LIQ HD (Lick Instance Quantifier Home Cage Device): An Open-Source Tool for Recording Undisturbed Two-Bottle Drinking Behavior in a Home Cage Environment

Assessment of modified beam-break two-bottle choice system

The initial two-bottle choice pilot experiment conducted in the lab used a modified infrared (IR) beam break system (Frie and Khokhar, 2019; Godynyuk et al., 2019; Extended Data Fig. 1-1A) and singly housed female C57BL/6J mice. Specifically, the design of our device was adapted to expand the volume the bottles could hold to 90 ml each and record 16 cages from a single Arduino microcontroller. Following a week of habituating to two bottles containing water, four additional days of water-only availability were measured by the experimenter in daily sessions. This period was followed by a solution series, during which the fluid in one of the bottles was replaced with the following solutions: two consecutive days of 1% sucrose, 1 d of 10% sucrose, and 1 d of 0.1 g/l quinine. During this series, fluid intake was measured daily by an experimenter. To assess accuracy of the beam break-based measurements, the daily weight change values determined by experimenter intervention were correlated with the IR beam-break number and duration of bream-breaks recorded during the same recording period. When the data across the water and solution series were compiled, we were able to closely replicate previous findings (Godynyuk et al., 2019; Extended Data Fig. 1-1B,C). The correlation between total beam-break bout number and change in bottle weight (R² = 0.3810, F_(1,85) = 52.31, p < 0.0001) as well as the correlation between total beam-break bout duration (seconds) and change in bottle weight (R² = 0.3844, F_(1,85) = 53.07, p < 0.0001), were statistically significant. Unfortunately, we encountered several issues in our redesign throughout the recording period. First, multiple devices required repair or replacement because of mice chewing and damaging the photobeam sensors. Second, because of our selected sipper type and the placement over the sensor, we found that drips hanging from the bottom of the sippers would trigger the sensor and overcount bout duration. Further, mice frequently attempted to bury the sensors with bedding, especially in the initial recording days, which also erroneously triggered the sensors. In the presented data, data points were removed in bins where the bout duration lasted an entire bin or where there was a recorded bout duration without a corresponding bout number.

Assessment of LIQ HD two-bottle choice system

While the data collected with the IR beam-break device significantly correlated with change in bottle weight and were nearly identical to published results, we sought to further improve on the accuracy and reliability of the behavioral recordings as well as increase the precision of the recorded data by designing a device, LIQ HD, that utilizes capacitive sensing technology to detect single licks. We took the modified design described above and replaced the IR beam-break sensors with a capacitance-sensing system. To validate the ability of LIQ HD to measure intake behaviors accurately, we performed a new series of two-bottle choice experiments with singly housed female C57BL/6J mice. Bottle weights were measured every 2–3 d (experiment 1) or every 7 d (experiment 2). Experiment 1 consisted of two groups of mice (8 mice each), where one group received a sucrose dose–response (0.5%, 1%, and 10% sucrose vs water) and the other group received a quinine dose response (0.01, 0.03, and 0.1 g/l quinine vs water). In experiment 2, 16 mice went through a water-only two-bottle choice session, followed by 8 of those mice advancing through an additional ethanol dose–response paradigm (3%, 7%, and 10% EtOH vs water). In both experiments, mice were first habituated to the LIQ devices with water in both bottles for one week, where no measurements were taken, and then given an additional week of water only. For experiment 1, the experimental solutions (sucrose and quinine) were changed every 2–3 d, which coincided with weight measurements and swapping the sides of the bottles. For experiment 2, the ethanol and water bottles were weighed, changed, and swapped sides every 7 d. The LIQ HD and bottle measurement data from experiments 1 and 2 were combined to correlate the total lick numbers and total lick durations from each recording period with the corresponding bottle weight measurements (Fig. 1C,D). We observed a strong, significant correlation between both total lick number and change in bottle weight (R² = 0.9174, F_(1,363) = 4034, p < 0.0001) as well as total lick duration and change in bottle weight (R² = 0.8623, F_(1,363) = 2273, p < 0.0001), substantially higher R² values than our previous IR beam-break based system (Extended Data Fig. 1-1). The fitted regression model for the correlation between lick number and change in bottle weight is Y = 736.4*X − 1172 (slope 95% CI, 713.6–759.2). Thus, on average, the mice took 736 licks to drink 1 ml of fluid, or 1.4 μL per lick. As expected, total lick number and total lick duration also strongly correlate (R² = 0.9720, F_(1,363) = 12 601, p < 0.0001) with a fitted regression model of Y = 0.05,132*X + 8.027 (slope 95% CI, 0.050–0.052; Fig. 1E). On average, the individual lick duration was 51 ms. These metrics are consistent with the literature (Mundl and Malmo, 1979; Parkison et al., 2012; Rossi and Yin, 2015; Bollu et al., 2021), providing further evidence that LIQ HD is detecting individual lick events with high fidelity. In our tests, the LIQ HD system ran reliably undisturbed between periods of bottle weight measurements for at least 7 d (Fig. 1F).

