1. Introduction

Understanding the structure-function relationships at cellular, circuit, and organ-wide scale requires three-dimensional anatomical and phenotypical maps with high spatial resolution. To achieve whole organ three-dimensional imaging, a series of tissue clearing methods have been developed [1,2], which can improve the optical homogeneity of a large volume tissue. These methods can be divided into three types, including hydrophobic organic chemical cocktails, such as DISCO [3–6]; hydrophilic reagents, such as Sca/e [7,8], CUBIC [9–12]; hydrogel-embedding fixation and dilapidation, such as CLARITY [13,14] and PACT [15]. Although each type of clearing method has its own capabilities for fluorescence preservation, morphology retention and tissue transparency, it is difficult to make tissue absolutely transparent and free of light absorption with the exiting methods [1]. Therefore, it is hard to acquire high quality images in deeper part of the tissue with the exiting clearing methods, especially for the large-volume samples. Combined with light sheet fluorescence microscope (LSFM), previous studies have used tissue clearing methods to generate datasets of whole mice brains and other organs [4,9,11,16], as well as a postnatal day 3 marmoset brain [10]. Recently, several novel LSFM with high spatial resolution and throughput such as COLM [17], MOVIE [18] and OTLS [19] have been developed. However, high-resolution objectives are more sensitive to RI heterogeneity in the deep regions of cleared tissue. So, it’s hard for the high-resolution LSFM to image big-volume cleared tissue with unsatisfied transparency in deep regions of cleared tissue.

Combining tissue clearing methods with mechanical sectioning, which can reduce the light scattering interference effectively, will overcome these limitations for imaging the fine structures in deep tissue. However, after being treated with clearing agents, the rigidity of the tissue is greatly reduced and the tissue can become very fragile [7,8]. Even if these cleared tissues were embedded with agarose or gelatin, the sectioning property of the biological tissue inside did not change since the macromolecule of agarose cannot penetrate the samples. During sectioning, the interior of the cleared tissue will show distortion, which could bring more difficulties for subsequent data analysis including registration among continuous sections [20].

We developed a new method called M-CUBIC with a tissue clearing method modified from CUBIC-L [11] and permeable hydrogel embedding to keep tissue rigid. This method can clear tissue effectively and enable it to be sectioned properly during datasets acquisition. Combined M-CUBIC with HLTP [21], first we acquired the cholinergic neuron distribution information from the whole mouse brain at the resolution of 2.6 × 2.6 × 1.84 µm³ in about 3.5 hours. Then we acquired the whole-brain inputome of cholinergic neuron in basal forebrain with dual color imaging at the resolution of 1.3 × 1.3 × 0.92 µm³ in about 5.5 hours. We also acquired the microstructures of blood vessels and glomeruli in mouse kidney at the resolution of 0.32 × 0.32 × 2 µm³ with fMOST [22]. These results proved that our new method can be used for mesoscale imaging of the intact rodent organs with high throughput, which can facilitate the understanding of structure-function relationship in biological tissues.

2. Materials and methods

2.1. Animal

All the animal experiments followed procedures approved by the Institutional Animal Ethics Committee of Huazhong University of Science and Technology. 8-week-old C57BL/6J mice, Thy1-GFP M-line transgenic mice (Jackson Laboratory, Bar Harbor, ME, USA), Chat-Cre:LSL-H2B-GFP transgenic mice [23,24], Chat-ires-Cre:Ai14 transgenic mice [23] were used.

2.2. Viral labeling

The stereotaxic coordinates for the target areas were based on the Mouse Brain in Stereotaxic Coordinates Atlas [25]. Using a pressure injector (Nanoject II; Drummond Scientific Co., Broomall, PA, USA), 225 nl of RV-DG-EnvA-GFP and 75nl AAV-DIO-TVA-BFP and AAV-DIO-RG were injected into the SI of a 8-week-old Chat-ires-Cre: Ai14 mouse (AP -0.38 mm, ML -1.75 mm, DV -5.50 mm). After surgery, the animals were return to standard living conditions for 14 days, and then they were sacrificed for brain sample preparation.

2.3. Vascular staining of the mice kidney

8-week-old C57BL/6J mice was handled in a homemade device. 75 µL tomato lectins (DyLight 594 labeled Tomato Lectin, 1 µg/µL, 10 mM HEPES, 0.15 M NaCl; Vector Laboratories) was injected through the tail vein. After 20 min, animal was anesthetized and subsequently intracardially perfused for kidney samples preparation.

2.4. NAGA synthesis

NAGA was synthesized with the previously described method [26]. Glycinamide hydrochloride (6.30 g, 0.05706 mol), 6 mL cold deionized water, 33.6 mL of 2 mol/L cool potassium carbonate, and 18 mL cold diethyl ether were mixed in a 100 mL three-necked glass flask which was placed in an ice bath. Subsequently, a solution of 5.70 g of acryloyl chloride in 24 mL diethyl ether was added dropwise under stirring at 0°C for about 1 h. Then, the mixture was further stirred for 4 h at room temperature. After that, 6 mol/L HCl was added to the solution until the pH was 2. Next, the mixture was washed three times with 150 mL of diethyl ether to remove the organic phase, and the remaining diethyl ether was evaporated under a vacuum. Again, the pH of the solution was adjusted to neutral using 2 mol/L NaOH, and the mixture was freeze-dried. The raw product was washed three times with 150 mL of ethanol/methanol mixture (4/1, V/V). Then, the ethanol and methanol were removed by rotary evaporation and the remaining solution was recrystallized at 0°C to obtain the resultant NAGA, which was dried in vacuo.

