It has been said that the human brain is the most complex thing in the universe. To study it is not unlike doing a thousand-piece puzzle of clear-blue sky with your eyes closed.
As I go about studying neuroscience I’d like to share some of the things I learn. I hope that you will find this useful. Please note that none of this is intended to be used as medical advice.
The First 2000 Years
A very long time ago, sometime in 460 B.C. a Greek physician named Hippocrates suggested that the brain is the seat of intelligence.
And men ought to know that from nothing else but from the brain come joys, delights, laughter and sports, and sorrows, griefs, despondency, and lamentations.
That guy was smart.
Around the same time an equally famous ancient Greek philosopher Aristotle didn’t quite share this point of view. He believed that the heart is the seat of our thoughts and that the brain is a mere radiator, a cooling device for the body.
History is littered with wrong assumptions.
In the 17th century Rene Descartes thought that pineal gland is what controls our mind. He would also advocate the “hydraulic view” of the brain suggesting that animal spirits in our nerves transferred information to and from the ventricles in the brain. Another famous Greek physician Galen believed that nerves were hollow and fluid or air passed through them to inflate our muscles. This became known as the balloonist theory and surprisingly this view survived for 1500 years.
We can laugh about it now. But for a very long time there was no way for us to really tell what’s going on inside the brain. Things started to change fairly recently.
In the 17th century microscopes stopped being a novelty and achieved a decent magnification of 300 (thanks to van Leeuwenhoek). Biologists started really digging into the fine details of the nervous system. Around the same time Luigi Galvani published his work on electrically stimulating frog muscles.
In the 19th century an Italian physiologist Angelo Mosso notices that the brain pulsated in patients with skull breaches and comes up with the first device to measure mental activity, calling it a ‘human circulation balance’. He would strap someone to a table delicately balanced so that a slight change in weight distribution would tip it in either direction. Mental processes would increase cerebral blood flow thus making the head heavier and the table would tip.
However most advances happened in the 20th century due to rapid advancements in physics, electronics, computing technologies and data science.
The human brain really looks like a giant wobbly pale pink walnut. A sight that sends some medical students running for the bathroom the first time they see it. It packs some 100,000,000,000 neurons within its many structures. And yet we often need to peek inside the brain. Does it have abnormalities? How is the human brain different from the dolphin brain? Does the size of the amygdala predict the level of anxiety we experience?
Thankfully there are plenty of methods helping us to see the brain anatomical structure:
- DWI / DTI
However seeing the brain structures tells us little about how the brain really works. Wouldn’t it be great to see which parts of the brain correspond to specific thoughts, feelings, sensory information or movements? Luckily we have a number of methods that show us sites of activity within the brain helping us to investigate the brain functions:
- EROS / DOT
As you will see later, even the best of these methods cannot tell us about the behaviour of individual neurons or how neurons connect to each other. Using the following methods we can see what’s happening at the level of individual neurons:
- Transsynaptic tracing
- Wallerian degeneration
- Membrane probes
- Axoplasmic transport
- Metabolic markers
So lets dive in.
Structural imaging is a way to visualise the anatomical structure of the brain. Typically the spatial resolution of structural brain imaging is between 0.5 to 5 mm³. There are close to 100,000 neurons in a cubic millimetre of the brain cortex and about a billion synapses so structural imaging provides us only with a “macro” view of the brain.
X-rays are short length high energy electromagnetic waves. They were discovered in 1895 by a German physicist Wilhelm Roentgen who noticed the silhouette of his own bones on the fluorescent screen when he placed his hand between the screen and an electron beam he was experimenting. What we see on the X-ray screen is essentially a shadow formed by objects that X-rays could not penetrate. On a typical X-ray the brightness of the area depends on the object density with denser parts being brighter.
The brain has about the same density throughout so we don’t see much of it on an X-ray. An American neurosurgeon Walter Dandy used X-ray for ventricular radiography when he injected air directly into the brain ventricles. Another relatively risky technique called cerebral angiography allowed researchers to see the brain’s blood vessels on an X-ray.
A plain X-ray can also help detect certain calcified lesions or skull fractures. Here’s a quick video explaining how X-ray radiography works. The technique is cheap, fast and easy to use. Radiation exposure from a single X-ray is usually low.
Computed Tomography (CT)
Computed Axial Tomography (CT or CAT) was invented by two scientists Godfrey Hounsfield and Allan Cormack in 1972. The idea is quite simple. A circular frame rotates around your head. On one side of the frame there’s a tube emitting X-rays and on the other there’s a curved X-ray detector measuring the attenuation of the emerging beam. In a single rotation a CT scan takes about a thousand snapshots and a computer can then put them all together to make a detailed image of the brain. Here’s a video explanation of how CT scan works.
