1. Introduction

Visualization and quantification of fluorescent markers harbor challenges for research in translational medicine. For example, in nanomedicine, fluorescent tags enable the investigation of drug delivery systems such as quantum dots [1], nanoparticles [2], micelles [3], or liposomes [4] concerning their biodistribution in cells and tissues. In addition, tracers are often applied in (bio)medicine to identify tumor metastases, inflammation, cell death, or organ functions in various tissues [2,5–8].

Light and fluorescence microscopy is well established and frequently performed on excised tissue sections. This method provides detailed tissue information at the cellular and subcellular levels. However, microscopy requires intensive pre-processing steps, such as fixation and permeabilizing a thin tissue section, followed by antibody staining to label the tissue slices fluorescently [9]. Most fluorescence microscopy imaging techniques are optimized for scanning a region of interest of tissue slices to obtain high-resolution images. Such images may provide detailed information down to the intracellular level. However, the resolution is traded in for the field of view, and the overall volume of the specimen processed [10,11]. In drug delivery systems research, a highly non-uniform distribution of drug carriers in specific organs is often observed [12,13]. For instance, Dreher et al. demonstrated that a change in the molecular weight of the nanocarriers has a significant impact on their distribution profile in tumors [14]. Lower molecular weight nanoparticles penetrated deeply into the tumor (more than 35 µm) and accumulated homogeneously, while larger molecular weight particles were highly concentrated near the vascular surface. Therefore, scanning only a comparably small region of the tumor slices at high magnification by fluorescent microscopy makes assessing the overall distributions of the carrier inside the whole tumor very difficult. Enlarging the scanned area by lowering the magnification may help but results in a loss of image resolution. Even if the loss in resolution may be acceptable, the scanned area is limited by the size of tumor slices and the technical limitations of the microscope.

In recent years, Multi-spectral Optoacoustic Tomography (MSOT) was introduced to provide tomographic 3D imaging and cross-sectional images in deep tissue in a noninvasive imaging manner. Animals were illuminated in a temperature-controlled water chamber with a 360° laser ring illumination. MSOT offers a 270° acoustic signal detection for a basis of the complete cross-sectional views of the animal, resulting in a challenge on their resolution in the areas that are not directly covered by acoustic detectors [15]. Furthermore, given the working principle of acoustics imaging excitation in the (near-) infrared (NIR) region and detecting acoustic frequency alterations, selecting appropriate markers or contrast agents is also challenging. Tracers for MSOT applications require optoacoustic properties, such that light is efficiently converted and released as vibrational and thermal energy. Therefore fluorophores with a relatively low quantum yield are utilized since they still can be detected by fluorescence imaging to validate results and, at the same time, give a specific optoacoustic signal. To allow for optoacoustic deep tissue imaging, the excitation/emission wavelength of those fluorophores commonly ranges from 680 to 980 nm. Thus, in MSOT, NIR probes are often utilized, limiting signal validation feasibility with fluorescent microscopy.

In vivo Imaging Systems (IVIS), another technique equipped with 3D tomographic reconstruction for both fluorescence and bioluminescence is used intensively to characterize drug pharmacokinetics and pharmacodynamics for cancer research [16,17]. IVIS illuminates fluorescent markers or bioluminescence in either epi-illumination (reflection) or trans-illumination (transmission). However, this imaging method results in limited information about the markers’ precise tissue distribution since the technique do not provide cross-sectional images. In addition, the sensitivity is usually of a limited depth range [8]. Thus, lower fluorescence signals are generally not detectable. Although some IVIS can provide a broader range of fluorescence probes and include bioluminescence detection, both IVIS and MSOT still suffer from low image resolution, light attenuation in deep tissues, and difficulties quantifying the fluorescent intensity. Sharpe et al. has published a new technology, optical projection tomography, which allows higher resolution bright field, darkfield, and imaging fluorescence imaging compared to IVIS, MRI, and MSOT. The technique provided virtually sectioned imaging in any orientation over 400 angles. However, it still suffers from signal scattering in the deep tissue and is limited to a sample size up to 15 mm [18,19]. To minimize the light attenuation and photon scattering within deep tissue, Pan et al. and Cai et al. have developed an impressive and robust sample preservation technique [20,21]. The advanced methodology on sample pre-processing of both uDISCO and vDISCO could overcome the general limitations of light-sheet microscopy by shrinking the sample and increasing the sample's transparency. However, high pressure transcardially perfusion and whole body immunolabeling are inevitable when employing uDISCO/vDISCO and light-sheet microscopy. In contrast, by slicing the tissues and performing block-face imaging, the proposed mesoscopy is an alternative to overcome the deep tissue problem by utilizing fluorescence markers without additional and extensive sample steps. Further, the setup combines fluorescence imaging with detailed RGB anatomical images of a large field of view of native samples, while uDISCO/vDISCO focuses on high-resolution fluorescence imaging exclusively.

In contrast, by slicing the tissues and performing block-face imaging, the proposed mesoscopy is an alternative to overcome the deep tissue problem by utilizing conventional fluorescence markers without additional and extensive sample steps. Further, the setup combines fluorescence imaging with detailed RGB anatomical images of a large field of view of native samples, while uDISCO/vDISCO solely focuses on high-resolution fluorescence imaging. The imaging system proposed here restores the original status of fluorescent marker staining in mouse's body without any intervention of sample processing steps.

