Physical mechanisms of emerging neuromodulation modalities

One of the ultimate goals of neurostimulation field is to design materials, devices and systems that can simultaneously achieve safe, effective and tether-free operation. For that, understanding the working mechanisms and potential applicability of neurostimulation techniques is important to develop noninvasive, enhanced, and multi-modal control of neural activity. Here, we review direct and transduction-based neurostimulation techniques by discussing their interaction mechanisms with neurons via electrical, mechanical, and thermal means. We show how each technique targets modulation of specific ion channels (e.g. voltage-gated, mechanosensitive, heat-sensitive) by exploiting fundamental wave properties (e.g. interference) or engineering nanomaterial-based systems for efficient energy transduction. Overall, our review provides a detailed mechanistic understanding of neurostimulation techniques together with their applications to in vitro, in vivo, and translational studies to guide the researchers toward developing more advanced systems in terms of noninvasiveness, spatiotemporal resolution, and clinical applicability.


Introduction
Neural stimulation is a clinically-approved tool for treatments of different neurological disorders such as Parkinson's disease, dystonia, essential tremor, major depressive disorder, retinal degeneration and hearing loss [1][2][3][4]. In addition, it is also an invaluable research tool that enables researchers to explore and unravel the working principles of neural structures from organelle and single-cell level to organ and system level [5,6]. Even though neurostimulation already constitutes a huge market size over 4 billion dollars and affects the lives of millions of patients, the interest in neurostimulators in the industry and academia is still growing. For example, the market size of neurostimulators is expected to reach 14 billion dollars in 2030 with an annual growth rate over 10% [7].
Neural stimulation aims for the modulation of the membrane potential, which originates from the ionic charge difference across the membrane due to the balance of the diffusion and drift of the ions such as sodium and potassium. The membrane potential can be fundamentally perturbed by the application of ionic currents, temperature change of the membrane, and mechanical deformation of the membrane. Today, there exists a wide variety of transduction techniques for the perturbation of the membrane potential from photothermal, photoacoustic to magnetoelectric, magnetoacoustic, etc that are also named as the type of input energy and output stimulation mechanism (e.g. photoacoustic where light is the input and acoustic waves are the output). Even though the type of input energy and output stimulation mechanism can be intuitively understood, understanding the transduction steps of various mechanisms is important to develop new, advanced and multimodal neurostimulation techniques. This understanding involves exploring the synthesis and characterization of transducing micro/nanomaterials for achieving desired electronic, optical, thermal, or acoustic properties to ensure efficient energy transduction. Moreover, the coupling of output energy to cellular membrane via voltagegated, mechanosensitive, or heat-sensitive ion channels and the spatiotemporal properties of resulting biophysical processes are critical for finding suitable in vitro or in vivo applications.
In this review, we discuss three fundamental stimulation routes, i.e. electrical, mechanical, and thermal stimulation, by presenting their direct and noninvasive application routes (e.g. transcranial electrical stimulation, focused ultrasound stimulation (FUS)), and their indirect and minimally invasive application routes via nanotransducers (e.g. photothermal, magneto-acoustic stimulation). We provide an overview of the operation principles for each stimulation modality by summarizing the input energy, energy transduction mechanism, biophysical and physiological processes, and its neurostimulation spectrum together with advanced applications. We conclude the paper with discussion and perspectives on the applicability and effectiveness of the reviewed mechanisms and possible future directions in neurostimulation research.

Electrical stimulation
The conventional way to modulate the electrical activity of neurons is electrical stimulation. Since the extracellular medium surrounding neurons and other excitable cells is an ionic solution, electric signals can be transferred via ionic movements from electronic devices to the cells. Stimulation electrodes induce a potential gradient in the extracellular medium, which results in current flow via the movements of ions. This injected current changes the extracellular electric potential near the cellular membrane (figure 1(a)) [8]. Since the membrane potential is equal to the difference between intracellular potential and extracellular potential, current-injecting electrodes drive the membrane potential either to more positive values (depolarization) or more negative values (hyperpolarization). Sufficient depolarization can elicit action potentials by opening voltage-gated ion channels (figure 1(a)), while hyperpolarization can inhibit action potential firing.
There are two main mechanisms of current injection at the electrode-electrolyte interface. Capacitive charge injection is governed by the ionic displacement currents occurring during the formation of electrical double layer at the interface (figure 1(b)) and Faradaic mechanism involves electron transfer between the electrode and electrolyte (figure 1(c)) [8]. Faradaic reactions can be irreversible, where the reaction products diffuse away from the interface without being recovered, or reversible, where the reaction couples are rapidly reduced and oxidized before moving away from the interface [9]. Irreversible processes can generate byproducts in the electrolyte, which may not be desired in a physiological environment, while the recovery of injected charges via reversible processes prevents formation of byproducts. Reversible Faradaic reactions are similar to capacitive charge injection in terms of inducing charge storage at the electrode-electrolyte interface with the advantage of having larger capacitances compared to double layer capacitors [9,10]. While capacitive stimulation electrodes with purely double layer capacitance have typically low charge injection capacity, increasing the ratio of electrochemical surface area/geometric surface area by porous coatings can significantly increase the charge injection capacity [10].
Electrical stimulation can be performed either epidermally or by implantation of electrodes near excitable tissues. Epidermal stimulation is advantageous because of its non-invasive nature, but it is not suitable for spatially-specific stimulation. On the other hand, implanted electrodes offer higher spatial resolution with the downside of enhanced invasiveness. Besides conventionally used implanted electrodes (e.g. platinum, titanium nitride, iridium oxide), nanotransducers that convert electromagnetic or acoustic energy into electric current emerged as effective electrical stimulation tools in the recent years since they enable building wireless and remotelyactivated, i.e. minimally-invasive neuromodulation systems.

