Outer Retinal Degeneration: An Electronic Retinal Prosthesis as a Treatment Strategy

Objective  To review progress toward an electronic retinal prosthesis for outerretinal degeneration.

Method  Literature review.

Results  Retinal degenerations such as retinitis pigmentosa result in a lossof photoreceptors. There is a secondary loss of inner retinal cells, but significantnumbers of bipolar and ganglion cells remain for many years. Electrical stimulationcan produce phosphenes in the eyes of individuals who are blind as a resultof retinitis pigmentosa. Several research groups are trying to exploit thisphenomenon to produce artificial vision with electronic retinal prostheses.Two groups, with private company sponsorship, have recently implanted first-generationdevices in subjects with advanced retinitis pigmentosa. They have reportedlimited preliminary results. This article seeks to put these results in abroader context and review potential obstacles to successful prosthesis development.These include inner retinal cell viability, high thresholds, signal encoding,power requirements, biocompatibility, and device encapsulation.

Conclusion  There has been substantial progress toward an electronic retinal prosthesis,but fully functional, long-lasting devices are not on the immediate horizon.

Retinal degenerations such as retinitis pigmentosa (RP) initially resultin photoreceptor loss. Later in the course of the disease, there is secondaryloss of inner retinal neurons. Postmortem studies demonstrate the presenceof inner retinal cells even in eyes severely degenerated by RP (Figure 1). Counts of cells in the inner nuclear layer suggest that40% to 88% are retained, whereas 20% to 48% are retained in the ganglion celllayer, depending on the severity of the degeneration and the retinal areasampled^1-3 (Table 1).

The mechanism of damage and loss of inner retinal neurons in RP is unknown.The conventional explanation for the loss of inner retinal cells is transneuronaldegeneration. Another hypothesis involves the action of retinal pigment epithelialcells that migrate to the inner retina. These cells may envelop, invade, andocclude retinal vessels. The resulting ischemia may cause loss of ganglionand bipolar cells.⁴ The proximity to the choroidenjoyed by the bipolar cell layer after loss of the outer retina may accountfor the relative preservation of these cells.

Because the genetic defects for numerous types of RP are now known,gene therapy for these conditions has been proposed.⁵ Althoughgene therapy may hold great promise for halting retinal degenerations, itdoes not seem likely that it could restore lost function. It is very difficultto insert genes into postmitotic cells.⁶ Thenumerous genetic subtypes of RP⁷ suggest thatindividually tailored approaches might be necessary for each mutation. Inaddition, many mutations have yet to be identified. Retinal or retinal pigmentepithelial transplantations have been proposed as therapy.⁸ Thusfar, extremely limited functional results have been obtained with these transplantations.^6,9 Antiapoptotic treatment shows somepromise but, again, results in animal models have so far shown limited success.^6,9

Preservation of the inner retinal neurons in RP raises the possibilitythat appropriate stimulation of these cells may produce vision. The ganglioncells are the "output" cells of the retina. Driving them with suitable stimulation,electrical signals for example, may mimic their usual signals to the brain(Figure 2). In addition, the ganglioncells have a topographic configuration that closely corresponds to the visualfield. This would suggest that it would be simpler to achieve patterns ofstimulation that correspond to objects in the visual field than would be possiblewith cortical stimulation, where the topographic relationship to the fieldis much more complex.

The concept of an electrical retinal prosthesis was introduced as earlyas 1956 by Tassicker¹⁰ in Melbourne, Australia.A variety of prosthetic designs have since been proposed. The simplest isa 1-piece subretinal device consisting of multiple small photodiodes (Figure 1 3A). When light is absorbed by aphotodiode, it is converted to electricity. Each photodiode acts like an artificialphotoreceptor, receiving ambient light and converting it to a graded electricalresponse.^11,12 This electricalresponse may then act to stimulate adjacent nerve cells, such as bipolar cells.Since the ambient light arrives in a topographically appropriate distributionand also supplies power, this type of device is very simple in design. Epiretinaldevice designs typically are more complex because they rely on external imagingdevices and power sources. Figure 3Bis an example of one type of epiretinal device. An external television cameraconverts ambient light to an electrical signal. This signal is transmitted,along with additional electrical power, by an induction coil on the temple.A second coil on the scleral surface receives the power and signal and transmitsthese via a small cable to a microchip inside the eye. This chip containselectrical circuits that distribute the signal appropriately to small electrodesthat contact the epiretinal surface. Other epiretinal device designs are illustratedin Figure 3C and Figure 3D. Designs vary according to how much of the required electroniccircuitry is contained in the intraocular device vs extraocular elements.Power and signal transmission may be accomplished by penetrating wires, inductioncoils, or lasers. Rizzo and Wyatt¹³ suggestedplacing the intraocular electronic components inside a modified intraocularlens, away from the retinal surface, and running a ribbon of electrodes fromthis device to the retinal surface.