2.5. Measurement of mechanical properties and machinability

The compressive stress test of the hydrogel samples was tested on WDW-05 electromechanical tester (Time Group Inc, China) at room temperature. In this study, all the samples were fully equilibrated in deionized water before test. The test method just followed the previous article [26].

We cut the PNAGA hydrogel sample with a vibratome (Leica VT1000 S, Germany). The sectioning amplitude and speed of the vibratome were 0.95 mm and 0.6 mm/s, respectively. We used a stylus profiler (Dektak XT, Bruker, Germany) to test the surface flatness of the sectioning face of PNAGA hydrogel.

2.6. Solution preparation

0.01 M phosphate-buffered saline (PBS) (P3813, Sigma-Aldrich) was dissolved in 1L dH2O to form 1x PBS.

40 g paraformaldehyde (PFA) (158127, Sigma-Aldrich) powder and 25 g sugar (V900116, Sigma Aldrich) were dissolved in 1L 1x PBS solution to form 4% PFA.

CUBIC-L: 10% N-Butyldiethanolamine (B0725, Tokyo Chemical Industry) and 10% Triton X-100 were mixed with dH2O.

M-CUBIC: 10% N-Butyldiethanolamine, 10% Triton X-100 and 10% N,N,N’,N’-Tetrakis(2-hydroxypropyl)ethylenediamine was mixed with dH2O.

Sca/eS: 50% urea, 22.5% sorbitol (85529, Sigma-Aldrich) and 5% Triton X-100 were mixed with dH2O.

The upper solution of CUBIC-L, M-CUBIC and Sca/eS were all stirred at 45°C until all the component dissolved.

PACT: 4% acrylamide (Sigma Aldrich Inc., St Louis, MO, USA) and 0.25% VA-044 initiator were mixed with 1x PBS to form A4P0 solution; 8% SDS were mixed with 1x PBS.

NAGA hydrogel solution: 3 g NAGA, 0.15 g MBA (Sigma Aldrich Inc., St Louis, MO, USA) and 0.025 g VA-044 initiator were dissolved in 7 g 1x PBS and the mixed solution was shaken continuously until all the powder dissolved to form hydrogel solution. The mixed solution was stored at 4°C in the dark.

2.7. Tissue preparation and clearing

The mouse was anaesthetized using a 1% solution of sodium pentobarbital and subsequently perfused with 1× PBS, followed by 4% PFA. The brains were removed and post-fixed in 4% PFA at 4°C for 24 h. Each intact brain was rinsed overnight at 4°C in 1× PBS solution after fixation. For PACT treatment, tissue was immersed in A4P0 solution for 24 h. The immersed tissue was polymerized at 37°C for 5 hours, then the tissue was removed from hydrogel solution. Then all the tissues were immersed in the clearing solution (M-CUBIC, CUBIC-L, Sca/eS and PACT (8%SDS)) at 37°C with gentle shaking (the whole mouse brain, 4 days). Clearing solutions were changed every 2 days. After clearing, tissues were washed three times (4 hour each) with 1× PBS until there is no clearing solution. Then all the tissue was immersed in hydrogel solution at 4°C with gentle shaking for 2 days. PNAGA hydrogel solution was changed every 12 hours. Then, permeated tissues were degassed with nitrogen for 2 minutes and then incubated at 40°C for 6 hours to initiate tissue-PNAGA polymerization. Certain concentration of the PI solution was dissolved in the delipidation reagent to perform the PI staining on biological tissue simultaneously with the tissue degreasing step.

2.8. Measurement of light transmittance and refractive index

The light transmittance of the cleared brain slices and embedded brain slices (Fig. 1(c)) was measured with a visible near-infrared optical fiber spectrometer (Lambda 35, PerkinElmer, USA). The brain slices were mounted on the glass slide. The test light was irradiated on the central part of cleared brain slice samples. We set the light transmittance of the clearing solution as the blank value, and the light transmittance of the samples were normalized to the blank value, and was defined as the relative transmittance. Each value was determined as an average of three samples.

 

Fig. 1. Development of modified CUBIC clearing method. (a) The pipeline of embedding of optical clearing tissue with PNAGA hydrogel and imaging with mechanical sectioning based whole organ imaging system. (b) Pictures of adult whole brains cleared with CUBIC-L, Sca/eS and PACT. (c) The transmittance curves of the brain tissue in different clearing methods and embedding states. (d) Quantification of the fluorescence preservation of EGFP and tdTomato after clearing with M-CUBIC, CUBIC-L, Sca/eS and PACT. (e) Fluorescence images of EGFP (Thy1-GFP mouse) and tdTomato (Chat-Cre: Ai14 mouse) in brain slices before and after M-CUBIC clearing. (f) The normalized fluorescence intensity varies with distance of purple and blue lines in (e). (g) The Z-axis fluorescence images of CX3CR1-GFP before and after M-CUBIC treatment, g1 and g2 is the XY direction images in 50 µm and 150 µm depth. (h) The normalized fluorescence intensity versus time curve in PBS environment and after M-CUBIC treatment.

