Serial Block Face-Scanning Electron Microscopy as a Burgeoning Technology

Abstract

Serial block face scanning electron microscopy (SBF-SEM), also referred to as serial block-face electron microscopy, is an advanced ultrastructural imaging technique that enables three-dimensional visualization that provides larger x- and y-axis ranges than other volumetric EM techniques. While SEM is first introduced in the 1930s, SBF-SEM is developed as a novel method to resolve the 3D architecture of neuronal networks across large volumes with nanometer resolution by Denk and Horstmann in 2004. Here, the authors provide an accessible overview of the advantages and challenges associated with SBF-SEM. Beyond this, the applications of SBF-SEM in biochemical domains as well as potential future clinical applications are briefly reviewed. Finally, the alternative forms of artificial intelligence-based segmentation which may contribute to devising a feasible workflow involving SBF-SEM, are also considered.

1. Introduction

Serial block face scanning electron microscopy (SBF-SEM), also referred to as serial block-face electron microscopy, is an advanced ultrastructural imaging technique that enables three-dimensional visualization that provides high x- and y-axis ranges.^[1] Techniques for higher volume are typically done through automating the collection of serial sections, such as automated ultramicrotome SEM (ATUM-SEM)^[2] or serial section transmission electron microscopy (TEM).^[3] While SEM was first introduced in the 1930s and the first microtome was established Lin an SEM system in the 1980s by Leighton,^[4] SBF-SEM began more standard as a novel method to resolve 3D architecture of neuronal networks across large volumes with nanometer resolution by Denk and Horstmann in 2004.^[5] SBF-SEM alternates scanning of the block face surface of a resin-embedded biological sample by a focused beam of electrons with the precise excision of the block face at set intervals by a diamond knife until serial sections of the compete region of interest (ROI) have been captured (Figure 1).^[1,5] Specifically, backscattered electrons (BSEs) are more specific for heavy metals in the sample, with electrons from the SBF-SEM beam (Figure 1) interacting with electrons in the sample on the basis of the material, generating a composition-dependent backscattering.^[5] BSEs are collected by a detector which allows for micrographs. These z-directional micrographs can be digitally aligned and reconstructed into a 3D volume through an array of software, allowing for revolutionary ways to visualize and analyze complex biological structures, from subcellular organelles to whole tissues, including spatial organization and interactions of cellular components. Quantifications can further be performed through a range of software including ImageJ, Reconstruct, Amira, or Dragonfly.^[1,6,7] Machine learning capabilities and integration with complementary techniques, such as correlative light and electron microscopy (CLEM), will expedite SBF-SEM workflows and enable researchers to combine spatial information obtained from SBF-SEM with functional and molecular data from fluorescence microscopy.^[8] Below we highlight several advantages and disadvantages for volume electron microscopy or specifically SBF-SEM.

Figure 1.

Project workflow of serial blockface-scanning electron microscopy with a representative image of 3D reconstruction and quantification of mitochondria.

2. Advantages

1. High resolution and volume: SBF-SEM provides high-resolution, 3D images of large-volume samples, enabling researchers to study the morphology and organization of cells and tissues with extreme detail.^[1] SBF-SEM is much faster at imaging large volumes than focused ion-beam (FIB-SEM) scanning electron microscopy, samples of up to 10⁷ μm³ can be surveyed with SBF-SEM.^[9] In comparison, while FIB-SEM is evolving, volumes are typically under 10⁵ μm^3[10,11] and processing speeds are 20 times slower.^[12] Additionally, different regions can be analyzed by scanning multiple regions of interest.^[13] This allows for organelles that are scattered across large cell regions to still be imaged and their connectivity to be considered. For SBF-SEM this is particularly useful for studying complex structures such as neural networks or organs.

2. Automation: SBF-SEM can be combined with automated acquisition^[14] and analysis software, which enables large-scale imaging projects. Automated software can assist with image acquisition, registration, and alignment, enabling high-throughput imaging projects with minimal user intervention, and offering robust data for quantitative morphometric analysis.