Previous works have presented wide field and fluorescence cryo-imaging systems mounted on an advanced whole-body sectioning cryo-microtome, which allowed slicing on large specimens with the size of 250 × 110 × 5 mm [22] and 100 mm × 100 mm × 50 mm [23]. However, both systems were equipped with technology typical for microscopy. In Roy et al., a 1.4 MPx charge-coupled device (CCD) camera was used with precise motorization to compose a 5300 × 2100 pixel image with 15.6 µm pixel size from 5 × 4 individual image tiles [22]. At the same time, Wirth et al. utilized a fast 4.2 MPx scientific complementary metal-oxide-semiconductor (sCMOS) camera to produce 93 × 93 mm images with 45 µm pixel size and 71 µm resolution only. Additional hyperspectral imaging and spectral unmixing remarkably improved the contrast of fluorescent markers over autofluorescence from the tissue. However, the quantification of such signals remains challenging [23]. Here, we use a 16 MPx CCD camera from the astronomy domain and a 1:1 macro lens from the industrial vision domain to cover a field of view up to 21.6 mm × 27 mm with as small as 6 µm pixel size in a single shot. Especially on a conventional cryo-microtome, the avoidance of tiled imaging allows for seamless images even in the presence of cooling unit vibrations. Since no modification of the cryo-microtome is required, the imaging setup is mobile and can be easily installed and removed depending on the user's need. The imaging system enables multi-sectional imaging of both fluorescence probe distribution and tissue anatomy, with high flexibility on the filter configuration and fluorophores. Our technique provides automated high-resolution (8.5 µm lateral for diagonal line-pairs) cross-sectional imaging of entire mice with different organs visible in the same planar cross-section. The cross-sections achieve a high level of anatomical details and fluorescence image data, further improving the quantification of fluorescence signals. The automated mesoscope may be used as an add-on to commercially available cryo-microtomes capturing the images directly during tissue sectioning, avoiding extensive hands-on time for processing, tissue slicing, and mounting for fluorescent microscopy imaging. Furthermore, with the possibility of exchanging the filter set on the mesoscope, this automated mesoscope can examine a wide range of fluorescent probes and provides high-resolution anatomical cross-sectional images of large volumes. In this article, we demonstrate the basic setup of this automated mesoscope as a cryo-microtome add-on and evaluate its performance on several fluorescent probes, such as Cyanine 3 (Cy3, λ_Em= 568 nm), Allophycocyanin (APC, λ_Em= 660 nm), Cyanine 5 (Cy5, λ_Em= 666 nm), DY-635 (λ_Em= 671 nm), and indocyanine green (λ_Em= 814 nm) with λ_Em corresponding to the vacuum wavelength of emission.

2. Methods

2.1 Opto-mechanical setup and control

The block-face episcopic imaging setup was demonstrated on a Leica CM3050s cryostat (Leica Biosystems, Wetzlar, Germany). The optical add-on consists of a monochrome CCD camera with 4500 × 3600 pixels, 6 µm pixel size, and fair NIR quantum efficiency of 30% at 800 nm (MicroLine ML16200, FLI Instruments, USA) (Fig. 1). This camera is equipped with 1:1 imaging optics with an 89 mm focal length using f/4.8, corresponding to an NA of ∼0.10 (XENON-ZIRCONIA 2.8/89, Schneider, Bad Kreuznach, Germany). According to the product specification, this lens shows optimum performance at f/4.8, close to the diffraction-limited performance. It features a modulation contrast of 45% at 72 lp/mm (line pairs per mm) across the entire field of view, maintaining uniform brightness. The tube consists of customized threaded adapters and stackable/adjustable SM2 tubes (Thorlabs, USA). The tube length is carefully adjusted to achieve 1:1 projection according to the given flange focal distance of the camera and the specifications of the lens. For automated focusing in the z-direction, the camera and lens are mounted on a motorized stage (C-663.12 Mercury Step Schrittmotor-Controller and VT-80 Lineartisch, Physik Instrumente, Germany). This stage is mounted on a self-made base, featuring pitch and yaw adjustment and easy attachment to the cryo-microtome frame. RGB color images are generated by illuminating the sample sequentially with a ring of red, green, blue, and white (RGBW) light-emitting diodes (LEDs) (24× SK6812RGBW-WS, Opsco, China). Fluorescence images are produced depending on the dye using a LED with a controller (T-cube LED Driver, Thorlabs, USA), a 1” excitation filter, and a 2” emission filter in a motorized filter flip mount (MFF102/M, Thorlabs, USA) (Table 1). The emission filter is toggled into the beam path for each slice to acquire the fluorescence image. The fluorescence excitation LED was adjusted to avoid the shadow of the RGBW ring. The illumination geometry is elementary and can omit the beam splitter that could create astigmatism, thus producing a homogeneous and gloss-free illumination of the surface. The cover glass of the cryostat was removed during the imaging system. The cryostat was covered with a thick black cotton cloth to shield ambient light, humidity, and heat during imaging.

