Cortical layering disrupts multi-electrode current steering

Objective. Blindness affects approximately 40 million people worldwide and has inspired the development of cortical visual prostheses for restoring sight. Cortical visual prostheses electrically stimulate neurons of the visual cortex to artificially evoke visual percepts. Of the 6 layers of the visual cortex, layer 4 contains neurons that are likely to evoke a visual percept. Intracortical prostheses therefore aim to target layer 4; however, this can be difficult due to cortical curvature, inter-subject cortical variability, blindness-induced anatomical changes in cortex, and electrode placement variations. We investigated the feasibility of using current steering to stimulate specific cortical layers between electrodes in the laminar column. Approach. We explored whether the multiunit neural activity peak can be manipulated between two simultaneously stimulating electrodes in different layers of the cortical column. A 64-channel, 4-shank electrode array was implanted into the visual cortex of Sprague–Dawley rats (n = 7) orthogonal to the cortical surface. A remote return electrode was positioned over the frontal cortex in the same hemisphere. Charge was supplied to two stimulating electrodes along a single shank. Differing ratios of charge (100:0, 75:25, 50:50) and separation distances (300–500 μm) were tested. Results. Current steering across the cortical layers did not result in a consistent shift of the neural activity peak. Both single-electrode and dual-electrode stimulation induced activity throughout the cortical column. This contrasts observations that current steering evoked a controllable peak of neural activity between electrodes implanted at similar cortical depths. However, dual-electrode stimulation across the layers did reduce the stimulation threshold at each site compared to single-electrode stimulation. Significance. Multi-electrode stimulation is not suitable for targeted activation of layers using current steering. However, it can be used to reduce activation thresholds at adjacent electrodes within a given cortical layer. This may be applied to reduce the stimulation side effects of neural prostheses, such as seizures.

The position of the electrodes within the cortex may determine the efficacy and quality of restored vision. V1 is partitioned into six layers that can broadly be divided into the supragranular (layers 2 and 3), granular/input (layer 4), and infragranular (layers 5 and 6) layers [19]. Targeting the granular layer with electrical stimulation may be more likely to generate phosphenes since the artificially induced activity will propagate through the visual system to higher visual areas [20,21].
The neuronal composition of the layers also affects the stimulation threshold [22,23]. A low stimulation threshold is desirable since it allows for spatially-precise stimulation, reduces the risk of seizures, and reduces power consumption [2,4,24]. Each layer has differing densities of neuronal axons, soma, and dendrites [19,25]. Stimulation preferentially activates axons and so those layers with higher densities of axons will likely have lower thresholds [23,26]. Detection thresholds were lowest in the supragranular and granular layers in awake macaques [23] indicating we likely want to target these layers. Additionally, layers 4 and 5 have the highest neuronal stability following electrode implantation [27]. This means neurons in layers 4 and 5 are most likely to consistently respond over the lifetime of the implant.
With all of these considerations, it becomes clear that electrical stimulation applied to the granular layer may improve safety and efficacy. However, inherent inter-subject variability in the thickness of the human visual cortex [28], cortical thickness variability induced by differing age of onset blindness [29][30][31][32], and variations in implantation depths can make targeted stimulation difficult. This means a post-hoc method of steering the activation towards layer 4 is needed.
One solution is to use an array of electrodes along a single shank to increase the likelihood of reaching the target neurons. However, the manufacturing cost for such an array will be substantially more than a single electrode. We propose a different, more costeffective solution known as current steering. Current steering involves supplying low-level, subthreshold stimulation at two electrodes simultaneously. The evoked electric fields overlap to create a suprathreshold response between the electrodes. The suprathreshold response location can be theoretically controlled by altering the distribution of charge between the physical electrodes. Hence, by having two electrodes per shank, we can potentially ensure that the granular layer is targeted despite inaccurate electrode placement for a minimal manufacturing cost.
Current steering has been shown to shift the phosphene location within the visual field [33,34] and we have previously shown that current steering can shift the peak of activity when all electrodes were at similar cortical depths (single-layer current steering) [35]. However, the cortical layers have differences in cell densities and types meaning that current steering across the cortical layers may not be as straightforward. In this paper, we investigate whether current steering can be applied across cortical layers (multilayer current steering) to mitigate errors in electrode placement. We tested current steering with three different separation distances between pairs of stimulating electrodes and eight total current levels. We could not control the peak location of neural activity with multilayer current steering. We believe the neural connections likely influence and defocus the steered response regardless of the location of the stimulating electrodes in the cortical column.