3. Inherent alignment: SBF-SEM is inherently aligned so less imaging processing is required for proper alignment than other 3D techniques and there is less chance of potential error incurred by incorrect orthoslice alignment.^[14]

4. Broadly applicable: SBF-SEM provides details of all cellular components on the serial section^[15] and can be applied to a wide range of models, from cultured cells to bones, making it an ideal technique to address a wide array of biological questions.^[6,13]

5. Combinable: Can be combined with other techniques such as CLEM for fluorescence labeling to further consider protein spatial orientation relative to the organelle organization.^[8]

6. Use of TEM samples: Samples previously prepared for transmission electron microscopy can be screened for their quality of staining and re-milled for SBF-SEM.^[15]

3. Challenges

1. Expertise: One of the main limitations of the technique is that it requires specialized equipment and expertise to operate. The proper tissue preparation for SBF-SEM, affected by factors such as the embedding quality and staining protocols, is also time-consuming and can be challenging. A heavy atom stain is primarily for BSEs contrast,^[13] while poorly defined or inconsistent imaging parameters can reduce rigor. Therefore, care must be taken to use validated methods (Figure 1 shows examples of imaging quantifications that may be utilized).^[6]

2. Destructive: SBF-SEM is typically a destructive technique, as the sample is physically sliced during imaging, preventing future analysis or reimaging at higher resolution.^[5]

3. Smaller: SBF-SEM requires samples to fit within the max stage range, which typically requires a less than 500 μm blockface.^[5] Depending on the size of the sample, the sample may need to be sectioned into smaller pieces, which can be time-consuming and technically challenging.

4. Staining-Limited: Samples must go through a harsh process of staining and resin embedding which may cause artifacts in soft tissues,^[16] while well-contrasted staining can be difficult to optimize, thus limiting the applicability of SBF-SEM.^[17]

5. Surface Charging: SBF-SEM generates sequential images, and variations in image quality can occur due to factors such as negative surface charging, bad sectioning, and contamination of the sample during fixation or imaging. These artifacts can affect the quality of the resulting 3D reconstruction and the results of the study, if systematic analysis techniques are not devised.^[14] Charging can be reduced by variable-pressure SEM to create a low-vacuum environment which also allows for control of humidity around the sample ;^[18] however, procedures need to take into account the electrons scattering affected by the gas, known as electron skirts, which can affect the resolution.^[19] Focal charge compensation also alleviates charging issues through a constant stream of nitrogen which neutralizes the charge.^[17]

6. Time-consuming: SBF-SEM generates large amounts of data, and image processing and analysis can be time-consuming and computationally intensive. Due to low contrast with certain organelles, automated image processing and analysis algorithms may not be available, and manual processing can introduce human bias into the analysis.^[1]

7. Static: Imaging dynamic processes remains a challenge due to the static nature of the technique. These can be alleviated by combining with techniques such as CLEM, but these increase the expertise and equipment requirements for SBF-SEM.^[8]

8. Low Z-resolution: Given the sensitivity of SBF-SEM samples to electron dose, for high image quality, cutting thickness needs to be increased (typically 50 nm).^[6,20] The microtome also dictates a minimum of 25 nm slice thickness.^[9] This results in lower z-axis spatial resolution compared to x- and y-axis resolution and to other techniques such as focused ionbeam scanning electron microscopy, which uses ion beams and can have thicknesses reported as low as 5 nm.^[9,21] This makes SBF-SEM unfit to survey certain organelles including cristae of mitochondria, nuclear pores, and endoplasmic reticulum (ER) cisternae. However, certain techniques such as the Monte Carlo simulation, which utilizes multiple primary beam energies, can slightly increase the effective z-resolution.^[9]

