Abstract
The temporal resolution of neuronal integration depends on the time window within which excitatory inputs summate to reach the threshold for spike generation. Here, we show that in rat hippocampal pyramidal cells this window is very narrow (less than 2 milliseconds). This narrowness results from the short delay with which disynaptic feed-forward inhibition follows monosynaptic excitation. Simultaneous somatic and dendritic recordings indicate that feed-forward inhibition is much stronger in the soma than in the dendrites, resulting in a broader integration window in the latter compartment. Thus, the subcellular partitioning of feed-forward inhibition enforces precise coincidence detection in the soma, while allowing dendrites to sum incoming activity over broader time windows.
At certain brain synapses, reliable transmission is ensured through large, rapidly rising, excitatory postsynaptic potentials (EPSPs), which are able to trigger a spike with little latency variation (1).
In hippocampal pyramidal cells (PCs), the small size of most unitary EPSPs requires that synaptic activity summate to reach the spike threshold (2). In principle, the relatively long membrane time constant of these neurons (3) may allow EPSPs to summate over large time windows. The occurrence of spikes would then reflect the average synaptic bombardment over time instead of being selectively time-locked to coincident synaptic activity. Thus, it is not known whether the timing of a spike in PCs reports the timing of the afferent activity triggering the spike (4–6). This issue can be addressed experimentally by determining the time window within which the activity of independent synaptic inputs must occur to trigger a spike.
We recorded from CA1 PCs in acute hippocampal slices from rat brains in cell-attached mode to avoid interfering with the intracellular ionic composition. Two stimulation electrodes were placed in the stratum radiatum at 300 to 600 μm on each side of the recorded neuron (7). Stimulation intensity was set so that when the two Schaffer collateral pathways were stimulated simultaneously, the PC fired a spike, detected as a capacitive current, in about 50% of the trials (threshold stimulation). The probability of spiking steeply decreased when one of the stimuli was shifted in time in 2.5- or 5-ms steps (Fig. 1A). A Gaussian fit of the data gave a SD of 1.4 ms (n = 9 cells).
Coincidence detection in PCs is abolished by GABAAR antagonists. (A) Current traces showing cell-attached recordings from CA1 PCs upon stimulation of two Shaffer collateral pathways. Each sweep shows a different interstimulus interval (ISI) between pathways. (Upper left) In control conditions, a spike is triggered only upon simultaneous stimulation of pathways 1 and 2. (Upper right) In bicuculline, spikes are also triggered with larger ISIs. (Bottom) Histogram showing normalized probability of spike generation plotted against ISIs (n = 9 cells). The solid columns show control conditions (bin width, 2.5 ms for ISIs between ±15 ms and 5 ms for longer ISIs). The open columns show conditions in GABAAR antagonists (bin width, 5 ms). Norm. prob., normalized probability; Δt, interstimulus interval. (B) Same cell as in (A). (Upper panels) Current traces; four superimposed sweeps were recorded upon simultaneous stimulation of both pathways without (left) and with (right) bicuculline. (Bottom) Histograms showing the probability distribution of the time of spike occurrence (n = 9 cells; bin width, 500 μs).
We then blocked γ-aminobutyric acid A receptors (GABAAR) with bicuculline (20 μM) or with the more selective antagonist SR95531 (3 μM) (8) and readjusted the stimulation intensity of both pathways to match the spiking probability observed under control conditions with simultaneous stimulation (51 ± 3% in control conditions versus 50 ± 5% in the presence of bicuculline, n = 5 cells; 62 ± 6% in control conditions versus 64 ± 11% in the presence of SR95531,n = 4 cells). This increased the delay between stimulus and spike (8.2 ± 0.6 ms in control conditions, 16.8 ± 1.2 ms in the presence of GABAAR antagonist; n= 9 cells).
GABAAR antagonists greatly prolonged the integration window (SD = 17.8 ms in bicuculline, n = 5 cells; SD = 15.6 ms in SR95531, n = 4 cells). In addition, although spikes triggered with simultaneous stimulation under control conditions showed submillisecond variability in spike delay (jitter), as described for intracellular current injections (9), in the presence of GABAAR antagonists the jitter increased almost threefold (the SD was 0.5 ms in control conditions versus 1.4 ms with GABAAR antagonists,n = 9 cells; Fig. 1B).
