Glutamate and Ionotropic Glutamate Receptors

Glutamate, used at approximately 80% of all synapses in the mammalian brain. Most neurons in the mammalian brain are not spontaneously active, and become excited through excitatory synaptic input. Glutamatergic synaptic transmission is therefore the main way neurons in the brain are driven to fire action potentials.

1. The action potential propagates down the axon and invades the presynaptic bouton.
2. Calcium influx into the bouton triggers exocytosis of docked, release-ready synaptic vesicles each containing approximately 100 mM of glutamate.
3. Glutamate is released into the synaptic cleft and diffuses across it, temporarily increasing the glutamate concentration in the synaptic cleft to a peak of 1 mM for approx. 1 ms.
4. Glutamate binds postsynaptic ionotropic glutamate receptors and increases their open probability.
5. Cation flux through those receptors depolarizes the postsynaptic cell, giving an excitatory postsynaptic potential (this is the most general level, there are also more complex dynamics with AMPA and NMDA receptors)

Approximately 100 mM. Glutamate is loaded by the vesicular glutamate transporter, which is present on every synaptic vesicle (it is a membrane protein in the vesicle membrane with its transport domain facing the cytosol so it can take up glutamate and move it into the vesicle lumen).

Approximately 1 mM, for a period of approximately 1 ms. Synaptic communication therefore relies on brief transient pulses of glutamate, and postsynaptic glutamate receptors must be sensitive to a pulse of this amplitude and duration. This is why the kinetics of glutamate receptors are studied using 1 ms pulses of 1 mM glutamate.

AMPA receptors and NMDA receptors, which are cation channels that are permeable to Na+, K+, and in some cases calcium. Ionotropic means the receptor is itself a ligand-gated ion channel, whose open probability depends on the binding of the neurotransmitter (in this case glutamate). They are therefore distinct from metabotropic receptors that act indirectly through second-messenger cascades.

Each subunit has four hydrophobic regions (M1-M4) that enter the plasma membrane, but only three alpha helices cross it completely; the second region enters the membrane from the intracellular side without crossing it all the way [re-entrant pore loop]. Because only three helices fully cross, the N terminal and C terminal domains lie on opposite sides of the membrane, with the N terminal extracellular and the C terminal intracellular. The N terminal domain has a large extracellular region where the glutamate binding domain is located. The binding of glutamate changes the structure of the ion channel and increases its open probability.

Four subunits come together around a central pore to form the functional ligand-gated ion channel. The glutamate binding domain is located in the large extracellular region of the N terminal domain of each subunit, so the assembled channel has four glutamate binding sites. Glutamate binding to a number of the four subunits changes the structure of the channel and increases its open probability. The resulting pore is a cation channel permeable to Na+, K+ and in some cases Ca2+.

Four genes encode AMPA receptor subunits and seven encode NMDA receptor subunits, meaning there are four distinct types of AMPA subunits (GluA1-4) and 7 distinct types of NMDA subunits (GluN1, GluN2A-D, GluN3A-B). The two major subtypes of ionotropic glutamate receptor are therefore products of distinct gene families. They are also distinguishable by their ion permeability, single-channel conductance, kinetics, and voltage dependence.

AMPA receptors are permeable to Na+ and K+, being slightly more permeable to Na+, and most are not permeable to Ca2+. The reversal potential is around 0 mV, well depolarized relative to action potential threshold, so at resting potential there is a net inward cation current and the AMPA conductance is excitatory. The single-channel conductance is approximately 5 pS, though the precise value depends heavily on which subunits form the channel.

NMDA receptors are permeable to Na+, K+ and, importantly, Ca2+, which most AMPA receptors are not. The reversal potential is also around 0 mV, so like the AMPA conductance it is depolarizing and excitatory (0 mV will depolarize most neurons which have RMP of -70 mV). The single-channel conductance is approximately 50 pS, roughly ten times larger than that of the AMPA receptor. NMDA receptors also require a co-agonist (glycine or D-serine) to open, and because of their Mg2+ block at negative membrane potentials they require sufficient postsynaptic depolarization as well.

A co-agonist: either glycine or D-serine must be present in the extracellular fluid. Glutamate alone will not open the NMDA receptor channel. These co-agonists are usually present, so they are not normally thought to be an important regulator of NMDA receptor activation, although there may be circumstances in which they act as genuine regulators.

Note that even with glutamate and co-agonist bound, little current flows unless the postsynaptic membrane is depolarized enough to relieve the Mg2+ block.

Because postsynaptic Ca2+ acts as a signalling molecule, just as presynaptic Ca2+ does in triggering neurotransmitter release. Ca2+ entering the postsynaptic cell through NMDA receptors plays a key role in governing synaptic plasticity. This is one of the main functional differences between NMDA receptors and most AMPA receptors.

AMPA receptors activate extremely quickly and also switch off extremely quickly (after a few ms), giving only a couple of milliseconds of glutamate-gated current. It is very hard to have long-lasting AMPA currents: prolonging the glutamate does not prolong the AMPA current (AMPA current remains short-lasting), because AMPA receptors desensitize strongly in the continued presence of glutamate. AMPA receptors are therefore only briefly activated and give transient currents lasting a few milliseconds.

A state in which the AMPA receptor closes and stops conducting despite glutamate still being bound. AMPA receptors desensitize strongly, so they give only transient currents lasting a few milliseconds. The mechanism is a conformational rearrangement of the glutamate-binding domain that uncouples glutamate binding from channel opening--->this is distinct from deactivation which is channel closure due to glutamate unbinding.

NMDA receptors respond slowly: it takes tens of milliseconds, approximately 20 ms, before the NMDA conductance is maximally opened, long after the glutamate has gone. The NMDA receptor currents then remain active for hundreds of milliseconds after the glutamate pulse, because once glutamate is bound it remains bound. NMDA receptor kinetics are therefore 1-2 orders of magnitude slower than AMPA receptor kinetics.

AMPA: rapid activation of a brief cation current that quickly shuts off, lasting a couple of milliseconds.
NMDA: slow activation, taking tens of milliseconds to peak, and a cation current lasting hundreds of milliseconds.

NMDA kinetics are therefore 1-2 orders of magnitude slower than AMPA kinetics (100s of ms vs a couple ms)

Two fluids are run side by side in laminar flow, one containing 0 mM glutamate and the other 1 mM glutamate, but otherwise the fluids contain identical concentrations of the other components of the ECF. The two flows are mounted on a piezo actuator, which rapidly moves the interface so that different concentrations reach a given point in space. This allows a 1 ms pulse of 1 mM glutamate to be applied to a small patch of membrane held on a patch clamp electrode.

1. Bring the patch electrode onto the dendrite and suck a bit of the membrane into the glass electrode to obtain a gigaohm electrical seal of the patch of membrane around the electrode (it is in the cell-attached configuration, recall from week 2)
2. Rupture the membrane with a brief pulse of pressure to obtain whole-cell access (it is now in the whole-cell recording configuration)-->we now have access to the inside of the membrane through the recording electrode. The pipette interior is now part of the inside of the cell.
3. Retract the electrode, pulling a strand of membrane away from the dendrite while the dendrite stays put. Since the pipette was inside the cell, the cytoplasmic face of the membrane faces into the pipette and the extracellular face points towards the bath.
4. Keep pulling until the strand breaks, leaving a small patch of membrane attached to the electrode, with the extracellular face of the membrane pointing out into the bath and the cytoplasmic face pointing into the pipette
5. The glutamate binding sites on the membrane patch are then in the correct orientation, facing the bath outside the electrode, so you can give the 1 ms pulse of glutamate through the piezo transducer, and we can then voltage clamp the patch to measure the currents flowing through the ligand-gated ion channels.

