Optogenetics 2.0: challenges and solutions towards a quantitative probing of neural circuits

1 Introduction – simulation or physiology?

The implementation of optogenetics now almost two decades ago has revolutionized preclinical neuroscience (Boyden et al. 2005). The optical control of genetically targeted neurons, with millisecond precision, allows identifying the contribution of individual neurons to network function. Optogenetics pioneered the causal interrogation of not only local but also brain wide functional connectivity. Particularly, it opened up the field of optogenetics-guided functional brain mapping, which led to the discovery of previously unknown functional connectivity motifs. It had been employed to precisely map long-range connections from cortical and thalamic regions onto pyramidal neurons within the mouse vibrissal motor cortex (Hooks et al. 2013) and between sensory and motor cortex (Mao et al. 2011), in addition to unravel the intricate inhibitory-to-excitatory wiring pattern in various cortical areas (Kätzel et al. 2011). Particularly the combination of optogenetics and fMRI allowed for the distinction between the functions of neuronal subtypes in the various layers of the motor cortex (M1), where the activation of layer II/III evoked only local responses while the activation of Layers IV/V/VI showed activity in subcortical regions including thalamus (Chan et al. 2022). This combination, termed ofMRI, has allowed not only efficient brain mapping in healthy subjects, but also helped advance the examination of pathological conditions, such as epilepsy and Parkinson’s among others (Lee et al. 2022a).

And while we made tremendous progress on our understanding of the role of neurons and nodes, an important aspect has not been fully addressed and understood: Applying an optogenetic stimulus can lead to a neuronal output which does not necessarily reflect the extent of the activation of the given neuron in an endogenous naturalistic activation. So, the task ahead would be to move from revealing what a neuron can do if maximally stimulated, to emulating the contribution of a neuron to network function in the naturalistic chain of signal dynamics in an endogenous activation. The development and application of step function opsins affording constant subthreshold depolarizing currents provide enhancement of naturalistic synaptic inputs rather than directly inducing them (Berndt et al. 2009; Gong et al. 2020). Similarly, the introduction of ultra-fast opsins aims to mimic temporally naturalistic activations (Chow et al. 2010; Mager et al. 2018). The simulation of endogenous network activation can also be achieved with older, commonly used opsins like Channelrhodopsin-2 (ChR2), simply by adjusting the transfection method and the titration of the light intensity and frequency, as we discuss in the next chapter. By ramping up and down the light intensity and activating only a fraction of the neuronal population at a given time (Yang et al. 2017), a sparse, passion-distributed firing can be achieved, which mimics naturalistic activations (Pillow et al. 2008).

Among these lines, it is critical to understand how the induced expression of light-sensitive proteins can affect the cellular physiological state. Indeed, overexpression of opsins could alter important biophysical and physiological properties and compromise cell function (Lin et al. 2009). For example, the prolonged activation of ChR2 in human melanoma cells has been found to induce mitochondria-mediated apoptosis (Perny et al. 2016). The activation of ChR2 has also been shown to cause transient K⁺ intracellular elevations and affect gene expression (Octeau et al. 2019). Overexpression of opsins can change electrophysiological properties of cells. It has been reported that high expression of ChR2 in mammalian human embryonic kidney 293 (HEK293) cells in vitro induces marked increase of the area-specific membrane capacitance (Cm) (Zimmermann et al. 2008). The change of Cm could be a consequence of an alteration of dielectric properties and membrane morphology.

Another important aspect to keep in mind is that the expression of opsins at the smaller compartments of the cell, such as axons and dendrites, could have different effects compared to the soma due to changes in pH-levels and shifts in the equilibrium potential of the given ion (Mahn et al. 2016; Rost et al. 2022). The presynaptic activation of proton pumps, such as archaerhodopsin, or chloride conducting opsins, such as channelrhodopsins, can cause a paradoxical release of neurotransmitter instead of the desired inhibition. Therefor it is important to combine the expression of light activated channels with fluorescent markers (Tye et al. 2011) to control the location of the expression or use optical imaging, by co-expressing calcium indicators, or electrophysiology to control the activity levels (Backhaus et al. 2023; Kim et al. 2017).

