The ability to quantify drinking behaviour in laboratory animals has been an essential part of behavioural and metabolic research for decades [1, 2, 3]. Often, the focus is on measuring the volume of liquid intake over relatively long periods of time (hours or days), which can provide essential information about, e.g., degrees of thirst or caloric intake. In other cases, however, it is important to be able to characterise lick-by-lick the precise timing of fluid intake rather than the exact volume consumed, because licking patterns can be a useful measure of ingestive behaviour [1, 4, 5, 6]. This requires the ability to quantify licking with a temporal resolution of milliseconds. Such ability is particularly important for experiments where licking is assessed in combination with recordings of fast changes in cellular activity, e.g., in vivo electrophysiology [7] or photometry [8], in order to investigate correlates between neural activity and behaviour.

Instruments for measuring licks (as opposed to volume of liquid intake) are commonly termed lick sensors or lickometers. Several manufacturers of scientific equipment offer lickometers for research. Often, however, these commercial lickometers can only be used in conjunction with proprietary hardware and software. For example, some commercially-available lickometers are designed to only fit predetermined, proprietary animal cages, and require specific control and data acquisition hardware and software supplied by the same manufacturer. This not only increases costs dramatically but can also be a barrier to using lick sensors in combination with additional laboratory instruments.

To increase flexibility and reduce costs, several researchers have proposed lick sensor designs that can be implemented in the lab using a few electronic components [9, 10, 11, 12, 13, 14]. These designs have in common that the animal is part of an electrical circuit that is completed with every lick, the tongue acting effectively as a switch: licks are detected by the opening and closing of the electrical circuit (Fig. 1A) [15]. One disadvantage of this design is that the experimental arena must be modified to include conductive material on the floor, on top of which the animal must stand when licking in order to complete the electrical circuit (Fig. 1A). Moreover, the data recorded by these methods is often an analog signal (i.e., voltage oscillations) that require a degree of post-processing, e.g., thresholding the signal to extract the licking events (Fig. 1B) [9, 14], and this is unsuitable for real-time experimental loop-control.

Fig. 1.

Advantages of capacitive lick sensors. (A) Many of the lick sensors currently available detect licks when the tongue of the animal, acting like a switch, closes an electrical circuit. To be able to complete the circuit, the animal must stand on a conductive platform (e.g., a metal grid) connected to the lick detection circuit. (B) Often, the output of these sensors is an analogue signal that requires some degree of postprocessing, e.g., setting a threshold (dashed line) to detect licks. (C) With capacitive sensors, the animal is not an integral part of the lick-detection circuit: no metal surfaces are required on the cage floor so the setup is simpler and more flexible. Physical contact with the spout induces changes in capacitance in the sensor’s circuit [16] and this is recorded as a lick event. (D) Lick events with a capacitive lick sensor are detected as on/off, binary events (i.e., a digital signal) in real time. Lick data is thus immediately available, and can potentially be used for, e.g., controlling an LED for in vivo optogenetics. LED, light-emitting diode.

Here, we describe an alternative, easy-to-build lick sensor that does not require modifications to the experimental arena (Fig. 1C, Ref. [16]), can be used with standard water bottles or drinking spouts, and detects licks in real time with millisecond precision. This system, based on off-the-shelf capacitive touch sensors, is inexpensive and easy to construct, can be connected to a variety of data acquisition systems, and can be used for real-time control of additional hardware such as light-emitting diodes (LEDs) for in vivo optogenetics.

We recorded licks in mice with access to one of two sucrose solutions (5% sucrose or 20% sucrose) on separate sessions for a total of n = 9, 20-minute long recordings. A second group of mice had access to both water and 2% sucrose drinks on the same session, and in these mice licks were recorded for 3 hours for a total of n = 5 recordings. Lick events were used for calculating interlick intervals and lick clusters (Fig. 3).

3.1 Lick Frequency and Interlick Intervals

Licking in rodents is rhythmic and occurs at a regular rate of about 8 to 10 Hz, depending on the strain [18, 19, 20]. As a first step towards validating our lick sensor we quantified lick frequency. Mean lick frequency ranged from 8.6 to 9.9 Hz across all 9 sessions in Experiment 1, and from 7.1 to 9.7 Hz across the 5 sessions in Experiment 2. The observed distributions of interlick intervals (Figs. 4A,5A) were as described previously [5, 20].

Fig. 4.

