Adult male and female mice (>56 days old) from BALB/cByJ were housed in standard distributional type of housing cages in a temperature and humidity-controlled sound proof box on a 12:12h light-dark cycle for 1 week before the beginning of the experiment. The intensity of the LED light that mice experience at the bottom of the housing cage is adjusted to 300 lux. Humidity and temperature for animal maintaining chambers were kept between 40~60% and 25~26℃. All experimental procedures were conducted according to protocols approved by the directives of the Institutional Animal Care and Use Committee of the Institutional Animal Care and Use Committee (Approval number :201603025) of KIST (Seoul, Republic of Korea).

For animal housing floor material, metal grid floor was optimal for long term recording compared to flat metal floor, although both materials provided similar conduction, resistance, and accuracy of electrical signals. This is because mice feces can be easily accumulated in flat metal floor. Moreover, flat metal floor has been shown to hamper adequate footing of animals. Therefore, metal grid floor is suitable for long term recording, considering animal's welfare.

For animal housing cages, we used metal grid floor type distributional mouse cage with metal lid (JEUNG DO Bio & Plant, Korea) (Fig. 1A). Sound proof LED light-dark chamber was custom ordered from the company (JEUNG DO Bio & Plant, Korea). For standard A/D converter, we have utilized a Digidata 1440a (Molecular Device) for simultaneous 16 channel recording. The recording environment of computer was Intel G2020 2.90 GHz dual core processer with 2 GB RAM and 1.5 TB hard drive running on Windows 7 OS. Our Circa Analysis software was developed on MATLAB version 2017b. For recording software, Axoscope ver. 10.6 (freely available from Molecular Device) was used. With our HKA system, it was possible to record up to 16 channel simultaneously with A/D converter of Digidata 1440a model (Up to 8 channels with 1320a model). Circa analysis software provides up to 16 channel simultaneous analysis.

Formula 1Filtered Data=HPFRaw Data

Formula 2Absolute Datan=Filtered Datan

Formula 3Reduced Datan=1Sf∑k=Sf(n−1)+1SfnAbsolute Datak

Formula 4When, h′Th= h(Xpeak)−h(Xend)Xpeak−Xend

Formula 5Bianary Datan= 1,   Reduced Datan>Th0,   Reduced Datan≤Th

Formula 6Ativity Countsn=1Bi∑k=Bi(n−1)+1BinBianary Datak

^*HPF: High Pass Filter, Butterworth 5^th order, 1 Hz cut-off

^*Rosin's: Rosin's thresholding algorithm

^*S_f: Sampling frequency

^*T_h: Threshold

^*B_i: Bin Interval

^*h(x): Histogram of samples, x: Amplitude

The A/D converter used in our HKA system has 16-bit resolution, 16 input channels, 1 to 250 kHz sampling rate range, −10.00 to +10.00 V of input dynamic range, 1MOhm of input resistance. Since the sampling frequency of our recording condition was 20Hz. We recorded 16 channel simultaneously, therefore the precise calculation of the single data size is 52.7 MB.

16(bit)×20 (Hz)×86,400 (second)×16 (channel)=442,368,000 (bit)

442,368,000 (bit)×0.125=55,296,000 (byte)

55,296,000 (byte)÷10⁶=52.7 (MB)

A previous study has shown that simply connecting standard A/D converter with metal sipper tube of water bottle and metal grid floor can precisely detect animal's licking behavior [24]. In detail, a junction potential is generated whenever animal licks the tube and closes the electrical circuit for the duration of the tongue-sipper tube contact [24]. Junction potentials occur wherever dissimilar conductors are in contact. Based on this idea, we hypothesized that connecting the metal grid floor and the metal lid of the animal housing cage would generate electrical signals composed of any lid-touching behaviors. To test this hypothesis, we have plugged one side of BNC cable with input of the standard A/D converter and another side with animal's housing cage, by connecting BNC core part at the metal lid and BNC shell part at the metal grid floor (Fig. 1A). Next, we placed laboratory mouse inside the cage and monitored animal's behavior and signal generation. The signals are generated when animals grab, eat and drink (Fig. 1C). Whenever animals are just crawling the floor, hanging on the cage ceiling, or sleeping, signals were not generated (Fig. 1D). For drinking behavior to generate electrical signals, the metal lid and the metal sipper tube must form metal contact (Fig. 1B). Note that the signals were generated only by direct physical contact between animal and metal materials, but not by the induction. This was confirmed by putting the thin plastic barrier between the cage between metal lid and cage body (data not shown). Taken together, the recorded signals reflect HKA of an animal which can serve as a marker for circadian rhythm.

