2.1.

Design of Hybrid Multiphoton Holographic Stimulation and Imaging System

Our approach toward a bidirectional optophysiological system is based on expanding a recently introduced hybrid multiphoton TF microscope, which can volumetrically record neural activity using calcium sensitive dyes,²¹ to also enable precise 2-D or 3-D optogenetic stimulation of neuronal networks. The hybrid microscope uses light from a regeneratively amplified ultrafast laser (RegA 9000, Coherent) and combines a scanning line TF mode with standard point-by-point 2P laser scanning microscopy. For targeted photostimulation, a HONS subsystem based on a phase-only spatial light modulator (SLM) was added [Fig. 1(a)], and the RegA wavelength was custom-switched from 800 to 905 nm,²³ closer to the 2P excitation peak of ChR2.^4,14 Both the hybrid microscope and the new hybrid 2P HONS subsystem use the same optical element, a dual prism grating (DPG or grism), in order to generate optically sectioned TF illumination patterns, while allowing an easy switch to a non-TF mode. The DPG is a custom transmissive diffraction grating mounted between two prisms and has three fundamental advantages over standard blazed gratings, which are usually used for TF. First, it can be easily placed in the optical path with minimal changes, since the diffracted light passing through the DPG continues to propagate along the original optical axis. Second, the DPG enhances diffraction efficiency (85% in our setup) relative to reflective diffraction gratings (typically 50% to 70%) accommodating for the limitation on the holographic projection system imposed by the pulse energy. Third, a DPG-based TF setup enables remote scanning of the focal plane,²⁴ thus allowing an independent axial movement of the imaging and stimulation modules.

Fig. 1

Hybrid-multiphoton holographic stimulation and imaging system. (a) Optical scheme shows the optical arrangement of the two-photon holographic optical neural stimulation (2P-HONS) subsystem and its integration with the hybrid multiphoton microscope, whose EMCCD camera can also be used with a fluorescence light source for conventional single-photon imaging. The dual prism grating provides the HONS system with an optional temporal focusing (TF) mode. (b) System’s resolution for single spot generation (w/o TF) shown as lateral and axial projected point spread function’s (PSF) cross-sections with full width at half maximum dimensions of $1.5\mu \mathrm{m}$ laterally and $13.9\mu \mathrm{m}$ axially, and $2.6\mu \mathrm{m}$ (lateral) and $25.2\mu \mathrm{m}$ (axial), for the $40\times$ and $20\times$ objective lenses, respectively. Dots and lines represent the raw measured data and Gaussian fits, respectively. (c) Holographically generated two-dimensional (2-D) light pattern projected onto a fluorescein film through a $20\times$ objective [neural interface engineering lab (NIEL)—lab initials]. Scale bar: $50\mu \mathrm{m}$.

For identical average powers, an amplified laser enhances 2P absorption by a factor inversely proportional to the pulsed laser’s duty cycle²⁵ (product of pulse duration and repetition rate): compared to a standard Ti:Sapphire laser operating at a repetition rate of $f=80\mathrm{MHz}$, an amplified laser that has roughly the same pulse duration ($\tau =100\mathrm{fs}$) but a repetition rate of $f=160\mathrm{kHz}$ provides an expected excitation gain factor of $\sim 500$ (excluding saturation effects). In HONS and TF, this gain in efficiency can be used to mitigate the effect of reducing the beam’s intensity by splitting it and increasing its focal diameter. However, due to the technical issues involved in switching the RegA laser wavelength from 800 to 905 nm, the modified laser generates pulses with an extended duration ($\sim 400\mathrm{fs}$) and reduced average power output ($\sim 150\mathrm{mW}$ at 150 KHz) in comparison to its 800-nm design wavelength (600 mW at 150 KHz), which was further attenuated to $\sim 25\mathrm{mW}$ near the objective focal plane mainly due to the SLM’s relatively low efficiency. To allow the entire laser power to be directed toward each of the subsystems, a flip mirror (not shown) enables an easy switch between the two setups; to maintain the ability to image the holographically stimulated networks’ responses in 2-D cultured cells with the ability to flexibly change the imaged wavelengths, a wide-field epifluorescence illumination path (single-photon) was integrated into the system: light from fiber illumination source (Intensilight, Nikon) was transmitted through a designated filter cube, which was replaced according to spectral specifications of the imaged fluorescent probe. Light was reflected to and from the sample by a dichroic mirror placed above the objective lens. The emitted fluorescence from the sample was detected by an electron multiplying charge coupled device (EMCCD, Andor, front-illuminated iXon DU-885, 1Mpixels, max full resolution frame rate: 31 fps).

