REVIEW ARTICLE Molecular Reproduction & Development 82:530–547 (2015)
Three Dimensional Reconstruction by Electron Microscopy in the Life Sciences: An introduction for Cell and Tissue Biologists KILDARE MIRANDA,1,2* WENDELL GIRARD-DIAS,1 MARCIA ATTIAS,1 WANDERLEY DE SOUZA1,2, 3 AND ISABELA RAMOS * 1
, rio de Ultraestrutura Celular Hertha Meyer, Instituto de , Biofısica Carlos Chagas Laborat, o ^, ncia e Tecnologia em Biologia Estrutural e Bioimagens Filho and Instituto Nacional de Ci, e Universidade Federal do Rio de Janeiro, Rio de Janeiro, Brazil 2 ^ncias da Vida, Instituto Nacional de Metrologia, Diretoria de Metrologia Aplicada a Cie , m, Rio de Janeiro, Brazil Qualidade e Tecnologia (INMETRO), Xer, e 3 dica Leopoldo de Meis Laboratorio de Bioquımica de Insetos, Instituto de Bioquımica Me Universidade Federal do Rio de Janeiro, Rio de Janeiro, Brazil
SUMMARY Early applications of transmission electron microscopy (TEM) in the life sciences have contributed tremendously to our current understanding at the subcellular level. Initially limited to two-dimensional representations of three-dimensional (3D) objects, this approach has revolutionized the fields of cellular and structural biologybeing instrumental for determining the fine morpho-functional characterization of most cellular structures. Electron microscopy has progressively evolved towards the development of tools that allow for the 3D characterization of different structures. This was done with the aid of a wide variety of techniques, which have become increasingly diverse and highly sophisticated. We start this review by examining the principles of 3D reconstruction of cells and tissues using classical approaches in TEM, and follow with a discussion of the modern approaches utilizing TEM as well as on new scanning electron microscopy-based techniques. 3D reconstruction techniques from serial sections and (cryo) electron-tomography are examined, and the recent applications of focused ion beam-scanning microscopes and serial-block-face techniques for the 3D reconstruction of large volumes are discussed. Alternative lowcost techniques and more accessible approaches using basic transmission or field emission scanning electron microscopes are also examined.
Mol. Reprod. Dev. 82: 530547, 2015. ß 2015 Wiley Periodicals, Inc. Received 1 July 2014; Accepted 10 December 2014
INTRODUCTION Understanding the functional organization of cells and tissues is a hallmark for understanding life. The description of cell and tissue organization and the discovery of novel
ß 2015 WILEY PERIODICALS, INC.
A number of techniques are available to generate volumetric ultrastructural models, and combination of a variety of strategies is now possible for tailoring to specific biological questions and applications.
Corresponding authors: Laboratorio de Ultraestrutura Celular Hertha Meyer Instituto de Biofısica Carlos Chagas Filho and Instituto Nacional de Ci^encia e Tecnologia em Biologia Estrutural e Bioimagens Universidade Federal do Rio de Janeiro Rio de Janeiro, Brazil. E-mail: [email protected] (K.M.); [email protected] (I.R.)
Published online 4 February 2015 in Wiley Online Library (wileyonlinelibrary.com). DOI 10.1002/mrd.22455
Abbreviations: 2/3D, two/three dimensional; ART, algebraic reconstruction techniques; ATUM, automated tape-collection ultramicrotomy; CCD, chargecoupled device; FIB, focused ion beam; SBF, serial block face; SEM, scanning electron microscope; SIRT, simultaneous iterative reconstruction technique; TEM, transmission electron microscope; WBP, weighted back-projection
THREE DIMENSIONAL RECONSTRUCTION BY ELECTRON MICROSCOPY
structures at the micro- and nanoscale occurred in parallel with the development of several microscopy techniques. Most of these techniques initially provided information on the two-dimensional (2D) organization of the object. Development of techniques that added a third dimension, however, allowed for a more complete assessment of biological structure and revolutionized the fields of cell and tissue biology. Like light microscopy, electron microscopy has taken a similar path of maturation, and continues to evolve through technology that provides subcellular access to the third and fourth dimensions (Lorenz and Zewail, 2014). Advances in three-dimensional (3D) electron microscopy have attracted those biologists who want to explore the volumetric architecture of their models at high resolution. A variety of new approaches are available, yet all options are grounded with basic knowledge generated in the early days of electron microscopy, which provided the foundation for the development of what is today considered cutting-edge methodology. Of course, the question of which method is best suited to a model emerges from the array of possibilities. The answer is sometimes not straightforward, and a combination of methods may eventually be chosen to address a single question. The introduction of the electron microscope as a research tool in biology, therefore, has provided a complete reassessment of the microanatomy of biological tissues, its cells, organelles, and molecules. One of the main challanges faced by the life-sciences research field in the early days of electron microscopy was the sensitivity of biological samples to the electron beam, which usually operates at high-acceleration voltages and high energies. Enormous improvements for biological electron microscopy came with advances in sample-preparation methods that provided stability and specific contrast to different samples according to the type of study required (e.g., routine methods, enzyme cytochemistry, immunogold labeling, and autoradiography). Different sample-preparation techniques were then developed, thus establishing a new research field that continues to evolve (Bozzolla and Russel, 1992). Before going into detail on the sample preparation and advances in analysis techniques, we must first briefly discuss the fundamentals of single-plane transmission electron microscopy, which is performed using the most basic instruments.
