Introduction

Vision is an enormously complex process. Seeing starts when light, passing the optical parts of an eye, reaches the retinal photoreceptors. Within the retina complex neuronal circuits compress visual information from 130 million photoreceptors into action potentials transmitted by 1.2 million axons of the optic nerve to higher visual centres of the brain [11, 33, 36, 38, 49]. Blindness can result when any step of the visual pathway—the optics, the retina, the optic nerve, the lateral geniculate body or the visual cortex—is damaged. In Germany, 17,000 people become blind every year, and about 50% of all these cases of blindness are caused by retinal damage [31, 32]. Abundant blinding diseases are retinitis pigmentosa or macular degeneration, which cause progressive degeneration of the outer retina. Up to now, no medical treatment exists to cure blindness. However, the restoration of vision could be achieved by electronic devices called visual neuroprostheses [1, 3, 4, 6, 7, 8, 9, 10, 14, 15, 16, 20, 24, 25, 26, 27, 34, 35, 37, 43, 46, 48, 51, 52, 53].

A retina implant for restoring basic visual perception by electrical stimulation in patients that are blind due to photoreceptor loss should not only evoke focal phosphenes at high resolution, but should also evoke the sensation of visual objects. Thus, we started to analyse the cortical representation of electrical retina stimulation. Here we present analyses of intracortical population responses evoked by electrical retina stimulation with basic form and motion stimuli.

Materials and methods

Surgical and electrophysiological procedures were performed with anaesthetised cats. The procedures were in accordance with the guidelines of the European Communities Council Directive (86/609/EEC) and were approved by an official German Animal Care and Use Committee. In addition, we followed the NIH Principles of Laboratory Animal Care (publication no. 85–23, revised 1985), the OPRR Public Health Service Policy on the Human Care and Use of Laboratory Animals (revised 1986), the U.S. Animal Welfare Act and the ARVO guidelines.

Anaesthesia and surgical procedures

Three adequately fasted adult cats (3 to 5 kg) were premedicated with atropine sulphate (0.03 mg/kg). Anaesthesia was induced by intramuscular injection of a mixture of ketamine hydrochloride (ketanest, 10–20 mg/kg) and xylazine hydrochloride (rompun, 1 mg/kg). After orotracheal intubation, anaesthesia was maintained by ventilation with N2O/O2 (70/30%) and isofluorane (0.5–1.5%) or with air and isoflurane (0.6–2.0%). During surgical procedures we additionally applied local anaesthesia (bupivacaine, lidocain) and/or supplementary doses of ketamine hydrochloride (ketanest, 3–10 mg/kg). Continuous monitoring of rectal temperature (38°C), end-expiratory CO2 (3.8–4.2%), ECG, EEG and reflexes were used to control anaesthesia. Nutrition was provided by a continuous infusion of Ringer's solution enriched with glucose (3–20 ml/h, 1.5%). For head fixation we positioned the cats in a standard Horsely-Clarke support. A craniotomy was performed above visual areas V1 and/or V2 for subsequent recording of intracortical activities, leaving the dura mater intact. After a lateral canthotomy of one eye the lateral conjunctiva was incised. Next we performed a sclerostomy 4–5 mm posterior to the limbus. An incision of about 1.1 mm was sufficient for transscleral positioning of the stimulation electrodes onto the retinal surface. If necessary, we applied alcuronium chloride to attenuate potential residual eye movements (alloferin, 0.03–0.1 mg/kgh). In general, at the end of a recording and stimulation session sclera and conjunctiva were closed with polygalactine sutures. Atropine, gentamycin, and dexamethason were topically applied. The opening of the skull was closed by implanting the previously removed and adequately supplied bone cylinder or a plexiglass window. After the sewing of the scalp incision the cats prophylactically received a streptomycin-penicillin combination (tardomyocel, 0.1 ml/kg) and were allowed to recover. However, in one case a cat was killed for histological analyses. A schematic drawing of our experimental set-up was recently published [42].