We monitored the underlying synaptic events using whole-cell recordings. In voltage-clamped cells, a stimulus elicited an excitatory postsynaptic current (EPSC)–inhibitory postsynaptic current (IPSC) sequence. The IPSC, which could be blocked by bicuculline, had a delayed onset with respect to the EPSC (1.9 ± 0.2 ms,n = 9 cells; Fig. 2A). Under current-clamp conditions, a stimulus elicited an EPSP-IPSP sequence (10), which was strongly reduced by bath perfusion of the α-amino-3-hydroxyl-5-methyl-4-isoxazole-propionate (AMPA)/kainate receptor antagonist NBQX [20 to 40 μM; EPSP, 114 ± 5% reduction; IPSP, 91 ± 2% reduction (11);n = 12 (six cells, 12 pathways); Fig. 2A], indicating that the stimulation of Schaffer collaterals leads to synaptic activation of GABAergic interneurons (feed-forward inhibition).
Feed-forward inhibition is responsible for coincidence detection. (A) Whole-cell recordings from CA1 PCs upon Schaffer collateral stimulation. (Left panel) Current traces recorded in control conditions and bicuculline (bicu) and their algebraic difference (holding potential, –73 mV). (Left inset) Delays between the onset of the EPSC and the IPSC (the solid circle represents the average; n = 9 cells). (Right panel) Different cell, from the one on the left. The voltage traces recorded in control conditions and in NBQX are shown. The membrane potential is −66 mV. (Right inset) The residual EPSP and IPSP (% of control) in NBQX [n = 12 (six cells, 12 pathways)]. (B) (Left) Responses to low-intensity stimulation of Schaffer collaterals from a CA1 PC voltage clamped at –68 mV. Illustrated sweeps were collected with the same stimulation intensity and were ordered according to whether the stimulus evoked an EPSC-IPSC sequence, the failure of either component, or the complete failure of transmission. (Right) IPSC conductance is plotted against EPSC slope (n = 14 cells; bin width, 5 pA/ms). In all experiments, the stimulation intensity was slowly decreased until the complete failure of transmission occurred. (C) EPSC-IPSC sequences recorded in a CA1 PC voltage clamped at –58 mV and evoked by applying brief potassium puffs (1M KCl for 10 to 40 ms at 0.1 Hz) with a patch pipette on the CA3 cell body layer. (D) Schematic diagram of disynaptic feed-forward inhibition. SC, Schaffer collateral; IN, interneuron; PC, pyramidal cell. (E) Voltage traces for current-clamp recordings from CA1 PCs upon stimulation of two Schaffer collateral pathways. (Left) Control conditions; the dotted line is the average response to stimulation of one pathway. Continuous lines represent single responses to three different ISIs. (Middle) Four superimposed responses to four ISIs in bicuculline. Spikes were truncated. (Right) Histogram showing normalized probability of spike generation plotted against the ISI (n = 8 cells). Bin width in control, 2.5 ms for ISIs between ±15 ms; 5 ms for longer ISIs (solid columns); bin width in GABAAR antagonists, 5 ms (open columns). (F) The same cell as in (E), showing simultaneous stimulation of both pathways without (left) and with (middle) bicuculline. (Right) Histograms (n = 8 cells; bin width, 500 μs).
This was not due to the stimulation of large numbers of Schaffer collaterals, because even the weakest stimulation intensities, still producing a detectable postsynaptic response, evoked a delayed inhibitory component (2.9 ± 1 nS; n = 14 cells;Fig. 2B) (12). In addition, brief puffs of potassium solution focally applied onto the CA3 PC layer evoked clear EPSC-IPSC sequences in CA1 PCs (n = 5 cells; Fig. 2C) (13).
These results are consistent with previous observations indicating that spikes in interneurons can be triggered by unitary EPSPs (14–16) and suggest that CA3 to CA1 signaling occurs via a canonical EPSP-IPSP sequence (Fig. 2D).
To determine the integration window in current-clamp conditions, we set the stimulation intensity at the threshold for spike generation when two pathways were stimulated simultaneously and readjusted the intensity to keep spiking probability constant after the perfusion of GABAAR antagonists (58 ± 5% in control conditions, 56 ± 2% in GABAAR antagonists; n = 8 cells (five in bicuculline, three in SR95531) (17). This decreased the initial slope of the EPSP by 80 ± 4% [n = 6 (three cells in SR95531, six pathways)], indicating that without feed-forward inhibition, a fifth of the excitatory fibers originally recruited are sufficient to reach threshold. The slower depolarization is likely to underlie the increased spike delay (6.9 ± 0.5 ms versus 16.1 ± 0.8 ms;n = 8 cells) and jitter (0.6 ms versus 1.8 ms;n = 8 cells; Fig. 2F) (18). The integration window increased from 1.6 ms in control conditions to 14.8 ms with GABAAR antagonists (n = 8 cells; Fig. 2E).