Experiment 1, LIQ HD validation in sucrose and quinine dose–response two-bottle choice tasks

To test the LIQ HD system in common two-bottle choice paradigms, female C57BL/6J mice were split into two groups. One group underwent a two-bottle choice paradigm with a sucrose dose–response curve, and the other a quinine dose–response curve. Bottles were weighed and swapped sides every 2–3 d. When the recorded data were aggregated, we observed a strong, significant correlation between preference score calculated by total lick number and preference score calculated by bottle weight change (R² = 0.8883, F_(1,110) = 874.5, p < 0.0001) as well as between preference score calculated by total lick duration and preference score calculated bottle weight change (R² = 0.8740, F_(1,110) = 763, p < 0.0001; Fig. 2A,B). Additionally, with the added temporal resolution of LIQ HD over standard experimenter-determined measures, we also investigated changes in drink preference over time (Fig. 2C). As expected, preference for the sucrose bottle over water increased with increasing concentrations of sucrose, and preference for the quinine bottle over water decreased with increasing concentrations of quinine (Fig. 2C). Moreover, we examined changes in drinking behaviors over the light and dark cycle to detect potential deviations from the typical circadian drinking patterns, with examples displayed in Figure 2D–I. Overall, for the percent of total licks occurring per cycle in the sucrose group there was a significant main effect of light cycle (F_(1,21) = 209.7, p < 0.0001) and interaction of light cycle × sucrose concentration (F_(2,21) = 17.64, p < 0.0001). While the mice increased their total licks at the sucrose bottle as the sucrose concentration was increased from 0.5% to 1% sucrose, the percentage of the overall licks that occurred during the light and dark phases did not change (Fig. 2J). In addition to further increasing total licks at the sucrose bottle during access to 10% sucrose (Fig. 2H), the mice also displayed a unique increase in percent of sucrose consumption occurring during the light phase (0.5% vs 10% sucrose, p < 0.0001, 1% vs 10% sucrose, p < 0.0001), and corresponding decrease in the dark phase (0.5% vs 10% sucrose, p < 0.0001, 1% vs 10% sucrose, p < 0.0001; Fig. 2J). Mice exposed to increasing concentrations of quinine with water did not display any changes in the light cycle drinking pattern (Fig. 2K; significant main effect of light cycle only, F_(1,21) = 485.5, p < 0.0001). Taken together, these data indicate that LIQ HD can accurately and consistently record overall preference scores, preference over time, and light cycle-dependent drinking patterns in classic two-bottle choice paradigms.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Using LIQ HD to investigate changes in drink preference and light/dark cycle drinking patterns in sucrose and quinine two-bottle choice paradigms. A, Correlation between percent preference calculated with lick number and percent preference calculated with change in bottle weight for each recording period. B, Correlation between percent preference calculated with lick duration and percent preference calculated with change in bottle weight for each recording period. Solid lines represent fitted simple linear regression models, and dashed lines denote 95% confidence intervals. C, Preference over time in 1-h bins for the sucrose and quinine two-bottle choice dose–response paradigm. The solid line represents smoothed mean (sliding window length of 6), and the shaded area signifies ±SEM (n = 8 cages for each group). Vertical dashed lines indicate when bottles swapped sides and when indicated, a change in experimental solution in one bottle. D–I, Lick number over time in 1-h bins for recording periods for the two-bottle choice sucrose and quinine dose response paradigms. The shaded purple area signifies the dark/active phase. J, Percentage of total licks that occur during the dark and light phases for recording periods of the sucrose dose–response paradigm (repeated measures two-way ANOVA, with the Geisser–Greenhouse correction and Tukey’s multiple comparisons test). During access to 10% sucrose, the percentage of licks that occurred during the dark phase was significantly decreased, and the percentage of licks that occurred during the light phase was significantly increased when compared with 0.5% and 1% sucrose availability. K, Percentage of total licks during the dark and light phases for the quinine dose–response paradigm (repeated measures two-way ANOVA, with the Geisser–Greenhouse correction and Tukey’s multiple comparisons test). Shaded areas and error bars represent ±SEM; ****p < 0.0001.

