Mouse models have become a powerful and de facto research tool for studying biological processes, pathologies, and therapeutics. With the advances in genetic manipulation, the number of mouse lines continues to increase with no end in sight. These advancements in genetic modification techniques (Bedell,1997) and interest in therapeutics and new imaging agents have been paralleled by development of imaging instruments specifically targeted to the small animal. Instruments include MRI, PET, SPECT, CT, in vivo fluorescence, in vivo bioluminescence, and intravital imaging. The advent of in vivo cellular and molecular imaging techniques enable one to visually, and sometimes quantitatively, evaluate biological processes at the cellular and subcellular level (Massoud and Gambhir,2003) in a living subject. However, each of these instruments tends to be limited with regards to one or more important features, including field-of-view, depth of field, contrast, resolution, and capabilities for cellular and molecular imaging.
Our group has been developing a cryo-imaging system, which gives contrast rich, brightfield anatomical, and fluorescence cellular and molecular imaging of an entire mouse with micron-scale resolution. The cryo-imaging system consists of a mouse-sized, motorized cryo-microtome with special features for imaging, a modified, brightfield/fluorescence microscope, and a robotic xyz imaging system positioner, all of which are fully automated by a control system. By alternately sectioning and imaging, the system acquires three-dimensional (3D), very high-resolution, large field of view (FOV), brightfield anatomy, and fluorescence molecular image volumes from sequential images of the tissue block face. Such images can be used in a variety of applications, including anatomical phenotyping, high-resolution vascular imaging, and fluorescently labeled cell tracking, among many. In this article, we describe the system, present the methods used in a cryo-imaging session and illustrate promising results from assorted imaging experiments.
Cryo-imaging is unique among all 3D in vivo (e.g., micro-MRI) and microscopic techniques, because it allows micron resolution and information-rich contrast mechanisms over large 3D fields of view. Imaging modalities like CT and MRI can image a whole mouse but lack adequate resolution and can produce only gray-scale images. Optical Projection Tomography is limited to small samples such as embryos and requires treatments to reduce scatter and increase transparency (Sharpe et al.,2002). Diffuse optical tomography tackles the light scatter problem with advanced algorithms (Boas et al.,2001) and can even image a whole mouse. Resolution, however, is limited and trans-illumination images provide little anatomical detail. Traditional approaches of imaging sections to create a 3D volume involve serially cutting the sections, applying histological processing, mounting on glass slides, digitizing the slides, and then creating a 3D volume from the two-dimensional (2D) images. Over the years, there have been reports on 3D reconstruction from serial histology sections (Ware,1975; Manconi et al.,2001) for a variety of biomedical applications (Griffini et al.,1997; Kaufman et al.,1997). However, such processing results in incomplete, torn, sometimes spatially distorted sections and is inherently prone to misalignment between sections. A block-face imaging system like ours alleviates much of these difficulties. Episcopic imaging of the block-face had its most famous exposition through the visible human projects in the USA, China, and Korea (Ackerman,1991; Spitzer and Whitlock,1998; Zhang et al.,2003; Park et al.,2005). These projects have provided extraordinary new information about human anatomy, and data have been used for a variety of applications. Block-face imaging setups on custom-modified microtomes have been used for imaging excised brains (Cannestra et al.,1997; Kenzie-Graham et al.,2004; Annese et al.,2006). Episcopic imaging in a wide variety of wax-embedded biological samples, most notably embryos, has been achieved through episcopic fluorescence image capturing (EFIC) which uses endogenous, broad-spectrum tissue autofluorescence to resolve tissue structures (Rosenthal et al.,2004). Cellular resolution has been achieved on resin embedded embryos and tissue samples in high-resolution episcopic microscopy (HREM) using histological stains (Weninger et al.,2006). Adult whole mouse images, on the other hand, have been limited to low-resolution photographs of sections taken for autoradiography studies (MacLaren et al.,1999) and CT and block-face imaging for creating a mouse atlas (Dogdas et al.,2007).