Photoelectric stimulation
Optoelectronic systems rely on the light-triggered exciton generation within the photoactive material, its dissociation inside the device structure depending on the energy band diagram, and generation of the electric fields for inducing membrane potential changes as a result of the dissociated charges. Advantageously, optoelectronic systems remove the need for electrical wiring and consequently reduce the foreign body response, which is one of the fundamental problems of long-term-use neural prostheses.
Moreover, optoelectronic conversion shows particularly similar function to the retinal cells converting optical energy to electrical signals. This similar functioning can also be utilized in the treatment of retinitis pigmentosa, which is a highly seen retinal degeneration disorder as many as 1/4000 individuals in the world [26]. Organic and inorganic semiconductor-based biointerfaces have been successfully shown as retinal prostheses with capability of photostimulation of neurons. In organic semiconductor-based systems, the organic photoactive polymers including poly(3-hexylthiophene-2,5diyl) (P3HT), phenyl-C61-butyric acid methyl ester and their derivatives or blends were widely used either as the active layer where photogenerated electronhole pairs are produced [15,18,[27][28][29]. One of the critical aspects of designing retinal implants is to have a functioning device within the ocular safety limits. To do so, enhanced photoconversion and charge separation properties are needed in order to reduce required illumination intensity threshold. With this purpose and thanks to constant advancements in material science, several different materials have been proposed and integrated into previously reported architectures to promote aforementioned optoelectronic properties. The study by Rand et al comprised a thin, 80 nm of tri-layer of metal and p-n semiconducting organic NCs showed successful neural stimulation under 660 nm light excitation (figures 2(a) and (b)) [16]. This electrolytic photocapacitor with p-n heterojunction bilayer on top of a metallic Cr/Au back-contact generates excitons upon illumination [16]. The electrons accumulated in the n-type material induce charged double layer at the device/electrolyte interface, which induces dominantly photocapacitive stimulation of neurons (figures 2(c) and (d)). The fluorescent calcium imaging experiments revealed that under 600 nm, 480 mW cm −2 , 100 pulses with duration of 5 ms and interpulse interval 10 ms, Cr/Au/H2Pc/PTCDI device successfully stimulates neurons (figures 2(e) and (f)) [16]. Moreover, the device also successfully stimulates light-insensitive retinas. To further enhance the charge injection performance of such photocapacitor devices, recent reports proposed integration of dielectric materials [31], plasmonic NPs [32], and supercapacitor materials [33] that can allow building more light-sensitive systems.
One recent study by DiFrancesco et al [30] utilizes graphene layer for effective photogenerated charge extraction in the biointerface (figure 2(g)).
The widely studied 2D material, graphene, shows high electrical conductivity and exceptional material properties such as ultra flexibility and transparency. These features have been successfully demonstrated in novel scaffolds [34], wearable electronics and also in implantable electrodes [35] for monitoring cortical brain activity. Inspired by these studies, graphene is used to replace PEDOT:PSS conductive layer for proper band alignment of the photovoltaic device silk fibroin base substrate, chemical vapor deposited (CVD) graphene for charge extraction and photoactive P3HT interfacial layer [30]. The enhanced work function of CVD graphene increases the device performance in comparison with PEDOT:PSS based interface and efficiently recovers light-sensitivity in blind retina explants of the dystrophic Royal College of Surgeons (RCS) rats (figures 2(h) and (i)). An interesting corollary of the reported study is the relation between the initial state of the membrane voltage with the modulation of the neural activity (figure 2(j)). The more initial hyperpolarization will lead more light-induced depolarization and similarly more initial depolarized neural state will yield more hyperpolarization under illumination. Depending on the targeted retinal pathway for neural modulation, the ON/OFF cycles need to be designed for specific excitation/inhibition balance of the targeted neural activity [30]. Such future studies might unravel the possible different applications of light-mediated signals in different neural networks, not limited to the retinal tissue.
On the other hand, nanomaterials represent unique structural, electrical and optical features to be used as the mediators of neuronal stimulation by transforming primary stimulus to spatiotemporally confined secondary stimulus [36]. Particularly, their small size compared to the targeted biological molecules or cells can affect these systems and processes at the molecular level. By utilizing several binding agents, such as antibodies and proteins, nanomaterials can potentially diffuse through the blood-brain barrier, which is normally difficult for milli/micron scale systems [37]. Moreover, optical absorption tunability from visible to the nearinfrared (NIR) region can offer many advantages such as lower absorption by hemoglobin [38] and high tissue penetration as compared to visible light [39]. However, toxic heavy metal content (e.g. cadmium, mercury), electrochemical coupling with the cells and low photon-to-current efficiency limit their effective use. Photoelectric transducers for neuromodulation. (a) Schematic of the photocapacitor consisting of sequentially deposited chromium/gold (Cr/Au) and phthalocyanine (H2Pc) (p-type) and N,N ′ -dimethyl perylenetetracarboxylic diimide (PTCDI) (n-type). (b) Molecular structures of the pigment semiconductors. Metal-free H2Pc functions as the primary light-absorbing layer and p-type electron donor, while PTCDI acts as the n-type electron-acceptor, which attains a negatively-charged surface upon illumination. (c) Energy band illustration of a metal-p-n photocapacitor during the start of the illumination pulse when the capacitor charges. (d) Mechanism of capacitive coupling of an illuminated photocapacitor with an adjacent cell. (e) Photostimulation of neuronal cultures with control cortical primary neurons cultured on poly-D-lysine (PDL)-coated Petri dishes, and cortical primary neurons cultured on type I devices. (f) Calcium imaging traces (dF/F) of neurons cultured on PDL-coated Petri dish and neurons cultured on type I devices. Vertical red lines indicate a light stimulation of 100 consecutive pulses (600 nm, 480 mW cm −2 , pulse duration 5 ms, interpulse interval 10 ms). For this purpose, NCs [16] (aluminum antimonide (AlSb) [18], perovskite [40]) and colloidal QDs (indium phosphide core (InP) [22], indium phosphide/zinc sulfide core/shell (InP/ZnS) [23,24], lead-sulfide [39]) have been proposed for neural stimulation. By employing the optical and electrical tuning properties of these nanomaterials, they can be utilized either as interfacial layer of thin films or mixed with photoactive polymers to achieve spectral control in visible and NIR optical windows and manipulate the charge injection mechanism at the device-electrolyte interface [41].
On the other hand, alternating strategies for optical stimulation are concentrated on the injectable nanomaterials. As the two-dimensional planar neural interfaces exhibit lower spatial resolution over one order of magnitude lower than the foveal cones, which are about 4-5 µm in diameter [42], nanoscale materials hold great promise for superior spatial resolution [43][44][45][46][47].

Magnetoelectric stimulation
In its general form, magnetoelectric effect is described as the coupling between magnetic field and electric field in matter [48]. This physical phenomenon can be an intrinsic property of certain materials, which are called magnetoelectric materials. These materials have non-zero magnetoelectric effect, which allows them to generate a local electric field when they are exposed to a magnetic field. Advantageously, magnetoelectric materials can operate with low intensity magnetic fields, 10 −2 -10 −1 T, compared to the field intensities used in transcranial magnetic stimulation (TMS) (typically >1 T) [49]. Besides generating a local electric field which can activate voltage-gated channels, they also amplify the local magnetic field, which can also generate local electric current due to mutual induction principle. Thus, the administration of magnetoelectric materials into the vicinity of neural structures might allow remotely controlled and localized neurostimulation via low magnetic fields.
Yue et al proposed the use magnetoelectric NPs for neurostimulation and investigated the potential of these materials for brain stimulation in a computational study [50]. They simulated the effect of 20 nm magnetoelectric NPs on the activities of neurons in the deep brain regions, where NPs are assumed to be injected, to find the optimal NP concentrations in solution and magnetic field frequencies that will result in normalization of neural activity in a model patient with Parkinson's disease. With the optimized NP magnetoelectric coefficient, solution concentration, and the intensity and frequency of the externally applied magnetic field, they demonstrated that the electrical activity in certain brain regions of patients with Parkinson's disease can be brought to similar levels with the neural activity of healthy people [50]. The same group later showed an in vivo animal study, in which 30 nm CoFe 2 O 4 -BaTiO 3 core/shell magnetoelectric NPs were injected into the bloodstream of mice [51]. The NPs were forced to cross the blood-brain barrier by applying a d.c. magnetic field gradient and distributed over the desired parts of the brain by application of spatially varying field (figure 3(a)). EEG recordings from the mice brain showed that activation of magnetoelectric NPs via application of external low intensity (10 −2 T) a.c. magnetic field led to the activation of neurons in the deep brain regions where NPs were delivered [51].
More recently, Kozielski et al reported the use of CoFe 2 O 4 -BaTiO 3 for wirelessly stimulating deep brain parts of freely moving mice [52]. In that nanostructure, CoFe 2 O 4 is the magnetostrictive material and BaTiO 3 is the piezoelectric material. Application of external magnetic field creates strain in magnetostrictive CoFe 2 O 4 , which in turn applies strain to piezoelectric BaTiO 3 , leading to electric field generation due to charge separation in BaTiO 3 ( figure 3(b)). The optimal magnetoelectric effect was observed when alternating and direct magnetic field were applied together (figure 3(c)). Injection and activation of NPs via alternating and direct magnetic field (figure 3(d)) modulates the electrical activity of neurons in the motor cortex and limbic thalamus of mice brain (figures 3(e) and (f)), which leads to behavioral responses due to the indirect modulation of thalamocortical circuits [52]. Nguyen et al reported a similar study in which cortical activities were evoked in mice brain using magnetoelectric NPs that were injected and forced to cross the blood-brain barrier [53]. These studies revealed the potential of magnetoelectric materials for selectively stimulating the neural tissues in deep brain regions owing to the strong local electric field generation of such materials upon exposure to remotely applied low intensity magnetic fields.
The low attenuation of magnetic field in the body also allows the use of magnetoelectric transduction for remotely powering the implanted neural stimulators inside the body. One recent study shows a novel application of this, where millimeter-sized magnetoelectrically powered neural stimulators can perform high frequency stimulation in the freely moving rat brain and treat the symptoms of Parkinson's disease [54].