Most AMPA receptors are voltage-insensitive: the conductance activated by a pulse of glutamate does not depend on the membrane potential, because the open probability is independent of membrane potential. So the conductance vs Vm plot is just a horizontal straight line. This gives a linear current-voltage relationship reversing at around 0 mV (the Erev for AMPA receptors). The same amount of current therefore flows at −60 mV as at +60 mV, but in opposite directions.

The conductance of NMDA receptors depends strongly on membrane potential, though not through a change in open probability. Conductance is strongly reduced at potentials more negative than about −40 mV, and reaches full activation at positive potentials. At a resting potential of around −60 to −70 mV, puffing glutamate onto NMDA receptors therefore produces very little current. The resulting curve resembles those of voltage-gated Na+, K+ and Ca2+ conductances, except that the reversal potential is around 0 mV rather than the equilibrium potential of a single ion. So the I-V relation is nonlinear over the whole range negative to the Erev, with inward current that is strongly suppressed at very negative potentials and grows from about −40 mV upwards, and then becomes linear and outward when Vm > Erev.

NMDA receptors are not intrinsically voltage sensitive, i.e. they don't contain an intrinsic voltage-sensing domain. Instead, the voltage sensitivity comes from a voltage-dependent Mg2+ block of the channel pore. With zero extracellular Mg2+, the NMDA receptor has a linear current-voltage relationship just like an AMPA receptor. Physiologically, Mg2+ is always present in the ECF, at approximately 1 mM. At negative potentials, the electric field points towards the inside of the cell, which drives the positively charged Mg2+ into the NMDA channel pore, where it binds a relatively high-affinity site. Mg2+ cannot permeate the channel and so it prevents the permeation of other ions. As the membrane depolarizes, the inward electric field weakens, so Mg2+ is no longer driven into the pore and is pushed back out of this internal site, and full current flows.

The field set up across the bilayer by the membrane potential itself. At a negative potential such as −70 mV, charge is separated across the membrane, negative on the inner leaflet and positive on the outer. Because the bilayer is only about 4 nm thick, this gives a field on the order of 10^7 V/m, strong enough to act on individual ions, pulling cations inwards. Depolarizing towards 0 mV discharges this capacitor, so the separated charge and the field weaken to zero. Depolarizing beyond 0 mV recharges it with the opposite polarity, positive on the inner leaflet and negative on the outer, so the field grows again in the reverse direction and now pushes cations outwards.

It does not affect open probability at all. The Mg2+ pops in and out of the channel pore in a so-called fast, flickering block, where it directly blocks the permeation of the other ions. In other words, the gate is opened normally by glutamate and stays open, but each Mg2+ sits in the pore for only microseconds before thermal energy knocks it out, current flows briefly, then the electric field drives another Mg2+ in. The current is therefore reduced by how much of the time the pore is plugged, not by the gate being shut. This is why, without Mg2+, the conductance of the channel has no voltage dependence--the open probability is set entirely by glutamate binding.

see image (be able to draw it): At +60 mV both currents are outward. The AMPA current is a brief outward spike lasting a couple of milliseconds, while the NMDA current rises slowly to a peak and decays over hundreds of milliseconds. At −60 mV the AMPA current is inward, with the same magnitude and time course as at +60 mV but opposite in direction, whereas the NMDA current is almost absent because Mg2+ blocks the pore at negative potentials. The AMPA current therefore only reverses direction with voltage, while the NMDA current is additionally suppressed at negative potentials.

Glutamate must be released from the presynaptic terminal, and the postsynaptic membrane must be depolarized to relieve the Mg2+ block. The NMDA receptor therefore acts as a coincidence detector, sensitive both to the activity of the presynaptic cell and to the state of the postsynaptic cell in which it is embedded. AMPA receptors, by contrast, conduct whenever glutamate is present, at any membrane potential. This coincidence detection, together with Ca2+ entry, underlies the role of the NMDA receptor in synaptic plasticity, because NMDA receptors are only active with coincident activity of the presynaptic and postsynaptic neuron, which then drives calcium-dependent synaptic plasticity.

Four genes, gria 1-4 (glutamate receptor ionotropic AMPA), encode the subunits GluA1, GluA2, GluA3 and GluA4. Most AMPA receptors contain the GluA2 subunit, which is what gives them a linear current-voltage relationship and no calcium permeability.

(In older papers these are called GluR1-4 or GluRA-D but not anymore)

They have a linear current-voltage relationship and no calcium permeability. Most AMPA receptors contain GluA2, typically as heteromers of GluA2 with GluA1 or with GluA3 (i.e. the receptor is not made up of just one type of subunit). So this is the standard AMPA receptor whose properties are described as voltage-insensitive. Some AMPA receptors lack GluA2, and these behave differently.

They show inward rectification, which means it is easier for current to flow into the cell than outwards. Current flows inward at negative potentials to depolarize the cell, but outward current at positive potentials is blocked by intracellular polyamines, so the I-V relation is linear below 0 mV and then flattens instead of rising (see image) They are also permeable to Ca2+, which gives them signalling roles. They can be homomers of a single subunit type or heteromers of any combination of GluA1, GluA3 and less commonly GluA4, as long as no GluA2 is present.

Which subunits a synapse puts into its AMPA receptors is tightly regulated, and changing that composition is one mechanism of synaptic plasticity and other signalling features. Because GluA2-lacking receptors are calcium-permeable and inwardly rectifying while GluA2-containing ones are neither, swapping subunits changes both the strength of the synapse and its calcium signalling.

A third family of ionotropic ligand-gated glutamate receptors, alongside AMPA and NMDA receptors. They are structurally more similar to AMPA receptors than to NMDA receptors. Their currents appear to be smaller and slower than AMPA receptor currents, and in some cases they have been reported at presynaptic specializations rather than postsynaptically. Much less is known about their function than about that of AMPA or NMDA receptors.

Five genes, grik 1-5 (glutamate receptor ionotropic kainate), encode the subunits GluK1-5 (older names are GluR5-7 and KA1-2).

Seven genes, grin1, grin2A-D and grin3A-B (glutamate receptor ionotropic NMDA), encode GluN1, GluN2A-D and GluN3A-B. GluN1 (aka NR1) is absolutely required for function, so at least one of the four subunits in every NMDA receptor is GluN1. GluN1 combines with other subunits, and the most common combinations in the mammalian brain are GluN1 with GluN2A and GluN1 with GluN2B.

GluN1 carries the binding site for the co-agonist glycine or D-serine, while the GluN2 subunits carry the glutamate binding site, so a receptor that responds to glutamate needs both. GluN1 is the one singled out as obligatory because every NMDA receptor contains it, whereas no individual GluN2 subunit is universal, and a small subset of NMDA receptors contain no GluN2 subunit (just GluN1 with a GluN3 subunit) and only open in response to glycine or D-serine alone.