In the HONS optical path, a collimated beam from the RegA laser passes through a Pockels cell (Conoptics, model 350-801A) for power modulation and is then expanded by a beam expander (6x) to match the laser beam cross-section to the input aperture of a liquid crystal on silicon-SLM (X10468-02, Hamamatsu), which operates in reflection mode. This nematic SLM has a resolution of $792\times 600\text{pixels}$ extending over an area of $16\times 12{\mathrm{mm}}^2$, with 256 phase gray levels and a 95% fill factor. A half-wave plate was placed between the beam expander and the SLM to horizontally polarize the input light in order to optimize the modulation performance, and to minimize the system’s distortion, the angle between the SLM’s input and output beams was minimized ($\sim 10\mathrm{deg}$). A $4f$ telescope ($L_1=300\mathrm{mm}$ and $L_{\text{tube}}=250\mathrm{mm}$) was used to image the SLM aperture to the rear aperture of the objective lens; the telescope’s magnification, $M=f_{\text{tube}}/f_1$, was chosen to fully utilize the objective lens’ numerical aperture (NA). An enlarged replica of the target hologram was obtained in the focal plane of $L_1$ (which is conjugated to the focal plane of the objective lens) and was used for filtering unwanted diffraction orders or other manipulations of the holograms. Specifically, a beam block made of a tin dot was used to physically block the zero order spot. The custom DPG (Wasatch Photonics) is optionally introduced in the focal plane of $L_1$ for obtaining TF and consists of a transmission diffraction grating ($1200\text{lines}/\mathrm{mm}$) mounted between two prisms ($58\times 32\times 90\mathrm{deg}$, BK7 glass).

An objective lens and tube lens mounted in a $4f$ configuration image the hologram generated by the SLM onto the objective focal plane, where a 2P holographic pattern is obtained. With the DPG introduced (TF mode), the first plane is on the grating’s surface and a TF pattern is obtained in the objective’s focal plane. The system was characterized (and applied here) using two water immersion objective lenses: a $40\times$ 0.8 NA lens (Nikon) and a $20\times$ 0.5 NA lens (Olympus), resulting in respective projection field dimensions of $275\mu \mathrm{m}\times 265\mu \mathrm{m}$ and $495\mu \mathrm{m}\times 485\mu \mathrm{m}$. Without the DPG installed into the optical path, the PSF full-width at half-maximum (FWHM) dimensions are $1.5\mu \mathrm{m}$ laterally and $13.9\mu \mathrm{m}$ axially, and $2.6\mu \mathrm{m}$ (lateral) and $25.2\mu \mathrm{m}$ (axial), for the $40\times$ and $20\times$ objective lenses, respectively [Fig. 1(b)]. Note that due to underfilling of the objective lens, the effective NA is $0.19\pm 0.05$ and $0.28\pm 0.05$, respectively. This underfilling enables a wider scan range and the use of fewer and larger spots for each holographic patch. Using a flexible graphical user interface based software which implements the weighted Gerchberg–Saxton (GSW) algorithm,²⁶ the holographic system is capable of generating flexible multispot holographic patterns [Fig. 1(c)], which can be dynamically updated at rates up to the SLM’s refresh rate of $60\text{frames}/\mathrm{s}$.

2.2.

Hologram Calibration and Calculation Process

The stimulation and imaging fields of view (FOVs) were calibrated to match the location of the excitation patterns’ coordinates to the neurons’ image locations using the following equation:

Calibrated holograms were calculated using GSW algorithm,²⁶ which produces holograms with high efficiency and uniformity but is relatively computationally demanding. We set the convergence values for efficiency and uniformity to 0.95 and limited the number of iterations to 10 to shorten the calculation time. Compensation over distortion caused by the SLM lack of flatness was done by adding a flatness correction phase to the projected computer generated holograms (CGHs), supplied by the SLM manufacturer.

The CGHs were controlled by MATLAB®-based psychophysics toolbox²⁷ and sent to the SLM via a digital video interface at a refresh rate of 60 Hz. The SLM was synchronized with other equipment by splitting the video output to an additional display and by using a photodiode to detect the illumination changes on the screen. This signal was used for recording of actual holograms’ timing.

3.1.

