Acousto-optic deflectors in experimental neuroscience: overview of theory and applications

Cutting-edge methodologies and techniques are required to understand complex neuronal dynamics and pathological mechanisms. Among them, optical tools stand out due to their combination of non-invasiveness, speed, and precision. Examples include optical microscopy, capable of characterizing extended neuronal populations in small vertebrates at high spatiotemporal resolution, or all-optical electrophysiology and optogenetics, suitable for direct control of neuronal activity. However, these approaches necessitate progressively higher levels of accuracy, efficiency, and flexibility of illumination for observing fast entangled neuronal events at a millisecond time-scale over large brain regions. A promising solution is the use of acousto-optic deflectors (AODs). Based on exploiting the acousto-optic effects, AODs are high-performance devices that enable rapid and precise light deflection, up to MHz rates. Such high-speed control of light enables unique features, including random-access scanning or parallelized multi-beam illumination. Here, we survey the main applications of AODs in neuroscience, from fluorescence imaging to optogenetics. We also review the theory and physical mechanisms of these devices and describe the main configurations developed to accomplish flexible illumination strategies for a better understanding of brain function.


Introduction
A central quest in neuroscience is to unveil complex neuronal dynamics and understand their correlation to pathological conditions in vertebrate organisms [1][2][3][4][5][6] to clarify equivalent human neural mechanisms [7,8].In this framework, advanced light microscopy represents the tool of choice [9,10] to non-invasively image extended neuronal populations [11,12] at both cellular and subcellular scales [13].This approach, in combination with genetically encoded calcium indicators, as reporters of neuronal activity [14,15] (such as GCaMP [16]), paved the way for important progress in neuroscience, including high-resolution anatomical reconstruction [17], monitoring of biological sample development [18,19], and advanced functional analysis [20,21].Furthermore, recent significant advancements in molecular biology have unlocked the potential for selectively altering the activity of neurons in the brain by means of light [22].This breakthrough technique, known as optogenetics, has revolutionized the field of neuroscience [23] by enabling researchers to precisely control neural activity with unprecedented accuracy [24,25].To do this, optogenetic actuators, i.e. specific proteins, that work as light-gated ion channels [26], are expressed in genetically modified neurons.This makes these neurons photo-responsive such that they can be activated and inhibited by an external light trigger [27].This remarkable technology not only offers a powerful tool for investigating the complex functions of the brain, but it also has the potential to transform our understanding of neurological disorders, with the goal of supporting new therapeutic interventions [28].As such, the development of optogenetics has been hailed as a revolution in experimental neuroscience and has opened exciting new paths for exploring the brain's inner workings.
However, imaging and optogenetics demand advanced tools to achieve more precise and efficient light delivery to specific regions of the brain and to obtain the possibility of analyzing fast and entangled neuronal events.In particular, optical microscopists strive for the best trade-off between precision, speed, and flexibility of targeted illumination [29,30].All these features are indeed necessary to selectively image specific groups of neurons and to control vast circuits in large brain volumes at millisecond time-scale, the typical dynamics of neuronal signalling.Wide-field illumination e.g. the most straightforward way to illuminate the sample, lacks spatial selectivity, making conventional microscopes unable to generate high-contrast or high-resolution images of thick samples.Point-scanning illumination, on the other hand, can obtain high-quality images with laser scanning confocal or multi-photon microscopy [31] but this approach is usually too slow to image the entire brain.This is a common issue with most point scanning procedures where a light beam needs to pivot throughout the whole field-of-view (FOV) for a complete reconstruction.Mechanical constraints, low illumination efficiency, and slow scanning speed limit the applicability of such devices in both imaging and optogenetics.For instance, galvanometer mirrors (GMs), due to their mechanical inertia, are rarely used in photostimulation but commonly in 2D raster scans [32,33].With their slow sweeping rate, they are not suitable for rapidly accessing connected neurons spread over large volumes.Making use of resonant GMs to increase the temporal resolution of such scanners up to frame rates of a few kHz is a possible solution for fluorescent imaging, but not for delivering custom stimulation patterns, as they do not provide the necessary flexibility for arbitrary excitation trajectories.