Animal preparation
Sprague-Dawley rats (male, 12 weeks old, n = 7) were anesthetized with Halothane (5% induction, 1%-2% maintenance, 80 breaths per minute, 4.25 ml stroke volume). Heart rate, temperature, and reflexes were monitored constantly. The animals' heads were positioned in a stereotaxic frame and a burr hole was drilled 2 mm posterior to sutura coronalis and 5 mm lateral from the midline. A silver wire was implanted in the burr hole to serve as the remote return electrode. A craniotomy was made in the same hemisphere ∼1 mm anterior to lambda and 2 mm lateral from the midline. The dura mater was resected and a four-shank 64-channel Neuronexus probe (177 µm 2 electrode area, 50 µm electrode spacing, 5 mm shank, 16 sites per shank, 200 µm shank spacing) was implanted perpendicularly to the cortical surface ( figure 1(A)). Animals were euthanized at the conclusion of the experiment with an intracardiac or intraperitoneal injection of pentobarbitone. These experiments were approved by the Animal Ethics Committee at Monash University and conducted in accordance with the National Health and Medical Research Council of Australia guidelines for animal experimentation.

Electrical and visual stimulation
Visual stimulation was used to approximate the electrode array location within the V1 cortical layers. Flash stimulation from a stroboscope was directed into the eye contralateral to the hemisphere where the electrode array was implanted (900 trials, 0.9 Hz, Lutron digital-stroboscope-ic-dt-2289).
Multiunit neural activity was recorded using an Intan RHS stim/recording controller (M4200) with Intan headstages (gain 192, sample rate 30 kHz). The Intan system is capable of simultaneous recording and electrical stimulation. Multiunit activity meant multiple neurons could be recorded at a single electrode. To test the current steering theory, electrical stimulation was applied to 1-2 electrodes per trial with a remote return electrode (300, 400, or 500 µm separation between stimulating electrodes, biphasic pulse, 100 µs interphase gap, 200 µs pulse per phase, 0, 1, 2, 3, 4, 6, 8, 10 µA total, 0:100, 25:75, 50:50, 75:25, 100:0 distribution of current). As an example, if 10 µA total was applied to the stimulating electrodes, the current distributions were 0:10 µA, 2.5:7.5 µA, 5:5 µA, Experimental setup to test current steering in the visual cortex of rats. A Neuronexus microelectrode array (four shanks 200 µm apart at the tips, 16 electrodes per shank 50 µm apart) was implanted orthogonal to the cortical surface. Two types of stimulation were delivered; 1. A full field visual flash from a stroboscope which was used to confirm the recorded neurons were visually responsive, and 2. Electrical stimulation at 1-2 electrodes which was used to test the current steering paradigm. We hypothesize stimulation delivered at two electrodes will result in overlapping electric fields that will summate between the two stimulating electrodes. For example, stimulating electrodes in the supragranular and infragranular layers may be able to target the granular layer. (B) Raster-plot of action potentials evoked in response to a flash stimulus across 200 trials recorded at a single electrode. An increase in neuronal firing rate can be observed around 75-100 ms. (C) Raster-plot of action potentials recorded at the same electrode as shown in (B) in response to electrical stimulation at 6 µA. Immediate increase in firing rate due to direct activation of neurons. (D) Local field potential (LFP) at all electrodes from a single penetration. Change in amplitude and polarity of the LFP waveform indicates a change in layer which was confirmed through both a current source density analysis (CSD) and histology shown in (E). Electrodes were classed as being positioned in the supragranular, granular or infragranular layers. CSD was constructed from the LFP using the CSDplotter tool. (E) The array was coated in DiI prior to implantation allowing the tracks of all four shanks to be clearly identified. This then enabled layer classification of the electrodes in the cytochrome oxidase and/or Nissl sections. 7.5:2.5 µA, 10:0 µA. The 100:0 ratios correspond with single-electrode stimulation. A maximum of 10 µA was implemented to keep the electrodes within their safe charge limits (safety limit of 1200 µC cm −2 for iridium oxide electrodes [36] with 200 µs pulse per phase and electrode surface area of 177 µm 2 means a maximum stimulation current of 10.62 µA). A hardware highpass filter was set to 300 Hz for 100 ms following electrical stimulation to reduce the stimulation artifact. Each trial timing was randomly jittered to prevent time-locking to noise frequencies and there were 40 trials per condition.