Consequently, the expression of opsins is a key element to consider as overexpression can lead to toxicity and compromised cellular physiology. To address the side effects of expressing opsins in optogenetic experiments, it would be of interest to evaluate toxicity by using accepted criteria or marker of toxicity which may include histological, electrophysiological or molecular profiling measurements (Allen et al. 2015). It is also important to underline that results should be compared to appropriate control cells or animals which are infected with viruses carrying a fluorescent protein with no opsin fused or an opsin with residues mutated to prevent light-activation. Ameliorating cell toxicity associated with opsin expression can be achieved through systematic titration of the response by manipulating the quantity of injected virus. This approach allows for fine-tuning expression levels of opsins, ensuring optimal functionality while minimizing any detrimental effects on the targeted region. Indeed it was shown that a modulation of the amount of virus, or photostimulation intensity trigger different behavioral responses (Lee et al. 2014). These approaches offer means to optimize experimental outcomes and enhance the safety profile of optogenetic intervention.

The level of opsin expression is dependent on gene copy number carried by the cells of interest. Opsin expression can be achieved by transgenic techniques, viral vectors, or electroporation. The choice of viral vector depends highly on the specific questions of the experiments. In neuroscientific research, the most widely used vectors are glycoprotein deleted rabies viruses (RV), adeno-associated viruses (AAV) or lentiviruses (Lentz et al. 2012). RV is a quite powerful tool for synaptic tracing and circuit mapping, although this can be now achieved by trans-synaptic AAV serotypes (Deshpande et al. 2013). AAV-mediated gene transfer has the advantage that the gene of interest will not be integrated into host genome, thereby reducing biosafety levels, but with the disadvantage of not being effective in dividing cells. For dividing cells, such as embryonic stem cells lentiviral approaches are advantageous (Stroh et al. 2011).

However, a stable expression for a long time period spanning several months is a crucial element particularly for longitudinal studies (Packer et al. 2013). The opsin expression via AAV typically takes up to three or four weeks to reach its maximum, while electroporation is significantly faster, increases over days or weeks until it reaches an equilibrium. The expression stability requires a particular consideration because high-level and long-term expression of ChR2-EYFP has been shown to cause axonal and synaptic abnormalities (Miyashita et al. 2013). In addition, the level of opsin expression across cells from the same region is highly heterogeneous because the copy number of gene varies from cell to cell (Schmid et al. 2016). Accordingly, it will be of great interest to find a solution to stabilize the level of opsin expression in the targeted cell type. A stable expression will allow to determine a standard for optogenetic experiments and to compare results obtained in different mice. Unfortunately, the stability of the expression is not only dependent on the opsin variant and viral vector, but also has many complicated intracellular mechanisms involved, most notably gene silencing (Fire 2007), which make this difficult to achieve.

The limitations reached by viral vectors and electroporation can be partly overcome by generation of mouse transgenic line expressing opsin under specific promoter. In this case, the expression is more homogenous across cells and between animals. For example, the Thy1:ChR2-EYFP mouse lines express ChR2 under control of the Thy1 promoter leading to an expression in projection neurons (Wang et al. 2007). It is also worth pointing out that the induction of a cell specific expression is more difficult and the choice of promoter needs to be considered more carefully (Zeng and Madisen 2012). More importantly, particularly the first transgenic mouse lines such as the Thy1-ChR2 line exhibits expression of opsin in various brain regions, such as layer 5 of the cortex, but also thalamic and subthalamic nodes, which renders one of the key advantages of optogenetics, i.e., not only a cell-specific, but also node-specific expression, obsolete.

Optimizing the expression of opsin in defined set of cells can also help to optimize the light delivery. Indeed, the intensity of delivered light is experimentally adjusted to evoke an optical activation of the opsin and this parameter is related to the level of opsin expression. For example, a low opsin expression will require a strong light intensity to activate targeted opsin-expressing cells which may induce light side-effects (Rungta et al. 2017; Schmid et al. 2017). It is also important to note that the light is scattered in the brain tissue and the light intensity reaching the tissue is not homogenous. Consequently, the extent of optical activated cells depends on the light delivery from one hand and expression level in another hand (Figure 1).

Figure 1:

Typical scope of network activation by optic-fiber-based single photon optogenetic stimulation in mouse cortex. Transfected neurons are colored in red and activated neurons are colored in green. Due to scattering and absorption of light, and limited expression of the opsin, only a subset of the local neuronal population will be optogenetically modulated.