Block strength is primarily determined by which GluN2 subunit is present, since GluN1 is in every NMDA receptor. Receptors containing GluN2A or GluN2B are very strongly blocked, with almost complete block at around −80 mV. Receptors containing GluN2C or GluN2D have a much weaker Mg2+ block, so a substantial fraction of their current still flows even at strongly hyperpolarized membrane potentials (see image --> the second plot keeps an inward current across the negative range instead of flattening towards zero). The degree of Mg2+ block therefore varies with subunit composition.

GluN2A-containing receptors give currents lasting approximately 100 ms. GluN2D-containing receptors are the slowest, with currents lasting around 1 s. GluN2B and GluN2C fall between these two with currents lasting a few hundred ms (~300 ms).

All of them show the slow activation and sustained current characteristic of NMDA receptors, and what varies is the degree. Because different cells and different synapses express different subunit combinations, there is a great deal of diversity in NMDA receptor time course.

Their functions are much less well characterized than those of GluN1 and GluN2. They may function presynaptically, or play other roles in signalling and synaptic plasticity.

They are 7-transmembrane G-protein-coupled receptors rather than ion channels. They signal through G proteins onto pathways such as phospholipase C and adenylate cyclase, with downstream effects including calcium signalling. Eight genes encode the metabotropic glutamate receptors. Their signalling functions are slower than those of the ionotropic glutamate receptors, which act on the millisecond timescale. Their roles are diverse and not yet fully explored.

The current flowing through postsynaptic receptors after release of an excitatory neurotransmitter, measured in voltage clamp. At glutamatergic synapses it flows through AMPA and NMDA receptors, which have a reversal potential of around 0 mV. Under physiological conditions the cell sits well below this, so the current is inward and depolarizes the cell towards threshold. An AMPA-mediated EPSC lasts about 2 ms.

An EPSP (excitatory postsynaptic potential) is the depolarization of the postsynaptic membrane produced by the excitatory synaptic current. An EPSC (excitatory postsynaptic current) is that current itself, flowing through the postsynaptic receptors. They are two aspects of one synaptic event: the current is what charges the membrane, and the potential is the membrane's response to being charged. The EPSP is much slower and longer-lasting than the EPSC, because its decay is set by the membrane time constant (which describes how Vm decays over time after a current pulse) rather than by the channels: e.g., a 2 ms AMPA-mediated EPSC gives an EPSP lasting around 20 ms.

A presynaptic neuron is stimulated to fire an action potential (for example by current injection) while a synaptically connected postsynaptic neuron is recorded from. The action potential invades the presynaptic nerve terminal and releases glutamate, and with a short delay of around 1 ms the postsynaptic glutamatergic response begins. To record the EPSC, the postsynaptic cell is held in voltage clamp: Vm is fixed, so the driving force (Vm − Erev) is constant and the recorded current follows the time course of the synaptic conductance directly. To record the EPSP, the cell is recorded in current clamp with the pipette injecting zero current, so that the synaptic current itself moves Vm and the trace shows the cell's natural response. The two cannot be recorded at once from one cell, since clamping the voltage abolishes the EPSP.

The EPSC is brief and fast, and is almost entirely dominated by the AMPA component, appearing through the voltage clamp amplifier as a rapid inward current lasting about 2 ms. The NMDA receptors are largely blocked at such negative potentials, because extracellular Mg2+ enters the pore, so at typical resting membrane potentials of around −60 mV the NMDA conductance is basically shut which is why the rapid transient AMPA current dominates.
===
Image labels the EPSC as I glutamate, the total current through both glutamate receptor types.

The AMPA component has similar amplitude and kinetics to that at negative potentials, except that the current is now outward rather than inward, since the reversal potential for both AMPA and NMDA receptors is close to 0 mV. In addition, the NMDA channels can now contribute, because Mg2+ has been expelled from the ion channel pore and no longer blocks it, so considerable outward current flows through the NMDA receptor. The summation of a fast AMPA component and a slower, longer-lasting NMDA component gives a large, long-lasting EPSC.

Some synapses are almost entirely dominated by AMPA receptors and others almost entirely by NMDA receptors, and at these the EPSC takes the shape of whichever component dominates: fast and brief where AMPA receptors dominate, slow and long-lasting where NMDA receptors dominate. In most synapses there is a fair mix, with both present in roughly even quantities in terms of the total synaptic currents, so both AMPA and NMDA receptors contribute importantly at most glutamatergic synapses.

The phospholipid bilayer is represented as a capacitor, between 0 mV (outside: reference) and Vm (inside). The individual ion channels as conductors (or equivalently resistors), each with a specific driving force relating to its reversal potential (represented as a battery with voltage at the Erev). Alongside the K+, Na+ and Cl− conductances that determine resting membrane potential, the AMPA and NMDA conductances are added, with reversal potentials close to 0 mV. These synaptic conductances then contribute to determining the membrane potential of the cell.

1. The capacitative component, the charging of the capacitor: starting from Q = CV and taking the derivative, the current equals Cm × dVm/dt.
2. The leak component, combining the K+, Na+ and Cl− conductances into a single term: (Vm − Eleak) × Gleak, where Eleak is simply the resting membrane potential of the cell in the absence of synaptic conductances (only taking into account the ionic conductances), typically around −70 mV.
3. The synaptic components, i.e. the ligand-gated AMPA and NMDA conductances, where the driving force is the membrane potential minus the 0 mV reversal potential (Vm - 0), so the current is simply Vm × Gsyn for each (so Vm*G_AMPA and Vm*G_NMDA). Note that G_NMDA is additionally voltage-dependent due to the Mg2+ block, unlike G_AMPA, which depends only on glutamate binding.

The total conductance of the cell, measured from the electrode as G = ΔI/ΔV. It is the sum of every conductance that is open at that moment, whatever its type. Given by G_input = G_leak + G_gated, where G_leak is the conductance of leak channels (always open) and G_gated is the conductance of voltage-gated and ligand-gated ion channels. When the gated channels are closed G_input ≈ G_leak.

The rising phase is driven by the synaptic current itself: the brief AMPA current, lasting about 2 ms, charges the postsynaptic membrane capacitance and causes a rapid depolarization.

The decay phase is determined by how fast that charge discharges the membrane capacitance through the leak conductances, governed by the membrane time constant (τ = Rm × Cm, where Rm = 1/G_input). In this case, G_input ≈ G_leak because the synaptic channels have already closed by the time the EPSP decays, and a typical unitary EPSP of around 1mV is too small to open voltage-gated channels (the membrane is still too hyperpolarized). However this assumption does not hold for large summated EPSPs in dendrites, which can recruit NMDA receptors and voltage-gated conductances causing G_gated =/= 0.

An AMPA-mediated EPSC lasts about 2 ms, while the same synaptic input gives an EPSP lasting around 20 ms. In voltage clamp the EPSC precisely follows the synaptic conductance, but the EPSP outlasts it because the charge that entered during the EPSC has charged the membrane capacitor, so the capacitor then must discharge through the leak conductances. The membrane time constant sets how fast that discharge (voltage decay) happens, and is typically on the order of many tens of ms for a neuron, so the EPSP is a much slower and longer-lasting event than the underlying conductance that drives it.
=======
Note on his terminology: he calls the membrane time constant the "conversion factor" between the EPSC and the EPSP. It sets the conversion because τ is the decay constant of the EPSP: once the synaptic current stops, Vm falls as e^(−t/τ), so the EPSP lasts a few τ regardless of how brief the current was. With τ = 10 ms, a 2 ms EPSC gives an EPSP lasting ~20 ms.