Data filtering and spike detection
The artifacts created by electrical stimulation were removed from the data by interpolating from −1 ms to 2 ms relative to the stimulation pulse [37]. A multitaper bandpass filter was then used to remove unwanted frequency components (0.3-7.7 kHz passband, 4 kHz center frequency, 2 ms length) [38]. Neuronal action potential spikes were detected using a stimulation threshold of −4.5 times the signal standard deviation with any spike that exceeded 500 µV being rejected as it was likely to be caused by electrical noise. The baseline neuronal firing rates were determined from −60 to −5 ms relative to the stimulation pulse of each trial while the poststimulation firing rates were determined from 2 to 8 ms. Baseline activity was subtracted from the poststimulation activity to determine the net increase.

Determining the changes in the neural activity induced by current steering
The firing rates from the 16 electrodes on each shank could be used to form a neural activity curve. The curves were smoothed with a zero-phase moving average digital filter (filter length of three electrode samples). The firing rates were normalized by the maximum firing rate recorded on the shank to reduce the impact of tissue anisotropy on the firing rates. As we are testing whether current steering can mitigate errors in electrode placement, we want to observe whether we can steer the locus of activity regardless of the electrode location within the column. Consequently, the curves with the same separation distance between the stimulating electrodes were then aligned at the stimulating electrodes and averaged across the entire population.
The peak of neural activity for each current ratio was found by determining the maximum firing rate position for each neural activity curve. Filtering removed abrupt changes in neural activity making the peak position less susceptible to stochastic fluctuations in firing rate. The peaks of neural activity and the neural activity curves could be aligned across the population relative to the stimulating electrode positions and averaged.
Testing the significance of the peak shift was done using both a non-parametric Friedman test and a permutation test which allowed for testing across the continuous current ratio variable and, additionally, the directionality of significance to be tested. For the permutation test, we randomly assigned current ratios (x variable) to peak locations of the neural activity curves (y variable). Then linear regression was performed on the data to determine the slope of the peak shift per change in the current ratio. We did this 1000 times so that we had 1000 slope variables (the null distribution) which we could then compare against the actual slope of the data (test statistic) to form a p-value.

Identification of electrode position
Electrode position across cortical layers was assessed using histological methods and/or by completing a current source density (CSD) analysis from the local field potential (LFP) using the CSDplotter tool [39]. The LFP is a low-frequency signal thought to correspond with population levels of activity [40,41]. The response amplitude of the LFP to visual stimulation is dependent on the population of neurons and, as such, changes based on the layer function/neuronal composition [37,41]. We used the aforementioned strobe light visual stimulus to evoke the LFP response (900 trials, 0.9 Hz, Lutron digitalstroboscope-ic-dt-2289). We low pass filtered (filter cutoff 300 Hz) and trial averaged the signal to observe the LFP. A CSD was constructed from the LFP and the current sources and sinks were used to identify the cortical layer boundaries [39]. As layer 4 is an input layer, it is associated with being a current sink and so the earliest electrode with a current sink was defined as the layer 3/layer 4 boundary. Based on rat visual cortex anatomical data [25], approximately three electrodes were positioned in layer 4 dictating the granular/infragranular boundary. We determined the infragranular and white matter boundary based on anatomical data [25] and implantation depth. We verified these methods through histology in three of the animals. We coated the electrodes in Vybrant CM-DiI prior to implantation so that the electrode locations could be histologically verified. Cytochrome oxidase and Nissl staining were used to confirm layer positioning relative to each electrode shank [42].

