The CMOS imaging chips, designed by our laboratory, were cleaned by submersion in acetone (Fujifilm Wako), twice, and then in isopropanol (Fujifilm Wako) for 5 min each. After drying, they were each mounted on a polyimide, gold-circuit-printed FPC substrate (Taiyo Industrial) taped on a glass slide using a thin layer of epoxy resin [low viscosity epoxy resin Z-1 (N), CraftResin]. The blue-light-emitting micro LEDs (ES-VEBCM12A, Epistar), with a central emission wavelength of 470 nm, were also mounted in the same fashion.

The blue-light filter was prepared by first dissolving Valifast Yellow 3150 dye (Orient Chemical Industries) with cyclopentanol (1:1, w/w) overnight in a light-proof vial. In the same container, the mixture was mixed with Norland Optical Adhesive 43 (Norland Products) (2:1, w/w) using a vortex. The resulting adhesive mixture was spin coated (Spincoater model: 1H-D7, Mikasa) on a silicon-coated [CAT-RG catalyst and KE-106 silicone (Shin-Etsu Chemical), 1:10, w/w] cover slip with a size of 23 mm × 23 mm under the following setting: 3 s to 500 rpm–5 sec at 500 rpm–5 s to 2,000 rpm–20 sec at 2,000 rpm–5 s to 0 rpm. The spin coated material was immediately heated on a hot plate at 100°C for 30 min then left to set in room temperature overnight. The filter film was cut with an Nd: YAG laser (Callisto VL-C30RS-GV, TNS Systems) to make a grid of 10,000 × 3,500 μm sheets.

Cut filter sheet were manually placed over the entire imaging area of the CMOS chips. The device was baked in a vacuum oven (AVO-250NS, ETTAS) for 2 h at 120°C and left to cool.

Using a needle, red resist resin (ST-3000L, Fujifilm) was applied on the sides of the CMOS chip. This resin prevented entry of stray blue LED light through the sides of the chip.

The CMOS chip and the LED were connected to the circuitry of the FPC with micro aluminum wires (Tanaka Electronics) using a wire bonder (7400C-79, West Bond). The wiring was sealed by a protective cover of epoxy resin [low viscosity epoxy resin Z-1 (N), CraftResin] and left to set overnight.

The processed device was incised off from the excess FPC material with a blade. The sides at implantable end of the cut FPC substrate of the device were coated with a thin layer epoxy [low viscosity epoxy resin Z-1 (N), CraftResin] to cover jagged edges.

The devices were lined and bound together with a ribbon of Kapton tape and were enclosed in a parylene coater (PDS 2010, Specialty Coating Systems). They were coated with 5 grams of dichloro-c-cyclophane (GalentisS.r.l.). The thickness of the parylene deposition layer is about 2.5 μm. This transparent, long-lasting coating serves as a bio-protective sheath to prevent infiltration of biological substances into the device that can damage the circuitry.

All animal handling procedures were approved by the Nara Institute of Science and Technology (NAIST) Animal Committees, and were performed in accordance with the institutional guidelines of the animal facilities of NAIST.

GCaMP6 mice [strain: FVB-Tg(Thy1-GCaMP6)5Shi., provided RIKEN BRC through the National Bio-Resource Project of the MEXT, Japan (Ohkura et al., 2012)], around 2 months of old of either sex, were implanted with two of our CMOS-based imaging devices (Figure 2A). They were positioned adjacent to the CeLC and the DRN, both in the left hemisphere (Figure 2B).

The mice were anaesthetized. Hair was removed from the top of their heads and the scalp was disinfected with 4% clorhexidine (Hibitane). They were restrained to a stereotaxic platform (SR-6M, Narishige) using earbars. A heating pad was provided underneath the animals to stabilize body temperature.

Skin was excised from the dorsal side of the head, just enough to access the implantation sites. The cranium was exposed by clearing away tissue and washing with PBS (Fujifilm Wako). After aligning the bregma and lambda, coordinates for the brain site targets were marked on the cranium [CeLC: AP: −0.8 mm, ML: −3.35 mm (left), DV: −4.0 mm; DRN: AP: −5.56 mm, ML: −0.35 mm (left), DV: −3.31 mm; all DV coordinates counted from the dura]. All coordinates were determined using the Paxinos Mouse Brain Atlas (Franklin and Paxinos, 2008) and calibrated in previous trials. Two small micro-screws were shallowly secured into the cranium, spaced some distance from the marked sites. Small cranial windows of around 1 mm in diameter were created on the targets using a dental drill (Hypertec II, Morita) and were cleansed of debris and blood with PBS.

