Activation and depression of neural and hemodynamic responses induced by the intracortical microstimulation and visual stimulation in the mouse visual cortex

Transgenic mice expressing the Ca²⁺ indicator in excitatory neurons (Thy1-GCaMP6s; 024275, Jackson Laboratory, ME, USA) were used for simultaneous recordings of Ca²⁺ and hemodynamic signals under the mesoscopic-scale microscopy (n = 8) and for two-photon Ca²⁺ imaging (n = 9) studies. Surgical procedures were conducted following previously established protocols [27, 30, 31]. Briefly, mice were anesthetized with a mixture of xylazine (7 mg kg⁻¹ b.w., i.p.; Covetrus, OH, US) and ketamine hydrochloride (75 mg kg⁻¹ b.w., i.p.; Covetrus). Craniotomies were then performed over the left visual cortex (AP = −1 to −5, LM = 0.5–3.5 mm). Custom stimulation microelectrodes (Q1x4-3mm-50-703-CM16LP, NeuroNexus Technologies; single 3 mm-long shank, 30 µm-diameter four channels, 50 µm site spacing, electrochemically activated iridium/iridium oxide sites [40]) were implanted at an angle of 20–30° slanted towards the posterior region. The electrode was securely fixed to the skull with UV-curable resin (e-on Flowable A2, BencoDental, PA, USA). Reference and ground lines on the electrode connector were wired to the bone screws placed in the skull over the frontal lobes. Kwik-Sil was used to fill the exposed cortical surfaces in the craniotomy to maintain visibility, and a cover glass was used to seal the craniotomy with UV-curable resin. Anesthesia level was periodically monitored and an update (ketamine, 45 mg kg⁻¹ b.w., i.p.) was administered as necessary. Throughout the surgical procedure and recovery period, the mice's body temperature was maintained using a heating water pad and respiration was monitored visually. Following surgery, antipamezole hydrochloride (Antisedan, Zoetis, NJ, US; 1 mg kg⁻¹ b.w., i.p.) and ketoprofen (Ketofen, Zoetis; 5 mg kg⁻¹ b.w., i.p.) were administered for reversal of anesthesia and for analgesia. After recovery, mice were allowed to recover for up to 2 h prior to the recording session. Animals were kept in a 12 h/12 h light/dark cycle. All procedures described here were approved by the Division of Laboratory Animal Resources and Institutional Animal Care and Use Committee at the University of Pittsburgh and performed in accordance with the National Institutes of Health Guide for Care and Use of Laboratory Animals.

Animals were placed in a custom treadmill and head-fixed throughout the imaging experiments. We employed a mesoscopic-scale widefield microscope (MVX-10 epifluorescence microscope; Olympus, Tokyo Japan; 6.3× zoom) for simultaneous recording of Ca²⁺ and intrinsic optical signals over a relatively larger field-of-view of 2.3 mm × 1.7 mm (figure 1(a)). The imaging was conducted with three sequentially interwoven LED illuminations, and barrier filters were placed in front of each LED light source to restrict the spectral band of each LED to the desired wavelength: (1) blue for GCaMP6s excitation (482.5 ± 15.5 nm; #67-028, Edmund Optics Inc., NJ, US); (2) green for measuring hemoglobin-sensitive imaging (532 ± 2 nm; FLH532-4, Thorlabs Inc., NJ, US); and (3) red for differentiating oxy- and deoxy-hemoglobin (633 ± 2.5 nm; FLH633-5, Thorlabs Inc.) [37, 41]. Fluorescence and reflectance intensities were collected through a long-pass filter (>500 nm; ET500lp, Chroma Technology Corporation, VT, US) placed in the emission path before the camera (QImaging Retiga R1; Cairn Research Ltd, Kent, UK) and images were sampled at 30 Hz (10 Hz for each color). The camera was controlled using Micro-Manager software [42, 43] with an imaging matrix size of 344 × 256 pixels (figures 1(b)–(e)).

Zoom In Zoom Out Reset image size

Two-photon microscopy (Bruker, Madison, WI, US) was performed using a laser tuned to 920 nm wavelength (Insight DS+, Spectra-Physics, Menlo Park, CA) for GCaMP6s excitation (laser power ≈ 10–20 mW, figure 11(a)). Resonant Galvo scanning was performed to capture higher resolution images at 30 Hz over a 413 × 413 µm² field of view (512 × 512 pixels) at a single z-plane focused around the stimulation site. In both imaging conditions, each recording was composed of a 30 s pre-stimulation period, a 10 s stimulation period, and a 60 s post-stimulation period.

ICMS was performed using an IZ2 stimulator controlled by an RZ5D base processor (Tucker-Davis Technologies, FL, US). Each pulse consisted of a 200 µs cathodic leading phase followed by a 200 µs anodic phase, with balanced charges between the cathodic and anodic phases. The current was fixed to ±20 µA (4 nC/phase), selected to be lower than the safety limit (k = log(Q) + log(Q/A) ≈ 0.36 < 1.7, where Q and Q/A represent charge per phase [µC/phase] and charge density per phase [µC cm⁻²/phase], and value k below 1.7 can be considered as a safe configuration [44]). Six different train frequencies (10, 25, 50, 100, 250, and 500 Hz) were tested. The order of ICMS trials was repeated two times in ascending-descending frequency order (10–500, 500–10, 10–500, 500–10 Hz).

For comparison, a single blue LED was positioned in front of the animal's eye contralateral to the imaging window as a more natural visual stimulus (n = 5 out of 8 mice in the mesoscopic-scale widefield imaging, n = 5 out of 9 mice in the two-photon imaging). The LED was controlled by the stimulator, applying a square wave at 10 Hz (50 ms on, 50 ms off, 50% duty cycle) with an amplitude of 2.5 V. All stimulation trials were repeated 4 times.

2.4.1. Pre-processing of the widefield and two-photon imaging data

2.4.2. Somatic region of interest (ROI) definition for the two-photon imaging data

2.4.3. Neuronal Ca²⁺ response quantification for widefield and two-photon imaging

To quantify the excitatory Ca²⁺ activity for two-photon imaging, the fluorescence change ratio (dF/F₀ = (F(t) − F₀)/F₀) was calculated, where F(t) represents the fluorescent intensity at time t and F₀ is the mean fluorescent intensity over the pre-stimulation period (−30 ⩽ t < 0, in s).

In the case of the mesoscopic-scale widefield imaging, the Ca²⁺ signals were corrected based on the green channel values to account for the hemodynamic effect that could potentially affect the green GCaMP signals [37, 41] (figures 1(f)–(h)). The corrected dF/F₀(t) was calculated using the formula:

$$\begin{equation}{\text{Corrected}}\,{\text{d}}F/{F_0}\left( t \right) = \frac{{\frac{{F\left( t \right)}}{{{F_0}}}}}{{\frac{{{I_{532}}\left( t \right)}}{{{I_{{{532}_0}}}}}}} - 1\end{equation} \tag{ 1 }$$

where ${I_{532}}\left( t \right)$ represents the reflectance intensity to the green illumination at a specific time point t, and ${I_{{{532}_0}}}$ is the mean of the reflectance intensity to the green illumination during the pre-stimulation period.