To deal with this issue, researchers developed more advanced scanners that enable the flexibility of addressing multiple target cells without losing spatiotemporal information [34].To address preselected target regions simultaneously, illumination had to be parallelized.Particularly, several systems adopted micro-LEDs [35][36][37], digital micromirror devices (DMDs) [38,39], or liquid crystal displays (LCDs) [40] for parallel and tailored illumination.However, even with refresh rates of up to tens of kHz, these systems suffer from poor stimulation efficiency in terms of laser power delivery when targeting far-spread targets.A more efficient parallel illumination approach has been achieved with computer-generated holography, which takes advantage of liquid-crystal-based spatial light modulators (LC-SLMs) [41][42][43].However, such devices are not exempt from drawbacks.Speckle patterns are inherently generated at the sample plane by the repeated modulation of the wavefront phase, which has an impact on the precision of point illumination.In addition, cross-talk of the inter-pixel driving voltage limits the smallest available pixel size to ∼3 µm [44] which affects high-resolution illumination patterns.Therefore, only a few models provide the high refresh rate needed for advanced optogenetic stimulation (kHz).Indeed, even with continuous optogenetic stimulation, they are limited by the single actuator duty cycle (10 ÷ 100 ms), which does not match the fast ion-channel opening dynamics of many optogenetic actuators (1 ÷ 2 ms) [45].Thus, successful studies concerning the activation dynamics of neuronal circuits that require faster illumination of multiple spots are not supported.
A promising step in this direction is the use of technologies that exploit the acousto-optic effect.In these cases, the interaction between ultrasound waves and light enables unprecedented control of light at microsecond time-scales [46].For instance, tunable acoustic gradient index of refraction lenses (or TAG lenses) are today employed as varifocal lenses in remote focussing systems [47][48][49][50][51], acousto-optic tunable filters are often implemented as electronically adjustable narrow-band-pass filters [52][53][54], and acousto-optic modulators (AOMs) are used for a fast modulation of the diffracted beam intensity [55].Arguably, though, the acousto-optic method that holds the greatest potential in neuroscience involves acousto-optic deflectors (AODs).First, these devices can rapidly deflect a light beam, control its intensity, and even its focusing without any mechanical movement.This allows AODs to attain MHz-order dynamics and meet the temporal requirements for concurrent activation of spatially distributed neurons.Second, the AOD response time for deflecting the light is constant, independent of the distance between successive targets.Thus, AODs guarantee excellent accuracy and consistency in beam positioning, offering high flexibility in targeting non-sequential areas of interest, better known as random-access scanning.Finally, it is also feasible to achieve the simultaneity promised by parallel approaches, creating several independent beams with a single device.This is possible by driving a single AOD simultaneously with multiple acoustic signals.
As shown in table 1, AODs stand out as the faster and more flexible option to target different regions in a volume, compared with the other aforementioned deflectors.Thanks to their versatility, AODs are increasingly exploited in experimental neuroscience, and we dedicate this review to collecting the most representative results achieved.Here, we first outline in detail the theory of AODs' physical functioning.Then, we explore the main configurations developed to accomplish flexible illumination strategies in various optical systems.Next, we discuss imaging and optogenetic applications realized with AODs.We also illustrate what can be achieved by driving a single AOD with multiple frequencies to parallelize the illumination.Finally, we discuss the possible drawbacks of these devices and solutions adopted to face them.

