Dynamic super-resolution structured illumination imaging in the living brain

Significance The brain is composed of cells that continually communicate with one another via electric and chemical signals. To understand nerve cells in a physiological context, we must study them in vivo, for which optical microscopy is an essential tool. In particular, much of this communication takes place at the nanoscale level and requires in vivo super-resolution microscopy. We applied adaptive optics to correcting sample-induced optical aberrations and optimized image acquisition and reconstruction to combat sample motion, which allowed us to adapt super-resolution structured illumination microscopy to in vivo imaging in the brains of zebrafish larvae and mice. With these optimizations, we were able to image dynamic processes at dendrites and synapses in the mouse brain at nanoscale resolution in vivo.


Minimization of aberrations via correction collar and adaptive optics.
Optical aberrations encountered for in vivo imaging of the mouse brain were minimized in two steps. First, the positioning of the mouse brain was optimized with direct wavefront sensing. Direct wavefront sensing was implemented as a Shack-Hartmann wavefront sensor of the multiphoton fluorescence signal positioned at a descanned conjugated pupil plane. For this step, the wavefront was measured immediately below the cranial window and decomposed into Zernike coefficients. We then iteratively adjusted the objective correction collar and the tilt of the brain to minimize spherical aberrations and coma introduced by the cranial window, respectively (3). Images obtained under this condition is considered as "No AO". Images taken under the "AO" condition had the residual wavefront error corrected with the deformable mirror.
SIM reconstruction algorithm. The raw data series with structured illumination were reconstructed into the super-resolution 2D linear SIM images using a custom algorithm. Our algorithm was based on diverse aspects of previously published strategies (Wiener deconvolution (4), illumination parameter estimation (5,6), non-iterative Wicker phase estimation (7), ungrading (8), OTF attenuation (9,10), the open-source FairSIM library (11)), and thus did not closely follow existing reconstruction algorithms. We therefore provide detailed flowcharts for our algorithms with and without the corrections for sample motion in SI Appendix, Fig. S4. Briefly, each raw image was pre-processed in the real space by subtracting the background using a rolling-ball algorithm (256 pixels in diameter) and by attenuating the edges with a sin 2 (x) function of 10-pixel half-period. If motion correction was required, images were registered to the average image of the series using cross-correlation and rigid translation. Illumination parameters were estimated once per experimental session using data obtained from a 2D sample of 0.1-µm beads as the reference. The parameters were then used to separate the positive and negative 1 st -order bands from the 0 th one in the Fourier domain for each orientation of the patterned illumination. After separation, the first-order bands were shifted to their appropriate location using the illumination vector p and a complex scaling was applied using the modulation depth a and phase φn information. At this point, the out-of-focus information was filtered out using the OTF-attenuation technique with a Gaussian notch filter (depth: 100%, full width at half maximum (FWHM): 5 cycles/µm) (9,10). This operation was implemented simultaneously to the ungrading (Wiener filtering without normalization) on individual bands (8). The bands were then added together for all the orientations of the patterned illumination. The Wiener denominator was applied only after this summation. Finally, the data was apodized with a power function at the cutoff frequency and Fourier-transformed to obtain the spatial domain super-resolution SIM image. For volumetric data, each axial plane was reconstructed independently.
Evaluating the illumination parameters. Illumination parameters (the illumination vector p, the modulation depth a, and the phase (φ)) were evaluated from raw image series of the reference bead samples. Two phase variables should be defined: 1) the global phase offset (φ global ), from the assumption of equidistant phase steps, describes the offset for all the phases for a given orientation of the illumination pattern, and 2) the phase (φn) is for an individual raw image. Following the pre-processing described above, bands were separated assuming equi-distant phase step. The illumination vector was first determined by finding the maximum of the complex cross-correlation of the overlap region between the 0 th and non-shifted +1 st bands. The coarse estimation was further refined to subpixel accuracy by maximizing the complex cross-correlation between the 0 th and shifted +1 st band with parabolic interpolation. This latter step was repeated 3 times. The cross-correlation for both the coarse and fine estimation of p was done on bands that were OTF-corrected (multiplied by the complex conjugated of the OTF) to minimize the noise contribution. For samples with no motion, a linear cross-correlation was then performed to estimate the modulation depth a and the global phase offset (φ global ). When necessary, the phase (φn) for each image was further refined with non-iterative Wicker phase estimation (7). If no sample motion was present, this step was not essential and had no visible impact on the final images. Finally, these parameters were applied to separate the bands for brain data. For samples with apparent motion, registering the background-corrected images was necessary, and so was evaluating the phases (φn) for each image from brain data with non-iterative Wicker phase estimation. Also, if repeated image acquisition was applied, a had to be estimated on beads from the bands newly separated with the updated phase information.
Mouse preparation. All experiments involving animals were conducted according to the National Institutes of Health (NIH) guidelines for animal research and were approved by the Institutional Animal Care and Use Committee at Janelia Research Campus, Howard Hughes Medical Institute. All mice (Thy1-GFP line M or C57Bl6/J) were at least 8-week old at the time of cranial window installation. In vivo imaging was performed on mice under isoflurane anesthesia (∼1.0% by volume in O2) at least 2 weeks after cranial window installation.