Our first efforts in block-face imaging methodology were through development of a system in our laboratory to validate MR image data from interventional MRI radiofrequency ablation experiments (Lazebnik et al.,2002; Breen et al.,2004). In this predecessor system, color block-face images of fixed, wax embedded tissues were obtained following the acquisition of each relatively thick slice––about 3 mm each. The process was entirely manual but the experience gained from the project and the image processing software tools that we developed prompted us to go forward with designing a fully automated, high-resolution block-face imaging system. We first developed an initial proof-of-concept system with a stationary FOV and minimal automation which yielded encouraging results (Roy et al.,2006). This system has now been vastly improved to provide a flexible, high-resolution, fully automated, multimodality (brightfield and fluorescence) cryo-imaging system.
CRYO-IMAGING SYSTEM AND METHODS
The cryo-imaging system (Fig. 1) consists of four major subsystems––the cryomicrotome, the microscope imaging system, the robotic xyz positioner, and the computer control system. The cryomicrotome is a motorized, large section, whole body cryo-microtome (Model 8250, Vibratome, St. Louis, MO) with section thickness adjustable between 2 and 40 μm and a maximum specimen dimension of 250 mm × 110 mm × 5 mm. For an automated cryo-imaging setup, we interfaced appropriate control signals within the cryo-microtome to communicate with the control computer over a single Ethernet cable. A computer/manual selector switch was added to the operator pushbutton box to allow manual override and control. The cryomicrotome frame was mechanically modified to connect an XYZ robotic positioner carrying the imaging system. The imaging subsystem comprises of a stereo microscope (SZX12, Olympus, Japan), coaxial fluorescent attachment with multiple filter cubes, low light digital camera (Retiga Exi, QImaging, Canada), and brightfield and fluorescent light sources. The long working distance of the stereo microscope allows for optional tissue collection. Using multiple microscope objectives and zoom settings, the field-of-view can be varied to cover an entire mouse or down to a small organ and image at in-plane resolution of ≈3 μm. To enable very high-resolution imaging over a mouse-sized FOV, we designed a 3-axis robotic positioner. The computer control system automatically pans the positioner over the specimen for a high-resolution tiled image acquisition.
The computer control system controls the sectioning and image acquisition sequence through the custom developed Programmable Sectioning and Cryo-Imaging (ProSCI) software. Through the Graphical User Interface (GUI), the operator enters specimen information, sets up the illumination sources, chooses imaging modality (brightfield, fluorescence or both), and defines the imaging regions-of-interest (ROI). In case a ROI needs to be imaged at a high-resolution, tiled image acquisition can be setup by defining the area and the software calculates the minimum number of tiles necessary to image the ROI. The Image Processing and Visualization system is a quad-core Windows 64-bit PC with 32GB of RAM (Dell Inc, TX) capable of handling large cryo-image volumes. A typical whole mouse, if imaged at ≈15 μm in-plane resolution and sectioned at 40 μm, generates typically >50 GB of raw data. A suite of MATLAB (Mathworks, Natick, MA) custom programs and custom AMIRA (Mercury Computer Systems, San Diego, CA) scripts are used for image preprocessing tasks and 3D visualization.
