We expressed high levels of channelrhodopsin-2 (ChR2) in a focal subpopulation of neurons in anterior piriform cortex by an intersectional infection with two viruses. Adeno-associated virus (AAV) encoding Cre-dependent ChR2-YFP was co-injected with lentivirus encoding Cre recombinase (). This strategy ensures high ChR2 expression that is limited by the spread of the lentiviral vector to a focal subset of excitatory and inhibitory neurons. Cre-positive, ChR2-expressing neurons were largely restricted to a focal cluster of layer II/III cells a few hundred microns wide (), although axons of YFP-expressing cells were observed throughout the rostrocaudal extent of the piriform ().
Recurrent excitatory synapses extend undiminished across piriform cortex
We prepared acute parasagittal brain slices through piriform cortex from 8–12 week-old mice. Typically, one slice per animal included a significant extent of piriform cortex along the rostrocaudal axis containing a focal area of YFP fluorescence (). Whole-cell recordings were then obtained from multiple layer II pyramidal cells (Figure S1A–C
) at different distances from the center of the infection site. A 500-ms light pulse centered on the somata of cells at the site of infection evoked robust and sustained photocurrents in a subset of these cells (, S1D
). At the center of the fluorescence cloud, 35% of neurons were ChR2-positive (ChR2+
, defined by the presence of a sustained photocurrent) but the frequency of ChR2+
cells diminished dramatically with distance from the center of the infected area. In contrast, the magnitude of the photocurrent in ChR2+
cells did not decrease with distance from the injection site ().
We next determined the connectivity of this focal set of ChR2+
cells with ChR2-negative (ChR2−
) pyramidal cells across the piriform. A light pulse focused on ChR2−
cells distant from the site of infection elicited transient inward currents in voltage-clamp recordings at −70 mV. These currents were blocked by AMPA and NMDA receptor antagonists, indicating that these were excitatory postsynaptic currents (EPSCs) evoked from the axons of ChR2+
neurons (). These light-evoked synaptic responses exhibited properties consistent with those described for “associational” piriform synapses (Franks and Isaacson, 2005
; Figure S2A–C
). Monosynaptic (Figure S2D, E
), light-evoked EPSCs were observed in 94 of 95 recorded ChR2−
cells across the piriform in slices from 11 animals (). Interestingly, the EPSC amplitude was largely independent of the distance of the recorded cell from the infection site (). These data indicate that extend, undiminished, across millimeters of piriform cortex. This pattern of long-range excitatory connectivity in piriform was also apparent when a modified rabies virus was used as retrograde tracer (Wickersham et al., 2007
) to map the cells that provide synaptic input to pyramidal cells (Figure S2F–I
The long-range excitatory recurrent projections in piriform contrasts with the pattern of connectivity we observed upon focal expression of ChR2 in primary somatosensory (S1) and visual (V1) cortex. In both S1 and V1 (not shown), there was a steep decrease in light-evoked EPSC amplitude in layer II/III pyramidal neurons with increasing distance from the infection center (). Thus, unlike neurons in S1 and V1 that preferentially connect to more proximal targets, a given piriform neuron forms synapses onto layer II pyramidal cells with similar probability across the cortex.
We next obtained a quantitative estimate of the number and strength of the intrinsic excitatory inputs onto a given piriform neuron. The amplitudes of the light-evoked EPSCs were large, but somewhat variable within a given animal (mean ± S.D.: 441 ± 334 pA; C.V., 0.76; n=95 cells from the 11 animals/slices, ). These large EPSCs were presumably mediated by inputs from many ChR2+ axons, each of which contributes a unitary response (uEPSC). To determine the number of ChR2+ axonal inputs underlying the light-evoked response we first determined the strength of a single recurrent input. We decreased the intensity and field of illumination of the light pulse to achieve threshold activation of a single ChR2+ axonal input, which was indicated by interleaved successful responses and failures to these light pulses (). The mean size of the uEPSC was 36.2 pA (± 20.3 pA, S.D.; range: 16–74 pA, n=10), though our measurements may be biased towards slightly larger, more easily resolved responses. The success rate (0.52 ± 0.047; n=10) places a lower bound on the probability of synaptic vesicle release at recurrent synapses.
