Synaptic Integration

Synaptic integration is the process by which a neuron combines many postsynaptic potentials — excitatory (EPSPs) and inhibitory (IPSPs) — arising at different synapses and different times into a single net change in membrane potential at the site of spike initiation. Because individual PSPs are small and graded, no single synaptic event normally drives a neuron to threshold. Instead, the neuron's membrane acts as a leaky integrator: it sums incoming charge, but that charge continuously bleeds away through resting ion channels and spreads out along resistive dendrites, so summation is a race between arriving inputs and passive decay. Two axes structure this summation. Spatial summation combines PSPs generated at different synaptic locations that overlap in time — for example, several dendrites receiving input within the same few milliseconds. Temporal summation combines PSPs generated at the same (or nearby) synapse by successive presynaptic action potentials arriving close enough together that the depolarization from one has not fully decayed before the next arrives. Real neural computation almost always involves a blend of both: spatiotemporal patterns of input across a dendritic tree. The site where this running sum is "read out" is the axon hillock, together with the adjacent axon initial segment (AIS). This region has the lowest threshold for action potential generation in the neuron because it possesses the highest density of voltage-gated Na⁺ channels (particularly the Nav1.6 subtype in many mature mammalian neurons) of any part of the cell, along with a high density of Kv7 (KCNQ) channels that stabilize threshold. Because dendritic and somatic depolarizations spread passively (electrotonically) toward the axon, and because the AIS is the most excitable patch of membrane, it functions as the decision point: whichever pattern of synaptic input manages to depolarize the AIS past threshold at a given moment triggers a spike that is then propagated actively down the axon according to cable properties and regenerative Na⁺ current, without further decay.

Step 1 — Generation of local postsynaptic potentials. At each activated synapse, neurotransmitter binding opens ligand-gated ion channels, producing a local conductance change. At excitatory glutamatergic synapses, AMPA and NMDA receptors admit Na⁺ (and Ca²⁺ through NMDA receptors) driving the local membrane potential toward roughly 0 mV, producing a depolarizing EPSP. At inhibitory GABAergic or glycinergic synapses, Cl⁻-permeable channels open; because the Cl⁻ equilibrium potential in mature neurons is typically near or slightly below rest (roughly −65 to −75 mV), this produces a hyperpolarizing or "shunting" IPSP. Each PSP begins as a local change in conductance and voltage confined to a small patch of membrane near the synapse.

Step 2 — Passive spread governed by cable properties. The synaptic current then spreads electrotonically through the dendrite toward the soma, obeying the same passive cable equations that describe subthreshold spread in axons. Two parameters dominate:

• Length constant (λ) — the distance over which a steady voltage decays to 1/e (≈37%) of its initial value, set by λ = √(rm / ri), where rm is membrane resistance per unit length and ri is axial (internal) resistance per unit length.
• Time constant (τ) — how quickly a voltage change rises and decays at a point, set by τ = rm·cm (membrane resistance times membrane capacitance).

A synapse on a thin, distal dendritic branch, far from the soma, produces a much smaller and slower somatic voltage deflection than the same synaptic conductance placed near the soma, purely due to cable filtering — this is why dendritic geometry is itself a computational variable, not just wiring.

Step 3 — Spatial summation at the soma/axon hillock. If two or more synapses are activated within roughly one membrane time constant (τm ≈ 10–20 ms in typical central neurons) of each other, their depolarizations arrive at the soma overlapping in time and add together, algebraically, on the membrane. For pure subthreshold summation of small EPSPs this addition is close to linear; but as the net depolarization grows, several nonlinearities appear:

• Driving-force reduction: as Vm depolarizes toward the reversal potential of the synaptic current (~0 mV for AMPA), the electrochemical driving force (Vm − Erev) shrinks, so each additional EPSP contributes less depolarization than the first — summation becomes sublinear.
• Conductance overlap / shunting: when synapses are close together and active simultaneously, the local increase in membrane conductance from one synapse reduces the input resistance "seen" by the other, again producing sublinear summation.
• NMDA receptor supralinearity: because NMDA receptor channels are blocked by extracellular Mg²⁺ at hyperpolarized potentials, coincident depolarization (from summing EPSPs) can relieve this block, unleashing additional inward current — an example of summation being supralinear, actively boosted rather than merely arithmetic.