Operating theory and main implementations
The AODs' basic mechanism of light deflection resides in a periodic modulation of the refractive index of their acoustic-optic medium, which typically is a crystal of high refractive index, e.g.tellurium dioxide, TeO 2 .This modulation is induced by a piezo-transducer directly bonded to the crystal and driven by an electrical signal at radio frequencies .The oscillating piezo generates a pressure wave-or sound wave-that propagates inside the medium and alters its optical properties.The induced periodicity of the refractive index defines a Bragg diffraction grating [56,57], and a monochromatic impinging laser beam will be deflected at the first-order beam angle: where λ and Λ are the light and sound wavelengths, respectively; f is the frequency of the driving signal; and ν is the soundwave propagation velocity in the medium, typically ν = 650 m s −1 in TeO 2 .A residual undiffracted beam, or zero-order beam 0th, will instead travel straight, carrying a small fraction of the incident optical power.Figure 1(a) shows schematically the beam diffraction by an AOD.
In any dynamic beam deflector, the two most important parameters are resolution and positioning speed.The maximum number of resolvable and distinct beam angles achievable by AOD light diffraction is given by the ratio of the angular deflector bandwidth ∆θ and the angular divergence of the diffracted beam ∆Φ [58]: where ∆Φ = ελ/D, with D being the incident beam's width and ε a nearly unitary factor that depends on the amplitude distribution of the incident beam (e.g.ε = 4/π for a Gaussian beam).From equations (1) and (2) it follows that: where ∆f is the frequency bandwidth of the AOD, i.e. the admitted range of acoustic frequency.It is noteworthy to mention that for non-monochromatic lasers, such as ultrafast pulsed lasers, the effect of spatial dispersion is to increase the angular spread of the beam as: This means that whenever the spatial dispersion of the beam is significantly greater than the beam divergence ∆Φ, the maximum number of resolvable points N is accordingly reduced [59].
The ratio between the beam diameter and the soundwave propagation velocity in the medium defines a notable quantity, the access time: This is the necessary time for the acoustic wave to travel through the diameter of the laser beam.In other words, it is the time required for the beam to commute from one position to another, leading to an upper boundary on the AOD speed.Therefore, the parameter expressed in equation ( 3) is also known as the time-bandwidth product for an AOD.Until now, we have only considered driving the piezo of these devices with a single and constant frequency (figure 1(d), first panel).However, the AOD's largest potential can be unlocked with several other driving configurations such as the one where an acoustic wave is characterized by a time-varying frequency.Figure 1(b) shows the schematic of AOD light diffraction with a chirped acoustic wave.Here, the angular deflection becomes a function of time [60,61]: where x indicates each point in the AOD's aperture with x = 0 being the laser spot centre.Then, a time-chirped acoustic frequency can be described as a function of its profile slope, or chirp α: where T scan is the time required to change the frequency from the minimum value f min to the maximum f max .Substituting this frequency expression in equation ( 6) we find: This equation describes a convergent cylindrical lens, or acousto-optic lens (AOL), where the first term indicates an off-set central propagation axis that varies with time, i.e. called lateral drift.The inverse of the second term is the chirp-dependent focal length: In other words, in a determined focal position, the beam moves laterally while the acoustic frequency is chirping.It is possible to compensate for this unwanted lateral drift by optically coupling the first AOD to a second one, driven with an opposite chirp (figure 1(d), second panel): Synchronizing the two AODs, the light beam will impinge on two subsequent devices driven by counterpropagating sound waves travelling through their crystal.The total angular deviation that affects the beam, in this case, is: The cylindrical lens described in equation ( 12) has a central propagation axis determined by the median driving frequency and a variable focal length that depends on the chirp.This configuration forms an active AOL capable of 3D scanning.Figure 1(c) shows a schematic of light beam projection on the sample through the illumination objective after AOL deflection.Moreover, when these devices are configured for volumetric targeting, to keep the spot at fixed coordinates for a longer time than T scan it will be essential to regularly repeat the frequency scan with equal slopes, such as a sawtooth profile (figure 1(d), third panel).This is necessary as the acoustic bandwidth and thus T scan is limited.In addition, several other frequency profiles can be used to modulate the acoustic wave propagating in the AODs.For example, it is possible to obtain a fast axial periodic scan with two AODs, by modulating the two counterpropagating waves with sinusoidal profiles [62] (figure 1(d), fourth panel).
Notably, it should be mentioned that acoustic frequency sweeping also makes the AOD resolution a dynamic quantity, reducing it compared to the static case: This means that the number of neuronal targets addressable by AOD-based systems depends on the acoustic wave frequency profile selected.
To summarize, AODs can be driven in different ways to freely deflect the light and project the beam on the sample.Using two AODs oriented orthogonally to each other, eventually optically conjugated via a 4-f telescope, provides a straightforward and fast way to control the beam in 2D, for example, for a planar raster scan (x-y scanning, figure 1(e)).Moreover, with two optically conjugated AODs oriented parallelly it is also possible to independently control the lateral deflection of the beam along one direction and its axial address (x-z scanning, figure 1(f)).To extend the light manipulation to a more general 3D scanning (x-y-z scanning, figure 1(g)), two more AODs are required to be implemented, for a total of four optically coupled devices.That is because, for each direction, it is necessary to compensate for the corresponding lateral drift discussed above.
Being intrinsically unaffected by inertia or any other mechanical constraints, these mentioned driving configurations provide an absolutely stable and steady way to achieve fast point scanning.Moreover, unlike any other standard beam scanning principles, each not-adjacent coordinate can be reached with the same access time, thus allowing for true random-access scanning.However, a finite time is still required to move from one coordinate to another, that is the access time needed to allow the soundwave to interact with the whole beam diameter, (typical access time: 10-30 us, see equation ( 5)).Therefore, it is important to highlight a different solution adopted by [63].Introducing a slightly different chirp in equation ( 11) for the counterpropagating wave in the second AOD results in a partial compensation of the lateral drift.Overall, this generates a movement of the beam spot along a line throughout the specified dwell time.In other words, by selecting proper acoustic frequencies, it is possible to generate fast and continuous lateral scans, with certain drift speeds, in a particular z-plane.
Even though with this approach the lateral beam position drifts continuously, interruptions are still required to jump from one plane to another.In 2016, the same group [64] further developed the earlier method by introducing a novel relationship between the focal spot coordinates and the chirp parameters of the four AODs.Briefly, this is a generalization of what was previously stated concerning the use of acoustic frequencies characterised by non-linear chirps, that vary in time according to parabolic profiles.Partial drift compensation with these particular acoustic frequency profiles allowed to define continuous trajectories in 3D with arbitrary directions and predetermined speeds [53].