BASIC PRINCIPLES OF TEM IMAGING The transmission electron microscope (TEM) is a projection instrument. It operates using an accelerated electron beam that passes through a series of magnetic lenses (condensers, objectives, and projectors), interacts with the sample, and projects an image to a reactive surface (e.g., a fluorescence screen, film, or digital camera). The image formed is the result of an interaction between the electron beam and a sample (usually a thin section). A number of phenomena are involved in this interaction in particular, the electrons can be differentially scattered in a predictable manner that depends on the structure of the sample (re-
Mol. Reprod. Dev. 82:530–547 (2015)
viewed in Bozzolla and Russel, 1992). From a reductionist perspective, regions of the sample that contain heavy elements (high atomic number) or have higher mass density will tend to scatter the incident electrons more, which will then be retained or blocked by microscope components (apertures), resulting in darker regions on the projected image.In the case of biological samples, this is achieved by using a number of staining agents during sample preparation (osmium tetroxide, uranile acetate, lead citrate, e.g.). In addition, except for some special techniques where high voltage and/or energy-filtered TEMs are used to image whole cells (Miranda et al., 2000, 2010), for most applications, ultrathin (below 100nm) samples have to be used (Bozzolla and Russel, 1992). In the case of room temperature analysis of suspensions of particles, filaments, isolated organelles and small cells (proteins, viruses, cytoskeleton elements, etc), negative staining is usually the method of choice (Brenner and Horne, 1959). For most cell types, ultrathin sectioning (Porter and Blum, 1953) of resin embedded samples is used (Glauert et al., 1956; Maale and Birch-Andersen, 1956). The implications of sectioning the samples and how it limits the amount of information that can be obtained from an originally 3D object will be discussed in the next sections.
TRANSMISSION ELECTRON MICROSCOPY-BASED APPROACHES FOR 3D RECONSTRUCTION The Classic Approach: 3D Reconstruction From Serial Sections Obtaining volumetric information by electron microscopy has been a challenge since the first demonstrated biological applications of this technique. The TEM has been the primary tool used for the ultrastructural analysis of different cell types. A combination of techniques has been developed that can provide additional information, including 3D reconstruction (Crowther et al., 1970; Hoppe and Grill, 1977; Frank, 2006). Examination of most cell types by a TEM usually requires a number of sample preparation steps that involve chemical fixation, embedding, and ultra-thin sectioning. Low penetration of the electron beam and aberrations (that increase with sample thickness) limits samples to a thickness of 60100 nm, which severely restricts the sampling volume. Planar (2D) images of 3D structures that can extend for several micrometers in all directions are obtained, thus presenting a small fraction of the whole sample volume while neglecting information on the 3D structural organization of the specimen (Frank, 2006; Chklovskii et al., 2010). A number of techniques to circumvent the limitation of image acquisition are available (summarized in Fig. 1), with some requiring the use of sophisticated equipment such as medium- and high-voltage microscopes or dual-beam scanning electron microscopes (SEMs). Such specialized equipment is not yet implemented in most laboratories or microscopy centers, so the application of more-traditional approaches using a regular TEM has become one solution that can result in valuable 3D information when combined
531
Molecular Reproduction & Development
MIRANDA ET AL.
Figure 1. Scheme showing different approaches to 3D electron microscopy image acquisition. A: Serial-section transmission electron microscopy. B: Electron tomography. C: Serial-section scanning electron microscopy. D: SBF scanning electron microscopy. E: Serial FIB-milling. In all cases, the steps of 3D image acquisition involve the acquisition of a misaligned image series (tilt-series in the case of electron tomography); image stacking (3D reconstruction); and alignment of the series, segmentation, and modeling. In electron tomography, it is necessary to perform the reconstruction of the tilt series after alignment, which can be achieved by different approaches such as WBP, SIRT, and ART. SEM methods (CE) are grouped by their final resolution (intermediate), where serial-section scanning electron microscopy (C) and SBF scanning electron microscopy (D) have roughly the same zeta resolution, whereas serial FIB-milling (E) produces slightly better zeta resolution.