Cortical implants work in a manner analogous to epiretinal implants.An external miniature television camera sends its signal to an induction coil,which transmits the signal and power to a secondary coil buried under thescalp. Power and signal travel via wires to a microchip, which distributeselectrical current to an array of electrodes that contact the primary visualcortex.

It has been known for many years that passing an electric current throughthe healthy eye can produce the sensation of light. Potts and Inoue¹⁴ demonstrated in 1969 that external electrical stimulationof the eye could elicit perception of light and a cortical response (electricallyevoked response) in some subjects with RP, even when the visual evoked potentialwas essentially absent. These results seem to indicate that at least someretinal ganglion cells and more central elements of the visual system retainsome function even in very advanced RP. A retinal prosthesis would dependon these remaining elements, limiting its use compared with a cortical prosthesis,which could potentially provide vision in instances where even the optic nervehas been destroyed. A cortical prosthesis might be able to help patients withblindness from diabetic retinopathy and glaucoma, 2 of the leading causesof blindness in the industrialized world, whereas a retinal prosthesis wouldbe unsuccessful because of the loss of the ganglion cells that make up theoptic nerve. A retinal prosthesis might also be useful in macular degeneration,where there is preservation of inner retinal elements.¹³ See Table 2 for a comparison of the major implanttypes.

Several groups in the United States, Germany, Japan, Australia, andKorea are actively pursuing research in visual prostheses for individualswith blindness. Prostheses have been suggested that electrically stimulatethe retina,^11,12,15-19 opticnerve,²⁰ or visual cortex^21-25 inan attempt to produce artificial vision. Two groups are designing devicesbased on the local release of glutamate.^26,27 Thereis currently no device that fulfills the mission but incremental progressand success with cochlear implants for individuals with deafness provide theimpetus to continue.

Chow et al^28-30 andHumayun et al³¹ recently inserted prototypeimplants in human subjects with advanced RP. The inevitable publicity surroundingthese events, and the inability of most media to convey a thorough contextfor it, prompted us to prepare this review to provide a balanced discussionof the current state of knowledge regarding retinal prostheses.

Major issues remain for prosthesis development. We will discuss problemswith inner retinal cell viability, stimulus threshold, signal encoding, powerrequirements, biocompatibility, encapsulation, and testing of implant subjects.

As noted previously, a retinal prosthesis requires the presence of someviable retinal ganglion cells. The number, type, and location of viable ganglioncells that would be required for a useful prosthesis is unknown.

Although the studies cited previously show that numerous retinal ganglioncells remain even in advanced RP, the viability of these cells is unknown.Counts of ganglion cells are difficult and may be artifactually elevated bythe presence of displaced amacrine cells. Counts of inner nuclear layer cellsperformed with standard histochemical stains include 5 cell types so the proportionof lost bipolar cells may be masked by preservation of other types.

The types of ganglion cells that remain are unknown. It is possiblethat selective loss of critical types may limit the results of electricalstimulation. In retinal areas where significant numbers of viable photoreceptorsremain, it is doubtful that a prosthesis could improve existing vision. Thisis likely to limit prosthesis use to areas of severe degeneration.

It is not known whether the reduction in the number of ganglion cellsin degenerated retinas is consistent with useful pattern vision. Some datafrom acute electrical stimulation experiments in humans with advanced RP thatbear on this question are discussed later. Experience with cochlear implantssuggests that these prostheses may be useful when only 10% of spiral ganglioncells remain.³² The relevance of this observationto the retina is uncertain.

