Transcranial Direct Current Stimulation (tDCS) Induces Adrenergic Receptor-Dependent Microglial Morphological Changes in Mice

Visual Abstract


Significance Statement
Transcranial direct current stimulation (tDCS) is a neuromodulation procedure in which a weak electric direct current is delivered through the brain for tens of minutes. Despite reported positive effects, the mechanisms of tDCS stimulation are not yet well understood. Here, we examined microglial morphology in the mouse cortex after tDCS. We find that the morphology and morphologic dynamics of microglia are altered by tDCS in a manner dependent on adrenergic receptors, supporting the notion that (nor)adrenergic signaling is involved in tDCS.

Introduction
Noninvasive neuromodulation is a subject of intense research because of its potential for treating patients with neuropsychiatric and neurologic conditions. Transcranial direct current stimulation (tDCS) is the application of a constant and weak electric current to the brain through the skull. Typical parameters applied in humans are 1 mA over ϳ30 cm 2 for 10 -30 min (Bikson et al., 2016). A fair sized body of published literature suggests that tDCS has positive effects on cognitive abilities and could be an alternative treatment for various brain disorders (Fregni and Pascual-Leone, 2007;Nitsche et al., 2008Nitsche et al., , 2009Brunoni et al., 2012;Dedoncker et al., 2016). On the other hand, there is a notable degree of skepticism due to mixed outcomes of tDCS experiments (Horvath et al., 2015a,b;Jalali et al., 2017;Medina and Cason, 2017;Kunzelmann et al., 2018;Turkakin et al., 2018). The skepticism has been, in part, strengthened by a recent study that suggested negligible tDCS-induced membrane potential changes in cerebral cortical neurons (Vöröslakos et al., 2018), implying limited involvement of neuronal discharge as the prevalent mechanism of tDCS.
The circuit and cellular mechanisms for tDCS remain to be understood. Glial cells represent electrically nonexcitable cells in the nervous system. They have been regarded as "support cells" for the normal function of neurons. Among glial cell types, astrocytes and microglia maintain the extracellular milieu by ion homeostasis and phagocytosis, respectively. Additionally, astrocytes and microglia have been reported to interact with neuronal synapses (Wake et al., 2013;Araque et al., 2014). We recently reported that astrocytic Ca 2ϩ surges occur during tDCS in mice. Moreover, tDCS-induced astrocytic Ca 2ϩ surges were shown to promote cortical plasticity and have beneficial effects in a mouse model of depression Hirase, 2016, 2018). The recruitment of Ca 2ϩ activities in astrocytes has prompted us to investigate another major glial cell type, microglia.
Microglia are sensitive to brain tissue damage and transform to reactive microglia on inflammation. Iba1 (ionized calcium binding adaptor molecule 1) immunohistochemistry (IHC) visualizes the morphology of microglia, which is profoundly altered in reactive microglia. Following the published observation that reported the lack of pronounced microglial reactivity after tDCS , here we investigated Iba1 IHC in detail by digital image analysis. We report subtle, but significant effects of tDCS in an awake condition, but not under anesthesia, that depended on adrenergic receptors. Subsequently, we examined microglial motility by in vivo two-photon imaging and found that tDCS reduces microglial motility.

Materials and Methods
All animal procedures were performed in accordance with the RIKEN animal experimental committee regulations.

Surgical procedures
Mice were deeply anesthetized with isoflurane (1.5-2.0%), and their scalps were exposed by shaving. Each mouse was fixed on a stereotaxic apparatus (Narishige) under isoflurane anesthesia. Throughout the surgery and experiments with anesthetized mice, the body temperature was kept at 37°C with a heating blanket (BWT-100A, Bio Research Center). After topical application of xylocaine ointment (2% lidocaine) on the scalp, the skull above the sensory cortex was exposed by incision of the scalp and temporal muscle. A custom-made chamber ring was glued to the skull with cyanoacrylate superglue. After the glue settled, we applied dental cement (Fuji LUTE BC, GC Corporation; Super-Bond C&B, Sun Medical) for reinforcement. For two-photon imaging, the inner cavity of the chamber ring was reinforced with additional dental cement to secure the interface for an objective lens. Once the chamber ring was rigidly attached, the mouse was fixed on a custom-made stage via the chamber ring. Thereafter, a small craniotomy ( ϭ 3 mm, with intact dura) was carefully made using a dental drill.