The implants were gently wiped with 70% ethanol (Fujifilm Wako) using cotton swabs and were secured to stereotaxic manipulators (SM-15M, Narishige). They were slowly implanted into the windows, with the CMOS chip facing medially toward the target, up to the predetermined depth. For the DRN, the path of implantation was angled 25° posteriorly to avoid damaging the major superficial blood vessel lying between the entorhinal cortex and the cerebellum (Figure 2B).

Once the targets were reached, the cranial windows were sealed with silicone elastomer (Kwik-Cast). The exposed cranium and the surrounding incised skin were sealed with dental cement (Super-Bond kit, Sun Medical). The cement was also used for stabilizing and anchoring the implanted device unto the skull surface and the inserted screws. The manipulators and other restraints were removed. The mouse was allowed to recuperate for at least 12 h in a heated enclosure.

A modified formalin test was done with the aim of recording brain activity during pain perception, indicated by fluorescence from calcium signaling due to neuronal action potentials. Behavior attributed to pain perception, such as licking, was used to support the validity of the fluorescence imaging data (Figure 3A).

The whole recording procedure was done within a darkened canvas tent. This is to prevent the interference of light from sources such as illumination lamps for animal handling, interior overhead lighting, and sunlight passing through windows. Light from such sources are of high enough intensity to permeate the very thin skull, and overlying layers, of the experimental mouse. Such filtered light can still be detected by the implanted devices due to the high sensitivity of the CMOS imaging chip. The orbit and the frontal cranial bone are especially exposed to light penetration. Though the parietal cranium bone is covered with reinforcing dental cement, the cement itself is translucent. Implanted mice were anaesthetized with isoflurane (Fujifilm Wako) using an isoflurane pump (410 Anaesthesia Unit, Univentor). The implants were then connected to cables that were then connected to a slip ring. The rotating slip ring prevented entanglement of the cable pair and served as a bridge between the implants and the power sources (6146 DC Voltage Current Source, ADCMT) and collecting equipment (custom-made, NAIST).

After securing the connections, the mice were kept in an observation enclosure and monitored with a webcam (CMS-V37BK, Sanwa Supply). Illumination was under blue-light that was blocked by the blue-light filter and the red resist coating of the device. Their brain activity and behavior were recorded on a personal computer using specialized custom-made software (CIS_NAIST) and a webcam software (Bandicam) (Figure 3B) for a minimum of 10 min to serve as a baseline. Afterward, they were subcutaneously injected using a 30G needle with either 20 μL 2% paraformaldehyde-PBS (Fujifilm Wako) [“Formalin” group (n = 3)] or PBS (Fujifilm Wako) [“PBS” group (n = 3)] at the plantar side of the right hind-paw, contralateral to the side of the implantations. Brain activity and behavior were recorded for a minimum of an hour. For the duration of the experiment, experimenters vacated the tent to prevent affecting behavior.

At the end of the experiment, the mice were sacrificed with an overdose IP injection of sodium pentobarbital (Somnopentyl, ∼0.2 mL, KS Medical). They were perfused with a tubing pump (TP-10SA, AS ONE) with normal saline (Otsuka) and then 4% formaldehyde (Fujifilm Wako). Their brains were extracted and stored in 4% formalin overnight. Coronal sections of the brains (100 μm) were prepared using a vibratome (Linear Slicer PRO7, Dosaka) and were used to confirm successful targeting of the brain sites (Figure 2C). Data from animals with unsuccessful implantations were disregarded.

The inner-brain temperature, in Celsius, of an implanted mouse was measured during activation of the blue-light micro-LED at 0.5 mA. Two thermocouples (Cu/constantan (Type T) thermocouple, Muromachi Kikai Co., Ltd., Tokyo, Japan) were bounded unto needle-type devices with Parafilm M (PM-996, Pechiney Plastic Packaging). One thermocouple terminal was located by the LED and another at the device’s insertion tip, serving as a reference site away from the LED. The thermocouples were connected to a microcomputer thermometer (BAT 700 1H, Physitemp Instruments LLC).