Multibeam diffraction with AODs
Until this point, we have only considered driving each AOD with a single acoustic frequency.Even though AOD represents the fastest device commercially available to deflect light (up to hundreds of KHz), it enables only sequential, or random-access, scanning when driven with a single frequency at the time.It means that in this configuration, concurrent illumination of different targets, or neurons, is never obtained, but they can only be addressed subsequently.However, there is no limitation on the number of radio frequencies applied concurrently to the piezo other than avoiding exceeding the maximum power.When the piezo is simultaneously driven by different frequencies, multiple sound waves propagate through the crystal.In consequence, the impinging beam will interact with a linear combination of these waves and be simultaneously diffracted by different gratings [65,66].This principle enables the use of an AOD to parallelize the illumination and simultaneously deflect the input light under different angles (figure 1(h)).While this comes at the cost of power reduction for each diffracted beam-a common shortcoming in each light parallelization approach-it can be compensated by simply increasing the input light.Several methods exist to generate multi-frequency signals, ranging from simply combining the outputs of different generators or programming a complex dynamic waveform through an arbitrary waveform generator.

Applications in experimental neuroscience
In this chapter, we retrace the main results achieved with AODs in experimental neuroscience.Herein, we followed the evolution of the different AODs implementations and configurations, explaining how their different features and targeting strategies were exploited by neuroscientists in the past years.In detail, we collect the results in three separate sections: the first regards the imaging applications; the second concerns the uses of AODs to improve the capabilities of well-established imaging methods, such as light-sheet fluorescence microscopy (LSFM); the third reviews all the successes obtained with AODs for photostimulation of neuronal samples and optogenetics.

Imaging of neuronal populations with AODs
The first interest in using AODs for neuroimaging was driven by their scan rate advantage over GM-based scanners.With the simplest implementation of a single AOD, notable performances were reached in raster scans beyond video rates, i.e. above 30 Hz, both with single-photon [57] and multi-photon excitation [67,68].Another notable property of AODs led to the development of single-photon random-access fluorescence microscopes, where the excitation beam can be freely and quickly repositioned within the FOV [69].For instance, by using voltage-sensitive dyes and Ca 2+ indicators to monitor functional neuronal activity, multiple sites were imaged in µm-sized processes of cultured hippocampal neurons with up to 200 000 samples per second and a repositioning time between 3-5 µs.After that, with the introduction of a dispersion compensation scheme for ultrafast laser pulses [59], natural steps forward were moved toward AOD-based random-access multi-photon (RAMP) microscopy [70,71], extending the range of applications.Among them, many different 2D and 3D scanning methods have been developed to concurrently or sequentially explore different regions of interest (ROIs) with high temporal and spatial resolution for structural imaging and functional neuronal activity recording, both for ex vivo and in vivo preparations.Figure 2(a) shows as examples, seven different targeting strategies: point-by-point, ribbon, snake and multi-3D-line scanning, chessboard and multi-layer multi-frame sampling, and multi-cube volumetric scanning [64].Remarkable neural functional and structural investigations have been realized with 2D RAMP microscopes based on two crossed AODs [70][71][72][73][74][75][76][77][78][79][80][81].Multi-ROI scan rates of 1 kHz or beyond [72,75] have been demonstrated to track calcium activation in different cultured neuron populations and imaging large FOVs up to 0,5 × 0,5 mm 2 [77].Here, we report two examples of the structural and functional measurements presented in Iyer et al [71,75] and Otsu et al [75] as typical examples of what is achievable with AOD-based 2D RAMP microscopes.Figure 2(b) displays the maximum intensity projection (MIP) of a CA1 pyramidal neuron, revealing many apical dendrites [63].Recordings at 500 Hz of Ca 2+ transients from single optical sections at selected sites are superimposed.Figure 2(c) shows the MIP of a Purkinje cell [75], obtained by a sequential raster scan of 40 optical sections spaced every 1 mm, while figure 2(d) illustrates a single-plane image of the apical dendrite and of a proximal oblique dendrite of a layer V pyramidal neuron.The red dots indicate the positions of optical calcium-activation recordings displayed in figure 2(e).The upper trace exhibits a depolarizing current step triggered by an early spike and a tonic discharge of additional spikes on top of a depolarized plateau, while the lower traces show fluorescence transients recorded along the dendrite.
Interestingly, with two crossed AOD RAMP configuration, multi-beam illumination has also been investigated to parallelize the sampling of different ROIs.In particular, D'Angelo et al recorded in vitro the electrical activity from clusters of Purkinje cells [83] and cardiomyocytes with voltage-sensitive dyes [79].
The first attempts at using AODs for axial scanning enabled imaging and fluorescence lifetime imaging of neurons in mouse hippocampal tissue in the volume [84].In detail, two planes of 100 × 100 µm, axially displaced by 40 µm were subsequently illuminated adjusting the frequency profile of the radio-signal driving the AODs, without moving the objective.Introducing an ETL or a piezo-scanner at the objective improved on this aspect but compromised the volumetric scan rate to 20-30 Hz for functional in vivo Ca 2+ recording in mouse neocortex [85] and to 6 Hz for capturing sensory-evoked activity of the entire neurons within the developing brain of tadpoles [86].
Nevertheless, the full potential of this technique was achieved by employing two optically conjugated pairs of crossed AODs in 4-f configuration, as a 3D AOL [4,63,64,82,[87][88][89].This configuration supports repositioning speeds ranging from 30 µs down to 10 µs [88], which is important for capturing fast spontaneous neuronal microcircuit activity in the mouse cortex, and access volumes up to 1 mm 3 [63].The very high sampling speed of 3D RAMP has enabled in vivo functional imaging studies in behaving mice using Ca 2+ indicators, even with motion compensation [4,64,88,89].Figure 2(f) shows a 3D rendering of in vivo AOL-based volumetric structural imaging of neuronal dendrites and spines in a 150 µm depth range, labelled with the fluorescent dye Alexa 594 [82].Another representative example of 3D AOL-based RAMP imaging is presented in figure 2(g), which displays an in vivo dendritic segment of a selected GCaMP6f-labeled neuron in a behaving mouse [64] recorded with the 3D ribbon scanning method in a volume of 140 × 70 × 80 µm 3 .Figure 2(h) shows the longitudinal (left) and transverse (right) raw fluorescence scans (in green) and the average Ca 2+ responses during spontaneous activity (in colour) [64], acquired along the blue ribbon denoted in g) and corrected for motion artefacts.
Finally, it is noteworthy the AODs' driving solution recently adopted by Villette et al [90] and Akemann et al [91], where the frequency-and amplitude-modulation of the acoustic wave is synchronized with the pulsed laser clock.In such a way, the focus drift associated with chirped AO gratings is abolished, and only two AODs are needed to access the sample very rapidly in 3D [92].In particular, they used the submillisecond temporal resolution provided by this approach to probe cortical and hippocampal neurons in awake-behaving mice expressing high-gain voltage indicators [90] at 15 kHz, and GCaMP6f [91] at 40 kHz, respectively.

