2.1.

Optical Setup

The optical setup consists of a custom-made upright two-photon holographic microscope (Fig. 1). Two-photon excitation is supplied by a Ti:Sapphire laser (Chameleon Ultra II, Coherent, USA) providing 4 W maximum power at $\lambda =800\mathrm{nm}$. Polarization of the laser light is set to be vertical with a Glan-Taylor polarizer, preceded by a $\lambda /2$ plate for power tuning. The laser is collimated and expanded by a $6\times$ beam expander ($f1=25\mathrm{mm}$, $f2=150\mathrm{mm}$) and impinges at 9 deg on the reflective SLM (X10468-7, Hamamatsu), which allows an encoding phase from 0 to $2\pi$ on the beam profile with a $800\times 600\text{pixel}$ resolution and 8-bit pixel depth. The SLM plane is conjugated through a beam reducer ($f1=400\mathrm{mm}$, $f2=250\mathrm{mm}$) to the entrance pupil of a microscope objective (Zeiss Plan-APOCHROMAT $20\times$ 1.0 NA or Zeiss Plan-APOCHROMAT $40\times$ 1.0 NA). The beam reducer focal lengths were chosen to fill the objective entrance pupil. In this configuration, the electric field distribution on the objective focal plane is equal to the Fourier transform of the electric field distribution on the SLM plane. Since only phases of the electric field can be encoded on the SLM, the Gerchberg–Saxon (GS) iterative algorithm^18,19 was adopted to compute the phase distribution that best reproduces the desired intensity pattern. It is important to notice that an image of the desired pattern is formed on the focal plane of the first lens of the beam reducer. A common problem in SLM-based systems is the presence of the “zeroth order” of diffraction, a fraction of the excitation light which is not shaped by the SLM. This light in turn produces a bright focal volume at the center of the field of view, inducing photodamage in the sample and generating artifacts on images. This undesired contribution can be removed with a beam blocker in the focal plane of the first lens of the beam reducer; however, this would create a dark area at the center of the field of view where no FP can be positioned. In order to remove the zeroth order of diffraction while keeping the ability to place FPs throughout the field of view, a Fresnel lens pattern was superimposed on each calculated phase mask. The image of the desired pattern after the first lens of the beam reducer is thus formed on a different plane, while the zeroth order remains focused in the lens focal plane and can be stopped by a beam blocker without affecting the pattern. The position of the second lens is slightly moved to compensate for the pattern shift and to generate an image of the desired pattern in the objective’s focal plane.

Fig. 1

Spatial light modulator-two-photon microscope (SLM-2PM) scheme: (1) Ti:Sa laser (Chameleon Ultra II, Coherent); (2) beam expander ($f1=2.5\mathrm{cm}$, $f2=15\mathrm{cm}$); (3) SLM (X10468-07, Hamamatsu); (4) beam reducer ($f3=40\mathrm{cm}$, $f4=25\mathrm{cm}$); (5) pattern formation plane; (6) copper wire for undiffracted light removal; (7) objective (Zeiss Plan-APOCHROMAT $20\times$ 1.0 NA or Zeiss Plan-APOCHROMAT $40\times$ 1.0 NA); (8) dichroic filter (750 nm, 750dcspxr, Chroma); (9) pixelated detector (Coolsnap, Photometrics or MicamUltima, Scimedia). (10) High speed pixelated detector (MicamUltima, Scimedia).

Fluorescence light from the sample is then collected by the objective and separated from the excitation light through a dichroic mirror (750 nm, 750dcspxr, Chroma) and a bandpass filter. An image of the objective focal plane is formed by a tube lens ($f=125\mathrm{mm}$) on the detector. A flip mirror is used to create the image on one of the two pixelated sensors, a high-frequency complementary metal-oxide semiconductor (CMOS) camera (MicamUltima, Scimedia) for data acquisition or a high-resolution cooled charge-coupled device (CCD) camera (Coolsnap HQ, Photometics) for image formation. A brightfield transmission image of the sample was obtained through Köhler transillumination.

2.2.