4. Applications of SBF-SEM

Given that ultrastructural changes may be concomitant with many clinical diseases,^[22] SBF-SEM has been increasingly utilized in various biomedical applications for understanding ultrastructural changes in pathophysiology (Figure 2). SBF-SEM has been instrumental in studying the ultrastructure of the choroid plexus in diseases such as Alzheimer’s.^[23] SBF-SEM offers unique applications to pathologies such as dentine hypersensitivity by offering the ability to quantify occluded tubules and their depth of penetration across different experimental treatments to better elucidate potential mechanisms for the treatment of bone-like materials.^[24] SBF-SEM can be employed to visualize surroundings, such as blood vessels, immune cells, and the extracellular matrix.^[6] This has been utilized to study tumor microenvironments and stroma in pancreatic ductal carcinoma liver biopsies.^[25] SBF-SEM has been used to study glomerular diseases, such as focal segmental glomerulosclerosis and membranous nephropathy including the ultrastructure of the glomerular basement membrane area and volume.^[26] While past studies on host-cell interactions in viruses are limited by the thickness of the sample,^[27] SBF-SEM alienates these issues and, especially when paired with light microscopy, allows for greater visualization of pathogens.^[28]

Figure 2.

Examples of potential applications of SBF-SEM 3D reconstruction and quantification.

Models from a range of organisms and in vitro alternatives including C. Elegans,^[29] Drosophila,^[30] and mice^[1] demonstrate the broad applicability of SBF-SEM. The applicability of SBF-SEM may be even further increased when combined with CLEM, as fluorescently-labeled structures can be used to guide the imaging of large-scale samples such as mouse mammary gland organoids.^[31] Currently, for disease states, there is a greater emphasis on imaging larger samples, such as entire nervous systems or connectomes, which has been bolstered by advances in multiple scanning beams and optimization of preparation.^[11] This information may be to help elucidate disease mechanisms, identify novel therapeutic targets, and facilitate early diagnosis, as well as inform the development of targeted therapies and provide insights into drug resistance mechanisms. In contrast, cell biology has increasingly focused on specific organelle structures, which have necessitated the development of reliable methods of organelle quantification.^[11]

5. Differing Segmentation Techniques

Quantification of organelles using SBF-SEM data, or other volumetric techniques, involves extracting relevant structural and morphological information from the 3D volume reconstructions. Several computational techniques and tools have been developed for this purpose, each with its strengths and limitations.

5.1. Manual Segmentation

The most commonly used method is manual segmentation and quantification. This approach involves manually tracing the organelles of interest in the electron micrographs, then using software tools to compute the organelle volume, surface area, and spatial distribution. For example, if utilizing manual segmentation, typically a user would go through and identify each mitochondrion, going through z-stacks segmenting the organelles. Although manual segmentation is considered the gold standard for accuracy, it is labor-intensive, time-consuming, and subject to human error and bias.^[1,32] While the exact additional length of time for manual segmentation depends on the skill of the operator, the density of the object being subjected, the size of the sample, and the quantity of z-slices to be analyzed,^[33] this will generally take far longer than other segmentation methods. One potential alternative to decrease the time of manual segmentation is the employment of stereology.^[34] Stereology is a method for estimating the volume, surface area, or number of organelles in a 3D tissue from 2D sections, by employing a systematic, random sampling approach, it provides unbiased estimates of these quantities without requiring a complete 3D reconstruction.^[6] Given this expedited workflow, stereology can be less labor-intensive and more efficient than full 3D reconstructions, but given one of the principal use cases of SBF-SEM is assessing complex structures and spatial distributions, stereology remains nonapplicable in many cases. Another alternative is an abstracted model generation which allows for minimal resources 3D reconstruction in comparison to manual segmentation.^[33]

5.2. Semi-automated Segmentation

Semiautomated involves the user providing some input (e.g., seed points or region of interest), with the software completing the segmentation based on user-defined parameters.^[35] For example, if utilizing semiaugmented, typically a user would go through an allow a machine learning program to identify mitochondria and segments. However, from there all mitochondria would be reviewed structures may need to be retraced. Alternatively, a program may allow the user to select specifically the organelles of interest and tracing will be performed from there. Generally, semiautomated segmentation techniques are found to greatly reduce the time from manual segmentation techniques, while large-scale quantitative analysis may only fall within several percentage points as manual segmentation.^[36]