 

Fig. 1. Schematic of the block-face episcopic imaging instrument inside the schematically depicted cryo-microtome (a) Side view of the instrument, showing a 16 Megapixel monochrome CCD camera (cyan), 1:1 macro lens with 89 mm focal length using f/4.8 ≈ 0.10 NA (blue), a switchable emission filter (light red) in front of the lens, an RGB LED ring for color images (green), a focus motor stage (yellow z-nob), a self-made base with pitch (orange) and yaw adjustment (golden). (b) Top view of the instrument (perpendicular to the optical axis), additionally showing a filtered high-power LED for fluorescence excitation (red) with an excitation filter (orange). The distance between the fluorescence excitation LED and the sample was adjusted to roughly 12 cm. The angle between lens and LED was adjusted to about 25° to avoid a shadow from the RGBW LED ring. The distance between the RGBW LED ring and sample is 3 cm. Customized parts are drawn in light grey, and purchased parts are displayed in dark grey. The cryo-microtome (a chamber in light blue) features a blade (magenta) and the motorized sample holder with the sample (brown). Arrows indicate possible directions of movement (see Dataset 1) [24].

Download Full Size | PDF

Table 1. Filter sets used in this study to detect different fluorescent probes. The excitation light source and filter sets may be adapted to the individual needs of the researcher.

View Table

The hardware control is realized by a laptop connected via a USB port to the camera, the motor controller, and an Arduino Mega 2560 Rev3 microcontroller (Fig. 2(a)). The Arduino sends transistor-transistor logic (TTL) signals to the T-cube LED driver as well as controls the RGBW LED ring and the emission filter flip mount. A simple C-code running on the Arduino controls the light, emission filter flip, and emulates the foot-switch for automated slicing by the cryo-microtome. In addition, a software tool has been implemented in C# to enable automated multicolor imaging and cut for the acquisition of serial block-face images. A usual acquisition cycle consists of six images: white, red, green, blue light (without the fluorescence emission filter), dark, and a fluorescence image with the emission filter toggled into the beam path in front of the objective (Fig. 1(a), b) (also see supplementary Dataset 1) [24]. With the additional 8 s sensor readout, one cycle (slice) with 6 images (red, green, blue, white, dark and fluorescence) took about 1 min, including the cutting, automated focusing, and insertion/removal of emission filter. After this, a cut is performed by emulating the manual switch in continuous cutting mode until the external limit switch indicates a complete hand wheel rotation. Afterwards, a usual acquisition cycle can be repeated with the number of cycles inserted; the operator can stop and intervene in a cutting error.

 

Fig. 2. Control diagrams for the block-face episcopic imaging instrument. (a) Electronic block diagram for hardware control. (b) Diagram of the imaging process. For episcopic color imaging, a short exposure time of about 100 ms per channel is sufficient (less than 100 ms is not possible with the central mechanical shutter of the CCD camera). For fluorescence imaging, 1-5 s of exposure is typically required. ML 16200 CCD camera requires an additional 8 s for sensor readout in low noise mode in this specific setup. The imaging system equipped with the proposed components requires roughly 1 minute for generating a complete image, including the slicing, adjustments, imaging, reading, and saving of 6 channels.

Download Full Size | PDF

2.2 Preparation of cryo-sectioning sample

The animals were maintained at the Jena University Hospital's animal facility at room temperature at 23°C and environment humidity of 30 to 60% with systemic artificial day and night light cycles control (12 h light-dark cycles). All experimental procedures on animals had been approved by the ethical committee and local government authority of Thuringia, Thüringer Landesverwaltungsamt, Germany, and proceeded strictly following the approved guidelines. In the experiment of indocyanine green (ICG) retention profile in mouse's liver, 8 to 14 weeks old FVB/N mice (male and female) were injected intraperitoneally with human stool suspension for induction of peritoneal contamination and sepsis (PCI group) or saline (SHAM group) as described previously [25]. PCI and SHAM animals were injected intravenously with approximately 30 nmol ICG and sacrificed after 45 min under anesthesia by cervical dislocation. While evaluating Cy3 and Cy5 labelled siRNA polymeric micelles, 8 to 14 weeks old FVB/N mice (male and female) were injected with 30 µg siRNA contained nanoparticles. The animals used to evaluate the fluorescent signal of DY-635 labeled liposomes were prepared according to the previous works [4]. Approximately 10 µg APC-labelled anti-F4/80 (Clone: BM8) (Biolegend, Koblenz, Germany) was administered to the mouse to examine fluorescence in different emission wavelengths. All the injected animals were euthanized by Ketamine (CP-Pharma GmbH Germany, 500 mg per kg body weight (BW)) and Xylazine (Bayer Vital GmbH Germany, 80 mg per kg BW) overdose. After confirmed death (no heartbeat, no breathing, no reflexes), animals are frozen in a conical tube (50 mL) on dry ice. The animals were sectioned into three parts, thorax, upper abdomen, and lower abdomen, with a maximum thickness of approximately 2 cm using a surgical saw. The mouse sections were then mounted on a Leica CM3050S cryo-microtome using TissueTek (OCT Compound, USA) for the cross-sectional imaging. Each sample group (Sham, PCI, Cy5 nanoparticles) consists of three mice. Average signal intensity from 10 images per organ was taken for fluorescent signal quantification.

2.3 Data processing

The downstream image 2D processing and data quantification were performed using an in-house developed graphical software tool written in C#. The 3D volume imaging and displacement correction were performed using python scripts specially developed for this research, as shown in Code 1 [26], Code 2 [27] and Code 3 [28].