Sample preparation and imaging
Several steps are required to acquire cryo-image data. Animals are euthanized in a method approved by Case Animal Resource Center (ARC) which consists of either inhalation of Carbon dioxide delivered from tank or anesthetization using an agent such as pentobarbital, at a dose prescribed by the Case ARC. Animals are covered under various IACUC-approved projects. Optionally, to enable visualization of vasculature, India ink is injected through left ventricle of the mouse before sacrifice. A cryo-embedding compound, OCT (Optimal Cutting Temperature, Tissue-Tek, Terrance, CA), is rubbed on the skin of the euthanized mice to ensure that the carcass is completely wet. This ensures that air will not be trapped between the fur and the skin which may impede heat exchange. The wet carcass is then put inside a custom mould made out of aluminum foil and filled with OCT. More OCT is poured inside the mould after placing the carcass so as to completely immerse the carcass in OCT. Finally, the entire mould is covered in more foil and placed inside a freezing chamber made of Styrofoam and filled with liquid Nitrogen. Typically, samples such as whole mice are removed from this coolant bath after 15 min. It is important to make sure air bubbles are not formed in the OCT bath as these will create holes inside the embedded block. We pour the OCT from the plastic bottles gently to prevent air bubbles. If some form, we use a pipette tip to move them aside. The mould assembly is then removed from the liquid nitrogen bath and placed inside the cryo-microtome chamber for equalizing the specimen temperature to the cryo-microtome temperature. In case the cryo-imaging is scheduled to be performed at a later date, the mould assembly instead is stored in air-tight bags in a −80°C freezer. For moulds kept inside the cryomicrotome chamber for immediate cryo-imaging, the mould is removed from the specimen after 2 to 3 hr and the frozen specimen is mounted on the microtome stage using more OCT. For different sized specimens, optimally-sized specimen stages are used. An initial “facing-off” cycle continuously slices the block at maximum thickness until animal tissue become visible. Then the desired slice thickness is set. Typically 20–40 μm section thickness is used for whole mouse study and 10–20 μm for smaller specimens such as excised organs or embryos. The imaging system is readied, imaging protocols are set and the system is started to alternately slice and image the block-face. The system can run completely unattended. Although automatic image acquisition sequence is in progress, a notification option notifies the user periodically about imaging progress through email and cell phone text message alerts. A reduced-size version of the latest cryo-image acquired is attached to the email
For brightfield anatomical cryo-imaging, wild-type C57BL6J mice were used. For fluorescent cryo-imaging a transgenic mouse line SMGA/EGFP was used that expresses Enhanced GFP (EGFP) under control of smooth muscle gamma actin (SMGA) promoter (Szucsik et al.,2004). For cryo-imaging of embryos, wild type and Cited2−/− transgenics as well as the SMGA/EGFP transgenics were used.
Following acquisition of 2D images on the imaging workstation, we perform a series of off-line image preprocessing tasks on the image processing and visualization system. Any nonuniform illumination pattern is compensated using a reference image of a white card. For tiled acquisition, individual tiles are composited together. A typical intensity-compensated stitched 2D whole-mouse image is 5,300 by 2,100 pixels. These final images are then automatically aligned to each other to correct for either operator adjustment of FOV due to a growing tissue or minor misalignments due to the repositioning error of the stage at the return point. In experiments involving fluorescence cryo-imaging, we have observed that bright fluorescence can be visible on the block-face image from a few sections below. This results in subsurface fluorescent structures being visible on multiple images and if not corrected, would lead to artifacts in 3D reconstruction. We have developed a “next-image” method to reduce subsurface features. In this method, we take the next section image, attenuate it on a tissue-specific basis, and subtract it from the block-face image to provide a corrected image. The process is repeated in a pair-wise fashion throughout the volume. We have experimentally determined the attenuation coefficients of various tissue types identified in mouse cryo-imaging and apply coefficients from this library in imaging experiments
Fully automated volume visualization
As compared to gray-scale MRI or CT data, color cryo-image data provides many more opportunities for fully-automated volume visualization. We have determined that in addition to surface rendering of tissues/organs delineated manually by the user, fully automated volume visualization can be performed using color vectors. For example, we developed volume visualization schemes where the color contrast coupled with suitably derived transparency values provide an “implicit delineation” resulting in a quick visualization of an entire 3D data set. The baseline scheme for visualization involves the use of true colors with the opacity set to be equal to the grayscale intensity (or its inverse) at any point in the 3D volume. This basic scheme provides a visualization of lighter (or darker) structures in the volume. Such baseline schemes have been augmented by designing color feature detectors which takes into account the dominant color of a region to segment it. For instance, regions containing a high proportion of red can be extracted using a feature detector shown below:
