Anodal transcranial direct current stimulation boosts synaptic plasticity and memory in mice via epigenetic regulation of Bdnf expression

The effects of transcranial direct current stimulation (tDCS) on brain functions and the underlying molecular mechanisms are yet largely unknown. Here we report that mice subjected to 20-min anodal tDCS exhibited one-week lasting increases in hippocampal LTP, learning and memory. These effects were associated with enhanced: i) acetylation of brain-derived neurotrophic factor (Bdnf) promoter I; ii) expression of Bdnf exons I and IX; iii) Bdnf protein levels. The hippocampi of stimulated mice also exhibited enhanced CREB phosphorylation, pCREB binding to Bdnf promoter I and recruitment of CBP on the same regulatory sequence. Inhibition of acetylation and blockade of TrkB receptors hindered tDCS effects at molecular, electrophysiological and behavioral levels. Collectively, our findings suggest that anodal tDCS increases hippocampal LTP and memory via chromatin remodeling of Bdnf regulatory sequences leading to increased expression of this gene, and support the therapeutic potential of tDCS for brain diseases associated with impaired neuroplasticity.

Numerous studies provided molecular and functional evidence linking synaptic plasticity in certain brain areas to specific motor and cognitive functions 1,2 . Additionally, altered synaptic plasticity has been associated to neuropsychiatric disorders 3 and cognitive impairment 4 . Thus targeting synaptic plasticity may provide a major breakthrough for therapeutic interventions.
A very promising approach is the use of non-invasive brain stimulation techniques and, in particular, transcranial direct current stimulation (tDCS) due to its ease of use and beneficial effects on brain functions in healthy subjects as well as in patients with neurological and psychiatric diseases [5][6][7][8][9] .
Changes in neuronal excitability and behaviors were reported to persist after tDCS [14][15][16] thus raising the question on whether synaptic plasticity phenomena are involved in tDCS's action 17 . So far, a direct link between tDCS and synaptic plasticity was proposed by very few studies, including ours. Fritsch et al. 18 showed that anodal DCS applied to mouse motor cortex in vitro induced NMDA receptor (NMDAR)-dependent long-term potentiation (LTP) when the stimulus was given concomitantly with synaptic activation. We demonstrated that anodal stimulation applied to hippocampal slices increased the magnitude of LTP at CA3-CA1 synapses whereas cathodal DCS decreased it 19 . More recently, Rohan et al. 20 reported that in vivo tDCS increased LTP and paired-pulse facilitation in rat hippocampus.
To advance our understanding of tDCS action on brain functions we studied the effects and the duration of anodal tDCS on hippocampal synaptic plasticity at electrophysiological, molecular and behavioral levels.
We focused our attention on molecular mechanisms potentially responsible for long-lasting effects. Bdnf was considered a good candidate given its involvement in LTP and memory 21,22 . The structure of Bdnf, i.e., eight 5′ non-coding exons and a 3′ coding exon (IX) under the control of multiple promoters, allows differential regulation of gene expression by several stimuli 23,24 . In particular, memory formation is accompanied by changes in chromatin structure involving enhanced histone 3 (H3) acetylation on specific Bdnf promoters, leading to increased expression of exon I and IX mRNAs 25 . These data grounded the rationale to test whether anodal tDCS affected Bdnf expression via epigenetic modifications, in a fashion resembling neuronal activation.

Results
Long-term potentiation is enhanced in mouse hippocampus following anodal tDCS. We previously demonstrated that in vitro DCS of hippocampal slices induced polarity-dependent modulation of LTP at CA3-CA1 synapses 19 . As cornerstone data for further analysis we first ascertained whether tDCS (350 μA) applied for 20 min to the skull of behaving mice and targeting the hippocampal formation elicited similar effects on hippocampal LTP.
Field evoked post-synaptic potentials (fEPSPs) were measured in the CA1 area after standard high-frequency stimulation (HFS) of Schaffer collaterals and LTP was measured in slices from both sham-(control) and tDCS-stimulated mice that were sacrificed soon after the tDCS (Fig. 1).
A set of experiments was also performed to check whether unilateral tDCS also affected LTP in the contralateral hippocampus. Results showed that LTP values in the right hippocampus were slightly smaller but not significantly different from those obtained in the left side (P = 0.83, unpaired Student's t-test), thus suggesting that tDCS also affected the hippocampus contralateral to the stimulating electrode. Control experiments were performed to assess the specificity of the observed effects. To this aim we recorded LTP at CA3-CA1 synapses in slices obtained from mice that had undergone sham stimulation (n = 8 slices from 2 mice) or tDCS (n = 6 slices from 2 mice) of the left primary motor cortex 26 . Under these experimental conditions no changes in hippocampal LTP were observed following tDCS (P = 0.59 vs. control, unpaired Student's t-test; Supplementary Fig. 1c). Further control experiments were performed to assess tissue viability following tDCS. Histological evaluation by hematoxylin-eosin staining was carried out in 6 mice (n = 3 control and n = 3 receiving anodal tDCS; Supplementary Fig. 1d) and it revealed no signs of neurotrauma 27 .
Our subsequent studies focused on anodal tDCS effects given their potential use to counteract impaired synaptic plasticity in normal aging or disease. Thereinafter animals receiving anodal tDCS are referred to as "tDCS-mice".
To check whether enhanced LTP following anodal tDCS was due to changes in basal synaptic transmission, we first examined input-output (I/O) curves obtained by plotting fEPSP amplitude against stimulus intensities. No significant differences were found between I/O curves of control and tDCS-mice (same slices used for LTP; P > 0.5 at all stimulus intensities, two-way repeated measure [RM] ANOVA followed by Bonferroni post-hoc test; Fig. 2c). Similar results were obtained by plotting the amplitude of presynaptic fiber volleys against the amplitude of fEPSPs (P = 0.86, unpaired Student's t-test; Fig. 2d).
Next, we studied the paired-pulse ratio (PPR) that provides information on neurotransmitter release probability 28 . Stimulation of the Schaffer collateral consistently produced paired-pulse facilitation in all slices tested and the PPR was not significantly different between the two groups (PPR: 1.68 ± 0.07 in slices from tDCS-mice vs. 1.64 ± 0.07 in controls; P = 0.72, unpaired Student's t-test; Fig. 2e). Finally, in whole-cell patch-clamp recordings we measured the ratio of AMPAR-mediated evoked post-synaptic currents (EPSCs) to NMDAR-mediated EPSCs, that is a standard test to detect changes in synaptic strength 29,30 . This parameter was not significantly affected by tDCS (1.0 ± 0.2 [n = 14 cells from 3 tDCS-mice] vs. 0.9 ± 0.3 [n = 12 cells from 3 control mice]; P = 0.13, Mann-Whitney test; Fig. 2f).