Stimulation and recording

We used quartz-isolated platinum-tungsten fibre electrodes (80 µm shaft diameter) in different designs for retinal stimulation and cortical recording. Stimulation electrodes were arranged in a hexagonal array and had ground metal cone tips with a geometrical surface of about 2,500 µm2 [42]. However, these electrodes remained stable for biphasic charge-balanced current impulses consisting of cathodic-anodic pairs of rectangular impulses of up to 100 µA amplitude and 200 µs duration. The ground tips of the tapered recording electrodes were 25 µm in diameter and had impedances of 2–3 MΩ. The positioning of the seven retinal stimulation electrodes was achieved with a specially adapted Eckhorn manipulator drive [42]. We verified both the movement of the fibre electrodes and their close retinal contact optically (Fig. 1) and by the recording of spontaneous or visually evoked retinal activity. Retinal and cortical receptive fields were estimated by projecting light stimuli on a tangent screen 1.3 m in front of the cat according to Barlow's approach [2]. The tip positions of the retinal stimulation electrodes were projected on the screen in order to achieve a match between retinal stimulation sites and cortical receptive fields/recording sites. Repetitive control of electrode and receptive field positions was used to check their positions and to detect potential eye movements. Small eye movements of up to 100 µm (0.3° visual angle) could be compensated by allowing our electrodes to be guided over a relatively large distance of about 6 mm by the vitreous gel/cortex instead of the stainless steel guiding tubes of the electrode manipulator (Fig. 1).

Fig. 1.
figure 1

Instance of fibre electrode positioning onto the retinal surface. Note that the fibre electrodes are guided over a relatively large distance by the vitreous gel

Full size image

Electrical form and motion stimulation were performed in four experiments by using computer-generated sequences of voltage wave forms and fast voltage-current converters with zero offset current [42]. The multifocally applied stimulation currents consisted of short charge balanced rectangular impulses (200 µs) with amplitudes ranging from 10 to 100 µA. Electrical stimulation was performed centrally or para-centrally (up to 15° visual angle). The currents were applied by the fibre electrodes in mechanical contact with the inner limiting membrane against a distant large reference electrode located at the rectal body temperature sensor. The form stimuli were composed by simultaneous stimulation with neighbouring electrodes. Simple examples of form stimuli are (contrast) edges or bars. For motion stimulation we used moving 'electrical bars'.

Recordings in V1 or V2 were made with up to 16 µ-electrodes placed in corresponding cortical locations to retinal stimulation sites. Amplified broadband cortical signals were hardware-filtered into single unit activities, multiple unit activities (MUA) and local field potentials (LFP) for on-line display on oscilloscopes and were made audible. The amplified and bandpassed (1–4,000 Hz, 12 dB/oct.) cortical signals were recorded at about a 40 kHz sampling rate with a PC-based 128 channel data acquisition system (Multichannel Systems, Reutlingen, Germany).

Data analysis

We used the digitally stored broadband intracortical recordings to remove electrical stimulation artefacts and to compute MUA and LFP signals with the aid of digital filters for subsequent analyses [42]. However, MUA and LFP extracted from one electrode represent the average spike density and the average post-synaptic activity of a local group of neurones within a sphere with diameters of about 200 and 500 µm, respectively [41]. Since the cortical activation distribution in cat primary visual cortex is about 1 mm in diameter for single-site retinal stimulation [42, 50], these signals can be used to analyse the cortical responses to electrical form and motion stimulation of the retina.

By virtue of the lucid retinotopic organisation of the first visual areas of the cortex, a multifocal retinal stimulation should evoke a broader cortical activity distribution compared to a single-site retinal stimulation. Thus, a form stimulus, consisting of at least two retinal stimulation sites, should be identifiable by its evoked cortical activity. Since a moving visual object evokes a related spatio-temporal cortical activity distribution, moving electrical stimuli applied to the retina should also be identifiable in cortical recordings.

Due to convergent and divergent as well as to lateral connections in the visual system [18, 19, 28, 36, 39, 40, 44, 45], the cortical representation of two neighbouring simultaneously presented stimuli should be different compared to the superimposed responses of the singly presented stimuli. Using this hypothesis, we can construct a measure for cortical representations of electrical form or motion stimuli. For this we have to give some definitions. A compound stimulus is made up of elementary stimuli. For example, an electrical form stimulus consists at least of two point stimuli, and an electrical moving stimulus may be constructed from a form stimulus by shifting its spatial retinal stimulation position in time. If the neuronal response to a compound stimulus is identical to the superimposed responses of the corresponding elementary stimuli, we cannot state whether there is interaction between the neurones or not. But if the signals differ, we can conclude that there is a neuronal interaction between the neurones activated by the elementary stimuli of the compound stimulus. In other words, if there is any signal difference, then the different stimuli may be differently processed by the neurones. In mathematical terms we have to show that:

$$r_c \left( t \right) \ne r_s \left( t \right) = \sum\limits_{k = 1}^K {r_{s,k} \left( t \right)} ,$$
(1)

where r c (t) is the response of the compound stimulus and r s (t) is the response constructed by superimposing the responses r s,k (t) of the K elementary stimuli. An example of this approach is shown in Fig. 2.

Fig. 2.
figure 2

Prediction of cortical responses. A An illustration of the generation of a moving electrical bar stimulus. B Cortical LFP response to a moving electrical stimulus (average response). C Cortical LFP responses to static bar stimulations. D Prediction of the moving bar response by superimposing the corresponding static bar responses

Full size image

Instead of comparing the time courses of the responses directly, we compared the square roots of the average signal power of the responses:

$$P\left( r \right) = \sqrt {{1 \over {NT}}\sum\limits_{i = 1}^N {\sum\limits_{t = 0}^{T - 1} {r_i^2 \left( t \right)} } } ,$$
(2)

where r i (t) denotes a single response and T is the number of its sampled values. However, the average power of a response is close to the spike count in single-cell response analysis. This gave us the possibility to compute tuning curves of the local neuronal groups for the orientation of simple electrical bars and for the direction of moving electrical bars. Standard deviations of mean values were computed according to the Gaussian law of error propagation [30]. The response powers of measured and corresponding synthetic responses were assumed to be different, if their absolute difference was larger than twice the sum of the corresponding standard deviations of the means (two-sigma rule).

The spatio-temporal sampling of cortical activities evoked by visual or electrical retinal stimulation will enable a backprojection of the cortical activations to retinal or visual space, if the receptive fields of the cortical neurones recorded from are known. If we know the positions of the retinal stimuli, we will be able to compare them with the backprojections of the cortical activities. We estimated these backprojections by computing the instantaneous power of the average responses by a sliding window technique and used the centres of the receptive fields as samples for the computation of visual/retinal space maps of the cortical activities. Temporal sequences of these maps were used for comparing moving electrical stimuli and their evoked responses.

Results

Figure 3 shows some retinal and cortical electrode arrangements, the time course of the stimulation current, the cortical receptive fields, stimulation electrode locations and cortical responses to single-site electrical stimulation. Note that all retinal electrodes evoked cortical activations detectable in MUA and LFP, respectively. Furthermore, we verified our previous findings [42] that, for a successful retina stimulation with 200 µs rectangular current impulses, current amplitudes in the lower µA range will be sufficient if the electrodes are in close retinal contact. Note that these values correspond to charges in the nano Coulomb (nC) range.

Fig. 3.
figure 3

Example of experimental parameters. A Arrangement of stimulation and recording electrodes and time course of the applied stimulation currents. B Retinal positions of stimulation electrodes and cortical receptive fields. Orientation preferences of the cortical neurones are coded by different grey scales; an oval plotted receptive field indicates no orientation preference to visual stimulation. C Cortical LFP responses to single site electrode stimulation with biphasic charge balanced currents consisting of consecutive cathodic-anodic rectangular 100 µA impulses, each with a duration of 200 µs. Note that the response amplitudes may vary considerably according to different retinal stimulation sites

Full size image

Depending on both the retinal stimulation and the cortical recording site, the evoked activities can vary vigorously. Calculations of the cortical activation distribution revealed sizes of about 1 mm in diameter, depending on the stimulation currents' amplitudes and the contact between the stimulation electrode and retina, which was, as the temporal resolution of about 20 ms, in conformance with previous results [42].