Piezoelectric stimulation
Another indirect generation of electrical stimuli is through piezoelectric materials. Certain piezoelectric NPs and polymers were studied as transducing agents that convert incoming acoustic energy into an electrical stimulus to wirelessly modulate the electrical activity of neurons [55][56][57]. Modulation of neural activity can occur as a result of internalization of piezoelectric NPs (figure 4(a)) [55], absorption of piezoelectric NPs on the cellular membrane (figure 4(b)) [56], or fabrication of thin film device architectures containing piezoelectric NPs/polymers (figure 4(c)) [57].
Ciofani et al first demonstrated that incubation of neuron-like PC12 and SH-SY5Y cells in the piezoelectric boron nitride nanotubes-loaded cell culture medium leads to cellular internalization of nanotubes. When ultrasound waves are applied to these cells, internalized piezoelectric nanotubes convert the mechanical energy of ultrasound into electrical current, which leads to enhanced neural activity and consequently greater neurite growth in the piezoelectrically stimulated PC12 and SH-SY5Y cells [55].
In a follow-up work, barium titanate nanoparticles (BTNPs) were introduced into the cell Particles are forced to cross the blood-brain barrier by applying a DC magnetic field gradient (left). The distribution of particles throughout the brain can be controlled by application of DC magnetic field in a desired pattern (middle). An electric field can be induced at the selected stimulation site via exposure to focused alternating magnetic field, which leads to activation of neurons at the desired spot. Reproduced with permission from [51]. © Future Medicine Ltd (b) The process of magnetoelectricity in strain coupled magnetoelectric material consisting of a magnetostrictive material that produces strain upon exposure to magnetic field and applies strain to piezoelectric material which results in charge separation and electric field generation. (c) Optimization of magnetoelectric coefficient (αME) by application of large direct magnetic field together with low intensity alternating field. culture medium of SH-SY5Y neuroblastoma cells. These NPs predominantly attached to the cell membrane with marginal cellular internalization, which was ascribed to their negative external charge [56]. Similarly, application of ultrasound induces piezoelectricity in the NPs, which stimulates the SH-SY5Y cells, and leads to improved neurite growth. Moreover, the same study showed that the stimulatory effects are predominantly mediated by both Na + and Ca 2+ ion fluxes, shedding light on the Reprinted with permission from [55]. Copyright (2010) American Chemical Society. (b) Extracellular stimulation mechanism. Piezoelectric stimulation resulting from adsorption of barium titanate nanoparticles (red colored) onto the plasma membrane (green colored) of SH-SY5Y cells. Nanoparticles convert ultrasound waves into electrical stimuli to alter the electrical activity of SH-SY5Y neuroblastoma cells. Reproduced with permission from [56]. CC BY-NC-ND. (c) Scanning electron microscopy images of SH-SY5Y cells grown on control (top row), piezoelectric copolymer P(VDF-TrFE) film (middle row), and piezoelectric P(VDF-TrFE)/barium titanate nanoparticles (BTNPs) composite film (bottom row). The cells grown on piezoelectric films show greater neurite growths under ultrasound application (US+ column) compared to the no ultrasound applied (US− column) condition, indicating the effect of piezoelectricity on enhanced neurite growth. Reproduced with permission from [57]. CC BY-NC-ND.
physiological mechanism of the stimulation [56]. Later, the same strategy was shown to be effective on primary neurons as well. BTNPs adsorbed on the cellular membrane of neurons increase the neuronal activity when exposed to MHz frequency ultrasound wave [58].

