4 SYNAPTIC PLASTICITY

Nerve cells are specialized to process information, and to communicate with one other, using the language of electrochemistry. They produce electrical impulses that encode information and carry them along their slender fibers, relaying these signals to each other by means of chemical messengers. Synapses are the junctions between nerve cells where this signaling (neurochemical transmission) takes place, and synaptic plasticity refers to the various ways in which synapses can be modified.

Most neurons have multiple dendrites, or branches, and a single axon. The dendrites receive signals from other cells, and begin to process them locally, before passing them on to the cell body. Here, the incoming signals are summated; a response signal is then generated at the initial axon segment close to the cell body and propagated along the axon to the nerve terminal. Nervous impulses cannot cross the synapse, and so when an impulse reaches the terminal, it is converted into a chemical signal.¹

Functional Architecture of Brain Synapses

Synapses have two structural and functional components, termed the pre- and postsynaptic membranes, which send and receive chemical signals, respectively. Neurons can form synapses with non-neuronal elements, such as skeletal muscle fibers and hormone-producing glands, the so-called “effector” organs. In the brain, however, nerve cells form connections exclusively with one another, with the nerve fiber terminal of one cell coming into close apposition with the axon, dendrite, or cell body of another.

Nerve terminals are often referred to as synaptic boutons, and the postsynaptic elements of excitatory synapses are arranged within tiny protuberances called dendritic spines, whereas those of inhibitory synapses are located in specialized areas of the postsynaptic membrane, found either on the dendrite shaft itself or around the cell body.² The synaptic cleft, the minuscule gap between the bouton and spine, is just 20 to 40 nanometers (nm, or billionths of a meter) wide. But despite being so small, synapses are highly organized three-dimensional structures—the boutons and spines are highly specialized to perform their functions, and the behavior of their respective components is tightly orchestrated.

Despite being so small, synapses are highly organized three-dimensional structures—the boutons and spines are highly specialized to perform their functions, and the behavior of their respective components is tightly orchestrated.

Broadly speaking, there are two types of synapses in the brain: excitatory synapses release the neurotransmitter glutamate, which increases the probability that the postsynaptic cell will generate a nervous impulse, and inhibitory synapses, which use the transmitter gamma-aminobutyric acid (GABA), and this decreases the probability that the postsynaptic cell will fire.

In resting nerve cells, neurotransmitter molecules are stored in tiny, spherical, membrane-bound structures called synaptic vesicles, which are “docked” at the 
“active zone” just beneath the terminal membrane, awaiting the arrival of a nervous impulse. When an impulse reaches the terminal, it causes an influx of calcium ions through the presynaptic membrane, which in turn causes some of the vesicles to fuse with the membrane and release their contents into the synaptic cleft. Once released, the transmitter molecules diffuse across the cleft, and then bind to receptor proteins embedded in the postsynaptic membrane, triggering it to generate its own impulses. This process is said to be “quantized,” as each vesicle contains a specific amount of transmitter molecules, and thus constitutes a quantum (meaning a “packet”) of the transmitter.³

Neurotransmitters released in this way bind to receptor molecules embedded in the postsynaptic membrane. Some of these receptors are ion channels, which form pores spanning the postsynaptic membrane, and these open upon binding, allowing electrical current—in the form of positively charged sodium, potassium, or calcium ions, or negatively charged chloride ions—to traverse the membrane, altering its conductivity. Others are coupled to so-called second messenger cascades, downstream pathways of enzymes and other proteins, and binding of a transmitter to these receptors brings about longer-lasting biochemical changes within the postsynaptic cell.⁴

Figure 3 Pre- and postsynaptic components of a synapse (https://commons.wikimedia.org/wiki/File:Synapse_Illustration2_tweaked.svg, CC BY-SA 3.0).

In the postsynaptic cell, the movements of neurotransmitter receptors and the various components of their downstream signaling cascades are regulated by an intricate network of scaffolding proteins called the postsynaptic density (PSD), which can be seen with an electron microscope as a fuzzy thickening immediately beneath the membrane. The PSD consists of dozens of different proteins, all of which cooperate to control the movements of receptors and their related molecules within the postsynaptic cell.⁵

Of all the known forms of neuroplasticity, one form of synaptic plasticity, called long-term potentiation (LTP), is the most intensively studied and, therefore, the best understood. LTP is a process that increases the efficiency of synaptic transmission, which is now widely believed to be the neural basis of most, if not all, forms of learning and memory. Modification of synapses also plays an important role in addiction, a maladaptive form of neuroplasticity that involves aberrant learning (see chapter 8).

