typical inactivity (afternoon, light period).

Prior to injection, feed intake and body weight showed marked circadian fluctuations, rising sharply at the start of the dark period, and being lowest during the light period. Time of day was the only significant factor affecting feed intake (F_11,88 = 8.17, p < 0.0001; Figure 2) and body weight (F_23,184 = 9.11, p < 0.0001; Figure 3). OPEN IN VIEWER

OPEN IN VIEWER

Figure 3.

Mouse body weight fluctuated with time of day over 24 hours pre-treatment. Weights increased at the start of the dark period (2000–0600 hours, shown by the black horizontal bar) and were significantly above average (h>) between 2100 and 0000 hours, and below average (h<) between 1100 and 1600 hours. The dashed line represents average body weight. Data points represent least square means for each hourly bin (n = 10 mice), plus or minus standard errors of partial residuals (obtained by removing the estimated effects of sex and individual baseline weight). Sexes are graphed together because their body weights did not differ significantly.

Feed intake declined precipitously after LPS injection, as shown by a significant three-way interaction between treatment (control vs treatment), experimental period (pre- vs post-injection) and time of day (F_11,66 = 2.98, p = 0.0029; Figure 4). Treatment mice largely stopped feeding over 24 hours following LPS challenge. During the 10 hours of the dark phase following the injection, control mice consumed 70 to 138 pellets (1.75–3.45 g) while treatment mice consumed 0 to 3 (0–0.075 g). Because only two of five treatment mice visited the feeder after injection, we could not statistically analyse the effects of LPS on body weight. OPEN IN VIEWER

Human observers failed to detect a difference between treatment and control mice. Whereas IBS was affected by a significant interaction between treatment, experimental period and time of day (F_1,8 = 7.58, p = 0.0249), post hoc tests were not significant: at no time did the scores of LPS-injected mice differ significantly from those of controls or from their own pre-injection scores. We could not statistically analyse the effects on BCS because only two mice, both LPS treated, showed any change on the day after injection: a one-point increase and a one-point decrease.

Discussion

As hypothesized, the automated health-monitoring apparatus successfully detected experimentally induced illness in laboratory mice. Decreased feeding became apparent 2 to 4 hours after LPS challenge, persisted throughout the 24-hour period, and allowed perfect discrimination between treatment and control mice. In contrast, human observers failed to detect any significant difference between sick mice and controls. Automated round-the-clock home cage behavioural monitoring thus allows sensitive and rapid detection of illness: flagging subtle cues that human observers may miss, and doing so well before they are scheduled to conduct daily health checks. Configuring software to alert caretakers when signs of illness are detected would allow them to respond promptly by increasing observation, administering treatment or euthanizing sick animals.

The apparatus enables minimally invasive health monitoring for any number of group-living mice. Sub-cutaneous RFID transponder implantation is itself invasive, but it is a one-time procedure from which mice typically recover readily and resume normal activity. The tag’s small size and location in the nape of the neck make it unlikely to interfere with behaviour or physiology. Monitoring is otherwise completely “hands-off.” The absence of a human observer and lack of handling may both reduce stress^12,13 and possibly decrease the likelihood that mice will suppress behaviours that indicate pain or illness.^3,9 Note that automated behavioural monitoring is intended only as a complement to brief daily visual checks by caretakers, as it would be unlikely to catch certain problems, such as cages flooded by leaky water bottles.

Because several animals completely stopped visiting the data collection tube (feeder) after LPS challenge, we could not test whether the apparatus can detect illness via other measured outcomes. With a less severe challenge, we may have observed progressive declines in body weight ²⁹ or changes in physical appearance (e.g. Table 1). Photographs of the dorsal and ventral surfaces recorded at each feeder visit could also aid in detecting and retrospectively examining progression of some of the most common mouse morbidities, such as fight wounds or ulcerative dermatitis, both of which are visually salient. ¹ Changes in behaviour and physical appearance could potentially aid in other applications, such as detecting pregnancy, parturition, and even estrous in breeding mice. ³⁶

Our apparatus adds to many existing efforts in automated tracking of mouse home cage behaviour.^8,20–23 RFID reader arrays have been used to track location and activity of individual mice in large enriched enclosures.^37–41 Video and machine vision have been used to track location and quantify social and other behaviour in mice.^39,42–46 Feeding bouts have been tracked using RFID-equipped feeders,^47,48 feeders with infrared presence detectors, ⁴⁹ and feeders with built-in scales that additionally measure feed consumption.^50,51 Fluid intake can be regulated and monitored using RFID-based systems,^26,41,52,53 including one developed in our lab (publication forthcoming). Scales at drinking stations or other locations can additionally monitor body weight.^41,54 Intraperitoneal telemetry implants can track locomotor activity and body temperature,^55–57 as can much smaller sub-cutaneous implants (similar in size to the RFID transponders we used).^58,59 Respiratory rate and various behaviours have also been tracked non-invasively using electric field sensors. ⁶⁰ Many of these approaches could readily be combined or integrated with our apparatus to extend monitoring capabilities.

Feed intake and body weight showed circadian fluctuations, rising markedly during the dark phase. This is consistent with previous findings of nocturnal (or crepuscular) circadian rhythms in laboratory mouse feeding,^48–51 drinking,^49,51–53 body temperature,^56,57 and general activity.^{22,49,51,56,61} Mice nonetheless obtained nearly 40% of their feed during the light period. Feeding patterns are sensitive to the energetic or macronutrient content of feed ⁵⁰ and mice in semi-naturalistic enclosures may feed without a clear circadian pattern or even primarily diurnally at certain times of year. ⁴⁷ Diurnal feeding did not seem to be a result of competitive displacement, as mice who shared a cage showed similar feeding patterns – it is unknown if mice in larger groups would attempt to monopolize the collection tube. ⁶² Circadian rhythms in body weight, meanwhile, have not to our knowledge been examined as closely as we have here. Weights declined by nearly 1 g on average from their peak early in the dark period to late afternoon. Frequent automated weighing thus provides a much more accurate picture than even the most diligent daily weighing schedule. In an analogous case, automated monitoring via intraperitoneal implants revealed circadian rhythms in mouse core body temperature that would have remained hidden with manual measurement, particularly as handling affects body temperature.^56,63

This experiment provides proof-of-concept that automated monitoring of health and sickness behaviour in laboratory mice is more sensitive than assessment by humans. Deployment for large-scale routine health surveillance will require adjustments to minimize cost^54,64 and space requirements, and/or large-scale modifications to conventional caging systems.^38,39 Overcoming these hurdles could drastically improve laboratory mouse health, ending the sad situation in which most sick mice “fall through the cracks” and die before caretakers realize there is a problem. ¹