Transcranial electrical stimulation
As an effective epidermal stimulation approach, transcranial direct current stimulation (tDCS) was first introduced to modern usage by Nitsche and Paulus, who showed its effects on excitability in the human motor cortex [59]. tDCS is the application of low-intensity direct current (DC) (1-2 mA) to the brain cortex through two sponge electrodes placed over the scalp. The polarity of electrodes is chosen according to the direction of the intended cortical modulation-increasing or inhibiting excitability. Two saline-soaked sponge electrodes with 5 × 7 cm size are used to apply current; however, in a modified version of tDCS, the current focality is aimed to optimized by increasing the number and decreasing the size of the electrodes (high definition tDCS).
The level of DC in conventional tDCS is a subthreshold current intensity, which means that it does not trigger a new action potential on the neuronal membrane. Voroslakos and colleagues showed that current fields equivalent to the dose of tDCS (1-2 mA) do not impact neural firings [60]. However, this subthreshold DC can influence the resting membrane potential and alter the state of excitability [61][62][63]. If the resting membrane potential is depolarized, a weaker current can easily induce an action potential; conversely, if the resting membrane potential is hyperpolarized, a stronger current is required to induce an action potential.
There are several kinds of research in animals, human and computational models that investigated the effect of tDCS-induced electric fields on neurons [64][65][66]. Modeling studies predict that as the applied current intensity increases, the electric field in the brain increases [67]. It has also been shown in the modeling, and human studies that 1 mA induces 0.2-0.5 V m −1 and 2 mA induces 0.8-1 V m −1 maximum electric fields [61, 64-66, 68, 69]. Bikson et al showed that the maximum level of electric field for 2 mA tDCS (1 V m −1 ) alters the resting membrane potential of 2 mV, which is not enough to initiate an action potential in the neuron [70]. Moreover, these low electric fields on the cortex can be amplified by coupling with simultaneous endogenous neuronal activities, which increase the efficiency of low-intensity tDCS [71]. The results of human studies support this amplification by showing the increased effect of tDCS with task-specific training (online tDCS) [72]. The soma, axon, and dendrites are susceptible compartments of neurons to the polarization effect of low electric fields. The large pyramidal cells in layer-V of cortex lie perpendicular to the cortical surface and project their dendrites upward and axon downward in deeper layers. Due to this typical vertical position of pyramidal cells, 'somatic doctrine' assumes that the effect of tDCS is directly related to the depolarization/hyperpolarization dipole created between soma and axon (figure 5(a)). While soma is depolarized, the axon is hyperpolarized or vice versa. According to 'somatic doctrine' , the polarization of the soma has a significant role in determining the activity of neuron since the action potential is triggered from here (axon hillock) due to the high intensity of voltage-gated Na channels on the membrane. On the other hand, the cortex has gyri and sulci, which change the orientation of cortical neurons according to electric currents. The axonal modulation should not be ignored as well as the somatic doctrine. Cathodal electrical stimulation increases the excitability of the axon, which lies tangentially under the electrode. Therefore, the effect of cathodal stimulation is related to the orientation of axons. If we consider the interneurons and axons projecting parallel to the surface and reaching the different cortical layers, the net effect of electrical stimulation of the cortex becomes more complicated than a simple somatic modulation. For this reason, it is not possible to accept that anodal stimulation is always excitatory and cathodal stimulation is always inhibitory. Furthermore, cerebellar application of tDCS, induced similar responses independent of polarity, which suggests that the mechanism of DC is not the same in all neural tissues and is not always polarityspecific [73][74][75]. In addition, due to the cortical foldings, the electric field generated on the cortical surface is not uniform and has many local peaks [64]. Also, these maximum electric field magnitudes are not always under the anodal electrode; instead, they are generated between the electrodes depending on electrode sizes and distance between the electrodes (figure 5(b)) [65].
Transcranial alternating current (tACS) is a recent method with a different waveform of weak current. The montages, equipment, and electrodes are the same as tDCS. However, the current is a sinusoidal current with a particular frequency. The term 'alternating' is used because the sinus is balanced between two polarities by a continuous change of current direction between anode and cathode. It is developed by Antal and colleagues in 2008 with the aim of driving endogenous brain oscillations [77].
Contrary to tDCS, the mechanism of action of tACS is still not clear. Since there is no polarity specificity of electrodes, the modulation is mainly related to applied frequency. In low frequencies (EEG frequency range), tACS modulates the amplitude, frequency, or phase of ongoing brain oscillations [78][79][80][81]. The modulation effect of tACS depends on the stimulation frequency and the dominant frequency of activity in the target cortical area [82,83]. Higher frequency tACS suggested to influence the biochemical mechanisms and synaptic plasticity, and this effect seems to depend on the intensity of current [84][85][86]. In conclusion, the modulation effect of tACS may not be linear with changing parameters, and the results of the studies investigating tACS are inconsistent because of different modulating effects of wide frequency and intensity range.

Temporal interference stimulation
One recently introduced non-invasive and noncontact transcranial deep brain stimulation method uses interference of electric fields. In this technique, two high-frequency (typically in kilohertz (kHz) range) electric fields that have a small frequency difference (∆f ) are delivered into neural tissues (figure 6(a)). The envelope of the interfering fields follows the small beat frequency ∆f, while the net field still has high-frequency ( figure 6(b)). In such case, neurons do not respond to kHz-frequency net field, but the neural activity is modulated with the beat frequency ∆f at a stimulation spot where two waves interfere constructively (figures 6(c)-(e)). The reason for this temporal preference was initially ascribed to intrinsic low-pass filtering property of neurons that prevents responding to kHz frequencies [87], and the envelope demodulation by neural membrane that causes neurons to react to the beat frequency [88]. Later studies emphasized that the envelope demodulation is a result of nonlinear membrane dynamics such as rectification [89] and that rectification needs to precede low-pass filtering for not losing the low-frequency content of the interfering fields [90]. Another work further adds on the nonlinearity explanation by addressing the current imbalance between inward and outward currents when there is an envelope-modulated electric field [91]. That imbalance was essentially the outcome of the gating time constant differences between the fastresponding sodium channels and slow-responding potassium channels that results in slow depolarization due to the inward sodium current and eventually leads to action potential firing [91].
This technique allows stimulation of deeper structures without affecting more superficial tissues because the vectorial sum of the two fields determines the strength of the envelope at a point, which can reach maximum at deeper parts of the brain. Indeed, the initial report by Grossman et al showed the transcranial stimulation of mouse hippocampus without activating the superficial cortical structures [88]. Moreover, the electric field distribution in the brain, i.e. the potential site of activation can be controlled by simply adjusting the electrode currents without changing the positions of the electrodes (figure 6(c)) [88].

Transcranial magnetic stimulation (TMS)
Although the physical principle of mutual induction had been elucidated during 1830s by Michael Faraday, the research on the stimulating effect of magnetic waves on biological tissues accelerated in the 1980s after groundbreaking experiments of Barker et al that reported the first-time non-contact and non-invasive stimulation of human motor cortex via magnetic field. Barker et al observed movements of hand or leg of subjects when large, pulsed currents are induced in the coil that is placed above the scalp. This pulsed alternating current creates a rapidly changing magnetic field, which in turn induce electric current in the tissues, evoking action potentials that provoke muscle twitches [92]. After that pioneering work by Barker et al, TMS became a widespread method among many laboratories and clinics for both research and therapeutic purposes.
TMS can be applied as a single pulse or for a certain duration with a repetition frequency. Repetitive TMS (rTMS) is useful for modulating the neural activity for longer times, thus, suitable for treating brain disorders [93,94]. The potential use of rTMS for the treatment of neurological disorders such as major depression, Parkinson's disease, schizophrenia, epilepsy, stroke was investigated by numerous works [95][96][97][98][99][100]. Today, TMS is an FDA-approved technique for the treatment of major depression disorder [101]. Figure 7 shows the operation mechanism of TMS. The time-varying magnetic field is typically produced in the form of pulsed fields. The strength of these fields ranges between 1 and 10 T, which lasts for few milliseconds or less. The electrical circuit generating such fields essentially consists of a parallel RLC circuit. The capacitor rapidly charges with high voltage (peak values around 400 V-4 kV) and discharges by providing high current (peak values around 5 kA-50 kA) to the inductor coil. This continuous high voltage-high current switching leads to the generation of a changing magnetic flux, which induces current in the stimulation site.
The fast kinetics of the stimulator device is important to achieve more efficient neuromodulation. The electrical model of a neural membrane is simply a parallel RC circuit [103]. The time constant of that RC circuit and the duration over which the charges are injected determines the amount of charge collected by the membrane capacitance. If the stimulation pulse duration is shorter than the membrane time constant, membrane capacitance holds basically all the charges, leading to the elevation of membrane potential without leakages. However, if the pulse duration is long compared to the membrane time constant, a certain amount of ionic leakage current occurs through the membrane resistance, which slows down the depolarization of the membrane [103].

