Disinhibitory Circuitry Gates Associative Synaptic Plasticity in Olfactory Cortex

Abstract

Inhibitory microcircuits play an essential role in regulating cortical responses to sensory stimuli. Interneurons that inhibit dendritic or somatic integration act as gatekeepers for neural activity, synaptic plasticity, and the formation of sensory representations. Conversely, interneurons that selectively inhibit other interneurons can open gates through disinhibition. In the anterior piriform cortex, relief of inhibition permits associative LTP of excitatory synapses between pyramidal neurons. However, the interneurons and circuits mediating disinhibition have not been elucidated. In this study, we use an optogenetic approach in mice of both sexes to identify the inhibitory interneurons and disinhibitory circuits that regulate LTP. We focused on three prominent interneuron classes: somatostatin (SST), parvalbumin (PV), and vasoactive intestinal polypeptide (VIP) interneurons. We find that LTP is gated by the inactivation SST or PV interneurons and by the activation of VIP interneurons. Further, VIP interneurons strongly inhibit putative SST cells during LTP induction but only weakly inhibit PV interneurons. Together, these findings suggest that VIP interneurons mediate a disinhibitory circuit that gates synaptic plasticity during the formation of olfactory representations.

SIGNIFICANCE STATEMENT Inhibitory interneurons stabilize neural activity during sensory processing. However, inhibition must also be modulated to allow sensory experience shape neural responses. In olfactory cortex, inhibition regulates activity-dependent increases in excitatory synaptic strength that accompany odor learning. We identify two inhibitory interneuron classes that act as gatekeepers preventing excitatory enhancement. We demonstrate that driving a third class of interneurons inhibits the gatekeepers and opens the gate for excitatory enhancement. All three inhibitory neuron classes comprise disinhibitory microcircuit motifs found throughout the cortex. Our findings suggest that a common disinhibitory microcircuit promotes changes in synaptic strength during sensory processing and learning.

Introduction

Throughout the cortex, the responses of individual neurons as well as neural ensembles are refined by sensory experience. One feature of experience-dependent plasticity is long-term changes in excitatory synaptic strength. Although the cellular mechanisms of synaptic plasticity have been extensively studied (Abbott and Nelson, 2000; Malenka and Bear, 2004), less is known about the role circuitry plays in promoting changes in synaptic strength. Inhibitory interneurons (INs) regulate dendritic integration and neural activity and, hence, limit the induction of synaptic plasticity. Conversely, circuit mechanisms that regulate inhibition could gate synaptic enhancement (Artinian and Lacaille, 2018; Lucas and Clem, 2018). In this study, we investigate the inhibitory and disinhibitory circuit motifs that govern synaptic plasticity at recurrent excitatory synapses in the anterior piriform cortex (APC).

The APC is a primary cortex for processing olfactory information. However, unlike other primary sensory cortices, APC lacks a topological representation of odor identity. Odors are represented by distributed cortical ensembles (Illig and Haberly, 2003; Rennaker et al., 2007; Stettler and Axel, 2009). This distributed activity is supported by diffuse afferent projections from mitral and tufted (M/T) cells of the olfactory bulb (Sosulski et al., 2011; Igarashi et al., 2012) as well as uniform intracortical excitation across the APC (Franks et al., 2011). It is postulated that long-lasting ensemble odor representations are constructed by strengthening excitatory synapses between pyramidal neurons (PNs) that are coactivated by M/T afferents (Haberly, 2001; Wilson and Sullivan, 2011). In support of this hypothesis, intracortical synapses are strengthened following odor learning in vivo (Saar et al., 2002) and through pairing of afferent and intracortical stimulation in vitro (Kanter and Haberly, 1993; Johenning et al., 2009).

The APC is ideal for investigating the synaptic and circuit mechanisms that give rise to ensemble representations. The afferent and intracortical excitatory pathways form two distinct fiber tracts that compartmentalize inputs on the apical dendrites of PNs. M/T cell afferents synapse on the distal dendrite, whereas intracortical axons form a proximal fiber tract (Haberly and Price, 1977, 1978). When strong activation of afferent tract is paired with weak stimulation of the intracortical tract, intracortical excitatory synapses are strengthened through NMDAR-dependent, associative LTP (Kanter and Haberly, 1993). The induction of associative LTP requires dendritic disinhibition by GABA_A receptor antagonists (Kanter and Haberly, 1993; Kanter et al., 1996). This suggests a circuit mechanism that provides dendritic disinhibition could gate associative LTP and ensemble formation. However, the INs and circuits mediating dendritic inhibition and disinhibition have yet to be fully elucidated.

