Background

Connectomics refers to the mapping of cellular connectivity in a physiological context, e.g., in organs and organ systems (Sporns et al., 2005; Lichtman et al., 2008; Knothe Tate, 2017). Seamless, multiscale and multimodal imaging provides a platform for interdisciplinary research and the trans-dimensional linking of biological structure and function (Sporns et al., 2005; Lichtman et al., 2008; Eberle et al., 2015; Hageman et al., 2016; Knothe Tate et al., 2016a, 2016b and 2016c; Pereira et al., 2016; Knothe Tate, 2017; National Institutes of Health, 2018; https://www.humanconnectome.org/). The U.S. National Institutes of Health's (NIH) Connectome Project, launched in 2009, aimed to map "the human brain...to connect its structure to function and behavior" (https://www.humanconnectome.org/). In 2018, NIH debuted a program to promote cross-cutting research mapping the cells within the human body (National Institutes of Health, 2018). Connectome maps of brain and other tissues of the body have typically been restricted to small organisms due to challenges associated with cross-scale imaging from nano- to mesoscopic length scales. New imaging technology platforms enable such seamless scale-crossing image acquisition, and current barriers include biophysical hurdles intrinsic to specimen preparation as well as digital hurdles inherent to massive data management, analysis, and curation to maximize sharing across platforms and among geographically far flung users.

Currently, there is a great imperative not only to study the cells themselves but also to refocus at the intersection of and across multiple length scales encompassing the cellular networks that inhabit complex biosystems comprising tissues, organs, and organ systems. The status of these organ connectomes provides a powerful early indicator of degenerative changes and disease states over the lifespan of the organism. Three-dimensional multiscale imaging methods allow unprecedented study of structure and function across length scales; from organ to tissue, cells to molecules (Anderson et al., 2008; Hageman et al., 2016; Knothe Tate et al., 2016; Pereira et al., 2016).

Traditional transmission electron microscopy (TEM), commonly used to visualize cellular ultrastructure, provides superior resolution and contrast. However, this is contingent on the often-laborious preparation of specimens into ultrathin sections (no more than 100 nm in thickness) for electron transparency.

The emergence of multibeam scanning electron microscopy (multiSEM), with 61 or 91 parallel beams, is heralded by a new age of organ connectomics and presents a suitable platform for imaging large areas at nanometer resolution (Eberle et al., 2015; Knothe Tate et al., 2016a and 2016c; Hageman et al., 2016; Pereira et al., 2016; Knothe Tate, 2017). The imaging rates and resolutions of multiSEM far supersede single-beam SEM, imaging an area of 1 cm² at 4 nm pixel size in less than 3 h. It is able to image specimens reaching 10 cm in diameter at up to 4 nm resolution. Despite advances in the imaging technologies, biological tissue specimen preparation has remained unchanged for decades, and hinders translation to cutting-edge imaging modalities (multiSEM). 

This protocol addresses the difficulties associated with specimen preparation workflows enabling multiSEM, as well as multimodal and correlative imaging methods. While having been implemented successfully in peer-reviewed publications (Figure 1), details of the protocol have been requested by the medical and scientific community alike, providing impetus for its publication (Knothe Tate et al., 2004; Eberle et al., 2015; Knothe Tate et al., 2016b; Hageman et al., 2016; Pereira et al., 2016).


Figure 1. Cellular epidemiology studies in large scale maps comprising stitched, scanning electron microscopy images. These high-resolution images are obtained using multiSEM and demonstrate the feasibility of the approach that greatly expands beyond fields of view previously acquired using confocal scanning microscopy. A. Geonavigational methods and the Google Maps Javascript API were utilized to create navigable maps of osteocytes within the human hip imaged using multiSEM and spanning the nano- to mesoscale. The world's first map of the human hip was created by imaging a sample approximately 2.4 x 2.4 mm. Imaging using a 61-parallel-beam Zeiss MultiSEM 505 prototype yielded over 55,000 SEM images, which were stitched into a single image, and rendered in a pyramidal structure for seamless zooming into and out of the map using Google Maps API. The dark area in the middle of the sample is an artifact of automated polishing, a sample preparation step that was later changed to precision CNC milling and hand polishing (see Procedure section). B-C. Google Maps Javascript API enabled the geographic coordinates of physiologically relevant landmarks to be recorded. Blood vessel edges, viable osteocytes and pyknotic (sick) osteocytes were marked with blue, green and red pins respectively. D. The position of each pin could be exported for post hoc analyses, e.g., to test the hypothesis that cell viability correlates to blood vessel proximity, path lengths of viable, pyknotic (sick) and interstitial cells to blood vessels were calculated and compared. E. MultiSEM images of a viable cell, with projections emanating outwards into the matrix and pyknotic cell, no projections observed. F. The role of cell health and connectivity on tissue health was first observed in 3D images of osteocyte networks from (left) healthy, (middle) osteoporotic and (right) severely osteoporotic cortical bone from the femoral head of human patients, using laser scanning confocal microscopy. A shortcoming of this approach is that volumes of interest are limited to e.g., 274 x 274 x 50 microns. Figures adapted and used with permission. (Knothe Tate et al., 2004; Eberle et al., 2015; Knothe Tate et al., 2016c; Hageman et al., 2016; Pereira et al., 2016).

