The Synapse

What a synapse is, and why there is a gap

A synapse is the specialised point of contact where one neuron passes information to another cell, which may be a second neuron, a muscle fibre, or a gland. It is the working joint of the nervous system. The neuron sending the signal is called the pre-synaptic cell, and the one receiving it is the post-synaptic cell. The two are not simply wired together end to end; they meet at a carefully organised interface built to convert, and often to modify, the message as it passes.

The striking feature of most synapses is that the cells do not actually touch. A microscopic gap, the synaptic cleft, separates the pre-synaptic membrane from the post-synaptic one. This gap is tiny, on the order of tens of nanometres, but it is real, and it changes everything about how the signal crosses. An electrical impulse cannot simply leap across empty space in the way it flows along a continuous membrane. Instead the sending cell converts its electrical signal into a chemical one, releasing molecules that drift across the gap and act on the receiving cell. This is the chemical synapse, and it is by far the most common type in the human brain.

It is worth stopping on that gap, because textbooks tend to present it as a curiosity, and it is not a curiosity. It is a design decision, and an expensive one. Consider what the gap costs. It adds delay: the transmitter must be released, must diffuse, and must bind, which takes roughly a millisecond, an eternity by the standards of a signal that has just travelled a metre in ten. It costs energy: the terminal must synthesise transmitter, load it into vesicles, and then clear it out of the cleft again, and every one of those steps consumes ATP. And it is less reliable: at many central synapses a single arriving impulse fails to release any transmitter at all a substantial fraction of the time, whereas a direct electrical connection would simply pass current. Slower, costlier, less reliable. On the face of it the gap is a bad idea.

So what does it buy? It buys modifiability. A direct electrical junction, and the brain does have these, is fast but rigid: current passes, and there is very little the cell can do to change how much. A chemical junction, by contrast, is a chain of separable, adjustable steps. The terminal can hold more or fewer vesicles and can release with higher or lower probability. The cleft can be cleared faster or slower. The receiving membrane can add receptors or take them away. Every stage of the chain is a knob, and a knob that turns can store a value. That is why the brain accepts the delay, the energy, and the unreliability: a rigid connection can carry a signal, but only an adjustable one can carry a signal and remember something about it. The cleft is the price of learning.

A single neuron may carry thousands of synapses on its dendrites and cell body, each a small inbound channel. The neuron continuously weighs all of these inputs together, and the overall balance of what arrives across its synapses determines whether it will, in turn, fire a signal of its own. Understanding the synapse is therefore the key to understanding how the brain computes: it is where one cell's output becomes another's input, and it is the only place in the whole system where that relationship can be rewritten.

Chemical transmission step by step

Chemical synaptic transmission is a rapid, tightly ordered sequence. It begins when an impulse arrives at the sending terminal and ends, a fraction of a millisecond later, with a change in the receiving cell. The following steps trace the message across the gap.

1. The impulse arrives

   An action potential travels down the axon and reaches the pre-synaptic terminal, the swollen ending of the sending neuron. Here the brief change in membrane voltage sets the whole process in motion.

2. Calcium channels open

   The arriving voltage opens voltage-gated calcium channels in the terminal membrane. Because calcium is far more concentrated outside the cell than inside, it rushes in through these channels, producing a sharp local rise in calcium just inside the terminal.

3. Vesicles release their transmitter

   That surge of calcium is the trigger. Calcium binds a sensor protein sitting on the surface of the synaptic vesicles, tiny membrane-bound sacs already loaded with neurotransmitter and docked ready at the release site. The bound sensor forces the vesicle's own membrane to fuse with the pre-synaptic membrane, so that the inside of the vesicle becomes momentarily continuous with the outside of the cell and its contents spill into the cleft. That fusion-and-dumping event is what exocytosis means: the vesicle does not squeeze its cargo through a pore, it merges with the wall and opens.

4. Transmitter crosses the cleft

   The released neurotransmitter diffuses across the narrow gap. Because the distance is so small, this crossing takes only a fraction of a millisecond, which is why chemical synapses can still be fast despite the extra chemical step.

5. Receptors bind the transmitter

   On the far side, the transmitter binds to specific receptor proteins embedded in the post-synaptic membrane. Each receptor recognises a particular neurotransmitter, the way a lock accepts one key, so the message is delivered only to cells equipped to read it.