Moreover, although some recent opsins, such as Chronos and ChroME have very fast kinetics (2–4 and 5 ms, respectively), several widely used opsins can conduct for long periods depending on their off kinetics after the light pulse stimulation (Chow et al. 2010; Mager et al. 2018). For example, the opsins C1V1 and VChR1 exhibit long off kinetics (156 and 133 ms, respectively) (Rein and Deussing 2012). A prolonged opening of channels is an advantage for strong and efficacy stimulation but passage of ions and membrane depolarization/hyperpolarization for such long periods can evoke non-physiological events. Prolonged activation can be useful in the case of stable step-function opsins (SSFO), where they induce subthreshold depolarization and shift the neurons into a more excitable state (Gong et al. 2020; Yizhar et al. 2011b). Thus, enhancing naturalistic synaptic inputs without the direct elicitation of action potentials. This has a pharmacological effect with the added benefit of targeting specific brain regions and neuronal subtypes.

The optical activation of an opsin expressed by neurons can change various intrinsic neuronal properties. But it is also important to understand how those changes can affect the entire local and global network activity and whether it may introduce artifacts. Cardin et al. (2009) demonstrated that activation of ChR2 in fast-spiking interneurons, but not in pyramidal neurons, amplifies gamma oscillations in barrel cortex. Herman et al. (2014) showed that extreme stimulation of ChR2 leads to a neuronal depolarization block, inducing an arrest of action potential elicitation in regular-spiking interneurons, but not in fast-spiking interneurons or excitatory neurons. Consequently, excessively activated ChR2 silenced the neuronal activity instead of activating it, which can lead to misinterpretations of experimental data. As also demonstrated by Baleisyte et al. (2022), the intense activation of ChR2 in GABA neurons in the medial amygdala in mice led to paradoxically increased aggression and altered social behaviors compared to controls with the kinetically faster channelrhodopsin variant ChETA. As mentioned before, simultaneous electrophysiology or functional calcium imaging could be used as a control to avoid such problems.

A change in the activity of a cell or a subcellular group can impact the activity of downstream neurons and circuits and it is essential to know whether these changes are part of the physiology. For example, the manipulation of the activity of tonically active neurons can compromise the basal activity level of downstream neurons and perturb information processing. In the cerebellum, the tonic inhibition of granule cells is a key element in the regulation of neuronal excitability and network function (Duguid et al. 2012). Indeed, the tonic inhibition regulates the firing rate of granule cells, and it modifies their operational range over a wide range of excitatory drives. This modulation reduces the temporal window for synaptic integration and participates in the ability of granule cells to detect sensory signal from spontaneous network activity. Changes in granule cells activity decrease their ability to discriminate sensory information and reduce sensory transmission to downstream interneurons. Thus, optogenetic stimulation or inhibition of granule cells activity will affect basal and sensory information processing and may create a new encoding pattern of signal which cannot be produced by natural stimulations.

In addition, the optogenetic activation or inhibition of a neuronal subpopulation will lead to the driving or the deprivation of downstream circuits which cannot be modulated in this way under physiological conditions. For example, the activation of excitatory ChR2 expressing neurons in frontal cortex of nonhuman primates induces an increase of firing rate during stimulation in many neurons (called excited units), but they also observed a decrease of firing rate in many other neurons at long latency after light stimulation (called suppressed units) (Han et al. 2009). They found that the suppression of the activity is due to the recruitment of downstream inhibitory interneurons driven by excited units. In the same way, the optical silencing of ArchT-expressing pyramidal neurons resulted in the silencing of most neurons but a small proportion of neurons exhibited an increase activity at long latency after light stimulation (Han et al. 2011). They suggested that the silencing of excitatory neurons inhibits the activity of downstream interneurons which leads to a reduced inhibition on some neurons. These two studies indicate that when a large neuronal population is being driven or deprived, another secondary population reacts and balances the network activity. The understanding of the functioning of neuronal circuits is therefore important when conducting optical modulation of a neuronal subpopulation activity and it is highly relevant to know whether downstream activated circuits can occur endogenously.

It is also important to note that information encoding can be altered by optogenetics stimulations. During information processing, there is a physiological temporal jitter when the synaptic information flow through the different elements of the circuit (Marsalek et al. 1997). This temporal jitter is important for spike-timing precision in response to dynamic sensory stimulation and it plays a role in coincidence detection of spike trains. But the temporal jitter created by optogenetics stimulation could be different from the physiological temporal jitter due to the activation and deactivation kinetics of opsins. Consequently, the signal integration can be different, and it can lead to alternative responses. It is therefore essential to determine whether the temporal jitter induced by light stimulation is similar (or close) to the temporal jitter induced by natural stimulations.