For detecting stomach and intestinal regions in the adult mouse data set, which were found to be brown in color due to food, we have exploited the fact that brown is made up from 0.5R + 1.0G + 0.0B. Therefore, we have employed the following feature detector for brown:
Subsequent to feature detection, opacity transfer functions (OTF) were used to assign an appropriate α opacity value to each voxel. Below, we show an example of color-based step OTF that we have employed in our visualizations:
where T is an empirically determined threshold. We have also employed ramp, sigmoidal and power law OTF and in other algorithms, we have tried to optimize opacity by imposing several simultaneous constraints on an opacity function which maps each voxel to a custom opacity value. Furthermore, edges in the 3D data set (which correspond to tissue boundaries) can be enhanced by setting the opacities to the gradients of any one of the color channels or the grayscale intensity channel. Color information can be exploited for gradient computation (Ebert et al.,2002) which leads to a better delineation of structures based on the vector-valued color data. We have developed a number of schemes utilizing combinations of color values, transparency values, and image statistics. Standard tools are used in AMIRA for extracting digital sections along arbitrary planes and for volume presentation such as snapshot capture, movie-making, and so forth. Although it becomes necessary to better determine interesting substructures in the volume, we use semi-automatic methods such as seeded region growing followed by some manual corrections/adjustments. An interactive GUI was developed for determining parameters for volume visualization enhancements
To demonstrate this unique imaging device, we include a variety of imaging experiments. We imaged an 8-week-old C57BL6J adult mouse using a high-resolution, brightfield, tiled acquisition with 15.6 μm pixels and 40 μm section thickness, giving 55 GB of image data. A coronal section image (Fig. 2) is composited from 20 (=4 × 5) individual tiles. High-quality image stitching was achieved without perceived seams. Major organs like eyes, heart, lungs, liver, stomach, small intestine and colon, as well as details like the optic nerves, the rectus muscles of the eyes, the septa in the nose and the ribs can be clearly discerned. Although the whole mouse figure has been compressed to fit to the page width, we show higher resolution, cropped regions from various organs (Fig. 3) detailing cardiac chambers and vessels, muscle architecture, the villi of the small intestine, muscle layers of the stomach, and sections of the eye. A total of 663 sections yielded 13,260 individual images.
Three-dimensional renderings are given in Fig. 4. A cutaway 3D volume rendering shows the three orthogonal planes inside the mouse (Fig. 4a). Using seeded region-growing and manual interaction, we marked the lungs from the same mouse and reconstructed in 3D (Fig. 4b). Using an advanced volume rendering scheme, we computed optimal opacity values at different regions in the 3D volume. This fully automated volume visualization (Fig. 4c) shows lungs, brain, spinal cord clearly visible without requiring any section-by-section manual interaction. A color feature detector was used to visualize automatically the stomach and intestine region (Fig. 4 inset). Using a combination of low resolution and the high-resolution dataset, 3D zooming allows high-resolution 3D reconstruction of hepatic vessels from a low-resolution complete volume visualization of a liver lobe (Fig. 4d).
Perfusion with colored or fluorescent dyes provides opportunity for imaging vasculature at high-resolution. We show results of a mouse perfused with India ink with the 2D cryo-images and the resulting 3D vasculature branches with branch sizes down to 10 μm in diameter (Fig. 5). Using the next-image subtraction processing algorithm, the subsurface vasculature was removed on a section-by-section basis and the vasculature was segmented using interactive 3D region-growing.
Fluorescent transgenic mice models provide another contrast mechanism for detailing anatomy. As an example, we imaged the Lessard SMGA/EGFP transgenic mouse which expresses EGFP under control of smooth muscle gamma actin promoter. This mouse is very useful for studying the smooth muscle cells of the urogenital and gastrointestinal tracts. A 3D reconstruction from the fluorescence image volume shows the continuity and morphology of the gastrointestinal (GI) tract and the genitourinary (GU) system with a section from the brightfield image volume inserted into the 3D view (Fig. 6). In this case the animal was sectioned at 40 μm, and brightfield and fluorescence images were both acquired.
The mouse sized, cryo-imaging system can also be used for smaller samples such as embryos (Fig. 7). A Cited2−/− transgenic embryo was sectioned to discern phenotypic differences with a wild-type. The mutant exhibited lack of adrenal gland in comparison to the normal wild type. The dataset from the mutant embryo was used for automatic volume visualization of liver, heart, vasculature, and the spine. The same transgenic mouse in Fig. 6 was also investigated in embryonic stage E16.5. A 3D reconstruction from the fluorescent cryo-image volume illustrates the expression of smooth muscles in entire GI and GU system as well as in the airways of the lungs.