Long-Term Potentiation and Long-Term Depression

The idea that memory formation involves the modification of synaptic connections is more than 200 years old. In their correspondences during the 1780s, the Swiss naturalist Charles Bonnet and the Italian anatomist Michele Vincenzo Malacarne discussed the idea that mental exercise can induce brain growth. Malacarne agreed to test the idea by taking pairs of dogs and birds and training one from each pair. A few years later, he dissected the animals’ brains, and found that the trained animals had more folds in their cerebella than the untrained ones.⁶

Nearly one hundred years later, the philosopher Alexander Bain suggested that “for every act of memory, every exercise of bodily aptitude, every habit, recollection, train of ideas, there is a specific grouping or coordination of sensation and movements, by virtue of specific growths in the cell junctions.”

In the 1940s, the Canadian psychologist Donald Hebb noticed that the lab rats he took home as pets for his children outperformed others on problem-solving tasks when returned to the lab several weeks later. This seemed to show that early experience can have dramatic and permanent effects on brain development and function. Hebb reported these findings in his influential 1949 book, The Organization of Behavior, concluding that “the richer experience of the pet group... made them better able to profit by new experience at maturity—one of characteristics of the ‘intelligent’ human being.”

In that book, Hebb postulated that memories are formed by the strengthening of synaptic connections. “Let us assume that the persistence or repetition of a reverberatory activity (or ‘trace’) tends to induce lasting cellular changes that add to its stability,” he wrote. “When an axon of cell A is near enough to excite a cell B and repeatedly or persistently takes part in firing it, some growth process or metabolic change takes place in one or both cells such that A’s efficiency, as one of the cells firing B, is increased.” In other words, neurons that fire together, wire together.

The idea was way ahead of its time—it was not until nearly 25 years later that Timothy Bliss and Terje Lømo observed a mechanism just like the one Hebb had described. Working on anesthetized rabbits, Bliss and Lømo used microelectrodes to electrically stimulate fibers of the perforant path while simultaneously recording the electrical responses of neurons in the dentate gyrus of the hippocampus, which are at the end of that pathway.

Stimulation of the perforant path fibers evoked an electrical response in dentate gyrus cells, as expected. Bliss and Lømo also found, however, that repetitive stimulation of the fibers (with a frequency of between 10 and 20 Hertz, or pulses per second) caused a massive increase in the size of the electrical response in the dentate gyrus. As well as being far larger, the responses also lasted longer, so that the cells took much longer to return to baseline.⁷

Repetitive stimulation had dramatically increased the efficacy of the neurochemical signaling between the perforant path fibers and neurons in the dentate gyrus, strengthening the synaptic connections between them. In Bliss and Lømo’s initial experiments, this strengthening lasted for periods of between 30 minutes and 10 hours, and so they named it long-term potentiation (LTP); but we now know that it can persist for days or weeks, and perhaps even longer.

The induction of LTP is dependent upon binding of the excitatory neurotransmitter glutamate to N-methyl-D-aspartate (NMDA) receptors. The NMDA receptor is an ion channel that is permeable to sodium, potassium, and calcium, but the central pore that allows these ionic currents to pass is blocked by a magnesium ion.

Under normal circumstances, this magnesium block remains, and the glutamate released from a nerve terminal works through two other receptor types, the AMPA and kainate receptors. High-frequency stimulation of the kind that induces LTP increases the amount of glutamate released by the nerve terminal and removes the magnesium block, allowing currents to flow through the NMDA receptors. The influx of calcium is particularly important, as it triggers various enzymes needed for the cellular processes underlying LTP.⁸

NMDA receptors thus have unique biophysical properties that make them perfectly suited to triggering LTP. The magnesium block ensures that they are activated only in response to high-frequency stimulation from the presynaptic cell, and the calcium currents that flow through them are highly localized, producing very discrete “microdomains” of elevated calcium-ion concentration, such that LTP can be restricted to individual dendritic spines, or subsets of them, on a given neuron.⁹

LTP involves changes in both the pre- and postsynaptic components of the connection that is being strengthened. At the nerve terminal, each active zone typically has a pool of several hundred vesicles, but only a small proportion of these are available for release at any time.

