How volume EM is transforming biomedical research
However, until now, only a fraction of the biopsy can be imaged using high-power microscopes, which means that features of interest to pathologists may be missed. This is because, as with any camera, including the one on your mobile phone, the further you zoom in, the smaller the field of view. Indeed, if we were to take enough images to map a single kidney biopsy in this way, it would take around 50 years. But a team of scientists at the Francis Crick Institute, Imperial College London and the European Molecular Biology Laboratory aims to get this down to just five days.
Their technique is called ‘correlative and multimodal imaging’, which means combining different imaging methods to image a single sample across scales. As well as X-ray and light microscopy, the team are using a fast-growing technique called volume electron microscopy, also known as volume EM. This exciting new method has started to gain traction across the bioimaging field, as it allows scientists to see inside cells and tissues in 3D at extraordinary resolution. But what is volume EM, why has it taken life sciences research by storm, and what are the current challenges?
When the Crick was built nearly ten years ago, head of the Electron Microscopy team Lucy Collinson was given the chance to house the sensitive high-resolution microscopes in the quietest place in the building – which turned out to be as far as possible from London’s busy St Pancras station.
Her team spend a lot of their time underground, four stories below the Crick’s entrance, operating microscopes on specialised platforms to counteract vibrations in the surrounding environment, like the rumblings of nearby trains. A large proportion of their work now involves using the volume electron microscopes.
Lucy Collinson leads the Electron Microscopy team at the Crick.
Find out more about how electron microscopy works.
Lucy says, “To understand what volume EM is, we need to step back a bit. Most people are familiar with light microscopes from school. They use a beam of light to image the cell and they’re very powerful but limited in resolution. You can see single cells but not always the fine details of their inner workings. A lot is happening at that scale that can go wrong in disease, and to get to that level, we need to use electrons. Electrons have a wavelength smaller than that of light, which is why you can see more detail.”
Electron microscopy can largely be divided into two types: transmission electron microscopy (TEM), which involves the beam of electrons passing through the sample, and scanning electron microscopy (SEM), which involves scanning the beam across the surface and imaging the signals that bounce back.
The Crick's electron microscopes sit on specialised floors to minimise interference from vibrations in the environment.
Eyelashes being used to move slices on water. Credit: Aaron Sait.
Preparing a sample for EM is a fine art, requiring patience and dexterity. Because electron beams are scattered easily, even if they hit air molecules, the samples have to be imaged in a vacuum inside the microscope. To do this, they’re embedded in a plastic block for protection – Lucy likens failing to do this to “sending an astronaut into space without a suit”.
The samples are then sliced thinly enough that the electrons can pass through. Thin, in this case, is around 100 nanometres, about 1000 times smaller than the width of a human hair. The only knife capable of cutting slices this thinly is made of a specially sharpened diamond, mounted in an ultramicrotome, which Lucy describes as a “fancy bacon slicer”.
As the slices are cut, they float onto water and are moved using a human eyelash stuck to the end of a cocktail stick, before being mounted in the electron microscope. The preparation process is so difficult that it severely limits the number of experts in the field and the volume of the sample that can be imaged manually. “Meticulous preparation is crucial,” Lucy says. “If you put rubbish samples into the microscope, you get rubbish images out.”
So, that’s how samples are prepared for 2D electron microscopy. What about volume EM?
“The big change came when the process of slicing and imaging started to become automated,” Lucy continues. “There are different volume EM techniques. For example, in ‘block face imaging’ techniques, either the ultramicrotome is placed inside the scanning electron microscope, so that the diamond knife takes a slice and throws it away, or a focused ion beam is fired at the sample to cut an even thinner slice of material away from the surface. In both cases, the surface is imaged after each slice of material is removed. You keep going and build up a stack of images that represent the entire volume of your sample, reconstructing what it looks like in actuality. And that’s where you move into volume EM.”
Lucy explains that the difference between 2D EM and volume EM is comparable to the difference between looking at a blueprint of a room layout compared to a 3D interactive visualisation of the same room in computer software. “It allows researchers to visualise compartments inside the cell in 3D and understand how they interact,” she says. “This is very hard to appreciate from a 2D image.”
The EM team at the Crick collaborate with around 60 research groups every year on over 100 individual projects. These collaborations are usually long-lasting, taking on average two to five years. And the questions researchers are hoping to answer are hugely diverse.