Preparation, Transfection, and Staining of Cortical Neural Cultures

Cortices were extracted from anesthetized P0 Sprague–Dawley rats, dissected in phosphate-buffered saline $\mathrm{(PBS)}+20\mathrm{mM}$ glucose and incubated with trypsin for 10 min. Trypsin proteolytic activity was stopped by the addition of horse serum. Cells were separated from the tissue by gentle trituration with a pipette, filtered, and isolated by centrifugation. Cells were resuspended in planting medium, $\sim 0.14\text{million}$ cells were planted on each cover glass/multielectrode array (MEA) coated with polyethyleneimine. Cultures were maintained at 37°C in a 95% air, 5% ${\mathrm{CO}}_2$ humidified incubator, and culture media was replaced twice a week. Cultures were used after $\sim 10\text{days}$.

Viral transfection of neural cultures with optogenetic actuators/indicators was done 1 day after cell planting by application of adeno-associated virus (AAV) vector carrying the gene of interest under the control of a neuron specific promoter, usually synapsin (UPENN). Specifically: ChR2-mCherry [AAV9.CAG.hChR2(H134R)-mCherry.WPRE.SV40], ChR2-YFP [AAV9.hSyn.hChR2(H134R)-eYFP.WPRE.hGH], and GCaMP3 [AAV1.hSyn.GCaMP3.WPRE.SV40]. Concentrations of AAV vectors added to the culturing medium were as follows: 0.25 to $0.5\mu \mathrm{l}/\mathrm{ml}$ for GCaMP3, 1 to $3\mu \mathrm{l}/\mathrm{ml}$ for ChR2-mCherry, and $0.75\mu \mathrm{l}/\mathrm{ml}$ for ChR2-eYFP.

For calcium imaging using red-shifted organic indicator Rhod-2 AM (Invitrogen), at the day of the experiment, cell cultures were incubated with the indicator (2 to $3\mu \mathrm{l}/\mathrm{ml}$ of a $1\mu \mathrm{g}/\mu \mathrm{l}$ solution) for 45 min, at 37°C, 5% ${\mathrm{CO}}_2$ humidified incubator. Afterward, the medium was replaced with dye-free culture medium.

3.2.

Probe Screening Experiments

Probe screening experiments were performed in a single-photon holographic stimulation system. Detailed description of this optical setup and data analysis methods applied in experiments using optical and electrical simultaneous recordings can be found in Golan et al.²⁸ and Reutsky-Gefen et al.²⁹ Briefly, to stimulate ChR2-expressing neurons, wide-field blue light flashes were delivered with a laser coupled to an optic fiber (Thorlabs, $200\mu \mathrm{m}$ in diameter, 473 nm, 2 to $4\mathrm{mW}/{\mathrm{mm}}^2$). Calcium single-photon imaging was done using a CCD camera (Hamamatsu C8484-05G, Japan) at $10\text{frames}/\mathrm{s}$. Initial recordings of GCaMP3 calcium traces resulting from driving ChR2-expressing neurons were done using a GFP filter set (XF100-2, Omega), which was later changed to modified Texas Red filter set for Rhod-2 calcium indicator imaging (570Ex/625Em, Nikon, with excitation filter replaced to HQ580/13x, Chroma).

3.3.

Multiphoton HONS Characterization Measurements

Measurements of the spots’ lateral and axial resolution were done by projecting a single spot (or several spots) on a thin layer of fluorescein (FLUKA Fluorescein sodium salt, Sigma-Aldrich). The spot’s lateral resolution was calculated from the spot’s intensity lateral cross-section, which was fitted to a Gaussian curve and used to calculate the FWHM of the fitted intensity. The spot’s axial sectioning, which determined the system’s axial resolution, was calculated by measuring the spot’s intensity at different axial locations by an additional epifluorescence detection path (using a uEye camera, IDS), situated beneath the fluorescent sample. The sample and the second objective lens were mounted on two micromanipulators (MP-285 and MP-225, respectively, Sutter), which were used to move the sample and the detection system to controlled distances from the upper objective lens’ focal plane. The spot’s average intensity was calculated for images taken at different distances and fitted to a square-root of a Lorentzian curve. The FWHM of the fitted intensity was calculated.

3.4.

All-Optical Probing Experiments and Calcium Movie Analysis

Cells were first imaged to identify ChR2-eYFP-expressing cells. When using holographic targeting, we manually chose the cells we would like to target inside the accessible stimulation field and calculated the matching holograms. The excitation locations were saved for later recognition of stimulated cells during data analysis. Stimulation occurrences and pulse durations were also saved. Calcium responses elicited by ChR2 stimulation were optically recorded by an EMCCD camera at $10.9\text{frames}/\mathrm{s}$. The fluorescence signals of selected neurons were extracted from the movies by calculating the mean pixel intensity over a radius of $5\mu \mathrm{m}$ around the neurons’ locations for the different time points and subtracting the dark current values. These fluorescent signals were normalized according to