A postsynaptic potential caused by a single action potential in a single presynaptic neuron. Unitary EPSPs can differ greatly in amplitude, kinetics and trial-to-trial variability depending on the type of synapse. To measure them we trigger a single action potential in the presynaptic neuron and then measure the membrane potential of the postsynaptic neuron at the soma using a current clamp.

They can differ in all three of amplitude, kinetics and variability. In the example given, an action potential in the presynaptic cell evokes a large depolarization lasting 10s of ms with a decay time of 30 ms in one postsynaptic cell, and a response about ten times smaller with a faster decay time of 10 ms in the other. Also, over ten consecutive trials the large response is relatively constant when the trials are superimposed, whereas the small one shows enormous variability, with successes on some trials and failures on others (i.e. trials where the presynaptic action potential occurred but no vesicle was released, so there is no postsynaptic response at all). So the bigger response was more reproducible while the smaller response was more variable.

For large-amplitude uEPSPs, the presynaptic axon may make multiple synaptic contacts onto the same postsynaptic neuron (maybe like 10 different synapses) each of which can independently release glutamate. This increases the amplitude of the response and decreases its variability, because with many synapses each with about a 10% chance of releasing a vesicle, the chance that some of the synapses release simultaneously is high and the variance decreases.

For small-amplitude uEPSPs there might be just a single synaptic contact, so the presynaptic action potential either causes release or it does not, giving all-or-none transmission with failures on a trial-by-trial basis.

Thus, in order to evoke a postsynaptic action potential, we need the summation of many inputs arriving at roughly the same time.

They may correspond to release of vesicles that have less glutamate in them than other vesicles. This is a source of amplitude variation separate from failures, which are trials where no vesicle is released at all.

Most unitary synaptic connections in the brain give small-amplitude uEPSPs, typically around 1 mV, lasting some tens of milliseconds. Given a resting membrane potential of around −70 mV and an action potential threshold of around −40 mV, a single uEPSP causes only a small subthreshold depolarization. Unitary inputs onto most cells therefore do not cause postsynaptic action potential firing.

The summation of many inputs arriving at roughly the same time, probably requiring the convergence of many tens or hundreds of presynaptic neurons firing with some degree of synchrony. The integration is temporal, occurring on the timescale of the PSP, which is somewhere around 10 to 20 to 50 ms depending on how much postsynaptic depolarization there is and to what extent the slower NMDA receptors are activated. Only when the membrane potential at the axon initial segment crosses threshold is an action potential fired.

A synapse in the auditory brain stem where an individual axon comes and envelops the postsynaptic cell, containing a large number of vesicle release sites, forming maybe > 200 synapses. Because so much glutamate is released onto a postsynaptic cell with many receptors, a single presynaptic action potential automatically causes an action potential to fire in the postsynaptic neuron. It is therefore a specific case where one action potential is converted directly into an action potential in the postsynaptic cell, unlike most synapses in the brain.

At −65 mV the membrane is relatively hyperpolarized, the NMDA receptors do not get activated, and the EPSP is AMPA-driven.

As the cell is depolarized (up to around -55 mV i.e. before AP threshold), two things happen at once: the driving force for the AMPA conductance decreases, which by itself would give smaller-amplitude EPSPs, while NMDA conductances become unblocked and are recruited which would make the EPSP larger and longer-lasting. Since these two forces compete, the net effect depends on how many NMDA receptors are present at that synapse, so depolarization has quite different effects at different synapses--if more AMPA than NMDA, the EPSP would be smaller than at RMP, if more NMDA it might be longer and larger.

With further depolarization the amplitude drops further because the driving force for the AMPA receptor has decreased more, but now the cell is much closer to threshold, and once threshold is crossed an action potential is fired.
====
Example in image:
0 pA injected → −65 mV, EPSP driven by AMPA current. 100 pA → −55 mV, EPSP driven by both AMPA and NMDA currents: here the peak is smaller because the loss of AMPA driving force wins, but the decay is much longer than at RMP due to the NMDA component. 200 pA → −45 mV, the EPSP reaches threshold and an action potential is fired on top of it.

The postsynaptic specializations are typically distributed over the dendrites, with relatively few present on the actual soma. Since action potentials are initiated at the axon initial segment close to the soma, an input arriving in a distal dendrite has to travel some distance before it can contribute to firing, and it is filtered in space and time by the cable properties of the dendrites on the way. So the soma sees a slower and smaller amplitude version of the synaptic potential recorded in the dendrite.

Because dendrites are thin, around 1 μm in diameter, so the local membrane surface area is small. Less membrane means fewer leak channels, giving a high local membrane resistance, and less area to store charge, giving a small local capacitance. A given synaptic current therefore produces a much larger voltage change there than the same current would at the soma, and produces it faster, since there is little capacitance to charge and the charge also drains quickly away along the dendrite. This matters because the local depolarization can be large enough to unblock NMDA receptors and open voltage-gated channels, which add current on top of the AMPA current, so a distal input that would otherwise be too attenuated to depolarize the soma can still contribute to reaching threshold at the axon initial segment.

Locally, inputs onto the same stretch of dendrite that arrive at the same time add together there, in the dendrite itself, before anything propagates; inputs arriving at slightly different times may not summate at all. Simultaneous inputs can depolarize the dendrite enough to open NMDA receptors there. Globally, whatever survives the journey from every branch then adds together at the axon initial segment, and that is what determines whether threshold is crossed and an AP is fired in the postsynaptic cell.

Because of how small dendrites are: the high local resistance and low capacitance mean even a few simultaneous inputs produce a very large local depolarization. That depolarization relieves the Mg2+ block, so NMDA receptors open and add current that enlarges and prolongs the local EPSP, and may further recruit voltage-gated sodium conductances. Distal dendritic regions may therefore communicate with the soma largely through NMDA receptor activation, since without it their EPSPs would attenuate too much on the way and wouldn't be able to trigger an AP with just AMPA receptor currents.

Because the dendrite adds current that depends on how depolarized it already is. One input alone gives only an AMPA EPSP, but two arriving together depolarize the dendrite enough to unblock NMDA receptors, so the extra NMDA current makes the combined EPSP more than twice the size of a single one. The result therefore cannot be predicted by adding up the individual inputs.

At the axon initial segment, close to the soma, where the inputs from synapses across the whole dendritic arborization arrive summated and filtered. When the membrane potential there crosses threshold, an action potential is fired, and that is the output of the cell. Taking many postsynaptic potentials, integrating them at one site, and deciding whether or not to fire is the basic computation a neuron performs.

Because every sensory pathway relays information from the periphery to the neocortex through a chain of glutamatergic synapses, and conscious perception is thought to be driven by the activity of neocortical neurons. For touch, mechano-gated ion channels in the skin drive action potentials in sensory neurons, which release glutamate onto brainstem neurons, which relay to the somatosensory thalamus, which relays to the neocortex. For vision, photoreceptors release glutamate onto bipolar cells in the retina, which release glutamate onto retinal ganglion cells, which relay to the thalamus and then to visual cortex. Each station in these pathways is a glutamatergic synapse.

The mouse is attractive for neurophysiology because of its extensive genetics and widespread use in biological research. Mice are nocturnal and live underground in tunnels, so vision is not one of their most developed senses, and they instead gather sensory information about their immediate surroundings through an array of whiskers on the snout, arranged in a highly organized manner. Using their whiskers they can navigate tunnels, judge whether an aperture is large enough to enter, and detect the shape of objects. Their texture discrimination with the whiskers rivals that of the human fingertip.