2.4.4. Hemodynamic activity quantification of mesoscopic-scale widefield imaging

The change in oxy- and deoxy-hemoglobin concentrations relative to baseline ($\Delta \left[ {{\text{HbO}}} \right]\left( t \right)$ and $\Delta \left[ {{\text{HbR}}} \right]\left( t \right)$, respectively; the solution for the modified Beer–Lambert law) were calculated as follows [37]:

$$\begin{equation}\Delta \left[ {{\text{HbO}}} \right]\left( t \right) = \frac{{{\xi _{{\text{HbR}},633}}\Delta {\mu _{532}}\left( t \right) - {\xi _{{\text{HbR}},532}}\Delta {\mu _{633}}\left( t \right)}}{{{\xi _{{\text{HbO}},532}}{\xi _{{\text{HbR}},633}} - {\xi _{{\text{HbO}},633}}{\xi _{{\text{HbR}},532}}}}\end{equation} \tag{ 2 }$$

$$\begin{equation}\Delta \left[ {{\text{HbR}}} \right]\left( t \right) = \frac{{{\xi _{{\text{HbO}},532}}\Delta {\mu _{633}}\left( t \right) - {\xi _{{\text{HbO}},633}}\Delta {\mu _{532}}\left( t \right)}}{{{\xi _{{\text{HbO}},532}}{\xi _{{\text{HbR}},633}} - {\xi _{{\text{HbO}},633}}{\xi _{{\text{HbR}},532}}}}\end{equation} \tag{ 3 }$$

$$\begin{align}\Delta {\mu _l}\left( t \right) & = - \frac{1}{{{\chi _l}}}\ln \frac{{{I_l}\left( t \right)}}{{{I_{l,{ }0}}}} = {\xi _{{\text{HbO,}}l}}\Delta \left[ {{\text{HbO}}} \right]\left( t \right) \nonumber\\ & \quad + {\xi _{{\text{HbR}},l}}\Delta \left[ {{\text{HbR}}} \right]\left( t \right)\end{align} \tag{ 4 }$$

where constant $\xi$ represents the absolute absorption spectra of HbR and HbO at 532 and 633 nm, constant $\chi$ represents the pathlengths at wavelength l (i.e. l = 532 and 633 nm). I_l(t) and I_l,0 are the reflectance intensities at a time point t and its mean value during the pre-stimulation period for wavelength l. The values of $\xi$ and $\chi$ were obtained from [37], and if an exact wavelength value was not provided, the mean value of the adjacent two values was used.

The change in the total hemoglobin concentration (${{\Delta }}\left[ {{\text{HbT}}} \right]\left( t \right)$) was calculated by summing the changes in oxy- and deoxy-hemoglobin (${{\Delta }}\left[ {{\text{HbT}}} \right]\left( t \right) = {{\Delta }}\left[ {{\text{HbO}}} \right]\left( t \right) + {{\Delta }}\left[ {{\text{HbR}}} \right]\left( t \right)$). The oxygen extraction fraction (OEF; relative concentration of deoxyhemoglobin in total hemoglobin concentration) was estimated using the following:

$$\begin{equation}{\text{OEF}}\left( t \right) = \frac{{\left[ {{\text{HbR}}} \right]\left( t \right)}}{{\left[ {{\text{HbT}}} \right]\left( t \right)}} = \frac{{{{\left[ {{\text{HbR}}} \right]}_0} + {{\Delta }}\left[ {{\text{HbR}}} \right]\left( t \right)}}{{{{\left[ {{\text{HbT}}} \right]}_0} + {{\Delta }}\left[ {{\text{HbT}}} \right]\left( t \right)}},\end{equation} \tag{ 5 }$$

where [HbR]₀ and [HbT]₀ represent baseline concentrations of the deoxyhemoglobin and total hemoglobin. For our estimations, we assumed baseline concentrations of 2 mM for hemoglobin in blood, 3% of blood in brain tissue (60 µM of total hemoglobin), and 75% of oxygenation (15 µM of deoxyhemoglobin) [37]. By definition, real OEF values range between 0 and 1. Therefore estimated OEF values outside this range were excluded as outliers (2.4 ± 1.0% pixels, mean ± SD across all animals).

2.4.5. Extraction of significant activation and depression in the widefield imaging data

To define activation and depression distances (from stimulation site), latencies (onset timing), and durations (offset timing–onset timing) of the spatiotemporal magnitude profile of the corrected dF/F₀, Δ[HbO], and Δ[HbR] while reducing the effect of data fluctuations, we employed hysteresis thresholding. As described above, this method requires two different threshold levels [46]. First, the original three-dimensional data (dF/F₀, Δ[HbO], and Δ[HbR]) obtained under the mesoscopic-scale imaging (X × Y × time) was converted into two-dimensional structure (distance from stimulation site × time). We calculated the mean ± 1 and 2 SDs of the pre-stimulation period (−30–0 s), and used them as the soft and hard thresholds. To determine the activation distance, we identified points on the distance profile with values above 2 SDs as well as points between 1 and 2 SDs that were connected to distances above 2 SDs as above threshold. The same procedure was applied in the time domain, resulting in a joint threshold image (distance × time) that contained multiple continuous regions. The regions extracted by the hysteresis thresholding were considered candidates for significant activation and depression. We selected at most one region for each activation and depression event based on specific criteria. For activation candidates in ICMS experiments, we selected single regions that (1) emerged within 10 s after stimulation onset, (2) occurred within 50 µm from the stimulation site, and (3) had the longest duration among the candidates. For depression candidates in ICMS conditions, we selected single regions that (1) emerged within 20 s after stimulation onset, (2) occurred within 50 µm from the stimulation site, and (3) had the longest duration among the candidates. Similarly, for visual stimulation conditions, we used criteria (1) and (3) regardless of whether the responses originated from the inserted probe. The onset timing of the extracted region is defined as latency, and the difference between onset and offset timings is duration. Activation and depression distances and mean amplitude across the distance are represented as functions of time from stimulation onset.

After defining the distance, latency, and duration of significant Ca²⁺ activation or depression and hemodynamic activities (d[HbO] and d[HbR]) using the hysteresis thresholding method described above, these measurements were averaged across time for each mouse (e.g. symbols in figure 4(d)). Also, the averaged measurements for each mouse were averaged across mice to evaluate these effects as a function of stimulation conditions such as ICMS frequencies and visual stimulation (e.g. bars in figure 6).

The balance of activation and depression was quantified using the following metrics: distance from the stimulated site, response magnitude, and response duration. Balance based on these metrics was calculated as follows:

$$\begin{equation}{\text{Balance}} = \frac{{{M_{{\text{act}}}} - {M_{{\text{dep}}}}}}{{{M_{{\text{act}}}} + {M_{{\text{dep}}}}}},\end{equation} \tag{ 6 }$$

where M_act and M_dep represent metrics for activation and depression, respectively. The resulting score ranged from −1 (biased to depression) to 1 (biased to activation); 0 indicated balanced activation and depression.

2.4.6. Statistical procedures

Although stimulation-induced depression of neuronal excitability (SIDNE) has been previously described [34], most imaging studies have focused on stimulation-induced activation [5, 26, 29, 47]. To investigate the extent of spatiotemporal activation and depression, we first analyzed ICMS-induced changes in Ca²⁺ activity within excitatory neurons using widefield imaging. We first show results from a representative animal to highlight the spatio-temporal changes (figure 2; see movies S1–S3 for entire frames). A 10 s long ICMS train at 10 Hz was delivered to the visual cortex and produced a clear increase in Ca²⁺ activity during the stimulation period, which spread to approximately 500 µm away from the stimulation site (0–10 s; figures 2(a)–(c)). Following the cessation of the stimulation train (at 10 s), the Ca²⁺ activity rapidly declined and dipped below pre-stimulus baseline ('depression') approximately 5 s after the ICMS offset (figure 2(b)) and persisted for several tens of seconds during the post-stimulation period although with a noticeably smaller magnitude than activation. This depression effect was particularly prominent in the region exhibiting an increase in Ca²⁺ (figure 2(c)). The Ca²⁺ data were corrected for hemodynamic absorption (as shown in figures 1(f)–(h)) to mitigate this potential signal source. Similar patterns of activation and depression were also observed with 100 Hz ICMS (figures 2(d)–(f)), but no such effects were observed during visual stimulation (figures 2(g)–(i)), where the visual stimulation train induced transient activation at the onset and offset periods. Visually-evoked activation and depression spread through almost the entire imaging field due to the use of uniform visual field stimulation.

Zoom In Zoom Out Reset image size

The magnitude, duration, and distance of activation and depression were quantified for each animal. The Ca²⁺ responses induced by the 10 Hz ICMS across all animals (figures 3(a)–(c)) showed that the average magnitude of Ca²⁺ activation increased over time, while the average magnitude of Ca²⁺ depression remained relatively low and constant (figure 3(a)). Since Ca²⁺ depression emerged in the distal region from the stimulation site (figure 2(a); see results 3.5) and prolonged Ca²⁺ activation remained around the center region, these temporal periods of Ca²⁺ activation and depression partially overlapped. The activation distance remained constant throughout the ICMS train, followed by varying degrees of depression distance ranging from 300 to 1000 µm away from the stimulation sites (figure 3(b)). Activation was observed in all mice (8/8 mice), while depression was observed in most cases (5/8) (see methods 2.4.5 for definition of activation and depression). Although the depression magnitude was attenuated by the hemodynamic correction (paired t-test, t = 1.75, p = 0.12), it was significantly greater than zero (one-sample t-test, t = 3.03, p < 0.05; figure S2(a)). The depression component in the uncorrected Ca²⁺ signals was significantly stronger than the change observed in the green channel (paired t-test t = 3.00, p < 0.05; figure S2(b)). These observations show that the hemodynamic correction works sufficiently with our data and the Ca²⁺ depression remained even after the hemodynamic correction, meaning that the depression was not a feint due to hemodynamic signal contamination.