Lick microstructure analysis for Experiment 1. Data from five male mice with access to a 5% sucrose or 20% sucrose drink (one drink per recording, n = 9 recordings). (A) Distribution of interlick intervals. (B) Mean of number of clusters and mean of cluster size. The whiskers represent SD. (C) Complementary cumulative distribution (CCD) of cluster size. (D) Cumulative probability of lick over time. The animals were more likely to lick for sucrose 20% than sucrose 5% during the first half but not the second half of the session.

Fig. 5.

Lick microstructure analysis for Experiment 2. Data from five female mice with access to both water and 2% sucrose (two drinks per trial, n = 5 trials). (A) Distribution of interlick intervals. (B) Mean of number of clusters and mean of cluster size. The whiskers represent SD. *p = 0.03, paired t test. (C) Complementary cumulative distribution (CCD) of cluster size. (D) Cumulative probability of lick over time. The animals were more likely to lick for sucrose than water during the first hour of the session.

We then calculated the mean primary interlick interval, defined by St John et al. [19] as “the average interlick interval of interlick intervals shorter than 160 ms”. When the two sucrose conditions tested in Experiment 1 were compared, mean (standard deviation, SD) primary interlick interval was 110 (5) milliseconds for 5% sucrose (n = 5 recordings), and 107 (6) milliseconds for 20% sucrose (n = 4). For Experiment 2, the mean (SD) primary interlick interval was 121 (16) milliseconds for water (n = 5), and 124 (6) milliseconds for 2% sucrose (n = 5). These values were very similar to those reported before for B6 mice [19].

There were no significant differences in mean primary interlick intervals between the two sucrose conditions in Experiment 1 (5% sucrose vs. 20% sucrose, Wilcoxon signed rank exact test on paired samples, V = 10, p = 0.125), or between the two drinks in Experiment 2 (water vs. 2% sucrose, Wilcoxon signed rank exact test on paired samples, V = 4, p = 0.875). This is in line with previous reports showing that lick frequency tends to remain relatively constant across different experimental conditions (e.g., drink palatability or water deprivation) [19, 20].

Lick microstructure is an important measure of intake behaviour, and it is particularly useful for studying food preference and motivation [1, 4]. To analyse lick microstructure it is necessary to record the precise timing of licks. This can be achieved with lick sensors (lickometers) [3]. Here, we described and validated a lick sensor for rodents that can be built in the lab easily with few off-the-shelf components.

The interest in measuring lick patterns has led several researchers to propose different lick sensor implementations. Early designs involved complex electronic circuits [1, 10, 11, 12, 21] that detect licks based on the completion of an electrical circuit (Fig. 1A). For this to happen the animals must stand on a conductive platform connected to the main circuit, thus requiring specific adjustments to the experimental cages. In addition, some lick-sensing systems also demand custom drink bottles. Moreover, lick sensors that rely on the completion of electrical circuits often capture data in the form of analog signals that need additional post-processing steps to isolate lick events [9]. This approach is not suitable for lick detection in real time.

By contrast, the sensor described here can be used with standard drinking bottles and does not require specific modifications to the cages. As a result, animal behaviour can be studied in the home environment, and since no novel bottles are introduced into this environment, no training of the animals is necessary before the experiments can commence. It is possible with our lickometer to detect lick events in real time (Fig. 1C,D) which can be useful for controlling additional laboratory hardware if required. In parallel, these lick events can be stored to a computer for analysis of lick microstructure, and this makes our design flexible.

The sensor described here works by detecting the changes in capacitance that occur when the tongue of the animal touches the drinking bottle (Fig. 1C). While capacitance-based lick sensors have been described before [22, 23], those early designs consist of elaborate electronic circuits that can be difficult to construct and operate. More recently, simpler capacitive lick sensors built with only a microcontroller and a few resistors have been described [24]. In contrast to those lick sensors, ours utilises off-the-shelf touch-sensing integrated circuits to detect changes in capacitance (Fig. 2), which offer important advantages. Touch-sensing integrated circuits are easy to use and configure, and include filters to make touch detection more reliable. They also calibrate automatically, which means that they can compensate for drifts in capacitance that can occur over time, and can be used in different environments or conditions with little need to make manual adjustments when these conditions change. Finally, a single touch-sensing integrated circuit can be used for detecting licks from many connected drink bottles at the same time.

