Genetic Transducers

Advances in genetic engineering have enabled optical stimulation or recording of neurons with high specificity and selectivity.^{47, 70} Optogenetics involves genetically modifying neurons to make them light sensitive for photic stimulation or to generate photic signals for measuring neural activity. For photic stimulation, neurons are genetically altered to express light-responsive channel proteins in the plasma membrane. The protein channels, originally discovered in photosynthetic algae and bacteria, open in response to light, permitting ions to flow into or out of the cell, thus triggering changes in the neuronal membrane potential.⁷¹ Depending on the specific protein and the corresponding ionic flux, light activation either excites or inhibits the neural action potential. Since its first demonstration in neurons in 2005⁷², optogenetics has expanded rapidly and advanced in lock-step with improvements in microbial opsin engineering, genetic methods for cell-type targeting, and optical strategies for guiding light through tissue⁷¹. One critical advantage potentially afforded by optogenetics is high cell-type specificity based on promoter selection. Cell-type specificity enables multiplexing based on wavelength, differential activity (firing vs. inhibition), specific neurotransmitter production, and other cell type-specific functions.

Early demonstrations with microbial opsins produced millisecond channel opening, thus enabling control of single spikes and even synaptic events.⁷² Despite such high temporal resolution, the early efforts suffered prolonged deactivation and rapid inactivation, which produced extra spikes, plateau potentials, and limited the firing rate to ~40 Hz.^{73, 74} Recent improvements in temporal precision and the ability to sustain higher frequency firing resulted from engineering microbial opsin variants with faster deactivation kinetics.^{73, 74} On the other end of the temporal spectrum, stabilized step function opsins (SSFO) permit on-off switching with a brief pulse of blue light to activate for up to 30 minutes, and deactivation by yellow or green light, thus enabling long timescale experiments.⁷⁴

Another genetic tool for visualizing and recording neuronal signals optically is through genetically encoded indicators. The two classes, Genetically Encoded Voltage Indicators (GEVIs) and Genetically Encoded Calcium Indicators (GECIs), both require transduction of an endogenous signal to a fluorescent indicator. These are notoriously difficult to calibrate due to nonlinearities of fluorescence. Typically, these include an analyte-binding or sensor domain combined with a reporter element containing a fluorescent protein.⁴⁷

GEVIs offer unique capabilities for large-scale recording⁴⁷ and exhibit better spatial resolution than traditional electrical recordings⁷⁰. However, current GEVIs are dim, resulting in only moderate sensitivity. Rhodopsin variants have made some improvement, but require further modification and will likely need to be paired with brighter retinal chromophores.⁷⁵ Another factor limiting sensitivity is that they must be paired with additional imaging techniques to visualize the signal. Unfortunately, GEVIs cannot be paired with two-photon imaging and must be performed with wide-field microscopy, thus limiting the SNR. One report of two-photon imaging with a rhodopsin variant exists, but required extensive trial averaging to detect the signal.⁷⁶

GEVIs typically exhibit poor kinetics⁷⁵ on timescales greater than 1 ms⁷⁷ leading to moderate temporal resolution. Protein engineering has incrementally improved the kinetics, but ultimately the tradeoffs between brightness, kinetics, and SNR require continued optimization. High-performance GEVIs have yet to be demonstrated in awake and mobile animals, although high SNR and fast Voltage-sensitive Fluorescent Proteins (VSFPs) have been applied in animal models.⁴⁷ These consist of a voltage-dependent conformational change of a voltage-sensor domain coupled to a pair of fluorescent proteins.⁷⁷ GEVI’s are also limited by their need to remain contained in the cell wall rather than the cytosol.⁴⁷

GECIs are the most widely used genetically encoded indicator for in vivo imaging⁴⁷ and were first reported in 1997⁷⁸. Despite their success over GEVIs, calcium indicators serve only as proxies for action potentials. Ca²⁺ ions can only enter the neuron through a subset of channel proteins, such as NMDA and nicotinic receptors. Therefore, Na⁺ ion influx may occur through other channel proteins to initiate an action potential without involving Ca²⁺.⁷⁸ Measuring Ca²⁺ instead of voltage also limits the temporal resolution because voltage changes rapidly during neuronal signaling whereas Ca²⁺ dynamics are much slower.⁷⁹ Extended Ca²⁺ transients can indicate an integrated signal⁷⁵ and make it difficult to discern spike rate, especially during high frequency or bursting activity. However, protein engineering has recently improved the kinetics and single action potential resolution has been reported.⁸⁰

Protein engineering has also produced bright GECIs⁸¹ with high enough SNR to exploit two-photon imaging⁸⁰. As such, these boast good spatial resolution and single action potential detection from hundreds of cells. In addition, expansion of the available color palette to at least five distinct wavelengths^{47, 75} greatly extends the specificity and experimental applications of GECIs within basic neuroscience.

