The Expansion of Volume Electron Microscopy

ABOVE: This image depicts the fruit fly nerve cord connectome. It highlights 930 neurons, a subset of the full set of reconstructed neurons. flyEM/Janelia Research Campus

With the inventions of transmission electron microscopy (TEM) in 1931 and scanning electron microscopy (SEM) shortly after in 1937, scientists gained an unprecedented ultrastructural view of the previously inaccessible subcellular world. No longer limited to the physics of light, researchers can now visualize samples at nearly 0.1nm resolution. However, this high-resolution perspective and the hypotheses it inspires are still bound to two dimensions.

A high-resolution 2D electron microscopy image of a cell might show the nucleus surfacing in the cytoplasmic sea or a cross section of the nucleus decorated with a few polyribosomes. However, a 3D volume of the cell reveals the whole nucleus draped in pearl-like necklaces of ribosomes. “There are a lot of weirdnesses that are there in 3D where if you take some random slice, you just wouldn’t be aware of it,” said Harald Hess, a microscopist at Howard Hughes Medical Institute.

There are a lot of weirdnesses that are there in 3D where if you take some random slice, you just wouldn’t be aware of it.

 —Harald Hess, HHMI Janelia Research Campus

To delve deeper into these biological complexities, scientists added a third dimension to both TEM and SEM, leading to volume electron microscopy (vEM).¹ “The community is defining volume electron microscopy as any method that allows you to collect serial images through a micron or more of resin-embedded cells and tissues at the ultrastructural level,” said Lucy Collinson, an electron microscopist at The Francis Crick Institute.

The early bird catches the worm

By the 1980s, microscopists had produced collections of a few hundred images comprised of serial sections of tissue. However, early efforts to visually capture the entire brain of the worm gave new impetus to technological innovation. In 1986, Sydney Brenner and his colleagues at the University of Cambridge published the first connectome of Caenorhabditis elegans.² To accomplish this monumental task, the team embedded a worm in resin, like a fly in amber, and shaved off incredibly thin sections—50nm to be exact—using a diamond knife ultramicrotome, an instrument akin to deli slicer that removes ultrathin slices from a sample. The researchers placed the slices on grids and imaged serial sections using transmission electron microscopy. Brenner and his team reconstructed the 302 neurons and more than 5,000 synaptic connections of the brain of the worm from more than 10,000 serial sections. 

It took Brenner and his team more than a decade to create the worm connectome, which they communicated in a 340-page paper.² Since then, a series of advancements in vEM technologies have revolutionized the field, facilitating faster imaging of larger tissues.³ Recently, researchers published the complete connectomes of the 23,000 neurons of the adult fruit fly nerve cord and 3,016 neurons of the fruit fly larva brain.^4,5

         

Getting faster, any way you cut it

Over the last two decades, quietly in the background, key advancements in sample preparation, imaging technologies, and scientific computing have gradually revolutionized the field and streamlined vEM. As a result, vEM is increasingly finding its way into the biosciences.

vEM modalities are based in either TEM or SEM.¹ For TEM, scientists place ultrathin sections on a film that allows electrons to pass through onto a downstream detector. In SEM, a focused electron beam scans and interacts with atoms in the sample, which generates backscattered electrons or secondary electrons. These electron signals carry information about the structure that scientists use to generate an image. 

The choice of imaging method largely depends on the resolution or field of view required for the project. In general, TEM modalities offer higher resolutions, whereas SEM modalities accommodate larger samples.¹ For example, serial section electron tomography, a TEM-based modality, achieves the highest resolution of 1-5nm³, but is limited to a volume of 1µm³. In contrast, focused ion beam SEM (FIB-SEM) can image volumes close to 100 µm³ but with less depth resolution of around 5-10nm³.

     

Brenner and his team endured years of manual slicing, slow imaging, and painstaking, hand-drawn reconstructions of every serial section. Now, scientists have developed innovative modalities for automation and parallel imaging to generate serial sections.³ In TEM, microscopists acquire images either by feeding them through the imaging column, as with GridTape TEM, or by tilting the sample in increments and reconstructing them with back-projection, as with serial section electron tomography. Similarly, the SEM based technique array tomography images ultrathin serial sections placed on coated substrates in a 2D array. 

Manually slicing hundreds and thousands of ultrathin sections is a grueling process that is vulnerable to human error. To address this, researchers have developed several automated methods, including GridTape TEM and an automated tape-collecting microtome (ATUM) for SEM.^1,3 Imaging speeds are also on the up and up. To image close to 100x the speed of a conventional SEM, microscopists equipped the system with 61 electron beams scanning in parallel.⁶ (Single-beam SEM is standard.) Although it comes with a big price tag, a handful of facilities interested in large-scale connectomics are trying multiple microscopes in parallel.⁷

Automated slicing and multibeam imaging are big throughput advancements in vEM. While it took researchers over a year to slice and image the drosophila larva brain, now, “they can be imaged with a multi beam machine and literally be done in less than a month,” Hess said.

Another major development over the last decade is the rise of serial block face (SBF) techniques and in situ slicing mechanisms in SEM. In contrast to TEM modalities, SBF-SEM uses in-chamber microtome sectioning to serially remove ultrathin sections from the resin-embedded sample before a scanning electron beam images the surface.² Driven by a need to slice even thinner sections, another block-face technique emerged: FIB-SEM.^1,7 

“The problem with diamond knife sectioning is that you end up with a certain minimum thickness that you can cut reliably,” said Hess. That thickness is around 40-50nm, but researchers working on connectomics were after even thinner sections because any loss in resolution complicates reconstruction. Alongside an electron beam, FIB-SEM microscopes shoot high energy ions like gallium at the surface of the sample. Because ions are heavier than electrons, the ion beam sandblasts away very thin layers of material, giving researchers the ability to go through the sample nanometers at a time.

The high resolution of FIB-SEM comes at a cost. For larger samples like fly brains, slow imaging speeds coupled with short run times disrupt imaging and muddle efforts for sample continuity. To address this, Hess and his team developed enhanced FIB-SEM and re-engineered the microscope to deal with malfunctions that arise during imaging to maintain continuity in the face of staccato imaging. Meanwhile, other researchers are innovating on slicing modalities, replacing gallium ion sandblasters with oxygen plasma beams due to the element’s superior compatibility with resin.⁸ The enhanced FIB-SEM system is just another example of technological advancements since Brenner’s efforts to capture the 1mm C. elegans, making vEM more user friendly and capable of handling larger samples.

Bigger brains, bigger data

If a single strand of hair is 80,000nm thick, the level of resolution achieved by modern microscopes imaging five nanometers at a time is incredible. This impressive feat of resolution ushers in new challenges, namely data wrangling and storage. “We are always playing catch up because the microscopes are getting faster and more automated,” said Collinson. To manage larger volumes, scientists across academia and industry are teaming up to share vEM data and to develop sophisticated computational and data storage solutions.^1,3 

The next decade promises faster imaging with further automation and parallelization. “This will push vEM towards quantitative science, which is where it needs to be to be able to compare a lot of samples in health and disease,” said Collinson.