Multilayer current steering
To confirm that the array was positioned in the visual cortex, we performed a full-field flash directed into the contralateral eye of the animal (figure 1(A)). We ensured that we were recording visually responsive neurons by observing the changes in neuronal firing rate evoked from the full-field flash (301 electrodes recorded neurons responding with a firing rate above baseline (67%)) ( figure 1(B)). We also verified that these neurons responded to electrical stimulation (figure 1(C)). For every stimulating electrode, 64 electrodes could record a firing rate significantly above baseline; that is 64 stimulating-recording pairs of electrodes. A total of 4780 stimulating-recording electrode pairs recorded neurons responding above baseline levels out of 7040 total pairs (68% responding). We subsequently ensured we were likely recording across the layers of the visual cortex by observing the visually evoked changes in the LFP at different cortical depths (98% of electrodes measured a significant variation of the LFP in a 100 ms window after the flash compared to baseline (paired t-test, p < 0.05)) (figure 1(D)) [37,43]. Lastly, we confirmed Normalized firing rate response to current steering when the array was implanted perpendicularly to the cortical surface. (A) Single animal example response to current steering (6 µA total stimulation current). Separation distance between the stimulation electrodes is 500 µm in the left plot, 400 µm in the middle and 300 µm in the right plot. Each color represents a current ratio between the stimulating electrodes indicated in the key. Opaque thin lines are the average normalized firing rate while the color-coded transparent area represents the standard error. The location of the peak firing rate is shown above the neural activity curves (error bars indicate standard error). The yellow and purple vertical lines denote the stimulating electrode locations. Neural activity peak locations for each current ratio were used to assess if there was a significant shift in activity (Friedman significance test). (B) Same as (A) but the average is taken over the seven-animal population. Results were randomly sampled to N = 60 shanks for comparability between separation distances. the array placement in the cortical column using a CSD analysis in all seven animals (figure 1(D)), and histology in three of the seven animals (figure 1(E)).

Multi-electrode stimulation does not steer the peak of neural activity across cortical layers
To test whether current steering can shift the peak of neural activity in a cortical column between two stimulating electrodes, we aligned the multiunit firing rates to each pair of stimulating electrodes (example array positioning is shown in figure 1(A)). A shift in the ratio of current between the pair of electrodes should produce a corresponding shift in the location and peak of evoked neuronal activity.
In some conditions for the single animal averages (figure 2(A)) and the population averages (figure 2(B)), we found that the peak of neural activity significantly shifted as the current ratio was transferred from one stimulating site to the other with both a normalized firing rate (Friedman test, p < 0.05. 3, 4, 6, and 10 µA total current with a 500 µm separation distance between the stimulating electrodes) and a raw firing rate (3, 4, 6, and 10 µA total current with a 500 µm separation distance between the stimulating electrodes, and 6 µA with a 400 µm separation distance). However, although the peak shifts were significant when all current ratios were considered, the peak shifts in  )). When we only tested the dual-electrode conditions, the peak shift was insignificant in 76% of the tested conditions at a population average (permutation test, p > 0.05). Hence, the single electrode conditions (100:0, 0:100) were driving the overall significance.
Additionally, there was no relationship between current ratio, separation distance, and current level with a significant peak shift in the dual-electrode conditions (permutation test, p < 0.05. 4 and 8 µA for a 500 µm separation distance, 6 µA at a 400 µm distance, and 1 and 6 µA at a 300 µm distance). This indicates that dual-electrode current steering does not work to consistently govern the peak position of neural activity when the electrodes are positioned perpendicularly to the cortical surface.