Collectively, these data indicate that effects of anodal tDCS on LTP are independent of changes in basal synaptic efficiency.
Spatial learning and memory are improved in mice subjected to anodal tDCS. Having shown that anodal tDCS increased hippocampal LTP, we determined whether hippocampal-dependent learning and memory were similarly affected by comparing the performance of control and tDCS-mice in the MWM and NOR tasks. These behavioral tests were started 24 h after the end of tDCS or sham stimulation (Fig. 1).
In the acquisition session of the MWM all mice successfully acquired the task with latency to reach the platform decreasing progressively across training days (main effect of days: F 3,48 = 37.8, P < 0.001, two-way RM ANOVA; Fig. 3a). Notably, tDCS-mice performed better than controls (n = 9 mice/group; F 1,48 = 21.2, P < 0.001 for overall acquisition). As far as the differences we observed on day 1 (main effect of tDCS: F 1,80 = 10.5, P = 0.005, two-way RM ANOVA), there were no significant differences between control and tDCS-mice in the first two (a) Escape latency to reach the hidden platform of control and anodal tDCS-mice (n = 9 for each group; F 1,48 = 21.2, P < 0.001 for overall acquisition; main effect of days: F 3,48 = 37.8, P < 0.001, two-way RM ANOVA). (b) Time spent in the four quadrants during probe test performed on day 5 of MWM test. In particular, tDCS-mice displayed a significant increase in the time spent in the target quadrant (black bars) compared to control animals (40.6 ± 3.0 s vs. 30.3 ± 3.3 s, P < 0.001; F 3,48 = 4.7, P = 0.006 for interaction between tDCS and time in quadrants; two-way RM ANOVA, Bonferroni post-hoc). (c) Preference toward the novel object in NOR paradigm (n = 9 for each group; P = 0.04, unpaired Student's t-test). Data are expressed as mean ± s.e.m. *P < 0.05; **P < 0.001. trials (P = 0.29 and P = 0.28 for trial 1 and trial 2, respectively, two-way RM ANOVA, Bonferroni post-hoc) indicating equal baselines. Similar results were obtained analyzing the length of the swim path (main effect of days: F 3,48 = 31.7, P < 0.001; main effect of treatment: F 1,48 = 14.1, P = 0.002; day 1: P = 0.67 and P = 0.41 for trial 1 and trial 2, respectively, two-way RM ANOVA, Bonferroni post-hoc; Supplementary Fig. 2).
In the probe test performed 24 h after the last training trial all mice showed a preference for the target quadrant (main effect of quadrant: F 3,48 = 71.8, P < 0.001, two-way RM ANOVA). Notably, tDCS-mice spent more time in the target quadrant compared to controls (40.6 ± 3.0 s vs. 30.3 ± 3.3 s, P < 0.001; F 3,48 = 4.7, P = 0.006 for interaction between tDCS and time in quadrants; two-way RM ANOVA, Bonferroni post-hoc; Fig. 3b), thus indicating improved spatial memory.
Furthermore, in the NOR tDCS-mice showed a greater preference toward the novel object than control mice (preference index: 75.1 ± 2.3% and 66.5 ± 3.4%, respectively; n = 9 for each group, P = 0.04, unpaired Student's t-test; Fig. 3c), supporting the beneficial effects of anodal tDCS on learning and memory.
Effects of anodal tDCS on hippocampal LTP, learning and memory persist one week after stimulation. To gain information on the duration of anodal tDCS effects on hippocampal plasticity we repeated LTP experiments and behavioral tests 1 week after tDCS (Fig. 1). Interestingly, LTP elicited by HFS in slices from tDCS-mice was significantly higher compared to that of control slices (95.5 ± 3.8% [n = 11 slices from 6 tDCS-mice] vs. 78.6 ± 6.1% [n = 9 slices from 5 control mice], P = 0.02, unpaired Student's t-test; Fig. 4a,b).
Consistent with the view that hippocampal-dependent learning and memory rely on LTP 31,32 , tDCS-mice also performed better than controls in MWM and NOR tests started 1 week after tDCS (Fig. 4c-e). In the MWM all mice displayed decreases in the latency to reach the hidden platform over training days (main effect of days: F 3,42 = 28.5, P < 0.001, two-way RM ANOVA), however tDCS-mice showed reduced latency to reach the hidden platform than controls (n = 9 control mice and n = 7 tDCS-mice; F 1,42 = 6.8, P = 0.021 for overall acquisition; Fig. 4c).
The persistence of effect of tDCS on spatial memory was then confirmed by enhanced time spent in the target quadrant by tDCS-mice compared to controls (39.7 ± 2.3 s vs. 30.7 ± 3.2 s, respectively, P = 0.0064, two-way RM ANOVA, Bonferroni post-hoc; Fig. 4d). Interaction between tDCS and quadrant was statistically significant (F 3,42 = 2.9, P = 0.045, two-way RM ANOVA).

Anodal tDCS induces long-lasting changes in
Bdnf expression by epigenetic mechanisms. Our next step was to determine the molecular mechanisms underlying the long-lasting effects of tDCS on hippocampal plasticity. Given the critical role of Bdnf in neuroplasticity-related events 22,23 , we checked if this neurotrophin was target of tDCS action.
These results were validated by quantitative Real-Time PCR (qRT-PCR) analyses focused on Bdnf common 3′ coding exon IX and on exons I and IV that are highly responsive to neuronal activity 23,24 . Results confirmed tDCS-induced enhancement of Bdnf exon I and IX mRNAs (Fig. 5b). In particular, exon I mRNA levels were ~9.5-fold higher in tDCS-mice than in controls 24 h after stimulation (n = 3 mice for each group; P < 0.001, Mann-Whitney test) and ~7-fold higher 1 week after tDCS (n = 3 mice for each group P < 0.001, Mann-Whitney test; Fig. 5b). At both time points we also found enhanced expression of Bdnf exon IX in tDCS mice (~3.7-fold and ~3.0-fold higher than controls at 24 h and 1 week, respectively; P < 0.001, Mann-Whitney test; Fig. 5b). Conversely, the levels of Bdnf exon IV transcripts were similar in control and tDCS-mice confirming that the induction of this exon was not affected by tDCS (P = 0.37 at 24 h and P = 0.7 at 1 week, unpaired Student's t-test; Fig. 5b).
