2.1.

Confocal Airyscan Imaging Is Not Practical for a Detailed Volumetric Reconstruction of Extended Fluorescently Labeled Connectivity Maps

In a first approach, we explored the capabilities of state-of-the-art confocal laser scanning microscopy to obtain detailed high-resolution volumetric images of extended DG neuronal networks. For imaging EGFP-labeled DG granule cells (GC), we used a Zeiss LSM 880 with Airyscan detector, which is able to achieve a lateral full width at half maximum (FWHM) resolution of 120 nm under very optimal conditions.²⁷ Figure 1(a) shows a single high-resolution confocal plane of a $70\text{-}\mu \mathrm{m}$-thick coronal section through the dorsal DG, containing EGFP-labeled granule cells. The Airyscan-processed image was acquired using a water immersion (WI) objective lens with a numerical aperture (NA) of 1.2 and a pixel size of 50 nm. This image was generated from $5\times 8$ individual tiles comprising $2048\times 2048\text{pixels}$ each. Tiles were acquired with 10% overlap and stitched using the ZEN software (ZEN Black 2.3 SP1, Carl Zeiss Microscopy GmbH). The magnification demonstrates the excellent resolution of the Airyscan processed image, that allows to separate densely packed dendrites [Fig. 1(b)]. However, we encountered two principal limitations for extended volumetric neural network reconstructions performing these imaging experiments. First and most importantly, it is well known that image acquisition of confocal microscopes is inherently slow due to the sequential, pixelwise data acquisition. Thus, the total imaging duration of a single confocal plane with a resolution of $2048\times 2048\text{pixels}/\text{image}$ covering a field of $102.4\times 102.4\mu {\mathrm{m}}^2$ required about 10.1 s, when no averaging was performed. Acquisition of a complete image field of $742\times 452\mu {\mathrm{m}}^2$, as shown in Fig. 1(a), required 400 s, since the tiles were acquired with 10% overlap for stitching. Thus, imaging of a $32\text{-}\mu \mathrm{m}$-thick volume with this field size using the optimal axial step size of $0.21\mu \mathrm{m}$ according to the Nyquist theorem required 1020 min or 17 h. Extrapolation of this data acquisition time to the total DG volume of $1{\mathrm{mm}}^3$ yields a total duration of more than 1900 h. Moreover, post-hoc 3-D rendering of a $96\times 165\times 32{\mu \mathrm{m}}^3$ large region of interest (ROI) demonstrated that the achieved resolution was not sufficient to allow the unambiguous evaluation of the 3-D structure of the densely labeled granule cells due to melding together of individual neurites, although the image acquisition was done as carefully as possible. An additional increase of the applied laser intensity from 0.65 to $1\mathrm{kW}/{\mathrm{cm}}^2$ improved the signal-to-noise ratio but not the effective resolution. In addition, the higher laser intensity caused notable bleaching of the sample during confocal image acquisition (see Sec. 4).

Fig. 1

Coronal section of a dorsal mouse dentate gyrus containing EGFP-expressing granule cells imaged by high-resolution confocal microscopy. The endogenous EGFP fluorescence was enhanced by antibody staining against EGFP (see Sec. 4). (a) A single confocal plane acquired using an Airyscan detector and a $40\times$ 1.2 NA WI imaging objective lens. The total field size in (a) was $742\times 452\mu {\mathrm{m}}^2$, achieved by combining $8\times 5$ single image stacks comprising $2048\times 2048\text{pixels}$ in each frame. The individual tile stacks were acquired with 10% spatial overlap to enable efficient stitching using the algorithm described in Ref. 28. (b) Magnification of the ROI marked in Fig. 1(a).

In conclusion, we here used the latest high-resolution confocal laser scanning instrument with Airyscan detector and were able to laterally resolve neuronal structures as small as 160 nm (FWHM, see Sec. 4). The achieved axial resolution was 810 nm.

2.2.

Light-Sheet Fluorescence Expansion Microscopy for Volumetric Imaging

LSFM is known for its high-speed imaging requiring only low irradiances.²⁹ Unfortunately, volumetric imaging of mouse brain tissue by LSFM is complicated since it requires transparent specimens due to the necessity to illuminate the object from the side with a thin sheet of light to generate the optical sections. To overcome this problem and to increase the effective resolution, we decided to expand the mouse brain sections before light sheet imaging to achieve virtual optical super resolution in large tissue volumes at an imaging speed outperforming the currently fastest point scanning devices. Note that expansion of tissue has the great advantage that the heavy adsorption of water renders the samples completely transparent.

