Treatments for neuropsychiatric disorders include drug therapies, surgical interventions, electrode-based deep brain stimulation (DBS), transcranial magnetic stimulation (TMS), vagus nerve stimulation (VNS) and psychotherapy. These methods represent different modalities for interfacing with the diseased brain, from molecules to cognition. Although seemingly disparate, the effects of these treatments converge at the level of neural circuits, altering the way neurons individually communicate and collectively compute. The brain consists of numerous subtypes of excitatory, inhibitory and modulatory neurons, and it has become increasingly clear that specific cell types have crucial roles in many neuropsychiatric diseases. For example, the death of dopaminergic neurons in the substantia nigra pars compacta leads to Parkinson's disease in humans and related symptoms in animal models1, the generation of new neurons in the hippocampus has been linked to the therapeutic efficacy of antidepressants2, and dysfunction of parvalbumin (PV)-positive 
-aminobutyric acid (GABA)-releasing interneurons in the prefrontal cortex (PFC) has been implicated in schizophrenia3, 4, 5.
To better understand the contribution of specific cell types to the physiology of neuropsychiatric disease, a pair of microbial light-sensitive proteins have been developed that, when used togther, can optically interrogate intact neural circuits and bidirectionally control animal behaviour6. As described below, co-expression of these proteins enables bidirectional modulation of electrical signals (activation or inhibition) with millisecond precision and thus allows probing of the downstream circuit-level effects of turning specific cells on and off. Moreover, the ability to control the activity of specific neural populations may make it possible to generate new models of neuropsychiatric diseases by directly mimicking errant electrical signals in the brain with millisecond precision. Finally, therapies using cell-specific modulation might eventually be employed to selectively regulate malfunctioning neurons, thereby optimizing treatment efficacy and reducing the side-effects associated with less specific therapeutic interventions. Many other elegant approaches have been developed recently for cell-type-specific control7, 8, 9, 10, 11; in this Progress article, we discuss the recent advances in genetically targeted optical neuromodulation with microbial opsins, and give some broad general perspectives relevant to both basic science research and clinical application.
Optogenetics: two microbial opsins
Neuropsychiatric diseases may result from circuit-level effects of a group of malfunctioning neurons; for instance, the PV-positive interneurons in the dorsolateral PFC of schizophrenic patients have reduced GABA immunoreactivity relative to controls, suggesting reduced inhibition of cortical pyramidal neurons4, 5. To best understand the role of a specific neuron subtype in living animals, we must be able to bidirectionally turn these neurons on and off with cell-type specificity, high temporal precision and rapid reversibility. To satisfy these requirements, the microbial light-sensitive proteins Chlamydomonas reinhardtii Channelrhodopsin-2 (ChR2) and Natronomonas pharaonis halorhodopsin (NpHR) have been introduced into neurons6, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 25.
ChR2 (initially cloned in Ref. 23) is a monovalent cation channel that allows Na+ ions to enter the cell following exposure to 
470 nm blue light, whereas the NpHR (described in Ref. 24) is a chloride pump that activates upon illumination with 580 nm yellow light (Fig. 1a). As the activation maxima of these two proteins are over 100 nm apart (Fig. 1b), they can be controlled independently to either drive action potential firing or suppress neural activity in intact tissue6 (Fig. 1c shows implementation of this protocol in hippocampal neurons using cell-attached and whole-cell patch clamp), and together may modulate neuronal synchrony17. Both proteins have fast temporal kinetics, on the scale of milliseconds, making it possible to drive reliable trains of high frequency action potentials in vivo using ChR2 (Ref. 15) and suppress single action potentials within high frequency spike trains using NpHR6.