Download Full Size | PDF

The refractive index of the 8 kinds of index-matching solution (Fig. S1b online) were measured with an abbe refractometer (WYA-2S)

2.9. Image preprocessing

(1) Confocal microscopy

The brain slices were mounted on glass coverslips and incubated in index matching solution (such as 50% glycerol). Then, different regions of the brain slices were imaged with an inverted confocal fluorescence microscope (LSM 710, Zeiss, Germany) equipped with a Fluar 10×/0.5 objective (dry; working distance, 2.0 mm). The brain slices were imaged at the same regions with the same imaging parameters at different processes for quantitative statistics.

(2) HLTP

We used HLTP to image the brains [21]. We choose the sampling interval along the x-axis to be the same as that along the y-axis, resulting in a voxel size of 0.65 × 0.65 × 0.46 µm³. For Chat-Cre: LSL-H2B-GFP mouse brains, we used the 4 × 4 binning mode of the sCMOS to obtain voxel sizes of 2.60 × 2.60 × 1.84 µm³ datasets. For Chat-ires-Cre: Ai14 mouse brain, we used the 2 × 2 binning mode of the sCMOS to obtain voxel sizes of 1.30 × 1.30 × 0.92 µm³ datasets.

(3) Visualization and reconstruction

We visualized the data set using Imaris software (Bitplane, Zurich, Switzerland 9.0) to generate the figures and movies. The data were sampled to 5.20 × 5.20 × 1.84 µm³ (Fig. 5) or 2.60 × 2.60 × 0.92 µm³ (Fig. 6) firstly, and then were converted into a format that Imaris software can process and import into Imaris software to generate 3D images or movies. For the data that requires quantitative statistics, we used Imaris to intercept the whole data to 3D data blocks and then used the spot module to render the data. We selected the appropriate rendering parameters to ensure that the cell nucleus is accurately identified and finally exported the statistical results.

To trace the single neuron morphology, we transformed data from TIFF to LDA type via Amira (FEI, Mérignac Cedex, France 6.1.1)and applied the filament editor module of NeuroGPS-Tree [27] software to brain-wide tracing of GFP-labeled neurons at by human–machine interaction. Briefly, we loaded the data block of interests into NeuroGPS-Tree software and assigned the initial and terminal points of the fibers in the block, so that NeuroGPS-Tree could automatically calculate the pathway between initial and terminal points. We repeated this procedure until the neuron morphology reconstruction was finished. We checked the reconstructed neurons back-to-back by three persons to prove the tracing results were right, and then saved the tracing results in SWC format. At last, we loaded the mouse brain outline and the SWC format into Amira simultaneously and used the moviemaker module of Amira to generate figures.

2.10. Quantifications

(1) Imaging depth quantification

We imaged the tissue from the surface to deep layer of the tissue until the detectable signal strength decreased significantly. We statistic the fluorescence intensity at different depth with ImageJ software and generate fluorescence intensity as a function of imaging depth, all the data points were mean ± SD (Fig. 1 h).

(2) Normalized mean fluorescence quantification

To assess the fluorescence preservation of the clearing methods, we quantitatively calculated the normalized mean fluorescence intensity of fluorescence images after different clearing methods. For each brain sample of both before and after clearing states, we used the same imaging parameters to image the same area, all the images were about 50 µm. Then we circled independent cells and calculated the fluorescence intensity after clearing as I and that before clearing as I₀. The fluorescence intensity preservation after clearing was normalized to the intensity before clearing as “I/I₀” (Fig. 1(d)).

(3) Statistical analysis

Data analyses and graph construction were performed using the GraphPad Prism software. All the data were presented as means ± SD. One-way ANOVA was used to compare more than two groups of data. (Figure 1(d), Fig. S1b online). In this study, P < 0.05 was considered significant (*P < 0.05, **P < 0.01, and ***P < 0.001).