Mice were anaesthetized and implantations were done into the DRN of one mouse and the CeLC of another as previously described, but with some differences. After implantation, elastomer sealant was not used as a protective cover. Instead, PBS-moistened pieces of Kimwipe were applied on the exposed region surrounding the implant and were held in place with Parafilm. The animal’s body temperature was allowed to stabilize. Temperature was recorded for 5 min before LED activation, for an hour during activation, and for 5 min afterward.

Videos of the brain fluorescence activity were collected using CIS_NAIST and screen recordings of the working computer. Images of the fluorescence in the CeLC and the DRN were extracted from the time immediately after the injection of the substance and every 10 min thereafter, until the end of the 1-h observation period. The clearest calcium imaging result was selected as a representative example of each sampling group.

Behavior recorded by the webcam was reviewed and pain-related behaviors were taken note of, specifically the licking of the injected site. Behavior was quantified by tallying the amounts of pain-induced licks on the affected paw per 2.5 min blocks within the 1-h observation period. The results were graphed as a number of licks per time block along the passage of time.

Imaging data was acquired using CIS_NAIST, and saved as RAW files. The data was extracted from the files and analyzed using MATLAB (MathWorks). Custom made codes were written in order to process and visualize the data. For processing, the data was first stored as a 3-dimensional matrix (2D spatial pixel array across time), then separated into two for the CeLC and the DRN data. Afterward, the period of injection was determined from webcam recordings, and also by looking at the offset frames in the averaged data set.

Removal of hum noise was performed by making a column vector containing the average values of each row in the pixel array. The average value of that was computed and subtracted to each element of the column vector. The resulting vector was subtracted to each column in the pixel array. This was done in each frame. That is, given a pixel reading F_t(x,y) at frame t, where x = 1 : n_x, y = 1 : n_y, the following steps are performed for each t:

The data was normalized (ΔF/F₀) by getting the average of the frames before injection as the baseline. That is, given F_t(x,y), the baseline F_0 is given by F0⁢(x,y)=1t*⁢∑t=1t*Ft⁢(x,y), where t^∗ is the frame number before injection. The normalized pixel reading Ft*⁢(x,y) is then given by Ft*=△⁢FF0=Ft-F0F0, where Ft*⁢(x,y)=Ft⁢(x,y)-F0⁢(x,y)F0⁢(x,y).

Based on a previously published study (Takehara et al., 2016), the approximate size of neurons was computed and the regions of interest (ROIs) were selected accordingly. Specifically, it was assumed that the maximum distance between a visible ROI and image sensor surface was 100 μm, and therefore; the full-width at half-maximum (FWHM) would increase 3–4 times compared to a distance of 0 μm. Given the soma and device pixel sizes (8–9 μm and 7.5 μm, respectively) and the 4 times increase in soma size due to the distance from image sensor, then the ROI was computed to be around 6 × 6 pixels. Regions that seemed like neurons based on fluorescent activity were selected as ROIs for further analysis. The average of each ROI was plotted and compared against background values outside the ROIs. A scale bar representing 5% change from baseline was generated. A color plot showing the intensity of each pixel in a frame was graphed to visualize the ROIs. The behavioral data was aligned with the calcium imaging data to see their relationship.

Afterward, the first-differenced calcium traces in each ROI were cross-correlated or auto-correlated. First-difference was applied to ensure stationarity of the data. That is, given a calcium trace reading z_t at time t, the first-differenced calcium trace △z_t at time t is △z_t = z_{t + 1}−z_t. First-differencing was implemented using diff() function of MATLAB. Cross-correlation between CeLC and DRN ROIs was calculated by shifting the DRN data across time. That is, given the CeLC calcium trace reading x_t at time t and the DRN calcium trace reading y_t at time t, the non-normalized cross-correlation coefficient R at lag m is given by

R x , y ⁢ ( m ) = { ∑ t = 1 T - m x t ⁢ y t - m if ⁢ m ≥ 0 R y , x ⁢ ( - m ) if ⁢ m < 0 .

The (normalized) cross-correlation ρ at lag m is then given by ρx,y⁢(m)=1Rx,x⁢(0)⁢Ry,y⁢(0)⁢Rx,y⁢(m). Here, R_x,x(m) and R_y,y(m) represents the non-normalized auto-correlation at lag m for x_t and y_t, respectively. The DRN time lag with the highest correlation with CeLC m^∗ = argmax_m ρ_x,y(m) was recorded as the best time lag. Cross-correlation analysis was implemented using xcorr() function of MATLAB. The three ROIs from CeLC and three ROIs from DRN were cross-correlated as follows: CeLC vs. DRN, CeLC vs. CeLC, and DRN vs. DRN. Therefore, a total of 27 cross-correlations were analyzed per mouse.

