Electrical activation of voltage sensitive channels is controlled by the voltage difference across the cell membrane, which drives ion currents. Sodium transient channel opening is activated by a membrane depolarization, thus by an inward current through the neural membrane. Other voltage sensitive channels have different responses to voltage and are more or less specific to other ions. The seven nanometer-thin lipid cell membranes form a relatively large electrical capacitance, requiring current to flow for some time before the activation threshold voltage can be reached. This explains the main part of the strength-duration relationship between activation threshold and stimulation pulse duration (Bostock, 1983). Assuming that the membrane conductivity is negligible compared to the extracellular space, Kirchoff's laws indicate that the current through the membrane is roughly proportional to the Laplacian (second spatial derivative) of the voltage along the membrane and a constant resistive factor. In peripheral nerves, the differential equations can be approximated by difference equations spanning over anatomically-defined distances determined by the nodes of Ranvier. These are considered as the only active spots along myelinated axons (Stephanova and Bostock, 1995). More complex volume conductor models have been developed for complex dendritic trees of the neurons in the central nervous system (Ranck, 1975). Not all identical neural structures are activated simultaneously but their recruitment depends on the anatomical position of the stimulation electrodes, the electrical properties (conductivity) of the intervening tissue and the anatomy of the neural structure itself. Variation of the waveform's shape can be used to improve the selectivity of stimulation in the peripheral nervous system by recruiting different groups of fibers. Readers are referred to the recent review of Grill et al. (2009) for update on the topic. Recent modeling studies point out some possibilities in the central nervous system in terms of neuromodulation by AC currents (Mahmud and Vassanelli, 2016).

Magnetic stimulation can be compared to electrical stimulation in the sense that the extracellular current generated by a pair of electrodes is now replaced by an induced current by a varying magnetic flux (Silva et al., 2007). In both cases, the extracellular current generates an extracellular potential field which induces the trans-membranous current. When that current has loaded the membrane capacitance up to a threshold potential, the voltage sensitive channels set up an all-or-nothing action potential.

Variables determining DBS outcome are the stimulation parameters and the positioning of the electrode. The stimulation parameters include the amplitude (current of voltage), frequency, and pulse width. These parameters play a role defining the volume of tissue activated (VTA) and in the therapeutic effectiveness of DBS (Butson et al., 2007). Typical clinically effective parameters for monopolar DBS are presented in Table 1.

VTA in human subjects has been estimated in computational models (Maks et al., 2009). The more complete versions use patient specific imaging data that enrich the models with anatomical information. They also incorporate electrical characteristics, such as impedance and anisotropy, of the different brain parts surrounding the electrode. The study by Maks et al. (2009), found volumes varying from 30 to 116 mm³ with an average of 71 mm³ (i.e., 5 mm diameter). The results from this study further suggest that smaller stimulation volumes in the subthalamic nucleus (STN) are better at alleviating Parkinson's symptoms. The model was experimentally verified indirectly by estimating the stimulation volume spreading into the capsula interna for different voltages and comparing this to muscle twitching of the arm or leg by patient stimulation of the corticospinal tract (Butson et al., 2007). Other studies from this laboratory measured the voltage spread in the brain of a rhesus monkey and compared it to the one predicted by the model (Miocinovic et al., 2009). The study found that the model accurately predicted the voltage distribution observed in vivo.

Within the tissue, photons interact with biological matter via various processes, which can be broadly categorized into scattering and absorption. An accessible overview on current photonics medical applications is given in Yun and Kwok (2017). Light works in a different way upstream of the activation of the membrane ion channels. Here, it is the optical properties of light across the tissues that will determine the threshold and target selectivity. In the case of the well-studied ion channel ChR2, the blue light will cause all-trans-retinal to isomerize to 13-cis-retinal, and so a conformational change in the protein will open the ChR2 channel to allow cations to enter and depolarize the cell. Switching the light off again will cause the 13-cis-retinal to revert to its original state closing the channel and thereby repolarizing the cell. These ion channels do not close spontaneously (as the voltage sensitive sodium channel does) and the resulting response dynamics is in the first place dependent on the opto-chemical properties and density of the photosensitive channels. The strength-duration relationship or the significance of the stimulus duration is now entirely different despite still linked to electrical charges on the membrane capacitance. When open, these charges depolarize the membrane by providing a leakage current. Only then the membrane capacitance starts loading and the density of the available channels is thus much more a limiting factor. The strength-duration relationship still corresponds to a membrane capacitance being loaded, but no longer through the same current.