Experiment 2, LIQ HD bout detection and microstructure analysis in a prolonged ethanol two-bottle choice task

Because LIQ HD was able to accurately quantify behavioral data from sucrose and quinine two-bottle choice tasks, we next sought to validate its use in longer duration tasks. To do this, we turned to a six-week continuous access ethanol two-bottle choice task. Further, given that LIQ HD can detect individual lick events, we sought to determine whether the system is able to detect drinking bouts and record bout microstructure. To do this, we coded bout detection directly to the main LIQ HD Arduino code, where a “bout” begins when an animal licks at least three times within 1 s and ends when no licks have occurred over 3 s (Siciliano et al., 2019). With this we can also determine the lick number and lick duration during bouts, which allows us to calculate lick frequency and an estimated interlick interval.

First, to test the LIQ HD bout detection, 16 female C57BL/6J mice underwent a two-bottle choice task with access to two water bottles. Mice were first habituated for one week with a LIQ device with water in both bottles, during which no measurements were taken. Water-related measurements were then taken during a subsequent week of water-only access. We also sought to determine whether LIQ HD is capable to run for prolonged undisturbed recording periods, thus experimenter measurements of bottle weights were taken only every 7 d. The LIQ data from each sipper (16 cages, 32 sippers) was binned into 1-h bins and the average individual bout duration, bout size, bout lick frequency, and bout estimated interlick interval were calculated. Estimated interlick interval values >300ms were excluded from the analysis. The mean and median individual bout duration (seconds) during the water drinking period (n = 2919) were 5.35 ± 0.06 (SEM) and 4.77 (IQR 3.34–6.67; Fig. 3A). The mean and median of individual bout size (licks per bout) during the water drinking period (n = 2914) were 33.9 ± 0.34 (SEM) and 31.0 (IQR 22.0–42.5; Fig. 3B). The mean and median of lick frequency (Hz; n = 2915) were 6.62 ± 0.03 (SEM) and 6.77 (IQR 5.81–7.63; Fig. 3C). The mean and median of estimated interlick interval (milliseconds) during the water drinking period (n = 2858) were 106 ± 0.79 (SEM) and 94.7 (IQR 79.1–118; Fig. 3D). These findings are consistent with results from previous studies (Boughter et al., 2007; Johnson et al., 2010; Parkison et al., 2012; Raymond et al., 2018; Bollu et al., 2021).

• Download figure
• Open in new tab
• Download powerpoint

Figure 3.

Histograms of average baseline bout microstructure measurements from 1-h bins during one-week access to two water bottles (N = 16 mice, n = 32 bottles). Histogram displaying bout duration (A), bout size (B), lick frequency (C), and estimated interlick interval (D). Red and blue dashed lines represent the median and mean, respectively.

Following the LIQ bout detection validation, eight of the female C57BL/6J mice underwent a two-bottle choice ethanol drinking paradigm. The two-bottle choice paradigm consisted of an ethanol ramp where one of the water bottles was replaced with an ethanol solution. During the ramp, mice received one week of 3% ethanol, one week of 7% ethanol, and four weeks of 10% ethanol. As stated above, the system can accurately and consistently record drinking behavior over a 7-d period (Fig. 1F). Moreover, LIQ HD withstood eight weeks (including water only access and habituation) of continuous usage without any devices needing repair or replacement during experimentation.

To determine the correlation of bout number and bout duration with change in bottle weight, as well as the correlation of preference score calculated by bout number and bout duration with the preference score calculated by change in bottle weight, the water two-bottle choice and ethanol two-bottle choice data were aggregated. We observed a strong, significant correlation between total bout number and change in bottle weight (R² = 0.8062, F_(1,156) = 648.8, p < 0.0001) as well as between total bout duration and change in bottle weight (R² = 0.8787, F_(1,156) = 1130, p < 0.0001; Fig. 4A,B). The fitted regression models for bout number versus change in bottle weight and bout duration versus change in bottle weight are Y = 21.51*X + 0.8541 (slope 95% CI, 19.84–23.17) and Y = 157.533*X − 37 270 (slope 95% CI, 148.276–166.790), respectively. These data indicate that on average mice drink 1 ml over 21.5 bouts, or 46.5 μl per bout, and on average take 157.5 s to drink 1 ml during bouts, or 6.35 μl/s. We also found a strong, significant correlation between preference score calculated by total bout number and preference score calculated by bottle weight change (R² = 0.8343, F_(1,69) = 347.4, p < 0.0001) as well as between preference score calculated by total bout duration and preference score calculated by bottle weight change (R² = 0.8791, F_(1,69) = 501.8, p < 0.0001; Fig. 4C,D). While bout number and bout duration may not be as reliable of a predictor of total change in bottle weight compared with total lick number (R² = 0.9174), these data suggest that the LIQ HD bout detection software can accurately detect individual clustered drinking events throughout long recording periods.

• Download figure
• Open in new tab
• Download powerpoint

Figure 4.