Phase Mask Calculation

SLM-2PM is capable of illuminating the objective focal plane in several diffraction limited points. This is achieved through a reflective phase-only SLM, a pixelated liquid crystal device shifting of an arbitrary phase the radiation reflected by each pixel. In order to illuminate the focal plane with a desired intensity pattern, the phase shift applied by every pixel of the SLM must be calculated. The matrix of phase shift values (each ranging from 0 to $2\pi$), called the phase mask, is usually generated with slowly converging iterative algorithms.^9,18,19 For the purpose of SLM-2PM microscopy, a low number of iterations can be performed since a phase mask must be computed in a time compatible with experimental needs (seconds). Nonetheless, the phase mask must be accurate enough to guarantee that every single POI is illuminated by a diffraction-limited focal volume, as the efficiency of two-photon excitation decreases as the excitation volume increases.¹³

The classical Fourier transform-based GS algorithm¹⁸ proved to be the fastest available algorithm capable of reliably creating diffraction-limited focal volumes. A custom developed software based on the Python Scipy libraries²⁰ was used for phase mask computation. After the GS algorithm, a Fresnel lens pattern is summed to the obtained phase mask in order to avoid undiffracted light illumination on the sample (see Sec. 2.1). Changing the focal power of the Fresnel lens summed to the phase mask allows the illumination of planes above and below the objective focal plane.

2.3.

Two-Photon Image Formation

Two-photon images of the sample were obtained by computing a sequence of phase masks, each corresponding to a square grid pattern (sliding grid) of diffraction-limited (beam waist $\omega 0=400\mathrm{nm}$ for 1.0 NA objective) focal points (FPs) regularly spaced at a distance dFP. The sequence of phase masks is calculated so that the sliding grid performs a raster scan movement in which each spot scans a small square region. The image acquisition is performed by illuminating the sample with the sequence of phase masks. The pixelated detector acquires a stream of images synchronously with the SLM refresh, so that each frame shows the two-photon fluorescence signal elicited by one phase pattern of the sequence. For every acquired frame, the fluorescence intensity of each FP is integrated and stored. The final image is obtained by assigning to each pixel the fluorescence signal of the corresponding FP (Fig. 2).

Fig. 2

Image acquisition and reconstruction: (a) A set of square grid patterns covering the entire scanning area in a raster-scan sequence is generated and (b) the corresponding holograms are computed with the Gerchberg–Saxton algorithm. The series of holograms is shown on the SLM surface, and the resulting wavefront is projected on the sample. (c) An image of the sample is acquired for every frame of the sequence, and the fluorescence of each excitation volume is stored. (d) The stored fluorescence signals are converted to grayscale values and assigned to the corresponding pixels of the reconstructed image. The numbers in (a), (b), and (c) indicate the frame index.

The maximum refresh rate of our SLM is limited to 60 Hz, thus making a raster scan procedure sensibly slower than using a standard galvanometric scanner. In order to limit the time for image formation, the number of frames was minimized by reducing the spatial sampling of the sliding grid motion and increasing the number of FPs per frame. A large area of the field of view of the objective ($200\times 200\mu {\mathrm{m}}^2$ for a $40\times$ objective) was covered by optically oversampling the sliding grid ($\sim 2\text{pixels}/\omega 0$); $600\times 600$ FP positions are needed. This image sampling can be achieved with a sequence of 400 phase masks, representing a sliding grid of $30\times 30$ FPs, each moving on a $20\times 20$ raster scan pattern. A further increase in FPs number per frame mask is not feasible, as the excitation power per FP would become too low ($\ll 1\mathrm{mW}$ per FP) to determine detectable two-photon fluorescence. The minimum time required with this mode of image acquisition with a 60 Hz refresh rate is around 10 s. Moreover, depending on the sensitivity of the detector and on the intensity of the fluorescence signal, longer exposure times per image may be required to obtain a better signal-to-noise (S/N) ratio.

2.4.

Live Imaging

Live imaging is needed for the selection of the plane on which measurements should be performed. To achieve two-photon live imaging, a slightly undersampled sliding grid ($\sim 0.75\text{pixels}/\omega 0$) is projected on the sample at the maximum SLM refresh frequency, while the pixelated detector acquires a single image with an exposure time equals to the entire sliding grid sequence duration. Using a sliding grid of $28\times 28$ FPs moving in a $6\times 6$ raster scan pattern, the result is a live stream of two-photon images at about 2 frames per second.