“We’ve imaged anything and everything, from immune cells and neurons, to cells infected with malaria, tuberculosis, toxoplasma or influenza, to mouse tissue and whole organisms like worms, zebrafish and fruit flies. We also image human tissues from people affected by neurodegeneration, cancer and kidney disease,” says Lucy.
“At the end of the day, we’re all built of 3D cells,” says Jenny Hounsome, a postdoc in the team who collaborates on projects with scientists across the organisation. “Seeing the architecture inside the cell in 3D is not only important – it’s fascinating. A common research question is how changes in the cell’s DNA impact what’s physically happening inside. Volume EM can help to answer this.”
Every project is custom-designed and will depend on the research question, what the sample is, how big it is, and what the region of interest is.
“Researchers come in wanting to look at one specific part of the cell, but then they see these amazing images, and think ‘we also need to look at that!’ Once a question is answered, there’s a new one waiting. But that’s also credit to how fast these techniques are developing,” says Jenny.
Cell biologist Federica Mangione works with the EM team to understand how fruit flies sense touch with specialised bristle structures similar to human hair follicles. This could help us understand how we have also evolved to pick up messages through touch.
Using volume EM and rebuilding the bristle into its 3D structure, Federica recently discovered an entirely new cell type that is crucial for fruit fly sensation.
“Because I’m trying to understand how these hair follicle-like structures are constructed, and they’re so tiny, I used volume EM to reconstruct the entire sensory organ. We preserved the tissue in a resin block, physically cut this using a focused ion beam, and then imaged each surface. Post-imaging, we aligned all the image layers and modelled the cell to build it back into its 3D structure,” says Federica.
“Excitingly, we saw that bristles send signals to neighbouring epidermal cells to recruit a new cell type, which adopts a unique shape and wraps around the bristle. When flies didn’t have this cell, they became insensitive to touch. I named it F cell – the fifth cell type in the fruit fly bristle but also ‘F’ for Federica!”
Federica is now working with Jenny and others in the EM team to reconstruct the 3D bristle structure at different stages of development as well as imaging the organ in adult flies to understand which areas are more or less flexible to mechanical stimuli.
The bristle mechanosensory organ mid-way through its full development in three orientations. The F Cell is in magenta, growing hair in light blue, socket cell in green, sheath cell in orange and sensory neuron in yellow. Credit: Federica Mangione.
Candice leads the Kidney Immune Cell Imaging Satellite Laboratory at the Crick.
Now that researchers have fine-tuned this advanced technique, how can it be applied to a major medical challenge?
For every 100 people receiving a kidney transplant, between 10 and 15 will experience rejection in the first year, which is when the body recognises the new organ as ‘non-self’ and raises the alarm. Clearly working out how and why the immune system attacks the new kidney is critical to maximise transplant success.
Five years ago, Lucy met Candice Roufosse, a clinician scientist at Imperial College looking after people with kidney disease. Together they established a new pipeline aiming to use different types of imaging to better understand transplant rejection.
The project is aided by the fact that kidney disease is one of the only areas of medicine that routinely uses electron microscopy for diagnosis. Candice wants to use these images to better understand a particular type of immune rejection involving antibodies. When people are given a transplanted kidney, there are often mismatches between certain molecules on the donated kidney cells and their own cells, and these mismatches tell the immune system that a cell is ‘not self’. Patients with a transplant can then develop antibodies against the mismatched proteins, causing inflammation akin to that in infection, and ultimately breakdown of the new kidney.
Kidney tissue under the microscope. Credit: Candice Roufosse and Alana Burrell.
“The theory is that the damage starts when the antibodies congregate on the cells lining the blood vessels of the kidney, which then leads to an influx of immune cells, damaging the blood vessels,” says Candice. “We can now overlay different scales of imaging of the kidney tissue, from the whole sample using X-ray down to individual immune cells using volume EM. With this, we can snap a picture of all the immune cell types and states, where they’re found and what they are up to. Applying machine learning to the outputs will allow us to build a timeline of rejection, from healthy kidney tissue to early signs of antibody-induced injury to full-blown rejection.”
The ultimate outcome is to use understanding from the new pipeline to slot patients somewhere on this timeline. The makeup of immune cells could determine whether an individual is likely to experience rejection, and even what treatments could prevent this.
Such extensive imaging at high resolution will produce an unprecedented amount of data, so how are the team going to analyse this gigantic dataset and work out the red flags signalling a likelihood of rejection?