The mechanism of ganglion cell loss in RP is unknown, as reviewed previously.Ganglion cell loss may continue in the presence of a prosthesis. There isno currently accepted method for noninvasively assessing the viability ofretinal ganglion cells in the absence of photoreceptors. Two groups have usedthe electrically evoked response as a method to evaluate subjects who volunteeredfor intraocular stimulation experiments.^16,33 Althoughthere was a correlation between extraocular and intraocular thresholds inone small study,³³ sufficient data are notavailable to rely on this test. Selection of the most appropriate candidatesfor a prosthesis will therefore be limited unless such a method is developed.

In summary, although significant numbers of cells remain in the innernuclear and ganglion cell layers of the retina in advanced RP, we have littleinformation about the types and subtypes of these cells. We do not have amethod for noninvasive determination of the viability of these cells. Studiesto identify the remaining cells and to noninvasively estimate their functionare needed.

The amount of electric current that will be required to stimulate innerretinal cells and produce perception is a critical factor in implant design.First, the safety of any device will depend on keeping stimulus charge levelsin a range that does not damage the retina. Second, the heat created by adevice is primarily driven by its power consumption. The largest amount ofpower is consumed at the electrode-tissue interface because of its electricalresistance. Heat must be kept at levels that will not damage ocular tissues.Third, methods must be developed to deliver sufficient power to a device toallow suprathreshold stimulation of a sufficient number of electrodes to createuseful vision. The stimulation threshold is a primary driving factor in determiningthese parameters.

Stimulation thresholds have been studied in a wide variety of modelsand species with very different methods, making direct comparisons difficult.The charge density required to produce a response is a critical variable.Many investigators do not calculate charge densities for their stimuli (orgive sufficient data for the reader to do so), compounding the problem. Afurther difficulty is that the calculated charge density based on injectedcharge and electrode geometry does not take into account local "hot spots"(eg, at electrode edges) where the charge density may be considerably higher.Thresholds also vary significantly with stimulus conditions. For the purposesof this review, we present results obtained from intact animals and humanswith fairly similar techniques in Table3. In the rabbit studies presented in the table, either epiretinal-or subretinal-stimulating electrodes were placed. Recording electrodes wereplaced at the visual cortex. The smallest amount of electricity at the retina(given in coulombs [C] per centimeter squared of charge density) that wouldyield an evoked response at the visual cortex (electrically evoked potential)is given. In the human experiments, humans who were awake were tested. Eitherhandheld intravitreal electrodes close to the retina or microfabricated electrodearrays in contact with the epiretinal surface were used. Electrode currentand duration was varied until the subject reported a barely visible perception(usually a small spot of light). The human experiments are described furtherin the "Signal Encoding" section.

Thresholds have generally been lower for subretinal than epiretinalstimulation. In retinas with outer degeneration, thresholds have been consistentlyhigher. Unfortunately, data in humans have not been available for subretinalstimulation.

The explanation for higher thresholds in degenerate retinas is not clear.One possibility is that the remaining cells are abnormal in some way thatelevates their thresholds. Another possibility is that a lower-threshold populationof cells (most likely the photoreceptors³⁹)is destroyed by the degeneration and the remaining classes of cells (perhapsbipolar or ganglion cells) normally have higher thresholds. Greenberg⁴⁰ measured latencies and chronaxies of responses toelectrical retinal stimulation in frogs and in humans with RP. He concludedthat for longer pulses, the bipolar cell is the most likely site of responseto epiretinal stimulation in degenerated retinas. Ziv et al,⁴¹ workingwith isolated rabbit retinas, found evidence that brief pulses on the epiretinalsurface stimulate retinal ganglion cells whereas longer pulses stimulate bipolarcells.

In summary, threshold measurements for perception have been obtainedin humans with advanced RP by surface electrical stimulation. These thresholdsare significantly higher than those from healthy humans or those for evokedpotential recording in animals with outer retinal degeneration. Perceptualthresholds for subretinal stimulation in humans with advanced RP would beof great interest, but these have not been done.

Further consideration of the implications of these results is discussedin the "Biocompatibility" section.