Two critical SNR-limiting factors hamper optogenetics and genetically encoded indicators: expression levels and a secondary method of signal detection. Low expression levels require higher power lasers and exposure times, but this inevitably leads to photo-bleaching or cytotoxicity. In contrast, high expression levels can also result in cytotoxicity. Despite these constraints, optogenetic neural interfaces have evolved rapidly and facilitated a wealth of knowledge including cell specific roles in complex behaviors⁷⁵ and neural dynamics⁴⁷. Their applicability for human use remains unclear due to the genetic modifications required.

Direct Approaches

In contrast to the previously described methods, which require optical transducers, direct optical interrogation and stimulation can occur without an intervening transducer. Non-invasive, non-contact, label-free approaches are appealing for their lack of direct physical contact with the neural tissue, thereby significantly limiting the technical challenges associated with biocompatibility that typically plague traditional, invasive approaches. In vivo direct optical approaches use near infrared (NIR) or infrared (IR) signals due to their transmissivity through water and tissue, while the penetration depth’s dependence on wavelength provides a potential means for targeting fascicles.^{82, 83} Despite the high spatial resolution afforded by focused illumination and the transparency of tissue to NIR and IR wavelengths, optical approaches are still limited to penetration depths on the order of ~10² µm, limiting their use to research applications.^{84, 85} Infrared neural stimulation⁸⁴ and refractive index-based neural recording⁸⁶ represent two non-invasive, direct optical approaches.

Direct optical recording of neural activity typically probes local changes in the membrane’s optical properties when active. Such changes, which manifest as perturbations to the refractive index^{79, 86}, have been probed using various optical scattering, birefringence, optical coherence tomography, or similar techniques.^87-90 Other approaches may also relay neural activity by exploiting nonlinear optical processes, such as Raman spectroscopy of membrane lipid components⁹¹ to monitor perturbations to the lipid vibrational modes as the lipid bilayer accommodates the membrane potential activation. A critical technical challenge to directly interrogating neural activity using optical approaches rests with the low sensitivity, thus requiring signal averaging to increase SNR, which fundamentally limits the temporal resolution. Optical system-level optimizations may mitigate the SNR technical challenges for example, searching for approaches to minimize the optical source noise.⁹² Spontaneous Raman scattering approaches cannot measure neural activity in vivo due to the damage-inducing power required to produce a measurable Raman signal. Optical coherence tomography (OCT) or optical coherence microscopy (OCM) approaches are appealing for their sensitivity to small perturbations to the target optical properties, especially phase-domain measurements, which more accurately and sensitively probe slight changes in nerve displacement resulting from action potential generation.⁸⁹ The in-plane resolution achievable through OCT is limited to approximately 10-20 μm, while OCM employs high numerical aperture objectives that enable 1 μm resolution. Signal integration is not the sole hindrance to high temporal resolution measurements. These techniques can be limited in their temporal resolution depending on mechanism of activity being probed by the system, such as the membrane swelling response to compound action potential generation or physical realignment of the membrane proteins.⁸⁹ Despite these limitations, as the state of the art in optical source, detector, and system design continues improving, the potential application of in vivo, real-time optical neural recording techniques seems increasingly possible.⁸⁸

Several groups have demonstrated direct IR stimulation of neural activity^{83, 87, 93}-⁹⁶, but the precise physical mechanism by which direct IR stimulation activates neural activity remains unclear. Several hypotheses exist, including thermal disruption of the membrane ion and TRPV channels, photon absorption by the membrane proteins, and perturbations to the membrane, with the consensus converging on a combination of thermal effects and the membrane capacitance stimulating nerve activity.^{87, 94, 95} Early experiments demonstrated nerve propagation to be temperature-dependent and governed by a process driven largely by temperature gradients localized to the nerve membrane, which may increase the Na⁺ ion channel conduction upon opening or activate heat-sensitive channels.^{83, 97} In contrast, more recent efforts by Shapiro, et al. suggest that the localized heat produced by IR absorption in the surrounding water increases the local temperature and reversibly increases the membrane capacitance, thereby depolarizing the cell.⁸² Along with the murky origins of thermally induced neural stimulation by IR excitation, a key technical challenge to safe, direct optical stimulation of peripheral nerves in vivo remains the tradeoff between stimulation and tissue damage energy. Early studies determined that at wavelengths coinciding with minimal tissue absorption (4 μm and 2.1 μm), stimulation thresholds are highest relative to the threshold intensities resulting in tissue damage. The authors conclude that the low absorption wavelengths offer the greatest safety margin between stimulation and damage.⁸³

Emerging approaches in both neural recording and stimulation through direct optical techniques suggest a potential avenue for achieving non-invasive, non-contact, non-destructive in vivo neuromodulation. Core to both technologies remains a thorough understanding of the mechanisms of action to inform engineered solutions that meet the spatiotemporal requirements while addressing the system-level challenges for delivering the light safely to the target.