Presented in this protocol is a detailed workflow for sample preparation of sizeable tissue specimens (up to 10 cm in diameter, Figure 2) of human and large mammalian origin. The method has been tested in human samples from the femoral neck and head of subjects undergoing hip replacement (all human samples collected and used in development of this protocol complied with all relevant ethical regulations of the Institutional Review Board of the Cleveland Clinic, IRB #12-335) as well from sheep (Ovis aries) femur and guinea pig (Cavia porcellus) knee to verify multimodal imaging (all animal tissue samples used in development of this protocol derived from studies that adhered to protocols approved through the respective Institutional Animal Care and Use Committees [IACUC] of the Canton of Grisons, Switzerland and Case Western Reserve University, in adherence with the Animal Research: Reporting of In Vivo Experiments [ARRIVE] guidelines). Here we adapted methods typically used for TEM and SEM tissue sample preparation to enable across length scales, from nano- to mesoscopic. The methods integrate techniques including precision CNC-milling and chemical etching methods developed for atomic force microscopy to enable three dimensional and multimodal imaging. This method can be applied to macroscopic human tissues and tissues of large mammals and can be integrated into a multimodal imaging workflow, in particular correlative light electron microscopy (CLEM).


Figure 2. Workflow for acquisition of multiSEM data using mesoscopic samples

Comparison with other methods
The key advantages of this method for biological specimen preparation are that it is streamlined yet maintains the flexibility to be applied to any tissue specimen using readily available materials (typically used in atomic force microscopy, EM and laser confocal microscopy specimen preparation). However, this flexibility and ease of implementation has its drawbacks, as the method is not entirely self-contained and the sample must be moved from platform to platform, if light microscopy is used in addition to EM. Advantageously, the use of heavy metal stains (uranyl acetate and lead citrate) are eliminated, as carbon sputtering provides sufficient contrast enhancement in the workflow. This further simplifies the process, avoiding the use of hazardous chemicals and their associated risks. Similarly, osmium tetroxide (OsO₄) is not used in the fixation contrast as carbon sputtering alone provides sufficient contrast. This provides further advantages in context of multimodal, correlative microscopy, as OsO₄ imparts a yellow hue throughout the specimen, impairing any form of light microscopy (Figures 2-4). 

This protocol is ideal for studies involving tissue samples from human or large mammals (up to 10 cm in diameter) where the organization of cells and tissues are of primary interest. While the resultant data encodes great structural detail, it is also extremely large (> 10 TB) and requires an appropriate and robust data management system or pipeline for efficient analytics and interpretation, which is outside the scope of traditional laboratory infrastructure. Considering the magnitude of data, it is improbable that analyses can be conducted using conventional means. This presents a disadvantage as well as an opportunity, i.e., through incorporation of cutting edge approaches including deep learning and machine learning, the economic and time costs of manual methods can be eliminated, providing a path to full automation in the future (Nathanson et al., 2019).


Figure 3. Seamless multi-scale imaging of hip joint tissue, prior to etching step. A. Transmitted light image of OsO₄ stained femoral head section, with greatest penetration of contrast agent (darkest) in cartilage (white dotted square) and marrow spaces of the subchondral bone (black dotted square). B. Subchondral bone exhibits less staining than cartilage and is highly cellular. Area corresponds approximately to black dotted square (in A). C. The tidemark, delineating the interface of unmineralized (rust-brown) and mineralized (orange) cartilage is visible in the upper right corner. Defects in the uncalcified cartilage are visible as light "cracks" running orthogonal to the tidemark line. Area corresponds approximately to the white dotted square (in A). Regarding suboptimal quality of images: these images are taken of a 300 μm thick, EPON^® embedded, undecalcified section without prior mounting and cover slipping, using an inverted epifluorescent microscope with 1.6x (A) and 10x (B, C) dry objectives. Specimens were not mounted, or cover slipped due to plans to image subsequently using multiSEM. After completion of multiSEM studies, microradiography and renewed optical imaging can be carried out. D. Laser scanning confocal image of specimen shows autofluorescence (light areas) of subchondral bone and marrow spaces (ellipsoids) as well as cartilage (dark), with two visible cartilage defects (bright cracks, arrows). E. Large scale, tiled multiSEM image of same area with single field of view showing (sub-) cellular resolution of bone (F) and blood (G) cells without additional etching. H. Traditional SEM studies on specimens containing soft tissue reveal intertwining muscle fiber architecture.