Expansion of a $70\text{-}\mu \mathrm{m}$-thick anti-EGFP-labeled mouse brain slice from a PROX1-cre mouse, injected with rAAV-DIO-EGFP-WPRE to achieve selective expression of EGFP in DG granule cells was performed according to Ref. 16. The refractive index of the sample gel matched almost that of pure water and the expanded samples became completely transparent, excellently suited for LSFM using WI detection objectives.^15,17 Sample expansion was isotropic¹⁵ and enlarged our DG sample by a factor of 3.9 (see Sec. 4). We employed a custom-built LSFM to image respective samples of expanded DG sections. A $25\times$ objective lens with a NA of 1.1 was used together with a $1.5\times$ magnification lens and yielded a field of view of $355\times 355\mu {\mathrm{m}}^2$. Thus, images of expanded DG samples had to be acquired in a mosaic fashion. For each mosaic tile, a $z$-stack of 300 images with a $\mathrm{\Delta }z=1\mu \mathrm{m}$ was acquired with a frame rate of 1.8 Hz. After imaging, the tile stacks were contrast-adjusted and stitched yielding views of the dorsal DG sections of unprecedented resolution (Fig. 2). Figure 2(a) shows a single LSFM plane of the stitched volume at a depth of $151\mu \mathrm{m}$. The field size was $3600\times 1240\mu {\mathrm{m}}^2$ after expansion corresponding to a size of about $920\times 320\mu {\mathrm{m}}^2$ before expansion considering the magnification factor of $3.9\pm 0.3$ (see Sec. 4). Figure 2(b) shows the magnification of the ROI marked in Fig. 2(a), and the possibility to detect individual spines demonstrates the remarkable resolution of the image. Note, that only small segments of the dendrites are visible in this single optical section. However, a maximum projection of the complete region that was imaged revealed the 3-D outline of entire dendritic fields [Fig. 2(c)]. The high virtual resolution of 100 nm laterally and 415 nm axially (see Sec. 4) and the superb contrast allowed to trace and segment the dendritic trees of three individual granule cells revealing their detailed morphoanatomical shape [Fig. 2(d)]. Thus, we achieved a lateral resolution about twofold lower than STED microscopy (25 to 80 nm),³⁰ but clearly better than structured illumination microscopy (130 to 160 nm)^30,31 or high-resolution confocal microscopy ($>120\mathrm{nm}$,²⁷) using an Airyscan detector. Our axial resolution was lower than achievable by STED and structured illumination microscopy (150 to 600 nm and 250 nm, respectively³⁰), but higher than for high-resolution confocal microscopy ($>550\mathrm{nm}$,^27,30 and Sec. 4).

Fig. 2

Expanded, antibody-stained mouse brain slice imaged by LSFM. The sample was expanded and imaged with a custom-built light-sheet microscope. In total 70 $z$-stacks with a step size of $1\mu \mathrm{m}$, covering a depth of $275\mu \mathrm{m}$ with 30% overlap were stitched to generate the image. (a) Single plane of the stitched volume at a depth of $65\mu \mathrm{m}$. The total volume shown after expansion was $3600\times 1240\times 275\mu {\mathrm{m}}^3$ (Video 1, MP4, 32 MB [URL: https://doi.org/10.1117/1.NPh.6.1.015005.1]). (b) Magnification of the region marked in (a), lateral field size $254\times 492\mu {\mathrm{m}}^2$. (c) Maximum projection of the selected region marked in (a) comprising 76 slices of the stack, lateral field size $254\times 492\mu {\mathrm{m}}^2$. Single granule cells and dendrites can well be distinguished and separated. The stack was median-filtered to remove staining artifacts before the maximum projection. (d) Segmentation and tracing of the neurites of three selected granule cells (Video 2, MP4, 19 MB [URL: https://doi.org/10.1117/1.NPh.6.1.015005.2]). Left panel: segmented GCs in the neuronal network, right panel: segmented GCs with traced neurites.