Nanoparticle Transducers

Light incident on nanoparticle transducers produces concentrated thermal energy, and their injection to the desired stimulation site localizes the effect and enables targeted stimulation. However, the frequency of incident light affects the signal penetration depth irrespective of how deeply the transducers are injected.

Plasmonic gold (Au) nanoparticles (NPs) have been applied in vivo for many medical fields, including targeted cancer therapy and neuromodulation. Au nanoparticles can produce heat when irradiated by light tuned to their localized surface plasmon resonance – the quantization of a metallic nanoparticle’s free electron density oscillations in response to incident electromagnetic radiation. Metallic nanoparticles exhibit enhanced absorption of light tuned to their localized surface plasmon resonance, which in turn generates heat due to electron scattering and subsequent phonon-phonon interactions.⁹⁸ For cancer therapies, the localized heat produced by the excited plasmonic nanoparticles destroys tumor cells. In contrast, application of plasmon resonant nanoparticles for optically induced neural stimulation decreases the required intensity by leveraging the intrinsic enhancement produced by the plasmon resonant modes to generate heat and trigger neural activity. The plasmon resonance of these nanoparticles can be tuned by their geometry (i.e., dimensions and shape) and owing to Au’s significant interband losses in the visible spectrum, its plasmon resonances tend to be limited to red and longer wavelengths (NIR to IR)⁹⁹. These longer wavelengths coupled with Au’s intrinsic chemical stability make these nanoparticles promising candidates for in vivo use. Furthermore, their small dimensions render Au nanoparticles ideally suited for minimally invasive delivery via direct injection to the target site. Such localized delivery prevents off-target neural activity as it ensures that the resonant nanoparticles act on only the neurons in their local vicinity.

Demonstrations of localized Au NPs coupled with IR or NIR stimulation can generate neural activity at lower laser powers than direct optical stimulation alone owing to the NP-mediated thermal enhancement.¹⁰⁰ Previous efforts have also demonstrated the importance of tuning the nanoparticle plasmon resonance to the incident light to maximize effect. Spherical Au NP off resonant from the NIR laser produce less neural activity relative to Au nanorods with longitudinal plasmon modes resonant with the incident field.¹⁰⁰ The spherical nanoparticles produce less heat and therefore stimulate less activity due to less efficient (non-resonant) coupling of the incident light to the plasmonic modes. The key advantages to combining plasmonic nanoparticles with NIR or IR stimulation are the lower laser power, which mitigates the risk of tissue damage¹⁰¹, more precise nerve targeting based on nanoparticle location, and the potential to combine the NPs with binding elements to target nerve membrane. Nanoparticle-based neural interfaces are a promising technology buoyed by early demonstrations of thermal stimulation of neurons, but their clinical application requires further investigation into their safety, delivery, targeting, persistence, and efficacy. System engineering considerations such as optimization of the temporal resolution and monitoring power requirements, heat dissipation, and potential downstream activation also warrant further study.

Nanoparticle Transducers

An alternative to photothermally induced neural stimulation, magnetothermal stimulation relies on magnetic nanoparticles that locally generate heat when exposed to alternating magnetic fields (AMF), thus causing nearby heat-sensitive TRPV channels to open and activate the neuron.¹⁰² This is an appealing technology capable of deep penetration in vivo since it exploits the transparency of the body to low frequency (100 kHz – 1 MHz) AMF to heat in vivo nanoparticles with limited attenuation. In a 2015 demonstration of this novel nanoparticle-based neural interface technology, magnetic nanoparticles delivered to the ventral tegmental area (VTA) in anesthetized mice yielded reversible neuronal activation in mice with TRPV1-expressing neurons.¹⁰² The minimal cytotoxicity of these iron oxide (Fe₃O₄) nanoparticles coupled with their in vivo persistence for over a month suggest their potential for chronic use¹⁰², but given the bulky magnet required to provide the 15 kA/m AMF this technology may remain limited to research applications.