Olfactory stimuli recruit both feedforward and recurrent inhibition of PN dendrites (Poo and Isaacson, 2009; 2011). Strong stimulation of the afferent pathway diminishes feedforward inhibition and increases recurrent inhibition through a combination of short-term plasticity and PN recruitment (Stokes and Isaacson, 2010; Suzuki and Bekkers, 2010a; Large et al., 2016b). Recurrent inhibition is mediated by INs that express somatostatin (SST-INs) or parvalbumin (PV-INs) (Stokes and Isaacson, 2010; Suzuki and Bekkers, 2010a,b; Large et al., 2016a). Inhibitory synapses from SST-INs are optimally located on the proximal apical dendrite to regulate intracortical excitatory synaptic plasticity (Suzuki and Bekkers, 2010b; Large et al., 2016a). Conversely, PV-INs could regulate spike activity, backpropagation, and LTP induction (Johenning et al., 2009). In other cortices, vasoactive intestinal polypeptide INs (VIP-INs) inhibit both SST- and PV-INs and potentially disinhibit PNs (Lee et al., 2013; Pfeffer et al., 2013; Karnani et al., 2016). VIP-INs are prominent INs in APC (Suzuki and Bekkers, 2010b), but their connectivity and function are unknown. We investigated the influence of SST-, PV-, and VIP-INs on associative LTP induction in APC. We find that activation of VIP-INs as well as inactivation of SST-INs or PV-INs promote LTP. VIP-INs strongly inhibit SST-INs during LTP induction but only weakly inhibit PV-INs and PNs. Together, our findings suggest a VIP-to-SST-to-PN disinhibitory circuit gates associative LTP in APC.

Materials and Methods

Mice

VIP-Cre (B6:Viptm1(cre)Zjh/J), SST-Cre (B6.Sst tm2.1(cre)Zjh/J) and PV-Cre (B6.129P2 Pvalbtm1(cre)Arbr/J) mice express cre-recombinase (Taniguchi et al., 2011). These mice were crossed with Ai32 mice (B6.129S-Gt ROSA)26Sortm32 (CAG-COP4*H134R/EYFP)Hze/J) to express channelrhodopsin (ChR2) or Ai35 mice (B6;129S-Gt(ROSA)26Sor tm35.1(CAG- aop3/GFP)Hze/J) to express archaerhodopsin (Arch) (Madisen et al., 2012). All mice are from The Jackson Laboratory. All animals were bred, handled, and treated in a manner that was evaluated and approved by the Animal Care and Use Committee at the University of Pittsburgh Institutional Animal Care and Use Committee, Protocol 17070877.

Slice preparation

APC brain slices were prepared from mice aged P19-P35. The mice were anesthetized with isoflurane, and the brain was removed and immersed in ice-cold oxygenated (95% O₂-5% CO₂) ACSF (in mm as follows: 125 NaCl, 2.5 KCl, 25 NaHCO₃, 1.25 NaH₂PO₄, 1.0 MgCl₂, 25 glucose, 2.5 CaCl₂, all chemicals from Sigma, unless otherwise stated). Parasagittal slices (300 µm) were cut on a vibratome (Leica Biosystems) in ice-cold ACSF. The slices were transferred to warm ACSF (36°C) for 30 min, then 20°C-22°C for 1 h, and recorded at 25°C-28°C.

Electrophysiology

Recordings were performed using a MultiClamp 700B amplifier (Molecular Devices). Data were low pass filtered (4 kHz) and digitized at 10 kHz using an ITC-18 (Instrutech) controlled by custom software written in IgorPro (Wavemetrics). Recording pipettes (4-10 mΩ) were pulled from borosilicate glass (1.5 mm, outer diameter) on a Flaming/Brown micropipette puller (Sutter Instruments). The series resistance (<20 mΩ) was not corrected. For PSPs, the intracellular solution consisted of the following (in mm): 130 K-gluconate, 5 KCl, 2 MgCl₂, 4 ATP-Mg, 0.3 GTP, 10 HEPES, and 10 phosphocreatine, 0.05% biocytin. For IPSC recordings, Qx-314 was added to the K-gluconate internal (holding potential 0 mV), or Cs-Glu-Qx solution was used as follows (in mm, 130 Cs-gluconate, 5 KCl, 2 MgCl₂, 4 Mg-ATP, 0.3 GTP, 10 HEPES, 10 phosphocreatine, 1 Qx-314, holding potential 30 mV). Neurons were visualized using infrared-differential interference contrast microscopy (Olympus). For all recorded neurons, subthreshold response properties were obtained using a series of hyperpolarizing and depolarizing current steps (−50 to 50 pA, 1 s duration). We used specific criteria for identification of putative PV and SST-INs based on intrinsic properties of identified PV and SST neurons in APC (for details, see Large et al., 2016a). L2 PNs were identified according to previously reported intrinsic properties (Large et al., 2016b). We attempted to confirm neural identity of all neurons, including PNs post hoc using biocytin fills; ∼70% of neurons were recovered. All PNs included in this study had intact apical dendrites. This was determined from biocytin fills and/or electrophysiology. PNs that did not have apical dendrites intact typically did not receive sufficient synaptic input from the afferent fibers in L1a. These were excluded during recording.