DISCUSSION AND CONCLUSION
Principal advantages of the Case cryo-imaging include micron-scale resolution, mouse-sized depth and breadth of field, color brightfield anatomy, and molecular fluorescence imaging. As results demonstrate, the tiled whole mouse dataset provide details and color contrast not obtained with traditional gray-scale medical imaging modalities (CT, MRI, etc.). The intestinal villi of a mouse are typically 50–60 μm in diameter (Abbas et al.,1989), which is visible clearly in the cryo-images. The muscle architecture as well as atypical muscle layers and organization in the stomach region illustrate the histology-like anatomy seen in cryo-images. The resolution capability of the system was tested using standard high-resolution test target (≈3.1 μm, USAF 1951 HI-RES, Edmund Scientific) and is also corroborated by results from the mouse perfusion experiment, which show vasculature segmented down to 10 μm.
Molecular fluorescence studies are aided greatly by anatomical context as well as 3D distribution. The fluorescent cryo-image volume of the transgenic mouse, both adult and embryo, have accompanying brightfield images which provide the anatomical context clues. Further, unlike traditional methods of evisceration of the organs and selective sectioning of localized areas, with this system the entire animal is imaged, thus allowing 3D mapping of the fluorescence across multiple organ systems.
There are many motivations for whole mouse cryo-imaging. First, 3D high-resolution, high-contrast anatomical imaging can greatly aid mouse phenotyping. A systematic study of the relationship of phenotypes and gene mutations is critical to understanding of the underlying molecular and cellular processes (Brown et al.,2006). Second, whole mouse fluorescence imaging at high-resolution allows one to image single fluorescent cells anywhere in a mouse. Applications include imaging of stem cell homing, engraftment, and differentiation in regenerative medicine and imaging of micro-metastases in cancer. Third, cryo-imaging allows one to map fluorescent imaging agents at very high-resolution and sensitivity with fused 3D anatomy, enabling unique determinations of agent specificity. Fourth, whole mouse cryo-imaging is applicable in multi-modality correlative studies. In-vivo imaging techniques such as MRI, bioluminescence, and so forth, allow longitudinal studies of the whole animal albeit at reduced spatial resolution. With dual and triple-reporter gene constructs, whole mouse cryo-imaging can provide both fluorescence and brightfield dataset of the entire animal at a high-resolution, serving both as a validation tool as well as providing the anatomical framework of the findings of the other modalities.
The features of the ProSCI software were developed keeping in mind the need for flexibility in experiment protocol and minimum user interaction during imaging. It allows one to create contiguous tiles and thin sections over the entire mouse at the expense of considerable image acquisition time on the order of a day. Alternatively, many times, one is interested in a few regions of interest. Hence, we have also created very flexible region of interest imaging. Instead of dissecting out organs and imaging them separately, the user can define multiple regions of the interest on the same whole mouse specimen and image at-once, increasing throughput and lowering overall imaging time. The user's productivity is further enhanced because the imaging sequence runs unattended with remote diagnostics provided through email and text messages.
The image processing and visualization modules substantially extend the capability of the system. Using multiplanar reformatting, it is possible to digitally resect along standard orthogonal planes (Fig. 4) or arbitrary oblique plane. This is a significant advantage as compared to histology where one has to be very careful to orient the sample for optimal sectioning. With cryo-imaging, physical sectioning can take place along any orientation and the optimal plane of section can be digitally extracted for analysis later on. Similarly, as the whole mouse visualization demonstrates, automatic volume visualization involves no tedious manual delineation and makes 3D reconstruction tasks simpler and faster. The 3D datasets allow 3D zooming, analogous to 2D zooming, where we use a low-resolution dataset for a quick visualization and then zoom in and reconstruct a specific region using the native high-resolution dataset, as shown with the liver reconstruction. A vessel tree can be rendered in its entirety using the low-resolution dataset to study the whole organ and a particular area can be zoomed in 3D using the original high-resolution cryo-data. Subtraction processing is an important step for reducing subsurface features that may interfere with a crisp image, and is utilized in the brain perfusion experiment (Fig. 5). Perfusion enhanced the contrast between blood vessel and white matter, but the light coloration and transparency of the white matter also meant subsurface blood vessels were partially visible from the adjoining slice. Subtraction processing through our next-image algorithm eliminated this artifact. Next-image processing is a modified neighboring deconvolution method that models the scatter and absorption of light in a given tissue between adjacent images to remove out-of-plane fluorescence. Currently, tissue identification is done manually, but a classification-based semiautomatic tissue identification scheme is being developed. Parameters for different tissue types are either chosen from a library of values or estimated using point-like features in the image data of interest.