2.3.1 Level and gamma correction for image reproduction

We store the six raw monochrome camera images (white, red, green, blue, dark, fluorescence) in a linear 16-bit multi-page tiff-file for each slice. These linear images are converted to the reproduction medium (screen, paper) using black and white level and gamma correction. Our software tools and scripts for level and gamma-correction are based on Eq. (1).

Equation (1) describes the application of two-level corrections in succession. Due to linearity, the parameters for black levels b_i and l_b and white levels w_i and l_w are redundant in principle and could be summarized into a single level correction, as it is done in the python code for performance reasons (Code 1 [26], Code 2 [27], and Code 3 (Ref. [28]). However, the two-level corrections have different meanings, resulting in other criteria for determining their parameters. The first level correction normalizes the raw data so that the black level of each channel is 0.0 (using b_i) and a gray object appears gray in the data (same numbers for the red, green, and blue channels using w_i-b_i). The last remaining degree of freedom can be used to set the image data's absolute brightness (highest numerical values). The parameters with index i are specific for each raw data's camera and imaging conditions channel independently (R, G, B, W, F). For linear level comparison, no more adjustment is necessary. When reproducing the images on the screen or print media, however, the general white level has to be adjusted not to exceed 255 (for 8 bit), and the gamma needs to compensate for the 2.20 monitor gamma. This correction is managed by the second level correction using the independent channel parameters for black (l_b) and white level (l_w) and gamma (γ). It gives the opportunity to optimize visual contrast while remaining the true color. For the FLI ML16200 CCD camera that we used in this setup, the parameters b_i are set to 1000 according to the average black level of the camera.

The fluorescence channel C_F long exposure times (1 s and more prolonged) may lead to an additional background due to ambient light and the camera's dark current. Therefore, we subtract the linear dark image C_D with the same exposure time directly from the fluorescence channel C_F instead of a constant black level b_F as expressed in Eq. (3).

For linear level comparison l_b = 0, l_w = 1 and γ = 1. For a reproduction of the images to the screen or print media, γ = 0.45 and l_w is adjusted to avoid clipping.

2.3.2 Concentration measurement

For tracking the fluorescent maker distribution in experiments, it is sometimes necessary to quantify the concentrations of fluorescent signals in a single organ or some specific structure. The fluorescent signal was assumed to be linear concerning brightness using a measured background estimate. For the applied front face technique, the brightness also depends on the depth of light penetration into the tissue, limiting the analysis to relative intensity distributions within one experiment and making assumptions on absolute concentrations difficult. Nevertheless, relative quantification remains possible for comparing different individuals of the same tissue type.

To further discuss the fluorescence intensity (we use gamma = 1), we considered regions of stained organs and areas that should not be stained. The comparison of both values helps to evaluate autofluorescence. In usual experiments, the inhomogeneity of the organ stain is higher than autofluorescence issues. Like an organ with blood vessels, inhomogeneous structures remain challenging for comparison since it is hard to define a criterion leading to a stable average brightness measure. Here we select structures by hand (Fig. 3(a)), calculate a histogram between the darkest and brightest pixels of the selection (Fig. 3(b)), and determine the position of the global maximum (Fig. 3(b), black vertical line). If the selection consists of n pixels, our histogram possessed 0.5 × sqrt(n) bins, proven precise and stable against noise. The expectation value of all bins gives the average brightness measure (Fig. 3(b) red vertical line) next to the histograms’ maximum being on top above 90% of the histograms’ maximum (Fig. 3(b) black horizontal line). The bins between the two vertical blue lines (Fig. 3(b)) thus contribute to this expectation value.

 

Fig. 3. Calculation of the expected gray value of inhomogeneous structures. (a) A representative region of interest (ROI) is extracted from (b) a histogram between the darkest and brightest pixels. Pixel counts are attributed to 0.5×sqrt (n) bins. Blackline: Estimate the central peak position. Redline: average brightness measure. The bins between the two vertical blue lines contribute to the average brightness measure.

Download Full Size | PDF

2.3.3 3D Volume imaging and displacement correction

The recording of consecutive slices enables 3D volume imaging with high resolution. However, due to technical repositioning inaccuracy of the sample in y-direction after rotation of the handwheel, the successive images are taken at slightly different lateral sample positions, leading to a misaligned 3D volume image. Therefore, the displacement correction without any additional fiducial is done in a three-step process using one python script for each step, as shown in Code 1 [26], Code 2 [27] and Code 3 [28].

First (ImageStack01Analyser.py), the relative lateral displacements (xy) between consecutive images of a 16-bit linear raw image stack are estimated by a cross-correlation between neighbor and higher-order neighbor images. With this redundancy of multi-order neighbor images, misjudgment on displacement due to impurities (e.g., slices stick on the surface, incomplete cut) can be avoided efficiently.

Second (ImageStack02SmartIntegration.py), the relative displacements, estimated in the previous step, are integrated into absolute displacements of each image, exploiting the redundancies to avoid disruptions.

Third (ImageStack03Converter.py), each slice of the 16-bit linear raw image stack is shifted back to a common origin using the estimated absolute displacement information of the previous step. Then, according to the reproduction medium, the images go through a level and gamma correction process using a parameter CSV file.