LTP induction

Electrical stimulation was delivered using concentric bipolar electrodes (FHC). The electrodes were placed in the L1a and the L1b/L2 border. Current pulse stimuli (100 µs pulse width) were delivered through a stimulus isolation unit. Electrical stimulation strengths for L1a and L1b were set for each PN individually. PN responses vary with intrinsic properties as well as location with respect to the stimulating electrodes. For electrical stimulation of the afferent inputs to the APC (L1a), intensity was set such that theta burst stimulation (TBS) depolarized the PN to near threshold potentials (∼−45 mV) often evoking action potentials at low rates (1-10 Hz). Stimulation intensity was sufficient to activate NMDARs as long-lasting after-depolarizations were evident during TBS. For electrical stimulation of L1b, intensity was set to evoke reliable EPSPs that ranged from 1 to 6 mV at baseline. Stimulation intensity for L1b was <30% of the maximum subthreshold EPSP. During the LTP induction protocol, L1a and L1b were costimulated. TBS of L1a consisted of 10 bursts of 4 pulses (100 Hz) delivered at 250 ms intervals (Kanter and Haberly, 1993). L1b was stimulated with a single pulse delivered between the third and fourth pulse of each burst. Pre and post induction test pulses were delivered to L1b every 30 s. Baseline was collected for ∼5 min, and LTP was induced within 10 min of patching the neuron. Input resistance was monitored throughout, and neurons with deviations >20% from baseline were excluded from analysis (n = 10 neurons). We also monitored membrane potential throughout recording. Small fluctuations (<5 mV) were offset with constant current to maintain driving force. However, neurons with drift >5 mV or exhibiting unstable membrane potential fluctuations were excluded from analysis (n = 8).

Pharmacology

The GABA_A receptor antagonist gabazine (GZ, 20 μm in ACSF) was loaded into a regular patch pipette and focally applied to L1b using a gentle positive pressure (<1-5 s duration) see. Slices were oriented such that bath flowed from the soma to dendrite of the pyramidal cell to maintain GZ in the region of the dendrite. Every effort was made to ensure stability of GZ application, including supplemental applications. In a subset of slices (n = 5) from mice that expressed ChR2 in SST-INs, GZ application to L1b diminished optically evoked SST-mediated inhibition onto pyramidal cells (n = 5). We verified that optically evoked IPSPs (Control: 6.3 ± 0.69 mV) in these slices were stably diminished by GZ application throughout the LTP protocol and did not differ between baseline (IPSP GZ: 0.89 ± 0.27 mV) and post induction test pulses (IPSP GZ: 0.92 ± 0.28, p > 0.05, Wilcoxon Signed Ranks test [WSR]). In additional experiments, the NMDAR antagonist, DL-APV (10 mm, Sigma-Aldrich) was bath-applied as indicated in the main text.

Optogenetic stimulation

Shutter controlled full-field stimulation with blue (473 nm) or green (520 nm) light (Prior) was delivered through the epifluorescence pathway of the microscope (Olympus) using a water-immersion objective (40×). Light intensity (5-10 mW) was adjusted to induce spike responses (ChR2 activation) or spike suppression (Arch inactivation). Light pulse duration varied by experiment as indicated in the main text.