It should be noted that even if the stimulus does not produce an action potential, a membrane potential shift occurs, such that very long stimulus pulses will modulate spontaneous activity. New phenomena can thus be expected to occur (Grossman et al., 2011). For example, on striatal brain slices, dopamine release (which is an effect pursued in treating Parkinson's disease) appears to be stable when electrically stimulated while a rundown effect is observed with optical stimulation (O'Neill et al., 2017). Once the membrane depolarization voltage has reached the threshold value, however, the same voltage-controlled mechanism as described for electrical stimulation will launch a propagated action potential. Activation selectivity still depends on the geometry and intervening tissues. However, whereas electrical stimulation is controlled by the electrode position, the membrane ion channel distribution and roughly constant resistivity, here, source orientation, source optic features (color, polarization), optic characteristics of intervening tissues, channel density and opto-sensitivity, are the key parameters defining the threshold, selectivity and recruitment.

Selectivity also arises from the use of cell-specific promoters, which drives the expression of the protein in the cells. The opsin is placed downstream of a strong promoter such as synapsin, CMV or CAG. This will lead to strong expression of the opsin in almost all of the cells where the construct is present. Alternatively, one can opt to target cell-specific promoters such as alpha-calcium/calmodulin-dependent kinase II (α-CamKII), which is expressed in forebrain pyramidal neurons. Where electrical stimuli can in some conditions block the propagation of action potentials, similar effects are now obtained with infra-red pulses (Walsh et al., 2016). Some opto-sensitive channels can work as cell silencers, which is not the same as the selective activation of inhibitory pathways, which both techniques could in principle achieve (Malyshev et al., 2017). Through color selectivity, a single optic device could now selectively inhibit or activate the same cells.

Geometric or anatomical selectivity is in principle available to optic stimulation as well as to electrical micro-electrodes (McCreery et al., 2006). Activating light oriented toward a specific points is a realistic goal in examples, such as the retinal prosthesis (Soltan et al., 2017). Optical techniques are being developed that could achieve similar or better selectivity and parallelism than the electrical equivalent (McAlinden et al., 2015; Conti et al., 2016). It should be emphasized that the placement of electrodes is entirely dependent on the surgery. Electrode placement is also a selectivity factor in the frame of optogenetics but selective transfection is another possibility so that only the targets are made photo-sensitive. Selective activation of specific cells, as can be achieved in the frame of optogenetics, could become a key to a successful therapy (Yekhlef et al., 2017). However, the impact on the therapeutic effect cannot yet be established on the basis of the rather preliminary information available today.

The architecture of the brain nuclei poses a problem for purely electrical stimulation because electrodes are relatively indiscriminate with regards to the underlying physiology of the neurons that they stimulate. Typically, physical proximity of the stimulating electrode contact to the neuron is often the determining factor as to which neurons will be stimulated (Deisseroth et al., 2014). Accordingly, it is not considered feasible to restrict stimulation to a single class of neurons with electrical stimulation while optogenetics promises to do just that and even to activate specific channels within the same neurons.

In optogenetics, the volume of tissue that gets activated is also determined by the interaction of several processes. Neuronal activation depends on the absolute amount of light that reaches the neurons, the efficiency of the transfection and the sensitivity of the opsin. Besides these, neuronal physiological properties and the network in which the neuron is embedded also influence how effectively light can control the activity. The propagation of light in the brain is determined by the light absorption and scattering (Vo-Dinh, 2003). How far light reaches inside the brain can be estimated using Monte Carlo simulations. Different studies have calculated that the light intensity drops to 1% at a distance 1 mm away from the emission point (Bernstein et al., 2008; Chow et al., 2010; Stujenske et al., 2015). Additionally, these simulations show that the light distribution inside the brain has an ellipsoidal shape depending on the wavelength, the size and type of light source and the emission aperture angle (i.e., numerical aperture). The simulations also show that a portion of the light back scatters and illuminates the tissue behind the optical fiber. This effect gets exacerbated at higher emission powers to the point that, in some cases, the tip of the fiber is at the center of the illuminated volume. Regarding the sensitivity, studies in acute slices and in vitro show that maximal channel activation for different channelrhodopsin 2 (ChR2) variants is achieved at a power density of 10 mW/mm², with only 50% of the channels getting activated at 1 mW/mm² (Wang et al., 2007; Lin et al., 2009). If an optgenetic DBS would deliver light through an optical fiber with a core diameter of 200 μm and a numerical aperture of 0.22, the data obtained by (Stujenske et al., 2015) can be used to estimate the power required to achieve an activation volume similar to electrical DBS (i.e., 5 mm sphere). For maximal neuronal acativation, a total of 1.5 W would be required to obtain a power density larger than 10 mW/mm² within the illumination volume. Assuming that the tip is indeed at the center. The required power gets exponentially higher if it is not. If the criteria are relaxed and only a power density of 1 mW/mm² is provided at the end of the volume only 150 mW should be delivered. The first value is prohibitively high and would probably create some direct damage in the brain. The second is also quite high, although the brain might be able to withstand it. However, a portable device that has to deliver that amount of power would probably not be practical. Tapered fibers have been tested recently by Pisanello et al. (2017) who showed that varying tapering angle allows for obtaining selectivity of stimulation in vertical direction. Authors claim that achieving uniform effective illumination of large brain structures with minimal invasiveness and light power. In addition, tapering permits smooth insertion into the brain, which also minimizes tissue reaction.