The Ca²⁺ activation was significantly stronger than the depression by almost an order of magnitude (comparison of absolute magnitude; figure 3(c), left; t = 3.22, p< 0.05). However, there was no statistical difference in the mean distances between activation and depression (figure 3(c), middle; t = 1.29, p= 0.24). The duration of the activation tended to be longer than the duration of depression (figure 3(c), right; t = 1.22, p= 0.26). It is worth noting that the Ca²⁺ activation lasted slightly longer than the 10 s stimulation period. These results indicate that the 10 s, 10 Hz ICMS train consistently evoked sustained activation throughout the stimulation period, followed by depression to a similar spatial extent. These trends were highly dependent on the stimulation parameters and modalities, where 100 Hz ICMS induced transient activation that was stronger than depression in magnitude (figures 3(d)–(f), t = 7.97, p < 0.05), and visually-induced activation was comparable to depression (figures 3(g)–(i), t = 1.07, p = 0.34).

Zoom In Zoom Out Reset image size

Given that ICMS-induced activation is followed by depression of the Ca²⁺ response, we then asked how the stimulation frequencies can impact features of the induced activation and depression of the Ca²⁺ response (figure 4). In a representative animal, we observed activation-dominant responses to the ICMS where lower-frequency stimulation induced an elevating (or sustained) response and higher-frequency stimulation elicited a transient and localized response. Similar profiles were also observed in all mice (8/8 mice). Meanwhile, the visually-evoked response exhibited transient activation and depression with similar magnitudes (5/5 mice with visual stimulation). These representative examples of the excitatory neural Ca²⁺ responses provide valuable inspiration about the complex interactions between stimulation frequencies and their effects on the Ca²⁺ response.

Zoom In Zoom Out Reset image size

Given the observed frequency-dependent relationship between the magnitude, area, and spatiotemporal aspects of ICMS-induced activation and depression, we conducted further analysis to quantify the differences between stimulation conditions and examine the correlations between activation and depression (figure 5). We determined the response distances and durations of significant activation and depression using the hysteresis thresholding of the mesoscopic-scale Ca²⁺ spatiotemporal dynamics (see Methods 2.4.5). Neither the activation nor depression distances showed any statistical differences among the stimulation conditions (figure 5(a), activation, one-way ANOVA, F = 1.71, p= 0.14; figure 5(b), depression, F = 1.50, p= 0.20). The balance of the activation and the depression distances (equation (6) did not differ among the stimulation conditions (figure 5(c), one-way ANOVA, F = 0.65, p= 0.69). Furthermore, we did not observe any significant correlations between the activation and the depression distances (figure 5(d)).

Zoom In Zoom Out Reset image size

Although the area of ICMS-induced activation did not differ significantly, the magnitudes of activation were significantly different among the stimulation conditions (figure 5(e), one-way ANOVA, F = 3.10, p< 0.05). Lower-frequency ICMS tended to induce stronger activation, while the visual stimulation tended to elicit weaker activation compared to ICMS (post-hoc Tukey's HSD test, 10 Hz ICMS vs visual stimulation, difference = −6.6%, p< 0.05). In contrast, the magnitudes of depression were not statistically different across all stimulation conditions (figure 5(f), one-way ANOVA, F = 1.39, p= 0.24). Consequently, the balance between these two magnitudes were significantly different among the stimulation conditions (figure 5(g), one-way ANOVA, F = 7.02, p< 0.05), with visual stimulation showing balanced activation and depression responses (post-hoc Tukey's HSD test, 10 Hz ICMS vs visual stimulation, difference = −0.75, p< 0.05; 25 Hz ICMS vs visual stimulation, difference = −0.52, p< 0.05; 50 Hz ICMS vs visual stimulation, difference = −0.56, p< 0.05; 100 Hz ICMS vs visual stimulation, difference = −0.52, p< 0.05; 250 Hz ICMS vs visual stimulation, difference = −0.61, p< 0.05; 500 Hz ICMS vs visual stimulation, difference = −0.67, p< 0.05). Moreover, the magnitudes of activation and depression exhibited significantly strong positive correlation in the 10, 50, and 100 Hz ICMS conditions (figure 5(h); Pearson correlation coefficient, 10 Hz ICMS, r = 0.73, p< 0.05; 50 Hz ICMS, r = 0.82, p< 0.05; 100 Hz ICMS, r = 0.75, p< 0.05). These findings indicate that ICMS at lower frequencies elicits stronger Ca²⁺ activation in excitatory neurons, while visual stimulation induces a balanced response of activation and depression. Additionally, the correlations of magnitudes between activation and depression suggest that there may be shared or related mechanisms underlying the processes of activation and depression in certain ICMS conditions.

The activation durations were significantly dependent on the stimulation conditions (figure 5(i), one-way ANOVA, F = 2.36, p< 0.05). All ICMS frequencies, particularly around 100–250 Hz, caused prolonged activation that lasted beyond the duration of the stimulation train (10 s, horizontal dotted line in figure 5(i)). In contrast, visual stimulation only caused transient activation at the onset and offset of the stimulus train (post-hoc Tukey's HSD test, 100 Hz ICMS vs visual stimulation, difference = −11 s, p< 0.05; 250 Hz ICMS vs visual stimulation, difference = −12 s, p< 0.05). Here, we quantified a static aspect of activation based on the thresholding (see Methods 2.4.5), which also extracted weak sustained activity. However, as shown in the representative example (figure 4), stimulation frequency and modality also modulated the activation in a different way, especially in terms of peak magnitudes at each time point (e.g. higher-frequency ICMS induced transient activation). This dynamic aspect is further investigated in the later section (see Results 3.4).

Similar to activation, the duration of depression varied depending on the stimulation conditions (figure 5(j), one-way ANOVA, F = 2.48, p< 0.05), with tendencies of longer depression at 25–50 Hz ICMS and visual stimulation compared to the other conditions (no significant pairwise difference by post-hoc Tukey's HSD test). Consequently, the balance between the activation and depression durations exhibited significant dependence on the stimulation conditions (figure 5(k), one-way ANOVA, F = 4.77, p< 0.05). Depression tended to be longer in the 25 Hz ICMS and visual stimulation compared to the other conditions, while activation dominated in the lower and higher-frequency ICMS conditions, especially at 500 Hz (post-hoc Tukey's HSD test, 10 Hz ICMS vs visual stimulation, difference = −0.89, p< 0.05; 25 Hz ICMS vs 500 Hz ICMS, difference = 0.71, p< 0.05; 250 Hz ICMS vs visual stimulation, difference = −1.00, p< 0.05; 500 Hz ICMS vs visual stimulation, difference = −1.08, p< 0.05). No significant correlation was found between the duration of activation and depression (figure 5(l)). These results suggest that ICMS at any frequency entrains the Ca²⁺ activity, while the degree of subsequent depression is dependent on the stimulation conditions. Together, these quantitative findings inform the selection of stimulation frequencies, where the neural responses include both activation and depression phenomena.

Since it was observed that lower-frequency ICMS caused greater sustained Ca²⁺ activation, while higher-frequency ICMS caused transient and localized activation followed by weak sustained activation (figure 4(i), leftmost panel, 2–10 s), we proceeded to investigate how activation developed during the stimulation train across different conditions (figure 6). In the 10 Hz ICMS condition, the activation gradually increased in magnitude over the 10 s ICMS duration and spread through the visual cortex. On the other hand, in the 100 Hz condition, the Ca²⁺ activity initially peaked, and then decreased over the 10 s ICMS duration, becoming more localized (figures 6(a)–(d)).