EPSP analysis

EPSPs were analyzed using custom software written in IgorPro (Wavemetrics). Because we do not block inhibition during L1b test pulses, we set narrow criteria to ensure that only the excitatory portion of mixed EPSP-IPSPs was analyzed. Recurrent inhibition is disynaptic and IPSPs are delayed with respect to excitation. Only EPSP onsets within 5 ms of the stimulus artifact were analyzed. Across all conditions, the average EPSP onset was 3.7 ± 0.41 ms. EPSP peak amplitude was taken as the max amplitude within 5 ms of EPSP onset. We also measured the slope of the rising phase of the EPSP from onset to 80% of peak amplitude to further isolate the EPSP. Our core findings did not differ between peak and slope measurements.

Statistics

All data are presented as mean ± SE. Statistical tests were performed using two-tailed, one- or two-sample, paired or unpaired Student's t test as appropriate. In cases of small sample sizes, nonparametric tests were used, including the Mann–Whitney U test for unpaired data and the Wilcoxon Signed Rank (WSR) test for paired data.

Results

We investigated the roles of SST-, PV-, and VIP-INs, in gating associative LTP at intracortical excitatory synapses onto L2 PNs in APC. Afferent input to PNs arrives via L1a on distal dendrites, while the intracortical fiber tract (L1b) runs proximal to the soma (Fig. 1A1). L1a and L1b are easily identified under infrared-differential interference contrast microscopy, and can be independently stimulated using concentric bipolar electrodes. Associative LTP is induced by pairing L1a and L1b stimulation using a TBS protocol (Kanter and Haberly, 1993). This protocol is consistent with respiration-coupled M/T cell spike patterns (Kepecs et al., 2007; Carey and Wachowiak, 2011). Briefly, strong TBS of L1a was paired with weak, single-pulse stimulation of L1b (see Materials and Methods; Fig. 1A2). This L1a+L1b pairing is hereafter called the induction protocol. L1a-TBS evoked low PN firing rates (FRs; <10 Hz), while weak stimulation of L1b evoked EPSPs that ranged from 1 to 6 mV. Pre and post induction, L1b stimulation was delivered every 30 s. Both raw and normalized (to baseline) L1b EPSP amplitude and rising slope were analyzed. Baseline was quantified as the average EPSP amplitude or slope in the 5 min before the induction protocol. To avoid drift in recording integrity, synaptic plasticity was measured 25-30 min following induction and quantified as the average EPSP amplitude or slope in this interval. However, in neurons where stability could be maintained, changes in synaptic strength typically lasted for the duration of recording (45-60 min). Recordings from PNs were excluded if input resistance or membrane potential measured at 25-30 min post induction varied from baseline more than ±20% or ±5 mV respectively.

Figure 1.

Associative LTP of intracortical synapses is gated by dendritic disinhibition. A1, Schematic of APC circuit and stimulation paradigm. GZ was focally applied to the dendrite region in L1b (tan oval). A2, LTP induction protocol: Strong (s) TBS of afferents (L1a) is paired with weak (w) single pulses at the intracortical pathway (L1b). B, Representative average traces pre (gray) and post induction in control (black) and with GZ(Tan). C1, Average EPSP amplitude (mV) pre (−5 to 0 min) and post induction (25-30 min) with inhibition intact (black) or with dendritic GZ (tan). C2, Average slope (mV/ms) of the rising phase of the EPSP pre and post induction. D, Normalized EPSP amplitude, and slope 30 min post induction in control (black) and with GZ (tan). E1, Time course of normalized EPSP amplitude pre and post induction (gray box, t = 0) in control (black) and with dendritic GZ (tan). E2, Time course of normalized EPSP slope. Colors as in E1.

Disinhibition of PN dendrites promotes LTP

In control slices with inhibition intact, the induction protocol did not induce LTP of L1b synapses. At 30 min post induction, neither EPSP amplitude (in mV, pre: 2.6 ± 0.60, post: 2.6 ± 0.51, p = 0.85, paired t test, n = 9) nor slope (in mV/ms, pre: 0.64 ± 0.20, post: 0.61 ± 0.18, p = 0.27, paired t test, n = 9) significantly differed from baseline (Fig. 1C1,C2). Likewise, normalized EPSP amplitude and slope did not differ significantly from 1 (Amplitude: 1.0 ± 0.071, p = 0.65; Slope: 0.96 ± 0.079, p = 0.65, one-sample t test, Fig. 1D). However, when we focally applied the GABA_A receptor antagonist GZ (20 μm) to L1b (schematic, Fig. 1A1), the induction protocol significantly enhanced EPSP amplitude (mV, Pre: 1.3 ± 0.32, Post: 3.1 ± 0.95, p < 0.050 WSR, n = 6, Fig. 1C1) and slope (mV/ms, Pre: 0.28 ± 0.070, Post: 0.52 ± 0.12, p < 0.050 WSR, n = 6, Fig. 1C2). Normalized EPSP amplitude and slope were significantly >1 (Amplitude: 2.3 ± 0.35, p < 0.05 WSR; Slope: 2.0 ± 0.27, p < 0.05 WSR, Fig. 1D,E). Consistent application of dendritic GZ for the duration of recording is difficult to maintain. In some PNs, EPSP amplitudes were strongly potentiated, but membrane potential recordings were unstable; these were excluded from analysis (n = 5). This highlights the need for a more selective and reliable optogenetic approach to investigating disinhibition.