Recurrent excitatory piriform synapses are sparse and weak
We next determined the number of synaptic contacts each ChR2+
axon makes onto a given layer II pyramidal cell by measuring quantal responses (qEPSC) evoked by replacing extracellular Ca2+
to desynchronize synaptic release. In slices bathed in Sr2+
, light pulses evoked a large, early synchronous response with a tail of many small events that are thought to represent quantal synaptic currents (Dodge et al., 1969
; Franks and Isaacson, 2006
; Goda and Stevens, 1994
; ). The similar amplitude of the light-evoked uEPSCs and qEPSCs (25 pA ± 10 pA, S.D.; n=11; ) suggests that a recurrent axon typically makes single, en passant synaptic contacts with a given pyramidal cell in piriform cortex, consistent with anatomical predictions (Datiche et al., 1996
; Johnson et al., 2000a
). Moreover, at this contact, a presynaptic action potential releases, at most, a single quantum of transmitter. The light-evoked qEPSCs were larger and had faster kinetics than qEPSCs evoked from electrical stimulation of mitral and tufted cell axons in the lateral olfactory tract (LOT) in the same cells (14 pA ± 4.0 pA, n=9, ). The amplitudes of qEPSCs from afferent and recurrent inputs are consistent with the range of amplitudes of miniature EPSCs we recorded in TTX (17.3 ± 7.1 pA, S.D.; n = 562 events, 9 cells). The difference in the size of the afferent and recurrent qEPSCs may reflect differences in their biophysical properties (Schikorski and Stevens, 1999
) or may simply reflect greater dendritic filtering of the more distal LOT inputs.
The ratio between the average EPSC (500 pA) evoked with a saturating light intensity that activates all ChR2+
inputs (see ) and the unitary ESPC (25 pA) suggests that a cell receives, on average, 20 active inputs from the population of ChR2+
neurons. From the distribution of ChR2+
cells, we estimate that we infected about 8000 excitatory neurons per animal (Figure S1E–H
). This implies that the connectivity between any two pyramidal cells is less than 1%, and this value is largely independent of the distance between two piriform cells. Moreover, given that we infected less than 1% of all piriform pyramidal neurons (8000 neurons out of a total of an assumed 106
pyramidal cells in the piriform), our observation of 20 activated ChR2+
inputs per cell implies that that each neuron receives, at least, 2000 recurrent excitatory inputs. In contrast, pyramidal cells are thought to receive only about 200 afferent inputs from the bulb (Davison and Ehlers, 2011
). These inputs, however, are multiquantal and can be quite large, with each axon typically making ~5 contacts per cell (Bathellier et al., 2009
; Franks and Isaacson, 2006
); but see (McGinley and Westbrook, 2011
; Suzuki and Bekkers, 2011
). Individual pyramidal cells may therefore receive strong inputs from 200 mitral/tufted cells in the bulb and weak inputs from more than 2000 pyramidal cells across piriform cortex.
Recurrent excitation drives local strong, scaled inhibition
This recurrent network would result in runaway excitation in response to odor unless its activity was tempered by inhibition. To investigate the role of inhibition in modulating the activity of the recurrent excitatory network, we isolated the inhibitory synaptic current by recording from pyramidal cells at a voltage near the equilibrium potential for EPSCs (Vm = +5 mV). We first recorded from ChR2− cells close to the infection site in the presence of NBQX and APV to block glutamatergic transmission. Under these conditions, light pulses evoked outward currents that were blocked by the GABAA-receptor antagonist gabazine (GBZ, ), indicating that these were inhibitory postsynaptic currents (IPSCs) originating directly from ChR2+ GABAergic neurons. Although all cells in or near the infection site showed direct IPSCs, direct inhibition rapidly decayed at distances >300 μm beyond the edge of the infected area, indicating that this direct inhibition is local ().
In contrast to the local direct inhibition, when inhibitory currents were recorded with excitatory transmission intact, we observed large IPSCs in almost every neuron, regardless of distance from the site of infection (85/87 cells; ). Because direct inhibition is local, inhibitory currents distant from the site of infection must result from the activation of long-range excitatory ChR2+ axons that synaptically activate local inhibitory interneurons. The long-range inhibitory responses lagged behind the onset of the light-evoked EPSCs recorded in the same cells by 1.6 ± 0.12 ms (n=21) and were abolished by NBQX and APV (), indicating that this inhibition was disynaptic and driven by axons of ChR2+ excitatory cells. Our methodology therefore allows us to selectively isolate disynaptic inhibition by recording from cells far from the infection site where the light-evoked IPSC is not contaminated by direct inputs from ChR2+ inhibitory neurons.