Zoom In Zoom Out Reset image size

To quantify the spatiotemporal dynamics of neural Ca²⁺ activation, we measured the distance from stimulation site to activation edge at each frame and calculated the spread/withdrawal speed based on the inter-frame difference of the distance (0.5 s interval). For the 10 Hz condition, the activation distance remained relatively stable over the 10 s ICMS period, resulting in a near-zero spreading speed (figure 6(e)). In contrast, the activation distance rapidly decreased after 2 s of the 100 Hz ICMS, leading to a negative speed (withdrawal of activation) observed 4–5 s after the onset. The mean spread/withdrawal speeds significantly varied across the stimulation conditions (figures 6(f) and (g), F = 2.78, p< 0.05). Specifically, the speed tended to be positive in the 10 Hz ICMS, while it was negative in the other stimulation conditions (post-hoc Tukey's HSD test, 10 Hz ICMS vs 500 Hz ICMS, difference = −26 µm s⁻¹, p< 0.05). This indicates that, apart from the 10 Hz ICMS frequency, the Ca²⁺ activation area gradually withdrew after the initial transient spread in most stimulation conditions.

We next investigated whether the activation magnitude followed the same trend. Analysis of the magnitudes revealed that the 10 Hz ICMS induced an increase in magnitude over the stimulation period, while the 100 Hz ICMS condition showed a relatively constant magnitude (figure 6(h), top). The change rates of magnitude between each time point remained mostly positive (indicating increasing magnitude) in the 10 Hz conditions, whereas the change rate fluctuated around zero (indicating stable magnitude) in the 100 Hz ICMS condition (figure 6(h), bottom). The mean change rates of magnitude showed significant differences among the stimulation conditions (figures 6(i) and (j), one-way ANOVA, F = 6.99, p < 0.05). Specifically, lower-frequency ICMS induced elevating activation magnitude (post-hoc Tukey's HSD test, 10 Hz ICMS vs 25 Hz ICMS, difference = −0.3% s⁻¹, p < 0.05; 10 Hz ICMS vs 50 Hz ICMS, difference = −0.5% s⁻¹, p < 0.05; 10 Hz ICMS vs 100 Hz ICMS, difference = −0.6% s⁻¹, p < 0.05; 10 Hz ICMS vs 250 Hz ICMS, difference = −0.5% s⁻¹, p < 0.05; 10 Hz ICMS vs 500 Hz ICMS, difference = −0.5% s⁻¹, p < 0.05; 10 Hz ICMS vs visual stimulation, difference = −0.5% s⁻¹, p < 0.05). Also, individual neurons exhibited similar temporal dynamics and dependency on stimulation conditions (figure S5). Taken together, our findings demonstrate that the lower-frequency ICMS increases both activation area and magnitude over the stimulation period, whereas the higher-frequency ICMS causes a decrease in the activation area and a relatively constant magnitude.

Given the stimulation-parameter-dependent spatiotemporal dynamics of activation during sustained ICMS, we next investigated the spatiotemporal dynamics of the depression (figure 7). As shown in previous analyses (e.g. figures 4(a)–(f)), the depression onset was initiated in a distal location from the stimulation site. To quantify the spatiotemporal dynamics of depression, we defined the nearest 'edge distance' as the distance to the closer edge of the depression to stimulation site by conducting the hysteresis thresholding (see methods 2.4.5). When examining the nearest edge distances, all ICMS conditions exhibited a decrease in the first few seconds of depression, especially at higher ICMS frequencies (figure 7(a)). This indicated that the depression initially occurred in distal regions and then spread backwards towards the stimulation site. We compared the nearest edge distances between the initial time point of depression onset and the end of depression (up to 10 s after the depression onset) among the different stimulation conditions (figure 7(b)). Even though the initial nearest edge distances varied among the stimulation conditions, the last nearest edge distances decreased in all conditions. Two-way ANOVA revealed significant differences in the nearest edge distances of the neural Ca²⁺ depression among the stimulation conditions and between the initial and final time points (stimulation condition, F = 4.80, p< 0.05; time point, F = 4.91, p< 0.05; stimulation condition × time point, F = 0.29, p= 0.94; post-hoc Tukey's HSD test, 10 Hz ICMS vs visual stimulation, difference = 479 µm, p< 0.05; 25 Hz ICMS vs visual stimulation, difference = 526 µm, p< 0.05; 50 Hz ICMS vs visual stimulation, difference = 610 µm, p< 0.05; 100 Hz ICMS vs visual stimulation, difference = 613 µm, p< 0.05; initial vs last, difference = −168 µm, p< 0.05). However, the speeds of the nearest edge distance change did not significantly differ across stimulation conditions (figures 7(c) and (d), one-way ANOVA, F = 0.70, p= 0.65). These differences in spatiotemporal dynamics of Ca²⁺ activation and depression may reflect differences in recruitment of excitatory and inhibitory neurons (see discussion 4.1.1).

Zoom In Zoom Out Reset image size

The previous analysis revealed a tendency for weakened visually-evoked activation in the vicinity of the probe implanted in the visual cortex (e.g. figure 2(g), whitish region [near-zero dF/F₀] around center [yellow circle] at 0–5 s after stimulation onset; figure 2(i), near-zero dF/F₀ close to center at 0–5 s; figure 4(h), whitish region around probe at 0–2 s [first row]). To further investigate this reduced activation, we conducted additional analysis (figure S3). By comparing the induced activation and depression near the probe (27.5 µm < distance ⩽ 150 µm, where 27.5 µm represents half of the probe shank width) and distant locations (150 µm < distance ⩽ 500 µm), we observed clear reductions in responses near the probe (figures S3(a) and (b)). Pairwise comparisons with the pooled dataset confirmed that the responses near the probe were significantly lower than those in the distal regions during the transient offset period (10–12 s) (figures S3(c) and (d); one-sided t-test, 0–2 s, t = −1.92, p= 0.06; 2–10 s, t = −1.95, p= 0.06; 10–12 s, t = −2.66, p< 0.05; 12–20 s, t = −0.65, p = 0.28). This reduction in response may reflect the tissue reactions associated with the probe implantation injury. Further investigations, especially focused on chronic degeneration and the recovery process of the visual response, are needed to elucidate the underlying mechanism.

We next turned our attention to the relationship between Ca²⁺ activity and the hemodynamic response. Metabolically sensitive metrics such as oxyhemoglobin and deoxyhemoglobin signals play a crucial role in estimating energy supply and consumption to local neural networks [48], which are not yet fully understood in terms of the ICMS-induced responses. Similar to what we observed in neural Ca²⁺ signals (figure 4), we found that oxyhemoglobin and deoxyhemoglobin signals exhibited distinct spatiotemporal dynamics in response to different ICMS frequencies in a representative mouse (figure 8 and movies S4–S9).

Zoom In Zoom Out Reset image size

During the 10 Hz ICMS train (0–10 s), the oxyhemoglobin concentration increased over a wider cortical area surrounding the stimulation site compared to the Ca²⁺ changes (figure 8(b)). This suggests that oxygen was metabolically supplied not only at the center of activation but also in the distal regions where clear excitatory Ca²⁺ activity was not observed. Even after the stimulation train ended (10–20 s), the oxyhemoglobin concentration remained elevated compared to the pre-stimulation baseline (increase in metabolic supply). With the 100 Hz ICMS (figure 8(g)), which induced more localized neural Ca²⁺ activation compared to the 10 Hz ICMS, the increase in oxyhemoglobin was confined to a narrower distance. Simultaneously, the deoxyhemoglobin showed a similar trend, decreasing in concentration (figures 8(c) and (h); decrease in metabolic demand, relative increase in metabolic supply). It is worth noting that the changes in deoxyhemoglobin (<−10 µM in the 10 Hz ICMS and <−5 µM in the 100 Hz ICMS) were about five times smaller than those in oxyhemoglobin (>50 µM and > 25 µM), indicating that the change in total hemoglobin concentration (sum of oxyhemoglobin and deoxyhemoglobin concentration changes, Δ[HbO] + Δ[HbR]) decreased over a wide cortical area during and after the stimulation trains. The hemodynamic signals emerged slightly slower but persisted substantially longer than the excitatory Ca²⁺ signals (figure 8(d)). Additionally, the spatial profile also revealed the evident difference between the Ca²⁺ and hemodynamic signals (figure 8(e)). Notably, the hemodynamic responses exhibited substantial changes over a wider area than the Ca²⁺ changes. This difference was attenuated with higher ICMS frequency (10 Hz ICMS vs. 100 Hz ICMS). Compared to the ICMS-induced responses, visually-evoked hemodynamic responses were weaker (figures 8(k)–(o)).