Validation of LIQ HD bout detection and bout microstructure in an ethanol two-bottle choice paradigm. A, Correlation between total bout number and change in bottle weight for each recording period. B, Correlation between total bout duration and change in bottle weight for each recording period. C, Correlation between percent preference calculated with bout number and percent preference calculated with change in bottle weight for each recording period. D, Correlation between percent preference calculated with bout duration and percent preference calculated with change in bottle weight for each recording period. Solid lines represent a fitted simple linear regression model, and dashed lines denote 95% confidence intervals. E, Preference over time in 1-h bins for the ethanol two-bottle choice paradigm. The solid line represents smoothed mean (sliding window length of 6), and shaded area signifies ±SEM (n = 8 cages). Vertical dashed lines indicate when bottles swapped sides and when indicated, a change of experimental solution in one bottle. F, Average preference for experimental bottle during ethanol two-bottle choice paradigm (repeated measures one-way ANOVA, with the Geisser–Greenhouse correction and Dunnett’s multiple comparisons test). Mice show significantly increased preference compared with baseline for 7% and 10% ethanol. Bout duration (G), bout size (I), lick frequency (K), and estimated interlick interval (M) over time in 24-h bins for water and ethanol bottles. Bout microstructure data were grouped into 24-h bins because there were no significant differences between the light and dark cycle (Extended Data Fig. 4-1). Average bout duration (H), bout size (J), lick frequency (L), and estimated interlick interval (N) during access to two water bottles and access to water with increasing concentrations of ethanol (repeated measures two-way ANOVA, with the Geisser–Greenhouse correction and Dunnett’s multiple comparisons test). At the ethanol bottle, mice display a significant decrease in bout duration (G), bout size (I), and interlick interval (N) with a significant increase in lick frequency (L) during access to increasing concentrations of ethanol. At the water bottle, mice display a significant increase in lick frequency only during access to 3% ethanol (L), a significant decrease in bout duration during access to 3% and 10% ethanol (H), and a significant decrease in bout size only during access to 10% ethanol (J). Shaded areas and error bars represent ±SEM; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

In addition to preference over time in the prolonged ethanol two-bottle choice paradigm, where mice showed expected significantly elevated preference for both 7% ethanol (p = 0.0481) and 10% ethanol (p < 0.0001) compared with water (Fig. 4E,F), LIQ HD can also be used to analyze drinking bout microstructure over time (Fig. 4G–N). Bout microstructure data were grouped into 24-h bins, as we did not observe differences between the light and dark cycle (Extended Data Fig. 4-1). Throughout this increasing preference for ethanol, the average bout duration, bout size, lick frequency and estimated interlick interval were altered. Although bouts at the ethanol bottle became more frequent compared with those at the water bottle, access to ethanol significantly decreased the average bout duration (significant main effect of ethanol concentration, F_{(2.129,29.80)} = 23.68, p < 0.0001) at both the water bottle (H₂O vs 3% EtOH, p = 0.0096, H₂O vs 10% EtOH, p = 0.0022) and the ethanol bottle (H₂O vs 3% EtOH, p = 0.0055, H₂O vs 7% EtOH, p = 0.0310, H₂O vs 10% EtOH, p = 0.0054; Fig. 4H). Bout size also significantly decreased (significant main effect of ethanol concentration, F_{(1.882,26.34)} = 17.01, p < 0.0001) at the ethanol bottle during ethanol exposure (H₂O vs 3% EtOH, p = 0.0288, H₂O vs 10% EtOH, p = 0.0189), but only decreased at the water bottle during exposure to the highest ethanol dose (H₂O vs 10% EtOH, p = 0.0047; Fig. 4J). Interestingly, lick frequency significantly increased (significant main effect of ethanol concentration, F_{(2.300,32.20)} = 8.236, p = 0.0008) at the ethanol bottle at all doses (H₂O vs 3% EtOH, p = 0.0005, H₂O vs 7% EtOH, p < 0.0001, H₂O vs 10% EtOH, p = 0.0017; Fig. 4L) with a concurrent decrease in the estimated interlick interval (significant main effect of ethanol concentration, F_{(2.300,32.20)} = 8.236, p = 0.0008, significant interaction ethanol concentration × bottle F_(3,42) = 3.801, p = 0.0169, H₂O vs 3% EtOH, p = 0.0002, H₂O vs 7% EtOH, p < 0.0001, H₂O vs 10% EtOH, p = 0.0236; Fig. 4N). We only observed an increase in lick frequency at the water bottle during access to 3% ethanol (H₂O vs 3% EtOH, p = 0.0434) without any significant differences in estimated interlick interval (Fig. 4L,N). Overall, these data replicated known effects of prolonged, continuous access to increasing concentrations of ethanol in an undisturbed home cage environment on preference for ethanol over water and revealed significant changes to the bout lick microstructure in female C57BL/6J mice.