Retinal prostheses are being designed to electrically stimulate cellsthat survive degeneration to produce artificial vision. Investigators mustdemonstrate that electrical stimulation excites visual cells in a predictablemanner in order for a prosthesis to be reliable. Once stimulation parametersgive reproducible visual effects, researchers must learn how to adjust theseparameters to create useful vision.

Retinal ganglion and bipolar cells are topographically arranged in anorderly distribution across most of the field of vision. This has led to thesimple concept that an array of electrodes can be placed against the retina,with rows and columns like lights on a scoreboard, and that activation ofelectrodes in the array in a given shape might yield perception of a similarshape. In several respects, this concept is an oversimplification. There aremany types of retinal ganglion and bipolar cells, which are small and closelyspaced. Electrodes in a prosthesis will have to be large to avoid exceedingsafe charge-injection limits. Each electrode is likely to stimulate many celltypes indiscriminately. These cell outputs would then be very different thanthe normal physiologic situation in which there is a different orchestrationof responses. Even worse is the possibility that focal stimulation might activateganglion cell axons representing many cells across a broad area of the retina.This is likely to produce diffuse rather than focal perceptions. Work in animalsand humans has therefore been undertaken to investigate the physiologic andperceptual consequences of electrical stimulation.

Animal experiments have shown that subretinal or epiretinal electricalstimulation in isolated retina^12,42-46 andeyecup³⁴ preparations influences ganglion cellactivity. The patterns recorded from ganglion cells in response to focal electricalstimulation in some experiments seem comparable with those evoked by focalspots of light. Retinas from animals with degenerated, chemically inactivated,or chemically decoupled photoreceptors still show ganglion cell responsesto focal electrical stimulation.

Cortical evoked potentials have been recorded in a variety of animalsin response to focal epiretinal^36,47-49 andsubretinal^11,50-52 stimulation.The cortical evoked response waveforms and amplitudes are also comparablewith those evoked by focal spots of light. Although these results are encouraging,they are a long way from demonstrating the production of useful pattern visionfrom electrical stimulation of the retina. It should be remembered that acomplex visual evoked response can be recorded in humans with stimulationby a diffuse flash of white light alone, showing that the evoked potentialsin animals are not necessarily indicators of pattern vision.

Because of the limitations of animal experiments, 3 groups have undertakenepiretinal electrical stimulation of the retina in human volunteers.^{16,35,37,53,54} Mostexperiments were done in volunteers with advanced retinal degeneration, although2 volunteers had healthy retinas and orbital cancer. Two others were undergoingvitrectomy for other conditions.

The first human tests¹⁶ were done withepiretinal stimulation and handheld intravitreal electrodes. Subjects reportedsmall spots of light in response to electrical stimulation. Two-point discriminationwas also reported using multiple electrodes. Correlation of the length ofperception with the duration of the stimulus convinced the investigators thatoperating lights were not the cause of perception. These results were veryencouraging.

Later experiments by others aimed at eliciting form perception showedresults that conflicted with these early reports.^{35,37,53,55,56} Onegroup, using handheld electrodes, showed that subjects could distinguish avertical from a horizontal line, a few letters, or a square shape. Anothergroup found that a subject was able to perceive a shadow from a handheld electrode,even when no current was delivered. The later investigators felt this approachcould lead to false-positive perceptions. Additional experiments were donewith electrode arrays in contact with the retina. With this device, subjectssometimes reported single round percepts to electrical stimulation with asingle round electrode. Frequently, however, they reported multiple perceptsto single-electrode stimulation (which did not occur in a subject with normalsight). Stimulation of multiple electrodes, in contrast, did not always producemultiple percepts. The percept reported from stimulation of the same electrode(s)with the same electrical parameters did not always produce the same percept.Results in the subjects with retinal degeneration were reproducible 66% ofthe time and in the healthy subject, 82% of the time. Clear-form perceptionas obtained by others was not obtained, even in the subject with normal sight.This group therefore determined whether perceptions met a "reasonable expectation"based on the configuration of stimulated electrodes. This expectation wasmet for the subjects with RP 48% and 32% of the time for single- and multiple-electrodetrials, respectively. For the subject with normal sight, this expectationwas met 57% of the time. Two-point discrimination could not be obtained inall subjects.