2.5.

Multispot Illumination

The SLM-2PM system allows the user to select an arbitrary number of POIs chosen from the acquired image and to generate a hologram illuminating each of them with diffraction-limited FPs. In order to detect the signal of each POI, a pixelated detector must be used. For a bright two-photon dye, such as Rhodamine B in methanol, the fluorescence emission rate under 10-mW excitation power on a diffraction-limited spot (collection efficiency 1%, $\text{quantum yield}=1$) is about $\mathrm{64,000}\text{photons}/\mathrm{s}$.²¹ Such a flow can be easily detected by single-photon counting systems such as photomultipliers or single-photon avalanche diodes (SPAD). However, high-performance CCD and CMOS cameras can detect such a flow with a high S/N ratio. As the used CMOS camera has a dark noise of $400{\mathrm{e}}^{-}$ for 1 ms exposure and 65% quantum efficiency, around 20 to 30 molecules per excitation volume ($\sim 0.1\mu \mathrm{M}$) are required for an S/N ratio of 2. The system is, therefore, theoretically capable of detecting signal variations at 1 kHz.

2.6.

Slices Preparation and Calcium Imaging

Acute cerebellar slices ($200\text{-}\mu \mathrm{m}$ thick) were obtained from 18- to 25-day-old Wistar rats.²² Rats were anesthetized with halothane (0.5 ml in 2 L for 1 to 2 min) before being killed by decapitation. All experiments were conducted in accordance with international guidelines from the European Community Council Directive 86/609/EEC on the ethical use of animals. The cerebellum was gently removed, and the vermis was isolated, fixed with cyanoacrylate glue, and immersed into a cold (2 to 3°C) cutting solution. Slices were cut in the sagittal plane. The cutting solution contained (in mM): 130 K-gluconate, 15 KCl, 0.2 EGTA, 20 HEPES, 10 glucose, pH 7.4 with NaOH. Slices were then transferred to an oxygenated Krebs’ solution containing (in mM): 120 NaCl, 2 KCl, 1.2 ${\mathrm{MgSO}}_4$, 26 ${\mathrm{NaHCO}}_3$, 1.2 ${\mathrm{KH}}_2{\mathrm{PO}}_4$, 2 ${\mathrm{CaCl}}_2$, 11 glucose, pH 7.4 when equilibrated with 95% ${\mathrm{O}}_2-5\%{\mathrm{CO}}_2$. For extracellular bulk loading, slices were deposited onto the bottom of a small Petri dish ($35\mathrm{mm}\times 10\mathrm{mm}$) filled with 2 ml of artificial cerebrospinal fluid, and ventilated with 95% ${\mathrm{O}}_2/5\%$ ${\mathrm{CO}}_2$. A $50\text{-}\mu \mathrm{g}$ aliquot of Fura-2AM (Molecular Probes, Eugene, Oregon) was prepared in $48\mu \mathrm{l}$ dimethyl sulfoxide and $2\mu \mathrm{l}$ Pluronic F-127 (Molecular Probes).²³ The dye aliquot was then placed on top of the slice in the Petri dish and slices were incubated in the dark at 35°C to 37°C for up to 60 min. Slices were maintained at room temperature for at least 30 min before transferring them to the recording chamber. Slices were positioned on the recording chamber and fixed with a nylon mesh attached to platinum $\mathrm{\Omega }$-wire to improve tissue adhesion and mechanical stability. Perfusion of oxygenated Krebs’ solution at room temperature was continued during the recording session. During acquisition, neuronal activity was elicited by mossy fiber stimulation, performed through a tungsten bipolar electrode at 20 to 30 V. Stimulation trains consist of 10 pulses of 0.2 ms duration at a frequency of 50 Hz. In order to prove that the acquired signals were in fact elicited by neuronal activity instead of artifacts created by stimulation (i.e., tissue or solution motion), the tissue was perfused with $20\mu \mathrm{M}$ Gabazine, a GABA-A receptor blocker, and subsequently with $50\mu \mathrm{M}$ APV, an NMDA receptor blocker, and $10\mu \mathrm{M}$ NBQX, an AMPA receptor blocker. As expected,^24,25 Gabazine perfusion leads to an increment in the number of responding granule cells ($+89.1\%$ $\pm 11.3\%$; $n=15$ slices; $p<0.01$) and calcium signals’ amplitudes ($+129.9\%$ $\pm 2.5\%$; $n=15$ slices; $p<0.001$). Moreover, APV and NBQX perfusions completely and reversibly blocked all neuronal activities elicited by synaptic transmission.¹¹

2.7.