To measure the relationship between brain imaging and behavior, the mutual information coefficient (MI) between these variables was computed. Basically, the mutual information coefficient (Cover and Thomas, 2006) between two (discrete) random variables X and Y is given by

I ⁢ ( X ; Y ) = ∑ x ∈ 𝒳 ∑ y ∈ 𝒴 p X , Y ⁢ ( x , y ) ⁢ log ⁡ ( p X , Y ⁢ ( x , y ) p X ⁢ ( x ) ⁢ p Y ⁢ ( y ) ) .

Similarly, if X,Y are continuous, then the summation is replaced with integration. However, since brain imaging is a continuous random variable, while the behavior is discrete, MI was measured using an adapted method (Ross, 2014). MI was computed using discrete_continuous_info_fast() function of Ross (2014). Since the imaging data was a continuous real-valued dataset, it was binned and averaged across the same 2.5 min window of the behavioral data. This enabled the measurement of mutual information between continuous imaging data and discrete behavioral data, where a higher value indicates more dependence between the two.

Finally, statistical analysis was done using MATLAB. Non-parametric tests were performed: Kruskal-Wallis test for the cross- and auto-correlation analysis and non-parametric two-way ANOVA (Friedman’s test) for the brain calcium imaging and licking behavior relationship analysis were performed. P-values less than 0.05 were considered significant. Boxplots with median and interquartile range were graphed also using MATLAB.

No notable complications on the welfare of the experimental mice were encountered during the duration of the study. Implantation surgery was accomplished without issues and all the mice recuperated fully the following day. No signs of distress that may have come from the implantation were observed. Furthermore, there were no indications of encumbrance of the head. Post mortem weighing of the cement-bound dual implants, excluding the parietal portion of the cranium, gave an average weight of 0.474 g (n = 5). The only signs of discomfort were the pain-related behavior and inflammation reaction of the Formalin group mice after injection and the slight agitation immediately resulting from animal handling.

Images of the calcium signaling from representative examples were chosen (Figure 4). The top right corner of the DRN in the Formalin group had an increasing intensity throughout the time course, from the 20-min mark onward, with distinct and prominent fluorescence. The CeLC of the same group, particularly the lower-left of the imaging area, initially displayed some fluorescence that gradually lessened over time.

On the other hand, the PBS group had more uniform and constant images across time. Slight changes in overall fluorescence can still be seen, such as a brighter DRN and CeLC at the 30-min mark. The changes are quite difficult to spot by eye alone. Therefore, a more quantitative approach was taken.

Readings in a double-implanted mouse depict very small increase in deep brain tissue temperature during activation of the blue-light micro-LED (Figure 5). At most, there was 0.5°C increase in the portion of the DRN next to the LED. The results also show that proximity to the LED does not necessarily lead to a higher tissue temperature.

To better visualize changes in CeLC and DRN fluorescence activity, the whole frame average across time was computed in all mice (Figure 6, top and middle). In addition, the paw-licking behavior of the mice was also measured simultaneously (Figure 6, bottom). Due to the small size of the implantable device, both CeLC and DRN can be visualized at the same time in a freely behaving mouse. Based on the ΔF/F₀ scale bar, the Formalin group generally had higher amplitudes than the PBS group. For example, Formalin Mouse 2 displayed higher fluorescence values compared to PBS Mouse 2 in the CeLC. Furthermore, Formalin Mouse 1 had higher fluorescence intensities than PBS Mouse 1 in the DRN. In addition, the Formalin group had higher frequency of fluorescent peaks for both brain sites. This was most apparent for Mouse 2 and 3 for CeLC and Mouse 1 for DRN. Conversely, the PBS group had graphs that are comparatively flatter. Since this was a proof-of-concept to show that the dual-implantable device works, more mouse samples and further analysis can be done to quantify the difference in amplitudes and frequencies between both groups.

Pain-related behavior was observed from all the mice of the Formalin group, as reflected by the average number of licks. A high number of licks was seen in the initial minutes after injection. The licking behavior then subsided after 10 min. Then, after 30 min licking behavior started to increase again. This bi-phasic response may correspond to the acute and inflammatory phase of formalin injection, as will be discussed later. Interestingly, peaks in licking behavior also corresponded to higher fluorescence activity in the CeLC and DRN. On the other hand, the PBS mice did not display any hindpaw-licking behavior. These results confirm the successful execution of the formalin test.