Mechanical stimulation
Many different ion channels are sensitive to mechanical forces [104,105]. The conformational changes induced by the mechanical forces on the neural membrane can increase the conductance of mechanosensitive ion channels, thus, modulate the electrical activity of neurons (figure 8). These forces can take different forms like expansion, compression, and bending and the sensitivity of membrane to each type of force is different. Mechanical deformation of the membrane leaflets can change the area and thickness of the membrane. In this case, membrane capacitance is changed and a capacitive current will be generated, which is another possible mechanism for modulating membrane potential [106]. Mechanical stimulation has been successfully used for controlling neural activity in both in vitro and in vivo conditions. Applying focused ultrasound to neural structures is a more conventional way compared to transducer-mediated mechanical stimulation using photoacoustic or magnetomechanical nanotransducers, although there is still not a consensus on the exact mechanism of FUS.

Focused ultrasound stimulation (FUS)
When ultrasound waves are delivered into tissues, the mechanical forces cause nanoscale vibrations in the focused spot. Depending on numerous factors, those vibrations result in thermal and/or mechanical neural coupling. Most FUS neuromodulation studies use low or intermediate-level ultrasound intensities that induce negligible temperature elevations in the tissues (e.g. brain) [107][108][109][110][111]. Thus, mechanical processes are thought to be the primary governing factors for stimulation or inhibition of neural activity via ultrasound, although thermal neuromodulatory effects cannot be ruled out completely especially in peripheral nerve studies, where higher acoustic intensities have been used [112]. The exact form of the mechanical processes, however, is still unknown. The theories mostly focus on the mechanical interaction of ultrasound with the cellular membrane and the resulting transient currents due to the conformational changes of the bilayer membrane or mechanosensitive ion channel structures.
Krasovitski et al proposed that the activation of mechanosensitive channels and/or alteration of membrane permeability occurs due to intramembrane cavitations, which are created by consecutive inflation and deflation of the bilayer membrane leaflets upon exposure to acoustic negative pressure and positive pressure, respectively [113]. The follow up modeling study by Plaksin et al posits that under sonication, continuously deforming intramembrane cavitations facilitate oscillatory variations in the membrane capacitance and induce capacitive displacement currents ( figure 9(a)) that hyperpolarize the membrane. On the other hand, fluctuating leak currents through the voltage-independent channels cause slow charging of the membrane. These two imbalanced ionic currents gradually lead to action potential firing because of charge accumulation on the membrane [114,115].
Another main theory is based on the activation of mechanosensitive ion channels [107,108,[116][117][118]. A recent study investigated this in detail and reported that the excitation of neurons via focused ultrasound occurs through the opening of calcium-selective ion channels that are activated by mechanical forces on the neural membrane [119]. Activation of mechanosensitive ion channels leads to calcium influx. Increase in the intracellular calcium activates nonselective TRPM4 channels, which results in sodium influx. As a result, depolarization of the membrane activates T-type calcium channels, which further causes calcium influx. Thus, activation of mechanosensitive channels and resulting calcium influx is amplified via indirect activation of other ion channels, which altogether results in stimulation ( figure 9(b)). This study also showed that the previously proposed mechanisms such as temperature changes, cavitation formation, and membrane deformation do not have a significant effect on the stimulation mechanism.
The seminal work by Tyler et al comprehensively investigated the influence of low intensity low frequency focused ultrasound (LILFU) on neural activity by conducting a detailed analysis of the effect of LILFU on Ca 2+ and Na + transients, and synaptic vesicle and transmission dynamics [107]. Later, the same group reported the first time application of transcranial neuromodulation, in which they successfully showed the safe stimulation of motor cortex and hippocampus of intact mice by transmitting a low intensity low frequency pulsed ultrasound through the skull with less than 10% attenuation [108,120]. Another study by Legon et al demonstrated the non-invasive stimulation of somatosensory cortex of intact human brains via low intensity transcranial focused ultrasound, being the first demonstration of intact human brain modulation with focused ultrasound [121].
The typical frequencies used in the welldeveloped ultrasound diagnostic imaging technology ranges between 1 and 15 MHz. Choice of frequency ( f ) influences the attenuation of the ultrasound wave in the medium. The dependency of attenuation coefficient (α) to frequency in tissues is described with α = α 0 f n , where n varies between 1 and 1.5. Therefore, higher frequencies will be more attenuated, meaning the penetration depth will be lower compared to lower frequencies. Besides, higher attenuation might also generate undesirable thermal effects in the tissues.
The applications targeting deeper structures and involving major barriers such as skull use low frequency ultrasound, typically lower than 1 MHz, to effectively reach the aimed spot. Thus, most in vivo studies, especially brain stimulation, were conducted with frequencies lower than 1 MHz. Indeed, both experimental and theoretical studies proved that the optimum frequency for the highest ratio of transmission to absorption for skull is less than 1 MHz [122,123]. This leads to a trade-off between acoustic frequency and intensity, which was addressed several times in the literature [108,124].
While low frequencies are favorable owing to better transmission through biological structures, high frequencies are advantageous in terms of providing superior spatial resolution. Thus, brain stimulation and most other in vivo applications typically suffer from low spatial resolutions. For example, in the frequency range that is best transmitted through the skull, 0.5-1 MHz, the lateral spatial resolution is on the order of 2-5 mm [125]. To address this problem, Ye et al reported effective in vivo stimulation of mouse brain with frequencies up to 2.9 MHz, while confirming the previous observations on the requirement of more intensity for higher frequencies [126]. For the more accessible parts of the body, such as retina, higher frequencies might be more effective. For example, Menz et al reported the reproducible activation of salamander retinal ganglion cells (RGCs) via atypically high 43 MHz focused ultrasound, which provided an unprecedented spatial resolution of ∼90 µm [110]. Jiang et al's work on rat retina with more moderate frequency of 2.25 MHz led to spatial resolution of 1.6 mm [127].

Photoacoustic stimulation
Transcranial ultrasound that is mostly used for modulation has a drawback of reflecting from bones causing the loss of spatial resolution as well as reaching to cochlea, which may change the stimulation pathway [128,129]. Alternatively, optoacoustic provides local stimulation of the targeted neurons and the generation of broadband acoustic waves at ultrasonic frequencies is based on pulsed light absorption that induces non-radiative processes like thermal expansion and transient heating [130].
Shi et al [131] developed a tapered fiber optoacoustic emitter (TFOE) generating an ultrasound field with high spatiotemporal resolution ( figure 10(a)). To offer highly efficient optoacoustic signal generation and conversion efficiency, they selected carbon nanotubes (CNTs) embedded in a polydimethylsiloxane (PDMS) matrix ( figure 10(b)). The architecture composes of multi-wall CNTs with strong light absorption embedded in PDMS with a high thermal expansion coefficient to efficiently transduce optical energy into acoustic waves. They showed the activation of a single neuron owing to the effective field size of 39.6 µm and high temporal response converting a 3 ns single laser pulse into acoustic stimuli (figure 10(c)) [131]. Moreover, this design is also applicable to patch-clamp measurement systems contrary to the previous ultrasound based stimulators since the ultrasound field easily disrupts the gigaohm seals during recording [107]. Likewise, inspired by the developing optoacoustic materials and their use in translational medicine, recently Jiang et al showed the optoacoustic stimulation both in culture and in vivo in a functional brain using a fiber-optoacoustic converter with submillimeter spatial precision ( figure 10(d)) [132]. The two layer architecture is composed of a diffusion layer of ZnO NPs and epoxy, and the outer absorption layer of graphite-epoxy mixture ( figure 10(e)). While the diffusion layer enables Rayleigh scattering of the light, the graphite absorbs the scattered light and generates heat. The generated heat leads the expansion and compression of the epoxy, which generates the acoustic waves omnidirectionally [132]. Although the main physiological mechanism is not fully investigated, the spatially confined acoustic waves generate intracellular Ca 2+ flux leading the activation of neurons ( figure 10(f)). The acoustic wave effect is spatially confined within 1 mm due to attenuation with the distance, which confines the effective stimulation radius with 500 µm (figures 10(g) and (h)) [132]. Thus, the miniaturized design ensures the high spatial confinement of the neural stimulation.
In conclusion, optoacoustic stimulation may offer single-cell resolution neural stimulation with polymeric and nanomaterial based architecture, which can also be used in assistance with electrophysiological or imaging systems such as magnetic resonance imaging (MRI) owing to their nonmagnetic nature. However, the main physiological mechanisms behind optoacoustic stimulation of neurons require more in-depth electrophysiological studies.