Although the system is designed to image large samples such as a whole mouse, it is easily adapted for smaller specimens such as embryos or excised organs. A smaller stage with appropriate mechanical adapters is used for smaller specimens. We demonstrated the applicability of the system for embryo phenotyping through two transgenic mouse embryo studies. Lack of Cited2 gene in mice can result in cardiac malformations, adrenal agenesis, abnormal cranial ganglia, and exencephaly, among others (Bamforth et al.,2001). We cryo-imaged a wild-type embryo and a mutant embryo and detected the adrenal agenesis. Note that such phenotype studies can be carried on an entire litter without the necessity of carefully orienting each sample and sectioning along similar planes. Digital resection along common planes, just like whole mouse specimens, makes analysis tasks easier. The 3D reconstruction of the mutant embryo has been performed by using specific combinations of color channels with no manual segmentation of organs from the 2D images. This volume visualization approach will allow rapid comparative phenotyping in 3D. We also cryo-imaged and produced a fluorescent volume from the embryo of the transgenic mouse model expressing EGFP in smooth muscles. In this case, the entire volume was reconstructed in 3D and is proof-of-concept for experiments involving study of development of such embryos across stages.
As demonstrated through these results, contrast in images is achieved through either exploiting endogenous color contrast in tissues in brightfield images aided by our visualization scheme or contrast aided by either a dye or fluorescent markers. We note that for embryo imaging, especially early-stage samples, the whole-mouse cryo-imaging system is currently suboptimal. Procedures such as EFIC or HREM offer superior contrast and resolution for early-stage embryos. For embryonic samples, although a tiled image acquisition over the sample may improve spatial resolution, obtaining endogenous color contrast which is critical for our volume visualization schemes, is a challenge. In other methods, contrast is achieved by introduced stains in the embedding medium and reducing sample size for better infiltration. In our future work, we will be refining experiments and protocols for enhancing embryonic studies.
The automated Case cryo-imaging system provides an opportunity to study whole mice in microscopic resolution using both brightfield and fluorescence imaging. The demonstration experiments show the efficacy of the system and assert the complementary role it can play alongside in vivo imaging modalities. The Case cryo-imaging system is also being used for validation of MR studies of human blood vessel lesions (Salvado et al.,2006) and for plaque characterization. From anatomy to molecular imaging to stem cell tracking, we hope cryo-imaging would fill a niche and allow imaging in a continuum from mouse to organs to tissues at multiple resolutions.
The authors wish to thank many collaborators for being early adopters of the cryo-imaging technology and providing samples. The mouse in Fig. 2 was provided by P. Hakimi (Dept of Biochemistry, Case Western Reserve University). The brain perfusion experiment in Fig. 5 was conducted in association with Dr. J. Cutter (Case Center for Imaging Research) and the transgenic mouse and embryo in Figs. 6 and 7b was provided by Dr. J. L. Lessard (Division of Developmental Biology, Cincinnati Children's Hospital Medical Center, Cincinnati, Ohio). The embryos in Fig. 7a were provided by Dr. Yu-Chung Yang and Shi Gu (Dept of Pharmacology, Case Western Reserve University School of Medicine). Thanks are due to J. Wikenheiser (Dept of Anatomy, Case Western Reserve University) for helping with anatomical identifications. We also would like to thank the undergraduates in the laboratory, who helped with the testing of the system. This investigation was conducted in a facility constructed with support from Research Facilities Improvement Program Grant Number C06 RR12463-01 from the National Center for Research Resources, National Institutes of Health. This research is supported by the Ohio Wright Center of Innovation and Biomedical Research and Technology Transfer award: “The Biomedical Structure, Functional and Molecular Imaging Enterprise,” and NIH R41CA124270, “Cryo-Imaging of Fluorescently Tagged Cells in the Mouse”. Dr. Wilson has an interest in BioInVision, Inc., which intends to commercialize cryo-imaging technology.