Who are we?

We are a group of computer and bioimaging researchers from the Rosalind Franklin Institute and neuroscientists from King's College London. We are interested in learning about the developing brain – especially the hippocampus, a crucial region for memory and learning. We want to take a closer look at the synapses, where neurons make connections with each other, and study the organelles (mitochondria and synaptic vesicles) they contain. This will help us understand better how the brain grows and functions.

What are we studying?

The hippocampus

The hippocampus, named after its resemblance to a seahorse (from the Greek words 'hippos', meaning horse, and 'kampos', meaning sea monster), is located in the temporal lobes of the brain. It plays a vital role in many cognitive processes, and its proper development is essential for healthy brain function.

In the hippocampus, like other brain regions, neurons form intricate networks through structures called synapses. Most of these connections are established during early life, laying the groundwork for the hippocampus's role in memory and learning.

Thanks to the advancement in imaging techniques and computational power, we now have a general understanding of how different parts of the hippocampus connect with each other, and how they connect with other parts of the brain. However, there are lots of details that remain unknown, especially the functions and interactions of the sub-cellular structures within synapses. To bridge this gap, we are focusing on two key components of synapses: synaptic vesicles and mitochondria.

The synapse

Neurons are specialized cells with a central body (soma) and two types of extensions: dendrites and axons. Dendrites are short, branching tubes that receive signals from other neurons, while axons are longer, single tubes that transmit signals to other neurons. At the end of an axon, there is a small swelling called a bouton, which forms a synapse with the dendrites of other neurons.

Synapses facilitate communication between neurons through a combination of electrical and chemical signals. Electrical signals travel within neurons, triggering the release of tiny chemical messengers called neurotransmitters at the synapse. These neurotransmitters are stored in small bubbles called synaptic vesicles within the bouton. When an electrical signal reaches the bouton, synaptic vesicles fuse with the neuron's membrane and release neurotransmitters, which then bind to receptors on the receiving neuron, propagating the signal.

Our tiny messengers need a lot of energy to properly do their job. This is where mitochondria, the power plants of the cell, come into play. Mitochondria produce ATP, the cell's energy currency, which is essential for many cellular processes, including the synthesis and release of neurotransmitters. Mitochondria within synapses ensure a steady energy supply to support synaptic transmission.

Synapses can look very different from one another. They vary in the number, size, and location of synaptic vesicles and mitochondria they contain, and they undergo changes as the brain develops. To understand what this means to brain function, we need to quantify and analyse the variation in synaptic vesicles and mitochondria. However, this hasn’t been easy.

How do we study the hippocampus?

To visualise these tiny structures, we used electron microscope, which can capture details much smaller than what's possible with a traditional light microscope. Specifically, we are using a technology called serial block-face scanning electron microscope (SBF-SEM). It allows us to see not only the surface of a biological sample, but the entire volume in 3D! It works like this: after acquiring an image, a very thin and sharp knife slices way a thin layer off the top of the sample, exposing the layer beneath. This new layer is then imaged, and the process is repeated. By continuously imaging and slicing, a series of 2D images is generated. These images, when stacked together in sequence, can be reconstructed to form a detailed 3D volume of the original sample.
Below is the complete process of using SBF-SEM to study biological samples.

We have acquired images of mouse hippocampus and did the pre-processing to make further analysis easier. However, there’s still one crucial step before we can visualise the synapses and understand its structure, and that is segmentation.

Why do we need your help?

To analyse synaptic vesicles and mitochondria, we need to identify and outline them in microscopic images, a process called segmentation or annotation. Because microscopic images are so detailed and cover a large area (that is, in comparison to the tiny organelles of interest!), manually segmenting them would be extremely time-consuming for a small group of scientists.

Researchers are developing automated segmentation methods, using image processing algorithms or machine learning. They have already made significant progress. However, a structure that is easy for the human eyes to detect might be confusing for the computer. This could be due to imaging artefacts that are difficult for the machine to filter out, or the various shapes that the organelles may have. As a result, machine-generated segmentations almost always require human "proofreading" to ensure accuracy and reliability.

With your help, we hope we can efficiently tackle the task of proofreading and correcting machine-generated segmentations, ensuring high-quality data for our research.

How can you help?

In this project, you will be working with high-resolution electron microscope images of the mouse hippocampus acquired by neuroscientists at KCL. These images have a resolution of 4.4 nanometres per pixel, allowing us to visualize synaptic vesicles and mitochondria.

Our team has already manually segmented the synaptic boutons in these images. Your task will be to:

1. Verify the accuracy of machine-generated segmentations of synaptic vesicles and mitochondria within these boutons.
2. Correct any errors in the machine-generated segmentations by deleting, adding or adjusting the outlines.

How was machine segmentation done?

The machine segmentation of synaptic vesicles and mitochondria in our images was performed using two different approaches:

Synaptic vesicles

Synaptic vesicles are relatively uniform in size (40-50nm in diameter) and shape (mostly spherical), and they appear as dark, circular structures in electron microscope images. These characteristics make them suitable for a technique called template matching, where a small template image of a vesicle is compared to different parts of the larger image to find matches.

However, the small size of synaptic vesicles (around 10 pixels in diameter) makes them susceptible to noise and imaging artifacts, which can confuse the template matching algorithm. Additionally, the membranes of the bouton and the mitochondria have a comparable thickness to vesicle diameter. They also appear dark in the images. As a result, sometimes the computer confuses the membrane with a string of vesicles.

Mitochondria

Mitochondria have much greater variability in size, shape, and appearance compared to synaptic vesicles, making traditional image processing methods like template matching ineffective. Instead, we used deep learning models, which are trained on large datasets of annotated images to learn the patterns and features associated with mitochondria.

Deep learning models consist of artificial neural networks that can automatically learn hierarchical representations of image features through exposure to a large number of examples. By training on diverse datasets, these models can learn to recognize mitochondria in different contexts and adapt to their variability.

We used a pre-trained model called MitoNet ¹, developed by researchers at the National Institutes of Health. This model has been trained on millions of microscopic images and can accurately segment mitochondria in many of our images.

However, even state-of-the-art deep learning models can struggle with the variability and ambiguity of biological images. As a result, human proofreading and correction remain essential steps in ensuring the quality and reliability of the segmentation results.

What is the impact of your contributions?

By participating in this project, you will be directly contributing to our understanding of the developing hippocampus and its synaptic structures. Your efforts in proofreading and correcting the machine-generated segmentations will ensure that we have accurate and reliable data for our analyses.

The insights gained from this research may shed light on the mechanisms underlying brain development, learning, and memory, and could potentially inform the development of new therapies for neurological disorders.

Come join us on this exciting journey!

 

---

References:
[1] Ryan Conrad and Narayan Kedar, Instance segmentation of mitochondria in electron microscopy images with a generalist deep learning model trained on a diverse dataset, Cell Systems 14.1 2023 58-71

Images:
Hippocampus and Synapse: Mark Bear, Barry Connors, and Michael A. Paradiso. Neuroscience: exploring the brain, enhanced edition. Jones & Bartlett Learning, 2020.
Volume electron microscopy: Peddie, C.J., Genoud, C., Kreshuk, A. et al. Nature Review Methods Primers 2, 51 (2022).