Magnetomechanical stimulation
Magnetic energy can be converted into mechanical forces or torque via magnetic micro/NPs [133]. Those mechanical effects can then alter the electrical activity of neurons via different routes such as mechanosensitive ion channel activation or receptor clustering [134]. Activation of mechanosensitive ion channels requires application of force, which should be on the order of piconewtons [135]. Magnetic micro/nanoparticles can apply such forces on the mechanosensitive ion channels when they are exposed to externally applied magnetic fields. Hence, the electrical activity of neurons can be modulated using magnetic NPs that are positioned near or on the mechanosensitive ion channels over the cellular membrane [136][137][138].
Gregurec et al reported that magnetic vortex nanodiscs with different diameters can stimulate primary neurons by exerting a torque on mechanosensitive ion channels under weak and slowly varying magnetic fields ( figure 11(a)) [139]. In vortex nanodiscs, the dipole-dipole interactions are minimized because of the negligible magnetic moment of the vortex state, and this yields improved colloidal stability [140], which motivates their use in bioapplications. The magnetization state of the nanodisc ( figure 11(b) top), i.e. vortex, in-plane, out-ofplane, can be controlled by applying a weak magnetic field ( figure 11(b) bottom). The transition between magnetic states generates a torque on the neural membrane proportional to the magnetic moment of the nanodisc and applied magnetic field intensity. After the incubation of dorsal root ganglion (DRG) neurons with the magnetic nanodiscs, nanodiscs positioned near the neural membranes stimulate the DRG neurons via the forces applied on the mechanosensitive ion channels that cause calcium influx ( figure 11(c)). The observation that larger magnetic nanodiscs applied greater torques on the cellular membrane was reflected in the greater calcium influx induced by 226 nm nanodiscs (figure 11(c) left) compared to 98 nm nanodiscs (figure 11(c) right) [139].
Recently, Lee et al demonstrated an alternative magnetomechanical stimulation platform that can stimulate mechanosensitive PIEZO1 channels and induce behavioral changes in freely moving mice [141]. The system includes a magnetic torquer nanostructure (called as m-Torquer in the study) and a circular magnet array (CMA) that generates a uniform, rotating weak magnetic field to cause torque force generation in m-Torquer that would stimulate mechanosensitive ion channels ( figure 11(d)). Notably, optimization of the number and positioning of the magnets in the CMA resulted in an effective working range of 70 cm for the m-Torquer toolkit, which is similar to the bore size of clinically-used MRI instruments. The effectiveness of m-Torquer system was first demonstrated in vitro by temporally precise and reproducible stimulation of Piezo1-expressing neurons ( figure 11(e)). To verify the in vivo performance of the system, m-Torquer nanostructures were first injected into the right hemisphere of Piezo1expressed mice motor cortex. Application of magnetic field via rotating CMA resulting in enhanced motor function leading to greater travel distance for mice compared to control case without CMA application ( figure 11(f)). Moreover, m-Torquer nanostructures injected into the deep brain regions of mice can stimulate Piezo1-expressed neural tissues when exposed to rotating CMA. Figure 11(g) shows that m-Torquer injected right hemisphere regions of hypothalamus displays higher c-Fos expression, indicating the successful deep brain stimulation with m-Torquer [141].
Such studies indicate the potential of magnetomechanical systems for noninvasive in vivo and animal studies on neurostimulation. As in magnetoelectric stimulation, the tunability of nanotransducers and magnetic field properties provides versatility to this neurostimulation modality. The faster response times of magnetomechanical systems compared to magnetothermal platforms is also an advantage for achieving superior temporal control.

Thermal stimulation
Temperature changes near the cellular membrane can modulate the electrical activity of neurons via two fundamental mechanisms. First, rapid temperature fluctuations cause variations in membrane capacitance, which induces capacitive currents that excite neurons. This thermocapacitive pathway was shown to be the mechanism behind pulsed infrared neural stimulation (INS) by the seminal work of Shapiro et al [142]. This finding later triggered several more studies showing successful control of neural activity with high temporal and spatial resolutions using metallic nanostructures [143], organic polymers [144], and inorganic nanowire-graphene complexes [145]. Although thermal transients were shown to change the membrane capacitance in these studies, a more detailed biophysical understanding of the thermocapacitive mechanism confirmed that the membrane expands under temperature elevation and this results in a capacitance change proportional to the rate of temperature change ( figure 12(a)) [146]. However, this study further emphasized that the resulting membrane current is not only due to the capacitance change; membrane surface-charge related potential variations resulting from intracellular and extracellular charge accumulation upon heating also generate displacement currents, which contributes to the total thermocapacitive effect ( figure 12(b)).
Local thermal variations can also activate temperature-sensitive ion channels when the temperature exceeds or falls below certain thresholds [147,148]. These ion channels play a crucial role in the operation of sensory neurons that sense high and low temperatures, i.e. perform thermoreception [149]. Their sensitivity to temperature makes them an ideal candidate for thermal stimulation of neurons. TRPV1 is one of the best characterized heat-sensitive ion channels that is activated at high temperatures (typically above 42 • C) [150]. Many neuromodulation studies focused on activating TRPV1 channel by generating a local heating near the cellular membrane through photothermal (figure 12(c)) or magnetothermal (figure 12(d)) nanotransducers that are either attached or positioned in close proximity to the membrane [151][152][153][154]. Activation of temperaturesensitive ion channels leads to calcium influx, which triggers action potential firing by depolarizing the membrane.