Inactivation of SST-INs promotes LTP

We have previously shown that SST-INs strongly inhibit PN dendrites in L1b (Large et al., 2016a). Although SST-INs receive minimal direct L1a input (Suzuki and Bekkers, 2010a), SST-INs are recruited by strong TBS through recurrent excitation (Fig. 2A1,A2). To investigate the influence of SST-INs on LTP induction, SST-INs that expressed Arch were optically inactivated during TBS. Inactivation reduced SST-IN FR from 6.8 ± 1.2 Hz to 1.8 ± 0.4 Hz (p < 0.05, WSR test, Fig. 2A2) and enhanced depolarization in PNs (Fig. 2B). There was a modest but significant increase in PN FR during SST-IN inactivation (FR_ON/FR_OFF: 1.22 ± 0.10, p = 0.04, one-sample t test). However, the average PN FR during TBS with SST-IN inactivation was low (FR_ON: 3.58 ± 0.82 Hz). These findings suggest that SST-INs inhibit PNs during TBS stimulation and could prevent associative synaptic plasticity at L1B synapses. To test this, we optically inactivated SST-INs solely during induction (2 s). EPSP amplitude and slope were significantly enhanced 30 min post induction (Amplitude (mV), Pre: 2.6 ± 0.41, Post: 4.3 ± 0.76, p = 0.0071; Slope (mV/ms), Pre: 0.55 ± 0.14, Post: 1.1 ± 0.23, p = 0.0048; n = 9, paired t test, Fig. 2C1,C2,E). Normalized EPSP amplitude and slope were significantly >1 (Amplitude: 1.7 ± 0.16, p = 0.00010; Slope: 2.0 ± 0.24, p = 0.0033, one-sample t test, Fig. 2D). EPSP traces from representative PNs pre and post induction are shown in Figure 2E. The normalized amplitude and slope averaged across neurons over 30 min are shown in Figure 2F1, F2. To confirm that associative synaptic plasticity was NMDAR-dependent, we bath-applied the NMDAR antagonist, DL-APV (10 mm) to another set of slices. Antagonism of NMDAR prevented LTP induction despite inactivation of SST-INs. EPSP amplitude and slope 30 min post induction did not significantly differ from baseline (Amplitude (mV): Pre: 4.7 ± 0.81, Post: 4.0 ± 0.73, p = 0.21; Slope (mV/ms), Pre: 0.83 ± 0.19, Post: 0.74 ± 0.18, p = 0.26; n = 10, paired t test, Fig. 2C, black circles; Normalized Amplitude: 0.88 ± 0.13, p = 0.38; Slope: 0.85 ± 0.090, p = 0.28, one-sample t test. Fig. 2D,F, black circles). These findings suggest that inhibition by SST-INs regulates NMDA-dependent associative LTP at L1b intracortical synapses.

Figure 2.

SST-IN inactivation promotes associative LTP. A1, Schematic, SST-INs express Arch. A2, SST-IN spike responses during TBS in control (black) and inactivated (green) conditions (*p < 0.05, WSR test). B, PN responses for a single TBS burst. Left, Inactivation of SST-INs enhanced PN depolarization (green vs black trace). Right, Depolarization during SST-IN inactivation (green) is reduced by APV (gray trace). C1, Average EPSP amplitude (mV) pre (−5 to 0 min) and post induction (25-30 min) with SST inactivation (magenta) or with SST inactivation plus bath application of the NMDAR antagonist, APV (black). C2, Average slope (mV/ms) of the rising phase of the EPSP pre and post induction. Colors as in B2. D, Normalized EPSP amplitude and slope following pairing with SST-IN inactivation (magenta circles). LTP is blocked by bath application APV (black). E, Representative average traces pre (gray) and post induction with SST-IN inactivation (magenta) or inactivation plus APV (Black). F1, Time course of normalized EPSP amplitude pre and post induction with inactivation of SST-INs (green box, t = 0) in control (magenta) and with bath application of APV (black). F2, Time course of normalized EPSP slope. Colors as in F1.