Figure 1 | Optogenetic tools: ChR2 and NpHR.
a | Schematic of channelrhodopsin-2 (ChR2) and the halorhodopsin (NpHR) pump. Following illumination with blue light (activation maximum 470 nm, Ref.23), ChR2 allows the entry of cations (mostly Na+ and very low levels of Ca2+) into the cell. NpHR is activated by yellow light illumination (activation maximum 580 nm, Ref. 6) and allows the entry of Cl- anions. b | Action spectra for ChR2 and NpHR. The excitation maxima for ChR2 and NpHR are separated by 100 nm, making it possible to activate each opsin independently with light. c | Cell-attached (top) and whole-cell current-clamp (bottom) traces from hippocampal neurons showing all-optical neural activation and inhibition. Blue pulses represent the blue light flashes used to drive ChR2-mediated activation and the yellow bar denotes NpHR-mediated inactivation. Part a was modified with permission from Nature Ref.52 © (2007) Macmillan Publishers Ltd. Part c was modified with permission from Nature Ref.6 © (2007) Macmillan Publishers Ltd.
Because NpHR remains active for many minutes when exposed to continuous light6 and deactivates quickly when light is turned off, it can be used to mimic lesions in a rapid, stable and reversible manner. Stability in general is particularly important; it has long been known that many archaebacterial species express red-shifted light-activated halorhodopsins that can pump chloride ions and modulate membrane potential (for example, H. salinarum, H. halobium, and N. pharaonis), but we selected the Natronomonas halorhodopsin to experimentally develop for neural control, as NpHR has enhanced stability and chloride affinity6. Moreover, halorhodopsin does not require the addition of the cofactor all-trans-retinal (ATR) to function in intact mammalian tissue6, a result that is consistent with previous findings with the homologous opsin ChR2 (Refs 13,25). The ability to functionally express ChR2 and NpHR without the addition of any exogenous cofactor is crucial for applications to basic science investigations, preclinical disease models and possible clinical translation.
Light delivery into the brain
To modulate the activity of ChR2- and NpHR-expressing neurons, light must be delivered to the brain region of interest. Recently a fibreoptic-based system was developed that is suitable for delivering light in vivo16 to both superficial and deep brain structures. In this approach, a thin optical fibre (0.1 to 0.2 mm in diameter) can be inserted through a cannula guide (Fig. 2a) that is targeted precisely to the light-sensitized neurons, to deliver effective photostimulation to specific neuronal subtypes in the brain of freely moving rats (for example, expressing ChR2 or NpHR; Fig. 2b). The optical fibre can be coupled to a bright light source such as a diode laser16. Alternative solutions may involve digital light processing (DLP) micromirrors for projection, or high intensity light-emitting diodes (LEDs) directly mounted on the head of mice or rats; moreover, in larger animals such as primates, millimetre-scale LEDs or LED arrays could be directly implanted into deeper brain structures. In animals, the LEDs can be either tethered to an external power source or powered by a battery pack carried by the animal to allow normal behaviour; the LEDs could even be powered wirelessly by radio frequency or magnetic induction.
Figure 2 | In vivo optical neuromodulation in animal models of neuropsychiatric disease.
a | A cannula is implanted into the head of the experimental animal to guide an optical fibre to the targeted brain region. b | The optical fibre is coupled to a strong light source (here a 488 nm laser diode16) to bring blue or yellow light into the brain. c | Genetic targeting of channelrhodopsin (ChR2) or halorhodopsin (NpHR) into defined classes of disease-model-relevant neurons may allow cell-specific neuromodulation and avoid inadvertent stimulation of disease-model-irrelevant neurons as occurs with electrical stimulation. Figure modified with permission from Ref. 16 © (2007) Institute of Physics Pub.
Protein expression in neuron subtypes
Different types of neurons are characterized by unique gene expression patterns26. Many of the neuron types implicated in neuropsychiatric disorders can be identified based on their immunoreactivity to antibodies that recognize neurotransmitters (for example, dopamine, cholecystokinin or neuropeptide Y) or neurochemical markers (for example PV, calretinin or calbindin), suggesting possible targets for experimental or therapeutic control. Similar principles extend to disorders lying outside conventional neuropsychiatric disease categories. For example, animals lacking melanin-concentrating hormone (MCH) are reported to be hypophagic and lean27, and optical inhibition of the MCH-secreting neurons in the lateral hypothalamus using NpHR might therefore be used to modify food intake or obesity. A wide array of techniques is now available for genetically specifying the expression pattern of probes like ChR2 and NpHR in mouse models (for thorough discussions on strategies for making transgenic mouse lines see Refs 28,29) or in other experimental systems7, 11.