Genetic Transducers

A recent genetic approach to magnetic neuromodulation combines TRPV4 channels with a paramagnetic ferritin protein.¹⁰³ Termed Magneto, this fused construct actuates neural activity in response to applied magnetic fields that the authors hypothesize torques the ferritin protein, which in turn places a mechanical strain on the mechanosensitive TRPV4 channel.¹⁰³ In vitro and in vivo experiments successfully demonstrated activation of nerve activity in response to the applied static magnetic field (~50mT-250mT), even in awake, freely behaving animals.¹⁰³ These magnetogenetic constructs exhibit faster dynamics and generate neural activity at physiologically relevant timescales – a key challenge hindering optogenetic techniques. Still in its nascent stages, the Magneto technology represents the newest contribution to the suite of emerging, non-invasive neural interface technologies.

Direct Approaches

A deeply penetrating modality capable of beam focusing and steering, acoustic or ultrasound signals may overcome the depth and scattering limitations that plague optical approaches while achieving spatial resolutions at the millimeter-scale. Early demonstrations of acoustic neuromodulation, typically inhibitory and performed in cortex, built off prior focused ultrasound uses for therapeutic and surgical interventions.¹⁰⁴ The intensities required to elicit nerve activity are several orders of magnitude lower than the high-intensity focused ultrasound used to lesion tissue (mW/cm² vs. kW/cm²). Legon, et al, stimulated activity in the primary somatosensory cortex using transcranial focused ultrasound (FUS).¹⁰⁵ In the periphery, FUS need not be limited to long acoustic wavelengths to accommodate the high acoustic impedance of bone compared to soft tissue and can achieve greater penetration depths. While the low intensities applied for FUS neuromodulation typically fall below the 720mW/cm² safety limit, long-term and/or repeated application remains unstudied and the safety of such use will affect its tractability as a treatment for chronic conditions.¹⁰⁶ Acoustic neuromodulation has achieved centimeter-scale penetration depth coupled with focusing capabilities without requiring any in vivo components; however, latencies on the order of ~10¹ ms reveal slower kinetics compared to electrical or optical modalities.^{69, 106} The slower kinetics likely reveal some information about the mechanisms of action; however, the exact mechanism remains unclear and optimization of the parameter space a challenge.^{106, 107}

Major hypothesized mechanisms include heating, cavitation, strain-induced opening of ion channels, and neuromechanical perturbation of membrane capacitance. Localized heating triggers nerve activity due to absorption of the applied acoustic energy.¹⁰⁸ Secondary effects¹⁰⁹ beyond the elastic heating mechanism can occur with focused ultrasound, but their impact tends to be small compared to absorption. A recent hypothesized cavitation-based mechanism¹¹⁰ suggests mechanical forces generate pores in the cell membrane and the subsequent intramembrane cavitation perturbs its capacitance, which produces a displacement current that triggers an action potential. This so-called neuronal bilayer sonophore (NBLS) model has been validated in vivo in mouse primary motor cortex via front limb EMG.¹¹¹ Interestingly, membrane bilayer capacitance perturbation has previously been described by Shapiro, et al. to describe the mechanism by which IR-induced heating stimulates neurons, and the possibility that mechanical pressure may also be involved in the phenomenon is left for future study.⁸² These membrane capacitance mechanisms may converge to elucidate two diverse stimulation approaches, which could generate a more fundamental understanding of both modalities. Other hypothesized mechanisms rely on mechanical perturbations of the bilayer membrane, such as strain-induced opening on mechanosensitive ion channels^{111, 112} and ultrasound-mediated effects that alter membrane conductance due to the viscoelasticity of the neurons¹⁰⁹. Despite an abundance of hypothesized mechanisms, limited consensus exists and further detailed study of the mechanism of acoustic neuromodulation is needed.

Nanoparticle Transducers

Analogous to transducing optical stimulation via metallic nanoparticles, Marino, et al. demonstrated the use of ultrasound-stimulated piezoelectric nanoparticles to produce neural activity.¹⁰⁸ While the precise mechanism is not yet clarified, the influence of the piezoelectric barium titanate nanoparticles is not caused by local heating. The presence and absence of the nanoparticles yielded nearly identical temperature increases, but produced significant variation in Ca⁺² transient amplitude and subsequent neurite outgrowth in vitro.¹⁰⁸ As with other modalities coupled to nanoparticle transducers, the targeting, safety, and efficacy of acoustic nanoparticle transducers requires further investigation.

Acoustic neuromodulation presents an intriguing potential avenue for deep, targeted, non-invasive or minimally invasive interface with the nervous system. Early demonstrations of its efficacy for cortical and peripheral modulation can inform further efforts to elucidate the biophysical mechanism of action and ultimately to develop technologies and protocols for potential clinical use.