Deflection of a whisker activates mechano-gated ion channels in trigeminal sensory nerve endings innervating the whisker follicle, initiating action potentials that propagate down the trigeminal nerve.

1. The trigeminal brainstem: the primary sensory neurons release glutamate onto brainstem neurons, EPSPs are generated and some of those neurons fire action potentials.

2. The somatosensory thalamus: thalamic neurons receive the glutamatergic input, postsynaptic potentials are generated and integrated, and action potentials are fired in subsets of those neurons.

3. The primary somatosensory neocortex, where whisker sensory perception is thought to take place. There are therefore three glutamatergic synapses relaying whisker information from the periphery to the neocortex.

They are laid out in a highly stereotypical pattern, with precise somatotopic mapping in the barrel cortex, the layout is the same in every mouse and also in rats. The whiskers are organized into rows, labelled A to E, and arcs, numbered 1 to 4. An individual whisker is named by its row and arc, so the C2 whisker is in row C, arc 2.

The barrel cortex is a somatotopic map in the primary somatosensory neocortex in which anatomical units are in direct one-to-one correspondence with the layout of the whiskers on the periphery, suggesting that sensory information relating to an individual whisker is primarily processed in the corresponding unit. Each unit is called a barrel, so the cortical representation of the C2 whisker is the C2 barrel. A barrel column is the tissue running through the depth of cortex beneath a single barrel, with a horizontal extent (width across the cortical surface) of around 250 μm. Anatomically identifiable representations of individual whiskers are also found at the brainstem and thalamic stations.

The neocortex is about 1.2 mm thick at the barrel cortex, and is divided by depth into layers 1 to 6. Layer 1 is the most superficial and has very few cells, being mainly composed of synapses, axons and dendrites. Most neocortical neurons are located in layers 2 to 6, with different cell densities in different layers. The main sensory input from the thalamus arrives in layer 4, which contains very small neurons.

Pyramidal cells are the main type of excitatory neuron in the neocortex. Layers 2 and 3 contain small pyramidal cells, layer 4 contains very small neurons, and the deeper layers contain much larger pyramidal cells with much more extensive dendrites. Although all of these are glutamatergic, their molecular, electrophysiological and synaptic properties differ across layers, so they are treated as different cell types. Laminar position (i.e. their position within the cortical layers) is therefore a useful way of subdividing excitatory neurons, because it predicts a lot about their synaptic connectivity. Intermingled with them are many different types of inhibitory GABAergic neurons.

The barrel map is obvious in anatomical sections, but in the living brain only the surface blood vessels are visible. We can map the location of a barrel column with intrinsic signal optical imaging, where red light is shone on the surface of the brain while a single whisker is stimulated. Neuronal activity in the corresponding localized region increases light absorption, because of increases in local blood flow, and reduces light scattering, because of changes in the optical properties of the tissue. Less light is therefore reflected from the region representing that whisker, localizing the barrel column with around 100 μm precision.

A fluorescent dye (DiI) is injected into the location identified by intrinsic signal optical imaging. Then, in postmortem analysis, horizontal sections of the brain are cut and stained, and the labelled spot is seen to lie inside the anatomical barrel map, corresponding to the barrel column of the stimulated whisker. The functional and anatomical maps therefore match.

The barrel column is located by optical imaging and marked with an injection of fluorescent dye, then brain slices are cut through the thickness of cortex for in vitro recording. The dye, together with the barrel walls (the dense cell borders outlining each barrel in layer 4), shows which part of the slice is the target column. Whole-cell recordings are then made simultaneously from several neurons within it. Current is injected through one recording electrode to make that neuron fire an action potential, while membrane potential is measured in all the others: those that respond with a uEPSP are synaptically connected to it, those that do not are not. Repeating this with each cell in turn gives a wiring diagram for the group (i.e. a directed graph with the recorded neurons as nodes and arrows showing which cell connects to which).

One out of five possible postsynaptic partners was connected, so cell 1 has a 20% probability of connection. Only a handful of the thousands of neurons in a barrel column can be recorded at once, so the actual wiring of any one column can never be mapped, and it differs between animals anyway. Instead, hundreds of pairs are recorded across many brains, all from the equivalent well-defined column, and the results are pooled to give the probability that a connection exists between neurons of two given layers (he says "two different laminar types" = same thing).

A matrix showing how often neurons in one layer connect to neurons in another. It is built from recordings of hundreds of pairs of neurons, with the presynaptic neuron classified by which layer it is in and the postsynaptic neuron ordered according to which layer its soma is located in. For example, in the image, neurons in layer 4 make connections with cells in layer 2,3,4,5A,5B, and a little bit in 6, but don't receive input from those layers and receive most of their input from other layer 4 neurons.

It tells us probabilistically where neurons in different layers are most likely to be getting their synaptic input from.

Layer 4 neurons send axons to and make synaptic connections with neurons in all other layers of the neocortex: layers 2, 3, 4, 5A, 5B and a little in layer 6. Postsynaptically, however, they receive almost no input from layers 2, 3, 5 or 6, and almost exclusively receive input from other layer 4 neurons. Layer 4 is therefore a tight local processing area whose neurons talk to each other and send signals onwards to the rest of the column, without receiving information back from those layers.

There is a strong diagonal component: neurons within layer 2, within layer 3, within layer 4, within 5A and within 5B preferentially connect to their neighbours, simply by being close to each other. Beyond that, superficial postsynaptic neurons in layers 2 and 3 receive their input mainly from presynaptic layers 2, 3 and 4. Deep neurons in layer 5 receive synaptic input from almost all other layers, including 2, 3, 4, 5 and a little from layer 6, so they act as integrators that collect information from the whole column and distribute further processed signals to other parts of the brain.

Around 10%. Excitatory neurons are therefore sparsely connected to each other, and the connection probability matrix shows this varies considerably by layer, with some layer-to-layer combinations reaching 20% or more and others close to zero.

Most synaptic connections are very small in amplitude, around 100 or 200 μV, so an action potential in a presynaptic neuron typically gives a small postsynaptic potential of only a few hundred μV. This is almost nothing compared with the roughly 30 mV of depolarization needed to make the postsynaptic cell fire. However, there are also rare synapses in the microcircuit that are much larger, around 5 mV (some even 10 mV) in response to a single action potential and so these synapses contribute very strongly to driving postsynaptic AP firing. The distribution is therefore heavily skewed: a very large number of weak connections and a sparse set of strong ones.

Because large connections have a considerable postsynaptic impact and contribute strongly to driving postsynaptic action potential firing, whereas the many small ones do not. Computational studies suggest that all of the small synapses can be removed without making much impact on neocortical microcircuit activity. Large connections are also more reliable: plotting the coefficient of variation (standard deviation divided by the mean of the trial-to-trial uEPSP amplitudes for a given connection) against amplitude shows that large connections have low variance while small connections are much more variable. Reliable sensory processing may therefore depend on a sparse set of connections that are both strong and reliable, with low variance from one trial to the next.

Partly genetic mechanisms, since different cell types express different genes and can be distinguished genetically, so genetic factors may skew the wiring diagram of the neocortex. Experience is also likely to make a big difference: we learn how to perceive the world around us, and some of that learning about how to interpret sensory information probably occurs in the wiring diagram of the primary sensory areas. The distribution of large and small connections may therefore partly reflect experience-dependent learning.