So-stated arguments indicate that stimulation paradigm with ChR2 analogous to DBS might not be the best approach with common ChR2 variants. An alternative could be to use recent variants like ChR2-XXL, which is 10,000 times more sensitive with the disadvantage that it is slower (Dawydow et al., 2014). However, this approach does not take advantage of the capabilities of the optogenetical tools for specific cell type stimulation. Selectivity of stimulation has been tested in rodent models with some positive results (Gradinaru et al., 2009). As an alternative, the inhibition modality of optogenetics could be used. A widely accepted hypothesis of DBS for Parkinson's disease is that excitatory neurons in the STN are inhibited (Shin et al., 2007; Sutton et al., 2013). This hypothesis has been tested by two different groups in rodent models of Parksinosn's disease (Gradinaru et al., 2009; Yoon et al., 2014, 2016). Unfortunately, the results obtained by each group contradict the other, although Yoon et al. (2014) attribute the difference to the animal model and the test used to evaluate Parksinosn's disease improvement by the other group. Tables 2, 3 show the efficacy of applying optogenetic stimulation or inhibition to rodent models in different studies. The tables have a line-to-line correspondence to facilitate comparison. Although, there is yet no definitive conclusion regarding the use of optogenetics of the treatment of Parkinson's disease, these studies demonstrate that optogenetics would be an indispensable tool to better understand mechanisms of Parksinosn's disease and DBS.

We have seen that electrical currents and light do not necessarily activate the same structures. In addition, photosensitive channels have been developed for specific ions and superposition of different light wavelengths allows for activation/inhibition combinations that are not possible with electrical stimuli. However, once action potentials are launched, they can no longer be distinguished on the basis of the triggering event. Parameters, such as pulse frequency, train duration, train rate, treatment session duration, duration of the therapy, event-triggering of the therapy and so on are expected to yield the same effects if the same structures were activated and, this is where a serious word of caution is warranted. In addition, the neural system is an adaptive network. Plasticity, as a general phenomenon can completely modify long-term effects.

Long-term consideration is thus not only important to check the lifetime of implanted device or to evaluate unwanted side effects but also to ensure effectiveness of the treatment strategy (Jarvis and Schultz, 2015). As “accelerated aging” methods do not work here, chronic studies and high sensitivity detection seem to be the only alternative.

Gene transfer experiments in humans started in 1990s but were hampered by initial failures and serious adverse effects, which occurred in the early trials. By present, viral vector technology advanced substantially because of the substantial improvement of our understanding of how different viruses interact with the organism and how their genomes can be designed to reduce biohazard and improve efficiency.

Historically, the first applications of gene therapy used adenoviruses and retroviruses. An adenovirus is a non-enveloped particle of size ranging between 70 and 100 nm. Adenoviruses (AV) have been isolated from a large number of species and tissue types, and in humans, they cause mild respiratory illnesses and gastroenteritis. The virus genome consists of linear double stranded DNA of approximately 36 kbp. An AV particle enters a cell via receptor-mediated endocytosis. The virus escapes the endosome and translocates to the nucleus where it delivers its DNA, which stays episomal. AV can efficiently infect a wide variety of cell types independent of the phase of the cell cycle (Harui et al., 1999).

While adenovirus vectors are very efficient at delivering genes, adenovirus vectors can cause toxic effects that limit their efficiency and safety. AV are highly immunogenic. Upon contact with the virus, the human immune system mounts a full-scale assault, including CD4+ T-helper cells, CD8+ cytotoxic-T cells, and NK cells, in order to clear the virus (Xu et al., 2010). Intravenous AV vector delivery for gene transfer purposes, especially at high doses, stimulates strong innate and adaptive immune responses and can be fatal for the host as notoriously found out in the first clinical trials. Further deletions of the adenoviral genes have yielded helper-dependent adenovirus (HD-Ads) or “gut-less” vectors (Kochanek et al., 2001), which are completely devoid of viral protein coding sequences. This has decreased immunogenicity and also prolonged transgene expression.