We then quantified the distance, latency, and duration of the Ca²⁺ activation, oxyhemoglobin, and deoxyhemoglobin signals across all stimulation conditions (figure 9). The distance was significantly dependent on the stimulation conditions, but was independent of the signal modalities (figure 9(a); two-way ANOVA, stimulation condition, F = 3.31, p< 0.05; signal modality, F = 0.92, p= 0.40; stimulation condition × signal modality, F = 0.52, p= 0.90). Post-hoc tests revealed the significant differences between several stimulation condition pairs (50 Hz ICMS vs 500 Hz ICMS, difference = −298 µm, p< 0.05; 500 Hz ICMS vs visual stimulation, difference = 326 µm, p< 0.05). Comparing the distances of the oxyhemoglobin and deoxyhemoglobin signals to the Ca²⁺ activation distance, we found that the oxyhemoglobin signal distances were 29 µm further than the Ca²⁺ signal, while the deoxyhemoglobin signal threshold distances were 73 µm further.

Zoom In Zoom Out Reset image size

Significant differences in latencies were observed across the stimulation conditions and signal modalities (figure 9(b); two-way ANOVA, stimulation condition, F = 3.95, p< 0.05; signal modality, F = 18.97, p< 0.05; stimulation condition × signal modality, F = 1.21, p= 0.28). Interestingly, the post-hoc Tukey's HSD test revealed significant differences in latencies between certain ICMS frequencies and visual stimulation conditions (25 Hz ICMS vs visual stimulation, difference = 2 s, p< 0.05; 50 Hz ICMS vs visual stimulation, difference = 2 s, p< 0.05; 100 Hz ICMS vs visual stimulation, difference = 2 s, p< 0.05). This indicates that visual stimulation resulted in significantly slower response emergence compared to ICMS conditions, potentially due to the transmission of visual signals from the eye to the visual cortex and the unsynchronized recruitment of neurons, causing a sluggish emergence of signals. Furthermore, the latencies in the hemodynamic signals were a few seconds slower than the Ca²⁺ signals (oxyhemoglobin-Ca²⁺ = 1 s, p< 0.05; deoxyhemoglobin-Ca²⁺ = 2 s, p< 0.05). Additionally, the deoxyhemoglobin signal showed a significantly slower latency compared to the oxyhemoglobin signal (deoxyhemoglobin-oxyhemoglobin = 1 s, p< 0.05), which may reflect a cascade of blood influx and deoxygenation.

Having revealed distinct activation distance and latency differences in Ca²⁺ and hemodynamic activity dependent on stimulation parameters, we investigated whether there were differences in the duration of these signals during and following ICMS. Similar to the latencies mentioned earlier, the durations were significantly influenced by both the stimulation conditions and signal modalities (figure 9(c); two-way ANOVA, stimulation condition, F = 2.51, p< 0.05; signal modality, F = 30.92, p< 0.05; stimulation condition × signal modality, F = 0.46, p= 0.94). The duration, particularly in oxyhemoglobin signal, tended to be longer around the 25 Hz ICMS condition (no significant differences in the post-hoc Tukey's HSD test among the stimulation conditions). The gross durations across all stimulation conditions in the oxyhemoglobin signal were significantly longer compared to the Ca²⁺ and deoxyhemoglobin signals (post-hoc Tukey's HSD test, oxyhemoglobin-Ca²⁺ = 24 s, p< 0.05; oxyhemoglobin-deoxyhemoglobin = 19 s, p< 0.05; deoxyhemoglobin-Ca²⁺ = 5 s, p= 0.23). This prolonged oxyhemoglobin signal was statistically significant (figure 9(d), differences in latencies and durations between Ca²⁺ and hemodynamic signals; Ca²⁺ vs. oxyhemoglobin, F_stim = 1.15, p_stim = 0.34, F_time = 91.65, p_time < 0.05, F_{stim × time} = 1.46, p_{stim × time} = 0.20, post-hoc Tukey's HSD test, latency vs duration, difference = 23 s, p< 0.05; Ca²⁺ vs. deoxyhemoglobin, F_stim = 0.40, p_stim = 0.88, F_time = 1.28, p_time = 0.26, F_{stim × time} = 0.54, p_{stim × time} = 0.77).

Therefore, although the degree of neural Ca²⁺ activity and the metabolic supply and demand in response to various ICMS frequencies and visual stimulation are highly dependent on the stimulation conditions, the prolonged increase in oxyhemoglobin concentration is consistently observed in most stimulation conditions. These results suggest that hemodynamic activity is involved in forming the post-activation Ca²⁺ response in excitatory neurons, in which the increased oxygenation may be associated with increased activity in inhibitory neurons and/or govern the depression phenomenon (see discussion 4.2).

Alternatively, these prolonged hemodynamic signals could potentially reflect a metabolic supply-demand imbalance, leading to metabolic shortage and a reduction in neural activity ('metabolic silencing'), and/or indirectly depressing excitatory neurons through inhibitory neurons (see discussion 4.1.2). To quantify the degree of metabolic demand during and after the stimulation within the range showing clear hemodynamic responses (<500 µm, figures 9(e) and (f)), we calculated the OEF (equation (5)) with several baseline concentration assumptions (see methods 2.4.4). OEF is defined as the fraction of deoxyhemoglobin concentration over the total hemoglobin concentration. We observed a significant reduction in OEF, particularly with the 10, 25, and 100 Hz ICMS during the stimulation and initial 20 s post-stimulation periods (two-way ANOVA, stimulation condition, F = 8.27, p< 0.05, timing, F = 36.45, p< 0.05, stimulation condition × timing, F = 1.69, p< 0.05; post-hoc Tukey's HSD test, 10 Hz ICMS vs 250 Hz ICMS, difference = 0.021, p< 0.05, 10 Hz ICMS vs 500 Hz ICMS, difference = 0.025, p< 0.05, 10 Hz ICMS vs LED, difference = 0.030, p< 0.05, 25 Hz ICMS vs LED, difference = 0.024, p< 0.05, 100 Hz ICMS vs LED, difference = 0.022, p< 0.05; see table S2 for results of post-hoc Tukey's HSD test across the timing). These findings suggest that the 25–50 Hz ICMS trigger hyperoxygenation in the local tissues.

Additionally, we also asked how the induced hyperoxygenation is localized in the visual cortex (figure 9(g)). The OEF distance profiles during the 0–10 s stimulation period showed extensive spreads, nearly encompassing the entire imaging field upon ICMS application. Additionally, higher-frequency ICMS resulted in a diminished reduction of OEF. Regarding visual stimulation, the altered OEF is not discernible at any distance. Statistical quantification (figure 9(h)) revealed significant OEF reductions with 10–250 Hz ICMS (two-way ANOVA, stimulation condition, F = 24.12, p < 0.05; see table S3 for results of post-hoc Tukey's HSD test across stimulation conditions). These large reductions of OEF were restricted within 1000 µm from the stimulation sites (two-way ANOVA, distance, F = 30.16, p < 0.05; see table S4 for results of post-hoc Tukey's HSD test across distances). No significant interaction between the stimulation conditions and distances was observed (F = 0.89, p = 0.65). Taken together, these spatiotemporal profiles suggest that the ICMS-induced change in OEF is sufficiently greater than the metabolic demand that is caused by the natural sensory stimulus (visual stimulation), which is enhanced especially by low or middle frequencies.

Further analyses revealed that the correlation between spatiotemporal dynamics of excitatory neural Ca²⁺ responses and oxyhemoglobin responses during the stimulation period (0–10 s) significantly depended on the stimulation conditions (figure S4(a), one-way ANOVA, F = 3.02, p< 0.05). Specifically, the Ca²⁺ responses in the lower ICMS frequencies, particularly at 10 Hz, tended to be positively correlated with oxyhemoglobin responses during the stimulation train period (post-hoc Tukey's HSD test, 10 Hz ICMS vs 500 Hz ICMS, difference = −0.31, p< 0.05). Furthermore, strong correlations were observed in the magnitudes extracted from the Ca²⁺ activation/depression and oxy/deoxyhemoglobin signals (figures S4(e)–(h), Spearman's rank correlation coefficient, p < 0.05). This suggests that the 10 and 100 Hz ICMS induced a positive coupling between neural activity and oxygen supply/demand, whereas visual stimulation resulted in a negative coupling between neural depression and oxygen demand.