532
Mol. Reprod. Dev. 82:530–547 (2015)
THREE DIMENSIONAL RECONSTRUCTION BY ELECTRON MICROSCOPY
with methods, such as serial sectioning (Williams and Kallman, 1955; Ward et al., 1975; Coombs et al., 1986; Attias et al., 1996; Shoop et al., 2002; Miranda et al., 2004). 3D reconstruction from serial sections is defined by the acquisition and reassembly of images from different profiles of the same cell in sequential sections (Figs 1 and 2). A sample structure can be followed along the depth of the stacked sections, and the subsequent application of image processing, reconstruction, alignment segmentation, and modeling steps help generate a 3D model (Figs. 3 and 4). Volumetric, 3D reconstruction from serial sections using a TEM is considered a low-resolution technique since the projection of structures distributed along the sections depth to a 2D plane results in a fuzzy image that contains mixed information of overlapping structures that can corrupt the quality of the individual images in a series. The zeta resolution (resolution in the z axis) of a model is, therefore, limited to the section thickness (McEwen and Marko, 1999). TEM analysis of serial sections also requires the acquisition of ribbons of consecutive sections that are obtained using an ultramicrotome and collected on specially coated electron microscope grids that contain a central slot (Fig. 2). The lenght of this slot limits the total volume that can be analyzed to the sum of the thicknesses of each section contained in the grid (typically 30), unless consecutive ribbons are collected in different grids (see below) (White et al., 1986; Bumbarger et al., 2013). For these reasons, TEMbased 3D reconstruction from serial sections is generally applied to what may be considered intermediate volumes when compared to the other techniques discussed in this review (see below). The shape of the block surface (face) used in routine ultramicrotomy is not ideal for the serial sectioning required in 3D electron microscpy. In normal ultra-thin sectioning for
2D analysis, the block narrows towards the plane parallel to where the sectioning knife enters the block, resulting in the rapid increase in cross-sectional area within a few consecutive ribbon sections; this increased area could ultimately cover the whole extension of the grid, thereby restricting number of sections per ribbon per grid and severely hampering the analyzable volume (Fahrenbach, 1984). To overcome this limitation, a wider face with a shorter height than usual, with the side faces and the basal angles less than 608 (508408), should be sculpted from the sample block, which allows more sections to be placed in one grid since the section area would slowly increase with the progression of ultramicrotomy (Fahrenbach, 1984). An alternative method for preparing the block face using a glass knife and an ultramicrotome can be used, resulting in up to 100 sections in a slit of 2 mm in length (Elliot, 2007). A few groups have gone through tremendous effort to collect an impressive number of consecutive sections (up to 3,000) in different slot grids, a process that is generally timeconsuming (White et al., 1986; Bumbarger et al., 2013). Once a ribbon of sections is obtained, simple and careful observation of profiles of the same structure in serial sections may, in many cases, lead to the understanding of its 3D structure. When querying long series of sections for complex structures, however, it is not easy to envision the 3D volume by simply observing micrographs; instead, it becomes mandatory to generate a 3D model that can be rotated in all directions so that a correct volumetric representation can be obtained. In early work, such models were generated by sculpting the structures found in each micrograph using a variety of materials such as clay, Styrofoam, acrylic, or any transparent material, followed by their assembly into a physical, 3D model (Gaunt, 1978; Fisher et al., 1979; Stephens, 1979; Kinnamon and Westfal,
Figure 2. Example of a serial TEM acquisition. A: Slot grid containing a ribbon of serial sections. BJ: A sequence of images acquired from serial sections of differentiating monocytes. Scale bars, 0.5 mm (A) and 5 mm (BJ). Adapted with permission from Miranda et al., 2011.
Mol. Reprod. Dev. 82:530–547 (2015)
533
Molecular Reproduction & Development
MIRANDA ET AL.
Figure 3. Manual alignment of serial images. A: The merged overlap of two serial images with different color codes (green and pink), which shows misaligned edges. B: Merged outcome of two aligned images, where the sum of the color codes provides a less-colored image. Scale bar, 50 mm.
1981). Although simple by design, implementation of these methods was extremely laborious and time-consuming, and the models were often limited. Nevertheless, important information was obtained using this approach, such as the concept of the single mitochondrion in some cell types (Paulin, 1977), chloroplasts and mitochondria in Euglenoids (Pellegrini, 1980), and the mitotic apparatus of Trypanosomatids (Solari, 1983). With the introduction of digital imaging (initially using scanned film negatives, followed by image acquisition with a charge-coupled device [CCD] camera), software for 3D reconstruction and modeling was developed such that volumetric modeling from digital images obtained from serial sections became an accessible technique for any electron-microscopy laboratory. Some of the best reconstruction software currently available is free (see below). 3D reconstruction and generation of digital models follows a routine pipeline of image acquisition and processing steps (summarized in Fig. 1) that comprise: (1) selection of the structure of interest in different sections of the series; (2) image acquisition of the structure of interest in the whole series; (3) alignment of the images; (4) segmentation; and (5) generation of models. The resulting model can then be rotated in different directions, structures can be shown or hidden for a better presentation of the results, and movies showing different angles of the structures of interest can be generated. As each image is obtained from sections that are physically separated, small angle differences as well as small deformations (namely shrinking) due to beam damage (bombardment with the electron beam) frequently lead to misalignments that can be corrected or minimized with the use of specific alignment tools available in the software. Some of these alignment tools may be applied automatically (e.g., cross-correlation) or manually by comparing two sequential images in a series (Fig. 3). Once an aligned sequence of images is obtained, a segmentation step that generates contours from the image is required to assign a pixel identity to the profiles of the structures of interest in each section. This step consists of outlining structures of the sample by manually segmenting structures of interest in each 2D image that contributes to the volume. Contours
534
representing the same type of structure (plasma membrane, nuclear envelope, and etcetera) are then interconnected in a process of volume rendering that gives birth to the 3D model (Fig. 4). When the contrast of the structures of interest is sufficiently different from all other structures in the volume, segmentation can be done automatically by assigning intensity values to groups of pixels according to certain threshold limits defined by the user (see below). Such automatic segmentation is the fastest method of generating 3D models, although they often do not work well in reconstructions of biological samples due to the heterogeneous distribution of intensities in the sample.