2 Solutions towards a quantitative network probing

To emulate endogenous network function, we need to transcend and surpass the already challenging task of expressing an opsin in a genetically encoded neuronal population. We need to achieve a quantitative, scalable stimulation approach, which can not only stimulate a given neuron with millisecond precision – the claim to fame for optogenetics – but also achieve the modulation of a defined fraction of the local network. While single neuron optogenetic stimulation has been achieved in vivo by the technique of 2-photon all optical physiology, described in the next chapter, on mesoscale level, the scope of network modulation is often ill-defined. First and foremost, besides the promoter, the choice of the appropriate viral serotype is paramount. Here, due to the overwhelming advantages of AAVs for the transfection of postmitotic neurons, we will concentrate on AAV variants (Figure 2) and will furthermore only focus on preclinical application in rodents. The AAV toolbox expanded considerably in the last couple of years. For the local transfection, we will highlight five serotypes: AAV1, AAV2, AAV8, AAV9, and AAV-DJ. All of these serotypes afford the efficient transfection of mammalian neurons (Aschauer et al. 2013; Lentz et al. 2012; Salganik et al. 2015). Here, we highlight the differences in expression density in mouse cerebral cortex. Note, that layer IV is very hard to transfect with AAV-based gene transfer (Yang et al. 2017, 2018). This might be due to the limited accessibility or limited diffusion ability of the viral particles. Indeed, when injecting in supragranular layers or infragranular layers, layer IV almost acts as a barrier, restricting expression to either layers II/III in the case of supragranular injections, or layers V/VI for infragranular injections. Currently, AAV9 will afford the highest density of opsin-expressing neurons (Aschauer et al. 2013). What is more, AAV9 seems to be most restricted in terms of limiting infection only to neurons, with a lesser tropism towards non-neuronal cells (Haggerty et al. 2020). This is of importance when more ubiquitous and therefore often stronger promoters have to be used, which might not be neuron specific. Therefore, tropism can add an additional layer of specificity beyond the chosen promoter. For using optogenetics for circuit mapping, two new AAV variants have significantly extended the optogenetics toolbox: AAV-jump allows for the transfection of postsynaptic neurons, i.e. when injecting AAV-jump in the infragranular cortex, among others, the corticofugal projections of the infragranular layers will result in an opsin expression in the respective thalamic nucleus. Injecting in primary visual cortex will result in expression on the dorsal geniculate nucleus, amongst others (Figure 2). The engineering of AAV-retro opened up the possibility to transfect presynaptic neurons, which had been previously only amenable with rabies virus approach (Lentz et al. 2012). Here, the injection of AAV-retro in the lateral geniculate nucleus of the thalamus (LGN) will result in the expression of opsins in the thalamo-projecting cortical layers, mainly in layers V/VI, among other targets.

Figure 2:

The effectiveness of different AAV serotypes in transfecting neurons in rodent brain. Depending on the promoter of the AAV vector, multiple types of cells can be infected, here we assume a neuron specific promoter and show only pyramidal and polymorphic neurons. Top: using a local injection in the cortex, AAV9 induces the highest expression level followed by AAV8, AAV1 and AAV2. Bottom: The injection of AAV-Retro in the thalamus allows mainly mapping and manipulation of corticothalamic projection neurons (layers V–VI). The opposite, mapping and manipulations of thalamocortical projection neurons mainly of layer IV, can be achieved by injecting AAV-Antero at the same site.

Once a specific AAV is chosen, the quantification of opsin expressing cells can provide a first estimate on the fraction of neurons which could be optogenetically modulated (Packer et al. 2013; Schmid et al. 2016; Yang et al. 2017). Typically, the fraction of neurons modulated ranges from 5 to 30 %. Next, another level of tailoring optogenetics to the respective signal of interest to be emulated represents the light delivery. In the early days of optogenetics, optical fiber where the overwhelming mode of delivering light into tissue (Boyden et al. 2005; Stroh et al. 2013; Zhang et al. 2007). Using multimode optical fibers, the volume of above threshold activation can be calculated by applying either the Kubelka-Munk-model, or a combination with a spherical and Kubelka-Munk model (Schmid et al. 2016, 2017; Yizhar et al. 2011a; Yang et al. 2018). More recently, light delivery methods such as implantable LED on brain surface afford the optogenetic control of larger cortical areas (Lee et al. 2018; Mitsuhiro et al. 2014). Ultimately, taking together these two factors, limited expression, and limited light delivery, only a fraction of neurons will be optogenetically modulated (Yang et al. 2017). Yet, for most cases, a sparse activation mirrors most naturally occurring network activation states in which also only subsets of neurons react to a given stimulus (Dalgleish et al. 2020; Lee et al. 2022b). Finally, methods such as ramping up the light density will avoid a non-physiological hyper-synchronous activation: As the expression levels of the opsin will vary between neurons due to the different AAV quantities entering the cells, so will the threshold in terms of light intensity or either evoke or inhibit an action potential. Modulating light intensity within a light pulse will therefore lead to a gradually, temporally dispersed activation or inhibition in the local neuronal population.