Figure 4. Seamless nano- to mesoscale imaging of the human hip, visualized using the Google Maps navigational platform. A. The Google Maps JavaScript API was used to visualize and create navigable maps of an imaged region. The arrowheads point to the eroded surface of the hip joint, toward areas internal to the femoral head. B. A single osteocyte on the sample surface with C. projections (osteocyte processes) into the underlying matrix. This sample was imaged using a 61- parallel-beam Zeiss MultiSEM.

The choice of poly (methyl-methacrylate) (PMMA) over epoxy, results in more resilient samples, with little damage observed after electron beam exposure. MultiSEM requires sample flatness and parallelism to the mount surface to prevent charging artifacts, which ideally requires precision CNC-milling, an unconventional technique for EM sample preparation. Similar results have been achieved with conventional polishing methods.

The protocol does not involve ultra-thin sectioning characteristic of TEM, which is often difficult and time consuming. As a result, use of the method may not perfectly preserve or enable visualization of internal cellular ultrastructure. For ultrastructure studies, TEM resolutions remain superior to SEM. Neither this protocol method nor any electron microscopy method enables live-cell imaging and images represent single points in time. Hence, the method does not capture the nuance and temporal dynamic of complex biological processes.

Novel techniques such as field emission scanning electron microscopy (FESEM) (Tapia et al., 2012) and serial block-face scanning electron microscopy (SBF-SEM) (Mikula and Denk, 2015) allow the three-dimensional visualization of and reconstruction of cellular ultrastructure, however samples are limited to a few millimeters on edge in size. Although the images acquired from this protocol are two-dimensional, they provide a birds-eye-view, at nanometer resolution, of cellular networks and populations over large areas. The method thus enables visualization of tissue architecture not readily achievable using two-dimensional EM.

Experimental Design
For each experiment, we defined the relevant research question and a testable hypothesis, relating independent variables (also referred to as outcome measures) as a function of dependent variables. Exclusion criteria were defined to reduce effects of unrelated disease states that could lead to misinterpretation of results, e.g., metabolic bone disease is excluded in a study of loss of bone density due to disuse.

Power calculations were made to determine sample size for each experimental group. Some novel studies do not have a representative dataset upon which, variance measures can be made and initial sample sizes have to be based on previous experience. Power analyses conducted for previous studies determined the need for a sample size of 5 patients per cohort. Post hoc analyses on these data indicated the robustness of study outcomes; from a translational perspective, even if more than 700 patients were tested to obtain statistical significance, differences (in periosteum derived stem cell regenerative capacity attributable to age and/or disease state) would remain small, due to the small to medium effect size (of age and disease), even if significantly different (Chang et al., 2014).

Specimen Acquisition
The critical phase research requiring human tissue samples is the acquisition of specimens (Figure 5). It is crucial to find a surgeon willing and able to collaborate in the study, acting as not only clinical investigator on the ethics protocol but ensuring the identification and screening of possible participants, alongside ensuring informed consent is obtained for each individual involved. It is crucial all specimens are de-identified, by removing all identifying information such as name, medical patient numbers, etc. by the tissue procurement and/or pathology department of the clinic prior to release of the tissues for scientific research. Age and sex of the patient are retained in sample documentation. Together with date of acquisition, specimens are separately identifiable without access to medical documentation.

In some experimental study designs, samples are given non-obvious identifiers and the key to decoding the identifiers is retained in a secure position by the clinical partner to enable later information from the patient's medical record, e.g., specific health history information, if needed. In obtaining patient consent while also striving to include diverse populations in cohorts, consideration should be made to patient groups whose beliefs may preclude sharing of tissues. It may be possible to encourage their inclusion through awareness of cultural sensitivities and inclusion of language that addresses culture-specific considerations in outreach brochures as well as informed consent documentation.