Habituation to head restraint
The postsurgical recovery period was at least 3 d for IHC experiments and 2 weeks for in vivo two-photon imaging experiments. Following the recovery period, mice were placed on a water restriction schedule and subjected to an acclimatization procedure for head restraint (Fig. 1B). Food was given ad libitum. The acclimatization procedure was performed for 7 d.
On day 1, each mouse was held in the experimenter's hands and water was given via a syringe (ϳ0.2 ml). During handling, we let the mouse explore until it entered into a body tube similar to the one used with the tDCS apparatus. If the mouse entered the body tube, we repeated the procedure four to five times. The total handling time was 10 min for each mouse. From day 2, the mouse continued to be acclimatized to the experimenter and apparatus with a water reward (0.1-0.2 ml) for each entry to the body tube. At this point, the head of the mouse was quickly (Ͻ10 -20 s) fixed to the apparatus via the chamber ring with its body in the tube. Additional water and sunflower seeds were provided during head fixation (10 -15 min). The total amount of water given during head fixation was 1.0 ml/d. In some mice, for in vivo two-photon imaging acclimatization was performed for Ͼ7 d. Transcranial DC stimulation tDCS was applied on mice under anesthesia (2% isoflurane) or in awake conditions. In either condition, the anode (stainless wire, ϭ 350 m) was placed on a sodium chloride-based conductive gel interface (Z101BA, NIHON-KODEN) spread over a circular area ( ϭ ϳ2 mm) above the primary visual cortex (anteroposterior, Ϫ2.9 mm; mediolateral 2.0 mm). The cathode was connected to the neck skin after topical application of xylocaine ointment. DC (0.1 mA, 10 min) was applied with a custommade isolated constant-current supply.

Histology
After tDCS application, mice were kept for 30 min or 3 h before they were killed. After deep anesthesia by urethane, they were first perfused with 0.9% NaCl and later with fixative solution (4.0% paraformaldehyde in 0.1 M phosphate buffer, pH 7.4). Following brain removal and overnight postfixation in the same fixative, coronal slices (60 m) were prepared using a microslicer (PRO 7, Dosaka). For Iba1 staining, sections were incubated in a buffer containing the primary antibody (Tris-buffered saline with 0.1% Triton X-1000; 1:2000; catalog #019-19741, Wako) overnight. The sections were subsequently washed in PBS and incubated with the Cy3-conjugated secondary antibody (Invitrogen) for 2 h for fluorescent labeling. To evaluate DSP4 [N-(2-chloroethyl)-N-ethyl-2bromobenzylamine hydrochloride] efficacy, noradrenergic fibers were labeled by anti-tyrosine hydroxylase (TH) antibody (1:1000; catalog #AB152, Millipore) using sagittal slices (60 m). For positive control of microglial reactivity, Escherichia coli lipopolysaccharide (LPS; 0.5 mg/kg) was administered by intraperitoneal injection 2 d before the mice were killed.

In vivo imaging of microglial morphology
Adult Iba1-EGFP transgenic mice (Hirasawa et al., 2005), in which EGFP is expressed exclusively in microglia, were used to monitor microglial morphologic dynamics. All mice were habituated to the experimental apparatus for Ͼ7 d. On the day of imaging, the mouse was set on a custom-made stage under a two-photon microscope (B-Scope, Thorlabs). Microglia located Ͼ50 m below the pial surface were imaged under awake conditions at a wavelength of 920 nm. The laser power was adjusted to ϳ12 mW at the preparation (Hines et al., 2009;Wake et al., 2009;Pfeiffer et al., 2016). Depth stacks (24 -26 slices, 2 m z interval, 512 ϫ 512 pixels corresponding to 101 ϫ 101 m 2 or 201 ϫ 201 m 2 ) were acquired every 60 s.