The currently available WRA-based circadian rhythm systems have cost issues and intrinsic limitations due to the dependence of a running-wheel. The cost issues arise from the need of purchasing additional sensors and recording equipment. Intrinsic limitation of WRA system is that wheel-running behavior itself alters the animal's activity patterns in some animals. To overcome these issues caused by wheel dependency, we have designed a HKA-based comprehensive circadian rhythm recording system with hardware and software (Fig. 2).

Circadian rhythm behavior experiments require relatively long period of recording time ranging from one month to a year. Thus, the stability of recording environment, temporal precision and convenience of management are important. In our HKA system, since the only required recording device is an A/D converter and itself serves as a sensor, signal quality and data acquisition stability were determined by A/D converter. For our HKA configuration, we utilized a standard A/D converter which was originally designed for electrophysiology (see details in Materials and Methods). This A/D converter, Digidata 1440a, can digitize 16 independent channels simultaneously, at sampling rates up to 250 kHz, and supported by AxoScope 10 software. AxoScope is a freely available, easy-to-use, full-featured data acquisition program for Windows. With these resources, we developed a HKA-based hardware configuration (Fig. 2A). We recorded signals in gap-free mode at the sampling frequency at 20 Hz, because this particular sampling frequency reliably detected “touching activities” while maintaining the reasonable size of the recorded files. Files were saved at 24 hour interval as “.abf (axon binary format)” file which is the default file format of Axoscope. For 16 channel recordings with duration of 24 hours, the size of files for 16 channels was 52.7 MB, indicating that at least 1.6 GB storage space is required for 30-day experiment.

To extract animal's touch activities from the raw electrical signals and visualize animal's circadian rhythmic activity patterns, we developed MATLAB-based user-friendly software, Circa Analysis (Fig. 2B). We also made a demonstration video to show step-by-step procedure of how to use the program (Supplementary Video 1). With “Extract” function, Circa Analysis converts raw signals to scalable “touch activities”. For example, when an animal touches the metal lid three times (for 2 s, 25 s and 15 s) during 0.1 hour (=360 s) of time bin, the total duration of touch activities is 42 s per bin. With “Display” function, Circa Analysis displays converted scalable “touch activities” to draw actogram by aligning the daily activities in a series (Fig. 2B). Taken together, with a unique combination of hardware and software based on HKA, we have developed a comprehensive circadian rhythm system.

To convert raw electrical signals into scalable touch activity, we applied various signal processing algorithms to the raw data. We recorded HKA with 20 Hz sampling frequency, and our raw data was masked by low frequency baseline noise (Fig. 3A, H, I) and this noise was completely removed when we applied Butterworth 1 Hz cut-off High Pass Filter (Fig. 3B and Formula 1). Although junction potential due to touch activity should be a positive value, our filtered data showed both positive and negative values. This is probably caused by HPF artifacts. To get rid of negative values, we have absolutized the signals (Fig. 3C and Formula 2). Next, we reduced the data size by averaging the number of samples at lower sampling frequency (Fig. 3D and Formula 3). For example, since our original sampling frequency was 20 Hz, averaging values of 20 consecutive samples made one data point per one second. This process simplified the handling of data. Next, to automatically discriminate the real touching signal from the baseline, we have applied a threshold based on Rosin's method [25] (Fig. 3G and Formula 4). Rosin's method is known as an effective auto-thresholding algorithm for unimodal histograms. It turned out that the amplitude frequency histogram of the reduced data showed a distribution of unimodal histogram, and therefore, optimal for applying Rosin's method. To explain briefly, the algorithm first draws a straight line from the peak to the end of the unimodal histogram. Second, the algorithm draws an orthogonal line from the peak-to-end line to each bin value of the histogram, until it finds the bin with a maximal distance from the peak-to-end line. When the distance is maximal, then Formula 4 becomes true. The threshold is determined as the amplitude value at this bin with maximal distance. Using this algorithm, we were able to search the optimal threshold automatically for each channel data. Using the threshold, we converted the reduced data into binary data, assigning the value of one of each data point if amplitude is greater than threshold, otherwise the value was set as zero (Fig. 3E and Formula 5). To quantify the duration of touch activities, we summated touch durations per bin interval (Fig. 3F and Formula6). In our analysis, we set the bin interval to 0.1 hour (=360 s). Thus the value of y axis represents total duration of touch activities per bin interval of 360 s. Taken together, with these six signal processing steps, we were able to automate the process of extracting numerical values for touch activities.