Inactivation of PV-INs promotes LTP

We investigated whether optogenetic inactivation of PV-INs expressing Arch also promotes associative LTP (Fig. 3A1). PV-INs are robustly activated during TBS (Fig. 3A2) and inactivation significantly decreased FR (Light OFF: 7.2 ± 2.5 Hz, ON: 2.8 ± 1.7 Hz, n = 5, p < 0.05, WSR, Fig. 3A2, right). Inactivation of PV-INs also enhanced depolarization in PNs during TBS (Fig. 3B1) and modestly increased in PN FRs (FR_ON/FR_OFF: 1.41 ± 0.17, p = 0.03, one-sample t test). The average PN FR during TBS with PV-IN inactivation was 4.85 ± 0.69 Hz. As seen for SST-INs, PV-IN inactivation during LTP induction promoted LTP of EPSP amplitude and slope (Fig. 3C1,C2, Amplitude (mV), Pre: 1.6 ± 0.34, Post: 2.4 ± 0.40, p = 0.00095; Slope (mV/ms): 0.38 ± 0.10, Post: 0.66 ± 0.18, p = 0.016; n = 9, paired t test). Normalized amplitude and slope were significantly >1 post induction (Amplitude: 1.6 ± 0.16, p = 0.0063; Slope: 1.8 ± 0.15, p = 0.00061, one-sample t test Fig. 3D–F). Antagonism of NMDAR also prevented LTP induction mediated by inactivation of PV-INs. EPSP amplitude and slope post induction did not significantly differ from baseline (Amplitude (mV): Pre: 3.3 ± 0.45, Post: 3.0 ± 0.37, p = 0.48; Slope (mV/ms), Pre: 0.44 ± 0.082, Post: 0.52 ± 0.10, p = 0.74; n = 8, paired t test, Fig. 3C, black circles; Normalized Amplitude: 0.92 ± 0.076, p = 0.65; Slope: 1.0 ± 0.13, p = 0.56, one-sample t test. Fig. 3D–F, black circles). Thus, PV-INs can also regulate the induction of NMDA-dependent associative LTP at L1b intracortical synapses.

Figure 3.

PV-IN inactivation during induction promotes associative LTP. A1, Schematic, PV-INs express Arch. A2, PV-IN spike responses during TBS in control (black) and inactivated (green) conditions. B, Representative PN responses for a single TBS for control (black) versus PV-IN inactivation (green). C1, Average EPSP amplitude (mV) pre (−5 to 0 min) and post induction (25-30 min) with PV inactivation (blue) or with PV inactivation plus bath application of the NMDAR antagonist, APV (black). C2, Average slope (mV/ms) of the rising phase of the EPSP pre and post induction. Colors as in B2. D, Normalized EPSP amplitude and slope following pairing with PV-IN inactivation (blue circles) or with inactivation plus bath application APV (black). E, Representative average traces pre (gray) and post induction with PV-IN inactivation (blue) or inactivation plus APV (black). F1, Time course of normalized EPSP amplitude pre and post induction with inactivation of PV-INs (green box, t = 0) in control (blue) and with bath application of APV (black). F2, Time course of normalized EPSP slope. Colors as in F1.

Inhibition of INs by VIP-INs

Both SST-IN and PV-INs can inhibit LTP induction, but the question remains as to which circuits provide a means for disinhibition and the promotion of LTP. In other cortices, VIP-INs inhibit SST-INs and PV-INs (Pfeffer et al., 2013; Pi, Hangya et al., 2013). Although there are numerous VIP-INs in piriform cortex (Suzuki and Bekkers, 2010b), their targets have not been identified. We crossed VIP-cre mice with Ai32 mice to express ChR2 in VIP-INs. We then used light activation to drive action potentials in VIP-INs while recording potential postsynaptic targets. Typically, VIP-cre mice are crossed with mice that express GFP in SST-INs (GIN-mice) or PV-INs (G42-mice). However, these lines sparsely label SST and PV-INs in APC (Large et al., 2016a). Instead, we used intrinsic properties to identify putative (p)SST-INs and pPV-INs according to criteria determined from recordings in genetically identified SST-INs and PV-INs in APC (Large et al., 2016a) (see Materials and Methods). INs that could not be confidently identified were excluded from analysis.