Iba1 IHC image analysis
Confocal images were used for pixel intensity analysis. Image stacks extending to 15 m thickness were collapsed into 2D images by maximum intensity projection. Pixel intensities were converted to z scores, and the cumulative distribution was computed for each collapsed 2D image.
For soma size analysis, confocal image stacks (45-50 m thickness) were first filtered with a 3 ϫ 3 ϫ 3 median filter. The resultant image stacks were collapsed into 2D images by maximum intensity projection. To correct for uneven background, the rolling ball method with a radius of 30 pixels was used for background subtraction. Thereafter, the images were subjected to a 3 ϫ 3 2D median filter followed by binarization with Yen's thresholding method (ImageJ, National Institutes of Health) for soma extraction. In some cases, manual adjustments of threshold were needed. Extracted somata were approximated to ellipses. Following these automated procedures in Im-ageJ, extracted somata were validated by manual inspection. The median of microglial soma size distribution from each mouse was taken as a data point for statistical comparisons.

TH image analysis
The efficacy of DSP4 was evaluated by calculating the mean intensity of posterior cortical layer 1 TH-positive (TH ϩ ) innervation using ImageJ. Briefly, sagittal brain section images (60 m thickness) were acquired by a Keyence microscope (BZ-X710; 0.37 m pixel size). Ten to 12 contiguous regions of interest (ROIs; 100 ϫ 100 m each) were allocated to occupy layer 1. A background intensity value was calculated from a neighboring parenchymal area that does not contain TH ϩ axons. The mean TH ϩ signal intensity of each ROI was computed as the mean pixel intensity minus the background intensity.

Microglial motility assessment
Quantification of microglial surveillance was performed using custom-written ImageJ and MATLAB programs (MathWorks). The maximum intensity projection image was computed for each time point of xyzt image stack. The resultant xyt image stack was registered for xy motion correction. Next, each slice of the xyt stack was processed by the ImageJ "Subtract Background" plugin to subtract smooth continuous background with a ball size of 30 pixels. Thereafter, images were treated with a 2D 3 ϫ 3 median filter. After this preprocessing, rectangular areas containing the morphologic extent of single microglia were extracted as separate image stacks. These cellwise image stacks were then binarized with a single threshold determined by Li's Minimum Cross Entropy method (ImageJ). Noise reduction was then performed by a cycle of erosion and dilation. The normalized surveillance area at time t was calculated as the number of pixels that were occupied by the microglia at least once since the beginning of imaging until a given time t, divided by the number of pixels occupied by the microglia at the beginning. Normalized surveillance area is therefore a monotonically increasing function of time (see Fig. 6C,D, example). The surveillance index is defined as the ratio of normalized surveillance areas of a microglia in two different sessions [e.g., control ("Before") vs post-tDCS ("After")].

Statistical analyses
Statistical analyses were performed using Igor Pro (WaveMetrics). Student's paired t tests and Wilcoxon-Mann-Whitney rank sum tests were used for the comparison of two sample populations with matched data and unmatched data, respectively, unless otherwise noted. Data are expressed as the mean Ϯ SEM, and p values Ͻ0.05 were considered statistically significant. Statistical values are reported in Table 1.

Results
First, we confirmed tDCS-induced cortex-wide Ca 2ϩ elevations  in the present setting using G7NG817 transgenic mice that express the G-CaMP7 Ca 2ϩ sensor in astrocytes and a subpopulation of neurons. Mice had been acclimatized to be rigidly fixed to a head-restraint platform, where tDCS (0.1 mA, 10 min) and transcranial fluorescence imaging were performed (Fig.  1A,B; see Materials and Methods). Cortical Ca 2ϩ signals elevated immediately after the passage of the DC current. The peak amplitude of the G-CaMP7 response measured ϳ3 mm anterior to the anodal position was 39.7 Ϯ 4.1% ( Fig. 1C; N ϭ 4 mice), showing that tDCS-induced Ca 2ϩ elevation is observable with the head chamber-ring configuration. Notably, tDCS-induced Ca 2ϩ elevations were not observed in isoflurane-anesthetized mice (Extended Data Fig. 1-1). Having demonstrated the effectiveness of tDCS, we used C57BL/6 mice to investigate microglial morphology after tDCS by Iba1 IHC. Mice were killed either 30 min or 3 h after tDCS for perfusion fixation.