Fig. 2

Selection of optical indicators for imaging stimulated channelrhodopsin-2 (ChR2)-expressing neurons. (a) Left: ChR2, GCaMP3, and Rhod-2 single-photon excitation spectra with overlaid excitation filter bandwidths. The wavelength overlap between ChR2 and GCaMP3 (GFP filter cube, light blue) makes simultaneous stimulation challenging, whereas Rhod-2 excitation can be easily spectrally separated using appropriate filters (green shaded region). Right: ChR2 and GCaMP3 2P excitation spectra, which have a complete overlap at 905 nm. (b) Raster plot of representative channels from multielectrode array (MEA) recordings of neurons expressing ChR2 and GCaMP3 shows that responses are abolished once GCaMP3 imaging is performed in parallel with ChR2 stimulation. (c) Combination of ChR2 stimulation with Rhod-2 calcium imaging: stimulus locked responses to 0.5 s wide-field stimulation flashes are seen both in electrical recordings raster plots (left, four representative channels) and as calcium transients (right, four other selected neurons, arrows indicate the stimulus timing). Scale bar: 10 s and $0.5\mathrm{\Delta }F/F$. (d) Stimulation field maps superimposed onto fluorescence image of a culture seeded on an MEA (black dots are electrodes). Maps are estimated using spike-triggered averaging of pseudorandom holographic patterns (20 patches each). Scale bar: $200\mu \mathrm{m}$.

Fig. 3

Stimulation of ChR2 expressing neurons with a 2P holographic pattern in 2-D culture. (a) Concept image of holographic stimulation: superimposed image of ChR2 expressing neurons in 2-D culture (green) with a 2P holographic stimulation pattern (blue) targeted to specific neurons (separately projected over a thin fluorescein layer and captured). (b) and (c) Multicell targeting of ChR2-expressing neurons (surrounded by yellow circles). On the right, calcium transients show stimulus locked responses to various stimulation patterns of distributed spots: (b) sequential stimulation of multiple cells in culture through $40\times$ 0.8 NA. Stimulation pulse duration: 50 ms. (c) Simultaneous targeting of cells through $20\times$ 0.5 NA. Stimulation pulse duration: 100 ms. Arrows indicate the timing of the holographic 2P stimuli. Scale bars for calcium traces: 20 s, $0.2\mathrm{\Delta }F/F$. Scale bar for cell images: $50\mu \mathrm{m}$.

4.1.

Screening Calcium Indicators for Imaging Activity Patterns of ChR2 Stimulated Neurons

To find an effective combination of probes for parallel stimulation and imaging and to validate our 2P-HONS system, we applied all-optical single-photon imaging and stimulation to cortical neurons seeded on MEAs and recorded both spontaneous and stimulated activity. This experimental approach, using the (single photon) optical system previously described in Reutsky-Gefen et al.,²⁹ enables simultaneous electrical and fluorescence-based neural activity recordings for validating the all-optical measurements. Using this system, we tested the possibility to stimulate and image cultures transfected with viral vectors carrying ChR2-mCherry and GCaMP3 in parallel (following Guo et al.¹⁰), as well as the combination of ChR2-eYFP with the organic red calcium indicator Rhod-2-AM (Invitrogen), whose single-photon excitation spectrum is well-separated from that of ChR2 [Fig. 2(a), left].

We first tested the possibility to stimulate and image cultures transfected with viral vectors carrying ChR2-mCherry and GCaMP3 in parallel. Neurons were stimulated using wide-field blue laser light flashes (1 s flash duration, 9 s OFF, $\sim 4\mathrm{mW}/{\mathrm{mm}}^2$, $\lambda =473\mathrm{nm}$, VD-IIA DPSS laser; Sintec Optronics Technology) either with or without low intensity epifluorescent illumination for calcium imaging of GCaMP3 ($0.05\mathrm{mW}/{\mathrm{mm}}^2$, $\lambda =475\pm 20\mathrm{nm}$). Robust electrical responses to the light flashes were recorded when no calcium imaging of GCaMP3 was done in parallel [Fig. 2(b), upper panel]. However, these activity responses were abolished once the epifluorescence light source was lit [Fig. 2(b), lower panel]. Excitation results for this probe combination were also not observed optically even when the GCaMP’s illumination intensity was reduced $<1\mu \mathrm{W}/{\mathrm{mm}}^2$ and imaged by an EMCCD camera at high gain. These results indicate that ChR2 stimulation and calcium imaging excitation wavelengths overlap prevented the use of ChR2 in combination with GCaMP3 under these experimental conditions [Fig. 2(b)].