Infrared neural stimulation (INS)
After the first studies using pulsed mid-infrared lasers to stimulate neurons [155,156], lasers have gained significant attention as the stimulation sources due to their high spatial resolution, the ease of coupling with optical fibers and predominantly well-developed pulsed laser technology. The main principle behind INS is the generation of localized heating due to absorption of the pulses of infrared light by water ( figure 13(a)) [157], which generates depolarization in the cell membrane. Rapid heat transients are required to change the membrane potential, which is effectively observed in pulsed, but not in continuous wave operation [158]. INS eliminates the need for genetic manipulation, any electrical or electrochemical junction between the stimulator and the tissue  ( figure 13(b)), which reduces the scar tissue formation due to mechanical mismatch on the devicetissue interface. Moreover, INS limits the stimulation area by the spot size of the laser [159] and does not cause additional electrical noise on the recordings. The main drawbacks of INS are the careful choice of stimulation parameters such as the repetition rate and pulse energy [160], which significantly varies for different tissues, and can lead to potentially harmful artifacts due to heat accumulation in the tissue. As the most common illumination wavelengths for INS are in 1400-2200 nm infrared window, optical penetration depth down to deep neural networks in tissue is limited ( figure 13(a)) [161]. Moreover, possible neural prosthetics utilizing INS require localized light sources particularly lasers, which limits the wide range of potential clinical trials, daily use, and availability. Therefore, the most promising idea for next generation neural stimulators to eliminate any genetic manipulation is the use of mediator devices to convert optical energy of photons to electrical, chemical, thermal or acoustic energy for stimulation of neurons.

Photothermal NPs
Remembering the idea behind INS, local temperature changes are also used in optoelectronic biointerfaces for photothermal stimulation [162]. The stimulation can be generated by heat-sensitive ion channels [163], resulting higher ionic currents [164] or due to thermocapacitive [142] effect by increasing cell membrane capacitance. The increase in local temperature can trigger heat-sensitive ion channels such as the TRPV-family, and generates higher ionic currents by potassium inward rectification for example using quinacridone nano-structures with 5 mW mm −2 , 100-800 ms illumination [164]. However, activation of heat-sensitive ion channels and corresponding stimulation require hundreds of milliseconds, which is not sufficient for high frequency applications. Another mechanism takes place when rapid heating in shorter time scales, a few ms, which is called thermocapacitive stimulation was discovered in the study Shapiro et al and studied in-detail afterwards [142,165,166]. This mechanism and resulting membrane depolarization depend on the rate of temperature change that can be generated by short, high intensity light pulses. Moreover, since the temperature change can be confined in a region of interest without the need for temperature rise in the whole surrounding medium reducing heating based potentially harmful effects. For this purpose, semiconducting NPs have drawn significant attention to generate neural stimulation with high spatial control as well as for selective stimulation [167].
In this context, gold NPs and nanorods (Au NPs and Au NRs) are particularly useful due to their plasmonic nature as well as due to size matching with the ion channels and cell receptors [36,170,171]. Upon illumination with visible-NIR light, the conduction electrons in Au NPs generate localized surface plasmon resonance, which is a resonant and coherent oscillation [172]. Then, this excited electrons dissipate their energy by the lattice collisions and induce thermal energy, which activates thermosensitive ion channels (figures 14(a) and (b)). By engineering the inter-particle distance and surface morphology, plasmon absorption peak can be tuned to the tissue transparency window of 600-1200 nm [170]. For instance, 20 nm spherical Au NPs can be bio-conjugated with Ts1, a neurotoxin blocking the voltage-gated sodium channels, to bind transient receptor potential vanilloid member 1 (TRPV1) channels with high localization [143] ( figure 14(c)). This high localization is achieved through binding neural membrane proteins, which avoid the diffusion of the NPs and corresponding need for high concentration of NPs. Moreover, when NPs are washed out their effect almost diminishes (figures 14(d) and (e)). Although the main effecting mechanism is based on photothermal effects just like in INS, optical tunability of the Au NPs enables the neural stimulation in visible wavelengths. The depolarization induced by the Au NPs depends on membrane capacitance change rather than the ion-channel openings (figure 14(e)) [143]. Therefore, faster membrane depolarization can be achieved. Another advantage of Au NPs is the lower heat dissipation into surrounding environment. Direct heating with IR illumination and INS result in relatively large absorption regimes and larger solution volumes. However, Au NPs only heat their immediate environment, which is also beneficial for cooling thanks to shorter heat diffusion distance. Moreover, there are several studies demonstrating that the magnitude of the depolarization depends on the derivative of the temperature but not on the actual temperature value. Therefore, desired depolarization levels can be achieved with short heating intervals without reaching the critical temperatures that potentially damaging the cells. Advantageously, conjugated Au NPs can be injected to the targeted tissue such as eye. Photostimulation of RGCs is targeted to overcome photoreceptor degenerative diseases [143]. Conjugated Au NPs bonded to the RGCs through injection can excite the RGCs by eliminating the need for photoreceptor cells which are inoperative in age-related macular degeneration or retinitis pigmentosa. However, there are several limitations for this method. The need for high power light source is not feasible for human prosthesis and the penetration depth of the light sources in visible window is not adequate for brain tissue [167].
Although photothermal processes have been widely investigated for neural activation, there are only a few studies concentrated on the neural inhibition possibilities. The main proposed mechanism for neural inhibition is based on slow and prolonged heating near the cell membrane, which activates hightemperature sensitive ion channels that induce membrane hyperpolarization and inhibit neural activity [173][174][175].

Magnetothermal stimulation
Superparamagnetic NPs generate local heat energy when they are exposed to alternating magnetic field due to the Neel-Brown relaxation [176,177]. The produced local heat can be used to modulate the conductance of temperature-sensitive ion channels on cellular membrane, which enables controlling the electrical activity of the cells. Indeed, several studies showed the effectiveness of magnetothermal systems on deep brain stimulation [153], evoking behavioral changes [178], and treatment of neurological disorders [154].
The pioneering study by Huang et al used superparamagnetic manganese ferrite (MnFe 2 O 4 ) NPs that were targeted to temperature-sensitive TRPV1expressed cells to evoke calcium influx through TRPV1 when NPs were stimulated via RF magnetic field ( figure 15(a)) [179]. The NP heating via magnetothermal effect elevates the local temperature and causes calcium influx in human embryonic kidney cells (HEK 293) at similar levels to capsaicin-induced calcium influx ( figure 15(b) left). The increase in local temperature can also evoke action potentials in hippocampal neurons ( figure 15(b) right). The same study also showed behavioral responses in Caenorhabditis elegans worms [179], paving the way for further studies focusing on in vivo magnetothermal neurostimulation. Stanley et al demonstrated that calcium-sensitive insulin release can be increased by magnetothermal activation of TRPV1 channels in vivo. This was achieved via two different routes: by injection of magnetic iron oxide NPs into the tumors expressing TRPV1, or by genetically engineering cells to induce intracellular NP synthesis. Although the former method showed higher effectiveness, engineering cells for endogenous synthesis provides a continuous source for the production of NPs [180]. The same group reported that genetic engineering of cells that allow intracellular synthesis of magnetic NPs can be used in remote regulation of glucose homeostasis in mice via both thermal and mechanical activation of TRPV1 [181], and the regulation of feeding and metabolism by stimulation and inhibition of glucosesensing neurons in hypothalamus [182].
Magnetic NPs can thermally stimulate deep brain structures of mammalians if the neurons in those structures express temperature-sensitive ion channels [153]. The robust expression of such ion channels is a critical step in the in vivo experimental scheme ( figure 15(c)). The confocal image in figure 15(d) shows a successful expression of TRPV1 in ventral tegmental area (VTA) of mice. The quantification of c-Fos expression in the different deep brain regions, where TRPV1 ion channels were expressed and magnetic NPs were injected, indicates successful magnetothermal stimulation of deep brain structures of mice under magnetic field (figure 15(e)). Although the latency of stimulation was significantly improved in this study [153] compared to the previous one [179], the reported 5 s latency still needs to be improved for achieving high frequency neurostimulation (tens or hundreds of Hz). The in vivo success of magnetothermal stimulation led to demonstration of behavioral changes in awake mammalians as well. Using the experimental setup shown in figure 14(f), Munshi et al reported magnetothermal genetic stimulation platform that can induce motor behaviors in freely moving mice [178]. The magnetic fieldinduced thermal stimulation of TRPV1-expressed three different deep brain regions, motor cortex, dorsal striatum, and the ridge between ventral and dorsal striatum, resulted in behavioral changes in awake mice (figures 15(g)-(i)) that are similar to the reported behavioral changes with optogenetic and chemogenetic stimulation of same deep brain regions [178]. A recent report revealed the potential of deep brain magnetothermal stimulation as a therapeutic tool as well. The remote control of electrical activity of neurons in the subthalamic nucleus of mice via magnetothermal effect mitigates the motor symptoms related to Parkinson's disease [154].