Experimental Procedure

The standard calcium imaging experiment was performed as follows:

3.1.

Lateral and Axial Resolutions

The generation of a phase mask through the GS algorithm is an iterative and slowly converging process. It is, therefore, difficult to obtain a phase mask corresponding to a truly point-like illumination. However, it is not necessary to create extremely accurate phase masks, as the dimension of the FPs is physically limited by the Abbe diffraction limit. The phase mask calculation software performed 20 GS iterations to create a single phase mask. To prove that the FPs were diffraction limited, an image of fluorescent polystyrene beads immobilized in agarose gel was acquired²⁶ (Fig. 3). The diameter of the polystyrene beads is 100 nm, while the theoretical lateral resolution of the system with 800 nm excitation light is 400 nm. A total of 25 microbeads profiles were fitted with a Gaussian curve on the $x$ and $y$ image axes. The mean FWHM of the curves was $420\pm 6\mathrm{nm}$, proving that the FPs’ dimensions were diffraction limited. The axial resolution was tested with the same polystyrene microbeads sample. 3-D images were obtained with a holographic method, shifting the axial position in $0.2\mu \mathrm{m}$ steps by adding a Fresnel lens to the phase masks of the sliding grid. A total of 25 microbeads axial profiles were fitted with a Gaussian curve. The mean FWHM of the obtained curves was $1.15\pm 0.08\mu \mathrm{m}$, proving that the system is diffraction limited on the axial direction. System resolution was slightly lower near the edges of the field of view, possibly due to the spherical aberration or the limited performance of the SLM. Fitting over 10 spheres at a distance $<20\mu \mathrm{m}$ from the edges of the field of view resulted in a resolution of $498\pm 8\mathrm{nm}$ laterally and $1.21\pm 0.10\mathrm{nm}$ axially.

Fig. 3

Optical resolution of SLM-2PM: (a) Image of a fluorescent polistyrene bead (100-nm diameter), immobilized in 1% agarose gel. The red line indicates the section used for fitting in (b). (b) Lateral profile of a fluorescent polystyrene bead, fitted with a Gaussian function. The arrow indicates the FWHM of the curve. (c) Axial section of a three-dimensional (3-D) image of fluorescent polystyrene bead. The focal plane was shifted in $0.2\mu \mathrm{m}$ steps by superimposing a Fresnel lens pattern on the sliding grid phase masks. The red line indicates the section used for fitting in (d). (d) Axial profile of a single bead, fitted with a Gaussian function.

3.2.

Point of Interest Selection Precision

The pointing accuracy of the system was assessed on a sample of under-resolved polystyrene beads with diameter 100 nm, immobilized in agarose gel. A single FP was positioned over each of 11 microbeads identified as bright spots on the full-frame acquired image. As a negative control, 11 FPs were randomly selected on the image. In response to the acquisition of a single exposure image, the FPs on the microbeads were all extremely bright. On the contrary, the randomly positioned FPs only showed a faint background fluorescence signal, proving that the POI selection accuracy is higher than half of the focal volume dimension (200 nm for 1.0 NA objectives). The confocal image of the microbeads and the multi spot illumination snapshot are shown in Fig. 4.

Fig. 4

Pointing precision of SLM-2PM: (a) Image (green) of fluorescent polistyrene beads (100-nm diameter, $\lambda \mathrm{em}=568\mathrm{nm}$), immobilized in 1% agarose gel. Red circles indicate the selected beads for the placement of focal point (FPs). Blue circles indicate the position of control FPs placed in dark areas of the image. (b) Single exposure acquisition of the pattern illuminating the beads indicated by red circles in (a). It can be observed that only the FPs placed over fluorescent beads elicit fluorescence signal. Grayscales were manipulated in both images to enhance the contrast. Both images were processes with a maximum intensity filter (4 μm radius) in order to increase the visibility of small objects.