3 Closing the loop with optogenetics

Over the years, the field of neuroscience has evolved from the model that large areas of the brain are responsible for specific cognitive or behavioral functions to the realization that brain functions are much more complex and integrated (Singer 2009). Today, we have a better understanding about the importance of microcircuits and the role of single neurons and their interactions. The processing and encoding of information are highly depended on the dynamics of microcircuits and their activity patterns (Oby et al. 2019), which is defined by the number of neurons in the network and the rate, timing and synchronicity of neuronal action potential firing. For example, one study was able to demonstrate that the primary visual cortex exhibits altered spontaneous and visually evoked activity patterns and attractor dynamics in mouse models of schizophrenia (Hamm et al. 2017). The activity in the primary visual cortex has been also shown to have a larger impact on decision making in a visual discrimination task compared to other cortical areas such as the frontal cortex (Zatka-Haas et al. 2021). To causally unravel such complex interactions, interweaving readout and manipulation of single neurons activity is critical.

One example in which closed loop optogenetic control in real time has already been achieved is epilepsy (Armstrong et al. 2013). Based on EEG signal, it was possible to detect seizures and then activate GABAergic neurons in the hippocampus to stop the seizures. However, for emulating physiological network activation of physiological rather than pathophysiological signals, which is also needed to tackle more complicated disorders of memory or consciousness, a high-resolution read-out of the microcircuit’s activity is required. Currently, the best approach would be to combine functional calcium imaging with optogenetics for a higher number of recorded cells in an all-optical approach. This can be achieved by careful titration of viral vectors carrying the genes for the opsin and the calcium indicator (Fu et al. 2021; Guimarães Backhaus et al. 2021). Optical imaging technics have allowed a significant advancement for examining neural networks with a high spatial and temporal resolution. Using genetically encoded calcium indicators, the activity of neurons can be analyzed with single action potential resolution (Chen et al. 2013; Zhang et al. 2021). This can also be achieved in unrestrained roaming mice with miniaturized microscopes (Miniscopes) (de Groot et al. 2020; Srinivasan et al. 2019; Stamatakis et al. 2018).With more recent advancements in wide-field imaging, the activity of thousands of neurons cortex-wide can be detected simultaneously (Ren and Komiyama 2021).

To study the complex interaction of network dynamics, not only high-resolution imaging is necessary, but also a reliable integration of the involved systems and robust analysis pipelines (Backhaus et al. 2023). Since the interactions in neural networks occur on very short time scales, a fully automated real-time analysis routine is required to transform the acquired imaging series to ideally binarized activity traces of the individual neurons in the field of view and generate a meaningful intervention vector for the optogenetic stimulation (Figure 3). Usually, these analysis steps are referred to as segmentation, source extraction, binarization, and finally, intervention design. Currently, many open-source analysis software such as CaImAn, using a traditional non-negative matrix factorization algorithm (Giovannucci et al. 2019), or CITE-On, employing an implementation of the machine learning model RetinaNet (Sità et al. 2022), can perform live or real-time segmentation, extract the activity of detected neurons and either perform the binarization via an integrated function or can pass the activity traces to a second script for the binarization. It is worth noting that a confusing trend started lately by labeling such function as “online” analysis instead of “live” or “real-time”, which makes it hard to differentiate if the analysis being done remotely on an actually online (on the internet) device or in real time. While these analysis tools can be very powerful, they could have a very significant latency if not implemented correctly, especially if they are not integrated into the imaging software directly and rely on the images being first written to disk before getting queued for analysis. What is more, most of those solutions are written in python, a high-level language that must be compiled into bytecode and then interpreted which is significantly slower compared to other languages and adds more latency. For an ideal real-time analysis pipeline, it should be implemented in a low-level language or in hardware to reduce latency. Another very important aspect to keep in mind is that they rely on deconvolution algorithms for the binarization which can result in significant levels of false positive signals, particularly in conditions of low sign-to-noise conditions (Fu et al. 2021). Therefore, we would advise assessing the specificity and sensitivity of the algorithm by combining optical calcium imaging with electrophysiological single cell recordings.