3. Results

3.1. Selection of tissue clearing methods

To combine the tissue clearing method with mechanical sectioning based automatic imaging system, we need to establish a sample preparation method with high fluorescence preservation and suitable tissue clearing capacity, as well as great sectioning property. The whole process contains three major steps: tissue clearing, embedding and whole organ imaging (Fig. 1(a)). The whole organ imaging systems was previously developed [21,22] with a custom-made vibratome for serial sectioning. In order to select the proper tissue clearing method, we first excluded the organic solvent–based clearing methods since it may cause the tissue rigid and brittle, which makes the sectioning process difficult. Then, we tested several hydrophilic reagent-based clearing methods, including CUBIC-L, Sca/eS, PACT, and evaluated their fluorescence retention, signal-to-noise ratio and transparency depth. To evaluate the clarity of different methods, we recorded bright-field images of the mouse brains at different clearing states with CUBIC-L, Sca/eS and PACT (Fig. 1(b)). CUBIC-L and Sca/eS show better transparency characteristics after reagent-treated for 4 days (Fig. 1(b)). After embedded in PNAGA hydrogel, the clarity of CUBIC-L cleared sample was better than that of Sca/eS treated sample (Fig. 1(b)). It’s possible that the degree of delipidation of CUBIC-L is higher than Sca/es, which could make the CUBIC-L-treated sample more transparent in the “embedded” state. We also measured the transmittance spectra of the mouse brain slices cleared with the different protocols (Fig. 1(c)). We found that the clarity of tissue cleared via PACT decreased seriously after hydrogel embedding (Fig. 1(c)), which means PACT was incompatible with PNAGA hydrogel embedding. The subsequent introduction of salt ion solutions and PNAGA hydrogels may not match the optical properties of acrylamide used in the PACT, resulting in a decrease in its transparency. In contrast, tissue cleared via CUBIC-L retained relatively high transparency after PNAGA hydrogel embedding (Fig. 1(c)). Then we tested the fluorescence retention rate of EGFP and tdTomato with different clearing methods. The results showed that CUBIC-L could preserve green fluorescence better than PACT and Sca/eS, while the fluorescent retention of tdTomato in CUBIC-L was still dissatisfactory (Fig. 1(d)).

In order to increase the fluorescence retention of tdTomato, we added 10% N,N,N’,N’-Tetrakis(2-hydroxypropyl)ethylenediamine to CUBIC-L solution. We found the modified formula has better fluorescence preservation in red fluorescent protein than CUBIC-L (Fig. 1(d)). We named the modified formula with NAGA embedding as M-CUBIC. Specifically, we compared the fluorescent signals of Thy1-EGFP and Chat- Cre: Ai14 mouse brain slices before and after M-CUBIC treatment (Fig. 1(e)). Quantitative results showed that the fluorescence intensity and signal-to-background ratio of GFP and tdTomato remained unchanged after M-CUBIC treatment (Fig. 1(f)). Furthermore, we obtained the Z-axis images of EGFP-labeled mouse brain sections with PBS and M-CUBIC treatment (Fig. 1 g), the curve of normalized fluorescence intensity along with depth showed that M-CUBIC treatment could increase the depth of imaging significantly (Fig. 1 h).

3.2. Reduction of distortion of mechanical sectioning of transparent tissue

Selecting suitable embedding methods for mechanical sectioning and maintaining clarity of cleared tissue simultaneously is another key issue. Previously, agarose and gelatin were employed to embed cleared tissue and for mechanical sectioning to obtain whole brain data set [20,28,29]. However, gelatin and agarose can only support the outer contour of the tissue and cannot change the sectioning characteristics of the cleared tissue. After the clearing processes, tissue disintegrates partially and becomes soft, which can cause distortion during mechanical sectioning. In order to solve the problem of sample sectioning distortion, we embedded the pre-cleared tissue with PNAGA hydrogel [26,30].

As small molecule, NAGA monomer could penetrate into the tissue and crosslink with biological macromolecules such as proteins and nucleic acids in cleared biological tissues (Fig. 2(a)). NAGA hydrogel then formed a 3D network of NAGA molecules with biological tissues after polymerization to stabilize the biological structure, which could avoid being further damaged in the subsequent sectioning processing (Fig. 2(a)). By adjusting the ratio of the PNAGA hydrogel system, we could obtain 3D network polymers with different mechanical properties suitable for certain sectioning parameters.

 

Fig. 2. The establishment of M-CUBIC. (a) The mechanism of NAGA embedding of PFA-fixed and cleared biological tissue. Biomacromolecules such as protein can be bonded on hydrogel network. (b) The mechanical property of PNAGA with different concentrations of monomers. (c)The sectioning unevenness of PNAGA hydrogel with different crosslinking agent concentration. X is the direction of sectioning and Y is perpendicular to the direction of sectioning. The periodic pattern of X direction may be caused by the periodic lag in the gear rotation of the vibratome, which leads to periodic fluctuation in the advance direction of the sample. (d) The uniform PI-stained coronal plane images of the mouse brain embedded in PNAGA hydrogel. The inclined illumination-detection scanning strategy adopted here enables us to image 1 inclined plane each time. With the stage moving along the x-axis, we can scan a block of width 0.9 mm. The stage then moves 0.9 mm along the y-axis to image the next block. In every block, the illumination is not uniform in the edge and center, which introduce stripe artifact in the coronal section. (e) Comparison of the sectioning stability and smoothness between PNAGA embedding and agarose embedding. The yellow arrows indicate the bumpy block face after sectioning when the sample was embedded in agarose. (f) Acquisition of neuron distribution in intact Thy1-GFP mouse brain with PI staining for cytoarchitecture information. (g)The rigid registration result during the pre-section and the post-section of the experiment group (brain sample was embedded in PNAGA hydrogel).