Step 4 — Temporal summation. A single presynaptic terminal firing repetitively delivers a train of EPSPs. If the interspike interval is shorter than the decay time course of the EPSP (which itself approximates τm), each new EPSP begins from an already-elevated baseline, so successive potentials "stack." The steady-state summed depolarization for a train at frequency f approaches EPSP amplitude × [1 / (1 − e^(−1/(f·τm)))] for an idealized exponentially-decaying PSP train — in practice this means high-frequency presynaptic firing is far more effective at driving a postsynaptic cell to threshold than the same number of spikes delivered slowly.

Step 5 — Inhibition's dual role. Inhibitory input can act in two computationally distinct ways:

• Subtractive (hyperpolarizing) inhibition: an IPSP whose reversal potential is well below rest and below threshold directly opposes depolarization by pulling Vm away from threshold — this is a simple algebraic subtraction from the summed EPSP.
• Shunting (divisive) inhibition: an IPSP whose reversal potential is close to rest produces little voltage change on its own but massively increases membrane conductance. Because voltage change for a given synaptic current is inversely proportional to the total conductance (ΔV = I·Rm, and Rm falls as conductance rises), shunting inhibition divides the effectiveness of concurrent excitatory input rather than subtracting a fixed amount. This is especially powerful when the inhibitory synapse sits between an excitatory synapse and the soma (on the same dendritic path), acting as a conductance "gate" — a phenomenon called on-path shunting inhibition, first analyzed quantitatively by Wilfrid Rall.

Step 6 — Readout at the axon hillock/AIS. All these spatially distributed, temporally overlapping contributions converge electrotonically on the axon hillock. If, at any instant, the net depolarization there exceeds spike threshold — typically around −55 to −50 mV in many central neurons, roughly 10–20 mV above rest — the high density of Nav1.6 channels at the AIS triggers a regenerative, all-or-none action potential. This spike then propagates actively (non-decrementally) down the axon, governed by the Hodgkin-Huxley mechanism and axonal cable properties, while it may also propagate backward into the soma and dendrites (the backpropagating action potential), which itself can influence subsequent integration by triggering voltage-gated Ca²⁺ influx and modulating synaptic plasticity. The whole system is best thought of as a leaky, spatially extended, partially nonlinear integrator: charge is continuously summed, continuously leaking away, differentially filtered by location, and read out where excitability is highest.

The founding demonstration that postsynaptic potentials are graded and summate came from John Eccles, Paul Fatt, and Bernard Katz in the early 1950s, using intracellular microelectrode recordings from cat spinal motoneurons — work that grew directly from Eccles's collaboration with Katz and helped earn Eccles a share of the 1963 Nobel Prize in Physiology or Medicine (with Hodgkin and Huxley).

What they did. Eccles and colleagues inserted fine glass microelectrodes directly into the cell bodies of spinal motoneurons in anesthetized cats, then electrically stimulated afferent nerves (notably group Ia sensory fibers from muscle spindles, which make monosynaptic excitatory connections onto motoneurons — the circuit underlying the stretch reflex). By recording the motoneuron's membrane potential while varying the number of afferent fibers stimulated, the strength of stimulation, and the timing between stimuli, they could directly observe the postsynaptic response to controlled synaptic input.

What they observed. 1. A single afferent volley produced a small, graded depolarization — the excitatory postsynaptic potential (EPSP) — typically far too small (often under 1 mV) to reach threshold on its own. 2. Increasing the number of afferent fibers stimulated (recruiting more synapses simultaneously) produced progressively larger depolarizations that added together — direct evidence of spatial summation — and eventually crossed threshold to trigger a spike. 3. Delivering two stimuli to the same afferent pathway in rapid succession produced a second EPSP that arose from an already-elevated baseline left by the first, summing to a larger total depolarization than either alone — direct evidence of temporal summation. As the interval between stimuli increased beyond roughly the membrane time constant, the summation diminished, tracking the passive decay of the first EPSP. 4. Stimulating inhibitory (Ia inhibitory interneuron-mediated, or direct antagonist-afferent) pathways produced a hyperpolarizing inhibitory postsynaptic potential (IPSP), and — critically — combining an IPSP with a concurrent EPSP reduced the net depolarization below the arithmetic sum expected from linear addition of independently-measured EPSP and IPSP amplitudes, revealing conductance-based (shunting) interactions rather than pure algebraic subtraction. 5. By systematically varying stimulus strength and using the relationship between measured EPSP size and the known number of afferent fibers activated, they showed the postsynaptic response scaled with the number of active synapses in a graded, summing manner — not in discrete jumps corresponding to all-or-none synaptic units, and not through some independent regenerative process at each synapse.