“One microscope can easily collect 250 gigabytes of images per day, adding up to terabytes of data generated across the many microscopes in the EM facility every day. Without automation of data analysis, it could take centuries to fully analyse that amount of data,” says Lucy.
Physicist Martin Jones joined the EM team just as volume EM was starting to come online. His role is to develop computational analysis to tackle the vast amount of data volume EM produces and work with software engineers to see if AI can make the task more manageable.
“Firstly, we’ve got to check the quality of the data and make sure all the images are aligned – which they often aren’t”, says Martin. “Then we’ve got to match a volume EM dataset and a light microscopy dataset, as usually we want to add colour from fluorescent tags in light microscopy to the grey images volume EM produces. Both of these steps are manual and time-consuming before you even get to the analysis stage.”
Martin explains that analysis of volume EM data involves a process called segmentation – delineating objects of interest, like mitochondria or the nucleus, from everything else in the image. And this is where, surprisingly, both people and AI can play a crucial role.
The Etchiverse is a collection of citizen science projects that ask volunteers to annotate images of cells and tissues, helping the EM team reduce the time needed for analysis.
Helen Spiers co-leads these projects with Martin. The first project they built together, Etch A Cell, was born out of a collaboration between the Crick team and the University of Oxford’s astrophysics department.
Helen says, “The potential of AI is limited by how well it can be trained, so human effort is still needed for lots of tasks. In the Etch A Cell projects, volunteers look for mitochondria, lipid droplets, the endoplasmic reticulum, amongst other things. Along the way, they might spot other features that could lead to new discoveries. One thing that is critical to the success of these, and other, citizen science projects is the willingness of our research teams to engage with volunteer communities who are giving their free time and enthusiasm to research.”
The most recently launched project, called Etch A Cell – ImmunoExplorers, asks volunteers to search for immune cells in transplanted kidney tissue and draw a box around them. The boxes produced by the volunteers will be fed into AI algorithms.
“Segmentation involves tracing around things, which is fine for ten slices, but not for 10,000,” says Martin. “The new model means that people can get through images much quicker, and the computer system can therefore learn a lot quicker from their inputs.”
The team hope that developments from the project can help with training future models for a whole host of challenges. The ultimate goal in Martin’s view is to get closer to ‘generalisation’ – a model that can be applied to all types of volume EM, analyse many different scenarios and characterise more than one element of a cell at once.
“The more data we collect and analyse, the easier it will be to train new models, which will hopefully start to crunch data from volume EM images quicker and quicker,” says Martin.
An image of kidney tissue that volunteers receive. The following features have been segmented in colour: an immune cell (green), blood vessels (light blue), red blood cells (red), endothelial cells (purple), mesangial cells (pink) and basement membrane (yellow). Credit: Helen Spiers and Alana Burrell.
A combination of powerful, multiscale imaging techniques and some hundred volunteers are hoping to make strides in our understanding of why kidney rejection happens and how to prevent it. But where else does Lucy see the potential for the power of volume EM?
“It’s already being used in the field of connectomics, and that’s going to lead to big discoveries,” she says. “So far, EM is the only way to visualise how nerve cells are connected at the nanoscale, and volume EM can now allow researchers to follow how these connections change over time.”
It’s been no mean feat getting here. In the 1980s, it took Sydney Brenner’s team at the Medical Research Council’s Laboratory of Molecular Biology ten years to map the nervous system of the nematode worm C. elegans using electron microscopy. And researcher Michael Winding, now leading a lab at the Crick, mapped the very first fruit fly larva brain using volume EM, a task that took over five years while he was at the University of Cambridge. Earlier this year, another Cambridge team added the adult fruit fly connectome to the portfolio too.
Lucy says, “If it’s taken us forty years to get from the worm to the fly, then the jump from the fly to the mouse will be enormous, and the mouse to the human currently impossible. But I see this like the rapid acceleration in genomics – it took the Human Genome Project ten to fifteen years, building on decades of technical development, as well as global investment and dozens of research teams, to map the first human genome. Now DNA sequencing of a human genome can take just 24 hours.”
While connectomics continues to accelerate, Lucy and the team are harnessing the power of volume imaging pipelines to bring some clarity to real-world problems like organ rejection. And building a 3D picture using volume EM is key to the challenge. As Lucy says, “it can allow us to see things we’ve never seen before.”
The connectome of the fruit fly larval brain. Credit: Science (2023).