For potential preclinical animal models and potential clinical applications, viral gene delivery also may provide a convenient and quick approach for mediating ChR2 and NpHR expression. Viral vectors carrying the genes encoding ChR2 or NpHR can be delivered stereotactically into discrete brain regions6, 16, 30 (for detailed discussions on viral gene transfer into the nervous system see Ref. 31). Recombinant lentiviral and adeno-associated viral (AAV) vectors32, 32 have been popular choices for gene transfer into patients. Lentiviral vectors are integrated into the genome of the target cell and confer permanent gene expression. AAV-mediated expression may be less stable because a much smaller percentage (less than 1%) of the virus is integrated. These vectors can also be rendered cell-type-specific by the choice of promoter34, 35, 36, viral receptor37 or spatial targeting strategy38. However, the limited packaging capacity of many viral vectors restricts the size of transgene cassettes that can be used (10 kb or less for lentivirus and 5kb or less for AAV31).
Identification of relevant neural circuits
The cell types pertinent to most neuropsychiatric diseases remain to be identified and in most cases even compelling hypotheses for candidate cell types are lacking. Therefore, to generate hypotheses that will elucidate the cellular basis of disease mechanisms, it will be advantageous to identify relevant neurons based on their activity or functional connectivity during the manifestation of disease symptoms (for example, during states of elevated stress or anxiety). One possible method is to take advantage of promoters from immediate-early genes (IEGs; for example, FOS and ARC) to drive expression39, 40of fluorescent-protein-tagged ChR2 and NpHR in neurons that are active during particular behaviours relevant to diseases.
ChR2 or NpHR expression can also be restricted to cells that are synaptically connected to a previously-identified relevant population, in order to determine which connections are most important to disease symptoms. For example, it is possible to use viral delivery to retrogradely label neurons with ChR2 or NpHR based on their projection patterns38, 41. A recombinant, glycoprotein-deleted form of rabies virus has been shown to mediate strong protein expression not only in the neurons that project to the site of viral delivery, but also in their presynaptic partners that may be located in distant neural circuits41. With this approach, it may be possible to stimulate only the cells in a particular brain region that connect to cells in another brain region of interest, and to determine how these connections are altered in a disease state. This approach may be especially useful in diseases such as schizophrenia and depression, where the implicated cell types receive input from and make functional connections with neurons in heterogeneous regions of the brain (such as the hippocampus and neocortex). Analogous development of anterograde labelling methods will complement this approach and enable more powerful analysis of relevant circuits.
Integration with output measurements
To be informative in an experimental setting, optical neuromodulation must be tightly coupled with an appropriate quantitative readout. Stimulation of neurons using light makes it easier to perform simultaneous electrical recordings in vivo without electrical artefact interference from stimulation. This unique aspect of optical neuromodulation allows electrical monitoring of activity changes in brain regions within, or downstream from, the site of stimulation with no stimulus artefact-induced delay. Therefore, it is now possible to map out the functional connectivity of disease circuits in real time.
It also will be important to combine optical stimulation/inhibition with non-invasive forms of output measurement, including optical imaging, positron emission tomography (PET) and functional magnetic resonance imaging (fMRI). In particular, optical recording of neural activity in vivo42, 43or in brain slices holds great potential for monitoring effects on the activity of large neural populations. Ultimately, as genetically encoded activity probes44, 45mature, it will be possible to perturb a specific population of neurons and examine their influence on another genetically defined population of neurons. Finally, particularly for the study of the cellular and circuit-level mechanisms of neuropsychiatric diseases, the effects of modulating specific cell types in vivo needs to be quantified through behavioural assays. Light delivery with long, flexible and lightweight fibreoptics16 or LED leads will allow experimental animals to behave freely during targeted photostimulation.
It also may be possible to establish novel, reversible behavioural models of neuropsychiatric diseases that are linked to specific cell types as the ability to control the electrical activity of specific populations of neurons in discrete regions of the brain has the potential to reversibly mimic underactivity or overactivity of candidate neurons. Additionally, when applied to existing behavioural models used as output measures, cell-specific neuromodulation will be useful to either exacerbate or rescue the behavioural phenotypes and thereby provide insights for future clinical translation.