Next, we checked whether increased expression of Bdnf transcripts following tDCS was accompanied by enhanced levels of the protein. Enzyme-Linked Immunosorbent Assay (ELISA) showed that 24 h as well as 1 week after tDCS hippocampal Bdnf levels were ~1.8 fold higher in tDCS-mice than in controls (n = 4 for each group/ time point; P = 1.08E-06 at 24 h, unpaired Student's t-test; P < 0.001 at 1 week, Mann-Whitney test; Fig. 5c). No changes in Bdnf levels were detected in non-stimulated areas such as the cerebellum ipsilateral to the stimulation side ( Supplementary Fig. 4a). Further control experiments showed that other neuronal genes (e.g., NeuN) were not affected by tDCS. Indeed Western blot analysis of hippocampal extracts from tDCS and control mice revealed no significant differences in NeuN expression, thus suggesting that Bdnf is specific target of tDCS ( Supplementary  Fig. 4b,c).
With regard to the mechanism leading to enhanced Bdnf expression it is well known that neuronal plasticity and memory formation are tightly modulated by the epigenetic regulation of gene expression via histone modifications at particular gene promoter regions 33 . We hypothesized that anodal tDCS induced membrane depolarization mimicking neuronal activation and triggered epigenetic changes at Bdnf, thus favoring its transcription. Notably, enhancement of Bdnf exon I, IV and IX transcripts and increased acetylation of histones at promoter I have been previously reported following neuronal activation 23,24,[34][35][36] . As such, we examined whether changes in H3 acetylation at lysine 9 (H3K9ac) on Bdnf promoter I occurred after tDCS.
All together these results indicate that Bdnf expression in the hippocampus is induced by anodal tDCS and that enhanced acetylation at Bdnf promoter I is likely responsible for such effect. We then asked whether tDCS effects might involve CREB/CREB-binding protein (CBP) signaling. Indeed, one pathway mediating neuronal activity-dependent histone acetylation involves activation of CREB that, once phosphorylated at Ser133 (pCREB 133 ), binds the transcriptional coactivator CBP which acts as a histone acetyltransferase (HAT) thus altering chromatin structure of several genes including CREB itself and Bdnf 36,37 . We first checked whether anodal tDCS increased pCREB 133 levels given that this stimulation has been reported to enhance intracellular Ca 2+ signals 38 , that play a critical role in promoting CREB phosphorylation 39 . Hippocampi from control (n = 3) and tDCS-mice (n = 5) were collected two hours after tDCS and pCREB 133 was assessed by Western blot analysis. Results showed that pCREB 133 levels were significantly higher in hippocampal extracts from tDCS-mice than control mice (P = 0.015, unpaired Student's t-test; Fig. 6a,b and Supplementary Fig. 5).
Bdnf promoter I contains a CRE region proximal to the transcription start site and within the sequence amplified by the primers we designed for ChIP experiments 23 (Supplementary Table 2). ChIP analyses were then carried out to assess CREB binding and CBP recruitment to Bdnf promoters. Results showed tDCS-dependent increases in CREB binding to the promoter I (+ 300%; P < 0.0001, Mann-Whitney test) and CBP recruitment to promoter I (+ 181%; P < 0.00001, Mann-Whitney test; n = 3 mice for each group; same samples used to evaluate H3K9ac; Fig. 6c). Collectively, these data suggest that anodal tDCS induced epigenetic changes at Bdnf promoters likely relying on a mechanism involving CREB activation, CBP recruitment and H3K9 acetylation.
To establish a causal link between enhanced acetylation and tDCS effects on plasticity we treated mice with either vehicle or the p300/CBP HAT inhibitor, curcumin 40,41 (50 mg/kg body weight, i.p.; see Fig. 7a and Methods for injection schedule), and subjected animals to anodal tDCS or sham stimulation. Twenty-four hours after stimulation animals were engaged in the NOR and MWM (different set of animals were used for each behavioral test). Control experiments showed that curcumin treatment efficiently inhibited histone H3 acetylation ( Supplementary Fig. 6).
A separate cohort of mice injected with curcumin was tested in the MWM. In line with data obtained with the NOR, curcumin treatment hindered tDCS effects. Indeed no significant differences were found between sham-stimulated and tDCS-mice (n = 9 and n = 8, respectively) injected with curcumin as to: i) the latencies to navigate to the platform (F 1,45 = 0.0007, P = 0.98 for overall acquisition, two-way RM ANOVA; Fig. 7c); ii) the time spent in the target quadrant (27.5 ± 3.2 s in tDCS-mice and 25.4 ± 3.4 s in controls; P = 0.51, two-way RM ANOVA, Bonferroni post-hoc; Fig. 7d). Statistical analysis revealed that overall performance of the two groups of mice in both acquisition and probe test phase was not significantly impaired by curcumin treatment (acquisition: main effect of days: F 3,45 = 28.83, P < 0.001; probe test: main effect of quadrant: F 3,45 = 22.36, P < 0.001; time in the target quadrant vs. the other quadrants: P < 0.05 for both groups), thus supporting the view that blockade of HAT activity specifically interferes with tDCS effects on cognitive performances.
To further understand the molecular mechanisms underlying tDCS effects on plasticity we sought to establish a causal link between the increased levels of Bdnf and tDCS effects. Mouse treatment with the inhibitor of the Bdnf receptor TrkB, ANA-12 43 (0.5 mg/kg body weight, Fig. 8a), reduced the facilitatory effects of tDCS on learning and memory assessed by NOR and MWM tests. In the NOR test performed on vehicle-injected animals the preference index was 68.0 ± 3.2% and 76.6 ± 1.3% in sham-stimulated and tDCS-mice, respectively (n = 8 for each group, P = 0.008 two-way ANOVA, Bonferroni post-hoc); in ANA-12-treated animals the preference index was 63.6 ± 2.0% and 67.8 ± 2.3% in sham-stimulated and tDCS-mice, respectively (n = 8 for each group; P = 0.18, two-way ANOVA, Bonferroni post-hoc; Fig. 8b). Two-way ANOVA showed that recognition memory was not significantly impaired by ANA-12 treatment (P = 0.16 vehicle-injected vs. ANA-12-injected control groups, Bonferroni post-hoc).