Note that a sample volume of $1{\mathrm{mm}}^3$ would be transformed by expansion to a volume of ($3.9\times 3.9\times 3.9{\mathrm{mm}}^3$). The step size of imaging should be $0.8\mu \mathrm{m}$ in order to fulfill the Nyquist criterion, since the FWHM of our axial resolution was $1.62\mu \mathrm{m}$ (see Sec. 4). This yields a total number of 5000 sections. The lateral extension of such a sample requires the use of $12\times 12$ tiles each having a field of view of $355\mu \mathrm{m}$. Thus, a total number of 720.000 images will cover the complete volume, what would require 111 h for imaging at a rate of 1.8 Hz. Thus, light-sheet fluorescence expansion microscopy (LSFEM) imaging of a complete dentate gyrus is faster by a factor of 17 compared to Airyscan confocal microscopy.

Figure 1, which was acquired by Airyscan confocal laser scanning microscopy, showed a brain region of similar spatial extension before tissue expansion. Visual comparison between Figs. 1 and 2 readily reveals the gain in contrast and axial resolution achieved by light-sheet expansion microscopy. This was demonstrated by plotting line profiles of the dense neurite region of Figs. 1(b) and 2(b) (Fig. 3). The figure demonstrates the lower number of dendrites containing in an optical slice of the LSFEM with its effective thickness of 415 nm compared to the optical slice of the Airyscan image with its axial extension of 810 nm.

Fig. 3

Contrast in Airyscan and LSFM microscopy images. (a) Line profile determined in a $0.3\text{-}\mu \mathrm{m}$-wide line with a length of $60\mu \mathrm{m}$. The right panel shows the position of the line in Fig. 1(b). (b) Line profile determined in a $3.9\times 0.3\mu \mathrm{m}$ wide line with a length of $3.9\times 60\mu \mathrm{m}$. The right panel shows the position of the line in Fig. 2(b).

2.3.

Sparsely Labeled Expanded Mouse Brain Slices

GC in brain slices of dorsal DG display an extremely narrow arrangement of cell bodies and neurites, as can be seen in Figs. 1 and 2. Individual neurites may optically be resolved, yet, tracing them over their entire spatial extension is possible only in rare cases. This is due to the fact that entire neurons are unlikely to be entirely contained within a single tissue section and, although we achieve effective super resolution, melding together of closely together lying individual structures can virtually not be avoided in densely packed structures. One solution to this problem is using a more “sparse labeling” approach. To this end, we infected the hippocampal CA1 region of wt mice with a rAAV-expressing EGFP under control of the human synapsin1 promoter by stereotaxic virus injection and stained afterward with an antibody against EGFP (for details, see Sec. 4). The use of low virus titers (10E8/ml) resulted in a stochastic and relatively sparse labeling of CA1 pyramidal neurons. This experimental approach allowed us to unequivocally identify individual dendritic networks and follow them over 1.3 mm in this experiment [Fig. 4(a) and Video 3]. Note that the image stack shown in Fig. 4 was obtained using an axial step size of only $0.3\mu \mathrm{m}$ revealing even the tiniest dendritic structures in nanoscale resolution. E.g., individual dendritic spines can be visualized in the high-resolution images [Figs. 4(b)–4(d) and Video 4]. Figure 4(e) demonstrates that even spine necks can be recognized.

Fig. 4

Dendritic segments of sparsely labeled pyramidal neurons in CA1. (a) Maximum projection of a total of 24 stacks with an axial step size of $0.3\mu \mathrm{m}$ covering a depth of $450\mu \mathrm{m}$ were acquired and stitched together (Video 3, MP4, 39 MB [URL: https://doi.org/10.1117/1.NPh.6.1.015005.3]). The sparse labeling and small axial imaging step size allowed to reconstruct the labeled granule cells and dendrites over distances of 1.3 mm after expansion. (b) Magnification of the large ROI marked in (a). The high virtual optical resolution allowed to identify finest detail, e.g., the morphology of individual synaptic spines. (c) 3-D surface rendering of a region of the data shown in (b), the dimensions were $256\times 152\times 205\mu {\mathrm{m}}^3$. (d) Deconvolution of the image data using the experimental PSF yielded a significant increase in data quality (Video 4, MP4, 4 MB [URL: https://doi.org/10.1117/1.NPh.6.1.015005.4]). (e) A maximum projection of 250 frames of a magnified region of the image stack shown in (a) (see dashed ROI) containing a single dendrite demonstrates that even dendritic spine necks (arrows) can be recognized.