We investigated the relationship between stimulation parameters and the activation and depression of excitatory neurons, as well as oxyhemoglobin and deoxyhemoglobin signals, using mesoscopic-scale widefield imaging. Although this imaging technique allowed us to capture spatial information across a large area of the visual cortex, it cannot differentiate signals in finer neural structures like soma and neuropil. The mechanisms of depression in the soma and neuropil are not well understood. To address this, we conducted two-photon imaging (figure 10 and movies S10–S12).

Zoom In Zoom Out Reset image size

Under two-photon microscopy, we observed strong scattered spots of somatic autofluorescence around the inserted probe and neurite blebs (figure 10(a-ii)). Despite these autofluorescent spots, we were able to observe both somatic and neuropil activities during and after the 10 Hz ICMS train (figure 10(b)). The somas exhibited relatively stronger activation compared to the neuropil. In the 100 Hz ICMS condition, the somas near the stimulation site showed sparse activation, and the neuropil activation was localized near the stimulation site (figure 10(c)). In both cases, we observed distinct depression in both the somas and neuropil, where the Ca²⁺ activity dipped below the baseline resting state before the stimulus. We manually defined the ROI for each neuron (see methods 2.4.2) and obtained somatic Ca²⁺ time courses by averaging dF/F₀ values inside single ROIs. We also obtained neuropil Ca²⁺ time courses by averaging dF/F₀ values outside the ROIs.

To examine the dependency of excitatory neural Ca²⁺ response on the stimulation condition, we selected one ROI as an example (figures 10(e) and (f)). These somatic and surrounding neuropil responses were dependent on the stimulation conditions (figure 10(e)).

For instance, during the 10 s stimulation trains, the ICMS-induced Ca²⁺ activity was highest at 10 Hz both in the soma and neuropil, and decreased as the ICMS frequency increased (figure 10(e), top, 0–10 s). This observation aligns with the results we obtained using mesoscopic-scale microscopy (figures 2 and 4). A few seconds after the 10 s stimulation train ended, dominant depression was observed in both the soma and neuropil (figure 10(e), bottom, 15–30 s). With cellular scale imaging, we were able to discern the visual stimulation preference of each neuron for the stimulation onset, offset, and onset/offset (figure 10(d), movie S12). Although the visual responses were remarkably weaker than the ICMS responses, similar to the mesoscopic-scale imaging findings, several neurons exhibited Ca²⁺ activation at different time points during the stimulation train.

The somatic Ca²⁺ signals (figure 10(f-i)) exhibited a partially similar time course compared to the results obtained from mesoscopic-scale imaging (figure 4(i)). In mesoscopic-scale imaging, lower-frequency ICMS resulted in 'developing activation' (activation that slowly increases with stimulus duration), while higher-frequency ICMS caused a transient one. In this example soma (figure 10), the higher-frequency ICMS initially induced a transient activation, followed by developing activation. With the exception for the visual stimulation condition, we consistently observed distinct somatic depression following activation (figure 10(f-ii)). We analyzed the neural Ca²⁺ time courses of 425 somatic ROIs and of neuropil in 9 mice (figures 10(g) and (h)). Overall, the pooled dataset revealed developing activation with lower-frequency ICMS, whereas higher-frequency ICMS triggered an initial transient increase in Ca²⁺ fluorescent intensity, followed by rapid reduction. However, all ICMS conditions maintained activation throughout the 10 s stimulation train. The depression phenomena were also observable in all ICMS conditions, regardless of the different magnitudes. Additionally, we observed weak transient on/off activation with a visual stimulation train. These trends were consistent in neuropils, although the activation magnitudes were noticeably smaller compared to somas. We compared the somatic activation magnitudes between the transient (0–2 s) and sustained (2–10 s) phases of the stimulation trains (figures S5(a) and (b)), and found that the transient-dominant responses were evident in the 50–500 Hz ICMS and visual stimulation conditions (50 Hz ICMS, paired t-test, t = 12.28, p< 0.05; 100 Hz, t = 13.62, p< 0.05; 250 Hz, t = 6.95, p< 0.05; 500 Hz, t = 10.88, p< 0.05; visual stimulation, t = 7.98, p< 0.05). On the other hand, the sustained-dominant responses were observed in the 10 and 25 Hz ICMS conditions (10 Hz, t = −28.92, p< 0.05; 25 Hz, t = −21.98, p< 0.05). Moreover, the change rates of somatic Ca²⁺ response magnitude during the stimulation train significantly varied among the stimulation conditions (figures S5(c) and (d), one-way ANOVA, F = 395.77, p< 0.05). Lower ICMS frequencies tended to induce an increasing activation, while higher-frequency ICMS and visual stimulation tended to induce relatively constant activation (table S1). Overall, these findings demonstrate that the somas and neuropils identified through two-photon microscopy exhibited response properties similar to those observed under mesoscopic-scale microscopy.

Given that the spatial extent of the mesoscopic-scale Ca²⁺ responses differed across ICMS frequencies (figure 4), we investigated how the response magnitude of somas at different distances from the stimulation site was influenced by ICMS frequencies (figure S6). Firstly, we quantified the distribution of distances for the activated somatic ROIs (figure S6(a)). The identified somas were mostly located around 50–100 µm from the center of the stimulation channel. Since the probe had a width of 55 µm, tissues within this range could be deformed, and might have displaced neurons. Consistent with the mesoscopic-scale results, the magnitudes of somatic activation and depression were dependent on distance and stimulation conditions (figure S6(b), activation; two-way ANOVA, distance, F = 52.29, p< 0.05; stimulation condition, F = 334.34, p< 0.05; figure S6(c), depression; distance, F = 5.14, p< 0.05; stimulation condition, F = 17.62, p< 0.05). Further analyses revealed significant differences in magnitudes between several distances for activation (post-hoc Tukey's HSD test, 0–50 µm vs 50–100 µm, difference = 52.24%, p< 0.05; 0–50 µm vs 150–200 µm, difference = −27.51%, p< 0.05; 0–50 µm vs 200–250 µm, difference = −41.91 µm, p< 0.05; 50–100 µm vs 100–150 µm, difference = −49.04%, p< 0.05; 50–100 µm vs 150–200 µm, difference = −79.75%, p< 0.05; 50–100 µm vs 200–250 µm, difference = −94.15%, p< 0.05; 50–100 µm vs 250–300 µm, difference = −112.77%, p< 0.05; 100–150 µm vs 150–200 µm, difference = −30.72%, p< 0.05; 100–150 µm vs 200–250 µm, difference = −45.12%, p< 0.05) and for depression (100–150 µm vs 150–200 µm, difference = −0.57%, p< 0.05; 100–150 µm vs 200–250 µm, difference = −0.63%, p< 0.05). Moreover, the interaction between stimulation condition and distance significantly affected the magnitudes of somatic activation and depression (two-way ANOVA, activation, F = 6.01, p< 0.05; depression, F = 2.24, p< 0.05), suggesting that ICMSs at different frequencies could activate/depress somas at different distances. These results demonstrate that somatic Ca²⁺ responses, akin to mesoscopic-scale responses (figure S7), preferred specific ICMS frequencies corresponding to the distance from the stimulation site.

As demonstrated, Ca²⁺ responses in soma and neuropil of excitatory neurons were influenced by stimulation conditions. Thus, we proceeded to statistically compare the magnitude and duration of activation and depression in both somas and neuropils, as well as their balance (ratios of magnitudes between soma and neuropil, and between activation and depression) across different stimulation conditions (figure 11(a)).

Zoom In Zoom Out Reset image size

The magnitudes of somatic activation were significantly affected by the stimulation conditions (figure 11(b), one-way ANOVA, F = 280.43, p< 0.05), peaking at the 25 Hz ICMS (see table S5 for pairwise comparisons with post-hoc Tukey's HSD test). In contrast, the duration of activation exhibited a relatively uniform distribution, except under the visual stimulation condition (figure 11(d), F = 697.07, p< 0.05; table S6). Similarly, the magnitudes of the somatic depression were significantly influenced by the stimulation conditions, displaying an unimodal distribution biased towards 25 Hz (figure 11(f), F = 35.17, p< 0.05; table S7). However, the distribution of somatic depression magnitudes was broader than that of the activation.