AOD in synergy with imaging systems
Recently, AODs have been used to enhance the capabilities of well-established imaging techniques.For instance, Tsyboulski et al [93] presented an innovative frequency-multiplexing imaging strategy based on AODs.In particular, they demonstrated the ability to encode multiple excitation spots with specific modulation frequencies and discussed promising perspectives for very fast monitoring of multisite calcium signals in brain tissues.Furthermore, 1D AOD-based slit-scanning has been used in combination with micro-optical sectioning tomography to confocally image thick mouse brain samples [94].Nevertheless, AODs have given their largest contribution to another foundational technique: the LSFM.In detail, AODs have been successfully used to generate a scanning light sheet that illuminates sample planes in the fastest way possible, up to 0.8 million frames per second [95].Such speed enabled plane-by-plane volumetric structural and functional imaging of ex vivo tissues and even of entire in vivo animals, including Caenorhabditis elegans, at tens of Hertz volume rates.In addition, using a crossed configuration of 2 AODs in the excitation arm and an acoustic varifocal lens in the detection path has allowed the realization of a fast inertia-free volumetric LSFM [96].In the same context, single AODs have been successfully exploited to pivot the illumination beam and remove striping artefacts from the collected neuronal functional images in zebrafish [97] and mice [98], a well-known problem in LSFM [99] caused by scattering and/or absorption of light incident upon the sample through uneven lateral illumination.Notably, in these experiments, enhancing angular diversity in the illumination direction through simultaneous multibeam and/or rapid  Moreover, the approach of driving the AODs concurrently with several radio-frequency signals has also been exploited to generate a stack of parallel light-sheets, enabling the simultaneous illumination of different sample planes and obtaining a fast volumetric image with high signal-to-noise ratio and contrast [101].Finally, this method has been used to double the LFSM acquisition velocity by generating simultaneously two light sheets in combination with the advanced rolling shutter acquisition modes of a CMOS camera [100], as displayed in figure 3(c).Notably, this approach prevents any negative impact on the image quality, as demonstrated in figure 3(d) by imaging zebrafish larva's neuron nuclei with activity observable in the left habenula within the diencephalon (inset).