Download Full Size | PDF

In order to obtain stable and good sectioning properties, we need to choose the suitable hydrogel ratio. Previous research pointed out that when the concentration of NAGA reached 30%, its equilibrium water content (EWC) only increased slightly, indicating that the hydrogel with 30% NAGA was more stable [26]. Therefore, we set the NAGA monomer concentration to 30% and tested the mechanical properties and sectioning flatness with different cross-linker MBA ratios (0, 0.5%, 1%, 1.5%, 2%). The results showed that when the crosslinking agent concentration is 0, the hydrogel showed large elasticity and low rigidity (Fig. 2(b)). The fluctuation degree of the sectioning block face was about 50µm (Fig. 2(c)), which was too rough for optical imaging. When we added cross-linker MBA, the hydrogel rigidity increased and the fluctuation degree of the sectioning block face decreased to 4-6 µm both along and across the sectioning direction in the case of 0.5-2% cross-linking agent added (Fig. 2(b), (c)), which met the optical imaging requirement, since we could start to image the sample below the sectioning surface for several micrometers. We chose 30% NAGA and 1.5% MBA as the application demonstration formula. To test our formula was suitable for biological tissue embedding and imaging, we performed a continuous 80µm imaging with HLTP from 20 µm above the sample surface to the deeper tissue, including 44 Z-stacks. Propidium iodide (PI) was used for real-time staining during imaging to reveal the unevenness of the sectioning block face. As shown in Fig. 2(d), the uniform PI staining at each stack and different areas indicated the excellent flatness of the sectioning block face. Thus, our results proved that PNAGA hydrogel embedding is suitable for precise mechanical sectioning.

Previous study also used gelatin to embed the optical cleared brains for whole brain imaging, which proved high degree of crosslinking between gelatin and tissue [20]. However, the sectioning deformation of gelatin embedded cleared tissue was severe and the fluctuation was about 50 µm, which required acquisition of repeated data for registration. We speculated that agarose and gelatin were macromolecules that couldn’t penetrate into the interiors of the tissues. They could only cross-link with the surface of the tissue, but not with the internal molecules of the tissue, which leaded to the poor mechanical sectioning property. We also compared the surface flatness of pure agarose and gelatin blocks and found that the surface flatness of gelatin was inferior to that of agarose (Fig. S1a, b). To compare the sectioning properties between PNAGA hydrogel embedded samples and agarose embedded samples, we cleared the brain samples with CUBIC-L and embedded the whole mouse brain in PNAGA and agarose respectively and performed the whole brain imaging with PI staining during the imaging process. As shown in Fig. 2(e), the agarose embedded brain showed uneven staining in the interior regions of the brain such as the regions around the lateral ventricle, indicating the sectioning property of these regions in agarose embedded brain was not good enough to provide flat block surface for optical imaging. On the other hand, the PI staining in the interior regions of PNAGA embedded brain was uniform, indicating the excellent sectioning property of PNAGA hydrogel embedding.

Another key issue was to choose the right refractive index (RI) matching solution for desired transparency effect. A great quantity of RI matching solution would be used in mechanical sectioning and imaging system. In the present study, we used a whole-brain continuous imaging system based on mechanical sectioning and a circulating water system was designed to pump away the biological tissue which was cut off to avoid the blockage of the objective. A solution with a lower viscosity can ensure that the tissue is easier to be pumped away by the circulating water system and will not block the objective lens and affect the imaging quality. Therefore, the ideal RI-matching solution is expected to have good fluorescence retaining ability and low viscosity. The original RI matching solution of CUBIC-L is expensive and difficult to prepare for large volume tissue imaging. Therefore, we decided to select more economical and safer RI matching solution for whole brain imaging. We selected eight kinds of refractive index matching solutions such as iohexol, dimethylformamide and glycerol for test. We mixed up these solutions with distilled water to make 50% concentration solution, and then we test the fluorescence retention of EGFP-labeled tissue after immersed in these solutions. Based on fluorescence retention test, we first excluded tween and antipyrine since their fluorescence preservation rate was only 20% while others reached 100% (Fig. S2a, b). We chose 50% glycerol as the RI-matching solution after testing RI and viscosity (Table S1 online). The price comparison of glycerol and RI matching solution of CUBIC-L was shown in Table S2. Finally, we established a complete transparent protocol with suitable delipidation reagent degreasing, hydrogel embedding and 50% glycerol refractive index matching.

In order to demonstrate that our M-CUBIC method was suitable for mechanical sectioning based automatic imaging system, we cleared a whole brain of Thy1-GFP mouse and embedded the cleared brain tissue in PNAGA hydrogel and agarose respectively. Then we imaged the whole brain with HLTP (Fig. 2(f)). The sectioning amplitude and speed of the vibratome were 1.0 mm and 0.5 mm/s. The total mechanical sectioning period for imaging the whole mouse brain was about 1 h. In the M-CUBIC treated brain tissue, we could overlap the fiber signal of pre-section and the post-section through rigid registration (Fig. 2 g). The quantified error in brain-wide was provided in Fig. S3. While in the control group (embedded in agarose), we found that there was no similar characteristic signal in pre-section and the post-section, which meant that the undulations of sample sectioning surface had exceeded the imaging depth (Fig. S4) and the fluorescence signals had become discontinuous across different sections. This discontinuity might cause massive information loss during data acquisition. In general, our results demonstrated that our M-CUBIC method can increase the rigidity of transparent tissue and reduce deformation during mechanical sectioning, which guarantees the signal continuity and stability during whole organ imaging.