VIP-INs were activated by brief blue light pulses (5 ms), and IPSCs were recorded in voltage clamp with either Cs-gluconate-Qx internal (IPSCs, 30 mV) or K-gluconate-Qx internal (IPSCs, 0 mV). IPSC recordings were most reliable with Cs-Glu-Qx, and the data presented (Fig. 4) are from this condition. VIP-INs inhibited most recorded pSST-INs (86%) as well as pPV-INs (90%) and PNs (88%). Although high connectivity with pSST and pPV cells was expected, this was unexpected for PNs (Pfeffer et al., 2013). VIP-INs strongly inhibited pSST-INs (IPSC amplitude: 303 ± 57 pA, n = 19). However, despite a high probability of connection, VIP-INs very weakly inhibited pPV-INs (35 ± 5.8 pA, n = 9, p = 0.0008) and PNs (61 ± 14 pA, p = 0.001 n = 14, ANOVA, Fig. 4B1). Similar results were found with K-Glu-Qx internal; however, IPSC amplitude and connection probability were considerably lower (% connected, amplitude, no. connected: pSST: 73%, 47 ± 14 pA, n = 19; pPV: 73%, 31 ± 7.6 pA, n = 11; PN: 29%, 21 ± 7.0 pA, n = 4). This would be expected as voltage is not as effectively clamped using this internal compared with Cs-Glu-Qx.

Figure 4.

Inhibition by VIP-INs in piriform cortex. A1, VIP-INs express ChR2. Optically evoked IPSCs were recorded in putative(p) pSST-INs (magenta), pPV-INs (blue), and PNs (black). A2, IPSCs recorded in response to 10 light pulses (100 ms duration, 4 Hz, same neurons as in A1). B1, IPSC amplitudes were stronger in pSST-INs versus pPV-INs (**p = 0.0008) or PNs (**p = 0.001), ANOVA. B2, IPSC amplitude diminishes by the fifth pulse of theta stimulation (**p = 0.002, *p = 0.02). C1, IPSCs from VIP-INs delay pSST-IN spike responses during suprathreshold depolarization (4 overlaid traces) Left, Control. Right, Activation of VIP-INs (100 ms pulse, blue). Magenta trace represents VIP-IN-mediated IPSC during subthreshold depolarization. C2, Interspike interval (ISI) was significantly increased during optical activation of VIP-INs (blue circles, p = 0.016, paired t test, n = 11) compared with light off trials (black circles). D1, Spike responses in pSST-INs during TBS in control (magenta trace) and during pulsed light (blue trace). D2, pSST-IN FRs decreased during TBS on light trials (blue) versus control (black circles, p = 0.002, paired t test, n = 11).

We find that VIP-INs typically show strongly adapting spike responses within the first 500 ms of suprathreshold depolarization. Therefore, long light pulses (2 s) during LTP induction would likely result in significant adaptation of VIP-IN responses. Instead, we drove VIP-INs with shorter light pulses (100 ms) delivered at theta frequency (10 pulses, Fig. 4A2). This pulsed-light stimulation evoked strong and relatively stable IPSCs in pSST-INs (Fig. 4B2, top). In all cell types, IPSC strength decreased by ∼30% by the fifth pulse then stabilized (SST: 31 ± 10%, p = 0.002; PV: 31± 5%, p = 0.06; PC: 28 ± 12%, p = 0.02, paired t test, Fig. 4B2).

To investigate how VIP-IN-mediated inhibition could influence action potential activity, we depolarized target neurons to near or suprathreshold membrane potentials in current clamp. In pSST-INs, activation of VIP-INs evoked strong IPSPs (2.02 ± 0.38 mV, n = 18, Fig. 4C1). These IPSPs were sufficient to delay spikes by an average of 77 ± 26 ms (p = 0.016, n = 11, paired t test) compared with non-light trials (Fig. 4C1,C3, left). We next evoked pSST-IN spike responses with TBS and found that spike rate was significantly decreased by pulsed activation of VIP-INs in conjunction with TBS compared with non-light trials (OFF: 5.0 ± 0.6 Hz, ON: 3.5 ± 0.5, p = 0.002, n = 11, paired t test, Fig. 4C). We could not record IPSPs in PNs (n = 7) or pPV-INs (n = 4) likely because of a combination of low input resistance and relatively weak VIP-mediated IPSCs. Together, these findings suggest that VIP-INs could strongly influence pSST-INs during LTP induction but would have considerably less impact on pPV-INs or PNs.