3.3.

Two-Photon Calcium Imaging

The possibility to apply holographic two-photon imaging to neurobiological processes was assessed by detecting calcium signals in the granular layer of acute rat cerebellar slices. Slices were bulk-loaded with cell permeant Fura-2AM (see Sec. 2). Two-photon fluorescence excitation was set at 800 nm, while fluorescence light was selected through a $515/30$ bandpass filter. The Zeiss Plan-APOCHROMAT $20\times$ 1.0 NA objective was used, leading to a camera field of view of about $400\times 400\mu {\mathrm{m}}^2$.

Live imaging acquisition mode was used for confocal plane selection which was typically 50 to $60\mu \mathrm{m}$ deep in the cerebellar slice. A standard two-photon image of the selected plane was acquired with a 400 frames $30\times 30$ FPs scanning grid, covering a rectangular area of $220\times 300\mu {\mathrm{m}}^2$. Laser power was set to $4\mathrm{mW}/\mathrm{FP}$. CCD camera and SLM refresh rate were set to 50 ms per frame in order to obtain an image with a high S/N ratio, for a total image acquisition time of 20 s. The POIs were selected within the soma of each visible granule cell. During signal acquisition, laser power was set to $\sim 10\mathrm{mW}$ per FP. Calcium transients were recorded by the high-speed CMOS camera at frequencies up to 1 kHz. Granule cells showed calcium signals variations in the order of 5% to 10% ($\mathrm{\Delta }F/F_0$) peaking around 100 ms after stimulation and showing an S/N ratio of around 10 at 1 kHz (Fig. 5).¹¹

Fig. 5

High-frequency acquisition of calcium signals from multiple neurons in cerebellar slices: (a) Reconstructed image of the granular layer in an acute cerebellar slice bulk-loaded with Fura-2AM. Red crosses indicate the user selected positions for high-frequency signals acquisition. (b) Example of three different traces acquired at 1 kHz sampling rate, showing different kinetics, otherwise impossible to be discrimated at low-time resolution. The red line indicates stimulation onset. (c) Ensemble response of all 27 active cells in the sample, nonresponding granule cells are not shown. The green line indicates stimulation onset.

3.4.

Three-Dimensional Imaging

The SLM-2PM enabled the shifting of the focal plane in the axial direction only through SLM programming, without changing the objective position. This feature can be useful for experiments requiring high-mechanical stability (i.e., patch clamp recordings). The maximum possible shift was limited by the fact that the acquisition focus remained on a fixed plane, while the excitation pattern was digitally moved. The available axial range, with a 1.0 NA objective, was limited to about $20\mu \mathrm{m}$.

Figure 6 shows a Purkinje cell in an acute cerebellar slice, intracellularly loaded with Alexa488 sodium salt. The image was acquired by (1) sequentially shifting the excitation plane in 10 steps of $2\mu \mathrm{m}$; (2) exciting at 750 nm with a laser power of $4\mathrm{mW}/\mathrm{FP}$; (3) using the Zeiss Plan-APOCHROMAT $20\times$ 1.0 NA objective; (4) selecting fluorescence light with a $515/30$ bandpass filter; (5) acquiring each confocal image with a 400 frame $30\times 30$ FPs scanning grid, covering a rectangular area of $220\times 300\mu {\mathrm{m}}^2$; and (6) acquiring each frame with a 20 ms exposure of the camera (total acquisition time around 10 s for each confocal plane.). The same procedure was performed with the Zeiss Plan-APOCHROMAT $40\times$ 1.0 NA objective on a single dendritic arborization to increase spatial sampling [see Fig. 6(c)] enabling the visualization of single dendritic spines.

Fig. 6

3-D image of a Purkinje cell, intracellularly loaded with Alexa488. (a) Three confocal images acquired at planes $4\text{-}\mu \mathrm{m}$ apart. (b) Overlay of 10 different confocal images showing the entire Purkinje cell. (c) Overlay of 10 different confocal images ($2\mu \mathrm{m}$ step; $40\times$ magnification) of a dendritic arborization in which dendritic spines are visible.