3.3. M-CUBIC improves the data acquisition speed of HLTP

Due to the transparent treatment of the tissue, the imaging depth could be increased from 40 µm of the non-transparent tissues [21] to 80 µm or 120 µm in transparent tissues, which significantly reduced the moving time of the imaging platform and the total sectioning times during data acquisition process (Fig. 3(a), (b)). The data acquisition efficiency can be further improved by imaging two samples simultaneously. We calculated the imaging time of control group (40 µm) and experiment group (80 µm or 120 µm)to demonstrate that the experiment group could greatly save imaging time, especially when we set the image depth at 120µm and image two samples at the same time (Fig. 3(c)). HLTP in our work offered a relatively high spatial resolution and high data acquisition throughput compared to some other custom-built systems [17–19,31–35] and commercial products [36] for whole organ imaging (Fig. 3(d)). Combined with our M-CUBIC method, we can further increase the data acquisition efficiency.

 

Fig. 3. M-CUBIC improves the data acquisition speed of HLTP. (a and b) The carton diagrams show the whole organ imaging process of mechanical sectioning based automatic imaging system with or without tissue clearing. For the control group, every section is 40 µm. For the experiment group, we can set every section as 80 µm (a) or 120 µm (b) without loss of information. Tissue clearing can increase the imaging depths and reduce the total mechanical sectioning times during the whole data acquisition time. (c) The table of different parameters and datasets acquisition time show that tissue clearing can reduce the data acquisition time of HLTP. (d) The data acquisition throughput and spatial resolution of different whole organ imaging system based on LSFM.

Download Full Size | PDF

3.4. M-CUBIC allows acquisition of three-dimensional microstructures of intact mouse kidney with subcellular resolution

To further prove the excellent mechanical property of PNAGA embedding for whole organ imaging, we embedded an intact mouse kidney with PNAGA hydrogel to acquire the microstructures. The kidney contains different kind of tissues such as connective tissues, fatty tissues, muscle tissues and blood vessels. The mechanical property of kidney is very heterogeneous, which makes it’s difficult to embed the kidney with agarose for whole organ imaging. The NAGA monomer can penetrate the kidney and form three-dimensional network to maintain the stability of different tissues, which makes the mechanical property of kidney homogeneous and suitable for vibratome sectioning. Lectin was used to label the blood vessels and glomeruli in kidney. We imaged the intact mouse kidney with subcellular resolution (Fig. 4). The sectioning parameters for kidney were similar to those for mouse brain samples. The glomeruli were mainly distributed in the renal cortex (Fig. 4(a)), and there were blood vessel branches and different scale blood vessels in the kidney (Fig. 4(b), (c)). Due to the high resolution of our datasets, we also identified the afferent and efferent glomerular arteriole, as well as the path of the afferent glomerular arteriole (Fig. 4(d)). We reconstructed the morphology of 549 glomeruli with Imaris (Fig. 4(e), (f)). The glomeruli displayed a wide range of differences regarding of diameters, areas and volumes (Fig. 4 g). These results demonstrated that PNAGA hydrogel embedding could be applied to acquire the microstructures of different organs in three dimensions with subcellular resolution for biological research.

 

Fig. 4. Acquisition of the microstructures of blood vessels and glomeruli in mouse kidney with PNAGA hydrogel embedding. (a) The overall vascular network labeled with lectin594 and continuous 200 µm maximum projection results. (b) The enlarge image of white box from (a), we can observe the blood vessel branches identified by the purple arrows. (c) The partial enlarged detail of (b) show the large-sized and small-sized blood vessels identified by the white arrows. (d) The partial enlarged detail of (b) show the afferent glomerular arteriole and efferent glomerular arteriole (blue arrow). The local enlarged view of the circle showed the path of the afferent glomerular arteriole (yellow arrow). (e) The reconstruction of the vascular of a 3.5 × 4.4 × 0.8 mm³ cube from the white box of (a). (f) The render of the cube with Imaris software, the color represents the volume size. (g) The statistics of glomerulus number, diameter, volume, and surface area parameters of this cube, there are 549 glomeruli, and the average diameter is 71.6 µm, the average volume is 3.3 × 105 µm³, the average superficial area is 2.7 × 104 µm².