Notably, these images were acquired using a $25\times$ NA1.1 objective lens combined with a 1.5× magnification, which achieved a theoretical FWHM lateral optical resolution of 310 nm and a nominal FWHM axial resolution of 1010 nm at the emission maximum of the used dye, ATTO647N (664 nm). The expansion process magnified all labeled object structures by a factor of 3.9, thus transforming these values into an expected virtual lateral and axial resolution of $\approx 80$ and 260 nm, respectively. Experimentally, we did not reach the theoretical expectation, but rather determined values of 135 and 590 nm for the FWHM values of the lateral and axial virtual optical resolution, respectively (see Sec. 4). Presumably, this discrepancy was partly due to the fact that a dipping objective was used for imaging because it featured a high NA and a long working distance. It contained a correction collar for adjusting it also for the use with a cover slip and was employed in this way. We suspect, however, that in this configuration the objective did not deliver its full performance. Hence, we assume that it is optimal for imaging expanded samples using long working distance dipping objective lenses in an upright optical detection path.

2.4.

Multicolor Imaging of Sparsely Labeled DG Neurons After Tissue Expansion

In order to show the possibility of neural connectivity mapping, we performed experiments demonstrating multicolor labeling of pre- and postsynaptic proteins. Briefly, we labeled virus-infected EGFP-positive pyramidal neurons with antibodies against EGFP to outline neurite morphology and subsequently stained for the postsynaptic protein shank2 (Alexa568, green) and the presynaptic protein Bassoon (Alexa 647, red). Figure 5(a) shows a region with an axial extension of $390\mu \mathrm{m}$, which was imaged using 1300 sections. Figure 5(b) shows a magnification of the ROI marked in (a) as maximum projection of 13 images extending over a distance of $3.6\mu \mathrm{m}$. Figure 5(c) shows a contrast-enhanced magnification of a region marked in (b) by a white arrow with a lateral field size of $17.3\times 17.3\mu {\mathrm{m}}^2$, indicating synaptic connectivity. Figure 5(d) shows a 3-D surface rendering of the synapse region shown in (c) to demonstrate the spatial arrangement of the fluorescent labels within the context of the synapse. Using the image data shown in Fig. 5, we measured the distances between 10 shank2-bassoon pairs [see Fig. 5(d)]. The maximum intensity projection of those regions was used from 10 to 15 planes. The intensity profiles—obtained by drawing a line perpendicular to the synapse surface—were fitted by Gaussian functions to determine the respective maximum positions of the label distribution. This yielded the expected distance of $160\pm 50\mathrm{nm}$ between pre- and postsynaptic proteins. Altogether, Fig. 5 demonstrates the capability of our approach to visualize details of synaptic connectivity, which is a prerequisite for systematic connectivity analysis.

Fig. 5

Multicolor imaging of sparsely labeled pyramidal neurons in CA1 imaged by LSFEM. (a) Maximum intensity projection of an expanded mouse brain slice expressing EGFP. The endogenous EGFP fluorescence was enhanced by antibody staining against EGFP, the secondary antibody was conjugated to Alexa 488. The total axial extension was $390\mu \mathrm{m}$. (b) Magnification of the ROI marked in (a). Dendrites were stained with Alexa488 (blue), postsynaptic proteins (shank2) stained with Alexa568 (green), presynaptic proteins (bassoon) stained with Alexa647 (red). The shown images were deconvolved. (c) Magnification a region indicated in (b) (white arrow), lateral field size $17.3\times 17.3\mu {\mathrm{m}}^2$, indicating synaptic connectivity. The right panel in (c) shows the intensity profiles along the dotted line. (d) 3-D surface rendering of the synapse region shown in (c).

In addition, we generated DG samples containing sparsely expressing EGFP-positive GCs in order to visualize the mossy fibers within hilus of the DG. We then performed antibody stainings against parvalbumin on these EGFP-positive, anti-EGFP stained sections to selectively label hilar GABAergic interneurons in red (Alexa 568). We then performed two-color LSFEM volumetric imaging and used differential color surface rendering to identify GABAergic cells as postsynaptic targets of mossy fibers boutons. Figures 6(a) and 6(b) show an overview of the DG hilus with EGFP-positive mossy fibers in green and parvalbumin-positive GABAergic interneurons in red. Figure 6(c) shows a zoom identifying filopodia on mossy fiber boutons that contact dendrites of GABAergic hilar interneurons. Collectively, we present evidence that the spatial resolution using LSFEM is sufficient to study neuronal connectivity within the context of large neuronal ensembles.