High-frequency stimulation enhances glutamate release at the nerve terminal, either by increasing the number of vesicles that fuse with the membrane, expanding the pool of available vesicles, speeding up the recycling process, or a combination of these.

Using methods such as confocal microscopy, it is now possible to tag individual receptor molecules with fluorescent marker molecules or quantum dots, then visualize their distribution and track their movements in living cells, isolated from the brains of animals and kept alive in Petri dishes. Using such methods, researchers have shown that neurons have mobile and immobile pools of glutamate and GABA receptors on their surface, and that receptor molecules can move rapidly around inside neurons.

This receptor trafficking can enhance the responsiveness of the postsynaptic cell. Induction of LTP mobilizes AMPA receptors, inserts them into the membrane, and then shuttles them within it so that they become highly concentrated at the synapse but not in other parts of the dendritic spine. In the same way, LTP is thought to awaken “silent” synapses by insertion of AMPA receptors, which they normally lack. Mobilized receptors are transported in spherical, membrane-bound structures that resemble synaptic vesicles, and are inserted into the membrane by exocyctosis, the same process by which vesicles fuse with the presynaptic membrane during neurotransmitter release.¹⁰

At excitatory synapses, the movements of AMPA receptors are orchestrated by the scaffolding proteins of the postsynaptic density, which is restricted to the tip of the dendritic spine and which keeps the receptors and their downstream signaling partners anchored in their proper place. Following induction of LTP, calcium currents that flow in through NMDA receptors activate enzymes that redistribute the receptors by rearranging the scaffolding.¹¹

Once LTP has been induced, the postsynaptic cell sends a signal back to its presynaptic partner. Once it enters the presynaptic cell, this back-propagated signal activates genes that synthesize the many cellular proteins needed to maintain LTP. The gaseous neurotransmitter nitric oxide has been implicated as this so-called retrograde messenger.

All of these mechanisms are reversible. The rate at which a nerve terminal recycles spent synaptic vesicles can be reduced, for example, leading to a depletion in the number of readily available vesicles at the active zone. And receptors can be removed from the postsynaptic membrane just as rapidly as they can be inserted. Together, these events have the opposite effect of long-term potentiation: they make neurotransmission less efficient, and therefore weaken synaptic connections, in a process referred to as long-term depression (LTD). LTD is also dependent upon the NMDA receptor, but it is induced by repetitive low-frequency stimulation of a presynaptic neuron in the absence of a postsynaptic response.¹²

Bliss and Lømo concluded their classic 1973 paper describing LTP on a cautionary note: “whether or not the intact animal makes use [of LTP] in real life... is another matter.” But the fact that it was discovered in the hippocampus, which by then was already strongly implicated in memory, strongly suggested that LTP underlies learning, and ever since then, evidence that synaptic strengthening is indeed necessary and sufficient for memory formation has been slowly accumulating.

For example, when mice are placed into a circular pool of water, they can locate submerged platforms, and quickly form spatial memories of their exact locations, so that they can swim directly to the platforms when placed back into the water later on. But treating the mice with NMDA receptor–blocking drugs during the learning process prevents the formation of spatial memories, so that they are unable to find the hidden platforms afterwards.¹³

Researchers now have more sophisticated methods at their disposal, and one method in particular—optogenetics—allows for the control of neuronal activity with unprecedented precision. Optogenetics involves introducing the genes encoding algal proteins called channelrhodopsins into specific types of neurons. The cells then use their new genes to synthesize channelrhodopsin protein molecules, and insert them into the membrane, making the cells sensitive to light. The cells can then be switched on or off on a millisecond-by-millisecond timescale, depending on which channelrhodopsin they are synthesizing.

Using this method, researchers can now label the hippocampal neurons that fire during memory formation, and reactivate them with pulses of laser light delivered into the animals’ brains by optical fibers. Reactivation of hippocampal neurons that fire when mice learn to associate an unpleasant experience with particular location of their environment produces fear responses in the animals, strongly suggesting that the reactivation leads to retrieval of the fearful memories. This same method can be used to manipulate memories in various ways—to switch fearful memories into pleasant ones, or vice versa, and to implant completely false fearful memories into the mouse brain.¹⁴

Studies such as these provide the most compelling evidence yet that synaptic modification is the neural basis of learning and memory, and it is now widely believed that both strengthening and weakening of synapses are essential for both processes. Current thinking holds that memories form when specific sets of synapses are strengthened and others weakened, within a distributed network of hippocampal neurons, and that retrieval requires reactivation of the same neuronal network.