These disparities between the groups may be due to differences in methodsor subjects. It is surprising that microfabricated electrodes in direct contactwith the retina would yield results that were seemingly less satisfactorythan those obtained by handheld electrodes. It is also difficult to explainthe disparities based on subject selection because Humayun et al^16,35 hadmany subjects with severely reduced vision, some of whom had no light perception.Although results obtained by Rizzo et al^37,56,57 inthe subject with normal sight were superior to those they obtained in theirsubjects with RP, they still did not obtain perceptions that consistentlymatched the stimulation patterns. This suggests that the stimulation methods,rather than the severity of the retinal degeneration, may be responsible.Eckmiller et al¹⁷ have suggested that the stimulusthat will be required to produce pattern vision by epiretinal devices is likelyto be very complex. They have designed a method to optimize stimulus parametersusing a learning neural network approach. It is likely that this device couldonly be tested in a chronic implantation, not an acute experiment. In a preliminaryreport of 2 acute human tests, Eckmiller et al⁵⁴ apparentlydid not use this approach, although they did use careful psychophysical methodswith control subjects. Their report (to date) concentrated on the technologyand methods and did not report results of attempts to produce pattern vision.

Proponents of subretinal stimulation argue that this approach will providepattern vision with much simpler stimuli because it takes advantage of anyremaining processing capability in the middle layers of the retina. It wouldbe of great importance to test this hypothesis with an active subretinal device.Safety and technical considerations have precluded doing this type of testingin an acute experiment. (See "Testing of Implant Subjects" section for discussionof preliminary results in subjects with chronic subretinal implants.)

In summary, visual cells can be excited in a predictable manner by electricalstimulation. Cortical evoked responses have been obtained in several speciesby direct electrical stimulation of healthy and damaged retinas. Perceptionof light has been demonstrated in numerous human volunteers with advancedRP to intravitreal and surface electrical stimulation. Efforts to producepattern vision in human subjects have yielded conflicting results.

Most current subretinal device designs consist only of an array of subretinalmicrophotodiodes. Although the simplicity and similarity to the natural situationare conceptually attractive, experiments show that such devices do not generatesufficient current from ambient light alone to stimulate inner retinal elementsin animal eyes.^58,59 Recent workby Gabel et al⁶⁰ showed that cortical activationsecondary to retinal stimulation with such a device required brightness comparableto 2 to 3 times sunlight levels. Simple photodiodes will also not producecharge-balanced pulses, which are the safest form of electrical stimulationof nerve tissue.⁶¹ Across time, pulses thatare not charge balanced will lead to dissolution of metal with toxicity toneural tissue and loss of electrode function. Methods to amplify these signalsand produce charge-balanced pulses are proposed but these add significantcomplexity.⁶⁰

It is not practical to use an intraocular battery to deliver power toa visual prosthesis for several reasons. Repeated intraocular surgery to replacebatteries poses unacceptable risks. Potential toxicity of chemicals in batteriesis unacceptable. Finally, the weight of batteries is likely to be prohibitive.Power will therefore have to be transmitted from outside the eye. Becausethere are severe limits on the amount of power that can be delivered thisway, the power requirements of a prosthesis are critical.

The power requirement for a retinal prosthesis depends primarily onthe threshold charge needed for perception and the number of electrodes inthe stimulating array. As noted in the "Stimulation Thresholds" section, wehave some measurements of epiretinal thresholds in subjects with advancedRP. These show considerable variability. Although it seems reasonable to assumethat subretinal thresholds will be lower, we do not yet have these numbers.The number of electrodes needed to achieve form perception, if this indeedcan be accomplished, is unknown. Simulations of prosthetic vision have beenattempted in human subjects with normal vision.^62,63 Thesestudies suggest that a 4 × 4 array (with 16 electrodes) is unlikelyto allow vision better than crude localization. A 6 × 10 array (with60 electrodes) might allow spot reading and object recognition. The relevanceof these simulations is uncertain because they were carried out in subjectswith normal visual systems. It is likely, however, that at least this manyelectrodes would be required in diseased retinas.