Targeted photostimulation and optogenetics with AODs
AODs can play a key role in direct optical stimulation as they enable precise light delivery to specific neurons or brain regions and the selective targeting of specific neural circuits in the brain.The main advantage of AODs is that they can position a laser beam within a volume with high temporal and spatial precision, making them ideal for optogenetic experiments.This means that researchers can activate or inhibit specific neurons with light pulses of varying duration, intensity, and location.However, while AODs have been widely used for imaging in experimental neuroscience, there have been relatively few reported applications of AOD-based optogenetic stimulation.
The first significant attempt to use AODs for photostimulation dates back to 2002 and demonstrated local and graded neuronal membrane currents following focal multisite UV illumination of a caged neurotransmitter [102].In the following, Shoham et al [103] used AODs for rapidly and precisely controlling the release of neurotransmitters in specific locations of the brain.The researchers demonstrated that this technique can be used to selectively activate or inhibit the activity of neurons in the brain using UV light (355 nm) by single-photon excitation with both high temporal (20 000 locations per second) and spatial resolution.Figure 4(a) reports a two-photon image of a pyramidal neuron showing the selected sequential uncaging locations (numbers 1-10) on the apical dendrite and the related integration of electrical responses from the cell at different uncaging intervals.
After that, the next step was taken by Losavio et al [106] developing the first AOD-based two-photon microscope optimized for activating single neurons by their post-synaptic receptors using caged compounds.Contrary to single-photon excitation, the combination of two-photon excitation and AODs presented distinctive technical challenges.Ultrafast lasers, indeed, have a wider spectral content than continuous lasers, and since the AOD scanning principle relies on diffraction, compensation is necessary.The authors describe a detailed and practical strategy to compensate for spectral and temporal dispersion, allowing for nearly diffraction-limited visualization of structures and the delivery of physiological stimuli.
A positive validation of using an AOD system for optical brain stimulation has been reported by direct activation of cell-expressing channelrhodopsins [104,107,108].In these studies, the ability to selectively activate neurons using a single-photon [104,107] and multiphoton [108] AOD-based system was demonstrated in cultured specimens (cells, neurons, and astrocytes) and the brains of mice, drosophila and zebrafish larvae.Figure 4(b) shows a HEK 293 cell expressing light-sensitive ChIEF-tdTomato together with the pattern of fixed-spot and whole-soma scanning photostimulation; membrane currents recorded in response to fixed-spot (dark) and whole-soma scanning (red) stimulations at different levels of laser power, together with such currents evoked by traditional Hg lamp illumination (cyan) is also shown.Four different AOD-generated stimulation patterns show the large flexibility of the AOD scanning system that allows optimizing the excitation scheme for selectively covering the whole cell and scanning the laser across the cell in an ultra-fast manner (10-50 µs/site) [104].
Starting from there, AOD-based microscopy combined with genetically targeted expression of photoactive proteins has been used to highlight functional connections and synaptic plasticity in the mouse brain.Huang et al showed that learning tasks can affect the connectivity of interneurons in the olfactory bulb and that active learning can promote plasticity in connectivity maps [109].Quast et al analyzed the development of inhibitory sensory maps and how they change over time, showing that inhibitory sensory maps become broader during development [110].Hernandez et al discussed the use of optogenetics to induce complex patterns of activity in the cerebellar granular layer, showing that this technique can reveal insights into the integrative properties of this brain region [111].
A detailed analysis of the functional synaptic architecture of callosal inputs into the mouse's primary visual cortex has been reported by Lee et al [4].The authors used optogenetic techniques to activate specific callosal inputs and found they could modulate the activity of both excitatory and inhibitory neurons in the cortex, showing that callosal inputs play an important role in shaping visual processing.Along the same lines, Conti et al analyzed functional connectivity in the brain using a large-scale double-path illumination system with a split FOV to enable the simultaneous measurement of inter-and intra-hemispheric functional connectivity [112].AODs have been used in a combined rehabilitation approach for promoting the recovery of structural and functional features of healthy neuronal networks after stroke [113,114].Here, the authors used optogenetic techniques to selectively activate neurons in the cortex after stroke and found that a rehabilitation strategy based on combined physical and direct optogenetic stimulation of neural cells improves functional recovery and increases cortical activity.Furthermore, to probe specific neural activation features associated to relevant movements in awake mice, transcranial optogenetic stimulation by means of AODs, allowed to highlight the underlying neural connectivity [105].Figure 4(c) shows representative average calcium responses to the optogenetic stimulus train at increasing laser powers in a mouse.The yellow line represents the calcium response threshold associated with complex movement execution.On the right, representative wide-field image sequences of cortical activation at different laser powers.The results suggest that this combined approach may be a promising therapeutic strategy for stroke recovery and to highlight cortex connectivity.