Download Full Size | PDF

3.5. M-CUBIC allows high throughput acquisition for distribution of specific types neurons in the whole brain

To prove the high efficiency of three-dimensional imaging with M-CUBIC treatment and HLTP imaging, we cleared and imaged Chat-Cre: LSL-H2B-GFP mouse brains with M-CUBIC treatment. PI staining was applied to provide the cytoarchitecture information. We achieved dual channel imaging of cholinergic neuron distribution and cytoarchitecture at the resolution of 2.6 × 2.6 × 1.84 µm³. To improve the throughput of our imaging process, we imaged two samples at the same time (Fig. 5(a)). The total data acquisition time was about 7 hours for two mouse brains. We reconstructed cholinergic neuron nuclei (green) and cytoarchitecture (red) in the same brain (Fig. 5(b)). Based on cytoarchitecture via PI staining, we could distinguish the landmark of the brain regions and segmented the fluorescent signals according to different brain regions (Fig. 5(c)). Consistent with previous reports [37], cholinergic neurons mainly distributed in several nuclei from the frontal brain to brainstem, such as the diagonal band (NDB), medial habenula (MH), pedunculopontine nucleus (PPN), laterodorsal tegmental nucleus (LDT), the pontine gray (PG), the parabigeminal nucleus (PBG), the oculomotor nucleus (III), trochlear nucleus (IV), motor nucleus of the trigeminal (V), facial motor nucleus (VII), nucleus ambiguus (Amb), dorsal motor nucleus of vagus (X), and hypoglossal nucleus (XII). Small percentage of cholinergic neuron were found in the cortex. We showed the three-dimensional distribution of cholinergic neuron in the brainstem (Fig. 5(d)) and automatically counted the number of cholinergic neurons in each brain area with Imaris software (Fig. 5(d), right). We counted cholinergic neurons in some brain regions, including MH, PG, PBG, Amb, III/IV, V and VII (Fig. 5(e), (f)), and found that MH and VII had the largest number of cholinergic neurons, which was consistent with the previous study [37].

 

Fig. 5. Quantitation of the distribution of cholinergic neurons in the whole brain of the Chat-Cre: LSL-H2B-GFP mice. (a) Two samples were embedded together and imaged by one imaging system. (b) Horizontal view of genetically labeled cholinergic neuron nuclei (green) and PI-stained cytoarchitecture (red) in the whole brain. A, anterior; D, dorsal; L, lateral; M, medial; P, posterior; V, ventral. The different fluorescence signals in left and right hemispheres is due to the unstable fluorescence expression in the transgenic mice. (c) Merged image of coronal sections (GFP and PI signals) shows cholinergic neuron nuclei in the facial motor nucleus (VII). (d) Overview of the 3D distribution of cholinergic neuron nucleus in VII (Left). The neuron nucleus was distinguished using Imaris software and reconstructed in 3D space (Right). (e) Visualization of the anatomical localization and cholinergic neuronal distribution in 3D-reconstructed subdivisions in brainstem. (f) Numbers of cholinergic neuron nuclei in brain regions of hemisphere (from three brains, means ± SD).

Download Full Size | PDF

3.6. M-CUBIC allows analysis of neuronal projection and single neuron morphology

To obtain both brain-wide cholinergic neurons distribution and the inputs circuit to cholinergic neurons in substantia innominate (SI) at the same time, we injected AAV helpers (AAV-DIO-TVA-BFP and AAV-DIO-RG) and subsequently RV-DG-EnvA-GFP into the SI of Chat-ires-Cre: Ai14 mice. After M-CUBIC treatment and HLTP imaging, we acquired continuous whole brain 3D datasets with a resolution of 1.3 × 1.3 × 0.92 µm³. We merged the dual channel information and reconstructed it in three dimensions. The results showed that the whole set of data was continuous and complete (Fig. 6(a)). We found that the distribution pattern of cholinergic neurons labeled by tdTomato was similar to GFP-labeled neurons in Chat-Cre: LSL-H2B-GFP mouse brains (Fig. 6(a)). We also discovered multiple inputs regions to cholinergic neurons in SI by detecting EGFP-labeled neurons, statistics showed that SI cholinergic neurons widely received input from the cortex, hippocampal formation (HPF), pallidum (PAL), striatum (STR), hypothalamus (HY), pallidum (PAL), diencephalon and midbrain (MB), as well as a small amount of single synaptic input from the pons (P), medulla (MY) and cerebellum (CB) (Fig. 6(a), (b)). Specifically, there were intensive distribution of input neuron in medial prefrontal cortex (mPFC), HPF, pedunculopontine tegmental nucleus (PPTg) and caudal linear nucleus of the raphe (CLi) (Fig. 6(c)). Taking advantage of our dual-color imaging, we identified that cholinergic neurons in diagonal band (VDB) and PPTg could send long-range inputs to cholinergic neurons in SI (Fig. 6(d)).

 

Fig. 6. Acquisition of whole brain distribution of cholinergic neurons and long-range input neurons to cholinergic neurons in substantia innominata simultaneously with M-CUBIC. (a) The 3D reconstruction results of dual channel information in the whole mouse brain, RFP signals show all cholinergic neurons in the whole brain, and GFP signals show the neurons that project to cholinergic neurons in SI. The merged images show the cholinergic neurons and input neurons distribution in different brain areas. (b) Quantification of input neurons that project to cholinergic neurons in SI from different grouped brain areas. (c) The input neuron distribution in mPFC, HPF, PPTg and CLi. (d) Cholinergic neurons in VDB and PPTg send long-range inputs to cholinergic neurons in SI. (e) Reconstruction of the morphology of input neurons in Ect and LHA. The detailed images show the axon pathways of reconstructed neurons.