6. Ion channels open and a potential forms

   Binding opens ion channels in the post-synaptic membrane, letting charged ions flow in or out. The moving charge shifts the voltage of the receiving cell, and that small, local, graded voltage shift is called a post-synaptic potential. It is not an impulse and it does not travel; it is a brief nudge to the cell's voltage that fades within milliseconds unless something reinforces it. It is excitatory if it moves the cell towards firing and inhibitory if it moves the cell away from firing or pins it in place.

The receiving neuron does not usually fire from a single synapse. A single post-synaptic potential is a fraction of a millivolt; the cell has to travel some fifteen millivolts to fire. So the neuron sums many post-synaptic potentials arriving from many synapses at once, adding them across space, because inputs at different dendrites converge, and across time, because inputs in quick succession pile up before the earlier ones have decayed.

If the running total crosses threshold, the cell generates its own impulse and the chain continues. If it does not, nothing happens at all. In this way each synapse casts a small vote, and the neuron tallies them moment by moment.

Why calcium, and not any other ion

Step two of the sequence says calcium enters and step three says calcium triggers release. That is true, and it is usually where the explanation stops. But it leaves an obvious question unanswered. Sodium also floods into the terminal when the impulse arrives, in far greater quantity. Why is calcium the trigger and sodium merely a carrier of charge?

The answer is a matter of arithmetic, and it is one of the most elegant facts in cell biology. The cell spends energy continuously pumping calcium out of itself and sequestering it into internal stores, with the result that the concentration of free calcium inside a resting neuron is held at something on the order of one ten-thousandth of the concentration outside. Intracellular calcium is, for practical purposes, near zero. Sodium is not: sodium is already present inside the cell in appreciable quantity, and a further influx raises it by a modest fraction.

Now open a channel. Sodium entering the terminal changes the internal sodium concentration a little. Calcium entering the terminal changes the internal calcium concentration by orders of magnitude, and it does so within microseconds and within nanometres of the channel mouth. In signalling terms, calcium has an enormous signal-to-noise ratio and sodium has a poor one. A protein that must decide, unambiguously and in a fraction of a millisecond, whether an impulse has just arrived can read calcium and be certain. It could not read sodium and be certain of anything.

The same logic governs the receiving side of the synapse, as the section on coincidence detection below will show. When a receptor needs to do more than pass current, when it needs to tell the cell that something worth remembering has just happened, the ion it admits is calcium.

Clearing the signal

A synapse that could only turn on would be useless. For the message to be crisp, the neurotransmitter must be removed from the cleft quickly once it has acted, so the synapse can reset for the next impulse. There are two main ways this clearance happens, and both are fast.

Clearance is not a footnote: it is central to how the nervous system is tuned, and it is where many medicines act. Drugs that block reuptake leave a transmitter in the cleft for longer, strengthening its effect, which is how a common class of antidepressants that block serotonin reuptake works. Understanding clearance therefore matters both for how the synapse works normally and for how it can be adjusted.

Electrical synapses

Not every synapse uses chemicals. At an electrical synapse the two cells are joined directly by clusters of channels called gap junctions. These channels line up across the membranes of the two cells and form open pores, so that ions and the electrical current they carry can pass straight from one cell into the next with almost no delay.

This direct coupling gives electrical synapses two distinctive properties. First, they are extremely fast: because there is no chemical step, the signal passes essentially instantly, which suits circuits that must act in tight synchrony. Second, they are typically bidirectional, so current can flow either way between the coupled cells, unlike a chemical synapse where the signal runs only from pre-synaptic to post-synaptic side. Electrical synapses are less common in the mammalian brain than chemical ones, and they offer far less scope for modifying the signal, but where speed and synchrony matter most they are invaluable.

Excitatory and inhibitory synapses

A synapse does not only pass a message on; it also determines in which direction the receiving cell is pushed. It is tempting to say that an excitatory synapse is one that excites, but that explains nothing. The sign of a synapse is not a property the synapse simply possesses. It is a consequence of physics, and it can be derived, so let us derive it.

Start with what an open ion channel actually does. A channel does not push the membrane in some fixed direction of its own. It opens a hole through which one kind of ion can move, and that ion then moves under two forces: the concentration gradient, which pushes it from where it is abundant towards where it is scarce, and the electrical gradient, which pushes it according to its charge and the voltage across the membrane. For every ion there is one particular membrane voltage at which those two forces exactly cancel and the net flow stops. That voltage is the ion's equilibrium potential, and it is fixed by the ion's concentrations on the two sides of the membrane.