In systemic application AV are sequestrated by the liver Kupffer cells and cause toxic effects there. AV vectors also activate the complement system (Manickan et al., 2006; Tian et al., 2009). In addition, high doses of adenovirus vectors rapidly induce a burst of platelet activating factor in vivo that can lead to shock.

From this overview it can be concluded that AV vectors are sub-optimal for gene delivery in the brain due to their immunogenicity (i.e., possibility ot trigger gliosis and neurodegeneration) and low specificity.

AAV forms small, non-enveloped virions with icosahedral symmetry. AAV belongs to the Parvoviridae family and has single-stranded DNA (see Table 5). Multiple serotypes of AAV have been described (AAV1–AAV9) with distinct tissue tropism and transduction efficiency (see Table 4). AV can not replicate by itself and is not related to any known human pathology. AAV can replicate only in co-infection with helper viruses, such as AV, HSV or papiloma virus (HPV) (Weitzman and Linden, 2011). For example, the adenovirus acts as a helper virus by supplying the E1a, E1b, E2a, E4orf6 and viral-associated RNA genes.

The AAV genome consists of two 145 base-pair inverted terminal repeats (ITR), rep and cap genes. The ITRs form loop structures and are the only cis-acting elements that are necessary for genome replication, integration and packaging in the capsid (Kantor et al., 2014). The wild type AAV genome encodes four rep proteins—Rep 78, 68, 52, and 40; three viral structure proteins forming the capsid—VP1, 2 and 3; assembly activating protein, which is involved in the translocation to the nucleolus—AAP (review in Smith, 2008).

The capsid determines the tissue specificity or tropism of a given virus by regulating the immediate cellular response to the virus, mediating pathways for internalization into the cell, and functions in the uncoating process within the nucleus. Specific regions of the capsid proteins interact with receptors and co-receptors on the host cellular surface to mediate the viral infection process and serotypes can differ with respect to the receptors that they bind to. AAV infects a host cell through receptor-mediated endocytosis.

Approximately 80% of the population is seropositive for anti-AAV antibodies (review in Weitzman and Linden, 2011). Also approximately 60% of the population has neutralizing antibodies at age 10, which persists into adulthood. Nevertheless, surprisingly little is known about the life cycle of AAV in humans.

The AAV vectors can stably transfect tissues in different species, including humans. Reports demonstrate durations of such tranfections of at least 6 years in primates (Rivera et al., 2005), 8 years in dogs (Niemeyer et al., 2009) and more that 10 years in the human brain (Leone et al., 2012).

AAV is unique among mammalian viruses in that it integrates into a distinct region of the human chromosome 19, the so called AAV integration site AAVS1 (review in Smith, 2008). The site-specific integration of AAV requires the presence of two viral elements: the inverted terminal repeats (ITRs) and nonstructural proteins Rep78/68. Accordingly, all current AAV vectors lacking the rep gene lack the capacity for site-specific integration. Wild-type AAV sequences are integrated into the host cell genome as tandem, head-to-tail repeats linked to genomic DNA sequences by the viral inverted terminal repeat elements (or ITRs). Specificity is achieved through the interaction of a glycine-rich loop that binds the major groove and an α-helix that interacts with a downstream minor groove on the same face of the DNA. In contrast, the integration of recombinant AAV genomic sequences in the absence of the AAV Rep proteins is inefficient and is not limited to chromosome 19. It is estimated that only about 10% of recombinant AAV sequences integrate into the host cell genome, thus indicating that the majority of vector genomes persist in an extrachromosomal form in vivo (Smith, 2008).

HSV is a member of the Herpesviridae family. There are two main viral types—HSV-1 and HSV-2. While HSV-1 causes orolabial lesions and resides in the trigeminal ganglion, HSV-2 causes genital lesions and resides in the sacral ganglia. About 40% of the adult population is seropositive for HSV-1 in the developed countries (review in Kantor et al., 2014). The HSV particles have icosahedral capsid covered by a lipid bilayer envelope. The envelope includes glycoproteins, which are essential for the viral entry into the cell. The viral genome consists of dsDNA of 152 kb (see Table 5). The life cycle of HSV is particular: the cycle has two alternative phases—lytic and latent. The lytic pathway leads to viral proliferation, emission and imminent cell death, while the latent pathway causes the virus to form an episomal particle within the nucleus in a dormant state. The latent viral particle is capable of lytic transformation upon the action of physical factors (i.e., cold, heat shock etc.). Viral replication is a multistage process, controlled by many genes, which can be removed from the genetic backbone of the vector. This ensures large packaging capacity of up to 125 kbp. What makes HSV suitable for CNS and PNS applications are the following properties: (i) the wild type virus propagates trans-synaptically in both anterograde and retrograde directions; (ii) the wild type virus is neurotropic and (iii) stability of the latent phase.