Fig. 6

Two-color imaging of mossy fibers and GABAergic interneurons in DG. (a) Mossy fibers in the hilus area expressing EGFP. The endogenous EGFP fluorescence was enhanced by antibody staining against EGFP, the secondary antibody was conjugated to Alexa 488 (green). Parvalbumin staining identified GABAergic interneurons shown in red (Video 5, MP4, 65 MB [URL: https://doi.org/10.1117/1.NPh.6.1.015005.5]). 1500 optical slices were acquired with a step size of $0.3\mu \mathrm{m}$, the shown data were deconvolved. Volume size $456\times 945\times 390\mu {\mathrm{m}}^3$. (b) Side view of the data shown in (a). (c) Segmented parvalbumin cells and mossy fibers reconstructed in 3-D. Magnification of the ROI marked in (b), showing connection between the cells.

For performing the differential color surface rendering, the volume was deconvolved using Huygens with the corresponding measured PSF for each channel. The deconvolved image data were loaded into Imaris to perform a 3-D rendering of the volume of interest, $171.4\times 343.2\times 253.5\mu {\mathrm{m}}^3$ for the GABAeric cells and $71\times 57\times 147\mu {\mathrm{m}}^3$ for the mossy fibers boutons. The connectivity area shown in Fig. 6 was obtained using the surface tool over both channels choosing a surface grain size of $0.346\mu \mathrm{m}$ and a manually defined intensity threshold.

4.1.

Mice

Mice were maintained on a 12-h light/12-h dark cycle with food and water always available. All the experiments were carried out in accordance with the German animal protection law (TierSCHG), FELASA and were approved by the animal welfare committee of the University of Bonn.

C57BL/6 mice were purchased from Charles River (Sulzfeld, Germany). The Prox-1 cre line³⁶ is maintained in the local animal facility (Haus für Experimentelle Therapie).

4.2.

Virus Injection

Viral injections were performed under aseptic conditions in Prox1::cre or C57BL/6 mice up to 10 months old. The mice were anesthetized with a mixture of Fentanyl (Rotexmedica, Germany), Midazolam (Rotexmedica, Germany), and Domitor (Orion Pharma, Finland) via intraperitoneal injection ($0.05/5.0/0.5\mathrm{mg}/\mathrm{kg}$). Analgesia ($0.05\mathrm{mg}/\mathrm{kg}$ of buprenorphine; Buprenovet, Bayer, Germany) was administered intraperitoneally prior to the injection, and Xylocain (AstraZeneca, Germany) was used for local anesthesia. Stereotaxic injections were performed using an injection frame (WPI Benchmark/Kopf) and a microprocessor-controlled minipump (World Precision Instruments, Sarasota, Florida), 1000 nl of the viral solution was injected bilaterally into the hippocampus (rAAV-DIO-eGFP; rAAV-syn-GFP; coordinates: rostrocaudal: $-2.1\mathrm{mm}$ from the Bregma; mediolateral: $\pm 1.2\mathrm{mm}$ from the midline; dorsoventral: $-2.1\mathrm{mm}$). After the injection, the scalp was sutured with PERMA-HAND Silk Suture (Ethicon), and an antibacterial ointment (Refobacin^®, Almirall, Germany) was applied, followed by the intraperitoneal injection of a mixture of Naloxon (B. Braun, Germany), Flumazenil (B. Braun, Germany), and Antisedan (Orion Pharma, Finland) ($1.2/0.5/2.5\mathrm{mg}/\mathrm{kg}$). To prevent the wound pain, analgesia was administered on the 3 following days.

4.3.

Perfusion and Slicing

Three weeks after the injection, the mice were anesthetized with a mixture of xylazine ($10\mathrm{mg}/\mathrm{kg}$; Bayer Vital, Germany) and ketamine ($100\mathrm{mg}/\mathrm{kg}$; bela-pharm GmbH & Co. KG, Germany). Using a peristaltic pump (Laborschlauchpumpe PLP33, Mercateo, Germany), the mice were perfused transcardially with $1\times$ PBS followed by 4% paraformaldehyde (PFA) in 1 × PBS. All solutions were stored on ice prior to the perfusion. Brains were removed from the skull and post-fixed in 4% paraformaldehyde overnight (ON) at $+4°\mathrm{C}$. After fixation, the brains were moved into PBS containing 0.01% sodium azide and stored at $+4°\mathrm{C}$.