CT angiography (when a contrast is injected into your blood) can be used to study the structure of blood vessels. CT resolution is around 2mm and is enough to see brain tissue, blood, bones and cerebrospinal fluid (CSF). Notice how the densee part (the skull) is white while fluid-filled ventricles are black.
CT scans are fast and relatively cheap however the major problem of computed tomography is the amount of radiation it uses which can increase the risk of cancer. The risk is very small so using CT is justified for medical reasons but isn’t ideal for research.
Magnetic Resonance Imaging (MRI)
Magnetic Resonance Imaging (MRI) is also known as Nuclear Magnetic Resonance (NMR). The reason it’s called MRI is due to the fact that it was introduced during the Cold War and the patients weren’t particularly thrilled to get scanned by a “nuclear” device.
Tissues in our body are jam packed with atoms of hydrogen. The hydrogen nuclei are basically protons. As we know protons are charged particles that “spin” around their axis. This spin makes them into tiny magnets. An MRI machine uses a very strong magnet to align those spinning protons along its magnetic field. Mostly they align in such a way that proton’s north magnetic pole faces MRI’s south magnetic pole. They also start precessing around the magnetic axis.
An electromagnetic pulse is then sent across your body that gives protons enough energy that they flip in the opposite direction. It also makes them precess in phase. In a second or two these protons flip back (relax) and emit radio waves. These are picked up by the MRI machine. Because different types of tissues relax at different rates we then know what type of tissue we’re dealing with at a specific location. This is called T1 sequencing. At the same time in the absence of the pulse protons start dephasing. Hydrogen nuclei in water dephase slower than in fat and therefore emits energy for a longer period of time. This is called T2 sequencing.
If this sounds like gibberish to you, then this, this and this videos explain how NMR actually works. MRI uses a number of techniques or sequences that highlight various properties of the brain. T1 sequence displays fat and central nervous tissue as bright and other tissues and fluids as dark with good contrast between white and gray (darker) matter, while T2 is the opposite with lower contrast between white and gray matter. FLAIR sequence nulls the signal from the cerebrospinal fluid making it easier to study white matter. MRI sequences and their clinical applications are discussed in this video.
MRI imaging usually takes longer than a CAT scan and it can be very noisy (personally I tend to fall asleep inside an MRI). On the plus side, it typically has a higher resolution of 0.2-0.5mm, and is safe. Well, unless you’re wearing a pacemaker.
DWI / DTI
One of the MRI techniques is called Diffusion Weighted Imaging (DWI). This technique essentially measures the rate of movement of molecules of water, i.e. diffusion. Remember that in T2 sequencing the strength of the signal depends on how fast hydrogen nuclei dephase in their precession.
Imagine we have protons precessing and aligned with the magnetic field. We send an electromagnetic pulse making them precess with the same phase. Now imagine we apply a tapered magnetic field making protons closer to it precess faster, and those further from it precess slower. We then apply an equal but opposite magnetic field for the same duration. Protons would go back to their original frequency and phase.
However that would only happen if the protons didn’t move. If they moved between the first and the second application of the gradient magnetic field, then they will no longer be in phase. DWI technique picks up this difference. No signal loss means that the protons are precessing in phase, meaning there is no or minimal water diffusion. Again if this sound too complicated watch this video and things will become a bit clearer.
Scientists noticed that on DWI images the white matter contrast changed based on the direction of axons. Turns out that in axons water movement is mostly along the axonal membrane (anisotropic). Using this property an algorithm was built that allows us to obtain beautiful colour coded images of white matter tracts. This modification was called Diffusion Tensor Imaging.
Functional imaging is the next step in the brain imaging where we not only see the anatomical structure of the brain but also changes in the brain activity over time. Again, the resolution of these methods is typically low, so we only see the activity of a very large number of neurons. Another drawback is that functional imaging tells us little about how neurons connect to each other.
Functional MRI (fMRI)
Having a good understanding of the brain structure is not enough to know what’s going on inside the brain when we think, feel, sense or act. Functional brain imaging methods are based on the concept of cerebral blood flow (or relative cerebral blood flow, rCBF).
In 1948 it was demonstrated that when neurons are active they require more oxygen (remember that mitochondria use oxygen to produce ATP). To get that oxygen they release chemical signals causing nearby blood vessels to dilate leading to an increased blood flow. Blood flow consists of oxygenated haemoglobin and deoxygenated haemoglobin.