Infection with HSV typically occurs at the cutaneous or mucosal epithelium where replication of the virus is initially lytic. The virus can also invade axons of sensory neurons in the affected area and undergo retrograde transport to the dorsal root ganglia where it can switch to a latent phase.

Once the virus has reached the nucleus of a cell, the linear genome circularizes and is maintained as an episome with minimal integration (Lentz et al., 2012). HSV vectors demonstrate extensive host cell range, high efficiency gene transfer, and enhanced safety, as persistence of the genome as an episome decreases the likelihood of insertional mutagenesis (Shayakhmetov et al., 2010). Due to cytotoxicity associated with viral gene expression non-replicating and amplicon vectors have been developed. HSV amplicons are eukaryotic expression vectors that harbor the HSV origin of replication and cleavage/packaging signals. HSVs do not require pseudotyping to increase neurotropism, as these viruses are inherently neurotropic. HSV-mediated gene expression has a rapid onset (< 1 day) (Penrod et al., 2015). Transgene expression mediated by HSV vectors has been demonstrated to persist for up to 7 months in the rat brain, but may not be stable (Zhang et al., 2000; Sun et al., 2003).

Replication-attenuated and replication-deficient HSV vectors have the transgene of interest inserted into the viral genome, with targeted deletion of specific immediate early (IE) genes to disable the lytic cycle and render these vectors non-toxic (Wu et al., 1996; Krisky et al., 1998).

Lentivirial vectors are derived from the HIV genome and there is abundant literature about their properties and safety. Interested readers can consult the recent review of Kantor et al. (2014). Lentriviral vectors lead to permanent genetic modification, which may not be desirable in all cases. Since neurons in the majority of brain areas do not divide, there is a little justification, from medical ethics perspective, why a permanent genetic modification of the patient's brain tissue is desirable.

The CNS has proven quite permissive to viral vector gene transfer and expression for many of the conventional delivery vectors. AAV, lentiviral, and to a lesser extent HSV vectors, are the most frequently utilized agents for brain and spinal cord gene delivery. At present, AAV vectors are the leading platform for gene delivery in CNS. Beyond the large number of preclinical and basic mammalian studies involving AAV delivery to the brain and/or spinal cord, the large majority (approximately 75%) of clinical trials that have been initiated for CNS gene therapy have utilized AAV, as contrasted to trials utilizing adenovirus (6%), lentivirus (12%) or retrovirus (6%) (Gray et al., 2010a).

In a recent clinical trial, 12 patients with advanced Parkinson's disease had an application of an AAV vector carrying a transgene encoding glutamic acid decarboxylase (GAD) (Kaplitt et al., 2007). The therapy was well tolerated, with no adverse effects attributable to gene therapy noted for any of the patients. Observed improvement in motor activity lasted for at least 1 year. The trial using AAV-GAD did not establish adverse effects for up to 12 months (Feigin et al., 2007). Another trial for Parkinson's disease using AAV-hAADC vector for dopamine replacement established favorable safety profile (n = 5 patients) but low efficacy (Christine et al., 2009). Similar safety profiles have been established in another trial (Muramatsu, 2010). A phase 2 study using 45 patients and AAV-GAD did not establish serious adverse affects attributed to the treatment (LeWitt et al., 2011). So-reported results demonstrate that AAV is a sufficiently safe vector for applications in the brain. In summary, the recombinant AAV vectors have emerged as a viable delivery method for human gene therapy as they can be designed to meet the precise treatment needs of a given disease by delivering a gene to specific cell types within the affected tissues with a minimal immune response.

Genetic modification has some inherent risks. A comprehensive treatment of the principles of risk assessment will require a dedicated publication. Interested readers can consult Baldo et al. (2013) for treatment of gene therapy cases. Briefly, the methodology of risk assessment consists of the following steps:

Considering viral vector applications the following principal hazards can be identified:

Various candidate viral vectors are compared in Table 5, based on the studies of Doherty et al. (2011) and Howarth et al. (2010).

Transfection-related properties of different vectors are summarized in Table 6.

We can identify some desirable properties of the ideal vector, considering the specific application:

The payload capacity is not a differentiating factor for optogenetics applications as the channelrhodopsins are in the range 1.7 kbp (i.e., hChR2-GFP). Based on these criteria, a preference table can be assembled (Table 7). From the table it appears that, given the present state of development of the viral vector technology, the best choice for optogenetic application in human is AAV followed by HSV.