Electron Tomography As mentioned above, 3D reconstruction from serial sections remains an essential tool for many laboratories, despite its limitations when compared to other 3D reconstruction approaches. Some of these drawbacks include limited sample volume due to the maximum number of sections that can be collected in a grid, thickness variations during the ultramicrotomy process, distortions caused by the uneven heating of sections by the electron beam, and low zeta resolution due to the thickness of the sections (Frank, 1995). The development of medium- (200400 kV) and high-voltage (10003000 kV) electron microscopes allowed for the observation of specimens up to 50 times thicker (0.22 mm) than the sections observed in conventional TEMs (Wilson et al., 1992; Martone et al., 2000). Although remarkably rich in the number and diversity of structures that can be observed in the thicker sections, the resulting images are difficult to interpret since the frequency of structure overlap is much greater, causing the single, projected image to be fuzzier (Bouwer et al., 2004). Visualization of 3D structures of large volumes in high-voltage TEMs was nonetheless carried out with the use of stereopairs, a combination of two images of the same object taken at small angle differences (usually 58158) (Ris, 1981; Clermont et al., 1995). High-resolution 3D analysis of cells and subcellular structures remained challenging until the development of electron tomography (Klug and De Rosier, 1966; De Rosier
Mol. Reprod. Dev. 82:530–547 (2015)
THREE DIMENSIONAL RECONSTRUCTION BY ELECTRON MICROSCOPY
Figure 4. Steps required to generate a 3D model from serial images acquired by serial-section transmission electron microscopy of a differentiating
monocyte. A: TEM image of a cell. B and C: Manual segmentation of the nucleus in different sections. DF: 3D model of the nucleus (N) shown in panel A. Scale bar, 5 mm.
and Klug, 1968), a highly sophisticated approach that some may consider an ellaborated extension of the stereopair approach employed in high-voltage TEM analysis. Electron tomography consists of acquiring a tilt series of projected images in the electron microscope (see below), followed by a number of image processing and digital reconstruction steps that generate a 3D volume. As the volume generated is composed of consecutive digital sections, it is possible to extract high-resolution information from individual planes (e.g., 1-nm digital sections) (Koster et al., 1997; Lucic et al., 2005; Frank, 2006; Barcena and Koster, 2009). The volumetric model can then be virtually sectioned, wherein the information above and below the sample is excluded to exclusively view the selected plane, resulting in very sharp images of the structures without the inference of structures contained in adjacent image planes. For this reason, electron tomography is considered a high-resolution 3D reconstruction technique. The principles of electron tomography are similar to those applied in computed tomography for medical diagnostics. Essentially a 3D body composed of regions of variable density is traversed by a radiation (X-rays, light, or electrons) tilted at different angles, from which the intensity before and after hitting the sample at these comparative angles can be measured. Differences between
Mol. Reprod. Dev. 82:530–547 (2015)
intensities result from absorption of the radiation by the material within the sample. The theoretical principles of tomography were established by Radon (18871956), a German mathematician who developed the ‘‘Radon transform’’. Haunsfield Cormack applied Radons concepts to develop computed tomography (or ‘‘CT’’), for which he was awarded the Nobel Prize for Medicine in 1979. Since it is possible to accurately calculate the relative intensity difference of the radiation that passes through an object, one can reconstruct the interior of a sample by irradiating it at different angles and collecting data from each angle. The mathematical concept proposed for tomography is the reconstruction of a 3D volume from 2D projections of an object at various angles. In electron tomography, semi-thin sections (0.22 mm) of a selected sample are used. Once the cell or region of interest is selected, the specimen is rotated around its yaxis, at defined incremental angles through an angular range (typically 708þ708). Images are acquired, usually via CCD cameras connected directly to the microscope, where the series will be recorded for later tomographic alignment and 3D reconstruction in specific software. Semi-thin sections, thicker than those used in regular 2D TEM, are used so that a larger volume of the sample can be analyzed and reconstructed, although the
535
Molecular Reproduction & Development
voltage of the electron microscope ultimately dictates the maximum section thickness that can be used. For instance, for section thicknesses of 200500 nm up to 1 mm, it is possible to use scanning TEMs (see below) with an intermediate voltage (200300 kV). This is due to the fact that as the tilt angle of the sample is increased during the tomogram acquisition, its thickness increases: a 200-nm thick sample at 708 tilt angle becomes 2.924 times thicker (585 nm). As higher voltages elicit better penetration of the electron beam in the sample, using more powerful TEMs will reduce some optical aberrations associated with excess inelastic interactions that come from the thicker specimens, thus providing an image with less chromatic aberration and, consequently, higher resolution and contrast. On the other hand, as the thickness increases, a sample will suffer more damage by irradiation. Special specimen holders with a conical shape have to be used for electron tomography since the use of regular holders may block the electron beam when the specimen is tilted at high angles ( 708). Likewise, if the spacing between the grid bars is too small (