4.4.

Immunochemistry

Immunohistochemistry was performed according to standard protocols. Briefly, the fixed brains were sectioned coronally (70 or $100\mu \mathrm{m}$) using a vibratome (Leica VT1000 S). To prevent unspecific binding of the primary antibody, the sections were incubated in blocking buffer ($1\times$ PBS containing 0.1% TritonX-100 and 5% normal goat serum) on a shaker for 6 h at room temperature. After blocking, the sections were incubated ON in primary antibody (chicken anti-GFP; 1:5000 in blocking buffer; Abcam, ab13970; rabbit anti-parvalbumin diluted 1:1000 in blocking buffer, Swant, PV27) at $+4°\mathrm{C}$. The following day, slices were washed at room temperature in blocking buffer three times for 20 min and incubated ON in Alexa Fluor^® 488-conjugated goat antibody against chicken IgY, or Alexa Fluor^® 568 goat anti-rabbit (1:400 in blocking buffer; Life Technologies, A-11039, A-11011) at $+4°\mathrm{C}$.

4.5.

Labeling of Pre- and Postsynaptic Proteins

For pre- and postsynaptic labeling sections were stained against presynaptic active zone protein bassoon (primary antibody mouse anti-bassoon diluted 1:200 in blocking buffer, Enzo Life Sciences, SAP7F407; secondary antibody goat anti-mouse conjugated with biotin diluted 1:200 in blocking buffer, Jackson ImmunoResearch, 115-067-003 followed by Alexa Fluor^® 647 Streptavidin, Jackson ImmunoResearch, 016-600-084) and postsynaptic scaffold protein Shank2 (primary antibody guinea pig anti-Shank2 diluted 1:200 in blocking buffer, Synaptic Systems, 162 204; secondary antibody goat anti-guinea pig conjugated with Alexa Fluor^® 568 diluted 1:200 in blocking buffer, Life Technologies, A-11075).

4.6.

Gelation and Expansion

The expansion microscopy protocol was adopted from Refs. 15 and 16. The immunostained sections were incubated with 1 mM methylacrylic acid-NHS (Sigma Aldrich) linker for 1 h. After washing three times in PBS, the sections were incubated for 45 min in the monomer solution (8.6% sodium acrylate, 2.5% acrylamide, 0.15% N,N’- methylenebisacrylamide, and 11.7% NaCl in $1\times$ PBS), followed by 2 h incubation at 37°C in a gelling solution. The gelling solution was prepared by adding 4-hydroxy-TEMPO (0.01%), TEMED (0.2%) and ammonium persulfate (0.2%) to the monomer solution. After the gel formation, the samples were incubated at 37°C ON in the digestion buffer (50 mM Tris, 1 mM EDTA, 0.5% Triton-X100, 0.8M guanidine HCl, and $16\mathrm{U}/\mathrm{ml}$ of proteinase K; pH 8.0). The next day, the digestion buffer was removed and the sections were washed with deionized water for 2.5 h. Subsequently, the samples were stored in deionized water at $+4°\mathrm{C}$ until imaging. For microscopic examination, the expanded gel sample was fixed on the bottom coverslip of the imaging chamber with poly-l-lysine to avoid movements during the measurement and the chamber was filled with deionized water.

4.7.