In the MWM treatment with ANA-12 did not affect overall performance in all mice (n = 7 tDCS-mice and n = 7 sham-stimulated mice; F 3,36 = 38.45, P < 0.001 for overall acquisition; two-way RM ANOVA) but prevented the tDCS-induced enhancement in learning the location of the hidden platform (P > 0.3 at each training days; tDCS-mice vs. controls, two-way RM ANOVA, Bonferroni post-hoc; Fig. 8c). Consistently, memory performance of stimulated mice treated with ANA-12 was not significantly improved by tDCS (24.7 ± 1.9 s vs. 22.2 ± 2.0 s in control mice; P = 0.19, two-way RM ANOVA, Bonferroni post-hoc; Fig. 8d). Statistical analysis showed that spatial memory was not impaired by ANA-12 treatment (probe test: main effect of quadrant: F 3,48 = 51.94, P < 0.001; time in the target quadrant vs. the other quadrants: P < 0.005 for both groups; Fig. 8d).
These results strongly support our hypothesis that increased histone acetylation promoting Bdnf transcription plays a major role in anodal tDCS-induced enhancement of synaptic plasticity.
To investigate molecular pathways downstream of Bdnf/TrkB signaling potentially responsible for tDCS-induced increased synaptic plasticity we examined phosphorylation of Glycogen synthase kinase-3β (GSK-3β ), that is a key regulated substrate of TrkB. Indeed, GSK-3β is highly expressed in the rodent hippocampus where it has been involved in the modulation of synaptic plasticity 44 . Notably, TrkB activation produces GSK-3β inhibition by its phosphorylation at Serine 9 leading to enhanced LTP 45 . Hippocampi from control (n = 7) and tDCS-mice (n = 7) were collected 24 hours after tDCS and pGSK-3βSer 9 levels were assessed by Western blot. Results showed that pGSK-3βSer 9 levels were significantly higher in hippocampal extracts from tDCS-mice than control mice (P = 0.004, unpaired Student's t-test; Fig. 8e,f). Interestingly, the tDCS-induced enhancement of pGSK-3βSer 9 was not observed in tissues obtained from mice treated with ANA-12 (n = 3 for each group; P = 0.52, Fig. 8e,f). The ability of ANA-12 to prevent TrkB activation in our experimental model was confirmed by Western blot analysis showing that the levels of TrkB phosphorylated at Tyrosine816 was reduced in hippocampi from mice treated with ANA-12 compared to vehicle-injected mice (n = 5 ANA-12 injected mice vs. n = 4 vehicle-injected mice; P = 0.0006; unpaired Student's t-test; Supplementary Fig. 7).

Discussion
The present study elucidated the impact of anodal tDCS on hippocampal synaptic plasticity and shed light on molecular mechanisms underlying the learning and memory improvement we observed in the stimulated mice.  Results showed that LTP was increased in hippocampal slices from mice subjected to anodal tDCS. tDCS-mice also performed better in the MWM and NOR tests. These effects were seen within hours after tDCS and, remarkably, they persisted 1 week. TDCS effects were due to enhanced exon-specific mRNA expression of Bdnf. The hippocampi of tDCS-mice showed increased pCREB Ser133 levels along with epigenetic modifications including CREB binding to the Bdnf promoter I, recruitment of the HAT CBP leading to enhanced H3K9ac on Bdnf promoter I. These results provide novel insights into molecular mechanisms underlying tDCS effects whose understanding is required for a rational use of this technique to potentiate/ameliorate brain functions.
The tDCS intensity we used was below the safety limits of this stimulation 27,38 , as such, no morphological alterations were found in brain tissues subjected to tDCS. The montage we used has been widely employed in rodents and it has been proven to specifically target the cortical brain area under the stimulating electrode and to reproduce results of tDCS in humans [46][47][48] . Furthermore, previous studies using similar devices in rodents demonstrated that the path of current flowing between the electrodes penetrates not only the cortex but also sub-cortical structures including the hippocampus 20,49,50 . Although tDCS likely affected the brain cortex underneath the stimulation electrode, the neurophysiological, behavioral and molecular changes investigated in our study were all related to hippocampal function. Indeed, anodal tDCS enhanced LTP at hippocampal CA3-CA1 synapses and improved spatial and recognition memory assessed by two validated behavioral tests of hippocampal-dependent memory, i.e., MWM and NOR 51,52 .
Our finding that LTP values in the hippocampus contralateral to the stimulating electrode were not significantly different from those of the ipsilateral one might be related to input conveyed by hippocampal commissural connections.
On the other hand in control experiments no changes in hippocampal LTP were found when stimulation was delivered over the primary motor cortex, thus supporting the specificity of the observed effects. The finding that cathodal tDCS induced opposite effects on hippocampal LTP confirms our previous observations suggesting polarity-dependent DCS effects on brain plasticity 19 , and further supports the specificity of the stimulation protocol we used. We attributed the effects of tDCS to modifications of synaptic plasticity rather than to changes in synaptic transmission since we found no significant differences between data collected in slices from tDCSand sham-stimulated mice as to: i) input-output curves obtained by plotting fEPSP amplitude against stimulus intensities or against presynaptic fiber volleys, indicating no changes in basal synaptic transmission; ii) PPR, indicating that glutamate release probability at CA3-CA1 synapses was unaffected by tDCS; iii) AMPA/NMDA ratio, indicating that tDCS per se did not elicit changes in synaptic strength as also reported by Fritsch et al. 18 in the motor cortex. Our findings are in agreement with data recently reported by Rohan et al. 20 showing that anodal tDCS enhanced hippocampal LTP in rats without affecting basal synaptic transmission. The same authors also reported a transient effect of tDCS on PPR that did not parallel LTP changes: indeed increased PPR was only found in rats subjected to 30 min recovery after tDCS but not in those housed for 24 hours post-stimulation. In our experimental model no changes in PPR were found, probably because our recordings fell out of the time window of the transient changes observed by Rohan and coworkers and/or because of differences related to the different species we used.