Non-viral alternatives for gene delivery have been developed in parallel to viral vector technology. These are for example, transfection by electroporation (Nomura et al., 2016) or focused ultrasounds (Wang et al., 2017). Another potential alternative is laser photoporation, currently developed in vitro (Antkowiak et al., 2013). For various reasons however, it is the viral vector-based gene delivery that stands out as the most mature transfection technology.

Micro and perhaps nanotechnologies are at stake here when considering the development of the appropriate stimulation system (Pisanello et al., 2016). Multi-functional devices can combine photo-stimulation and electrical recording (Wang et al., 2012) including micro-electrode arrays placed in the vicinity of the target (Buzsaki et al., 2015; Naughton et al., 2016).

Designing probes that can be safely inserted in the brain is a significant challenge. Generic tissue reactions to the implant device have been discussed elsewhere (Prodanov and Delbeke, 2016a). For any implanted device, power dissipation and thus local temperature must be kept at an acceptable level (Arias-Gil et al., 2016; Shin et al., 2016). The strong light power used in photogenetic could transform photochemically some cell molecules into toxic agents. Chronic aspects are important to consider: a good example is the mitochondria-mediated apoptosis induced by prolonged ChR2 activation (Perny et al., 2016). Whether electric, microfluidic or optical, any application must take into account the diffusion and depth of penetration. Although only mentioned in animals, light leaks might influence an implant (Eckmier et al., 2016), considering that near infra-red spectroscopy is a technique based on light transmission through scalp, skull and brain (Benaron et al., 2000). Optic implants are new and require a specific encapsulation still lacking chronic evaluation (Rossi et al., 2015). On the other hand, it is much easier to make optic devices Magnetic Resonance Imaging (MRI) compatible than their electrical counterparts. This is an important factor for the modern clinical practice.

DBS is presently used in Parkinson's disease, epilepsy, pain, dyskinesia, tremor, dystonia, major depression and obsessive compulsive disorder (OCD). The exact mechanism of action is unknown and it is assumed that all neurons within a given volume are being activated. Because the cell type activated can be very different and sometimes non-neuronal (Nam et al., 2016), optical stimulation of the same volume might not perform an equivalent job. Epilepsy is presently the object of many attempts to apply neuro-modulation. This is justified by the large number of therapy resistant cases. However the pathophysiology of this condition remains obscure. The selectivity of optogenic stimulation might help sort out the pathophysiology and arrive at an efficient clinical treatment (Xu et al., 2016; Yekhlef et al., 2017). Also for conditions, such as retinal diseases, light stimulation would probably make the generation of meaningful visual perceptions easier in a retinal prosthesis (Nirenberg and Pandarinath, 2012; Al Atabany et al., 2013). Stimulation and sensing can be implemented exclusively with optical elements (Inagaki et al., 2017). A combination of techniques, such as electrophysiology and optogenetics (Chen et al., 2017) or micro-fluidics (Rubehn et al., 2013; McCall et al., 2017), can offer additional perspectives in feedback controlled systems.

In a standard optogenetics protocol the channelrhodopsin gene is delivered by viral vector injection. However, as already mentioned, innate immunity and antigen-specific adaptive immune responses against vector-derived antigens could reduce the efficacy and stability of the gene transfer when this technique is translated to the clinic (for an overview, see Bessis et al., 2004; Nayak and Herzog, 2009). In the case of human gene therapy, AAV vectors are frequently put forward as optimal solutions due to their good safety profiles (Carter, 2005). Innate immune response are limited, but have been observed at high doses or with specific serotypes (Lowenstein et al., 2007; Hadaczek et al., 2009). Adaptive immune responses, on the other hand, are more common. Anti-AAV antibodies are predominantly directed at the vector's capsid protein (Mingozzi and High, 2013). Neutralizing antibodies constitute one of the main challenges of successful gene delivery, as they impact transfection efficiency in animal (Arruda et al., 2010; Haurigot et al., 2010; Jiang et al., 2013) and clinical studies (Manno et al., 2006), even at low titers (Scallan et al., 2006). Especially persons with pre-existing anti-AAV antibodies would be at a disadvantage, though there is a difference in neutralizing effect depending on the serotype (Xiao et al., 1999). Other studies in the brain (Lo et al., 1999; Mastakov et al., 2002), muscle (Kay et al., 2000) and retina (Anand et al., 2002) revealed, however, no relation between the presence of the anti-capsid antibodies and the amount of transgene expression.

At present, it is seems too early to choose between systemic vs. local viral vector application. Systemic application of viral vectors could be favorable in the case mechanisms of DBS turn out to be more delocalized. On the other hand, in this case the volume of illumination can turn out to be the performance limiting factor. Localized application by the device itself may pose conflicting engineering requirements (e.g., microfluidics), which could decrease reliability.