Remember that MRI picks up the signal emitted by protons when they flip to their low energy state. Turns out that the deoxygenated haemoglobin is paramagnetic and blocks the signal emitted by protons. Active neurons increase the local blood flow washing away this deoxygenated haemoglobin therefore increasing the MRI signal. This is called the blood oxygen level-dependent (BOLD) signal.
Here’s how an fMRI machine can be used. The researcher will ask the person in the fMRI machine to think about something funny or sad, or wiggle a toe, or do something else and observe which parts of the brain “light up” in response. By observing the response you can roughly tell what the person is thinking about.
The scans can be made rapidly at intervals of about 50 ms and with a resolution of about 3 mm³. This by no means tells us what’s going on inside individual neurons but it’s enough to show, for example, that faces are recognised in one part of the brain while bodies in another.
Positron Emission Tomography (PET)
We can also see how the brain functions using PET scans. They are based on the same concept of measuring an increase in cerebral blood flow. Only this time a special type of radioactive glucose called fluorodeoxyglucose is injected into the blood stream. Active brain cells love glucose. The radionuclide breaks down and emits positrons which collide electrons and produce gamma rays. Those rays are picked up by detectors.
Compared to fMRI, a PET scan is invasive, may require fasting prior, and carries a small risk due to exposure to radiation. PET scan resolution is not great at 5-10 mm³ and the scans are usually more expensive.
Single-Photon Emission Computed Tomography (SPECT)
This is another functional imaging technique that is based on the cerebral blood flow. Again, a radioactive tracer is injected into blood stream. A gamma camera is then rotated around the head taking a number of 2D pictures with the computer calculating the 3D image. Gamma rays bounce off denser tissues so areas deeper in the body will tend to be distorted. To fix this a SPECT CT technique is used where a CT component is added to SPECT. X-ray attenuation is measured and taken into account by the computer. SPECT is cheaper, but uses a different radionuclide that emits photons directly and has an even lower resolution of about 10 mm³.
EROS / DOT
EROS stands for Event-Related Optical Signal. There’s nothing hot about sitting while wearing a helmet with a bunch of wires sticking out. However this helmet studded with tiny LED lights and detectors allowing scientists to peek a few centimetres into your brain. The idea here is that the sensors measure changes in the infrared light scattering that occurs with neural activity. It has a good spatial and temporal resolution of millimetres and milliseconds, however it’s limited as to how deep the infrared light can go into the brain tissue. Note that this method measures light scattering properties of the neurons rather than rely on the cerebral blood flow. Watch Brumback Peltz demostrating this method in action.
DOT (short for Diffuse Optical Tomography) is similar and uses near-infrared spectroscopy to build a 3D model of the brain structures. Light scattering and absorption depends on local changes in the amounts of circulating oxy- and deoxy-haemoglobin. Again, this method can only penetrate the brain tissue so much, and that limits its usefulness. Both methods are portable and inexpensive compared to SPECT, PET or fMRI.
Imaging Neural Pathways
While CT, PET and MRI allows us to see the anatomical structure of the brain and which areas of the brain are used for specific purposes, they tells us nothing about the way neurons are connected. There are roughly 100 billion neurons in the human brain and many more connections between them. Consider Purkinje cells in the cortex that can have as many as 200,000 connection per single cell. Axons reach distant parts of the brain, are curved along the way and have a very small diameter. Add to that the fact that axons often run in bundles with other, unrelated axons occupying the same space. This makes it extremely difficult to understand the organisation of neural networks inside the brain. There are a few methods currently in use that allow us to see connections between neurons.
Rabies is an especially nasty virus. If I had to pick a virus that can mutate and make everyone a zombie, this will it. By the time you notice symptoms you’re definitely going to die. Yet this virus is remarkable in that it can jump across synapses. So when researchers need to trace connections between an axon of one neuron and dendrites of another, they can immunohistochemically stain a virus (they use pseudorabies that doesn’t infect humans) and then introduce it into the cell membrane waiting for it to move along the axon and jump across synaptic clefts.
Tracing Based on Wallerian Degeneration
In 1850 Augustus Waller found that severing an axon from the cell body leads to the axon rapidly degenerating. Axonal skeleton disintegrates and the axonal membrane breaks apart.
By introducing a destructive lesion at a desired site in the animal brain scientists cause the axons connected to that region to start degenerating. After about 72 hours Marchi staining can be used on that axon which makes it stand out among others and makes it possible to trace it.
Lesions can be introduced physically however there’s always the risk of slicing through neural pathways that just happen to be passing through the same location. Instead an injection of toxins (e.g. ibotenic acid) can be used to kill neurons selectively.