Our findings are supported by previous reports suggesting that: i) DC stimulation modulates LTP elicited by other NIBS techniques [53][54][55] ; ii) tDCS influences LTP-related learning and memory processes 5 ; iii) tDCS effects on motor and cognitive functions are greater when stimulation is paired with "activity" engaging the subject's stimulated brain area 56 .
Remarkably, we found that one 20-min tDCS elicited electrophysiological and behavioural effects clearly detectable 1 week later. These findings suggested us the engagement of molecular pathways retaining memory of stimulation and the possible involvement of epigenetic regulation of plasticity-related genes. Bdnf was the most likely candidate since it is highly activity-regulated through both transcriptional and epigenetic mechanisms especially in the hippocampus, and the neurotrophin Bdnf is a key regulator of synaptic plasticity 34 . Of note, several lines of evidence indicate that the regulation of Bdnf activity is a correlate of hippocampus-dependent learning in vivo 21 and converging studies implicate Bdnf in the responses to tDCS in humans and rodents 18,57,58 .
ChIP experiments performed on the hippocampi of control and tDCS-mice revealed that anodal stimulation increased H3K9ac at Bdnf promoter I. This effect was observed 24 h as well as 1 week after tDCS and was paralleled by enhanced expression of Bdnf exon I and common 3′ coding exon IX specific mRNAs. Consistently with our results histone acetylation levels on Bdnf promoters, especially promoter I, have been shown to affect LTP and long-term memories 59 . Consequently, disruption of HAT impairs several forms of learning and memory, including novel object recognition, spatial learning in the MWM and fear conditioning 60 . Of note, recent work revealed a correlation among impaired LTP, reduced histone acetylation and altered Bdnf/TrkB signaling during aging. Such dysfunctional plasticity and impaired epigenetic pathways have also been observed in the prefrontal cortex and hippocampus of mouse models of depression and stress 25,61 . These data highlight the potential relevance of our findings for using tDCS as tool for epigenetic intervention strategy to treat diseases associated to impaired synaptic plasticity.
The link between the electrophysiological/functional outcomes of anodal tDCS and epigenetic changes at Bdnf promoters is supported by the inhibitory effects we observed following treatment with curcumin that is a specific p300/CBP HAT inhibitor at the concentration we used 41 . In curcumin-treated mice tDCS failed to induce significant increases in: i) learning and memory at both MWM and NOR; ii) LTP at CA3-CA1 synapses; iii) hippocampal Bdnf levels.
The role of Bdnf pathway activation in the improved learning and memory we observed in tDCS-mice is demonstrated by results showing that in vivo administration of the Bdnf TrkB receptor inhibitor, ANA-12, abolished the facilitatory effects of tDCS in both MWM and NOR.
Of note we found that tDCS enhanced GSK-3β phosphorylation at Ser 9 inhibiting the enzyme activity 44,45 and this effect was prevented by ANA-12, indicating its dependence on activation of TrkB. Therefore inhibition of GSK-3β activity is likely one of the mechanism contributing to the LTP enhancement we found following tDCS. Interestingly, a recent study demonstrated that exogenous application of the Val66pro-form but not the Met66pro-form of Bdnf to hippocampal slices inhibited LTP and facilitated long-term depression through dephosphorylation of GSK-3β at Ser 9 62 . This study takes on particular relevance in light of our findings given that Bdnf Val66Met polymorphism is known to affect responses to tDCS in humans 57 .
Collectively, these data support the view that anodal tDCS boosts synaptic plasticity by enhancing Bdnf expression in response to LTP induction protocols and learning. Nevertheless, other plasticity-related genes, that are worth studying in future research, might cooperate with Bdnf in mediating tDCS action. If operating in other brain areas, the molecular mechanism we uncovered here is a plausible substrate for tDCS action on cognitive and motor functions documented in the literature. It also provides the rationale for using tDCS in combination with other training/stimulation protocols converging onto the same target genes. Within this clinical perspective elucidation of the molecular machinery whereby tDCS induces epigenetic changes at Bdnf favoring its transcription is a very critical issue. In this regard we demonstrated that anodal tDCS increases CREB phosphorylation at Ser133 and CREB binding to Bdnf promoter I, which is hyperacetylated following tDCS. Tabuchi et al. 63 reported that a CRE-like response element contributes to the activity-mediated induction of rat Bdnf promoter I whereas Pruunsild and coworkers 24 showed that mutation in CRE did not alter the induction of human Bdnf, thus suggesting differential regulation of promoter I activity by CREB at transcriptional level in humans and rodents. Our study linked tDCS-induced CREB binding at CRE element on promoter I to epigenetic changes likely involving recruitment of the HAT CBP at the same region. TDCS-dependent upregulation of Bdnf exon I mRNA is fully consistent with enhanced transcriptional activity at promoter I. Interestingly, exon IV transcript, that is the most induced Bdnf mRNA in response to neuronal activity, was not sensitive to tDCS, supporting the view that this non-pharmacological brain-treatment elicits exon-specific modulation of Bdnf.
Our data involving CREB/CBP in tDCS-mediated effects on plasticity are consistent with the critical role of CBP in CREB activity and epigenetic regulation of learning and memory 37 . Furthermore, they provide a plausible, yet not necessarily unique, mechanism for the enhanced acetylation and increased exon-specific Bdnf mRNA expression we observed following tDCS.
As for the molecular events linking tDCS to the above reported cascades leading to enhanced acetylation, it is reasonable to hypothesize the involvement of increased Ca 2+ levels. TDCS effects have been directly or indirectly associated with membrane depolarization-dependent increases in intracellular Ca 2+ levels via NMDAR and Figure 9. Model of anodal tDCS-induced chromatin remodeling leading to enhanced hippocampal synaptic plasticity. Based on our data we propose that transient increase in intracellular Ca 2+ during tDCS initiates molecular cascades leading to persistent changes in chromatin structure of Bdnf. These include the phosphorylation of CREB, its binding to Bdnf promoter I and recruitment of CBP. CBP, in turn, promotes H3K9 acetylation of Bdnf (specifically at promoter I) because of its HAT activity. As a result, stimuli such as LTP induction protocol in slices or learning and memory in vivo are more effective in promoting transcription of Bdnf previously primed by anodal tDCS.
Enhanced acetylation at the promoters of Bdnf and other potential target genes provides a molecular substrate whereby tDCS influences synaptic plasticity in the hippocampus leading to enhanced LTP and cognitive performances. This mechanism might likely be engaged in other cortical areas targeted by tDCS protocols used in humans. Given that Bdnf and epigenetic regulation of gene expression are considered promising targets to potentiate synaptic plasticity in physiology and pathology our data ground and support the use of tDCS as viable therapeutic approaches in neuropsychiatric and neurodegenerative diseases.