Similar neural activity is evoked throughout the laminar column regardless of stimulation location
To examine if there was a particular layer combination driving the lack of shift, the data was split based on the layer that each stimulating electrode was in. Each electrode of a pair could be positioned in different cortical layers or white matter (e.g. supragranular and infragranular, or infragranular and white matter, etc). There were ten possible layer combinations; however, the shank length and electrode separation distance affected which combinations were plausible. We focused on electrode pairs with a 300 µm separation distance because it yielded the most samples across different layer combinations.
If the stimulating electrode pair were positioned in the supragranular and infragranular layers, the infragranular electrode activated the column more strongly than the supragranular electrode (∼0.2 normalized firing rate units higher on average, Wilcoxon signed rank test, p < 0.05) but there was no observable shift in the neural activity peak (Friedman test p > 0.05) ( figure 3(A)).
When the pair were positioned in the supragranular and granular layers, we saw more shifts in the peak of activity although the sample size was too low to be conclusive (Friedman test, p > 0.05) ( figure 3(B)). Granular stimulation also activated the column more strongly than supragranular stimulation (∼0.35 normalized firing rate units higher on average) (Wilcoxon signed rank test, p < 0.05).
When the pair of electrodes covered the granular and infragranular layers, the peak shift was insignificant (Friedman test, p > 0.05) ( figure 3(C)). Activation of the column in the single-electrode conditions was indistinguishable between the stimulating electrodes (Wilcoxon signed rank test, p > 0.05).
When the deepest electrode was in white matter, the firing rate reduced with depth on the array as expected (∼0.5 normalized firing rate unit drop) ( figure 3(D)). Stimulation in the white matter still evoked firing, likely because the electrode was close to the layer border and the current spread into the infragranular layers. It is also possible that stimulation in the white matter activated the input axons; however, the myelin sheath around the axons is insulative and impedes current flow making activation less likely. There was no distinguishable difference in the neural activity amplitude when the current was shifted between the infragranular layers (current ratio 100:0) and white matter (current ratio 0:100) (Wilcoxon signed rank test, p > 0.05).
When both electrodes were in the infragranular layers, the neural activity distribution was shifted towards the region with a higher percentage of current. This shift in distribution was most clear at the edges (figure 3(E)); however, the peak shift was still inconsistent (p > 0.05 Friedman test). The peak shift became more consistent and significant as the distance between the stimulating electrode pair was increased with both electrodes in the infragranular layers (e.g. 500 µm figure 3(F)) (Friedman test, p < 0.05).

Both the distance between stimulating electrodes and the current level are correlated with firing rate
Since there was no apparent consistency of the results when each stimulating electrode of a pair was in a different layer, we subsequently wanted to test the ground truth information such as the positive relationship between firing rate and current, and the positive relationship between the increased spread of the response with increased distance between stimulating electrodes [44,45].
In the 50:50 dual-electrode condition, the width of the response increased with the separation distance of the electrodes (Friedman test, p < 0.05). Per 100 µm increase in separation distance, there was a 3.6 ± 1.2% (mean ± SEM) average increase in the number of electrodes with a firing rate above half the maximum rate. This average was computed across all current levels. There was a 14.3 ± 2.37 sp s −1 (mean ± SEM) decrease in the firing rate with each 100 µm increase in separation on average across all current levels (Friedman test, p < 0.05). This is due to a decrease in charge density resulting in a decrease in peak firing rate. The peak firing rate also significantly increased with increasing current as expected since the firing rate should increase with charge density (Friedman test, p < 0.05).
As all of these trends were expected, we wanted to ensure that the lack of peak shift was not due to experimental hardware or data processing. We chose to compare our multilayer current steering results to our previously obtained results when current steering within a single layer [35].

Current steering in a single layer does steer the neural activity
We have previously implemented current steering [35] in a single layer by changing the array orientation to that shown in figure 4(A). The evoked activity was much closer to the expected output of Gaussianshaped curves with a peak location that was dependent on the current ratio between electrodes. This was true in both the single animal ( figure 4(B)) and population averages ( figure 4(C)). The shift in the peak of neural activity was significant at all separation distances in the population averages (Friedman test, p < 0.05. Permutation test, p < 0.05).
We then tested whether the peak shift was significant when only the dual-electrode stimulation conditions (75:25, 50:50, 25:75) were considered using the permutation test. The shift was significant at 43%, 71%, and 71% of the current levels for a 300, 400, and 500 µm separation distance respectively. The shift was more significant with increasing current and increasing separation distance: 8-10 µA at a 400 µm and 6-10 µA at a 500 µm separation distance were highly significant (p < 0.01). This demonstrates a positive relationship between shift, current level, and spread of the stimulating electrodes in the dual conditions as expected. Hence, the clear current-steering effects in the single-layer conditions indicate that the lack of peak shift and change in neural activity in the multilayer condition is not due to experimental hardware or data processing.