Besides vector-specific antibodies, the body might also mount an immune response against the transgene itself. Transgene-specific immune responses are dependent on the target organ, the administration route and dosage (Toromanoff et al., 2010; Mingozzi and High, 2013). Target tissue such as the liver display a high level of tolerance to the transgene product (LoDuca et al., 2009), while muscle (Ross et al., 2006) is more prone to activation of the immune system. In the case of neural tissue, opsins do not seem toxic to human neurons in the brain or retina (Busskamp et al., 2010; Valtcheva et al., 2016). However, high levels of opsin expression might still induce cell death (Klein et al., 2006), alteration in electrical membrane properties or form aggregates in neurons (Gradinaru et al., 2008; Zimmermann and Dours-Zimmermann, 2008; Diester et al., 2011), while no observable damage has been reported (Han et al., 2009). The diversity in responses points to the difficulty of determining the correct levels of expression (Allen et al., 2015; Jarvis and Schultz, 2015). Therefore, to counter potential toxicity, different modification strategies have been developed, leading to safe expression even at high titers (Gradinaru et al., 2008). Finally, as in any gene therapy case, there is a risk that the opsin might insert randomly into the genome and thereby result in oncogenesis (Hacein-bey abina et al., 2008).

The light-delivering device can either be implanted into or remain external to the brain. The necessity of opening a pathway in the first case bears the risk of tissue damage due to the insertion or chronic presence in the body as with any implantable device (for a review, see Prodanov and Delbeke, 2016a). Acute vascular damage can be controlled to some extent by the shape of the implant and the surgical trajectory; while the amount of micromotion is a function of the implant inertial properties and the mode of attachment to the skull. Both parameters are difficult to modify, once an application has been chosen. To prevent mechanical damage, the device should be as small as possible, while providing sufficient light to the target deep brain structures. For this purpose, multiple solutions have been proposed: for example, novel probe designs or adaptations of high-density fiber arrays (Zorzos et al., 2010; Han, 2012; Ozden et al., 2013; Hoffman et al., 2015).

Light absorbed by brain tissue can produce non-specific effects due to heating or photodamage. Heating may alter the activity of neurons or even behavior (Moser and Mathiesen, 1996; Long and Fee, 2008). The risk of overheating constraints the power that can be used by the device and hence limits the illumination volume. A possible solution is to place the device further from the neurons of interest or even outside the brain and use red light in combination with red-shifted opsins (Lin et al., 2013).

Whereas most of the optogenetic protocols do not induce excessive tissue damage (Gysbrechts et al., 2016), heating effects were also observed at standard light powers (Christie et al., 2013; Stujenske et al., 2015) and even in non-transduced cells (Han, 2012). On the other hand, heat itself has been used as well to perform optical stimulation of wild-type cells, using infrared pulses (Shapiro et al., 2012; Carvalho-de Souza et al., 2015). The light might also have non-thermal effects on neural tissue. Light-sensitive pathways naturally present in cells could produce unwanted effects to both targeted and surrounding cells (Koyanagi et al., 2013; Cheng et al., 2016).

Finally, a favorable acute response does not necessarily predict the chronic stimulation effects on neuronal tissue. Changes at single-cell and network level need to be considered when translating optogenetics to the clinic. Long-term exposure to light pulses has been known to plastically modify the behavior of transfected neurons, changing their response to stimulation (Schultheis et al., 2011; Lignani et al., 2013), or induce long-term potentiation (Zhang and Oertner, 2007). Some of these chronic effects may prove to be beneficial in a clinical context, leading for example to a reduction of the stimulation periods. On a single-cell level, changes in neuron morphology have been observed, due to the prolonged expression (Miyashita et al., 2013) or stimulation (Goold and Nicoll, 2010; Grubb and Burrone, 2010).

Four main aspects contributing to compromised long-term compliance can be outlined:

The disruption of BBB leads to the deposition of plasma proteins foreign to the CNS, such as albumin, globulins, fibrin/fibrinogen, thrombin, plasmin, complement (Kozai et al., 2015). The vascular damage is accompanied by fluid displacement, dragging of the blood vessels and eventual hemorrhage observed after implantation (Bjornsson et al., 2006). Hemorrhages have been shown to be particularly detrimental for long term recording (Stensaas and Stensaas, 1976; Turner et al., 1999; Grand et al., 2010). The cerebral vasculature is particularly susceptible to the action of ROS, which is of great importance since cerebral endothelial cells play a major role in the creation and maintenance of BBB (review in Obermeier et al., 2013). Such BBB dysfunction can result in an imbalance of ions, transmitters and metabolic products in the interstitial fluid, causing abnormal neuronal activity. In the implantation setting ROS can be linked to the catalytic function of Fe³⁺ present in the blood clot, which can lead to formation of stationary diffusion-limited concentration gradients around the implant (Prodanov and Delbeke, 2016b).