Three Dimensional Reconstruction by Electron Microscopy in the Life Sciences: An introduction for Cell and Tissue Biologists KILDARE MIRANDA,1,2* WENDELL GIRARD-DIAS,1 MARCIA ATTIAS,1 WANDERLEY DE SOUZA1,2, 3 AND ISABELA RAMOS * 1
, rio de Ultraestrutura Celular Hertha Meyer, Instituto de , Biofısica Carlos Chagas Laborat, o ^, ncia e Tecnologia em Biologia Estrutural e Bioimagens Filho and Instituto Nacional de Ci, e Universidade Federal do Rio de Janeiro, Rio de Janeiro, Brazil 2 ^ncias da Vida, Instituto Nacional de Metrologia, Diretoria de Metrologia Aplicada a Cie , m, Rio de Janeiro, Brazil Qualidade e Tecnologia (INMETRO), Xer, e 3 dica Leopoldo de Meis Laboratorio de Bioquımica de Insetos, Instituto de Bioquımica Me Universidade Federal do Rio de Janeiro, Rio de Janeiro, Brazil
SUMMARY Early applications of transmission electron microscopy (TEM) in the life sciences have contributed tremendously to our current understanding at the subcellular level. Initially limited to two-dimensional representations of three-dimensional (3D) objects, this approach has revolutionized the fields of cellular and structural biologybeing instrumental for determining the fine morpho-functional characterization of most cellular structures. Electron microscopy has progressively evolved towards the development of tools that allow for the 3D characterization of different structures. This was done with the aid of a wide variety of techniques, which have become increasingly diverse and highly sophisticated. We start this review by examining the principles of 3D reconstruction of cells and tissues using classical approaches in TEM, and follow with a discussion of the modern approaches utilizing TEM as well as on new scanning electron microscopy-based techniques. 3D reconstruction techniques from serial sections and (cryo) electron-tomography are examined, and the recent applications of focused ion beam-scanning microscopes and serial-block-face techniques for the 3D reconstruction of large volumes are discussed. Alternative lowcost techniques and more accessible approaches using basic transmission or field emission scanning electron microscopes are also examined.
Mol. Reprod. Dev. 82: 530547, 2015. ß 2015 Wiley Periodicals, Inc. Received 1 July 2014; Accepted 10 December 2014
INTRODUCTION Understanding the functional organization of cells and tissues is a hallmark for understanding life. The description of cell and tissue organization and the discovery of novel
ß 2015 WILEY PERIODICALS, INC.
A number of techniques are available to generate volumetric ultrastructural models, and combination of a variety of strategies is now possible for tailoring to specific biological questions and applications.
Corresponding authors: Laboratorio de Ultraestrutura Celular Hertha Meyer Instituto de Biofısica Carlos Chagas Filho and Instituto Nacional de Ci^encia e Tecnologia em Biologia Estrutural e Bioimagens Universidade Federal do Rio de Janeiro Rio de Janeiro, Brazil. E-mail: [email protected] (K.M.); [email protected] (I.R.)
Published online 4 February 2015 in Wiley Online Library (wileyonlinelibrary.com). DOI 10.1002/mrd.22455
Abbreviations: 2/3D, two/three dimensional; ART, algebraic reconstruction techniques; ATUM, automated tape-collection ultramicrotomy; CCD, chargecoupled device; FIB, focused ion beam; SBF, serial block face; SEM, scanning electron microscope; SIRT, simultaneous iterative reconstruction technique; TEM, transmission electron microscope; WBP, weighted back-projection
THREE DIMENSIONAL RECONSTRUCTION BY ELECTRON MICROSCOPY
structures at the micro- and nanoscale occurred in parallel with the development of several microscopy techniques. Most of these techniques initially provided information on the two-dimensional (2D) organization of the object. Development of techniques that added a third dimension, however, allowed for a more complete assessment of biological structure and revolutionized the fields of cell and tissue biology. Like light microscopy, electron microscopy has taken a similar path of maturation, and continues to evolve through technology that provides subcellular access to the third and fourth dimensions (Lorenz and Zewail, 2014). Advances in three-dimensional (3D) electron microscopy have attracted those biologists who want to explore the volumetric architecture of their models at high resolution. A variety of new approaches are available, yet all options are grounded with basic knowledge generated in the early days of electron microscopy, which provided the foundation for the development of what is today considered cutting-edge methodology. Of course, the question of which method is best suited to a model emerges from the array of possibilities. The answer is sometimes not straightforward, and a combination of methods may eventually be chosen to address a single question. The introduction of the electron microscope as a research tool in biology, therefore, has provided a complete reassessment of the microanatomy of biological tissues, its cells, organelles, and molecules. One of the main challanges faced by the life-sciences research field in the early days of electron microscopy was the sensitivity of biological samples to the electron beam, which usually operates at high-acceleration voltages and high energies. Enormous improvements for biological electron microscopy came with advances in sample-preparation methods that provided stability and specific contrast to different samples according to the type of study required (e.g., routine methods, enzyme cytochemistry, immunogold labeling, and autoradiography). Different sample-preparation techniques were then developed, thus establishing a new research field that continues to evolve (Bozzolla and Russel, 1992). Before going into detail on the sample preparation and advances in analysis techniques, we must first briefly discuss the fundamentals of single-plane transmission electron microscopy, which is performed using the most basic instruments.