Light Microscopy

For light-sheet microscopy, we used a custom-built setup based on a Nikon Eclipse Ti-U inverted microscope (Nikon, Düsseldorf, Germany). A previous version of this instrument was described by Ref. 26. Scanned illumination and sample stage were custom-designed, whereas the Eclipse Ti-U provided the detection path. For fluorescence excitation, three fiber-coupled lasers emitting at 488, 638 (Obis LS, Coherent, Santa Clara), and 532 nm (LasNova Green 50 Series, Lasos, Jena, Germany) were employed. Laser light was regulated by an acousto-optical tunable filter (AOTF, TF-525-250-6-3-GH18A, Gooch&Housego, Ilminster). The horizontally scanned light sheet was generated by a galvanometer system with silver-coated mirrors. The adjustment of the beam waist position within the sample chamber was realized by a relay optics mounted on a linear precision stage. The beam waist in the object plane was adjusted to a $1/{\mathrm{e}}^2$ diameter of $7.9\pm 0.02\mu \mathrm{m}$ for the 488 nm, $8.3\pm 0.02\mu \mathrm{m}$ for the 532 nm, and $9.5\pm 0.02\mu \mathrm{m}$ for the 638-nm laser lines. The irradiance at the sample plane amounted to 240, 340, and $380\mathrm{W}/{\mathrm{cm}}^2$ at 488, 532, and 638 nm, respectively. Our custom-designed sample chamber featured an illumination window with coverslip thickness (0.17 mm) and a bottom glass coverslip for light detection. The sample chamber could be moved in three spatial directions by motorized microtranslation stages. For illumination we used a Mitutoyo $10\times$ NA 0.28 air objective. The detection objective was a $25\times$ NA 1.1 WI objective with cover slip correction (Nikon). The Eclipse Ti-U had a built-in optional $1.5\times$ magnification, which was employed as indicated. We used a sCMOS camera ($2048\times 2048\text{pixels}$, pixel size $6.5\mu \mathrm{m}$, Orca Flash 4.0 V2, Hamamatsu Photonics K.K., Hamamatsu City, Japan) for data acquisition in rolling shutter mode. The object field pixel size was ${(173\mathrm{nm})}^2$ corresponding to a field of view of ${(355\mu \mathrm{m})}^2$. All electronic components were controlled by a custom-written LabView program.

The sample was mounted in a custom built sample chamber. The excitation laser beam entered the chamber through the window at the left-hand side, which was formed by a conventional $24\times 24\mathrm{mm}$ coverslip with 0.17 mm thickness glued to the outer chamber wall. The sample was mounted on a coverslip, which was slid into the interstice in the chamber walls from the front. Eventually, the chamber was closed by a third coverslip at the front side and filled with deionized water. For the waterproofing of the chamber, the slits between the coverslips and the chamber walls were sealed with 2% agarose. The chamber was mounted on a three-axis motor.

The image acquisition speed of our instrument was not optimized for high speed, since it was an experimental system. Image acquisition time depended on the line exposure time and the slit width of the rolling shutter, respectively, the waiting time before line activation. For a typical exposure time of 20 ms per line and a confocal slit width of 256 pixels, we reached a total exposure time of 180 ms for one frame. Due to various signal processing tasks of the instrument, the total time required to acquire one frame amounted in summary to about 500 ms. Altogether, the acquisition time for a stack of 400 frames amounted to 222.7 s corresponding to a frame rate of 1.8 Hz. Acquisition of a complete stack comprising 1500 images using a step size of $0.3\mu \mathrm{m}$ at the irradiance of $240\mathrm{mW}/{\mathrm{cm}}^2$ resulted in a bleaching of 2% in a central image.

The sample size greatly exceeded the lateral object field size of ${(355\mu \mathrm{m})}^2$. Therefore, the data were acquired in a mosaic-like fashion. To this end, we acquired $N$ image tiles in the direction perpendicular to the laser beam ($y$ direction). The propagation direction of the laser beam corresponded to the $x$ direction. This ensured that the focus of the beam remained always in the center of the field of view. To enable a subsequent stitching of the tiles, two neighboring tiles had 30% spatial overlap. When the border of the sample was reached, the sample chamber was shifted along the optical axis to start the acquisition of the next $N$ stacks, but displaced along the optical axis. Due to the air/water interface in the illumination beam path, the laser focus shifted along the optical axis when the sample chamber was moved in $x$ direction, and it had to be readjusted prior to image acquisition. So far, this readjustment was done manually. To obtain a mosaic of $N\times M$ stacks, the laser focus had to be adjusted $M$ times. The total acquisition time for a mosaic comprising $16\times 5$ stacks with 300 images per stack was 5 h. This mosaic covered a volume of $3.6\times 1.3\times 0.3{\mathrm{mm}}^3$.

High-resolution confocal microscopy was performed using a Zeiss LSM 880 equipped with Airyscan detector. Images were acquired using a $40\times$ WI objective lens with NA 1.2 according to standard protocols. For Airyscan imaging, a laser power of $0.65\mathrm{kW}/{\mathrm{cm}}^2$ was used routinely (0.3% of the available power). After imaging of a complete stack comprising 160 images, the topmost frame was bleached by 25%.