Figure 3:

An illustration of a closed-loop application by 2-photon functional calcium imaging and 2-photon optogenetic manipulation. The first step is to acquire imaging data by 2-photon microscopy. The extraction of the neuronal signal of interest is conducted in real time. Based on the current neuronal signal of interest, a tailored optogenetic intervention is designed, and executed by optogenetic manipulation with single-neuron accuracy.

The next step is to design the intervention or activation vector for optogenetic stimulation. Recent advancements in light delivery technics such as two-photon holographic stimulation allows for a single cell targeting (Shemesh et al. 2017). This method uses a spatial light modulator to restrict the photostimulation to specifically targeted regions using a computer-generated hologram. In combination with very fast opsins, such approaches are capable of reliably reading and mimicking network activation patterns and are a very important tool for understanding microcircuit functions. Such all-optical closed-loop system was demonstrated in 2018 (Zhang et al. 2018), but used a pre-defined threshold of activity of selected neurons to initiate the photostimulation. For more complicated experimental designs or complex disorders, this intervention vector must be designed considering the activity of the network as a whole instead of few selected cells and tailored to best emulate a physiological network activation, which is dependent on the brain region and the nature of the endogenous activation dynamics. This would require a considerable amount of data collection before the start of the closed loop experiment, especially if the endogenous activation dynamics varies among individuals.

The development of closed-loop applications could be useful not only in research but also in the treatment of complex disorders such as movement, memory and consciousness disorders, where a single neuron manipulation is required. In a recent case report, the ability to communicate was restored in a patient with locked-in syndrome by using a one-way brain computer interface, i.e. neural activity can only be read but not be manipulated (Chaudhary et al. 2022). By placing microelectrode arrays in the patient’s motor cortex, the computer was able to decode neural activity and play auditory tones as feedback for the patient. After some training, the patient was able modulate his neural activity to spell letters and words and communicate with the external world. Optogenetics is already a very powerful toolbox for targeting neuronal sub-populations using specific promoters and could be used in a two-way BCI.

4 Clinical applications

Optogenetics can be used as a new powerful biomedical tool to treat or to attenuate disease progression and symptoms. This technology has been already used with success in animal models of neurodegenerative disorders. For example, in Parkinson’s disease (PD), Deisseroth and coworkers have dissected the disease circuit by optical deconstruction (Gradinaru et al. 2009). Different elements of the circuit were systematically targeted. Indeed, optical stimulation of layer V projection neurons in the primary motor cortex influences the neuronal activity in the subthalamic nucleus in freely moving parkinsonian animals. In another study, it has been reported that optical activation of basal ganglia circuitry through the stimulation of the direct-pathway medium spiny projection neurons expressing ChR2 in the striatum of PD mouse model reduced motor behavioral deficits (Kravitz et al. 2010). Therefore, the optical activation of a particular cell type in an identified area can help to balance PD symptoms. Another disease amenable for optogenetics treatment is retinal degeneration which is characterized by the loss of photoreceptors leading to complete blindness. In this case, optogenetics can be applied to restore photosensitivity in ON bipolar cells of degenerated retinas (Lagali et al. 2008). The activation of ChR2 expressed by ON bipolar cells in a mouse model of the disease is sufficient to induce light-evoked spiking activity in ganglion cells and photoresponses are transmitted to the visual cortex. These results were also obtained when expressing ChR2 in inner retinal neurons (Bi et al. 2006). In addition, the optical activation of ON bipolar cells drastically increases performance in visual behavioral task. Optogenetics has been also used to highlight elements of disease circuit in psychiatric disorders such as autism (Nakai et al. 2021; Yizhar et al. 2011b), depression (Covington et al. 2010; Fakhoury 2021), anxiety (Haubensak et al. 2010; Jarrin and Finn 2019) and schizophrenia (Patrono et al. 2021; Sohal et al. 2009; Wolff et al. 2018). The activation or the silencing of subset of neurons in animal model can highlight cellular- or circuit-level targets for treatments and it allows a pre-screening before to apply this strategy to overcome human disease.