A comparison of the magnitudes of the excitatory and disynaptic inhibitory currents in a given cell revealed that the inhibitory response was much larger than the excitatory response (). We compared the input-output relationship of excitation versus inhibition by recording the excitatory and inhibitory responses to a series of light pulses of increasing intensity (). Increasing the intensity of the light pulse increased the excitatory responses from a level at which we failed to observe any synaptic response to a level at which the EPSC amplitudes saturated and failed to increase with increasing light intensity. The IPSC scaled with, and dominated, the EPSC across the entire range of stimulus intensities (). We also determined the laminar organization of the recurrent excitatory and inhibitory synaptic inputs onto layer II pyramidal cells using focal illumination along the cell's apical-basal axis in the presence of TTX/4-AP (Petreanu et al., 2009
). These experiments indicate that pyramidal cells receive the majority of their recurrent excitatory input onto their proximal apical dendrites in layer Ib, whereas feedback inhibition is preferentially recruited by their axons projecting through layer III (Figure S3
How do these recurrent circuits shape the response of piriform neurons to bulbar inputs? We paired a brief train of electrical LOT stimuli that mimics the burst firing of a mitral cell to odorant stimulation (Cang and Isaacson, 2003
; Margrie and Schaefer, 2003
) with a brief train of light pulses in piriform cortex (both stimuli, 5 pulses at 40 Hz; i.e. a 100 ms burst) and recorded the responses in pyramidal cells in current clamp. The stimulus strengths were adjusted to evoke spiking in 10% of the trials when either stimulus was presented alone (probability of spiking was 0.10 ± 0.38 following electrical stimulation of the LOT and was 0. 10 ± 0.054 with light-activation of piriform; n=6). In contrast to the low probability of spiking when LOT or piriform was activated alone, action potentials were evoked in 90% of the trials (0.90 ± 0.056) when the two inputs were presented simultaneously ().
Feedback inhibition shapes piriform activation
We next examined the effect of altering the temporal relationship between the pairing of bulbar and recurrent inputs. No increase in spiking was observed when the onset of the two 100 ms-long bursts of stimuli were 150 ms apart. However, when the LOT train was delivered 100 ms before the piriform train, such that the last LOT-evoked input coincided with the first light-evoked input, the cell fired action potentials in 75% of the trials (0.75 ± 0.098; ). In contrast, no enhancement in spike firing was observed when the piriform train arrived 100 ms before the LOT input (0.20 ± 0.073; unpaired t-test vs. LOT alone, p=0.423; vs. PCx alone, p=0.315; ).
We then examined the role of inhibition in this pairing paradigm by repeating these experiments in the presence of GBZ and the GABAB antagonist, CGP55845. Blocking inhibition broadened the time window over which spiking could be enhanced by pairing the inputs (). Furthermore, the efficacy with which the pairing of the inputs enhanced the response was less dependent on the order in which the two inputs were presented (skewness of control distribution, 0.64 ± 0.17, n=6; skewness of distribution in GBZ, 0.21 ± 0.04, n=4; unpaired t-test, p < 0.05). This result implies that much of the asymmetry we observed in the efficacy of pairing order is a consequence of inhibition.
We hypothesized that the response to LOT inputs might be suppressed by prior activation of the cortical circuitry because of the recruitment of strong feedback inhibition. This prediction was tested by delivering a short train of LOT stimulation (3 pulses at 40 Hz) to achieve spiking on half the trials (0.56 ± 0.042). Indeed, when a similar train of piriform stimuli (3 pulses at 40 Hz; probability of spiking; 0.36 ± 0.16) preceded the LOT input by 100 ms, we observed an 18% reduction in the probability of spiking. (LOT train following PCx train, 0.46 ± 0.049; n=9 cells; paired t-test comparing two LOT trains, p = 0.017; ).
Two forms of inhibition have been described in the piriform cortex. Feedforward inhibition is mediated by interneurons in layer I that receive direct input from the LOT and synapse on apical dendrites of pyramidal cells, whereas feedback inhibition is mediated by the layer II/III interneurons that are activated by pyramidal cells and synapse onto pyramidal cell bodies (Luna and Schoppa, 2008
; Neville and Haberly, 2004
; Stokes and Isaacson, 2010
; Suzuki and Bekkers, 2010
). Two experimental approaches were employed to demonstrate that feedback inhibition is significantly stronger than feedforward inhibition. We observed a dramatically greater effect of gabazine on synaptic responses following subthreshold recurrent stimulation versus LOT stimulation (Figure S4A
). We also determined the lowest stimulation intensities of either the LOT or recurrent inputs that reliably drove spiking when inhibition was blocked. LOT stimulation at this intensity could still generate spiking when inhibition was intact (Figure S4
), consistent with a relatively small role for feedforward inhibition. In contrast, piriform stimulation at this intensity always failed to evoke spikes in downstream piriform neurons when inhibition was intact. These data support a dominant role for feedback versus feedforward inhibition in controlling the activation of piriform cortex pyramidal cells.