Using thresholds obtained from acute human trials, Caulfield et al⁶⁴ estimated the number of electrodes that could bedriven by epiretinal devices. One device would use an external laser to deliverpower to an intraocular receiver. They estimated from a model that 15 mW couldbe delivered and that this would be sufficient for 208 electrodes. Using adevice with power transmission via induction coils, up to 3.4 mW could bedelivered, sufficient to drive 47 electrodes. Power transmission via inductioncoils would be simpler and more reliable than transmission with a laser. Thelatter requires precise alignment of the laser with an intraocular targetand clear media.

In summary, the total power required for a retinal prosthesis is currentlyunknown because we have only a few threshold measurements from epiretinalelectrodes in humans with RP and we do not know how many electrodes must bedriven to produce useful vision. Using induction coils for power transmission,the best current estimates suggest that there will be barely enough power.A laser delivery system can provide more power but will be more difficultto develop and use. Simple subretinal devices consisting only of microphotodiodesare unlikely to produce sufficient power.

The ideal implant material would be nontoxic to the retina and wouldnot elicit rejection, inflammation, or fibrosis from the host. For an electronicimplant, 3 factors must be considered: (1) chemical, biophysical, and immunologicalreaction to the implant materials and surgery, (2) reaction to electricalstimulation, and (3) heating of the tissue. These elements cannot be entirelyseparated because reaction to electrical stimulation depends in part on thematerial that the electrodes are composed of, whereas heating depends on powerconsumption, materials used, and implant location.

A variety of materials that might be used for implants have been testedin rat retinal cell cultures.⁶⁵ Cell survivalon these materials was poor. Coating of the materials with poly-D-lysine,poly-L-lysine, or laminin greatly improved cell survival on all materialsexcept titanium nitride. Chronic, inactive implants have been observed forbiocompatibility in animal eyes. Epiretinal implants^66,67 didnot dislocate, and histologic changes around them were minimal. Electroretinogramresults and visual evoked potentials remained normal in these eyes. Subretinalimplants^42,58,68 alsodid not dislocate. These eyes, however, show marked loss of outer retinalcells, with fibrosis and/or retinal pigment epithelial changes in many cases.Explantation of electrically active devices with gold electrodes from animaleyes showed dissolution of the gold after 8 months. A careful histochemicalstudy suggested that the outer retinal degeneration across subretinal implantsin cats with normal vision resembles naturally occurring degeneration in Abyssiniancats.⁶⁹

In summary, epiretinal devices secured with tacks over normal retinasshow little histologic change in the underlying retina, although the extentto which these arrays actually contact the retina is not obvious. The functionalintegrity of the underlying retina has not been determined. Subretinal devicesimplanted under normal retinas cause severe degeneration of the outer layersand a variable degree of changes in the inner retina. Modification of thesubretinal devices with perforations may decrease the degree of overlyingdegeneration. Histologic features have not been studied after chronic placementof a subretinal device in a retina with preexisting outer degeneration, whichis the clinically relevant situation.

Charge injection into tissue may be accomplished by faradaic or capacitivemechanisms. The capacitive mechanism is ideal because it allows charge transferwithout transfer of electrons, which will cause potential damage to the hosttissue or the electrode. Unfortunately, the amount of charge that can be deliveredper unit area (ie, charge density) is too low for the purposes of a retinalprosthesis. In faradaic injection, electron transfer occurs across the electrode-tissueinterface and chemical species are oxidized or reduced. Faradaic charge injectionmay be irreversible or reversible. Irreversible injection results in new chemicalspecies being introduced into tissue, with the potential for toxicity. Italso leads to electrode corrosion. Reversible faradaic injection relies oncharge-balanced current pulses to minimize or eliminate this problem. Completelyreversible faradaic injection is, for several reasons, not obtainable. Hence,some degree of potentially destructive chemical reactions will occur. Thegoal is to minimize these reactions. Judicious selection of electrode materials,electrode geometry, and stimulus parameters can reduce the amount of chargethat is injected into the tissue.