Possible clinical applications
The ability to control the activity of a defined class of neurons has the potential to advance clinical neuromodulation. Existing electrode-based DBS methods indiscriminately stimulate all neurons within a given volume, including cells that are not implicated in the disease state, thus leading to unwanted side-effects and even reduced efficacy as opposing excitatory and inhibitory cell types are affected by the electrodes. Genetic control makes it possible to develop more precise therapies by restricting the excited or inhibited neurons to the disease-relevant population (Fig. 2c). Moreover, the ability to simultaneously record electrical activity during optical stimulation without electrical artefacts makes it possible to engage in responsive neuromodulation by dynamically adjusting the stimulation or inhibition intensity based on feedback from the activity state in the brain. This feature may be especially useful for diseases characterized by sporadic fluctuations in electrical activity such as epilepsy. The combination of recording and optical control could also be used to bridge severed connections, for example to relay information from the brain to distal limbs in the case of severed spinal cord. Box 1 describes several potential scenarios for optical neuromodulation therapies.
Although the direct use of optical neuromodulation in humans has tremendous potential to control disease circuits with a precision unparalleled by drugs or electrode-based techniques, two key issues require additional research and development. First, the genes encoding ChR2 and NpHR must be safely and stably transferred into the patient's neurons. Progress is being made on this front as several viral and non-viral-based gene delivery methods are currently in late stages of clinical testing32. Second, a lesser, but still significant, problem is the engineering of an appropriate optical device. To achieve effective optical control, it will be necessary to deliver sufficient intensities of light to the correct local circuit, perhaps using analogous methods to those developed for light delivery in animal models. Recent advances in microelectronics, high-capacity batteries and optoelectronics have made the development of clinical-quality optical brain stimulators a tractable engineering problem.
Most importantly, long before optical neuromodulation is ready for use in neuropsychiatric clinics, the knowledge gained from the use of these tools in animal models is likely to have an impact on our understanding of neuropsychiatric circuit pathology. Moreover, further understanding of brain diseases in this way is likely to influence future development of existing pharmaceutical agents and medical devices; for example, knowing the cell populations responsible for the efficacy of DBS treatments could lead to better electrode placement and design.
Future perspectives
The convergence of genetic, optical and engineering advances holds substantial potential for mechanistic investigation of disease aetiology and treatment. The ability to restrict neuromodulation to genetically specified neuronal subtypes will enable more precise analysis of brain function. However, for these optogenetic methods to reach their full potential it is still necessary to improve the tools of their application. Achieving more precise genetic control through the identification of novel promoters together with enhanced understanding of transcriptional regulation, and developing less invasive modulation (for example, by harnessing infrared light) may lead to new breakthrough technologies.
A final perspective arises from the viewpoint of the practising psychiatrist. We do not yet know the precise neural 'code' that is relevant to symptoms of psychiatric disease, so in most cases we can not test hypotheses by delivering precisely timed pulses despite having the tools to do so. However, we predict that key psychiatric symptoms like those associated with depression (for example, hopelessness, anhedonia or low energy) may have less to do with coding or computation, and more to do with alterations in the spatiotemporal spread (also referred to as percolation) of activity through bottlenecks in neural circuits controlled by specific cell types46. Indeed many aetiologies and treatments could be classified as positive or negative modulators of activity propagation that do not affect coding per se46. By contrast, disease states with borderline and schizophrenic features in which aspects of the self can be experienced as alien (as with auditory hallucinations or dissociative symptoms) might be due to a high-speed coding problem in which neurons reporting self-generated stimuli are 'mis-tagged' with a non-self millisecond-scale timing tag (for example, oscillating at the wrong frequency or without correct synchronization). Both the 'tagging model' and the 'percolation model', which are based on, and fit well with, clinical psychiatry experience, correspond to theoretical extremes in the range of psychiatric disease mechanisms. Ultimately both models may be susceptible to testing with optical6, 46and other high-speed neuroengineering technologies.