Another two mice were selected as representatives, one per sampling group, to visualize analysis of multiple regions of interests (ROIs) (Figure 7). Slight differences in timing and intensity of fluorescence can be seen in ROIs of the same brain region. For instance, in the Formalin-injected mouse, DRN ROI2 displayed more fluorescence activity than DRN ROI1 at the 20-min mark. It is also observable that the ROIs of the DRN have higher and more distinct spikes than the background. Furthermore, ROI 1 and 2 of the CeLC had higher fluorescence amplitudes than ROI 3. However, CeLC ROI 3 still had higher fluorescence than the CeLC background values.

On the other hand, the PBS-injected mouse had ROI values that are relatively similar with the background. Similar to Figure 6, the PBS mouse had a more constant fluorescent activity where its peaks do not deviate too far from the average noise value. The peaks of the Formalin-injected ROIs were more distinct and had higher amplitudes compared to its baseline level before injection.

The relationship between each ROI was explored. The small size and large field of view of the device made it possible to measure cross-correlation of multiple ROIs within and between the CeLC and DRN. First, cross-correlation analysis of the Formalin-injected mouse showed distinct and clear peaks mostly at time lag 0 (Figure 8). The only exception was the cross-correlation between CeLC ROI 1 and the DRN ROI 1. The data indicated that when the first-differenced DRN data is shifted 1 frame back, then the correlation with first-differenced CeLC data increases. This means that when the CeLC fluorescence increases, there is a positive correlation that the DRN fluorescence will also increase after 0.094 s. However, this was only true for one ROI of the CeLC. The other two ROIs of the CeLC showed a definite peak at 0 time lag, which indicated that the cross-correlation was highest at 0 s delay.

In contrast, the PBS-injected mouse had less clear and less distinct cross-correlation peaks. It can be seen that most of the cross-correlation graphs are broader, especially for DRN ROI 2. Furthermore, the best time lag is less consistent among the ROIs, with one pair having the highest cross-correlation at 6 frames (0.562 s) and two pairs of ROIs at 5 frames (0.468 s). This may indicate less synchronization between CeLC and DRN neuronal firing in PBS-injected mouse.

To provide more insight and see if the results are consistent, the cross-correlation of ROIs in each mouse was analyzed (Figure 9). Majority of the peak cross-correlations was at time lag 0 (Figure 9A). However, it can be observed that PBS mice had several non-zero and non-one time lags. For example, PBS Mouse 1 had a high cross-correlation (ρ = 0.69) at −8 frames when analyzing between its CeLC ROI 1 and DRN ROI 2. Then its CeLC ROI 2 and DRN ROI 2 had the best lag at +3 frames.

Furthermore, even within the DRN, there was a high cross-correlation at 8, 15, and 4 frame lags for PBS Mouse 1, 2, and 3, respectively. On the other hand, more cross-correlations at time lag 0 can be seen in Formalin-injected mice. For instance, when comparing within the same brain region (i.e., CeLC vs. CeLC or DRN vs. DRN) all ROIs displayed a best time lag of 0, with the exception of DRN ROI 2 and 3 of Formalin Mouse 2. Taken together, this may indicate less synchronization within the DRN of PBS-injected mice.

To compare the difference between the number of frame lags higher than 1 for each mouse group, a two-tailed unpaired t-test was performed. The number of frame lags higher than 1 was statistically higher in the PBS group compared to the Formalin group (p = 0.00773, Unpaired t-test). This confirmed the hypothesis that PBS-injected mice had a higher number of ROIs that were not as synchronized compared to the Formalin-injected mice.

Finally, to measure the relationship between the brain calcium imaging data and licking behavior of mice, we calculated their mutual information (MI). The MI between calcium imaging of PBS mice and their licking behavior is close to 0, showing a lack of a relationship. On the other hand, the Formalin mice displayed higher MI levels across different ROIs and brain areas (Figure 10A). This was further confirmed using two-way ANOVA, wherein the Formalin group was significantly different compared to the PBS group (p < 0.05). However, no difference was detected between CeLC and DRN within each group.

By using this MI metric, the capability of the dual-implantable device to correlate brain activity and behavior was demonstrated. This showed the potential of the device for use in further experiments and analysis.