Perspectives and conclusions
The critical aspects of different electrical, mechanical, and thermal neuromodulation techniques discussed in this review are summarized in table 1.
A multimodal approach can be used by simultaneously injecting nanotransducers for photoelectric, magnetoelectric and piezoelectric stimulation at different depths from the surface of the nerve tissue. While photoelectric nanotransducers can be positioned close to the superficial part of the tissue for stimulation with high spatial resolution, to reach deeper parts acoustic waves can be simultaneously used and focused on piezoelectric nanotransducers. Even though the spatial resolution is not as good as optical waves, acoustic waves can be localized better than the magnetic waves. For much deeper parts magnetoelectric nanotransducers can be activated. With such an approach, the parts of the tissue at different depths can be simultaneously controlled. Moreover, the design of the transducers has also significant room for improvement. For example, for the photoelectric stimulation the introduction of the local return electrodes instead of a global return facilitates a significant improvement in the field confinement [198]. Such novel designs that are valid at microscale can be also translated to nanoscale transducers as well.
From a material science perspective, if nanotransducers would be composed of fully biocompatible and biodegradable content, they can be safely  [190] Not required NA FDA-approved treatment of depression [191] Mechanical FUS Acoustic wave Capacitive membrane current [114], mechanosensitive ion channels [119] Not required NA Nonhuman primate [192] and human deep brain stimulation [193] Photoacoustic Light Capacitive membrane current [131], mechanosensitive ion channels [194] Nanomaterial injection, light source implantation Nanoscale, microscale Mouse motor cortex stimulation [194] Magnetomechanical Magnetic wave Mechanosensitive ion channels [139,141] Micro/ nanomaterial injection Nanoscale, microscale Mouse deep brain stimulation [141] Thermal INS Light Capacitive membrane current [142], surface charge accumulation [146] Light source implantation NA Human brain stimulation [195] Photothermal Light Capacitive membrane current [143,145,196], temperaturesensitive ion channels [151,152,197] Nanomaterial injection, light source implantation Nanoscale Mouse deep brain stimulation [197] Magnetothermal Magnetic wave Temperaturesensitive ion channels [153,154,178] Nanomaterial injection Nanoscale Mouse deep brain stimulation [154] absorbed by the tissue and degraded in the body after they finish their function [199]. We consider that such a biodegradability is an important requirement for nanoscale transducers since their removability from the body is not as easy as in the case of a macroscale implant. On top of the biodegradable nanotransducers, optical stimulation with high spatial resolution is possible by using biodegradable optical fibers that can reach deep inside the targeted tissue [200]. In recent years, implantable microlight-emitting-diodes (µLEDs) gained popularity for performing deep tissue photostimulation with wireless optogenetic systems [201][202][203][204][205]. Using µLEDs together with wireless power transfer and wireless communication units, implantable wireless optoelectronic systems allow optogenetic neuromodulation of freely-behaving animals [206,207]. Moreover, upconversion NPs can convert NIR-light to visible light and this phenomenon can be used for improving the effective range of light-based systems [45]. From a fundamental perspective, electromagnetic, mechanical, and thermal energy can be interplayed to find novel routes of neurostimulation. One recent approach is sono-optogenetics, which uses mechanoluminescent NPs that are delivered to deep brain and then activated via focused ultrasound to emit visible light that stimulates genetically modified cells [208,209]. Another example combines photoelectric mechanism with temporal interference (TI) stimulation. In that work, pair of photocapacitor electrodes were used as wireless light-driven electric current generators at two different frequencies to perform TI stimulation without requiring wired connections to deliver the electric fields [210].
In terms of safety, off-target heating is a common concern because even small temperature changes in off-target neural tissues might result in undesired outcomes at cellular, behavioral and physiological levels [211][212][213][214], while larger variations can induce tissue damage. For example, ultrasound can stimulate off-target auditory structures and light can activate off-target neurons in its pathway during optical neuromodulation [128,215]. To prevent such unintended effects, different focusing strategies and power distribution techniques have been developed to minimize the energy reaching to off-targets, while delivering sufficient energy to the stimulation tissues [216,217]. In case of substrate-based stimulation electrodes, the mechanical compatibility of substrate to the implanted tissues is vital for minimizing the tissue damage. Soft substrates like parylene-C can provide conformable contact with tissues, enabling fabrication of cuff electrodes that surrounds the stimulation tissue [187]. Other types like PDMS and silk can be used for building foldable prosthetic devices or as a soft substrate that embeds nanotransducers [218,219]. It is also possible to embed tiny electronic, chemical, and/or mechanical components into soft substrates that are implanted or injected into the tissues for performing multimodal neural operations [220,221].
In terms of clinical use, although optogenetics is the predominant technique in research settings, its clinical translation is hindered by ethical concerns and technical limitations [222]. Currently, one optogenetics treatment is under clinical trials for restoring vision of blind patients and the results of the treatment led to partial vision restoration in patients [223]. As a nongenetic alternative, photovoltaic retinal prosthetic devices showed successful performance in clinical trials [186]. Deep brain stimulation and direct cortical stimulation are in common clinical use despite of their invasiveness. TMS and tDCS (and other tDCS modalities) stand out as the most common non-invasive methods used for clinical therapeutic purposes [224,225]. Besides, FUS is the newest promising method translated into human research and has a growing interest [121]. The potential of nanomaterials for clinical neuromodulation is still under investigation [226]. The therapeutic micro/nanomaterials or implantable prosthetic devices need to offer noninvasive removability from the body for potential human use. Similarly, irreversible geneediting operations may raise concerns and reversible genetic modification strategies can be valuable for clinical applicability of genetic techniques [227].

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).