Comparison of multilayer to single-layer current steering
When the stimulating electrodes were in different layers, there was no significant current steering effect (figures 3(A)-(D)). When both electrodes were in the infragranular layers with the array orthogonal to the cortical surface, we had the clearest current steering effect (figures 3(E) and (F)). This result is consistent with the clear single-layer current steering effects seen when the array was implanted parallel to the cortical surface (figures 4(B) and (C)). However, the influence of current steering was much stronger with the array implanted parallel to the cortical surface than orthogonal. The differences between single-layer and multilayer current steering become more pronounced with a direct comparison between the peaks of neural activity (figures 5(A)-(C)). The peaks do not consistently transfer in the multilayer condition as the ratio of charge changes whereas they do in the singlelayer condition regardless of the separation distance between the electrodes.
The firing rates are similar in both conditions regardless of the ratio of charge as is shown in figure 5(D) (each point on the graph represents the firing rate from one of the current ratios). This indicates if the firing rate is high for single-electrode stimulation, it will also be high for dual-electrode stimulation which was expected. Hence, unconventional trends in the firing rate are not the reason for the differences between single-layer and multilayer stimulation.
However, the firing rate was higher on average in the single-layer condition (superficial layer) which is in agreement with previous literature where the superficial layers have the highest firing rate (figure 5(D)) [37]. As the multilayer population averages include samples from supragranular, granular, and infragranular layers, the average firing rate is reduced. We attempted to mitigate the differences in firing rate across the shank by normalizing the firing rate so that it did not affect the current steering result.
Subsequently, we investigated the differences in current steering between shanks as we hypothesized that current steering may have a stronger effect when the recording electrodes were further away and less likely to be influenced by local columnar connectivity. The results were consistent with this hypothesis. Both at the stimulating shank and 200 µm away at the consecutive shank, the neural activity curves had the same shape regardless of the current ratio with multilayer current steering. Meanwhile, the expected shift of the neural activity curves towards the electrode with the highest charge became more distinct 400-600 µm away (2-3 shanks) ( figure 6(A)).
Contrastingly, the single-layer results show distinct Gaussian curves with the trends getting noisier with distance from the stimulating shank ( figure 6(B)). The number of samples is greatest one shank away from the stimulating shank since stimulating on a middle shank resulted in two samples at this distance. This is the same reason that there are fewer samples three shanks away. As the only difference between the single-layer and multilayer conditions are the neural connections and neuron types, we began to postulate what the influence of the neural architecture is on stimulation-evoked activity.

Differences between multilayer and single-layer current steering
The efficacy of targeting certain layers within the laminar column using current steering is heavily influenced by the columnar neural network (figure 2). When the stimulating electrodes were placed in different layers, the same distribution of neural activity was evoked regardless of the charge ratio between the two electrodes. There was preferential activation in response to stimulation of the granular and infragranular layers (figure 3) which resulted in the neural activation being skewed towards these layers during dual-electrode stimulation of any layer. However, even when the pair of stimulating electrodes was split between the granular and infragranular layers effectively eliminating the preferential activation (figure 3(C)), the charge distribution between the stimulating electrodes did not affect the evoked activity. Hence, regardless of the layer combination, the recorded neural activity did not follow the pattern demonstrated when current steering was applied parallel to the cortical surface in the same layer (comparison of figures 2-4).
We did observe a trend towards significance as the separation distance increased when both stimulating electrodes were implanted in the same layer orthogonal to the cortical surface. However, the changes in the neural activity curves and shift were much less clear compared to when the array was implanted parallel to the cortical surface. Additionally, current steering within a single layer perpendicular to the cortical surface is not useful for correcting electrode placement when the electrodes are not situated in the desired layer.
Although the distribution of neural activation from stimulation is not well understood, it is often assumed that artificial electrical stimulation evokes neural activity spherically from the point of stimulation which lead to the theory of current steering [44][45][46][47]. However, we have shown that the influences of stimulation are different when electrodes are positioned perpendicular to the cortical layers compared to parallel.
Unlike a cortical column, single layers have consistency in neuron type, uniformity in connection density and length (horizontal connections), and homogeneity in the extracellular composition. This would cause stimulation within the same layer to behave more similarly to the spherical stimulation model.
Contrastingly, the connectivity between the layers of the cortical column appears to cause activation throughout the column regardless of the current ratio supplied between electrodes. We demonstrate this hypothesis in figure 7 where typical stimulation patterns are influenced by the intrinsic connections of the laminar column. This results in non-spherical firing rates being evoked from single-electrode stimulation and consequently, unconventional responses to current steering with dual-electrode stimulation.
The hypothesis in figure 7 is supported by our results where there was minimal change in the neural activity curves with changes in the current ratio. In the dual-electrode conditions, the neural activity curves and peak positions overlap considerably resulting in insignificant shifts of the neural activity peak ( figure 2).
Assuming the current is dissipating omnidirectionally with distance from the stimulating electrode, there are two explanations for this result. The first Figure 7. The influence of neural connections within the cortical column on current steering. Connectome data sourced for rat somatosensory cortex from the Blue Brain Project [62,63].
is that, in addition to measuring action potentials evoked directly from the stimulation, we are recording secondary postsynaptic potentials throughout the column which are triggered by the primary response. The second is that the recorded action potentials are from the axons of directly activated neurons which are carrying the signal away from the stimulation site. Both scenarios produce a dispersed response recorded across the electrodes. Previous studies suggest that at our largest current, we would directly activate neurons passing within 30-200 µm of the electrode indicating that something is transferring the response outside of the expected region [44,45].
Both of these phenomena would be mitigated by a change in the electrode array orientation to parallel with the cortical surface since the horizontal connections are smaller in length resulting in a less distributed response [48]. This theory is supported by shank level differences in the current steering result ( figure 6). In the multilayer condition, the effects of current steering (changes in the Gaussian shape of neural activation) became more pronounced and significant away from the stimulating shank. Comparatively, on the stimulating shank and 200 µm away, we see similar neural activity patterns with the entire column being activated. The single-layer condition has the opposite trend with the stimulating shank having the most clear result. Hence, this could be due to strong local connections in the cortical column and limited spread in the horizontal connections parallel to the cortical surface.
Other factors that may influence the measured firing rate include the densities of neurons and the type of neurons in each layer. Indeed, we observed a skewed distribution in the firing rates with stimulation in the infragranular layers, which could have been caused by the neurons within different layers having varying stimulation thresholds and/or firing rates in response to stimulation. The effect of different neuron types is unaccounted for in the spherical model of stimulation activation and is a possible explanation for the differences we observe between single-layer vs multilayer current steering.