Consequently, the balance between somatic activation and depression (activation magnitude/depression magnitude) exhibited an unimodal preference, with a peak at the 25 Hz ICMS condition (figure 11(l), F = 147.35, p< 0.05; table S10). This indicates that lower-frequency ICMS, particularly at 25 Hz, elicited activation-dominant somatic Ca²⁺ responses. Interestingly, the activation/depression balance in the visual stimulation condition was nearly one, which suggests that visual stimulation induced a balanced activation of excitatory and inhibitory/suppressive responses. To replicate neural responses induced by visual stimulation, higher-frequency ICMS could be utilized in the field of visual cortical prostheses (see discussion 4.3). Moreover, the duration of somatic depression also exhibited a clear bias toward the 25 Hz ICMS condition (figure 11(h), F = 47.01, p< 0.05; table S8), and tended to be shorter than the duration of activation.

Given the clear dependency of somatic activation and depression on ICMS frequencies and visual stimulation, we proceeded to quantify the responsive properties of neuropil across various stimulation conditions and compared them with somatic properties. The magnitudes of activation in the neuropil displayed a preference similar to that observed in somas (figure 11(c)), was significantly influenced by the stimulation conditions (one-way ANOVA, F = 4.42, p< 0.05), and exhibited a unimodal distribution peaking at the 25 Hz ICMS (post-hoc Tukey's HSD test, 10 Hz ICMS vs visual stimulation, difference = −79.78%, p< 0.05; 25 Hz ICMS vs 250 Hz ICMS, difference = −64.53%, p< 0.05; 25 Hz ICMS vs 500 Hz ICMS, difference = −72.61%, p< 0.05; 25 Hz ICMS vs visual stimulation, difference = −92.38%, p < 0.05).

When comparing the magnitudes of activation between soma and neuropil, we observed a significant difference in the soma/neuropil balance across stimulation conditions (figure 11(j), F = 28.99, p< 0.05; table S9), where higher-frequency ICMS elicited more soma-dominant activations. The duration of activation in the neuropil exhibited a uniform distribution except for the visual stimulation, similar to somas (figure 11(e), F = 11.88, p< 0.05), in which visual stimulation induced significantly shorter activation compared to all ICMS frequencies (10 Hz ICMS vs visual stimulation, difference = −6.93 s, p< 0.05; 25 Hz ICMS vs visual stimulation, difference = −7.21 s, p< 0.05; 50 Hz ICMS vs visual stimulation, difference = −6.77 s, p< 0.05; 100 Hz ICMS vs visual stimulation, difference = −6.60 s, p< 0.05; 250 Hz ICMS vs visual stimulation, difference = −6.71 s, p< 0.05; 500 Hz ICMS vs visual stimulation, difference = −6.38 s, p< 0.05). These results indicate that both somas and neuropils are capable of sustained activation throughout the ICMS train, at least for a duration of 10 s, within 10–500 Hz, although the magnitudes may vary depending on the frequency.

The depression magnitude in the neuropil also exhibited a slight bias, peaking at 25 Hz (figure 11(g), one-way ANOVA, F = 2.72, p< 0.05; post-hoc Tukey's HSD test, 25 Hz ICMS vs visual stimulation, difference = −4.58%, p< 0.05). However, the visual stimulation did not induce significant depression in the neuropil. When comparing the depression magnitudes between somas and neuropils, we observed that the soma/neuropil balance of depression magnitudes were unaffected by the ICMS frequencies (figure 11(k), F = 1.18, p= 0.32). This suggests that the mechanism underlying the depression phenomenon selectively affects the soma and its selectivity is not modified by the frequent repetition of the ICMS pulses. The activation/depression balance in the neuropil also peaked at the 25 Hz ICMS condition (figure 11(m), F = 6.71, p< 0.05; 10 Hz ICMS vs 100 Hz ICMS, difference = −8.10, p< 0.05; 10 Hz ICMS vs 250 Hz ICMS, difference = −8.90, p< 0.05; 10 Hz ICMS vs 500 Hz ICMS, difference = −10.40, p< 0.05; 25 Hz ICMS vs 100 Hz ICMS, difference = −9.27, p< 0.05; 25 Hz ICMS vs 250 Hz ICMS, difference = −10.07, p< 0.05; 25 Hz ICMS vs 500 Hz ICMS, difference = −11.57, p< 0.05). This finding is similar to somatic activation, suggesting that the 25 Hz ICMS could be the preferred choice for inducing an activation-dominant neural response in the field of neuroprosthetics. The depression duration in the neuropil exhibited a peak at around 100 Hz (figure 11(i), F = 2.98, p < 0.05; post-hoc Tukey's HSD test, 100 Hz ICMS vs visual stimulation, difference = −19.94 s, p < 0.05), which tended to be longer than in the somas. These results demonstrate that ICMS alone has the capacity to elicit depression, even in the neuropil, implying that ICMS and visual stimulation do not share the same mechanism for inducing the depression phenomenon.

Overall, the activation magnitudes exhibited a bias towards 25 Hz. The depression magnitudes also peaked at 25 Hz, but were less sensitive to differences in ICMS frequencies compared to the activation magnitudes. This discrepancy indicates that the mechanism underlying depression differs from that of activation, which shows a preference for higher-frequency ICMS.

We observed that the depression of excitatory Ca²⁺ signals occurred following the activation (figure 2), and its characteristics varied depending on the stimulation parameters (figures 4 and 5). Similar depression phenomena, where neural activity is reduced compared to the pre-stimulation baseline, have been described in previous studies [6, 31, 33, 34], where the specific methods used for induction and quantification varied each other but the underlying mechanisms might be shared. For instance, McCreery and colleagues investigated a phenomenon called SIDNE [34, 49, 50]. In their experiments, repetitive electrical pulses (100–500 Hz, < 3 nC/phase) were applied for 7 h. They observed an increase in the threshold currents required to evoke compound action potentials after the pulse train. This SIDNE phenomenon could be reversed within several days, and no histological tissue injury attributable to the stimulation was observed. The mechanism underlying SIDNE is thought to be related to the influx of Ca²⁺, triggering intracellular mechanics affecting membrane ion channels [34], and/or transient changes of extracellular Ca²⁺ and potassium ion concentrations induced by prolonged stimulation [51].

Cortical spreading depression (CSD) is another example of an induced neural depression phenomenon, which can be induced by various forms of stimulation, including electrical pulses [52, 53]. CSD has often been investigated using hemodynamic signal imaging [54], where CSD is accompanied by a reduction in reflectance (indicating increased blood flow). Unlike SIDNE, CSD occurs on a shorter timescale, emerging a few seconds after the stimulation and lasting from several minutes [55–57] to an hour [58]. The formation of CSD involves both presynaptic and postsynaptic mechanisms. Presynaptically, there is a reduction in the probability of transmitter release, while postsynaptically there is a relative increase in inhibitory current, contributing to the generation of CSD [58].

In other scenarios, there have been reports of neural activity reduction occurring on an even shorter timescale, lasting in the range of hundreds of milliseconds. In these cases, the gamma-aminobutyric acid B (GABAb) receptor has been found to play a role in mediating the reduction of action potentials [33] and population-level membrane potential [6] following transient activation induced by ICMS. Furthermore, numerous other potential mechanisms can contribute to decreased neural excitability and related phenomena. These include energy supply shortage due to repetitive neural activation (metabolic limitation), neurotransmitter vesicle depletion, network inhibition involving vascular-astrocyte-neural coupling, network inhibition through recurrent inhibition (excitatory neuron—inhibitory neuron—excitatory neuron feedback loop), long-term depression, contrast enhancement through lateral inhibition, and sensory adaptation. When summarizing these possible mechanisms underlying the Ca²⁺ depression phenomenon, they can be classified into several groups based on different aspects such as spatial and temporal considerations.