Touch-sensing integrated circuits have become widely available in recent years because they form the basis for human interfaces to many consumer products such as touch screens. Touch sensing technology has become reliable, inexpensive, and simple to use, making it easy to adapt these integrated circuits for touch-detection tasks in the research laboratory. In fact, in a recent paper, Petersen et al. [25] describe a capacitance-based lick-sensing setup that follows very similar principles to those we present here. This corroborates that lickometers designed with off- the-shelf capacitance sensors are a reliable and simple alternative to commercially-available lick sensors. One important difference, however, between the system by Petersen et al. [25] and the one described here is that theirs requires custom 3D-printed drink bottles and bottle holders, which complicates the implementation and limits their setup to placing 2 bottles per cage. Moreover, their system is not designed to provide real-time lick signals for loop-control of external equipment (although it should be possible to adapt their system for that purpose).

Capacitive sensors like the one used here detect touch by measuring the charge-discharge time of a capacitor, then digitally filtering the data (see the MPR121 data sheets, available at https://www.nxp.com/products/MPR121, for details). This necessarily introduces a small time delay between the touch event and its detection. With the default settings, the typical response time of the MPR121 is 16 ms. This delay is likely short enough to be of little relevance for the purposes of lick detection. It is possible, however, to modify the MPR121 settings to decrease this value if necessary.

To validate our lick sensor we carried out two independent sets of experiments in mice. Our analysis of lick microstructure from data recorded during these experiments (Figs. 4,5) matches what is already known about licking behaviour in mice tested under similar conditions. Thus, lick frequency and its counterpart interlick intervals were within the range previously reported and, as others have found [18, 19, 20], did not change significantly with the different drinks tested. This demonstrates that our lick sensor is capable of performing at least as well as commercially-available sensors used by others.

Our analysis of lick clusters also mirrored previous findings under similar experimental conditions [18, 20]. It is interesting to note that previous reports of cluster size in mice vary widely. For example, lick clusters in C57 mice drinking water have been found to consist of as little as 4 licks per cluster [18] or as many as 40 or more licks per cluster [19, 20]. This suggests that cluster size should not be used to compare lick microstructure across different studies. Instead, cluster size is likely a measure only useful for comparing different conditions within the same study. This is evident here, where cluster size in Experiment 1 (Fig. 4B) cannot be compared directly to that from Experiment 2 (Fig. 5B) likely because of the different experimental designs. Finally, differences in cluster size observed across and within our two experiments may also be a reflection of the different training schemas implemented. Our animals in the two experiments experienced sucrose drinks for dissimilar periods of time before the experiments commenced. This may be relevant because it has been reported that the degree of previous exposure to sucrose drinks can affect the number of clusters measured during experiments [26], perhaps mediated by learning processes.

We described an open-source lick sensor for behavioural experiments in rodents that is inexpensive, reliable, and easy to build. It is flexible, because it can be used with any standard drinking bottles (up to 12 bottles per sensor; a second sensor can be added for 24 bottles), and does not require specific modifications to the animal cage or experimental arena. This latter characteristic means that licking behaviour can be monitored in more naturalistic (e.g., home cage) environments. Finally, lick events detected with this sensor can be stored for future analysis or, importantly, they can be used for real-time loop-control of additional laboratory instruments. The sensor described here therefore lowers the barrier to implementing a wider range of experiments involving lick behaviour, which should be of great benefit for neuroscience and metabolism research.

Diagrams, code, and detailed instructions for building and operating the lick sensor can be found at https://github.com/antgon/lick-sensor-for-rodents. A C library for using the Raspberry Pi Pico microcontroller with the MPR121 touch sensor is available at https://github.com/antgon/pico-mpr121. Data from behavioural experiments is available on request.

JAG conceived the project, designed and built the lick sensor, carried out animal experiments, analysed data, and wrote the manuscript. YM contributed to designing and performing animal experiments. Both authors contributed to editorial changes in the manuscript, and read and approved the final version of the manuscript. Both authors have participated sufficiently in the work and agreed to be accountable for all aspects of the work.

All experimental procedures were performed in accordance with the UK Animals (Scientific Procedures) Act 1986 and the University of Aberdeen Ethical Review Committee (ERC21-22:06) approval, and complied with the ARRIVE 2.0 guidelines.

The authors would like to thank Jenna Hunter and Giuseppe D’Agostino (University of Manchester) for their assistance with animal experiments.

Research funded by the Biotechnology and Biological Sciences Research Council (BBSRC), UK (grant BB/V010557/1 to J.A.G.).

The authors declare no conflict of interest.