Iba1 IHC patterns are affected by tDCS in awake mice
Iba1 IHC visualized highly ramified microglial morphology throughout brain slices of sham-operated, LPStreated, and tDCS mice ( Fig. 2A,B,I,J). To investigate the impact of tDCS on the wide-field appearance of Iba1 IHC, we computed the pixel intensity distribution, which is a proxy of global morphologic changes. We analyzed layer 2 and 3 of the visual cortex located below the anode, since a previous study demonstrated that tDCS-mediated plasticity occurs in these layers . Pixel intensities were converted to z scores with which the cumulative distributions were plotted. We compared head ring-implanted, unrestrained control mice (Ctl group) with head ring-implanted, acclimatized, 25 min headrestrained mice (Sham group) to evaluate the possible effects of head restraint. In Figure 2C, we demonstrate that cumulative pixel intensity distribution is similar between the Ctl and Sham groups, whereas the pixel intensity distribution shifted significantly in mice with reactive microglia caused by LPS. These results suggest that the head-restraining procedure in acclimatized mice does not cause reactivity in cortical microglia.
Next, we compared tDCS and sham-treated mice. The combination of two conditions [isoflurane-anesthetized (isofl) or awake] and two time points (30 min and 3 h after tDCS) were investigated (Fig. 2D-G). Pixel intensity distribution was similar between sham and tDCS for isofl 30 min, isofl 3 h, and awake 30 min experiments; however, the awake 3 h tDCS data exhibited a visible deviation from the sham group (Fig. 2G). This deviation was caused by a higher proportion of pixels at the high-intensity end. For instance, tDCS had a relatively large presence of pixels that had a z score Ͼ0.6 (21.2 Ϯ 1.6% vs 18.2 Ϯ 1.4%; p ϭ 1.7e-5, t test). Moreover, a high-intensity cluster that has a z score Ͼ2 was apparent in the pixel intensity histogram (Fig. 2H). Consistent with this observation, thresholding with z Ͼ 2 preserved more microglial structures in awake 3 h tDCS images than the sham counterpart (Fig. 2I,J). While the cumulative pixel intensity histogram of awake 3 h tDCS deviated in the same direction as LPS, microglial morphology appeared normal with fine ramified processes throughout the extent of the cor-

tDCS enlarges microglial somata in awake mice
While the z score-based pixel intensity distribution analysis detected changes in the global appearance of images, it falls short of providing information on specific aspects of morphologic alterations. Microglial soma size has been reported to be sensitive to brain environmental changes (Kongsui et al., 2015). Therefore, we measured microglial soma size from Iba1 IHC images (Fig. 3A-C; see Materials and Methods). First, we compared the median microglial soma size of individual animals (43 cells per mouse on average) for unrestrained control and headrestrained sham groups as we did in Figure 2C. Figure 3D indicates that microglia soma sizes are similar between

D-G, Intensity was compared between tDCS-and Sham-treated groups under the isoflurane-anesthetized (D, E) or awake (F, G)
conditions, perfused at 30 min or 3 h after sham/tDCS. H, In awake mice, the pixel intensity histogram indicates that there is a cluster at z score Ͼ2 (i.e., mean ϩ 2 SDs) region in the tDCS group (dotted red square). I, J, Representative images from a sham-treated mouse and a tDCS-treated mouse. Images in the red squares correspond to the thresholded images on the left at the mean ϩ 2 SDs. Red scale bars, 20 m. ‫‪p‬ءء‬ Ͻ 0.01, ‫‪p‬ءءء‬ Ͻ 0.001. the control and sham groups (Ctl group: 45.4 Ϯ 1.0 m 2 , 7 mice; Sham group: 43.4 Ϯ 1.0 m 2 , 7 mice; p ϭ 0.16 b , Mann-Whitney Wilcoxon test), suggesting that the microglial soma size of the Sham group serves as a valid control for tDCS experiments.