3. Results

The proposed block face episcopic imaging system's functionality was established on a Leica CM3050s cryo-microtome. The high-resolution camera and all the filter sets are mounted on a self-designed stage which can be easily attached and removed from the cryo-microtome. The camera and lens are set up on a motorized stage that the software can control to adjust the focus in the z-direction. The camera shifted approximately 1 mm backwards automatically from the sample whenever the emission filter was activated and inserted. Thus, the focus offset of the filter is compensated with the motorized stage during acquisition. A ± 58 µm depth of field (using f/4.8 and assuming 6 µm as the acceptable circle diameter of confusion) combined with a 34.6 mm field-of-view (21.6 mm x 27 mm) diagonal requires a very straight alignment of the optical system relative to the sample. The optical focus could be adjusted by fine-tuning the pitch and yaw control on the stage. The acquisition software calculates the focal plane angles from three distinct focus positions at the sample (using 2D linear regression). It provides instructions to the operator for adjusting the pitch and yaw on the optical system (Supplementary Document File 1). We note that our images at 0.1 NA should yield a diffraction-limited resolution of 5 µm at 500 nm illumination. Consequently, the systems under-sample our images by about a factor of two, a compromise we had to accept to achieve the required imaging volumes.

To further demonstrate the features of our proposed automated multicolor mesoscopic imaging system, we performed a series of sectioning on FVB/N mice. To evaluate the functionality of the imaging system in different excitation and emission wavelengths, we employed several fluorescence markers and dye-labeled nanoparticles with various visible and near-infrared (NIR) excitation wavelengths. In vivo stained 8 to 14 weeks old mice were euthanized and frozen. The frozen mice were mounted onto the holder of the cryo-microtome and subjected to slicing. The in-house designed software fully automated the cryo-sectioning and imaging process. The software controlled the camera and cryo-microtome through the Arduino Mega 2560 Rev3 microcontroller. Image acquisition parameters such as the numbers and sequence of images acquired in the RGB, bright field, and fluorescence channels and their individual exposure time are adjustable. The system initiated the cutting actions, interleaved by image acquisitions. The sectioning and imaging procedure were run in a complete automated mode with the number of cycles inserted. The focus and brightness of each channel were optimized before starting the cycle and maintained throughout the whole process loop. The thickness of each slice can be modified on the cryo-microtome. After each cut, 2D images were acquired continuously from the remaining sample piece mounted on the holder. Subsequently, 3D reconstruction of the whole sample could be obtained from the series of 2D images.

RGB images were taken from different partitions of the mouse, i.e., thorax, upper, lower abdomen, and pelvic region, visualizing the mouse's cross-sectional anatomy. All RGB images shown in Fig. 4 were acquired with 4500 × 3600 pixels and 6 µm pixel size. All organs, including the lung, heart, liver, stomach, gastrointestinal tract, spleen, kidney, and bladder, can be easily identified from different mice's sections’ transverse planes. Also, smaller structures such as the trachea in the lung, blood vessels, gallbladder, intestinal villi, or the white and red pulp in the spleen are well resolved (Fig. 4). The internal structure of the lung (Fig. 4(a)), the tiny blood vessels in the fat tissue surrounding the intestines (Fig. 4(b)), the rugae in the stomach, and the cortex of the kidney (Fig. 4(c)) can also be recognized clearly from the high-resolution cross-sectional images.

 

Fig. 4. The anatomy of a mouse is resolved by the mesoscope. Continuous anatomical cross-sections were recorded automatically by cryo-sectioning. Images depict different cross-sections to elucidate the resolution. (a) thorax, showing a circle of rid bones surrounding lung and heart, *the enlarged regions represent the lung and trachea. (b) This imaging technique captured the upper abdomen with stomach and liver, **gallbladder filled with bile acid. (c) Lower abdomen with various organs, kidneys, intestines, spleen. (d) In the pelvic region, the image displays a urine-filled bladder. *** Villi in the small intestines and smaller blood vessels inside the fat tissue surrounding the intestines are resolved. ****white and red pulp of the spleen can be differentiated. More than 20 images at each section were taken, and representative photographs (where most anatomical features were visible) were chosen.

Download Full Size | PDF

Continuously recording multiple slices encountered the risk of relative lateral displacements between the consecutive images. These lateral displacements mainly result from the repositioning inaccuracy of the sample in the y-direction after rotation of the handwheel. They can be corrected using our python scripts, as shown in Code 1 [26], Code 2 [27], and Code 3 [28]. After several tens of unattended and automated cuts, cut debris often adhered to the edge of the block specimen, making it difficult to perform a simple nearest-neighbor cross-correlation and automatic displacement correction. We utilized up to fourth-order neighbors for cross-correlation to robust the displacement correction against disturbing images. A later comparison of the redundant displacement information enables us to identify problematic slices and avoid disruptions in the 3D data stack during displacement correction. The utilization of higher-order neighbors for cross-correlation might be compelling with image stacks harboring unwanted debris. In the experiment depicted in Fig. 5, debris and poor-quality images were minimized by removing them manually and replacing the cryo-microtome blade regularly every 150 slices before becoming dull. Here, nearest-neighbor-based alignment results in a visual smoothening of the image that is also depicted by the mid-line intensity plot in Fig. 5.