Cranial window installation.
Under isoflurane anesthesia (∼1.5% by volume in O2) and following aseptic techniques, a craniotomy of 3.5 mm in diameter was made over the left cortex. The dura was left intact. The hat-shaped cranial window, consisting of a 3.5 mm disk and a donut-like ring with 3.0-mm inner diameter and 4.5-mm outer diameter, both made from coverglass (Fisher Scientific, No. 1.5, 160-190 µm thick), was prepared by applying a UV adhesive between the two components and curing it with a UV lamp. The disk was embedded into the craniotomy and the window fixed in place with cyanoacrylate glue and dental acrylic applied on the ring. In order to keep the head stable, a titanium head-post was attached to the skull with cyanoacrylate glue and dental acrylic. For membrane imaging experiments, C57Bl/6 mice were injected during cranial window installation with viral vectors (AAV2/1-FLEX-ChR2-GFP and AAV2/1-SYN-CRE) into layer 4 of the primary visual cortex (30 nL, 500 µm below dura) with a glass pipette with a 20 µm opening beveled at 45 • and back-filled with mineral oil. For calcium imaging experiments, the same procedure was followed but mice were injected with AAV2/1-SYN-FLEX-GCaMP6s and AAV2/1-SYN-CRE. Imaging was performed at least 3 weeks following viral vectors injection.

Potassium chloride and bicuculline injection into mouse cortex.
For potassium chloride (KCl, Sigma-Aldrich, No. 7447-40-7) injection, the window was installed as described above, except that it consisted of a disk of 3.0-mm diameter with a 0.1-0.2 mm opening approximately 1.0 mm away from the center (drilled by a laser cutter, Universal Laser Systems, PLS6.75). The opening was sealed with silicone elastomer (WPI, Kwik-Cast sealant) prior to experiment. KCl was diluted to 50 mM in sterile buffered saline solution and loaded into a glass micropipette. An injector attached to a syringe and controlled by a syringe pump was used to administer 200 nL of the compound intraparenchymally at a depth of 50 µm. The animal was imaged immediately after injection at a location ∼1 mm away from the injection site. Imaging was also performed every 5 min for 10 min prior to KCl injection and only minor structural changes, consistent with organelle motion, were observed. The same procedure was followed for bicuculline injection (TOCRIS, No. 0130) for which 200 nL of a solution at 500 µM bicuculline in sterile buffered saline solution was administered.
Fixed mouse brain slices. Mice were completely sedated with isoflurane before being transcardially perfused with 10 mL of phosphate-buffered saline (PBS), followed by 50 mL of paraformaldehyde (PFA) at 4%. After perfusion, the brain was dissected and post-fix in 4% PFA for 24 hrs at 40 • C, then washed with PBS 3 times. The brain was embedded in 5% agarose in PBS, then cut on a vibratome (Leica Vibratome 1200) at a thickness of 100 µm. Sections were directly mounted on slides for drying (24 hrs), then were rehydrated with PBS before being mounted and covered with Vectashield Hardset™ (H-1400). A coverglass (Fisher Scientific, No. 1.5, 160-190 µm thick) was also placed on top. Brain slice samples were placed on a goniometer platform (Thorlabs, GNL10) for imaging and were orthogonal to the objective optical axis unless otherwise mentioned.
Zebrafish. Zebrafish larvae of a pigmentation mutant, casper, were used. At 5 days postfertilization, larvae were briefly anesthetized by 0.02% tricaine (MS-222, Ethyl 3-aminobenzoate methanesulfonate) in fish system water. A fine-tip tungsten needle was dipped in fluorescent dye (20% weight/volume) of Alexa Fluor 488 conjugated with dextran, MW 10,000 (Thermo Fisher Scientific), and the fluorescent dye was injected into larvae by the needle after removing the excessive anesthetic solution. Injection sites were located in the spinal cord at the 10 th spinal segment to label spinal projection neurons. After overnight incubation in fish system water, larvae were anesthetized by 0.02% tricaine, embedded in 1.6% (weight/volume) low-melting-point agar and the agar covering the imaging target was removed by a pair of fine forceps.
Beads. Fluorescent beads (Invitrogen, FluoSphere™ carboxylate-modified microsphere, yellow-green, 505/515) of 0.1 µm in diameter were used for illumination parameter estimation. The stock solution was diluted to 1:7000 in deionized water and 2 µL of the dilution was added on a microscope slide covered with a thin layer of poly-L-lysine hydrobromide (Sigma, P1399). 1.0-µm beads were prepared in the same manner, but at a 1:100 dilution in deionized water. When putting a coverglass (Fisher Scientific, No. 1.5, 160-190 µm thick) over a bead sample, a drop of water was added on the beads to avoid air gaps. Bead samples were placed on a goniometer platform (Thorlabs, GNL10) for imaging and were flat unless otherwise mentioned. An example image from a SIM raw data series for a 2D bead sample that was carefully aligned, so that all the beads were in focus simultaneously, and (Right) the OTF of the reconstructed SIM image using illumination parameters derived from the raw data series. We obtained an accurate estimate of the illumination periods and orientations, leading to a SIM OTF closely resembling that predicted by theory and SR images of beads that were symmetric and artifact-free. (B) (Left) An example image from a SIM sequence for a 0.1-µm bead sample that was tilted at 6 • to generate a 3D fluorescence distribution and introduced out-of-focus background, and (Right) the OTF of the reconstructed SIM image of the tilted bead sample using illumination parameters derived from the raw data series obtained from this tilted sample. Substantial errors were introduced in the SIM reconstruction, yielding artifacts such as images of split beads.        Movie S1. Dynamic structural imaging. Time-lapse movie with deconvolved widefield microscopy (dWF, left) and SIM (right) of a dendrite and its spines after KCl injection in the brain of a GFP-expressing mouse (Thy1-GFP line M). Scale bar: 2 µm.
Movie S2. Functional calcium imaging. Movie of GCaMP6 fluorescence variation with deconvolved widefield microscopy (dWF, left) and SIM (right) in a dendrite and its spine after bicuculline injection. Scale bar: 2 µm.