The evidence that slightly different electrical form stimulations of the retina can result in related different cortical activation distributions is shown in Fig. 4. The simplest electrical form stimulus is a simultaneous stimulation with two neighbouring electrodes. Comparison of the responses to form stimuli and to corresponding predicted responses showed that the amplitudes of the predicted responses can be larger or smaller than then those of true form stimuli, indicating an inhibitory- or excitatory-like interaction, respectively. Analyses for 15 different two-point electrical stimuli, achievable with our electrode array, revealed that the interactions in the visual pathway (retina, geniculate body, visual cortex) depend on the distance of the retinal stimulation electrodes. A more detailed analysis revealed that the superimposed LFP responses were larger than the measured responses for retinal electrodes with distances ranging from 3 to 4° visual angle (Table 1) and vice versa for smaller and larger distances between the stimulation sites.

Fig. 4.
figure 4

Cortical LFP responses to different form stimuli and their projections into visual space. The upper half of the charts (A) were obtained from stimulation with true electrical form stimuli, whereas the lower half of the charts (B) are the computed response predictions for the two different stimuli, as shown between the upper and lower contour plots on the left and right side, respectively. Note the small projected activation zones, calculated at t=14 ms post stimulation, with diameters of about 1°. However, the time course and the projected activations were different for both stimuli. The comparison of the responses evoked by form stimulation with the corresponding predicted responses revealed that predicted response amplitudes were often larger, indicating an inhibitory-like interaction between neighbouring stimulation sites. In addition, the spatial activation distributions are slightly broader for the predicted than for the measured data

Full size image
Table 1. Differences of normalised LFP response powers for two point form electrical stimuli and the predicted response powers obtained by superpositioning the responses to corresponding single-site electrical retina stimulations (P form–P pred). Class denotes the distance of the retinal stimulation electrodes; n is the number of the analysed data sets, and σ denotes the standard deviation of the mean. Note that all power differences are highly significant (P<0.05)
Full size table

The backprojection of the cortical activation to visual or retinal space revealed activation spots with diameters of about 1° visual angle. The comparison of the response profiles to form stimuli and their slightly rotated counterparts showed that the stimuli generated different cortical activations and differentiable and stimulation pattern-related visual space projections. This also indicates that electrical form stimuli can be transmitted to the visual cortex by electrical stimulation and that the retinotopic projection was somewhat preserved.

The tuning of local groups of cortical neurones to moving electrical form stimuli is shown in Fig. 5. The prominent differences in direction-related response amplitudes between measured and predicted data show that slightly different moving stimuli can lead to distinguishable cortical response distributions. This indicates that different moving electrical stimuli are processed differently in the early stages of the visual system. Here we analysed 288 MUA data sets (including the reference data used for prediction) and found that the number of significant neuronal interactions depended on stimulus velocity: 46 of 96 for 35 °/s and 20 of 96 for 91 °/s. However, we expected to find by chance about five significant interactions. In addition, we found that the tuning to moving electrical stimuli is two to three times less pronounced for LFP recordings acquired in parallel with the MUA recordings.

Fig. 5.
figure 5

Tuning of multiple unit activities to moving electrical bars. A Parameters of the moving electrical bars. Bars moved in six different directions. The interval between two subsequent form stimuli was 23 ms. B Schematic illustration of the time window used for the subsequent root mean square analysis. C Tuning of 16 simultaneously recorded spike activities of local intracortical groups of primary cortex neurones (MUA) to six different directions of the moving electrical bar. Note the different scaling of the normalised amplitudes. The thick black lines signify the mean response amplitudes for true stimuli/responses, the thin black lines indicate the distance of the respective standard deviations of the means. The means and standard deviations of the means for predicted responses are drawn in grey lines

Full size image

Discussion

It is well known that single neurones of the primary visual cortex exhibit orientation and direction tuning or selectivity, i.e., they respond best to a specific orientation and/or movement of a visual contrast edge or a light bar [22, 23, 33]. Thus, we used elongated electrical stimuli generated by synchronous stimulation with retinal electrodes. Since neighbouring cortical neurones have similar coding preferences we restricted our analysis in this first investigation to the relation between electrical retinal form and motion stimuli and intracortical group signals.

In accordance with our previously published data [42], we found that single-site retinal stimulation evoked single localised cortical activation zones of about 1 mm in diameter. However, as a result of retinal anatomy epiretinal stimulation may evoke multiple cortical activation spots if a retinal stimulation electrode stimulates neighbouring ganglion cells and passing axons of distant ganglion cells. The stimulation of passing axons can be detected by comparing retinal stimulation sites with the positions of cortical receptive fields; i.e., if the stimulation electrode's retinal location and the cortical recording site are in non-corresponding positions, we can assume that axon stimulation will evoke the appendant cortical activation. This can be tested in future work by a sophisticated re-positioning of stimulation and recording electrodes or by increasing the number of recording electrodes to extend the cortical area simultaneously recorded from.