Tracing Based on Membrane Probes
If you have cyanine dye and a few months to spare you could inject the dye into the cell body and wait for it to diffuse along the membrane. Axons can reach one meter in length therefore diffusion of the dye takes a very long time. The benefit of this method is that diffusion happens even in dead tissue.
Tracing Based on Axoplasmic Transport
In tiny cells proteins and organelles move around easily through the force of diffusion. However axons can be a meter long and cannot rely on diffusion to move organelles, proteins, lipids and synaptic vesicles from one end to another. Neurons solve this via axoplasmic transport. All those useful goodies are packed in so called transport vesicles and moved by kinesins along microtubules running the length of the axon. To understand what it looks like check out the swag on this little guy. This is called anterograde transport. Stuff that needs to be destroyed is shuttled back via retrograde transport back to the neuron cell’s lysosomes.
A radioactive tracer or even a tracer with a fluorescent marker is introduced at either end of the axon and after some time the tracer is distributed across the axon and can be examined under a microscope.
This method was developed by Kwanghun Chung and Karl Deisseroth at Stanford. The idea here is to make the brain tissue transparent. This is achieved by chemically removing lipids in a post-mortem tissue sample and applying acrylamide-based hydrogel to support the preexisting protein structure.
Obviously all of the neurons will be transparent as well. So to be able to tell one from another, stains and fluorescent labels are used. Once imaged, the stains can be removed and different stains applied to view a different set of cells. The resulting picture looks amazing as you can see in this video. The downside is that acrylamide used is highly toxic and staining can take weeks before the sample can be imaged.
This method was originally developed in 2007 by a team lead by Jeff W. Lichtman and Joshua R. Sanes at Harvard. Instead of manually staining a single neuron (like in the axoplasmic transport method) what if we made neurons fluoresce in various random colours. This is achieved by inserting fluorescence genes that can randomly combine to produce up to 100 colours. The brain then lights up in bright colours under a microscope with each individual neuron fluorescing in a different colour.
Mice with their 75 million brain cells were the first to be examined using this method. In order to introduce random colours to neurons we first create a transgenic mouse that carries genes encoding fluorescent proteins of different colours (they are called CFP, YFP and RFP). Offsprings of this mouse and another mouse that carries Cre-recombinase enzyme would have their fluorescent genes flipped randomly creating hundreds of combinations of different colour. This so called Cre/Lox mechanism is explained in this video.
Brainbow. A great name, there needs to be a song about it.
Trans-Tango is a brand new brain imaging method (though long time in the making) that was announced in 2017 by the group of Gilad Barnea of Brown University. It also uses anterograde transsynaptic tracing by introducing a synthetic signalling pathway into all neurons. This pathway then acts like a switch that can be flipped by exposure to a triggering protein. Researchers present this triggering protein together with the green fluorescence protein to the first neuron which then makes the next neuron to light up in red.
Metabolic Markers in Neuronal Tracing
Active neurons require a lot of energy and therefore need a constant supply of glucose and oxygen. Radioactively labelled 2-deoxyglucose enters neurons but is not metabolised and can be used to reveal active neurons. Unlike methods based on axoplasmic transport, pseudorabies or degeneration, this technique allows researchers to see which connections play the most important role in particular brain activities.
Optogenetic Pathway Tracing
Imagine that you’re trying to see how activating a particular neuron affects the activity in the nearby neuronal circuitry. There’s no easy way to selectively activate a specific neuron (electrical activation is not very precise). Unless you use optogenetics. Optogenetics is an amazing and a little scary technology that allows us to activate neurons deep in our brain with a flash of a light. Boris Zemelman in 2002 introduced rhodopsin (a light sensitive protein) in neurons and managed to activate them with light. This video explains the basics of how optogenetics works.
What if we wanted to trace every single neuron in the brain? One obvious solution would be to get a sample of the brain tissue, record what we see, slice a very thin layer off the sample, map the next layer and so on. By connecting the maps at the end of the process we can visualise the entire neuronal map (connectome). And this is what EyeWire attempts to do.
EyeWire is essentially a massive multiplayer online game to map the brain designed by Sebastian Seung’s at Princeton. Currently EyeWire is mapping retinal neurons. The activity of these neurons was mapped using two-photon microscopy. The sample was then sliced using microtome and scanned with an electron microscope.
EyeWire attempts to trace neurons using an machine learning algorithm. But the algorithm can only go so far, while the rest of the tracing is done by us, humans. Check out this I am more than my genes video by Sebastian Seung.