Experiments to determine safe charge-injection limits for a varietyof materials have been performed in neural tissue. For platinum, a limit of10⁻⁴C/cm² has been proposed.⁷⁰ Oxidizediridium has a significantly higher limit (1-2 × 10⁻³C/cm²).^61,71 The most recentevidence, however, is that the product of charge density and charge per phasedetermines the safe charge-injection level.⁷² Safecharge-injection limit determinations have been made primarily in brain tissue,and although it seems reasonable to use them as estimates for the retina,actual charge-density safe limits for the retina are unknown. In addition,the stimulus paradigms used to determine safe limits (typically experimentswere performed across several hours on 1 day) do not closely mimic what theretina would be exposed to with long-term stimulation by a prosthesis. Weilandet al⁷³ recently reported results of long-termepiretinal stimulation in dogs. Using 90 or 180 µA, 1-ms-biphasic pulses,they stimulated the retina through an array of 16 electrodes, each 500 µmin diameter. The stimulation was done for 10 to 12 hours per day for up to60 days. Seven of the dogs had normal retinas, and 2 had retinal degeneration.Clinical examination, fluorescein angiography, and histologic study resultsof some eyes showed no damage attributable to the electrical stimulation.As noted in the "Stimulation Thresholds" section, acute experiments in humanswith RP show thresholds between 2.8 × 10⁻³to 2.8 ×10⁻⁴C/cm² with an epiretinal microelectrode arraydirectly on the retinal surface.³⁷

In summary, with present materials and stimulus paradigms, long-termepiretinal stimulation requires charge levels close to or higher than theestablished safe limits for neural tissue. Further experiments such as thosereported by Weiland et al⁷³ are needed to determinesafe levels that are specific to the retina, particularly with conditionsthat more closely resemble what the retina would be exposed to with an implantedprosthesis. We do not know whether safe charge levels for subretinal stimulationwould differ from those for epiretinal stimulation. Thresholds for subretinalstimulation in humans with RP have yet to be determined.

A prosthetic device that consumes electrical power will create heat.The creation of heat will elevate temperature proportional to the amount ofheat created per unit of time and the ability of the local structures to conductaway the heat. In most prosthesis designs, the major areas of power consumptionare in the microelectronic chip (if any) and at the electrode-tissue interface.

Liu et al⁷⁴ developed a computationalmodel to assess the heating effects of prototype prostheses and concludedthat the power dissipated by the implanted chip (vs at the electrode-tissueinterface) would have the greatest effect on temperature increases in theeye. These results suggest that a microelectronic chip may have to be locatedaway from the retina to avoid damage by heating. Locating such a chip in amodified intraocular lens or platform is one possible design, as mentionedpreviously. Caulfield et al⁶⁴ suggest thatmost power will be dissipated at the electrode-tissue interface, creatingthe most heat at this site. According to this model, there would be less advantagein avoiding damage from heat by locating the microelectronics away from theretina.

The degree of temperature elevation that the retina can tolerate overthe long-term, however, is unknown. Piyathaisere et al⁷⁵ examinedthe histologic features after heating dog retina. With the heat probe on theretinal surface, dissipation of power of 50 mW or greater for 1 second damagedthe retina acutely. If the same parameters were used but the animal was allowed4 weeks to recover, damage was not found by standard histochemical analysis.Twenty milliwatts for 1 second in this location did not cause changes, whereas100 mW showed damage immediately and at 4 weeks. Placing the probe in themidvitreous did not produce damage even at power levels of 500 mW for 2 hours.Accordingly, the safe limit for power consumption at the electrode-retinainterface may be between 20 and 50 mW. It is therefore possible that the numberof electrodes that can be stimulated at one time may be severely limited bythe potential of damage from heating.

In summary, heat can damage the retina. The heat produced by a prosthesisdepends on its power consumption, which in turn depends on the stimulus currentneeded for each electrode, the number of electrodes, and the operation ofmicroelectronics that drive the electrodes. Compromises may have to be madebetween the desire for resolution with as many electrodes as possible andcomplex microelectronics vs the need for safety.

Just as implant materials may harm the eye, the saline environment ofthe eye may lead to corrosion of implant materials. A surface coating of anencapsulant that is resistant to the saline environment will therefore berequired.⁷⁶ Encapsulant materials are hydrophobic,preventing entry of water molecules.⁷⁷ Thepresence of even minute amounts of water inside a microelectronic device willeventually lead to failure.⁷⁸ Biocompatibility,on the other hand, generally requires hydrophilic materials because proteinswill denature on contact with hydrophobic materials. Implants thus will probablyrequire a hydrophilic, biocompatible coating over a hydrophobic encapsulant.