Methods
Ethics and Animal Use Statement. Male C57bl/6 mice (30-45 day-old) were used and randomly assigned to two main groups: (i) tDCS-treated and (ii) sham-stimulated animals (controls). Different groups of mice were used for each experimental test and time point from tDCS.
All animal procedures were approved by the Ethics Committee of the Catholic University and were fully compliant with Italian (Ministry of Health guidelines, Legislative Decree No. 116/1992) and European Union (Directive No. 86/609/EEC) legislations on animal research.
The methods were carried out in strict accordance with the approved guidelines. The animals were housed under a 12-h light-dark cycle at room temperature (RT: 19-22 °C). Efforts were made to limit the number of animals used and to minimize their suffering.
Electrode implantation and transcranial direct current stimulation. For tDCS stimulation we adopted an electrode montage widely used and well validated in both rat and mouse models consisting in an unilateral epicranial electrode arrangement 48,49 . Specifically, the active electrode consisted of an epicranial implanted tubular plastic jack (inner area = 6.25 mm 2 ; model GB3305-00, Star electronic SPA, Italy) filled with saline solution (0.9% NaCl) just prior to stimulation; the counter electrode was a conventional rubber-plate electrode surrounded by a wet sponge (5.2 cm 2 ; Physiomed Elektromedizin AG, Germany) applied over the ventral thorax by a custom corset. According to the literature this unipolar arrangement prevents currents from bypassing between two juxtaposed epicranial electrodes.
For the epicranial electrode implant, animals were anesthetized with a cocktail of ketamine (70 mg/Kg, intramuscular injection [i.m.]) and medetomidine (1 mg/Kg, i.m.) and temperature during surgery was maintained at 37 °C. The scalp and underlying tissues were removed and the electrode was implanted using a carboxylate cement (3 M ESPE, Durelon, 3 M Deutschland GmbH, Germany). The centre of the active electrode was positioned on the skull over the left hippocampal formation 1 mm left and 1 mm posterior to bregma 26 . After surgery, all animals were allowed to recover for 3-5 days before undergoing tDCS.
TDCS was applied to awake and freely moving mice at a current intensity of 350 μA for 20 min using a battery-driven, constant current stimulator (Stimulus Isolator, model: A385, World Precision Instruments, USA). This intensity corresponded to a current density of 56 μA/mm 2 (0.35 mA/0.0625 cm 2 ) that was within the range of that used previously in rats 48 (57.1 μA/mm 2 ) or mice 38,49 (55.5 μA/mm 2 ). The current intensity was ramped for 10 s instead of switching it on and off directly to avoid a stimulation break effect. The animals were observed during tDCS and in vivo experiments to determine possible abnormal behaviors related to the stimulation. Control animals received sham stimulation in which no current was applied but the animal underwent the same manipulations as in the stimulation condition. We chose to apply tDCS over the left hippocampus given that, experimental evidences suggest that long-term memory processing is strictly dependent on the left hemisphere 65 .
Experimental design and protocol timeline of experiments are summarized in Figs 1, 7a and 8a. Electrophysiological, molecular and behavioral analyses were performed either at short-term (2-24 h) and long-term (7 days) intervals from sham stimulation or tDCS, to assess the duration of effects elicited by a single tDCS session.
Investigators were blinded to the identity of the groups during experiments and analysis.
Histological processing. Histological evaluation was carried out to detect possible current-induced neurotrauma (e.g., oedema, necrosis, haematoma, cellular alterations). At the end of the stimulation session animals were deeply anesthetized with a cocktail of ketamine (80 mg/Kg, i.m.) and medetomidine (1 mg/Kg, i.m.) and perfused transcardially with saline followed by a fixative containing 4% paraformaldehyde in 0.1 M PBS. After post-fixation, brains were removed from the skulls and stored at 4 °C in a high sucrose solution (30% sucrose in 0.1 M PBS) for 2 days. A vibratome (VT1000S, Leica Microsystems) was used to collect serial coronal 40-μm thick sections containing the hippocampus. All sections were further processed for hematoxylin-eosin staining. Images were acquired with Olympus BX3-CBH microscope.
Electrophysiology. Whole-cell patch-clamp and field recordings were performed on hippocampal coronal slices (300-and 400-μm-thick, respectively) as previously described 28,66,67 . Briefly, mice were anesthetized by halothane inhalation (Sigma) and decapitated. The brain was rapidly removed and put in ice-cold cutting solution (in mM: 124 NaCl, 3. Slices were transferred to a submerged recording chamber and continuously perfused with aCSF (flow rate: 1.5 ml/min). The bath temperature was maintained at 30-32 °C with an in-line solution heater and temperature controller (TC-344B, Warner Instruments). Identification of slice subfields and electrode positioning were done with 4× and 40× water immersion objectives on an upright microscope equipped with differential interference contrast and fluorescence optics under infrared illumination (BX5IWI, Olympus) and video observation (C3077-71 CCD camera, Hamamatsu Photonics).
All recordings were made using MultiClamp 700 A amplifier (Molecular Devices). Data acquisition and stimulation protocols were performed with the Digidata 1440 Series interface and pClamp 10 software (Molecular Devices). Data were filtered at 1 kHz, digitized at 10 kHz, and analyzed both online and offline.
Field recordings were made using glass pipettes filled with aCSF (tip resistance 2-5 MΩ ) and placed in the stratum radiatum of the CA1 region. FEPSPs were evoked by stimulation of the Schaffer collaterals with a bipolar tungsten electrode (FHC, USA) connected to a S11 Grass stimulator (Grass Instruments).
The stimulation intensity that produced one-third of the maximal response was used for the test pulses, LTP induction and paired-pulse facilitation protocols. The fEPSP amplitude was measured from baseline to peak.
Before the LTP induction protocol, to check for a possible effect of tDCS on the basal synaptic transmission, I/O curves were obtained i) by recording fEPSPs induced by presynaptic stimulation at intensities ranging from 0 to 50 V (in 5 V-steps); ii) by plotting fEPSP amplitudes against presynaptic fiber volley amplitudes and then comparing the angular coefficients of I/O slopes. On the same synapses paired-pulse facilitation was also examined at 50 ms interpulse intervals.