In 1999 Humayun and colleagues [26] reported on pattern electrical retinal stimulation in humans. They were able to evoke the perception of phosphenes by a hand-held intraocular electrode array. They stated that charge thresholds were in the µC range, which is about three orders of magnitude larger than in our findings. Despite this, the subjects were able to differentiate between different stimulation patterns. Unfortunately, they were not able to record cortical responses to electrical retina stimulation. However, our results indicate that electrical form and motion stimulation of the retina can be assessed by the distribution of cortical responses. We found initial hints for temporal and spatial processing of electrical form and motion stimuli. The highest inhibitory-like neuronal interaction was obtained for form stimuli consisting of two stimulation sites being 3.5° apart, for other distances we found an excitatory-like interaction, and for motion stimuli with an interstimulus interval for two sub- or elementary stimuli of about 60 ms (35 °/s). However, for higher velocities (91 °/s, 23 ms), the interaction decreases. Note that this is in correspondence with Humayun's [26] result that flicker fusion occurs for stimulation rates of about 50 Hz =1/(20 ms) as well as with our data for temporal resolution for single-site electrical retina stimulation [42].

A closely related issue concerns the minimal time difference for which two stimuli can be resolved in cortical responses as single events. This topic, which addresses both repeated identical stimuli and subsequent elementary stimuli of a compound stimulus, is of importance for the estimation of an upper limit of transferable information to the visual system by retinal implants. We have addressed this in part by wave and information theoretic analyses for single-site stimulation [42] and found a temporal response resolution as far as 50 ms or 30 bit/s. If we contemplate on the multiple unit response shown in Fig. 5B, we see that the single responses to the elementary stimuli can be clearly differentiated. In order to find the minimal time difference for which two stimuli can be separated by their corresponding responses, the interstimulus time interval must be decreased until the single responses merge. Preliminary, analyses of our data indicate that interstimulus time intervals of much less than 23 ms have to be used and that the stimulation current's amplitude and the temporal neuronal resolution are correlated, i.e., the higher the stimulation current the higher the resolution. However, this important issue should be addressed in future experiments in combination with more realistic electrical stimulation sequences mimicking the activities of retinal neurones to more natural visual stimuli.

Our estimates for spatio-temporal interactions on the basis of cortical activation may also depend on the effective coupling among neurones in the retina, the geniculate body and the cortex. This effect can be due to divergent excitatory feed-forward connections (cat V1:0.5–1 mm; [18, 19, 40]) and lateral intracortical facilitatory, inhibitory, and disinhibitory connections [28, 29] or by feedback projections form higher cortical areas [21, 39]. However, all these connections can lead to interaction zones in cat V1 of about 2 to 3 mm as measured by signal correlations of intracortical activities evoked by visual stimulation [5, 12, 13, 17, 47] that are compatible with our results for electrical retina stimulation.

The comparison of measured (true) and predicted (synthetic, superpositioned) responses to electrical form and motion stimuli revealed stimulus-dependent interactions in the visual pathway. Since our analyses have merely allowed to test whether there is an interaction or not, a more detailed analysis of the cortical recordings should clarify the nature of the interactions, i.e., whether they are inhibitory or excitatory, as presumed by us, or not. In addition, the strongest interaction for electrical stimulation occurred for velocities that were about two to three times higher than those expected for visual stimulation [36]. This could be due to the fact that we performed a coarse spatial electrical stimulation and recording. Thus, this issue should be addressed again with many more and more narrowly spaced electrodes and more sophisticated stimulation waveforms.

Our data have shown that electrical form stimuli and moving electrical form stimuli can generate different cortical activation zones, indicating that different stimuli can be differently represented. Furthermore, our results indicate that the stimulation of the retina with electrical form and motion stimuli can lead to spatio-temporally related cortical activations. However, the selective activation of single cortical neurones with specific tuning or classical receptive properties by electrical retinal stimulation and the potential adaptation of the visual system to long-term stimulation with retina implants should be addressed in future work.