In contrast, ChR2-eYFP transfected cultures stained with Rhod-2 presented a high signal baseline and strong calcium response signals. In order to image calcium changes with Rhod-2 while completely avoiding ChR2 activation, we used a Texas Red filter cube (570Ex/625Em, Nikon) where the excitation filter was replaced to a narrow band filter around 580 nm (HQ580/13x, Chroma) since ChR2 produces no photocurrent at wavelengths between 550 and 650 nm [Fig. 2(a)]. With this filter set, it was possible to perform both optical and electrical recording of the stimulated neural activity simultaneously without causing ChR2 desensitization [Fig. 2(c)]. Cells were illuminated with average powers of $0.09\mathrm{mW}/{\mathrm{mm}}^2$ ($\lambda =580\pm 7\mathrm{nm}$) for imaging and of $\sim 2\mathrm{mW}/{\mathrm{mm}}^2$ ($\lambda =473\mathrm{nm}$) for wide-field stimulation (0.5 s ON/ 10 s OFF pulses). These stimuli caused responses to be observed on multiple electrodes [Fig. 2(c), left], as well as in the cellular calcium signals, where robust stimulus-locked calcium traces were observed [Fig. 2(c), right]. Additional study of these optogenetic responses included mapping of stimulation fields for different units by spike-triggered averaging of fast pseudorandom holographic patterns [Fig. 2(d), procedure previously described in Ref. 29]. The resulting stimulation fields were relatively large, including both cells’ somata and dendritic trees, and were clearly not disrupted by the constant illumination for Rhod-2 calcium imaging throughout the mapping experiments.

4.2.

All-Optical Probing of Cultured Neural Networks

We tested the basic 2P-HONS system’s ability to stimulate cultured neurons by imaging their stimulated responses using the single-photon epifluorescence imaging subsystem. Planar neuronal networks were transfected with ChR2-eYFP and bath-loaded with the fluorescent calcium indicator Rhod-2 [Figs. 3(b) and 3(c) left, see also relevant spectra and their separation in Fig. 2(a)]. Several stimulation protocols were used: sequential stimulation of multiple neurons in the culture at a constant rate [Fig. 3(b) right, 50 ms stimulation] and simultaneous stimulation of a number of cells at a constant rate [Fig. 3(c) right, 100 ms stimulation]. Single spots were illuminated on the sample through a $40\times$ or $20\times$ objective lenses at selected locations and the power directed to each spot was $\sim 14\mathrm{mW}$ on average with minimal excitation power of 9 mW. The cells were successfully excited using the 2P HONS system, causing a rise in the calcium signal following the localized excitation [Figs. 3(b) and 3(c) right].

4.3.

Improving the Axial Resolution of Cell Matched Patches

To robustly stimulate ChR2-expressing neurons, it is desirable to create illumination patches that will effectively cover the cell’s membrane^4,29 and simultaneously open a larger amount of channels. This capability is demonstrated in our system by selectively projecting cell matched $10\text{-}\mu \mathrm{m}$ diameter patches onto $10\mu \mathrm{m}$ beads [Fig. 4(a)]. Different-sized patches projected onto a fluorescein layer have the expected linear relationship between the target patch size and their lateral dimension [Figs. 4(b) and 4(c)]. These holographic patches are inherently accompanied by holographic speckle noise, and therefore we averaged over five circularly shifted holograms for each patch size to reduce the speckle noise.^30,31 Increasing the patch’s diameter strongly affects its axial size, which quickly exceeds cellular dimensions—calling for TF-based improvement of the axial resolution of the light patches.

Fig. 4

Projection of cell-size illumination patches through $40\times$ objective in conventional (non-TF) and TF modes. (a) Excitation patches matched to a selection of $10\mu \mathrm{m}$ fluorescent beads in order to cover their whole cross-section. (b) Images of different sized non-TF patches projected on thin fluorescein layer and averaged over five different holograms to reduce speckle noise. (c) Correspondence between illumination patch target diameter and their corresponding lateral dimensions (shown for non-TF mode). (d) Variable-sized illumination patches and their corresponding axial dimensions for both modes. Scale bars: $10\mu \mathrm{m}$.

In order to generate optically sectioned holographic illumination patterns by TF, the 905-nm optimized DPG was inserted into the optical path. Consistent with previous studies,¹⁹ when TF is introduced, the strong coupling seen between the patch lateral size and its axial dimension, becomes only weakly dependent on patch size [Fig. 4(d); best fit linear slopes of 3.77 versus 0.83, respectively], providing a regime with suitably cell-matched patches in our system.