For LTP recordings, stable baseline responses to test stimulations (once every 20 s for 10 min) were recorded and then HFS was delivered (4 trains of 50 stimuli at 100 Hz, 500 ms each, repeated every 20 s) 19,68 . Responses to test pulse were recorded every 20 s for 60 min to assess LTP. LTP magnitude was expressed as the percentage change in the mean fEPSP peak amplitude normalized to baseline values (i.e., mean values for the last 5 minutes of recording before HFS, taken as 100%). The occurrence of LTP was statistically verified (paired Student's t-test) by comparing mean fEPSP amplitude measured at 55-60 min after HFS relative to baseline responses (Supplementary Table 1). Slices were prepared soon after the delivery of stimulation protocol and LTP recordings started ~2 hours after tDCS or sham stimulation.
Whole-cell voltage-clamp recordings were performed to measure the AMPA/NMDA ratio, expressed as the peak AMPA evoked post-synaptic currents (EPSCs, mean values of 30-50 events) at − 70 mV divided by the peak NMDA EPSCs (mean values of 20-40 events) measured at 50 ms after the onset of the dual EPSC component at + 40 mV.
For this set of recordings the electrodes were filled with internal solution containing (in mM): 135 CsMeSO 3 , 8 NaCl, 10 HEPES, 0.25 EGTA, 2 Mg 2 ATP, 0.3 Na 3 GTP, 0.1 spermine, 7 phosphocreatine and 5 QX-314 (pH: 7.25-7.30; 294-298 mOsm/l). We monitored the access resistance and membrane capacity before and at the end of the experiments to ensure recording stability and the health of studied cells. Recordings were considered stable when the series and input resistances, resting membrane potential, and stimulus artifact duration did not change > 20%.

Behavioral tests.
A set of animals was tested in the MWM and NOR that are the most widely used and standardized tests to investigate hippocampal-dependent learning and memory 51,52 . To better link changes of learning and memory performances to tDCS we avoided challenging the animals in more than one behavioral test. Therefore different groups of mice were used for each behavioral test and time point from tDCS (Fig. 1).
Mice were trained in the MWM to find a platform hidden 1 cm below the surface of a pool (127 cm in diameter) filled with water made opaque with white nontoxic paint. The acquisition training session started 4 days before the test session (probe test) and consisted of six trials a day for 4 consecutive days, during which the animals were allowed to reach the platform within 40 s. Starting points were changed daily and different starting points were used for each trial. A trial lasted either until the mouse had found the platform or for a maximum of 40 s. Mice rested on the platform for 10 s after each trial. Time (latency) needed to navigate to the platform and swim path length were recorded by an automated video tracking system (Panlab Harvard Apparatus). The probe test session was performed 24 h after the last day of the training. In this session, the platform was removed and each mouse was allowed to swim for 60 s; the time spent in each quadrant was measured 69 .
The NOR protocol lasted three consecutive days including a familiarization phase, a training phase and a test phase. On the 1 st day, mice were individually submitted to a single familiarization session of 10 min, during which they were introduced into the empty arena (45 × 45 cm). On the 2 nd day, animals were submitted to a single 10 min session (training phase) during which two identical objects were placed in a symmetric position from the centre of the arena. An explorative behavior was scored when the head of the animal was facing close (> 2 cm away) to the object or any part of the body except the tail was touching the object. The time spent exploring each object was recorded. The animals were returned to their home cages immediately after training. On the 3 rd day, during the test phase, one of the familiar objects used during the training was replaced by a novel object and the animals were allowed to explore freely for 10 min. All objects were balanced in terms of physical complexity and were emotionally neutral. The open-field and the objects were cleaned by 70% alcohol after each session to avoid possible odorant cues. Preference index, i.e., the ratio of the amount of time spent exploring any one of the two items or the novel object over the total time spent exploring both objects, was used to measure recognition memory 70 .
RNA extraction and cDNA synthesis. Total RNA was extracted from the left hippocampi of control and tDCS-mice using QIAzol Lysis reagent (Qiagen) according to the manufacturer's instructions. RNA sample integrity and concentrations were evaluated with the BioPhotometer plus (Eppendorf, Germany). Reverse transcription reactions were performed on equal amounts of RNA (2 μg) with a high capacity cDNA reverse transcription kit (Applied Biosystems, USA).
Scientific RepoRts | 6:22180 | DOI: 10.1038/srep22180 Semiquantitative RT-PCR and quantitative Real-Time PCR. Semiquantitative PCR of the cDNA was performed using Taq Polymerase (Fischer) and primers described by Aid et al. 23 (Supplementary Table 3). The values of the control samples were set to 1.0, and the others were expressed as fold changes relative to the controls. Analyses were performed in triplicate on the left hippocampi obtained from controls and tDCS-mice.
Quantitative Real-Time PCR (qRT-PCR) amplifications were performed using Power SYBR ® Master Mix on AB7500 instrument (Life Technologies) according to the manufacturer's instructions. The thermal cycling profile featured a pre-incubation step of 94 °C for 10 min, followed by 40 cycles of denaturation (94 °C, 15 s), annealing (55-57 °C, 30 s), and elongation (72 °C, 20 s). Melting curves were subsequently generated by heating amplified products at 94 °C for 15 s, cooling to 50 °C for 30 s, followed by slow heating to 94 °C in increments of 0.5 °C. Melting-curve analyses confirmed that only single products had been amplified. The primer sequences coding the Bdnf exons I, IV and IX are shown in Supplementary Table 4. All data were normalized by reference to the amplification levels of the glyceraldehyde-3-phosphate dehydrogenase (Gapdh) gene; a reference dye was included in the SYBR master mix. The thresholds calculated by the software were used to calculate specific mRNA expression levels using the cycle-at-threshold (Ct) method, and all results are expressed as fold changes (compared to control) for each transcript, employing the 2 -ΔΔCt approach.
Chromatin Immunoprecipitation. Chromatin immunoprecipitation (ChIP) assays were performed as previously described with minor modifications 69 . Animals were sacrificed 24 h or 1 week after the end of the stimulation protocol (Fig. 1). Mice were anesthetized with a cocktail of ketamine (80 mg/Kg, i.m.) and medetomidine (1 mg/Kg, i.m.) and transcardially perfused with an oxygenated Ringer's solution (pH: 7.3), followed by 4% freshly depolymerized paraformaldehyde in 0.1 M PBS (pH:7.4). The brain was post-fixed overnight at 4 °C, and then transferred to a solution of 30% sucrose in PBS for 2 days. Coronal brain sections (45-μm-thick) containing hippocampi were then cut with a vibratome (VT1000S) and floated in ice-cold PBS.