 

Fig. 5. 3D reconstruction with different levels of displacement corrections (view: zy, x-position: 8.37 mm). (a) Left (no alignment): the misalignment of the image at z position results in distortions on the image stack. (b) Center (alignment by the nearest neighboring slice cross-correlation, 1st order): the simple correlation followed by position correction yields a fair result. (c) Right (alignment by fourth-order neighboring slices cross-correlation, 4th order): smoother alignment at tiny organ structure was observed compared to the center image. (a-c) The mid-line (red) intensity plots below the images depict the visual smoothening by reducing noise and spikes for the different alignment methods.

Download Full Size | PDF

A mouse's head with great sensory organs, tiny vessels, and nervous systems was subjected to 2D imaging and subsequently concatenated into 3D stacked images and videos. A series of fine slicing with a thickness of 10 µm was performed, and images were taken after each slicing. A total of 1700 RGB images were acquired continuously to reconstruct the structure of the mouse head. After the automated displacement correction described above, 123 slices were removed due to large debris attached to the slices. The 3D reconstruction from 1577 RGB images of a mouse head is presented in several videos from three different planar views, as shown in Visualization 1, Visualization 2 and Visualization 3 (Ref. [29]). Figure 6 displays single images from the 3D transverse (Fig. 6(a)), sagittal (Fig. 6(b), c), and coronal view (Fig. 6(d), e). The structures and location of the brain, eyes, nasal meatus, tongue, and teeth were visible in a three-dimensional view.

 

Fig. 6. 3D reconstructions images of a mouse head. The whole mouse head was subjected to cryo-sectioning with a thickness of 10 µm. Images were taken at every slice from the transverse direction (XY), around 1700 slices. 3D reconstructions were done on the 1577 slices and displayed in a multiplanar direction. (a) XY direction, the original slicing 2D image, showing the transverse view of the mouse's head. (b-e) The 3D reconstructed image stack from 1577 2D images is depicted in two planar directions. (b, c) Images of ZY direction from 3D reconstruction display (b) left and (c) middle sagittal cross-section view. (d, e) Images of XZ direction from 3D reconstruction, (d) top and (e) bottom coronal planar view of the head (see Visualization 1, Visualization 2 and Visualization 3, Ref. [27]).

Download Full Size | PDF

The current setup of this mesoscopic camera, which can be equipped with different LED light sources for excitation (here: 530 nm, 625 nm, or 780 nm), allows the detection of a wide range of fluorescence markers (Fig. 7). To evaluate the feasibility and sensitivity of this system in capturing the fluorescence signal inside the tissues, we performed several experiments with the administration of different fluorescent dyes and dyes labeled nanoparticles. Each fluorescent marker accumulates in separate compartments of a mouse in vivo, which can be displayed and quantified using this imaging system (Fig. 7,8). Animals injected with Cy3 labeled nanoparticles were excited at the wavelength of 530 nm. The Cy3 fluorescence signal detected in the kidney and gallbladder disclosed these polymeric micelles’ renal and hepatobiliary clearance pathways. The polymeric micelles from the same neat material and a different fluorescent-labeled (Cy5) were injected into another animal and subjected to imaging at excitation at 625 nm. Cy5 labeled nanoparticles again distinctly accumulated in the gallbladder, kidney, intestines, and bladder. The same results were obtained even with different dyes at different excitation LED sources and emission filter sets, validating the functionality and accuracy of this system in tracking fluorescence tag nanoparticles in tissues. We further examined other dye-labeled lipid nanoparticles conjugated with hepatocytes targeting ligand DY-635. The promising result was obtained with strong fluorescent signals detected in the gallbladder, confirming that DY-635 liposomes have been taken up by hepatocytes and further eliminated and accumulated in the gallbladder [4]. The Harderian glands of murine that exhibit natural red fluorescence due to porphyrins were also visualized with the strong signal by this mesoscopic camera. At the same excitation wavelength, we also characterized another fluorescent molecule Allophycocyanin (APC), frequently used to label antibodies for their detection under flow cytometry. Here we injected an APC-F4/80 antibody conjugate into mice intravenously. The F4/80 antibody exclusively binds to the murine macrophages, which present abundantly in the spleen, and the far-red fluorescence intensity was observed in the red pulp of the spleen [30,31].

 

Fig. 7. Combine imaging of fluorescent markers and anatomical structure. The fluorescent molecules analyzed depend on the installed fluorescent excitation and emission filter. It is also feasible to evaluate the distribution profile of dye-labeled polymeric and lipid nanoparticles. For instance, the fluorescent signals from Cy3 and Cy5-labeled polymeric nanoparticles are excited using two different excitation wavelengths (530 nm and 625 nm). The same accumulation behavior was observed in the kidney (Kd), gallbladder (Gb), intestines (Int), and bladder (B) for this same type of nanoparticles. DY-635 conjugated liposomes also show accumulation in the gallbladder [4]. Another commonly used fluorescent molecule is Allophycocyanin (APC), tagged to an F4/80 antibody that binds specifically to macrophages. APC fluorescence is present in organs with a high density of immune cells, such as the red pulp of the spleen (Sp). The 625 nm excitation can also be employed to detect auto-fluorescence signals, for instance, the Harderian Glands (Hd) surrounding murine eyes with porphyrin-rich cells. This imaging system is also feasible for evaluating near-infrared (NIR) dyes (excitation 780 nm). Indocyanine green (ICG), a clinic used dye for assessing liver function with emission wavelength at 814 nm, can be visualized in the liver (L) and intestines (Int). Six mice were used to generate the images with 5 fluorescent markers and 1 autofluorescence. Each mouse was injected with one fluorescent marker. More than 20 images for each fluorescent marker section were taken to select representative images of critical anatomical features.