Different classes of GABAergic interneurons selectively innervate different subcellular compartments of PCs (19). To reveal the subcellular target of feed-forward inhibition on CA1 PCs, we simultaneously recorded from the soma and apical dendrite (20). The Schaffer collateral-evoked EPSP recorded in the dendrites was larger by 368 ± 112% [4.2 ± 0.5 mV in the dendrite versus 1.2 ± 0.3 mV in the soma; n = 10 (nine cells, 10 pathways; the average distance between pipettes was 211 ± 19 μm, and the range was from 120 to 341 μm)] and had a steeper initial slope compared with the somatic response [1.5 ± 0.2 mV/ms versus 0.6 ± 0.1 mV/ms, P < 0.01;n = 10 (nine cells, 10 pathways); Fig. 3A]. In bicuculline, EPSPs recorded in the dendrite were only 15 ± 5% larger than those recorded in the soma [6.2 ± 0.7 mV versus 5.5 ± 0.6 mV in the soma;n = 10 (nine cells, 10 pathways)] and EPSP half-decay times did not significantly differ between compartments [38.6 ± 6.9 ms versus 40.9 ± 6.2 ms; P > 0.28;n = 10 (nine cells, 10 pathways); Fig. 3, A and B]. In bicuculline, initial slopes remained unchanged, confirming the lack of direct activation of inhibitory fibers [0.6 ± 0.1 mV/ms in control versus 0.7 ± 0.1 mV/ms in bicuculline (measured in the soma); P > 0.26; n = 10 (nine cells, 10 pathways); Fig. 3B]. The bicuculline-sensitive area (21) was significantly greater in the soma by 24 ± 5% (P < 0.001; n = 9 cells; Fig. 3B) (22). In addition, the bicuculline-sensitive area in dendritic recordings decreased with increasing distance from the soma (Fig. 3B). To test whether an exclusively somatic inhibition could account for the above observation, we imposed an inhibitory conductance (dynamic clamp) through the somatic pipette 2 ms after the onset of Schaffer collateral-evoked EPSPs in the presence of the GABAAR antagonist SR95531 (3 μM; the average distance between pipettes was 200 ± 14 μm, and the range was from 143 to 255 μm; n = 9 cells; Fig. 3C) (23). The attenuation of the simulated IPSP-sensitive area was not significantly different from the decrease in the bicuculline-sensitive area (P > 0.26; Fig. 3D) (24).
Feed-forward inhibitory inputs selectively impinge on the soma of PCs. Drawing at top shows the recording configuration. (A) Voltage traces showing somatic (s, black) and dendritic (d, blue) responses to Schaffer collateral stimulation, in control conditions (left) and in bicuculline (middle). (Right) Summary graph showing dendritic EPSP amplitudes plotted against somatic EPSP amplitudes in control conditions (solid symbols) and in bicuculline (open symbols). Different symbols are shown for different experiments [n = 10 (9 cells, 10 pathways); the average distance between pipettes was 211 ± 19 μm;n = 9 cells]. (B) Same cell as in (A). Voltage traces; somatic (left) and dendritic (middle) recordings with and without bicuculline and their algebraic difference. (Right) Summary graph in which the integral of the algebraic difference between traces recorded in the presence and absence of bicuculline in the dendrites is divided by the corresponding integral in the soma and plotted against the distance between the two recording sites (n = 9 cells). (C) Different cell from that in (B). Voltage traces; somatic (s, black) and dendritic (d, blue) responses to Schaffer collateral stimulation in SR95531, without (left) and with (middle) dynamic current injection (lower trace) through the somatic pipette. (Right) Summary graph showing dendritic EPSP amplitudes plotted against somatic EPSP amplitudes under control conditions (open symbols) and with dynamic current injection (solid symbols). Different symbols are shown for different experiments (the average distance between pipettes was 200 ± 14 μm; n = 9 cells). (D) Same cell as in (C). Voltage traces showing somatic (left) and dendritic (middle) recordings in the presence and absence of dynamic current injection and their algebraic difference. (Right) Summary graph showing the ratio between the dendritic and somatic integrals of the algebraic differences between traces recorded in the presence and absence of dynamic current injection plotted against the distance between the two recording sites (solid symbols; n = 9 cells). Note the close match (apart from one experiment) with the data points obtained with the physiological activation of feed-forward inhibition [open symbols, taken from (B)].