BASIC PRINCIPLES OF TEM IMAGING The transmission electron microscope (TEM) is a projection instrument. It operates using an accelerated electron beam that passes through a series of magnetic lenses (condensers, objectives, and projectors), interacts with the sample, and projects an image to a reactive surface (e.g., a fluorescence screen, film, or digital camera). The image formed is the result of an interaction between the electron beam and a sample (usually a thin section). A number of phenomena are involved in this interaction in particular, the electrons can be differentially scattered in a predictable manner that depends on the structure of the sample (re-
Mol. Reprod. Dev. 82:530–547 (2015)
viewed in Bozzolla and Russel, 1992). From a reductionist perspective, regions of the sample that contain heavy elements (high atomic number) or have higher mass density will tend to scatter the incident electrons more, which will then be retained or blocked by microscope components (apertures), resulting in darker regions on the projected image.In the case of biological samples, this is achieved by using a number of staining agents during sample preparation (osmium tetroxide, uranile acetate, lead citrate, e.g.). In addition, except for some special techniques where high voltage and/or energy-filtered TEMs are used to image whole cells (Miranda et al., 2000, 2010), for most applications, ultrathin (below 100nm) samples have to be used (Bozzolla and Russel, 1992). In the case of room temperature analysis of suspensions of particles, filaments, isolated organelles and small cells (proteins, viruses, cytoskeleton elements, etc), negative staining is usually the method of choice (Brenner and Horne, 1959). For most cell types, ultrathin sectioning (Porter and Blum, 1953) of resin embedded samples is used (Glauert et al., 1956; Maale and Birch-Andersen, 1956). The implications of sectioning the samples and how it limits the amount of information that can be obtained from an originally 3D object will be discussed in the next sections.
TRANSMISSION ELECTRON MICROSCOPY-BASED APPROACHES FOR 3D RECONSTRUCTION The Classic Approach: 3D Reconstruction From Serial Sections Obtaining volumetric information by electron microscopy has been a challenge since the first demonstrated biological applications of this technique. The TEM has been the primary tool used for the ultrastructural analysis of different cell types. A combination of techniques has been developed that can provide additional information, including 3D reconstruction (Crowther et al., 1970; Hoppe and Grill, 1977; Frank, 2006). Examination of most cell types by a TEM usually requires a number of sample preparation steps that involve chemical fixation, embedding, and ultra-thin sectioning. Low penetration of the electron beam and aberrations (that increase with sample thickness) limits samples to a thickness of 60100 nm, which severely restricts the sampling volume. Planar (2D) images of 3D structures that can extend for several micrometers in all directions are obtained, thus presenting a small fraction of the whole sample volume while neglecting information on the 3D structural organization of the specimen (Frank, 2006; Chklovskii et al., 2010). A number of techniques to circumvent the limitation of image acquisition are available (summarized in Fig. 1), with some requiring the use of sophisticated equipment such as medium- and high-voltage microscopes or dual-beam scanning electron microscopes (SEMs). Such specialized equipment is not yet implemented in most laboratories or microscopy centers, so the application of more-traditional approaches using a regular TEM has become one solution that can result in valuable 3D information when combined
531
Molecular Reproduction & Development
MIRANDA ET AL.
Figure 1. Scheme showing different approaches to 3D electron microscopy image acquisition. A: Serial-section transmission electron microscopy. B: Electron tomography. C: Serial-section scanning electron microscopy. D: SBF scanning electron microscopy. E: Serial FIB-milling. In all cases, the steps of 3D image acquisition involve the acquisition of a misaligned image series (tilt-series in the case of electron tomography); image stacking (3D reconstruction); and alignment of the series, segmentation, and modeling. In electron tomography, it is necessary to perform the reconstruction of the tilt series after alignment, which can be achieved by different approaches such as WBP, SIRT, and ART. SEM methods (CE) are grouped by their final resolution (intermediate), where serial-section scanning electron microscopy (C) and SBF scanning electron microscopy (D) have roughly the same zeta resolution, whereas serial FIB-milling (E) produces slightly better zeta resolution.