Synapse Formation

LTP is a form of functional plasticity that involves transient molecular changes on both sides of the synapse, but learning and memory also involve structural changes that can significantly alter neuronal architecture. As well as modifying the strength of existing synaptic connections, experience and learning lead to the creation of entirely new synapses.

The vast majority of excitatory neurotransmission in the brain takes place at dendritic spines, and so researchers have focused their attention on understanding how learning and experience alter the form of these tiny structures. Dendritic spines were discovered by Cajal over a century ago, in the cerebellum of birds, but it was not until the development of electron microscopy in the 1930s that researchers could study them in any great detail.¹⁵ By cutting brain tissue into a series of ultrathin slices, imaging each one, and then painstakingly reconstructing all the images, they began to get a better idea of how spines and synapses are arranged on the dendrites of postsynaptic neurons, and also of how they can be rearranged in response to sensory experience.

Early studies yielded conflicting evidence. Some showed that spines increase in size by about 15% within 2 to 6 minutes of induction of hippocampal LTP, and then grow even larger at between 10 and 60 minutes, while others showed that LTP causes a marked increase in the surface area of the postsynaptic density. Some researchers observed an increase in the number of spines and synapses, but no changes in size, following LTP induction, and yet others noted significant increases in spine volume but not numbers.¹⁶

The development in the 1990s of high-resolution time-lapse imaging techniques such as two-photon laser scanning microscopy enabled researchers to examine these processes in even greater detail. Initially, experiments like this were performed in brain tissue dissected from animals and maintained in Petri dishes, but they can also be done in live animals through “cranial windows,” or thinned sections of the animal’s skull. Combined with the use of sensor molecules, which fluoresce in response to the localized increases in calcium ion concentration produced by NMDA receptor activation, in vivo imaging can be used to monitor these processes for prolonged periods of time during sensory experiences or learning of a new motor skill.

These newer methods confirm earlier findings, showing again that sensory experience can produce structural changes to dendritic spine morphology, and that LTP can induce rapid changes in the size, shape, and number of synapses. Following induction of LTP, new spines form on the dendrite, sometimes forming connections with the same synaptic bouton that triggered their formation. The heads of existing spines grow larger, while their necks become shorter and wider. Spine head volume can increase threefold within one minute of repeated electrical stimulation. All of these changes facilitate the trafficking of receptors into the spine heads, making them more sensitive to glutamate.

Learning and experience likely lead to the patterned formation of new spines along the same dendrite branch and also across other branches of the same dendritic tree. Motor learning induces clusters of new spines to form in adjacent locations on the dendrites of cells in the mouse motor cortex, and causes weakening and shrinking of neighboring clusters; the new clustered spines are more persistent than spines that form alone.¹⁷

It is tempting to speculate that the persistence of memory is related to the stabilization of new dendritic spines and to synchronized activity in neighboring synapses. Structural changes to dendrites involve reorganization of the filamentous proteins that make up the postsynaptic density, by the same signaling pathways triggered by the NMDA receptor following induction of LTP. Furthermore, different motor tasks activate NMDA receptors to produce calcium microdomains on different branches of individual pyramidal neurons in the mouse motor cortex. Thus, individual branches of dendritic trees, or subsets of spines on them, might serve as basic units for storing information. Such mechanisms could help to explain the brain’s extraordinary capacity for memory storage.¹⁸

Yet, the precise relationship between synaptic modification, spine formation, and memory is still unclear, and there is some evidence to suggest that new spines are not actually necessary for memory. For example, spine density in the squirrel brain decreases dramatically during hibernation and increases again afterwards, but the animals can still remember tasks they learned before they started hibernating. Similarly, spine density in the hippocampus is reduced by 30% in female rats in estrus, but they can still remember items they learned earlier in the menstrual cycle.

Findings like this suggest that the persistence of dendritic spines is not necessary for long-term storage of memories. But the conflicting findings on exactly how experience and learning alter dendrite architecture could be due in part to differences in the type of stimulation used, or the brain area being studied. There is even some evidence suggesting that merely handling brain tissue in preparation for experiments can alter the density of spines within it.