4.1.1. Inhibitory circuits on spatiotemporal inhibition and depression

4.1.2. Metabolic stress and fatigue on neuronal depression and decrease in GCaMP signal

4.1.3. Inhibition in visual processing

In relation to the preceding discussion, quantifying hemodynamic changes in conjunction with induced neural activity provides a better understanding of neurovascular coupling [86] and the contribution of oxygen supply/shortage in neural activity depression. Previous studies have examined hemodynamic changes in the cerebral cortex induced by ICMS [38] and superficial electrical stimulation [87], demonstrating that stronger and longer stimulation leads to greater blood flow changes. Particularly, larger blood flow responses have been reported with optogenetic stimulation of inhibitory neurons [88]. These results are consistent with preferential activation of inhibitory neurons by engaging relatively large blood flow responses that drive large increases in oxyhemoglobin.

Regarding the temporal aspect of the hemodynamic signals, they exhibited notably longer time courses compared to local neural dynamics. During and after the ICMS train, the oxyhemoglobin (reflecting metabolic supply) increased, while the deoxyhemoglobin (reflecting metabolic demand) decreased within the same period. The magnitude of the oxyhemoglobin increase was approximately five times larger than the deoxyhemoglobin decrease. Estimating the relative metabolic demand (OEF) revealed a reduction during ICMS with lower frequencies, particularly at 25 and 50 Hz (figures 9(e)–(h)). This reduction persisted even after the ICMS train, despite the excitatory neurons exhibiting fewer Ca²⁺ signals and requiring less oxygen compared to the stimulation period. These findings suggest the presence of a metabolic over-supply (hyperoxygenation) environment around the stimulation site that lasts for tens of seconds. The existence of this prolonged hyperoxygenation implies that inhibitory neurons work dominantly during the post-stimulation period. In contrast, such a significant hyperoxygenation was not observed in the visual stimulation condition. As we mentioned above, visual stimulation recruits many sub-cortical circuits, and the full-field LED blinking induced balanced activation and depression. Given that, depression in the full-field visual stimulation condition may reflect the reduction of input from the sub-cortical pathway, requiring relatively less inhibitory neural contribution in the local cortical circuit and relatively less post-stimulation hyperoxygenation in the visual cortex. To achieve a biomimetic response in the field of cortical visual prostheses, it may be necessary to design improved ICMS patterns that effectively balance metabolic supply and demand, at least when representing an environmental illumination.

We discovered that the ICMS trains elicited sustained activation, whereas the visual stimulation resulted in only transient responses (e.g. figure 2). These findings raise questions about the underlying reasons for such stark differences in the induced responses. Moreover, the observed discrepancy in response patterns between ICMS and visual stimulation poses potential challenges in the practical implementation of visual prostheses.

The neurons in the mouse visual cortex exhibited a preference for frequencies around 3 Hz or lower [89, 90]. Thus, the 10 Hz visual stimulation utilized in our study exceeded their neural preferences and could be considered as constant-on visual stimulation for such neurons. Considering the temporal receptive field structures that demonstrate transient on, off, and on/off properties, it becomes evident that the 10 Hz visual stimulation train cannot elicit continuous visual responses following each light flickering. Instead, it generates on and off responses at the onset and offset of the stimulation train. In contrast, ICMS can override these visual preferences in individual neurons. If the aim is to mimic visual responses in visual prostheses, a better approach would involve generating transient responses by adjusting the parameters of ICMS. The simplest method would be to modify the duration of the stimulation train, limiting it to a short period that ensures the necessary activation of neurons [91]. Another option involves exploring the potentially intertwined nature of frequency dependency and the depression effect. Our findings indicate that higher-frequency ICMS induced transient activations that dropped to the weak sustained level within a few seconds after the onset of stimulation (figures 6 and S5). Also, in terms of induced response magnitudes, such higher-frequency ICMS led to a smaller activation/depression balance, which was relatively close to the value for visual stimulation (balance ≈ 1; figures 11(l) and (m)). Further increasing the frequency might enhance this trend. However, the perception associated with weak sustained neural responses remains poorly understood. Future studies should address these aspects by employing animal behavior paradigms or by assessing induced perceptions in human subjects.

Paradoxically, natural vision is capable of perceiving constant light. Hence, it is crucial for the visual prosthesis system to generate a constant neural response. Our findings indicate that most of the ICMS frequencies we tested were able to induce significant neural activation throughout the 10 s stimulation train, albeit with varying magnitudes and sizes depending on the frequency employed (figures 6 and S5). Some middle frequencies (100 and 250 Hz) resulted in relatively longer activation compared to the other ICMS frequencies. Regarding the magnitude, it remained relatively constant during the sustained period (2–10 s) for middle to high frequencies (e.g. 500 Hz), despite a decreasing trend in the size of the cortical response. Conversely, in terms of the distance, low-frequency ICMS (e.g. 10 Hz) elicited a response of constant size, although the magnitude exhibited an increasing trend. Consequently, it is possible that higher frequencies with increasing pulse intensity or lower frequencies with decreasing pulse intensity could be effective in the ICMS paradigm to generate the constant neural response.

Furthermore, the utilization of depression in the field of visual prostheses can be considered. While depression may prevent continuous light perception or induce perceptual adaptation through the introduction of a refractory period, it could also aid in generating temporally accurate perceptions by rapidly halting unexpectedly prolonged activation. This, in turn, has the potential to enhance the temporal resolution of prosthetic vision, which is currently relatively slower in contemporary visual prostheses or exhibits a lower critical fusion frequency [92]. By effectively inducing depression through the implementation of a higher-frequency ICMS paradigm (figures 11(l) and (m)) in prosthetic visual systems (e.g. high-frequency burst ICMS evoking temporally localized activation at each burst phase, like blinking-light perception, leading to high temporal resolution and high critical fusion frequency), we could potentially overcome this temporal limitation.

In the current study, we constantly used the same stimulation intensity at 20 µA/phase. However, neural activity is also expected to be modulated by the change in stimulation intensity. For example, using two-photon microscopy in mouse somatosensory cortex, Ma et al evaluated neural responses to various stimulation intensities from 25 to 225 µA and found that neural activity amplitude exhibited a sigmoidal profile to the intensity change, in which around 80 µA/phase stimulation pulse train elicited half-maximum response [28]. Similar intensity-response trends, which exhibit a sigmoidal curve or at least a monotonic increase in neural response along with stimulation intensity increase, were reported in multiple other studies [6, 35, 59, 93, 94]. Sombeck et al explored the decrease in neural activity by changing stimulation intensity from 5 to 100 µA and found that such depression duration and fraction of depressed neurons increased along with stimulation intensity increase [94]. These previous results suggest that, at least within the range used in prosthetic systems [95], greater stimulation pulse intensity drives both excitatory and inhibitory neurons more in terms of response magnitude and number of cells recruited [93]. This is different from what we observed in the current study by increasing stimulation frequency (figures 5 and 11) even though both greater stimulation pulse intensity and more frequent stimulation inject greater total charge into the cortex. Future studies should consider the spatial current spread inducing direct activation [8] and the impedance spectrum of cortical tissue [96, 97] together to draw a detailed picture of the mechanism underlying the difference in induced responses to varying stimulation intensity and frequency.

Although this study did not specifically investigate the chronic response properties induced by ICMS, certain results suggest that there were effects of tissue damage caused by probe insertion even during the acute experimental phase (see figure 10(a-ii)), which could impair the normal visual response (see figure S3). Our recordings were conducted approximately 1–2.5 h after the probe insertion, a period during which processes such as protein binding, neutrophil recruitment, and monocyte recruitment and differentiation into macrophages occur around the surface of the inserted probe [98]. These cellular events can result in the production of reactive oxygen and nitrogen species, which have the potential to harm neurons, as well as cytokines that promote immune responses [99]. Additionally, the traumatic brain injury resulting from probe insertion can lead to acute local ischemia [100]. While these various factors may attenuate normal sensory responses in the sensory cortices, our study demonstrated that ICMS still processes significant capability to induce neural responses, at least during the acute phase. However, during the chronic phase, which includes days following probe insertion, other processes such as foreign body giant cell formation, fibrotic encapsulation around the probe surface [98], and vasculature remodeling [101] occur. These events can have additional effects on neurons and the probes themselves [24]. Further investigation is necessary to examine the potential chronic alterations in neural Ca²⁺ and hemodynamic signals in response to ICMS and sensory stimulation.