Ontogenetical therapeutical treatment may offer many advantages compared to pharmacological and electrical treatments usually used to treat neurological disorders. The major advantage of the optogenetics is that it allows targeted activation or inhibition of a specific cell population or circuit (Gradinaru et al. 2009). In contrast, pharmacological and electrical medical treatments act unselectively on the activity of different cell types. Deep brain stimulation (DBS) has been used to regulate the abnormal activity of targeted structures in psychiatric disorders (Kopell et al. 2004), epilepsy (Zhong et al. 2011), PD (Okun 2013) and obesity (Dupre et al. 2015). But DBS is less precise as it will activate surrounding cells and processes within the electrical field generated by the stimulus which can lead to unintended side effects. In addition, electrical stimulation nonspecifically stimulates both excitatory and inhibitory neurons (Kim et al. 2011). Moreover, optogenetics displays a fast response compared to pharmacological treatments. Optical stimulation can impact on neuronal activity in millisecond timescale temporal resolution (Boyden et al. 2005) whereas drug treatments take several minutes before to induce an effect due to their time of diffusion, absorption and action. The fast effect induced by optical stimulations is as good as the one obtained with electrical stimulation.

However, optogenetics therapy also exhibits significant potential risks. The first obstacle to confront is to achieve stable expression of opsin in the human brain. Indeed, the human immune system can react against viral transfection methods used for the delivery of genes encoding light-sensitive proteins. Although AAVs have been already used in primate studies to express ChR2 (Han et al. 2009) successfully and did not induce immune reaction or cellular abnormalities even after months, the immune reaction to AAV has proved to be different in humans. The first clinical trial for the use of AAV for gene therapy was approved in 2017 with many similar trials following in recent years. Recombinant AAVs for gene transfer are derived from wild-type AAV which can infect humans in the early years of childhood. Therefore, the immune system could be primed, preventing rAAV-mediated gene transfer (Au et al. 2022; Wang et al. 2019). The proportion of the human population caring neutralizing antibodies against all serotypes of AAV could be as high as 60 %. What is more, clinical trials showed that AAV could provoke both innate and adaptive immune responses (Ronzitti et al. 2020).

Moreover, optogenetics therapy raises ethical concerns. Indeed, this therapy will modify human gene expression in an irreversible way via introduction of non-human DNA to externally regulate cell activity and neuronal network. This control of brain activity can lead to govern human behaviors through motor and psychological actions. This will imply a third part in the control of human actions and decisions. In this context, some limitations must be set to avoid abuses and misuses of this technics.

Recent advancements in light delivery can minimize the invasiveness of optogenetic treatment or even eliminate the need for a surgical implant for a completely non-invasive approach. A group of researchers developed a method of using mechanoluminescent nanoparticles that can reach the brain through the blood stream without the need for any surgery (Wu et al. 2019). Those nanoparticles could then be activated using ultrasound to deliver nanoscopic light with millisecond latency. Another potential approach could be the use of red or near infrared light for transcranial stimulation (Pouliopoulos et al. 2022). Since red and near infrared light is less scattered and absorbed by the skull and tissue, the use of red shifted opsins could be activated via an external device and eliminate the need for surgical implants.

Optogenetic trials have already started in humans. At the GenSight startup, a team recruited volunteers with retinitis pigmentosa, a type of inherited blindness due to the loss of rod photoreceptors. They used gene therapy to deliver modified AAV2 encoding photoactivatable opsin protein to confer a photoreceptor-like function to the retinal cells combined with biomimetic goggles they developed to stimulate retinal cells with specific wavelength. The goggles mimic the normal retinal activity of capturing vision information, then amplify the light signal to the retina. The results of this therapeutical strategy was successfully able to restore a sensory response in the visual cortex and to restore the vision in one patient as the first reported case of partial function recovery in a neurodegenerative disease using optogenetics (Sahel et al. 2021).

5 Outlook

Optogenetics as it stands right now is already a very powerful tool to manipulate microcircuit activity in many animal models – also in humans. Looking ahead, we are on a path to use optogenetics for a true emulation of endogenous network activation, particularly taking advantage of the parallel development of optical imaging and systems integration concepts. With the development of new opsins and light delivery methods, the accuracy of cell targeting, both on a microcircuit scale and mesoscale can be improved. The development of closed loop applications can lay the foundations to network-informed clinical approaches to treat disorders of higher complexity, currently beyond the scope for any causal treatment.