We compared the time window of EPSP summation between the soma and the dendrites of CA1 PCs (the average distance between pipettes was 229 ± 30 μm, and the range was from 120 to 323 μm;n = 6 cells; Fig. 4A). The graph illustrates that although the soma effectively summates EPSPs over a time window ranging from –1.5 to +2.4 ms, effective summation of EPSPs recorded in the dendrites occurs over a broader window (from –8.6 to +12.3 ms) (25). This difference was due to a preferential inhibitory input on the soma, first, because in bicuculline the window of EPSP summation was indistinguishable between both compartments (>60 ms; the average distance between pipettes was 200 ± 12 μm, and the range was from 168 to 236 μm; n = 4 cells; Fig. 4B). Second, imposing a somatic inhibitory conductance in the presence of SR95531 mimicked the somatic and dendritic integration windows observed under control conditions (soma, +1.8 ms; dendrite, +12.5 ms; the average distance between pipettes was 184 ± 20 μm, and the range was from 143 to 248 μm; n = 4 cells; Fig. 4C) (26).
Dendrites effectively summate EPSPs over broader time windows compared with the soma. Simultaneous somatic (black) and dendritic (blue) recordings from CA1 PCs upon stimulation of two Schaffer collateral pathways. The drawing at the top is as inFig. 3. (A) Voltage traces showing seven superimposed somatic (left) and dendritic (middle) responses to seven different ISIs in control conditions. (Right) Summary graph showing the ratio ofb to a plotted against the ISI, wherea is the amplitude of the EPSP evoked by the stimulation of one pathway alone, and b is the maximal positive deviation from the resting membrane potential upon stimulation of that same pathway when preceded by the stimulation of the other pathway. Black symbols represent somatic recordings; blue symbols represent dendritic recordings (the average distance between pipettes was 229 ± 30 μm; n = 6 cells). (B) A different cell from that in (A). Conditions were as in (A), but in the presence of bicuculline (the average distance between pipettes was 200 ± 12 μm; n = 4 cells). (C) (Upper panels) Voltage traces showing seven superimposed somatic (left) and dendritic (middle) responses to seven different ISIs in the presence of SR95531. (Right) Summary graph (n = 4 cells). (Lower panels) Voltage traces showing the same cell with dynamic current injection in the soma (left and middle). (Right) Summary graph (n = 4 cells; the average distance between pipettes was 184 ± 20 μm).
Our data indicate that feed-forward inhibition limits temporal summation of EPSPs far below the mean interspike interval of PCs (27), thus making them precise coincidence detectors (28). The presence of feed-forward inhibition in several major projections in the central nervous system (29) may thus serve as a means to maintain timing across brain areas (30).
The recruitment of GABAergic interneurons whose terminals preferentially impinge on or close to the soma rather than the apical dendrites leads these two compartments to integrate EPSPs over different time windows. Thus, summation of EPSPs to reach the threshold for action potential generation has to occur within less than 2 ms in CA1 PCs, whereas dendritic summation of the same EPSPs can occur over longer time periods. Cooperativity between separate inputs over long time windows may enable the expression of slower, voltage-dependent signaling [e.g., through N-methyl-d-aspartate (NMDA) receptors], as may be necessary for the induction of lasting changes in synaptic strength.
Although interneurons innervating PC dendrites also receive excitatory inputs from Schaffer collaterals, providing the anatomical basis for dendritic feed-forward inhibition (31), their mode of activation remains to be clarified.
Noradrenaline and acetycholine can selectively regulate the excitability of subsets of GABAergic interneurons (32,33) and are released during different behavioral states (34, 35). The excitability of interneurons will determine the delay and size of feed-forward inhibition and hence the size of the integration window. It is conceivable that according to the behavioral state of the animal, the operation mode of PCs shifts from precise coincidence detection to integration over large time windows.
↵* To whom correspondence should be addressed. E-mail: massimo{at}hifo.unizh.ch