4.8.

Characterization of Optical Resolution

The lateral resolution theoretically achievable with an objective lens is given by the Rayleigh criterion, $d_R$, which quantifies the distance between the maximum and first minimum of the point spread function (PSF):

Here, $\lambda$ denotes the wavelength of the detected light and NA is the numerical aperture of the objective. Experimentally, this value is difficult to determine. Commonly the full width at half maximum (${\mathrm{FWHM}}_{xy}$,) is used instead, which is related to $d_R$ as

Similarly, the axial detection resolution $d_z$ is given as

The axial width of the LSFM PSF is given by the product of the excitation and detection probability distributions.

In order to measure the resolution realizable with our LSFEM setup, we used fluorescent beads with subresolution diameters and the PSF extraction feature of the deconvolution software Huygens (Scientific Volume Imaging, Hilversum, The Netherlands).

We prepared samples of green and red fluorescent beads with diameters of 100 and 176 nm in 1% agarose gels, respectively. We acquired $z$-stacks of these samples with both the Airyscan confocal microscope and the light-sheet microscope and determined the PSF using Huygens (Table 1). From the lateral and axial intensity profiles of the PSFs, we derived the respective FWHM values as follows:

Table 1

Optical resolutions of the utilized microscopes.

Note that the PSF was measured using a lateral object field pixel size of 173 nm. We are restricted to this value, because it is determined by the objective magnification ($25\times$), the supplemental magnification lens ($1.5\times$) and the camera pixel size ($6.5\mu \mathrm{m}$). Simulating the imaging of subresolution objects with this pixel size shows that this slight violation of the Nyquist theorem leads to an underestimation of the resolution by 15% to 20%.

For comparison, we also determined the PSF of the Airyscan LSM 880 using a $63\times$ oil immersion objective with NA 1.4. Here, we obtained FWHM values for the PSF of 135 and 515 nm laterally and axially, respectively.

4.9.

Image Processing

3-D stacks of raw 16-bit images were processed using custom-written MATLAB scripts, which allowed parallel data processing. In a first step, the intensity histograms were adjusted to homogenize brightness and contrast throughout the complete data set. Every 3-D stack was first scanned to find its minimum and maximum intensity values. With the respective values, a linear intensity adjustment was performed to cover the full dynamic range.

To achieve complete representations of the mouse DG, several 3-D data sets were stitched together. For this purpose, the stitching plugin of Fiji was used.²⁸ Stitching was performed in two steps to optimize the processing speed. First, substacks of the 3-D data sets were created using a MATLAB script. Each substack contained 15% of the full stack. In a second step, each substack was stitched to its respective neighboring substack yielding the best overlap in terms of the cross correlation measure. Based on this information, the full 3-D stacks were stitched.

A final step to improve the contrast throughout the 3-D data was performed after stitching. This was done to compensate possible intensity variations of the sample in axial direction. To this end, a histogram equalization was performed in every image plane of the stitched data set. For calculation of $z$-projections, the maximum intensity projection algorithm of Fiji was used.

Selected image stacks as outlined in the results section were spatially deconvolved using Huygens software (Professional version 17.04, Scientific Volume Imaging, The Netherlands) using theoretical PSFs and the classical maximum likelihood estimation algorithm. The deconvolution software provides as output the PSF, which was determined by analysis of fluorescent microbeads embedded in 1% agarose gel.

The 3-D representation of the data was achieved using the Surpass view in Imaris (Version 9.10 Bitplane Inc., Zurich, Switzerland). Data processing was performed on a workstation equipped with two Intel Xeon Platinum 8160 CPU (2.1 GHz, 24 cores), 512 GB memory, and an Nvidia Quadro P5000 GPU (16 GB GDDR5X) running under Windows 10 Pro.

4.10.

Determination of the Expansion Factor

The magnification factor of expansion microscopy was experimentally determined by comparing the short width of the cell soma of labeled granule cells in images obtained from samples before expansion (Fig. 1) with those after expansion (Fig. 2). From Fig. 1, we obtained a soma width of $10.1\pm 0.6\mu \mathrm{m}$ ($n=25$). For three different expansion experiments, we determined a soma width of $35.6\pm 1.7$, $43.3\pm 1.7$, and $39.8\pm 2.6\mu \mathrm{m}$. Thus, the average expansion factor was $3.9\pm 0.3$.