Stimulation artifact impact on results
As a result of the stimulation artifact, up to 2 ms of the response was missed following electrical stimulation. Due to the extracellular nature of the recordings, the recorded action potentials were possibly postsynaptic [49]. This would mean the influence of neural connections is exacerbated in these results. Regardless, a higher concentration of activity was expected around the stimulating electrode with a larger proportion of neurons responding to the artificial stimulation as can be seen when current steering in a single layer [50].

Effect of separation distance and current level on the neural activity trends
Increasing the current and the separation distance increased the likelihood of the neural activity peaks being significantly different in the single-layer condition but not with multilayer current steering. It was expected that increasing both current and separation distance would create a larger significant difference in the neural activity with current steering since the firing rate would be larger overall and the Gaussian curves would have less overlap respectively. We believe this effect was masked in the multilayer condition by the dispersed response throughout the column. It is expected that increasing the current above 10 µA would eventually saturate the firing rate with both single and multilayer stimulation [44,45,49] making the curves less distinguishable. Unfortunately, due to the safe charge limits of the electrodes, we were unable to test higher current levels. This would have been beneficial as often higher stimulation currents are used to evoke phosphenes [51][52][53].

Stimulating electrode configuration
Although we used the more common monopolar stimulation configuration, bipolar current steering has been implemented previously in cochlear implants [54,55]. This is effectively phase-shifted current steering since the pair of electrodes are in opposite phases to one another, that is, one cathodic and one anodic electrode make up the steering pair. This type of steering reduces the activity around the anodic electrode by hyperpolarizing the tissue. By positioning an anodic electrode in the infragranular layers, it might be possible to reduce infragranular activity and steer the locus of neural activity towards the granular/supragranular layers. However, a non-trivial issue with phase-shifted current steering is that the electrode with the anodic phase can create cathodic side lobes [56] meaning that depolarization occurs well outside the target region. For this reason, we chose to use monopolar stimulation although we believe the phase-shifted current steering method warrants future investigation given the lack of steering effects seen across layers with monopolar stimulation.