Refractory epilepsy and Parkinson's disease present different cases for an eventual application of optogenetics. While the Parkinson's disease is an established target for DBS since 1997, long-term efficacy of electrical stimulation for refractory epilepsy is still an ongoing investigation. Despite maximal antiepileptic drug therapy, more than 30% of patients with epilepsy suffer from persistent seizures, while up to 40% of patients are not candidates for surgical resection (Halpern et al., 2008). The SANTE trial³ fond out that the median percent seizure reduction from baseline at 1 year was 41%, and 69% at 5 years. The long-term follow-up of anterior thalamic nucelus (ANT) DBS showed sustained efficacy and safety in a treatment-resistant population (Salanova et al., 2015). The readers are directed to the recent review of Laxpati et al. (2014) for a summary of the clinical trials in epilepsy.

While for Parkinson's disease an optogenetic stimulator must demonstrate superiority to eventually become an accepted therapy, in epilepsy the target can be somehow lowered to non-inferiority. This indicates that optogenetic modulation of refractory epilepsy could in fact be achieved first.

Optogenetics also offers the possibility to work with cells other than neurons (Cho et al., 2016). Systems combining optic simulation and electrical recording have been developed (Wang et al., 2012; Rubehn et al., 2013; Hoffman et al., 2015, 2016; Naughton et al., 2016; Segev et al., 2017). Targets have not been limited to the brain and include the spinal cord and peripheral nerves (VNS for epilepsy and depression, motor nerves for various palsies, hypoglossal nerve for sleep apnoea and many more). For some peripheral nerves, optic stimulation could be realized without implant or transfection (Maimon et al., 2017). Alternatively, non invasive techniques are being developed with Transcranial Direct Current Stimulation (tDCS), transcutaneous VNS (tVNS), Repetitive Transcranial Magnetic Stimulation (rTMS) and others.

Despite its acceptance, there is still much room for progress in the practice of DBS—for example, by providing bi-directional (i.e., including sensing) interfaces, which would open the path for closed-loop approaches and titration of stimulation. Neural tissue does not necessarily provide a stable response to chronic stimuli (McCreery et al., 1997). Poorly controllable brain plasticity can turn out to be an ally or an enemy in chronic applications. Along the same line, cellular biology homeostasis (Sinnen et al., 2017) should be considered. It is known that long term potentiation or depression (LTP, LTD) (Nabavi et al., 2014; Correia et al., 2017) can be induced by some stimulation conditions. Specific strategies might be essential in the control of epileptic activity (Ching and Ritt, 2013). Cell-type and even sub-cellular specificity hold the promise for real progress (Dvorzhak et al., in press) but we lack the knowledge to benefit from such developments. Also stimulus phase timing (Witt et al., 2013) can be important. Appropriate stimulation can shift the excitability of the cortex (Heitmann et al., 2017). Stimulation regimens are important (Shiri et al., 2017) but mostly established on empirical basis because of a lack of neurophysiological and pathophysiological understanding (Gradinaru et al., 2009; Dvorzhak et al., in press). Safe limits for the various stimulation parameters have not been firmly established. Despite the brain being essentially a control system, most rational explanations are still in terms of a simple balance between excitatory and inhibitory influences. Closed loop system are being presented (Grosenick et al., 2015) but they still represent “event-triggered” stimulation rather than network integrated devices.

DBS is a recognized therapy in Parkinson's disease and it has also benefited some cases of major depression. An optogenetics equivalent could do equally well or perhaps even better but there is no clinical evidence yet. Present findings point to a number of pathways along which optogenetics is likely to play a role in the near future, namely in the treatment of epilepsy (Mantoan Ritter et al., 2014). With the assumption that various issues related to genetic manipulations of adult patients can be resolved, optogenetics could soon become an alternative to DBS (Karas et al., 2013) in the treatment of Parkinson's disease (Beitz, 2014) and epilepsy (see for example Boon et al., 2007; Cox et al., 2014). Many more conditions might soon enter the therapeutic perspective including Alzheimer's disease (Kastanenka et al., 2017), the cochlear prosthesis (Richardson et al., 2017), visual pathologies (Scholl et al., 2016; Sengupta et al., 2016) and perhaps cardiac (Entcheva and Bub, 2016) or other non-neural conditions.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The reviewer WW declared a past co-authorship with one of the authors JD to the handling Editor.