What it proved. This body of work established, for the first time with direct intracellular evidence, that:

• postsynaptic potentials are graded (not all-or-none) and decremental,
• multiple postsynaptic potentials combine algebraically (with measurable deviations from perfect linearity),
• summation occurs both across different synapses active simultaneously (spatial) and across repeated activation of the same synapse over time (temporal),
• a spike is generated only when the summed depolarization reaches a fixed threshold, consistent with a genuine integrate-and-fire-like decision process rather than each synapse independently able to trigger a full response.

A complementary and mechanistically deeper contribution came later from Wilfrid Rall, who developed the compartmental cable-theory models of dendrites (from the late 1950s through the 1960s) that explained why summation deviates from simple linear addition — predicting sublinear summation from driving-force and conductance interactions, and specifically predicting the powerful, nearly divisive effect of inhibitory synapses placed on the path between an excitatory synapse and the soma. Rall's models were later confirmed experimentally by dual and multi-site intracellular/patch recordings (from the 1980s onward, e.g., work by Nicholas Spruston, Greg Stuart, and Bert Sakmann using dendritic patch-clamp recordings in cortical pyramidal neurons) that directly visualized EPSP attenuation with distance and confirmed the AIS/axon hillock as the lowest-threshold, spike-initiating zone.

Example 1 — Quantifying spatial and temporal summation with numbers. Suppose a cortical pyramidal neuron has:

• resting potential Vrest = −70 mV
• spike threshold Vth = −55 mV (so 15 mV of net depolarization needed)
• a single EPSP at the soma has peak amplitude 0.5 mV and decays with time constant τ = 15 ms
• synapse "X" fires four times at 100 Hz (10 ms intervals)

Temporal summation calculation. For an idealized EPSP decaying exponentially and stacking linearly (a reasonable first approximation for small EPSPs), the peak after the nth input is approximately the sum of the residual amplitudes of all previous EPSPs plus the new one:

• EPSP 1 at t=0: amplitude 0.5 mV, decays as 0.5·e^(−t/15)
• At t=10 ms, EPSP1 has decayed to 0.5·e^(−10/15) = 0.5 × 0.513 ≈ 0.26 mV. EPSP 2 adds 0.5 mV → running total ≈ 0.76 mV.
• At t=20 ms, EPSP1 residual: 0.5·e^(−20/15) ≈ 0.13 mV; EPSP2 residual (10 ms old): 0.5·e^(−10/15) ≈ 0.26 mV; EPSP3 adds 0.5 mV → total ≈ 0.89 mV.
• At t=30 ms, residuals from EPSP1 (≈0.07), EPSP2 (≈0.13), EPSP3 (≈0.26 mV), plus EPSP4 (0.5 mV) → total ≈ 0.96 mV.

So four EPSPs at 100 Hz from a single synapse sum to under 1 mV — nowhere near the 15 mV needed for threshold. This models why a single active synapse, even firing rapidly, essentially never fires a cortical neuron alone. Adding spatial summation. Now suppose 40 similar synapses are activated together within a 5 ms window (well within τ), each contributing an average of 0.5 mV (assuming near-linear summation for this small a depolarization, i.e., ignoring saturation for this estimate): Net depolarization ≈ 40 × 0.5 mV = 20 mV (if purely linear) — which would exceed the 15 mV needed and trigger a spike. In reality, sublinear summation (driving-force reduction, dendritic conductance loading) typically reduces the effective sum to roughly 60–80% of the linear prediction for this many coincident inputs, giving an effective ≈12–16 mV — placing the neuron right at the threshold boundary. This illustrates the essential point: spatial coincidence across dozens of synapses, not repetitive firing of one synapse, is what typically drives cortical neurons to threshold, and the neuron operates close enough to threshold that modest changes in coincident synapse number determine whether or not it fires.

Example 2 — Shunting inhibition changing the outcome of a clinical/experimental scenario. Consider an experiment (conceptually parallel to dendritic recording studies) on a CA1 hippocampal pyramidal neuron. Two excitatory synapses on the same apical dendrite, each producing a 4 mV EPSP at the soma when activated alone, are stimulated simultaneously, driving the soma from rest (−70 mV) to −62 mV — 8 mV of net depolarization (slightly sublinear vs. the naive 8 mV expected from simple addition of 4+4, here assumed to still be additive for simplicity), just shy of the −60 mV threshold in this cell. Now an inhibitory synapse, mediated by GABA_A receptors with a reversal potential of −65 mV (close to rest), is activated on the dendritic segment between the two excitatory synapses and the soma — an on-path shunting configuration — at the same time as the two EPSPs.