Discussion
As presented above, AODs respond to demands for light-addressing precision, high temporal resolution, random-access scanning, and multi-target illumination, making them appealing for a very wide variety of experiments.Table 2 summarizes several representative experimental results achieved in neuroimaging and optogenetics with AODs according to the specific configurations, light excitation modalities and characteristics.
Thanks to their high flexibility, these devices have been used to push the performance of different instruments designed for applications in experimental neuroscience.For instance, several light-sheet microscopes have benefited from the AODs' fast scanning rate.
However, despite their fundamental advantages, there is a drawback when AODs are used for axial focusing.This is because light transmission has an intrinsic dependency on the driving acoustic frequency.In particular, as a function of the chirp imposed on the driving frequency, the beam will spend different effective times illuminating spots in different planes, making the AODs in principle not well suited for uniform intensity axial photostimulation.To improve the power delivered within the pixel-to-pixel dwell time, the frequency ramps driving the AODs can be triggered repeatedly, effectively multiplying the minimum energy deposited on different focal planes [108].Alternatively, to overcome this lack of uniformity in the axially released power distribution, it is possible to apply a pre-calibrated tuning of the driving signal amplitudes for each point addressed within the volume, flattening the distributions.With these approaches, we can expect an increased use of AODs in 3D optogenetic applications, which is lacking so far.
Moreover, 1P excitation strategies have been preferentially implemented with AOD-based microscopes.The use of AODs with femtosecond pulses raises, indeed, several technical difficulties.This is because ultrafast laser pulses needed for multiphoton illumination, experience significant spatial and temporal dispersion while propagating through AO materials.Recalling equation ( 1), the angle of deflection is wavelength-dependent and an ultrafast and thus wideband laser pulse is spatially dispersed by the AOD (equation ( 4)).At the same time, ultrafast lasers are affected by temporal dispersion when they propagate through dispersive materials, with a frequency-dependent index-of-refraction n(λ).This is particularly evident for crystal materials commonly used in AOD, causing the temporal broadening of an ultrashort pulse.These effects of beam distortion and pulse broadening reduce the illumination efficiency, image resolution and signal-to-noise, making 2P excitation less attractive.Fortunately, several solutions have been proposed to face this issue.Iyer et al [59] developed a compact design introducing an additional diffraction grating after the AOD to strongly reduce the spatial dispersion, and a four-pass stacked-prism 'pre-chirper' for total compensation of the temporal dispersion.After that, it has been widely demonstrated [115,116] how introducing a highly dispersive prism before the AOD(s), can simultaneously provide a large compensation of both temporal and spatial dispersion.Other alternatives rely on the implementation of a single AOM placed at 45 • before the AODs [72], or of a special Keplerian telescope [117].
Notably, it has been demonstrated that AODs can be successfully applied in vivo experiments, where the sample animals are free to move.Even though light scanning over several targets usually assumes animal immobilization to avoid motion artefacts, this is not strictly required with AOD-based illumination.Indeed, artefacts and misleading results can be straightforwardly avoided by targeting each neuron with a tailored pattern of illumination covering the cell body [91] or by quickly drifting the excitation spot anticipating the displacement [64].
Furthermore, special attention must be paid when the AODs are used for multitarget illumination.The light power is not equally distributed among the different beams, as expected from an ordinary beam splitter.Each generated beam depletes the source beam approximately equally, but in this multifrequency condition, intermodulation orders can appear [65].In fact, the light in each principal beam may be re-diffracted by another acousto-optic grating generated in the crystal at the same time.This produces a slight light-intensity cross-modulation and can generate unwanted beam deflections at frequencies defined by a combination of the main frequencies.However, it has been demonstrated [118] that the intermodulation beams can be strongly attenuated.They showed that when the acoustic oscillation period is shorter than the access time (see equation ( 5)), the distribution of light power of the diffracted beams can be considered proportional to the power distribution over the corresponding frequency components.A complementary approach is to generate generic multi-frequency driving signals for the AODs with highly attenuated intermodulation through an arbitrary waveform generator.

Conclusions
In this review, we provide the theory of AODs and a wide range of applications in experimental neuroscience where they have been successfully employed, significantly advancing the field.This technology represents a viable alternative to other well-established light-addressing approaches developed in recent years.Indeed, the fast responsiveness, accuracy in light-targeting and random-access capability of AODs gradually made them a tool of choice for an increasing number of researchers utilizing optical approaches in neuroscience.In addition, the versatility and adaptability of AODs are continuously creating unexplored implementations and scenarios, such as 3D random-access optogenetics.In conclusion, we believe these devices will offer valuable possibilities to researchers, aiding in the study of increasingly complex neural systems, their disorders and pathologies, both in-vivo and ex-vivo.