According to Nyquist-Shannon theorem, one's sampling rate should be twice greater than the target signal's bandwidth frequency (f_S≥2f_B). Therefore, to set the optimal range of recording sampling rate, it is crucial to understand the property of signals that we are handling. In our HKA system, we are recording the animal's touching activity. Since HKA signals are generated by animal's touching or licking of metal materials, the expected HKA signal bandwidth would be determined by maximal frequency of those behavior. Previous studies have reported that animal's licking frequency range is 4~7.5 Hz [24,26] and touching frequency range is 0.1~1 Hz [27]. Therefore, we expected the bandwidth of HKA signals to be less than 10 Hz and 20 Hz of sampling rate would be sufficient for detecting the HKA signals.

With 20 Hz sampling rate, the HKA signals were accurately detected along with animal's metal contact. However, as shown in the Fig. 3A and 3I, our signals of interest were contaminated with unidentified low frequency noise (<1 Hz). We found that the source of this noise was 60 Hz powerline interference signal which was aliased down to low frequency signal. The reason why 60 Hz signals were aliased down to under-1 Hz-signals was because our 20 Hz sampling rate is subnyquist sampling rate for 60 Hz signals. Since theoretically calculated aliasing frequency of 60 Hz signal sampled with 20 Hz is 0 Hz, we should not observe any baseline noise in recordings. However, 60 Hz powerline in real life has been shown to vary as much as ±1 Hz [28], and this drifts in 60 Hz frequency seem to be the major cause for our HKA system's baseline noise generation.

To confirm this idea, we simulated recording of drifting 60±1 Hz powerline interference with 20 Hz sampling rate (Fig. 3H). When 60 Hz signal was sampled with 20 Hz signal, the aliasing frequency was 0 Hz. Along with increase or decrease in 60 Hz powerline frequency, aliasing frequency increased exactly same with the step size. As long as drifts of 60 Hz signals are restricted to a range of ±1 Hz, our baseline noise frequency is less than 1 Hz. Based on simulation result, it is possible to remove low frequency baseline noise by 1 Hz HPF.

Before applying 1 Hz cutoff HPF to remove baseline noise, we had to confirm that whether our HKA signals would retain after filtering. By investigating the frequency domain and comparing “Baseline only” window and “Baseline with Signal” window of raw data traces, we noticed that both of the window shares similar and high amplitude in the frequency range of 0.01~0.1 Hz, whereas the range of 0.1~10 Hz showed marked difference in power level (Fig. 3I). This was consistent with the findings of simulation result that low frequency domain less than 1 Hz seem to be baseline noise. On the basis of our observations, we assumed that the frequency domains of 0.01~0.1 Hz as baseline noise window, 1~10 Hz as HKA signal window, and 0.1~1 Hz as mixed window. Therefore, the cutoff frequency for removing baseline noise, should be selected in the range of 0.1~1 Hz. In our recording condition, we selected 1 Hz as a cutoff frequency of HPF, and filtering with this condition showed clear HKA signals (Fig. 3I).

Although our post-processing method provide much simpler way to setting up HKA system, note that adding a low-pass filter with cutoff frequency less than 60 Hz before ADC could have removed powerline interference at the beginning.

To verify that our system is working properly, we applied various protocols for circadian rhythm behaviors. First, we performed a phase advance experiment by applying 6 hour light pulse stimulation at Zeitgeber time (ZT) 15 of 11th day of constant dark period (Fig. 4A). Our HKA recording system successfully recorded mouse's phase advance (CT 3.1 h) of circadian rhythm without changing taus (before:23.5, after:23.69). Similarly, phase delay experiment by light pulse stimulation at 10th day of constant dark period showed mouse's phase delay (CT 3.5 h) of circadian rhythm without changing taus (before:23.7, after:23.8) (Fig. 4B). Next, we tested a jet-lag protocol (Fig. 4C) and showed actogram and Gaussian-filtered activity center plot. During jet-lag delay and advance period, animal spent 5 days and 4 days each to adapt. Lastly, seasonal variation protocols with light hour:dark hour ratio of 16:8 (Fig. 4D) and light hour:dark hour ratio of 8:16 (Fig. 4E). All of these experiments were stably and reliably recorded with HKA system. In addition, we confirmed that at least it is possible to record simultaneously from 16 mice over 150 consecutive days (data not shown). This indicates that HKA system can provide highly stable recording system for circadian rhythm behaviors.

To assess the cost-effectiveness of HKA system, we compared required components for WRA system and HKA system. As shown in Table 1, one can save the cost required for physical parameter counter (running-wheel) and transducer sensor (wheel-counter) with HKA configuration. Moreover, since HKA configuration utilizes distributional types for housing cage and A/D converter which are readily available in markets, HKA system can also benefit from not using company-tailored types which are optimized only for specific WRA system. Lastly, since we have developed a software which is optimized for this system, we were able to save additional cost for analysis software which is generally provided by company with a certain price. Overall, due to its simplicity of hardware composition, HKA system can be a highly cost effective system for recording circadian rhythm behavior.