Download Full Size | PDF

 

Fig. 8. Biodistribution profile and quantification of indocyanine green (ICG) and Cy5 labeled nanoparticles in mice. Quantification of ICG (fluorescence in green): (a) Cross-section images from the upper abdomen of SHAM and PCI animals showing that ICG stains the liver (Li) and quantification of the fluorescent signal in the liver depict a higher uptake by PCI animals (b) Transverse section of the lower abdomen gives a clear view on the intestines (Int). ICG significantly accumulates in the intestines of the healthy animal (SHAM), revealing a faster elimination of ICG to bile acid in the SHAM group than in animals suffering from sepsis (PCI group). Quantification of Cy5 labeled nanoparticles (fluorescence in green): (c) Cross-section images of the upper and lower abdomen of injected mice revealed the clearance pathway of these nanoparticles. This Cy5 labeled nanoparticles significantly uptake by hepatocytes and accumulate in the gallbladder and kidneys, followed by hepatobiliary and renal clearance. The outliners indicate the interested organs for fluorescence signal quantification. Three animals for each group (Sham, PCI, Cy5 nanoparticles). An average of 10 images per organ were taken for fluorescent signal quantification. The significant test is carried out using Pairwise Wilcoxon (Rank Sum) Test. Significant level, **** P ≤ 0.0001

Download Full Size | PDF

We further examined the feasibility of this mesoscopic system in detecting NIR dyes (Fig. 7,8). ICG is a NIR fluorescence tracer applied, for example, in clinics to evaluate patients’ liver function [32,33]. ICG administered to healthy (SHAM) animals or animals that suffer a life-threatening infection causing liver dysfunction, induced by the model of peritoneal contamination and infection (PCI) measured liver failure in vivo [4,34]. In a well-functioning liver, ICG is quickly taken up by hepatocytes, the organ's primary metabolic cells, eliminating it into the bile. It is drained into the intestine and finally excreted [34]. ICG fluorescence was detected in the murine liver at 780 nm. The elimination rate of ICG by the liver is the critical indicator for physicians to identify hepatic malfunctions. In sepsis, hepatocytes lose their ability to eliminate endo- and xenobiotic such as ICG from the liver into the bile. In humans and mice, as a consequence, the ICG accumulation in the failing liver is significantly higher in sepsis than in healthy controls [35]. Hepatocytes in the liver drain ICG into the intestines through the biliary system consisting of bile ducts and the gall bladder.

Consequently, the ICG concentration in the intestine is low when the liver fails to function in sepsis, and quantification of the ICG intensity in the liver and intestine can provide information about liver function [35]. We analyzed ICG in the liver tissue and intestine of healthy (SHAM) and animals with sepsis, induced by peritoneal contamination and infection (PCI), 45 min after intravenous ICG injection (Fig. 8(a), b). We imaged and quantified ten slices of the mid-abdominal region showing the liver, gall bladder, and intestine with a step size of 0.1 mm. For each animal, 50 areas of interest (ROI) within the boundaries of the liver and intestine are drawn. The quantification of the ICG signal in all the ROIs was done simply by Fiji ImageJ. ICG signal intensities in those ROIs are higher in animals’ liver tissue from the PCI group than in healthy controls (SHAM) (Fig. 8(a)). The ICG signal detected in the intestine was significantly lower in the PCI compared to the SHAM group (Fig. 8(b)). The results aligned with the above-described elimination route of ICG and its retention in hepatocytes when the liver fails. By examining and quantifying the well-established diagnostic dye ICG, the proposed mesoscopic imaging system validates its capability in providing a comprehensive and quantifiable characterization approach to tracking fluorescent dyes. In addition, we quantified the fluorescent signal of Cy5 labeled nanoparticles in the mice's organs, as shown in Fig. 7. Quantification was performed using the in-house developed software (see Method, Data processing II). Quantification results aligned with our previous finding using intravital microscopy (IVM) on these polymeric nanoparticles were preferably taken up by hepatocytes and eliminated through the gallbladder (Fig. 8(c)). Despite elimination through the hepatobiliary, the result further reveals that the renal route also cleared these polymeric nanoparticles.

4. Discussion

The imaging setup utilizes block-face episcopic imaging of the sample mounted onto a commercial cryo-microtome. Compared to previously published cryo-imaging systems [22,23], we designed a scanning-free mesoscopic imaging system assembled and structured on a handling stage that can be mounted onto the cryo-microtome and easily removed with minimal setup effort. Although this automated imaging system reported in this work was established using a Leica CM3050s cryostat, it can also be installed onto any conventional cryostat by adapting the handling stage. Our automated multicolor mesoscopic imaging system allows direct tracking and imaging of fluorescent dyes and markers in various organs on a large area scale. The proposed system significantly reduces the efforts on demanding formalin fixation steps. In addition, since the images are taken directly from the specimen, slicing and mounting samples on a glass slide may become redundant, particularly with the possibility to image fluorescently tagged antibodies or other probes injected pre- or directly post-mortem. This high-resolution camera performs block-face imaging right after the sample sectioning without any additional changes in standard procedures of usual cryo-sectioning. The animal model treated with fluorescent dye can be directly subjected to slicing and imaging. The complete automated process can significantly reduce the workload and conveniently acquire large sample blocks. This work's anatomical 3D model of a mouse head is reconstructed from 1577 high-resolution 2D images after displacement correction. The algorithm allowed the sectioning without manual steps or the operator's interference, eased obtaining 3D reconstructions.