Left hippocampi were isolated under optic microscope and minced through a 10 ml-syringe with decreasing needle size (18 to 22 gauge). Tissue lysate was resuspended in 200 μl lysis buffer containing SDS (1%), Tris-HCl (50 mM, pH 8.1), and EDTA (10 mM) and sonicated on ice with six 10-s pulses with a 20-s interpulse interval. Sample debris was removed by centrifugation, and supernatants were precleared by incubation with protein-G Sepharose 4B beads (Sigma) for 1 h at 4 °C. Beads were collected by centrifugation and supernatants were subjected to immunoprecipitation. Two μg of specific antibody (anti-pCREBSer 133 , anti-acetyl histone H3K9 from Millipore, anti-CBP from Abcam) or control IgG were added overnight at 4 °C. Immune complexes were collected by incubation with protein-G Sepharose 4B beads for 2 h at 4 °C and subjected to a series of seven sequential washes. Immune complexes were eluted from beads by vortexing in elution buffer (1% SDS and NaHCO 3 0.1 M; pH 8.0). NaCl was added (final concentration 0.33 M), and cross-linking was reversed by incubation overnight at 65 °C. DNA fragments were purified by using the PCR DNA fragments purification kit (Geneaid). The primer sequences for promoter I were designed on the basis of mouse Bdnf structure described by Aid et al. 23 (Supplementary Table 2).
PCR conditions and cycle numbers were determined empirically and each PCR reaction was performed in triplicate. Data are expressed as percentage of input calculated by the "Adjusted input value" method according to the manufacturer's instructions (ThermoFisher Scientific ChIP Analysis). In particular to calculate the Adjusted input the Ct value of input was subtracted by 6.644 (i.e., log2 of 100). Next, the percent input of control and tDCS samples was calculated using the formula: 100*2^(Adjusted input -Ct(ChIP). In the same way, the percent input of IgG samples was calculated using the formula 100*2^(Adjusted input -Ct(IgG).
Semiquantitative PCR amplification of Bdnf regulatory sequences performed on chromatin immunocomplexes showed that tDCS induced enhancement of H3K9 acetylation at promoter I but not at promoters II, III, and IV, thus supporting specificity of tDCS effects (Supplementary Fig. 3b and Supplementary Table 3; same samples used for ChIP assays showed in Fig. 5d).
Protein expression was evaluated and documented by using UVItec Cambridge Alliance. Experiments were performed in triplicate.
Images of Western blots have been cropped for presentation and full-size images are shown in Supplementary  Figs 5  Curcumin and ANA-12 administration. Curcumin (Sigma) was diluted in DMSO 40 (i.e., vehicle). Mice received curcumin 50 mg/kg or vehicle via intraperitoneal (i.p.) injection. Mice were randomly assigned to sham or tDCS-group. Curcumin or vehicle was given 24 h prior to tDCS or sham stimulation, soon before stimulation Scientific RepoRts | 6:22180 | DOI: 10.1038/srep22180 and the following day (Fig. 7a). Two hours later some animals were engaged in: i) the habituation session of the NOR test; ii) the 1 st training session of the MWM; iii) tissue explant for ELISA assay. A week after stimulation (i.e., 6 days after the last curcumin injection) the other mice were used for: i) slice preparation for LTP recordings; ii) tissue explant for ELISA assay (Fig. 7a).
For experiment with ANA-12, mice were divided in two groups receiving sham stimulation or anodal tDCS and subjected to NOR or MWM tests. A volume of 10 μl/g body weight was i.p. injected for vehicle and ANA-12 (0.5 mg/kg body weight) solutions 43 . TDCS was applied 24 h before the 1 st day of either NOR or MWM test (Fig. 8a).
Mice subjected to the NOR test were further distributed into ANA-12 and vehicle-injected groups (1% DMSO dissolved in 0.9% NaCl solution). ANA-12 was injected 4 h before and soon after the training day of the NOR test as well as 4 h before the test phase.
Animals subjected to the MWM received ANA-12 injection 4 h before: i) each of the 4 training sessions and ii) the probe test session.
With regard to behavioral tests, given that statistical analysis revealed no significant effect of vehicle administration on the NOR test (P = 1 vs. naïve, unpaired Student's t-test), we enrolled in the MWM test only two groups of animals, including tDCS-mice and sham-stimulated mice all injected with curcumin or ANA-12.
ELISA measurements. Brain tissues were obtained as described for Western blot and stored at − 80 °C.
Prior to analysis, samples were thawed and then weighed. Lysis buffer (100 mM PIPES pH 7, 500 mM NaCl, 0.2% Triton X-100, 2% BSA, 2 mM EDTA, 200 μM PMSF, 1 × protease, phosphatase-1, and phosphatase-2 inhibitor cocktails from Sigma) was then pipetted into each tube (100 μl per mg of tissue for each left hippocampus). Samples were homogenized, sonicated and centrifuged for 30 min at 16,000 × g at 4 °C. Supernatants were then removed and frozen at − 80 °C until analysis. The concentration of Bdnf was determined using the E-Max ImmunoAssay system (Promega) according to the manufacturer's instructions.
Statistical analysis. Sample sizes were chosen with adequate statistical power (0.8) according to results of prior pilot data sets or studies, including our own, that used similar methods or paradigms. Sample estimation and statistical analysis were performed using the SigmaPlot 12 software. Data were first tested for equal variance and normality (Shapiro-Wilk test) and then the appropriate statistical tests were chosen. The statistical tests used (i.e., Student's t-test, Mann-Whitney test, one-way ANOVA, two-way ANOVA, two-way RM ANOVA) are indicated in the main text and in the corresponding figure legends for each experiment. Post-hoc multiple comparisons were performed with Bonferroni correction.
All statistical tests were two-tailed and the level of significance was set at 0.05. Results are presented as mean ± s.e.m. In the graphs, y-axis error bars represent s.e.m. Analyses were performed in blind.
Criteria of animal exclusion/inclusion were pre-established according to Ethics Committee guidelines but no data were excluded from analysis.