Synaptic plasticity is the process of changing synaptic strength, and it is thought to be where the changes that occur during learning largely take place, so long-lasting changes in synaptic strength are likely to encode memories.
Presynaptic plasticity (as we've discussed) dominates on the millisecond and second timescales. Long-term forms, the kind that might underlie memories lasting days or a lifetime, are thought to occur largely through postsynaptic plasticity: changes in the receptor composition of the postsynaptic density at excitatory glutamatergic synapses. These changes are largely induced by the NMDA receptor, and expressed as changes in AMPA receptor composition.

A long-lasting increase in synaptic strength, seen as an increase in the amplitude of excitatory postsynaptic potentials, induced by a brief period of activity at the synapse. It is expressed postsynaptically, mainly through recruitment of additional AMPA receptors into the postsynaptic density, and induced by strong activation of postsynaptic NMDA receptors. The increase can last hours in a slice recording and many days in vivo, which is why LTP is thought to be a cellular basis for memory.

Bliss and Lomo discovered it in the hippocampus in 1973. They first established a stable baseline EPSP response by stimulating axonal fibres at a low rate (every 10 or 20 seconds), and recorded the resulting EPSPs, which are primarily mediated by AMPA receptors. They then applied a brief burst of high-frequency stimulation to the same axons, and on returning to the baseline stimulation found that the PSP amplitude had dramatically increased.

In each case, a baseline is first established by stimulating the axons at a low rate (every 10-20 seconds, so 0.1-0.2 Hz) and recording the resulting EPSPs.

1. High frequency (100 Hz) stimulation of many presynaptic axons (most common): field stimulation electrodes stimulate hundreds of axons at once at 100 Hz for one second, possibly repeated a few times. With a presynaptic stimulus every 10 ms, the incoming EPSPs summate, giving strong postsynaptic depolarization and often postsynaptic action potential firing.

2. Injecting depolarizing current through the whole-cell recording pipette during synaptic stimulation: the low-rate baseline stimulation continues unchanged, and instead depolarizing current is injected to take the postsynaptic cell to a depolarized potential such as 0 mV, which unblocks the NMDA receptor so that the same presynaptic stimulation now induces long-lasting NMDA currents in addition to the fast AMPA currents.

3. Pairing postsynaptic action potential firing with EPSP input: the low-rate baseline stimulation again continues unchanged, and brief currents are injected into the postsynaptic cell to make it fire an action potential during an incoming EPSP, so the spike coincides with the synaptic input rather than occurring at rest.

They all induce strong activation of the NMDA receptor, which is the key event for most forms of postsynaptic LTP. Each works by producing strong postsynaptic depolarization at the same time as the presynaptic terminal is releasing glutamate. The depolarization comes either from the presynaptic side, via high-frequency presynaptic stimulation whose summating EPSPs depolarize the postsynaptic cell, or from the postsynaptic side, via depolarizing current injected into the postsynaptic cell or postsynaptic action potentials evoked by current injection. Either way, the depolarization removes the Mg2+ block from the NMDA receptor, and since glutamate is still bound, Ca2+ influx can then occur.

1. NMDA receptor activation. 2. Postsynaptic cytosolic calcium increase. 3. Activation of protein kinases (CaMKII).

Because AMPA receptors are not permeable to Ca2+, whereas NMDA receptors are, and it is the rise in postsynaptic cytosolic Ca2+ that activates the downstream protein kinases such as CaMKII. The voltage dependence matters too: because of the Mg2+ block, Ca2+ enters only when glutamate is bound and the postsynaptic cell is depolarized, so the NMDA receptor converts the coincidence of presynaptic and postsynaptic activity into a biochemical signal.

Calcium/calmodulin-dependent protein kinase II, a Ca2+-dependent protein kinase present at particularly high concentration in the postsynaptic density. Ca2+ entering through the NMDA receptor binds and activates it, and the activated kinase then adds phosphate groups to a variety of different targets, including AMPA receptors. It also acts on postsynaptic density proteins to increase the number of binding sites available for AMPA receptors, leading to recruitment of more AMPA receptors into the synapse. CaMKII activation is thought to be the most important signal leading to an increase in synaptic efficacy.

It increases the conductance of the AMPA receptor, so the same amount of glutamate binding lets more depolarizing current into the cell, which enhances the amplitude of the postsynaptic potentials. This is one of CaMKII's targets, but it is not the more important contribution to postsynaptic long-term plasticity.

The recruitment of additional AMPA receptors into the postsynaptic density. AMPA receptors are very mobile: they diffuse around in the two-dimensional plane of the plasma membrane at many locations across the dendritic arborization, but they become tightly bound when they reach the postsynaptic density. CaMKII increases the number of binding sites in the postsynaptic density, which allows more of these diffusing receptors to be captured and held in place at the synapse. e.g., where a synapse might previously have had 10 AMPA receptors, after LTP it might have 20.

Because the strong depolarization used to induce the plasticity was only temporary, but the increase in AMPA receptor number at the postsynaptic density persists. The cell is back at its resting potential with the NMDA receptors blocked again, yet the same glutamate release now activates more AMPA receptors, so there is more AMPA conductance postsynaptically and the postsynaptic potentials are larger. The synaptic connection is therefore stronger: the same presynaptic action potential now brings the postsynaptic cell closer to threshold and is more likely to make it fire.

A long-lasting decrease in synaptic strength, in which the baseline EPSP amplitude (measured at low-rate stimulation before and after induction) is made smaller in a stable way across many hours or days. It is the converse of LTP: where LTP adds AMPA receptors to the postsynaptic density, LTD removes them. A process like this is necessary because otherwise synapses could only ever be strengthened, and every synapse in the brain would grow without limit, which clearly isn't feasible.

In each case, a baseline is first established by stimulating the axons at a low rate (around every ten seconds, i.e. 0.1 Hz) and recording the resulting EPSPs.

1. Repetitive 1 Hz stimulation of many axons (100x lower frequency than for LTP induction): the same axons are stimulated constantly at 1 Hz for many minutes (maybe 5-10 min) after which the low-rate baseline stimulation is resumed.

2. Injecting weak depolarizing current through the whole-cell recording pipette during synaptic stimulation: the low-rate baseline stimulation continues unchanged, and instead weak depolarizing current is injected to take the postsynaptic cell to an intermediate potential such as −50 mV (much less than the strong depolarization of ~0mV for LTP). This intermediate potential gives some degree of NMDA receptor unblock, but nowhere near as much as at 0 mV, so the NMDA receptors are only partially activated.

The requirements are NMDA receptor activation, postsynaptic cytosolic calcium increase, and activation of protein phosphatases, in particular the protein phosphatase calcineurin. Phosphatases do the opposite of kinases: where a kinase adds a phosphate group to a protein, a phosphatase removes it.

What is less clear is whether Ca2+ is the whole story: the NMDA receptor may have additional signalling capabilities, and protein-protein interactions between NMDA receptors and other signalling proteins in the postsynaptic density may be what drives the expression of LTD.

Calcineurin appears to directly dephosphorylate AMPA receptors, which are then removed from the postsynaptic density, at least in part. A synapse that started with many postsynaptic AMPA receptors ends up with fewer, so there is less AMPA conductance and the postsynaptic potentials are smaller.