To complicate matters, dendritic spines exist in a variety of forms, and its thought that any individual spine can morph between and adopt all of them. There are mushroom-like spines with large, round heads attached to their parent dendrite by a narrow neck; long spines, which appear as thin, finger-like protuberances; and small spines, which are short and stout and have no noticeable neck. It’s possible that each of these forms contributes to different aspects of memory storage, or that different types of memory cause different types of structural changes to dendrite architecture.¹⁹

Synapses can also be weakened, and the spines associated with them can shrink, pull away from their presynaptic partners, or even retract and be eliminated altogether. Synapse elimination, or synaptic pruning, occurs extensively during brain development, and is critical for shaping and fine-tuning neural circuits as they form (see chapter 3). Pruning also takes place widely in the adult brain, and, like LTP and synapse formation, is thought to be necessary for learning and memory.

Thus, learning, memory, and other experiences probably produce widespread patterns of synaptic modification throughout entire networks of neurons in particular regions of the brain, depending on the type of experience. Synaptic modification takes place continuously throughout the brain, and it is likely that millions of synapses are modified in the human brain every second in one way or another. Current imaging methods are rather limited in their field of view, being restricted to several branches of a dendritic tree, but emerging techniques such as super-resolution microscopy will undoubtedly reveal more about dendritic spine dynamics and their contribution to long-term memory.

Glial Cells: Partners in Plasticity

Glial cells are the nonneuronal cells of the nervous system, outnumbering neurons by about ten to one. They were discovered at around the same time as neurons, but were believed to play only supportive roles such as providing nutrition and insulating nerve fibers; hence their name glia, meaning “glue.” Glial cells do perform these roles, but we now know that they also make important contributions to—and are just as crucial for—information processing in the brain and spinal cord.

Traditionally, synapses were thought to consist of just two elements, the presynaptic bouton and postsynaptic membrane. In the early 1990s, however, evidence began to emerge that they are in fact tripartite structures, and that glial cells called astrocytes regulate the chemical signals that are transmitted between neurons.

Astrocytes are star-shaped cells that were initially thought to fill the extracellular spaces in brain tissue. But it is now clear that they not only respond to neuronal activity but can also produce their own electrical signals, and they synthesize and release a whole host of neurochemical transmitters, including glutamate and GABA.

Astrocytes are by far the most numerous cell type in the brain. Each one has many fine branches that come into contact with hundreds of dendrites and up to 150,000 individual synapses. These processes are highly motile, and rapidly extend toward and envelop active synapses. Electron microscopic examination of brain tissue reveals that their fibers interact with large dendritic spines in response to neuronal activity, and that these fibers are less motile than those associated with small spines.

Large spines tend to be more persistent than smaller ones, and so it seems that astrocytes help to stabilize those spines with active synapses. There is also some evidence that astrocytes can modulate synaptic signaling by clasping synapses to restrict the diffusion of neurotransmitters, or loosening their grip to allow them to flow more freely.

Astrocytes form networks with each other and with their neuron neighbors. Whereas neurotransmission takes place over a timescale of milliseconds, astrocyte activity lasts for a few seconds. When an astrocyte releases glutamate, it excites whole clusters of neurons, and their prolonged activity may be a way of synchronizing activity of entire populations of neurons. The prolonged activity of astrocytes may also contribute to LTP by persistently activating postsynaptic membranes to coincide with incoming signals.²⁰

Microglial cells also play important roles in synaptic plasticity. These are the brain’s resident immune cells, which provide the first line of defense against infection and injury. They are deployed to damaged sites, where they engulf pathogens and cellular debris by enveloping them in a small segment of membrane and then internalizing them, a process called phagocytosis, or “cell eating.”

Unwanted connections are “tagged” for destruction with immune system molecules called complement proteins. Microglia recognize this as a signal saying “eat me,” and engulf all the tagged synapses they come across.

It turns out that the developing brain treats unwanted synaptic connections in exactly the same way. Unwanted connections are “tagged” for destruction with immune system molecules called complement proteins. Microglia recognize this as a signal saying “eat me,” and engulf all the tagged synapses they come across. It’s now thought that microglia are responsible for synaptic pruning throughout the developing brain, as well as for the extensive pruning that occurs in adolescence (see chapters 3 and 9).

Synapses are also constantly being eliminated in the adult brain, and it seems that microglia are responsible for this, too. They continuously patrol their patch of brain tissue, and preferentially contact stubby spines, which are usually the least persistent of newly formed spines. Thus, microglia seem to monitor the status of synapses in their patch and engulf the unwanted ones.²¹