tDCS decreases microglial surveillance area in vivo
One of the striking features of microglia is the motility of their ramified processes. Here, we directly examined the New Research morphologic dynamics of individual microglia in the cortex of awake mice using a two-photon microscope (Fig.  6A). We used the Iba1-EGFP mouse, in which EGFP is exclusively expressed in microglia (Hirasawa et al., 2005). We confirmed that microglia showed surveillance activities by continual extension and retraction of their processes in all directions (Davalos et al., 2005;Nimmerjahn et al., 2005). For example, the overlay of 60 min imaging resulted in an extensive coverage of the area within ϳ60 m from the soma, while the soma position remained unmoved (Fig. 6C). We defined normalized surveillance area as the proportion of cumulative microglia-occupied area at a given time relative to a start time (Fig. 6D). To check whether laser scanning has an impact on microglial morphology, we compared the occupied area of each monitored microglia at the beginnings of Before and After imaging sessions of the Sham-treated group (Fig. 6E). We found no significant difference, suggesting that the effect of laser irradiation on microglial morphologic dynamics is negligible.
While the evolution of a normalized surveillance area varied considerably among individual microglia, the average trace converged to a gradually decelerating curve (Fig. 6F). The mean surveillance area after 60 min did not differ significantly between before and after sham stimulation. Remarkably, the mean surveillance area index curve of tDCS mice (i.e., After session) is plotted lower than the control condition (i.e., Before session). We assessed the surveillance area change of individual microglia by taking the ratio of the surveillance area indices during Before and After sessions, demonstrating a significant decrease of surveillance area by tDCS (t ϭ 40 min, p ϭ 0.014 j , paired t test; Fig. 6G).
Furthermore, we addressed whether noradrenergic signaling is involved in this tDCS-induced microglial surveillance reduction by prazosin or ICI181551 pretreatment in awake mice (Fig. 7A). As a reference, we computed the surveillance index comparing Before and After sessions at 40 min after the start of respective sessions. As expected from the previous analysis (Fig. 6G), the surveillance index of tDCS experiments was significantly reduced (Fig. 7A). Prazosin-treated mice showed a similar significant reduction of surveillance index after tDCS (Fig. 7B). By contrast, ICI181551 treatment abolished tDCS-induced reduction of microglia surveillance, and a trend for increased surveillance was apparent (Fig. 7C). These results point to a significant role of the B2AR in the inhibition of microglial surveillance activity after tDCS.