• If this IPSP were purely subtractive (as if it summed algebraically like the EPSPs), one might predict the net effect is small, since −65 mV is close to Vrest and the "driving IPSP amplitude" alone is under 1 mV.
• But because the GABA_A conductance increase is large and positioned on the path the excitatory current must cross to reach the soma, it substantially increases local membrane conductance right where the excitatory current is flowing. This shunts (short-circuits) a large fraction of the excitatory current to ground before it reaches the soma.
• Empirically (as in Rall's models and confirmed by dendritic recordings), on-path shunting inhibition of this kind can reduce the effective somatic depolarization from the two EPSPs by 50% or more — turning an 8 mV depolarization into perhaps 3–4 mV — even though the IPSP's own voltage deflection at the soma is barely measurable.

Clinical/interpretive reasoning: this explains why some inhibitory synapses are so much more powerful at silencing a neuron than their own voltage amplitude suggests, and why loss of this specific inhibitory geometry (e.g., through interneuron dysfunction, as hypothesized in some models of epilepsy) can produce hyperexcitability disproportionate to the apparent loss of "inhibitory drive" measured as a simple IPSP amplitude — the deficit is a loss of divisive gain control, not merely a loss of a subtractive offset.

Sir John Carew Eccles (1903–1997) was an Australian neurophysiologist whose career bridged the correlational, indirect neurophysiology of the early 20th century and the direct, quantitative intracellular recording era that transformed cellular neuroscience. He trained under Charles Sherrington at Oxford — inheriting Sherrington's concept of the synapse as an integrative junction — but for years argued (against Sherrington's own later leanings and against Henry Dale) that synaptic transmission was primarily electrical, not chemical. Confronted with his own increasingly precise intracellular data from cat motoneurons in the early 1950s, gathered with Paul Fatt and (in related work) Bernard Katz, Eccles changed his position — a celebrated case of a scientist's own experiments overturning his prior theoretical commitment. The technical breakthrough that made this possible was the glass capillary intracellular microelectrode, fine enough to penetrate a single neuron's soma without killing it, allowing direct measurement of resting potential, EPSPs, IPSPs, and spikes from an identified cell for the first time in the central nervous system.

Understanding synaptic integration underlies practical work across medicine, neuroscience research, and computational technology.

Epilepsy and excitability disorders. Many antiepileptic strategies target the balance of excitatory and inhibitory summation directly — GABA_A receptor-potentiating drugs (benzodiazepines, barbiturates) enhance shunting inhibition to raise the effective threshold for spatial/temporal summation to reach spike threshold; loss-of-function mutations in inhibitory interneuron populations or in dendritic ion channels (e.g., Nav or Kv channelopathies affecting AIS excitability) are studied as mechanisms of hyperexcitability in epilepsy syndromes.

Local anesthetics and channelopathies. Because spike initiation depends on a threshold density of functional voltage-gated Na⁺ channels concentrated at the axon initial segment, local anesthetics (Na⁺ channel blockers) and certain inherited channelopathies alter the effective threshold for integrated synaptic input to trigger a spike — clinically relevant in pain management and in inherited epilepsy/ataxia syndromes involving AIS channel mutations.

Dendritic computation and artificial neural networks. Modern computational neuroscience uses detailed multicompartment models (descended directly from Rall's cable theory) to simulate how real dendritic morphology and active conductances (NMDA spikes, Ca²⁺ spikes, dendritic Na⁺ spikes) allow single neurons to perform nonlinear, near "multi-layered" computations — findings that have influenced the design of more biologically realistic and more computationally powerful artificial neural network units beyond the simple linear-summation-plus-nonlinearity point neuron model.

Neurodegenerative and psychiatric disease research. Changes in dendritic spine density and synaptic weighting (well documented in Alzheimer's disease, schizophrenia, and autism spectrum models) are interpreted mechanistically through their effects on spatial and temporal summation — loss of spine density directly reduces the number of coincident inputs available for spatial summation, altering circuit-level computation even before overt cell death occurs.

Brain-machine interfaces and closed-loop neuromodulation. Devices that stimulate or record from neural tissue (deep brain stimulation electrodes, cortical microelectrode arrays) are designed and interpreted with explicit reference to how stimulation-evoked postsynaptic potentials will spatially and temporally summate in target neurons, since the therapeutic or decoded effect depends on whether the stimulation pattern can reliably push integrated membrane potential across threshold.

Ask the tutor: How does the relative position of an excitatory and an inhibitory synapse on a dendrite change whether inhibition acts subtractively or divisively?