5.3. Fully-automated Segmentation

An increasingly, utilized approach is fully automated segmentation In contrast, machine learning, particularly deep learning techniques such as convolutional neural networks, has been increasingly used for increased efficiency of imaging^[37] and diagnostic purposes.^[38] However, developments in SBF-SEM remain limited. One prominent example is Python-based Human-In-the-LOop Workflows, developed by Suga and colleagues, which allow for deep-learning of mitochondrial and cristae networks for their rapid analysis.^[39] Fully automated segmentation can significantly reduce analysis time and minimize human error, but the accuracy of the segmentation depends on the quality of the training data and the performance of the algorithm. For example, ideally, in a deep learning method, a user can simply input the data set into the algorithm and adjust parameters and verify the output of completed 3D reconstructions of selected organelles. CDeep3M, a model for the segmentation of nuclei, membranes, and mitochondria, was found to have a similar accuracy to a human expert, of around 98, after 50 000 instances of training and validation.^[40] This highlights the applicability of these for clinical diagnosis, as many machine-learning models utilized for cancers, kidney disease, and other pathologies typically have an accuracy of around 90–99%.^[38] Once validated, studies have found that the proofreading time required to verify machine-learning approaches is quite low.^[11,39,40] Yet, many studies using fully-automated methods are employing these approaches for a limited set of organelles not yet expanding to the full spectrum of biopsies and pathophysiology. However, the diverse phenotypes of mitochondria, or other organelles, limit the ability of SBF-SEM machine learning. Specifically, for low-contrast images such as those obtained from SBF-SEM,^[6] data is ideally very homogenous for proper analysis by automated segmentation.^[33] For example, machine learning models may need to be retrained depending on the pathology or type of tissue, and there may be limitations in the ability of machine learning models to render abnormal structures.^[33]

In summary, the quantification of organelles in SBF-SEM data involves a trade-off between accuracy, efficiency, and user control. Manual segmentation is the most accurate but also the most time-consuming, while fully automated segmentation can be highly efficient but may have limited accuracy depending on the algorithm’s performance. Underscored is the fact that new mechanisms to segment organelles are necessary. CLEM is increasingly viable as an alternative for SBF-SEM usage by itself (Figure 3). For example, immunogold labeling alongside SBF-SEM allows for augmented identification of organelles and segmentation.^[41] While this technique is powerful in allowing for the identification and locations of 3D reconstructions of immunolabeled antigens, limitations exist including artifacts caused by antibodies binding to the face of resin sections.^[42] Future systems may investigate employing machine learning to have mechanisms that use fluorescence from CLEM to guide neural networks.

Figure 3.

Correlative light and electron microscopy (CLEM) workflow as an alternative to bolster the applicability and information of serial block-face scanning electron microscopy (SBF-SEM).

Once segmented, SBF-SEM organelles can easily be quantified in a variety of ways. The authors have previously used quantifications through manual segmentation and quantification to measure various parameters of mitochondria including area, surface area, volume, perimeter, complexity index, sphericity index, and complexity index (Figure 1).^[1] Mitochondria-specific phenotypes can be observed and include aspects such as length and diameter.^[43] In neurites, the distribution and phenotypes of specific neuron networks can be analyzed in 3D.^[44] Axon distance and myelin thickness can further be measured, alongside other parameters.^[45] Porosity of cell membranes,^[46] and fractal dimension and porosity of material pores have also been quantified using SBF-SEM.^[47] Quantifications go far beyond this, and novel quantifications are further constantly being developed.

6. Clinical Diagnosis and Future Outlook

Given SBF-SEM’s application in a greater understanding of organelle volume, connectivity of neurons, and details of cellular ultrastructure, we believe SBF-SEM has a multitude of implications for clinical diagnosis. The usage of 3D medical imaging, such as in whole-body magnetic resonance imaging or computerized tomography scans are established,^[48] however SBF-SEM usage in clinical settings remains limited. SBF-SEM can be used to study the ultrastructure of neurons in the brain,^[49] which is important for understanding how the brain works and how it is affected by disease.^[44] For example, it may be able to be applied to identify dysregulated disease states. A key hallmark of disease states is mitochondrial dysfunction due to ultrastructural changes,^[50] including in Alzheimer’s Disease where mitochondrial show altered structure.^[51] In the future, it is possible this technique could be used in pathological departments to study the connectivity of different regions of the brain and to map the synaptic connections between neurons, especially in cancer cases or other biopsies which may necessitate a greater understanding of the pathophysiology.