Both require postsynaptic depolarization and NMDA receptor activation, but LTD needs much less of each: around −50 mV rather than 0 mV, giving only partial NMDA receptor activation. Downstream, both require a rise in postsynaptic cytosolic Ca2+, but LTP activates protein kinases (CaMKII) while LTD activates protein phosphatases (calcineurin), which do the opposite of kinases. The expression is also opposite: LTP phosphorylates AMPA receptors and recruits more of them into the postsynaptic density, while LTD dephosphorylates them and removes them. They are not direct reversals of each other, though: each uses its own machinery, so removing AMPA receptors is an active process in its own right rather than simply undoing what LTP did.

STDP is a form of synaptic plasticity controlled by the precise temporal relationship between action potential firing in a presynaptic neuron and action potential firing in the postsynaptic neuron, discovered by Henry Markram. It is more physiological than the sustained depolarizations and prolonged stimulus trains used to induce LTP and LTD, since neurons really do fire action potentials with precise timing, whereas the other protocols may not occur normally inside the brain. It can induce both synapse potentiation and weakening, and it has been found at many synapses, so it appears to be a general rule.

A synaptically connected pair of neurons is recorded. Current is injected into the presynaptic neuron to make it fire first, releasing glutamate and evoking a uEPSP in the postsynaptic cell. Then, 10-20 ms later, current is injected into the postsynaptic cell to make it fire an action potential, so that spike lands on top of the EPSP that is still in progress. This pre-before-post order is described as a positive delta-T, where delta-T is the postsynaptic spike time minus the presynaptic spike time. The pairing is repeated many times (maybe 100 times) to produce a robust increase in the efficacy of the connection between the two neurons, and afterwards the baseline EPSP has grown.

The same paired recording and pairing protocol as for potentiation, but with the order of the two spikes flipped: the postsynaptic neuron is made to fire first, and only a few milliseconds later is the presynaptic neuron made to fire and release glutamate, so the postsynaptic spike is over before the EPSP arrives rather than riding on top of it. After the pairing is repeated many times, the baseline EPSP has shrunk rather than grown.

Delta-T is the timing difference between the two spikes: the postsynaptic spike time minus the presynaptic spike time. A positive delta-T means the postsynaptic cell fires after the presynaptic cell, which produces potentiation. A negative delta-T means the postsynaptic cell fires first, which produces depression. The sign of delta-T, i.e. the order of firing, therefore determines the direction of the plasticity.

(be able to draw this!!!)
The x axis is delta-T (the ISI between the postsynaptic and presynaptic spikes) and the y axis is the change in EPSP amplitude, with more positive values indicating stronger potentiation and more negative values indicating stronger depression. Potentiation occurs at positive delta-T, where the postsynaptic cell fires after the presynaptic cell, and is greatest when that delay is only a few milliseconds. Depression occurs at negative delta-T, where the postsynaptic cell fires first, and is likewise greatest at the shortest intervals. Both decay towards zero as the interval grows, so the curve jumps across delta-T = 0 and then falls off on each side. Potentiation is confined to a window of roughly 10-20 ms, while depression extends out to perhaps 50-100 ms.
====
His note: what is remarkable is that which cell fires first, decided by only a few milliseconds, is what determines whether the synapse is strengthened or weakened

Extremely sensitive. Moving the postsynaptic spike from around 10 ms after the presynaptic spike to around 10 ms before it is enough to reverse the sign of the plasticity, turning potentiation into depression. The order of action potential firing on a millisecond timescale therefore makes a very big difference to the type of plasticity produced.

That the synapse strengthens inputs that appear to be useful and weakens ones that do not. If the EPSP arrived just before the postsynaptic cell fired its action potential, that input plausibly helped drive the cell to threshold, so it is worth keeping and is strengthened. If the postsynaptic cell had already fired before the glutamate (and subsequent EPSP) arrived, that postsynaptic action potential cannot have been caused by this input, so the input appears not to contribute to the cell's firing and is weakened.

Two things: how strongly the postsynaptic cell is depolarized, and whether that depolarization coincides in time with glutamate release from the presynaptic cell.

The synapse is strengthened when the postsynaptic cell is strongly depolarized during the incoming EPSP, for example when a postsynaptic action potential fires on top of it. This works at least in part by recruiting diffusing AMPA receptors, which bind to postsynaptic density proteins and are held there. The synapse is weakened when the postsynaptic cell is only weakly depolarized during the EPSP, or when it fires at a time unrelated to when the synaptic input arrives. Useful inputs are therefore strengthened and inputs that do not contribute to postsynaptic firing are weakened.

Dendritic spines are small protrusions, about 1 μm in size, that grow out from a dendrite to make excitatory synapses. For most cells the excitatory glutamatergic inputs arriving across the dendritic arborization do not contact the main parent dendrite directly, but instead contact these spines, which reach out to make synaptic contact with a presynaptic bouton that releases glutamate. Each spine receives at least one glutamatergic input.

By counting the spines, since in mature pyramidal cells almost all glutamatergic synapses arrive on spines rather than on the parent dendrite. A typical nerve cell has on the order of 1,000 spines, so this gives a first-order approximation of around 1,000 glutamatergic inputs.

A whole-cell recording electrode containing a fluorescent dye is introduced onto a neuron. The dye fills the cell, and a fluorescence image then shows the soma and the dendritic arborization with much higher contrast than the ordinary video image. The spines appear as small spots decorating the sides of the dendrites.

Because light microscopy cannot resolve them, having a resolution limit set largely by the wavelength of light. Fluorescence imaging is enough to see where spines are and to count them, but the soma (about 10 μm across) and dendrites (1-2 μm across) are the only things clearly resolved, and spines appear only as fuzzy protrusions with no visible internal structure. Electron microscopy resolves the spine neck, the spine head, the postsynaptic density and the spine apparatus. The tradeoff is that EM requires fixed tissue that has been stained, dehydrated and embedded in resin, so it gives one static snapshot, whereas fluorescence imaging works in living tissue and allows the same spine to be followed over time.

A narrow stalk called the spine neck, which connects the spine to the parent dendrite, and a head at the far end, which carries the synapse. The parent dendrite is about 1 μm in diameter, while the neck is much narrower, 100-200 nm across, and 1-2 μm long. The head is around 1 μm in diameter on the largest spines (mushroom spines).

Spines form a continuum, differing mainly in the head. Mushroom spines have a large head (about 1 μm in diameter) and typically carry large synapses. Other spines are very thin with no obvious head, staying at 100-200 nm along their whole length like filopodia, and typically carry very small synapses. In between there are small, stubby spines.

A very dense matrix of proteins at the postsynaptic membrane, directly opposite the presynaptic bouton, at the tip of the dendritic spine. It appears as a dark region because the heavy metal stains used for EM bind proteins, so a dense concentration of protein scatters many electrons out of the beam. Besides the ligand-gated ion channels themselves (the AMPA and NMDA receptors), the PSD contains molecules that anchor those receptors in position, including PSD95, and a host of signalling molecules.

Postsynaptic density molecule with 95 kilodaltons (its molecular mass), one of the most highly expressed proteins in the postsynaptic density. It is a scaffolding molecule that links to the ligand-gated ion channels and, together with other signalling molecules, holds them in the right place, i.e. it holds them directly opposite the presynaptic specialization from which glutamate is released, so that glutamate reaches them at high concentration.