Hammerle et al⁷⁹ studied the stabilityof microphotodiode arrays based on silicon-oxide substrates. Devices placedin phosphate-buffered saline for up to 21 months were undamaged when examinedby scanning electron microscopy and energy dispersive x-ray analysis. Whenthe same type of device was implanted in the subretinal space in animal eyes,the silicon-oxide layer was completely dissolved within 6 to 12 months. Thisstudy highlights the need for encapsulation with suitable materials. It alsoshows that in vivo testing of materials is essential. Encapsulation of intraoculardevices for long-term implantation remains a major challenge.

The first retinal implants have appropriately been placed in subjectswith very advanced RP and very poor vision. These early implants are probablynot capable of very high resolution. It is thus not possible to simply testpreimplant and postimplant Snellen acuity to assess the results of implantation.In addition, there is now considerable experience from clinical trials thatindicates the significant subjectivity of measurements of visual function.Subjects in trials therefore routinely have baseline and periodic postinterventionvisual function measurements performed in a standardized manner, after a protocol-drivenrefraction, by a masked examiner. While the specific methods used in mostophthalmic clinical trials are not appropriate for subjects undergoing implantation,the general principles should be followed. This important experience fromprevious trials has not yet been applied to the testing of subjects undergoingimplantation.

Humayun et al³¹ presented brief findingsfrom a single subject implanted with a chronic, epiretinal device (SecondSight, Valencia, Calif) with 16 electrodes. Preoperatively, the subject hadno light perception vision from retinitis pigmentosa. Postoperatively, theauthors demonstrated that the subject could perceive light in response toelectrode stimulation. The subject perceived individual spots in responseto stimulation of each of 16 electrodes. The size and brightness of the spotsvaried with electrode position and stimulus current. In general, the locationof the perceived spot corresponded to the retinal position of the stimulatingelectrode. With a camera connected to the implant, the subject could detectroom light and locate a flashlight. The subject could perceive motion. Formperception was not reported. Chow et al^29,30 recentlyreported results from chronic implantation of a 2-mm-diameter subretinal arrayof microphotodiodes (Optobionics Company, Chicago, Ill) in the superotemporalretina of 6 subjects with advanced RP. The results were given mainly via videotapedinterviews of the subjects by Alan Chow, MD. Some visual acuity and fielddata were reported. Subjects reported increases in vision, central visualfields, and color vision that did not correspond to the retinal area overthe implanted device.

Many groups have contributed to the effort to produce a retinal electronicprosthesis during the last decade. We now know that at least some perceptioncan be obtained in individuals with nearly blind eyes with RP by epiretinalelectrical stimulation. The quality of the perception that can ultimatelybe obtained is unknown. Small devices can probably safely be placed on theretinal surface for extended periods. Heat-producing components will haveto consume very little power or be located away from the retina. The optimalarrangement, fixation, and connection of components of a prosthesis have yetto be determined. Sufficient power can probably be delivered to an intraoculardevice by appropriately placed induction coils or by a laser. Subretinal devicescan be placed and remain in a stable location. The long-term biocompatibilityof subretinal devices must be demonstrated, but there is promise that improvementscan be made in this area. Encapsulation of a device to protect it from thesaline environment of the eye remains a challenge. Stimulation thresholdsmay be sufficiently high that new electrode materials and/or configurationswill have to be developed.

To date, psychophysical evaluation of subjects preimplantation and postimplantationhas not been sufficiently quantitative or objective. We would encourage thedevelopment of a standard, psychophysically rigorous protocol for the preoperativeand postoperative testing of subjects who have retinal prostheses. This willallow more reliable presentation and interpretation of data by the ophthalmiccommunity.

We can thus tell our patients with outer retinal degenerations thatthere is progress toward an electronic retinal prosthesis but fully functional,long-lasting devices are not on the immediate horizon.

Corresponding author: John I. Loewenstein, MD, Retina Service, MassachusettsEye and Ear Infirmary, 243 Charles St, Boston, MA 02114 (e-mail: john_loewenstein@meei.harvard.edu).

Submitted for publication July 2, 2002; final revision received February18, 2003; accepted August 18, 2003.