The consequence is a single rule, and every excitatory and inhibitory effect in the brain follows from it: opening a channel drags the membrane voltage towards the equilibrium potential of whatever ion that channel passes. The channel is not a pump or a lever. It is a leak towards a target value, and the target value is a property of the ion, not of the synapse.

Now apply the rule.

That last point is the one usually skipped, and it is the interesting one. Because chloride's equilibrium potential sits near the resting potential, an inhibitory synapse frequently does not drive the cell downwards at all. What it does is open a large conductance that holds the voltage where it is. Any excitatory input arriving at that moment is now trying to drag the membrane towards plus 60 while an open chloride leak is dragging it back towards minus 70, and the excitatory input is substantially blunted. The inhibition works not by subtracting a voltage but by short-circuiting the attempt to change one. This is called shunting inhibition, and it explains why a synapse can be powerfully inhibitory while producing almost no visible deflection in the membrane voltage.

Note what this derivation buys. It means the sign of a synapse can change without anything about the transmitter changing. In immature neurons the chloride gradient is reversed, so chloride's equilibrium potential sits above threshold, and GABA is excitatory in the developing brain. The transmitter is the same molecule. The receptor is the same protein. Only the ion gradient has changed, and the sign flips with it. That is the clearest possible demonstration that excitation and inhibition are properties of ions and gradients, not labels attached to chemicals. For the underlying gradients and pumps, see ion channels and electrolytes.

Healthy brain function depends on a careful balance between these two forces. Too little inhibition and activity can spread unchecked, as in the runaway firing of a seizure; too much and signalling is damped down. Because each neuron sums excitatory and inhibitory inputs continuously, the synapse is not just a relay but a genuine site of computation, weighing evidence for and against firing before the cell commits.

How a synapse detects coincidence

Everything so far describes a synapse that works. Nothing so far describes a synapse that learns, and there is a hard problem standing in the way. If a synapse is to strengthen when the sending and receiving cells are active together, then the synapse has to be able to tell that they were active together. But a synapse is a patch of membrane a fraction of a micrometre across. It has no vantage point, no clock, and no view of two cells at once. So the question that the whole of learning turns on is a small and concrete one: how does a synapse know that the two cells fired together?

The answer is that it does not need to know. A single protein computes it, and the computation falls out of the protein's physical structure. The protein is the NMDA receptor, and understanding it is the most valuable thing on this page.

The starting point is that glutamate, the main excitatory transmitter, does not bind one kind of receptor. It binds two, and they sit side by side in the same post-synaptic membrane, listening to the same released glutamate, and they behave completely differently.

The obstruction is a single magnesium ion, Mg2+, lodged in the pore. Nothing holds it there mechanically. It is held there by the electric field across the membrane. The inside of the resting cell is negative, at about minus 70 millivolts, and magnesium carries two positive charges, so it is drawn inwards and settles in the mouth of the channel, where it fits and sticks. It is a plug held in place by voltage.

Now work through the consequences, because the whole of Hebbian learning is contained in them.

1. Glutamate arrives, and the cell is quiet

   The pre-synaptic cell fires and releases glutamate. AMPA receptors open and produce a small depolarisation, too small to matter. Glutamate also binds the NMDA receptors and opens their gates. But the cell is still near minus 70, the field is still strong, and the magnesium is still in the pore. The NMDA receptors conduct essentially nothing. The synapse has transmitted, and it has not learned. This is what happens at almost every synapse, almost all of the time.

2. The cell is already depolarised, but no glutamate arrives

   Other inputs elsewhere on the neuron have driven it well above rest. The membrane field is weak, and any magnesium sitting in an NMDA pore is no longer held: being positive, it is now repelled from the positive interior and expelled. The pore is clear. But an unplugged NMDA receptor still conducts nothing, because glutamate is what opens its gate and no glutamate has been released at this synapse. Again, nothing.

3. Both at once, and only then

   The pre-synaptic cell releases glutamate while the post-synaptic cell is already depolarised, which in practice means while enough other converging inputs have driven their AMPA receptors hard enough to lift the membrane. The gate is open because glutamate is bound. The pore is clear because the depolarisation has expelled the magnesium. Both conditions are satisfied simultaneously, and the channel conducts.

That is a molecular AND-gate. It is one protein whose output reports the conjunction of two facts that occur in two different cells, and it is worth pausing to appreciate how strange and how economical that is. The NMDA receptor does not measure correlation, does not compare, and does not remember. It simply has two locks that must both be picked, one by a chemical and one by a voltage, and the chemical can only come from the sending cell while the voltage can only come from the receiving one.