Effects of anesthesia on stimulation results
An experimental concern is that anesthesia can reduce synaptic activity and hence, the firing rate of neurons [57][58][59]. As we recorded neuronal firing rates significantly above baseline (figures 1(B) and (C)), we believe the same significant effects would be seen when awake but with an increased rate/amplitude; however, this was not tested. The same experimental setup conditions were kept between the single and multilayer experiments indicating that these differences were not due to anesthesia.

Effects of hemorrhaging on results
There was hemorrhaging observed around penetrations in the histology. We believe this was due to the explantation process which involved reattaching the electrode array to the holder and/or pulling on the polyimide cable to explant the array. This caused array movement in the medial-lateral direction. We could have used a fixed array without the polyimide cable to partially mitigate this issue, but the arrays with the cable are much more stable allowing for recording longevity from a single penetration which outweighed the benefit of an easier explantation. We cannot guarantee that the hemorrhaging was caused by explantation; however, we did not observe bleeding until after explanting the array. If the hemorrhaging occurred during implantation, this would have affected the results by creating a barrier at the electrode-tissue interface. We continued to observe spiking activity until the conclusion of the experiment also indicating that this was less likely to be the cause of our results. Additionally, the same protocol was applied between the single and multilayer experiments.

Histology and LFP/CSD analysis
We confirmed electrode positioning in three out of seven animals using histology. A potential caveat of the study is the lack of histology completed in the first four animals. However, we estimated the electrode positioning within the layers using the LFP/CSD analysis in the last three animals prior to confirming the electrode positioning with histology. The histology was concurrent with our LFP/CSD analysis indicating that mislabeling electrode positioning is unlikely to be the cause of the current steering effect.

Implications for cortical prostheses
The spread of neural activity throughout the cortical column implies that the stimulating electrode location in a laminar direction is less important than the horizontal location for prostheses. Additionally, variations in the surgical depth of electrodes in the cortical layers are more tolerable than that of the horizontal location. However, there may be differences in the downstream activation that is layer dependent.
We did not observe significant changes in the peak location when the data was split into layers. Evidently, the cortical structure is a very prominent effector of the dual-electrode stimulation response across the layers. Multilayer dual-electrode stimulation for prosthetics would cause the entire column to activate which could make the propagation of the artificial stimulation through the visual hierarchy more likely as more feedforward circuits become active. It has been previously shown that the perelectrode detection thresholds with dual-electrode stimulation decrease regardless of the stimulation layer in rat somatosensory cortex [47]. Given the lack of change in the neural activity curves and peak locations of our dual-electrode data in comparison to the single-electrode data, less charge can be input at two single electrodes while achieving the same level of neural activity in the column. The less concentrated charge input would reduce the likelihood of stimulation-induced tissue damage at the stimulating site, and the probability of inducing a seizure [24,60,61]. Furthermore, dispersing the charge along the column would reduce the horizontal spread of charge delivered by any single electrode potentially allowing for more punctate percepts. It should be noted that we did not test visual perception and are inferring these influences based on neural activity.
To this end, we observed stimulation of the infragranular layers eliciting a larger response when current steering orthogonal to the cortical surface, however, it has previously shown in macaques that detection thresholds were lowest in the supragranular and granular layers [23]. This is a clear disconnect between the activity in V1 and perception indicating the effects of artificial stimulation are much more complex than the simple spherical models. Further investigation linking these neural activity effects to perception is needed.

Conclusions
Unlike current steering parallel to the cortical surface, current steering perpendicularly to the cortical surface does not evoke a controllable peak of neural activity. This has implications for controlling the effects of multi-electrode stimulation in neural prostheses. The underlying cause of the differences in neural response is undetermined and suspected to be related to neural connections and/or neuronal morphology. Further research could be conducted with synaptic blockers to determine if the effect still exists without the impact of these connections. Additional tests to investigate the impact of the tissue homogeneity and neuron type on stimulation characteristics are also required. However, this is the first work to our knowledge showing a difference in dualelectrode stimulation effects depending on the orientation of the electrode array within a cortical column. We did find that dual-electrode stimulation evoked similar activation of the whole cortical column at a lower current level than single-electrode stimulation per electrode. This could potentially improve the safety of intracortical prostheses by reducing the charge required at any one site to evoke a neural response.

Data availability statements
The data cannot be made publicly available upon publication because they are not available in a format that is sufficiently accessible or reusable by other researchers. The data that support the findings of this study are available upon reasonable request from the authors.