Alongside this, it is possible that in the future SBF-SEM will better be implemented within hospitals to improve diagnosis. It can be used to study the structure of cancerous tissue or to investigate the changes that occur in the brain during neurode-generative diseases such as Alzheimer’s disease.^[52] This may potentially allow for different ways of studying disease states than past mechanisms. For example, the hallmarks of kidney disease are currently investigated by pathologists after biopsies through a multitude of mechanisms. Histochemical stains measure the glomerular structure and look for hallmarks of glomerular inflammation, such as hematuria.^[53] Additionally, cell proliferation and other gross morphology aspects are also typically examined by pathologists, while immunohistochemistry can detect the presence of specific proteins.^[53] Using SBF-SEM, glomerular basement membrane biopsies have been investigated.^[54,55] Past results found evidence of cytoplasmic penetration and glomerular basement membrane disruptions, which are a preceptor for future hematuria, were able to be observed with SBF-SEM 3D reconstruction but not TEM.^[54] Further, SBF-SEM can serve as a mechanism to provide greater details on the spatial relationships with extracellular glomerular components, with novel protrusions observed only through SBF-SEM.^[56] Given that glomerular basement membrane structures’ morphological changes are frequently observed in kidney failure,^[26,56] utilizing SBF-SEM for clinical diagnosis may allow for the earlier detection of inflammatory glomerular diseases.

Importantly, SBF-SEM may also be used to study the effects of new drugs or therapies on tissues at the cellular level. For example, bone growth and pathologies of bone can be considered through a much more involved lens than previously allowed by traditional pathology techniques.^[57] Thus, SBF-SEM can be used to study tissues in greater detail than is possible with other imaging techniques to help doctors develop more personalized treatment plans for patients and potentially identify dysregulated disease states earlier on. Notably, Oregon Health and Sciences University (through the Serial Measurements of Molecular and Architectural Responses to Therapy program) is currently beginning to utilize high-throughput methods of SBF-SEM to evaluate tumor 3D dynamics as a clinical diagnostic tool.^[52] Importantly, this research has demonstrated that SBF-SEM can be utilized in addition to traditional pathological techniques for cancer such as protein analyses and cyclic immunofluorescence,^[58] to glean more information from tumor microenvironments.^[52]

We believe the implementation of SBF-SEM in hospitals and pathology departments is a promising potential avenue; however, several challenges must be overcome. SBF-SEM is a relatively specialized technique, educating pathologists and clinicians on its principles, routine applications, and interpretation is essential to coincide with implementing SBF-SEM facilities within hospitals. In its current form, SBF-SEM is limited in time-consuming and labor-intensive processes, involving fixation, dehydration, resin embedding, and staining.^[6,21,25] Additionally, biopsy preservation for SBF-SEM is typically best performed with high-pressure freezing, which is often not available in clinics.^[52] Therefore, we believe, developing faster, more automated, higher volume, and standardized protocols would enable a quicker turnaround time and broader use in a clinical setting. Advances in electron detectors, automation of data acquisition, and machine-learning-based segmentation of organelles could significantly increase the imaging speed, making it more suitable for real-time diagnostics.^[5,13,21] Already, developments in automated sampling technology have allowed for the analysis of breast cancer to be reduced to under 24 hours.^[59]

Importantly, while this article focused principally on SBF-SEM as an isolated technology, SBF-SEM is also powerful in other applications such as in CLEM setups (Figure 3). Combining SBF-SEM with other imaging techniques, such as fluorescence microscopy, can provide complementary information for a more comprehensive diagnosis.^[8,17,60] New imaging modalities technologies are constantly evolving and offer a multitude of potential throughputs which may increase the applicability of SBF-SEM.^[8,17,60] Developing robust correlative workflows that integrate data from multiple imaging modalities may enhance the clinical utility of SBF-SEM.^[8] By addressing the challenges of sample preparation, imaging speed, data analysis, and integration with other diagnostic modalities, SBF-SEM has the potential to become a more actively utilized tool in pathology and real-time diagnosis in hospital settings.