There is, in fact, a third lock, and it belongs to neither cell. Alongside the site that binds glutamate, the NMDA receptor carries a separate co-agonist site that must also be occupied before the channel will conduct, and the molecule that occupies it, glycine or D-serine depending on the region, is supplied in large part by the surrounding astrocytes. The consequence is remarkable and easy to miss: a glial cell, which fires no impulses and sends no messages of its own, holds a veto over whether a given synapse is able to change at all. Coincidence detection is the neuron's business; whether the detector is switched on is not entirely up to it.

And here is the payoff, which is the point of the whole section. Donald Hebb, writing in The Organization of Behavior in 1949, proposed that when one cell repeatedly takes part in firing another, the connection between them should strengthen. He had no mechanism; he was writing a specification for one, because a learning brain would need something like it. The popular compression of that idea, that neurons which fire together wire together, is usually offered as an explanation. It is not an explanation. It is a restatement of the phenomenon, and by itself it is no better than saying that things fall because they are heavy.

One thing remains, and it is what makes the AND-gate useful rather than merely clever. When the NMDA receptor finally does conduct, what comes through it is calcium. For the reasons set out earlier on this page, calcium is not simply charge. Intracellular calcium is held near zero, so its arrival is an enormous proportional change and therefore an unambiguous message. It binds enzymes. It is a second messenger, and it does not merely shift the voltage: it starts a chemical cascade that can change what the synapse is made of. So the NMDA receptor does not just detect that two cells fired together. It converts that detection into the one chemical signal the cell reads as an instruction to rebuild itself.

Synaptic plasticity and learning

The most remarkable property of the synapse is that it does not stay fixed. Its strength, meaning how large a response the pre-synaptic cell produces in the post-synaptic one, can change with use, and the change can outlast the activity that caused it by hours or by years. This capacity is called synaptic plasticity, and it is widely held to be the physical basis of learning and memory.

The previous section supplied the detector. This section supplies what the detector does, because a coincidence signal that led nowhere would be of no use at all. Calcium has entered through the NMDA receptor, and calcium is an instruction. The question is what gets built.

The answer is disarmingly concrete. What changes is the number of AMPA receptors in the post-synaptic membrane. Recall that AMPA receptors are the ones doing the ordinary work: glutamate binds, sodium enters, the cell depolarises. Put more of them in the membrane and the identical puff of glutamate opens more channels, admits more sodium, and produces a larger post-synaptic potential. Take some away and the identical puff produces a smaller one. The strength of a synapse is, to a first approximation, a count of the AMPA receptors facing the cleft. Learning is the editing of that count.

It is worth asking why the brain needs the downward direction at all, because the answer is not obvious and the usual accounts skip it. Suppose a synapse could only ever be strengthened. Every episode of correlated activity would push it up, nothing would ever push it back, and given enough experience every synapse in the network would arrive at its ceiling. A network in which everything is maximally connected to everything carries no information whatsoever. It has forgotten by remembering too much. Long-term depression is not an afterthought or a failure mode: it is what keeps synaptic strengths within a usable range, what allows connections that turn out to predict nothing to be quietly turned down, and what makes memory a pattern of strong and weak synapses rather than a uniform saturated blur. A brain that could only learn would be no better than a brain that could not.

Long-term potentiation was first observed by Tim Bliss and Terje Lomo, working in Oslo and reporting in the Journal of Physiology in 1973. They stimulated a pathway into the hippocampus of an anaesthetised rabbit with a brief burst of high-frequency shocks, then returned to the same routine test shock they had been using all along. The response was larger, and it stayed larger for hours. Nothing had been damaged and nothing about the measurement had changed: the synapse itself had become more effective. That result, arriving twenty-four years after Hebb had specified what such a mechanism would have to look like, is where the modern neurobiology of memory begins. The full account, including how the early phase differs from the late phase and what the evidence does and does not establish, is on the long-term potentiation page.

Chemical versus electrical synapses

The two kinds of synapse solve the same problem, passing a signal between cells, in very different ways. The contrast below sets out where each excels.

The trade-off is clear. Electrical synapses win on raw speed and are ideal where cells must act as one, but they offer little room to shape or store a signal. Chemical synapses pay a small price in speed for an enormous gain in flexibility, and it is that flexibility, above all the ability to change their strength, that makes them the dominant currency of thought.

Related topics