Download Full Size | PDF

Meanwhile, we also reconstructed the morphology of input neurons in contralateral ectorhinal area (Ect) and lateral hypothalamic area (LHA) (Fig. 6(e)). We showed the tracing results in the 3D brain contour. Neurons in the Ect that project to SI could send projection to the contralateral Ect through anterior commissure, posterior (acp). In the LHA brain region, we found that the axons of the input neurons in LHA to SI projected through the ipsilateral anteromedial thalamic nucleus (AM) and mediodorsal nucleus of thalamus (MD). Then the axon passed through habenular commissure (hbc) to the contralateral side and eventually projected to cholinergic neurons in SI through the AM brain region (Fig. 6(e)).

4. Discussion and conclusion

In the present study, we developed an optical clearing method named M-CUBIC that gave tissue suitable transparency to increase imaging depth and excellent mechanical properties to perform mechanical sectioning with less distortion and better stability.

Although great progress has been made in optical clearing field and different types of optical clearing methods have been developed [1,2], it is still difficult to make biological tissue absolutely transparent. The light scattering and absorption are still the main obstacle when imaging the deep layers of large volume tissues. Combining tissue clearing methods with mechanical sectioning can decrease light scattering in the deep layers of tissue to increase imaging quality, especially for large volume tissues. Previous study had tried this combination to image the whole mouse brain [20]. However, the gelatin embedding method used in previous study couldn’t improve the mechanical property of brain sample, which might cause distortion and deformation during sectioning. Our M-CUBIC method used small molecule NAGA to embed the biological tissues, which could penetrate the tissue and form three-dimensional network to stabilize the proteins and nucleus acids in biological tissues. The M-CUBIC method can provide excellent sectioning property, which can facilitate the sectioning and imaging process (Fig. 2).

On the other hand, the spatial resolution of light sheet microscopy for large volume tissue is generally not high enough to map the detailed structural information [2]. For example, to our best knowledge, optical tissue clearing combined with light sheet microscopy hasn’t been able to achieve the reconstruction of fine morphology of single neurons [1].Only a few previous study achieved reconstruction of axon pathway with tissue clearing and LSFM [38].The mechanical sectioning based whole organ imaging system could achieve unprecedented volume of biological tissue with an optical diffraction limited resolution in three dimensions [39,40]. However, the imaging speed of mechanical sectioning based whole organ imaging system was generally lower than that of light sheet microscopy. By combining optical clearing with mechanical sectioning based whole organ imaging system, our M-CUBIC method can achieve subcellular resolution imaging (Fig. 4 and Fig. 6) and significantly improve the imaging speed of mechanical sectioning based whole organ imaging system (Fig. 3). Specifically, PNAGA-hydrogel empolyed in M-CUBIC method is hydrophilic and compatible with multicolor fluorescent proteins, which make it possible to obtain different information of various structural components at the same time (Fig. 6). Thus, we can make direct comparisons between different biological information and investigate their interactions from the individual biological samples.

The monomer of NAGA can penetrate the whole organ and form three-dimensional network to provide homogeneous and excellent sectioning property. Therefore, PNAGA hydrogel embedding can be applied on other type of organ such as kidney (Fig. 4). Previous studies have applied optical clearing on various types of organs [6,15]. However, the light sheet microscopy couldn’t capture fine enough details such as the individual capillary in different glomerulus due to low spatial resolution, which may limit the application of these methods on biological and clinical research. The fine detailed microstructures of various types of organs can be acquired by mechanical sectioning based imaging system. However, few studies have reported such results probably due to lack of proper embedding methods for these organs. The agarose embedding can only support the exterior of the organ, the sectioning property of interior of the organ cannot be improved. The heterogeneity of sectioning property of such organs makes it difficult to be sectioned by vibratome. Technically, the resin and paraffin embedding can be applied on various types of organs [41–43]. However, the dehydration process during resin and paraffin embedding may cause severe deformation when performing the whole organ embedding. The PNAGA embedding we used skipped the dehydration process, which was much milder to biological tissues. Combined with mechanical sectioning based whole organ imaging system, we can acquire the detailed microstructures of different types of organs with subcellular resolution in both physiological and pathological conditions.

It is noteworthy that in the present study we used a stylus profiler to measure the surface flatness of the PNAGA hydrogel. However, the results of measurement didn’t represent the absolute flatness of the sectioning surface since it was possible that the hydrogel may deform a little bit when the stylus contacted and moved along the soft surface, which would make the surface appear more even than it is. Nevertheless, the stylus profiler result can reflect the surface flatness partly and help us select the optimal formula of hydrogels with better mechanical performance. In combination of our imaging results, we demonstrated the feasibility of M-CUBIC.

In general, we developed a new tissue process method called M-CUBIC, which combined tissue optical clearing with mechanical sectioning. This method is compatible with multiple fluorescent proteins and can reduce the data acquisition time of whole organ imaging system to acquire the microstructures of biological organs with subcellular resolution. This excellent sample processing method can be also applied on different types of biological organs, which can facilitate the biological and clinical research.