532
Mol. Reprod. Dev. 82:530–547 (2015)
THREE DIMENSIONAL RECONSTRUCTION BY ELECTRON MICROSCOPY
with methods, such as serial sectioning (Williams and Kallman, 1955; Ward et al., 1975; Coombs et al., 1986; Attias et al., 1996; Shoop et al., 2002; Miranda et al., 2004). 3D reconstruction from serial sections is defined by the acquisition and reassembly of images from different profiles of the same cell in sequential sections (Figs 1 and 2). A sample structure can be followed along the depth of the stacked sections, and the subsequent application of image processing, reconstruction, alignment segmentation, and modeling steps help generate a 3D model (Figs. 3 and 4). Volumetric, 3D reconstruction from serial sections using a TEM is considered a low-resolution technique since the projection of structures distributed along the sections depth to a 2D plane results in a fuzzy image that contains mixed information of overlapping structures that can corrupt the quality of the individual images in a series. The zeta resolution (resolution in the z axis) of a model is, therefore, limited to the section thickness (McEwen and Marko, 1999). TEM analysis of serial sections also requires the acquisition of ribbons of consecutive sections that are obtained using an ultramicrotome and collected on specially coated electron microscope grids that contain a central slot (Fig. 2). The lenght of this slot limits the total volume that can be analyzed to the sum of the thicknesses of each section contained in the grid (typically 30), unless consecutive ribbons are collected in different grids (see below) (White et al., 1986; Bumbarger et al., 2013). For these reasons, TEMbased 3D reconstruction from serial sections is generally applied to what may be considered intermediate volumes when compared to the other techniques discussed in this review (see below). The shape of the block surface (face) used in routine ultramicrotomy is not ideal for the serial sectioning required in 3D electron microscpy. In normal ultra-thin sectioning for
2D analysis, the block narrows towards the plane parallel to where the sectioning knife enters the block, resulting in the rapid increase in cross-sectional area within a few consecutive ribbon sections; this increased area could ultimately cover the whole extension of the grid, thereby restricting number of sections per ribbon per grid and severely hampering the analyzable volume (Fahrenbach, 1984). To overcome this limitation, a wider face with a shorter height than usual, with the side faces and the basal angles less than 608 (508408), should be sculpted from the sample block, which allows more sections to be placed in one grid since the section area would slowly increase with the progression of ultramicrotomy (Fahrenbach, 1984). An alternative method for preparing the block face using a glass knife and an ultramicrotome can be used, resulting in up to 100 sections in a slit of 2 mm in length (Elliot, 2007). A few groups have gone through tremendous effort to collect an impressive number of consecutive sections (up to 3,000) in different slot grids, a process that is generally timeconsuming (White et al., 1986; Bumbarger et al., 2013). Once a ribbon of sections is obtained, simple and careful observation of profiles of the same structure in serial sections may, in many cases, lead to the understanding of its 3D structure. When querying long series of sections for complex structures, however, it is not easy to envision the 3D volume by simply observing micrographs; instead, it becomes mandatory to generate a 3D model that can be rotated in all directions so that a correct volumetric representation can be obtained. In early work, such models were generated by sculpting the structures found in each micrograph using a variety of materials such as clay, Styrofoam, acrylic, or any transparent material, followed by their assembly into a physical, 3D model (Gaunt, 1978; Fisher et al., 1979; Stephens, 1979; Kinnamon and Westfal,
Figure 2. Example of a serial TEM acquisition. A: Slot grid containing a ribbon of serial sections. BJ: A sequence of images acquired from serial sections of differentiating monocytes. Scale bars, 0.5 mm (A) and 5 mm (BJ). Adapted with permission from Miranda et al., 2011.
Mol. Reprod. Dev. 82:530–547 (2015)
533
Molecular Reproduction & Development
MIRANDA ET AL.
Figure 3. Manual alignment of serial images. A: The merged overlap of two serial images with different color codes (green and pink), which shows misaligned edges. B: Merged outcome of two aligned images, where the sum of the color codes provides a less-colored image. Scale bar, 50 mm.
1981). Although simple by design, implementation of these methods was extremely laborious and time-consuming, and the models were often limited. Nevertheless, important information was obtained using this approach, such as the concept of the single mitochondrion in some cell types (Paulin, 1977), chloroplasts and mitochondria in Euglenoids (Pellegrini, 1980), and the mitotic apparatus of Trypanosomatids (Solari, 1983). With the introduction of digital imaging (initially using scanned film negatives, followed by image acquisition with a charge-coupled device [CCD] camera), software for 3D reconstruction and modeling was developed such that volumetric modeling from digital images obtained from serial sections became an accessible technique for any electron-microscopy laboratory. Some of the best reconstruction software currently available is free (see below). 3D reconstruction and generation of digital models follows a routine pipeline of image acquisition and processing steps (summarized in Fig. 1) that comprise: (1) selection of the structure of interest in different sections of the series; (2) image acquisition of the structure of interest in the whole series; (3) alignment of the images; (4) segmentation; and (5) generation of models. The resulting model can then be rotated in different directions, structures can be shown or hidden for a better presentation of the results, and movies showing different angles of the structures of interest can be generated. As each image is obtained from sections that are physically separated, small angle differences as well as small deformations (namely shrinking) due to beam damage (bombardment with the electron beam) frequently lead to misalignments that can be corrected or minimized with the use of specific alignment tools available in the software. Some of these alignment tools may be applied automatically (e.g., cross-correlation) or manually by comparing two sequential images in a series (Fig. 3). Once an aligned sequence of images is obtained, a segmentation step that generates contours from the image is required to assign a pixel identity to the profiles of the structures of interest in each section. This step consists of outlining structures of the sample by manually segmenting structures of interest in each 2D image that contributes to the volume. Contours
534
representing the same type of structure (plasma membrane, nuclear envelope, and etcetera) are then interconnected in a process of volume rendering that gives birth to the 3D model (Fig. 4). When the contrast of the structures of interest is sufficiently different from all other structures in the volume, segmentation can be done automatically by assigning intensity values to groups of pixels according to certain threshold limits defined by the user (see below). Such automatic segmentation is the fastest method of generating 3D models, although they often do not work well in reconstructions of biological samples due to the heterogeneous distribution of intensities in the sample.