Discussion
The present experiments report that tDCS induces subtle, but significant, alterations of Iba1 distribution and microglial motility in the cerebral cortex in awake mice. Furthermore, these alterations were dependent on (nor-)adrenergic receptors, which is in line with the results of an earlier study that described tDCS-induced A1ARdependent astrocytic Ca 2ϩ surges . Notably, while astrocytic Ca 2ϩ responses occur during tDCS, morphologic alterations of microglia occurred after a few hours.
We demonstrated that the microglial soma is enlarged after tDCS. Remarkably, the soma enlargement occurs only in awake mice. It is well established that microglial morphology is radically altered by LPS-induced inflammation (Kondo et al., 2011;Kozlowski and Weimer, 2012;Kongsui et al., 2015). LPS-induced microglial alterations are obvious even with a low dosage of 100 g/kg, whereby ϳ20% soma enlargement has been reported in the prefrontal cortex (Kongsui et al., 2015). The tDCSinduced microglial soma enlargement of a mere several percentage points in the current study is relatively modest. Moreover, no obvious change was detected in ramified processes. As general anesthesia compromises astrocytic Ca 2ϩ activation, in particular noradrenergically driven large-scale and synchronized Ca 2ϩ surges (Thrane et al., 2012;Ding et al., 2013), microglial changes by tDCS conceivably depend on the elevated noradrenergic tone during awake states. On the other hand, some studies have reported significant changes in anesthetized mice that underwent tDCS. For instance, one study reported that enhancements of GFAP and brain-derived neurotrophic factor (BDNF) in anesthesia changed gene expression ( de Souza Nicolau et al., 2018). Another study showed long-lasting antidepressive behavioral effects (Peanlikhit et al., 2017). However, these studies used stronger stimulation in terms of stimulus current, duration, and/or frequency. Moreover, the anesthesia condition used in the current study is deeper than that in the study by Peanlikhit et al. (2017). Considering the lack of astrocytic Ca 2ϩ surges in this condition (Extended Data Fig.  1-1), our results support the involvement of volumetransmitted neuromodulators in tDCS.
A few studies have examined cortical microglia after tDCS. For instance, Rueger et al. (2012) reported that multisession tDCS of 5-10 d induced a mild sign of microglial activation as observed by an upregulation of Iba1 immunohistochemical signals. The current density used in the study by Rueger et al. (2012) was ϳ150 A/m 2 , whereas that used in the current study is Ͻ30 A/m 2 . Considering the study by Gellner et al. (2016), which reported a microglial activation threshold of 30 -50 A/m 2 with light isoflurane anesthesia, it is conceivable that our experiments were performed in near-threshold conditions. The tDCS-induced microglial soma enlargement and Iba1 signal intensity distribution shift are different from the microglial morphologic alterations reported in a rodent model of electroconvulsive therapy (ECT), in which obvious reductions in process ramification and Iba1 expression occur (Jinno and Kosaka, 2008). The pronounced alterations of microglia by ECT are most likely caused by the high-intensity electric stimulation that induces seizures. By contrast, cortical neuronal discharge activity remains undisturbed by tDCS Vöröslakos et al., 2018).
induced astrocytic Ca 2ϩ surges  support the idea that astrocytic activation exerts effects on microglia. While this is intriguing, neither the prazosin nor the IP 3 R2 KO mouse experiment is cell type specific; therefore, it is possible that direct noradrenergic activation of microglia causes soma enlargement. Indeed, B2AR inhibition by ICI181551 also disrupted microglial somatic enlargement. Functional and transcriptomic evidence underwrites the enriched expression of B2ARs in microglia (Tanaka et al., 2002;Gyoneva and Traynelis, 2013;Zhang et al., 2014). By imaging microglial morphology in awake mice, we found that tDCS attenuates microglial motility. This effect was also dependent on B2ARs, but not on A1ARs. The inhibitory effect of microglial B2ARs on motility is consistent with the in vitro observation by Gyoneva and Traynelis (2013) and recent in vivo observations in awake mice (Liu et al., 2019;Stowell et al., 2019). It is tempting to speculate that the brake on microglial surveillance creates an opportunity for relevant synapses to establish an initial stage of synaptic plasticity. Microglia have been demonstrated to be a source for BDNF (Parkhurst et al., 2013), a pivotal neurotrophin for synaptic plasticity and neurogenesis. Interestingly, tDCS upregulates Bdnf (de Souza Nicolau et al., 2018), promotes BDNF-dependent synaptic plasticity (Fritsch et al., 2010), and causes epigenetic modification to Bdnf genomic regions (Podda et al., 2016). It remains to be shown whether BDNF synthesis is promoted by (nor)adrenergic activation as is reported in astrocytes (Juri et al., 2008). In addition to astrocyteneuron interactions Cocco et al., 2018), our results advocate for the inclusion of microglia as a functional component of the tDCS mechanism via adrenergic receptor activation.
One of the limitations of the current study is the lack of microglia-specific molecular manipulations. While it remains undetermined whether the microglial changes observed in this study have causal roles for positive outcomes of tDCS, several groups have consistently reported inflammation-associated microglial soma enlargement (Chen et al., 2012;Kozlowski et al., 2012;Kongsui et al., 2015). Brain inflammation activates microglia and leads to the production of proinflammatory molecules such as TNF-␣, IL-1␤, and IL-6 (Hanisch, 2002). It is possible that these cytokines are involved in the synaptic plasticity induced by tDCS. For instance, it has been demonstrated that the glial TNF-␣ has a pivotal role in the regulation of homeostatic synaptic plasticity (Stellwagen and Malenka, 2006). Future studies should address the causal relationship, for instance by microglial B2AR knock-out mice combined with tDCS and behavioral performance.