Figure 1 .
Figure 1.AOD illumination strategies.(a) AOD driven with single static frequency diffracts the incident laser beam at a frequency-dependent angle θ d (f).0th order is the residual un-diffracted beam.(b) AOD driven with time-varying frequency diffracts the laser beam both laterally and axially with frequency-dependent displacement.(c) Schematic of AO deflected laser beam projected on the sample through the illumination objective.(d) Main frequency profiles used to drive the AODs.(e)-(g) Default configuration of AODs used for lateral scanning in 2D (XY), axial scanning in 2D (XZ) and 3D scanning (XYZ), respectively.Relay optics between deflectors omitted.(h) An AOD simultaneously driven by multiple acoustic frequencies diffracts the input beam at corresponding angles, parallelizing the illumination.FP: focal plane.

Figure 2 .
Figure 2. RAMP imaging applications of AODs.(a) Example of seven 3D scanning methods developed for in vivo imaging with motion compensation.The red points, lines, surfaces and volumes represent the addressed ROIs.Reproduced from [64].CC BY 4.0.(b) MIP of a CA1 pyramidal neuron, revealing many apical dendrites.Recordings at 500 Hz of Ca 2+ transients from a single optical section at selected sites are superimposed.Scale bar, 50 µm.Reproduced with permission from [71].(c) MIP of a Purkinje cell, obtained by a sequential raster scan of 40 optical slices spaced every 1 µm.Scale bar: 50 µm.Reprinted from [75], Copyright (2008), with permission from Elsevier.(d) Single plane image of the apical dendrite and of a proximal oblique dendrite of a layer V pyramidal cell, the red dots indicate the positions of the optical recordings displayed in (e).Reprinted from [75], Copyright (2008), with permission from Elsevier.(e) Upper trace: a depolarizing current step triggered an early spike and a tonic discharge of additional spikes on top of a depolarized plateau.Lower traces: fluorescence transients recorded at the red dot positions in (d).Reprinted from [75], Copyright (2008), with permission from Elsevier.(f) 3D rendering of an AOL-based volumetric structural imaging of neuronal dendrites and spines over a 150 µm depth range, labelled with the fluorescent dye Alexa 594.Reprinted from [82], Copyright (2014), with permission from Elsevier.(g) 3D in vivo image of a dendritic segment of a selected GCaMP6f-labeled neuron in behaving mice recorded with the ribbon scanning method using 3D DRIFT AO.Red cube, 140 × 70 × 80 µm 3 .Reproduced from [64].CC BY 4.0.(h) Longitudinal (left) and transverse (right) raw fluorescence scans (green) along the blue ribbon shown in (g), with motion artefacts elimination.Average Ca 2+ responses along the ribbon during spontaneous activity (syn.) are colour-coded.Reproduced from [64].CC BY 4.0.

Figure 3 .
Figure 3. LSFM imaging applications of AODs.(a) LSFM images obtained with different AOD-generated light-sheet configurations (1 or 7 static light sheets at different angles and single-plane sweeping mode).The sample contains 15 µm polystyrene beads embedded in 1% agarose gel immersed in water.The insets show the fluorescence intensity profiles along the y-axis, averaged over 6 µm around the white dotted line.The line plots depict how the background becomes more uniform with an increasing number of illumination directions.The green arrows indicate the examples of striping artifacts in the three panels.White scale bar 50 µm.Reproduced from [97].CC BY 4.0.(b) Confocal LSFM image of an area of a mouse brain expressing tdTomato, obtained by AOD-pivoted illumination of the sample.The image shows the effectiveness of the approach to remove striping artefacts from the images.Scale bar size: 100 µm.Reproduced with permission from [98].(c) An AOD is used to generate simultaneously two light sheets that are concurrently imaged on a camera to double the acquisition speed of a confocal LSFM while preserving the properties of confocal detection.Reproduced from [100]CC BY 4.0.(d) Example image of zebrafish larva's neurons taken with the AOD-based dual LSFM, colour-coded in purple.The inset shows a 4x magnified left habenula area within the diencephalon where neural activity can be observed.Scale bar size: 200 µm; 20 µm in the inset.Reproduced from [100].CC BY 4.0.

Figure 4 .
Figure 4. Applications of AODs for photostimulation and optogenetics.(a) Image and temporal traces of sequential photo-induced uncaging in different locations on an apical dendrite.Reproduced from [103], with permission from Springer Nature.(b) Optogenetic activation of HEK 293 cell expressing ChIEF-tdTomato and different activation patterns generated with AODs.Reproduced from [104] CC BY 3.0.(c) Calcium responses to the optogenetic stimulus train in mice cortex at different excitation power.Reproduced from [105].CC BY 4.0.

Table 1 .
Comparison of typical illumination strategy, dwell time, and cost of implementation between different devices.

Table 2 .
Representative experimental results achieved in imaging and photostimulation of neuronal samples by using different AOD configurations, specimens, and parameters; 1PE and 2PE indicate one-photon and two-photon excitation of the sample, respectively; SHG: second harmonic generation.