Electron Tomography As mentioned above, 3D reconstruction from serial sections remains an essential tool for many laboratories, despite its limitations when compared to other 3D reconstruction approaches. Some of these drawbacks include limited sample volume due to the maximum number of sections that can be collected in a grid, thickness variations during the ultramicrotomy process, distortions caused by the uneven heating of sections by the electron beam, and low zeta resolution due to the thickness of the sections (Frank, 1995). The development of medium- (200400 kV) and high-voltage (10003000 kV) electron microscopes allowed for the observation of specimens up to 50 times thicker (0.22 mm) than the sections observed in conventional TEMs (Wilson et al., 1992; Martone et al., 2000). Although remarkably rich in the number and diversity of structures that can be observed in the thicker sections, the resulting images are difficult to interpret since the frequency of structure overlap is much greater, causing the single, projected image to be fuzzier (Bouwer et al., 2004). Visualization of 3D structures of large volumes in high-voltage TEMs was nonetheless carried out with the use of stereopairs, a combination of two images of the same object taken at small angle differences (usually 58158) (Ris, 1981; Clermont et al., 1995). High-resolution 3D analysis of cells and subcellular structures remained challenging until the development of electron tomography (Klug and De Rosier, 1966; De Rosier
Mol. Reprod. Dev. 82:530–547 (2015)
THREE DIMENSIONAL RECONSTRUCTION BY ELECTRON MICROSCOPY
Figure 4. Steps required to generate a 3D model from serial images acquired by serial-section transmission electron microscopy of a differentiating
monocyte. A: TEM image of a cell. B and C: Manual segmentation of the nucleus in different sections. DF: 3D model of the nucleus (N) shown in panel A. Scale bar, 5 mm.
and Klug, 1968), a highly sophisticated approach that some may consider an ellaborated extension of the stereopair approach employed in high-voltage TEM analysis. Electron tomography consists of acquiring a tilt series of projected images in the electron microscope (see below), followed by a number of image processing and digital reconstruction steps that generate a 3D volume. As the volume generated is composed of consecutive digital sections, it is possible to extract high-resolution information from individual planes (e.g., 1-nm digital sections) (Koster et al., 1997; Lucic et al., 2005; Frank, 2006; Barcena and Koster, 2009). The volumetric model can then be virtually sectioned, wherein the information above and below the sample is excluded to exclusively view the selected plane, resulting in very sharp images of the structures without the inference of structures contained in adjacent image planes. For this reason, electron tomography is considered a high-resolution 3D reconstruction technique. The principles of electron tomography are similar to those applied in computed tomography for medical diagnostics. Essentially a 3D body composed of regions of variable density is traversed by a radiation (X-rays, light, or electrons) tilted at different angles, from which the intensity before and after hitting the sample at these comparative angles can be measured. Differences between
Mol. Reprod. Dev. 82:530–547 (2015)
intensities result from absorption of the radiation by the material within the sample. The theoretical principles of tomography were established by Radon (18871956), a German mathematician who developed the ‘‘Radon transform’’. Haunsfield Cormack applied Radons concepts to develop computed tomography (or ‘‘CT’’), for which he was awarded the Nobel Prize for Medicine in 1979. Since it is possible to accurately calculate the relative intensity difference of the radiation that passes through an object, one can reconstruct the interior of a sample by irradiating it at different angles and collecting data from each angle. The mathematical concept proposed for tomography is the reconstruction of a 3D volume from 2D projections of an object at various angles. In electron tomography, semi-thin sections (0.22 mm) of a selected sample are used. Once the cell or region of interest is selected, the specimen is rotated around its yaxis, at defined incremental angles through an angular range (typically 708þ708). Images are acquired, usually via CCD cameras connected directly to the microscope, where the series will be recorded for later tomographic alignment and 3D reconstruction in specific software. Semi-thin sections, thicker than those used in regular 2D TEM, are used so that a larger volume of the sample can be analyzed and reconstructed, although the
535
Molecular Reproduction & Development
voltage of the electron microscope ultimately dictates the maximum section thickness that can be used. For instance, for section thicknesses of 200500 nm up to 1 mm, it is possible to use scanning TEMs (see below) with an intermediate voltage (200300 kV). This is due to the fact that as the tilt angle of the sample is increased during the tomogram acquisition, its thickness increases: a 200-nm thick sample at 708 tilt angle becomes 2.924 times thicker (585 nm). As higher voltages elicit better penetration of the electron beam in the sample, using more powerful TEMs will reduce some optical aberrations associated with excess inelastic interactions that come from the thicker specimens, thus providing an image with less chromatic aberration and, consequently, higher resolution and contrast. On the other hand, as the thickness increases, a sample will suffer more damage by irradiation. Special specimen holders with a conical shape have to be used for electron tomography since the use of regular holders may block the electron beam when the specimen is tilted at high angles ( 708). Likewise, if the spacing between the grid bars is too small (