Figure 5. Procedures for acquiring and handling human tissues. Two samples from different patients, acquired on the same afternoon. Femoral heads (see photo, rounded structures divided into two halves) were sectioned along the plane of interest, e.g., coronal or sagittal plane, to allow for infiltration of fixation and embedding chemicals. A-G. Work flow chart for sample acquisition, with chemicals indicated in blue.

Sectioning and Fixation
With samples of size larger than a millimeter on edge, it is important to take extra measures to ensure samples undergo adequate fixation before proceeding. In this protocol, we are limited to diffusion fixation; perfusion of fixatives is not an option in human tissues. The femoral head and neck are sectioned, with the precision cutter, from the proximal and distal end before further sectioning of the medial and lateral portions, transversely across the bone. Following this, 1 mm slices in the plane of interest are taken with the saw microtome, slices at this thickness allow the complete infiltration of glutaraldehyde and paraformaldehyde fixatives in cacodylate buffer in a fume hood. The specimens are then washed in cacodylate buffer before sequential ethanol dehydration (30%, 50%, 75%, 95%).

Contrast Enhancement
Biological specimens for EM typically require post-fixation contrast enhancement to resolve subcellular detail. Contrast enhancing agents such as osmium tetroxide are commonly used but impart exogenous color throughout the entire sample, making it unsuitable for multimodal imaging (Figure 2). Heavy metal staining is another mode of contrast enhancement, whereby uranyl acetate and lead citrate are routinely employed. The use of heavy metals is a more labor intensive and hazardous process, which was not desirable. Additionally, uranyl acetate introduces instability into the sample, making it light and UV sensitive and prone to forming precipitates over time. Instead, we opted to highlight subcellular detail post hoc, with the use of both chemical etching (detailed below and adapted from protocols used for AFM) and carbon sputtering.

Resin Embedding
Conventionally, biological specimens are embedded in epoxy resins for EM. Initial attempts at using epoxy, EPON^®, proved unsuccessful as it is typically used for samples that are 1-2 mm on edge. Under inspection with a light microscope, the epoxy embedded specimens exhibited variations in surface topography of up to 100 μm making it unsuitable for multiSEM. The specimens exhibited incomplete and non-uniform resin polymerization, most likely caused by insufficient resin penetration in such large specimens. Additional contributing factors to poor polymerization, include insufficient time for immersion and perfusion, inadequate dilution steps and excessive resin viscosity, all resulting in poor resin infiltration into tissue pores, leaving residual ethanol and propylene oxide within the sample.

Due to the difficulties encountered with EPON^® and prior experience with PMMA, we used lower viscosity PMMA resin as the embedding medium. To ensure complete infiltration of the resin into tissue pores and uniformity of the sample surface, dehydration time was increased, and the tissues were bulk embedded under vacuum to enable gradual polymerization.

Planarity and Sample Size
The development of multiSEM enables rapid throughput of specimens that far exceed traditional dimensions, up to 100 x 100 x 30 mm. However, it requires specimens to be flat and parallel to the stage, with a surface flatness less than or equal to 500 nm per 100 μm (peak-to-peak). This minimizes edge effects and specimen damage resultant of increased electron emissions at high points in the surface topography.

Initially, specimens were manually and semi-automatically mechanically ground and polished, with both methods exhibiting varying degrees of success. Automated polishing proved too aggressive, rapidly removing the sample surface. Conversely, manual polishing was not time effective and resulted in the emergence of microscopic cracks in the sample surface, most likely due to the release of residual stresses. These cracks could later cause edge effects along the edges of the crack, interfering with SEM imaging quality. We progressed to micro- CNC milling and manual buffing with acrylic polish, the sample surface was milled with a great level of precision without the introduction of surface cracks. Specimens were precision milled in a CNC-lathe using a carbide insert with the unhoned edge against the sample surface. Following this, the surface was lightly buffed with a soft microfiber cloth, acrylic polish and ultra-pure water.

Chemical Etching and SEM Preparation
As noted above, this protocol does not use heavy metals based post-fixation contrast enhancement. Instead, chemical etching methods adapted from techniques developed for AFM were used to uncover underlying molecular detail and cellular networks (Reilly et al., 2001; Knapp et al., 2002). Dilute hydrochloric acid (HCl) and sodium hypochlorite (NaClO) enable controlled etching of the organic, collagen, and inorganic, apatite composite matrices, revealing cells and projections embedded in the extracellular matrix. This method also allows for selective etching of regions of interest by masking around areas to be etched with electrical tape (geometric shapes) or nail polish (free form regions). Following this step, the sample can be imaged with light microscopy. To increase sample conductance and contrast, the specimen requires a thin carbon coat (20 nm) before it can be imaged using SEM.