Because synaptic plasticity requires receptors to be moved in and out of the synapse: AMPA receptors diffusing on the plasma membrane must be recruited into the synapse for LTP, and removed from it for LTD. Holding the relevant signalling molecules at the postsynaptic density places them exactly where they are needed, right next to the AMPA and NMDA receptors they act on.

A specialization of the endoplasmic reticulum found inside some but not all spines, located in the spine neck region. Its two functions may be:

1. Local protein synthesis and delivery, so some proteins may be made within the spine itself.

2. Handling calcium: as an intracellular calcium store it can take up or release Ca2+, so it may shape the calcium signal that follows NMDA receptor activation. Both functions could be important for synaptic plasticity.

1. They act as isolated compartments, both biochemical and electrical, because the narrow spine neck restricts what passes between the spine head and the parent dendrite. This keeps signalling at one synapse separate from its neighbours on the same dendrite.

2. They provide flexibility for rewiring: many axons pass within one or two micrometers of any given dendrite, so by growing a small protrusion out to one of them a neuron can gain an input from a cell it was not connected to, and by retracting a spine it can lose an input it no longer needs, reconfiguring how the network functions.

The spine neck is only around 100 nm across, so it acts as a diffusional barrier and signalling molecules cannot easily leave the spine head. Calcium entering through NMDA receptors is the key example. The spine head is such a small volume that only a few calcium ions are needed to raise the local concentration dramatically. That calcium then struggles to escape: the narrow neck restricts its diffusion, calcium-binding molecules bind it so that less of it is free to diffuse, and the spine apparatus can take it up into the ER. Since each spine carries its own synapse, a calcium signal confined to one spine is confined to one synapse. This is what allows plasticity to be induced at that synapse without affecting its neighbours on the same dendrite.

The narrow spine neck may also offer considerable electrical resistance, measured at some synapses at around 1 gigaohm. Such a resistance makes it relatively difficult for synaptic current to flow from the spine head to the parent dendrite, so the membrane potential at the spine head may be quite different from that at the dendrite. Spines with a high neck resistance might therefore have larger and longer-lasting EPSPs in the head than spines with a low neck resistance, which would relieve more of the Mg2+ block and enhance the contribution of NMDA receptors, and so plausibly make that synapse more readily potentiated.

Staining a neuron for microtubules (MAP2) and for actin shows that the main parent dendrite is full of microtubules while the spines are full of actin. This makes spines highly motile (unlike the much more stable parent dendrite) since actin readily polymerizes and depolymerizes, which is what allows cells of all kinds to move: filaments can lose actin at one end and add it at the other in a treadmilling fashion, pushing the membrane around. This actin polymerization also is thought to underlie the formation of new dendritic spines.

Actin polymerizes and pushes the plasma membrane out from the dendrite towards a nearby presynaptic bouton, beginning the process of forming a new synapse. This happens under conditions that induce LTP (i.e. high frequency presynaptic stimulation), so new spines can appear on stretches of dendrite that had none before. The new spine starts thin and filamentous, with perhaps only one or two glutamate receptors, so it is more an exploratory synapse with little impact on the soma. If it turns out to be useful and is further strengthened by LTP, more actin polymerization enlarges it into a mushroom spine with a large postsynaptic density and very many glutamate receptors, which has a major impact.

Because many axons pass within one or two micrometers of any given dendrite. By growing a small protrusion out to one of them, a neuron can start receiving input from a cell it was not connected to, and by retracting a spine it can lose an input it no longer needs. Since spine growth and removal are driven by activity, this rewiring happens in an experience-dependent way, reconfiguring which cells talk to which.

Take one presynaptic axon whose bouton sits close to dendrites from two different postsynaptic neurons, call them cell A and cell B. At first, only cell A has grown a spine out to that bouton, so an action potential in the axon evokes a postsynaptic potential in cell A and nothing in cell B. Then the animal learns something, and the connection to cell A is no longer useful: cell A's spine retracts and disappears. Cell B grows a spine of its own out to the same bouton and forms a synapse there. Now an action potential in the axon evokes a postsynaptic potential in cell B and nothing in cell A. So the wiring diagram has changed in an experience-dependent manner.

Spines are continually appearing and disappearing at a certain rate. A new spine appearing means a new synaptic connection has been formed, and a spine disappearing means a synaptic connection has been lost. During learning experiences the rate of this turnover increases dramatically.
Recall the approximate values for:
-Glutamate: vesicle concentration, cleft peak concentration and duration, fraction of synapses
-AMPA and NMDA receptors: gene numbers, single-channel conductances, reversal potentials, kinetics after a 1 ms pulse, Mg2+ concentration and block threshold
-NMDA subunit differences: Mg2+ block strength, current durations for GluN2A/2B/2C/2D
-EPSCs and EPSPs: synaptic delay, AMPA EPSC duration, EPSP duration, membrane time constant
-uEPSPs: typical amplitude, amplitude range in the barrel column, depolarization needed to fire
-Barrel cortex: cortical thickness, barrel column width, imaging precision, connection probability, number of relay synapses
-LTP/LTD/STDP: induction frequencies, depolarization levels, pairing number, STDP windows, AMPA receptor number change

Glutamate: Vesicle ≈100 mM; cleft ≈1 mM for ≈1 ms; ≈80% of all synapses
Receptors: 4 gria genes (GluA1 to 4), 7 grin genes (GluN1, 2A to D, 3A to B), 5 grik genes (GluK1 to 5), 8 mGluR genes; AMPA ≈5 pS, NMDA ≈50 pS; both Erev ≈0 mV; AMPA ≈2 ms, NMDA peaks ≈20 ms and lasts 100s of ms; Mg2+ ≈1 mM, block below ≈−40 mV
NMDA subunits: GluN2A/2B almost fully blocked at ≈−80 mV, GluN2C/2D much weaker; GluN2A ≈100 ms, GluN2B/2C ≈300 ms, GluN2D ≈1 s
EPSC/EPSP: Synaptic delay ≈1 ms; AMPA EPSC ≈2 ms → EPSP ≈20 ms; τm 10 to 20 ms (notes: many tens of ms); PSP timescale for summation 10 to 50 ms
uEPSPs: Typically ≈1 mV lasting tens of ms; barrel column most ≈100 to 200 μV, rare ones ≈5 mV (up to 10); ≈30 mV of depolarization needed to fire (rest −70, threshold −40)
Barrel cortex: Neocortex ≈1.2 mm thick; barrel column ≈250 μm wide (≈200 x 300 μm); optical imaging ≈100 μm precision; excitatory connection probability ≈10% (up to 20%+ for some layer pairs); 3 glutamatergic relay synapses periphery→cortex
Plasticity: LTP: 100 Hz for 1 s, depolarize to ≈0 mV; LTD: 1 Hz for 5 to 10 min, depolarize to ≈−50 mV; baseline stimulation every 10 to 20 s; STDP pairing ≈100x, potentiation window 10 to 20 ms, depression window up to 50 to 100 ms; AMPA receptors ≈10 → ≈20 after LTP
Spines: ≈1 μm long; neck 100 to 200 nm wide, 1 to 2 μm long; mushroom head ≈1 μm, giving a volume of ≈1 μm³ = 1 fl (most spines considerably smaller); ≈1,000 per pyramidal cell; neck resistance ≈1 GΩ; axons within 1 to 2 μm of any dendrite; PSD95 = 95 kDa