We developed a robust and fast method to algorithmically align the individual images of a 3D stack with smooth and error-free image re-slicing in all spatial directions to concatenate all images. The inaccuracy and deviation of sample position in the microtome could be compensated algorithmically. Although most noninvasive approaches such as magnetic resonance imaging (MRI), computed tomography (CT), ultrasound imaging could provide three-dimensional anatomical information in animal models, these approaches harbor difficulties in visualizing the detailed structures of the organs or fluorescent tracer distributions [36]. Tracer-derived techniques such as positron emission tomography (PET) provide sensitive and high quantitative images, but it often suffers from limited spatial resolution and lack of detailed anatomical information [37]. In contrast, our imaging system enables 2D image acquisition in large areas (21.6 mm x 27 mm), which can be rearranged into 3D models, providing excellent spatial resolution without losing the image details in the tiny structures. Our camera undersamples the optical image even at infrared wavelengths with a numerical aperture of 0.1 and a pixel size of 6 µm × 6 µm. This forces us to define the practical resolution limit based on the achieved image sampling. Using the Nyquist criterium, we obtain an aliasing-free lateral resolution up to horizontal and vertical line pairs of 12 µm and diagonal line pairs of 8.5 µm. The axial resolution for 3D imaging depends on the penetration depth of light into the sample and the mechanical slice's thickness. The slice's thickness is adjustable through the cryo-microtome. A reduced slice thickness improves the axial resolution on the 3D image stacks. However, thin slices can result in almost identical consecutive images due to the transparency of the sample. Based on repeated trial testing, the thickness of 5 and 10 µm gave the best results for 3D image stacks of mice. Since we do not employ optical sectioning, the axial resolution must not be confused with the ability to suppress out-of-focus light. Mesoscopic imaging visualizes gross and fine tissue structures such as intestinal villi (diameter: 200-500 µm [38]), the trachea in the (lung diameter: 1.5 mm [39]), or the rugae in the stomach. Thus, this automated imaging system facilitates potential access to provide a detailed 3D reconstruction view of desired organs or parts of the body. Unlike the other techniques that only image a single excised organ or selected localized parts of the animal, the entire cross-section of the mouse is projected into one single image, which allows 3D reconstructions of the torso or head's whole segment. This mesoscopic camera equipped with an exchangeable LED excitation light source and filter set allows detection and visualization of a broad range of fluorescent markers. Fluorescent signals were captured precisely in specific organs based on the distribution nature of the marker; spread-over signals on surrounding tissues were not detected. These findings confirmed the applicability of this mesoscopic camera in visualizing the distribution of administrated dyes or drugs in mice.

Moreover, quantification of fluorescence intensity in microscopic images often faces an issue with the deviation of fluorescence signals taken in different images. This mesoscopic camera facilitates extensive area imaging. Thus fluorescence signals from various tissues/organs can be captured in the same image, reducing the discrepancy during quantification. Quantification done on the NIR dyes, ICG injected animals further validate the accuracy and sensitivity of this system.

It was shown that brightness-based concentration measurements on the same tissue type are comparable across individual samples. However, brightness depends on the depth of light penetration in different tissue types because of the missing out-of-focus fluorescence suppression in widefield imaging. Steyer et al. reported that sub-surface fluorescence could be corrected by additional assumptions and data processing, given that the next slice shows the pure sub-surface fluorescence [40]. Unfortunately, the transmissivity of the first cut is still unknown, so a reconstruction of transmissivity and fluorescence without accurate optical sectioning is not possible in general. A sensitive and low noise 16 MPx CCD camera based on an ON Semi KAF-16200 sensor was used for this work. This camera has sensitivity in the visible and near-infrared range. However, a disadvantage of the CCD is the high readout time of 8 s in low noise mode. Camera development continues, and fast, cooled back-illuminated CMOS cameras with suitable characteristics will be more affordable. For instance, cameras based on the Sony IMX455 sensor can take four 60 MPx images with 16 bits in one second, featuring even lower read noise and significantly speeding up 3D volume imaging. Thus, further adaption may be incorporated in the future to increase the sensitivity and specificity of the system, which is presented here as an open platform for research and development.

5. Conclusion

We presented an automated multicolor mesoscopic light- and fluorescence imaging system compatible with commercial cryo-microtomes. Here presented protocols and tools allow block-face imaging of large specimens resolving gross organs and more delicate structures in RGB color and fluorescence. The complete automated process gives advantages in acquiring large numbers of images that could be applied subsequently for 3D reconstruction. Furthermore, the system is designed as an expandable platform containing the essential system calibration functions, imaging settings, image reconstruction, and analysis, providing the opportunity for customizations toward individual experimental questions.