Activation of VIP-INs promotes LTP

VIP-INs are numerous in APC and strongly inhibit pSST-INs. This suggests that a VIP-to-SST circuit motif could mediate dendritic disinhibition of PNs. So, why isn't LTP induced without antagonism of inhibition? We recorded VIP-INs during TBS stimulation of L1a. We found that, in contrast to SST-INs or PV-INs, VIP-INs were only weakly driven to fire action potentials by TBS (0.7 ± 0.6 Hz, n = 6, Fig. 5A2). This lack of VIP-IN recruitment is consistent with the inability to induce LTP under control conditions (Fig. 1B). However, light activation of ChR2⁺ VIP-INs enhanced FR during TBS (14 ± 4.0 Hz, p < 0.05, WSR test Fig. 5A2) and enhanced EPSP summation in PNs (Fig. 5B). VIP-IN activation did not alter PN FR during (FR_ON/FR_OFF: 1.02 ± 0.09, p = 0.79, one-sample t test). The overall average PN FR during TBS with VIP-IN activation was 3.42 ± 1.10 Hz. Activation of VIP-INs during induction resulted in robust LTP of EPSP amplitude and slope (Fig. 5C1,C2, amplitude (mV), Pre: 2.1 ± 0.50, Post: 3.4 ± 0.48, p < 0.02; slope (mV/ms) Pre: 0.47 ± 0.081, Post: 0.76 ± 0.072, p < 0.02, n = 7, WSR). Normalized amplitude and slope were significantly >1 post induction (Amplitude: 2.0 ± 0.33, p < 0.02; Slope: 1.9 ± 0.32, p < 0.02, WSR, Fig. 5D). Antagonism of NMDARs prevented LTP induction even with the activation of VIP-INs. EPSP amplitude and slope post induction did not significantly differ from baseline (Amplitude (mV): Pre: 6.0 ± 1.6, Post: 5.5 ± 1.5, p > 0.05; Slope (mV/ms), Pre: 1.2 ± 0.23, Post: 1.1 ± 0.28, p > 0.05; n = 5, WSR, Fig. 5E,F, black circles; Normalized Amplitude: 0.93 ± 0.072, p > 0.05; Slope: 0.99 ± 0.17, p > 0.05, WSR. Fig. 5C,E,F, black circles). Thus, activation of VIP-INs mediates the induction of NMDA-dependent associative LTP at L1b intracortical synapses. This finding, combined with the demonstration that SST-INs inhibit LTP induction and that VIP-INs strongly inhibit pSST-INs, provides strong support for a VIP-SST-PN disinhibitory circuit that gates associative LTP at intracortical synapses.

Figure 5.

VIP-IN activation promotes associative LTP. A1, Circuit schematic, VIP-INs express ChR2 and were activated using theta pulsed light during L1a+L1b pairing. A2, VIP-IN responses during TBS without (black) and with light (blue). FRs increase during pairing with light (blue circles, p < 0.05, WSR, n = 7). B, Top, Activation of VIP-INs enhanced PN depolarization during TBS stimulation (blue vs black trace). Bottom, PN depolarization during VIP-IN activation (blue trace) is reduced by APV (gray trace). C1, Average EPSP amplitude (mV) pre (−5 to 0 min) and post induction (25-30 min) with VIP-IN activation (green) or with VIP-IN activation plus bath application of APV (black). C2, Average slope (mV/ms) of the rising phase of the EPSP pre and post induction. Colors as in C1. D, Normalized EPSP amplitude and slope following pairing with VIP-IN activation (green circles) or with activation plus APV (black). E, Representative average traces pre (gray) and post induction with VIP-IN activation (green) or activation plus APV (black). F1, Time course of normalized EPSP amplitude pre and post induction with activation of VIP-INs (blue box, t = 0) in control (green) and with APV (black). F2, Time course of normalized EPSP slope. Colors as in F1.

Discussion

In the APC, recurrent intracortical excitation comprises a substantial proportion of odor responses (Poo and Isaacson, 2011). It is hypothesized that long-term ensemble representations are formed by strengthening intracortical connections between PNs coactivated by odor inputs (Haberly, 2001; Wilson and Sullivan, 2011). Consistent with this hypothesis, intracortical excitation is stronger in animals that have learned an olfactory discrimination task (Saar et al., 2002, 2012). Further, the induction of LTP at intracortical synapses in vitro is occluded following odor learning (Lebel et al., 2001). This capacity for synaptic enhancement in a highly recurrent excitatory network necessitates strong inhibition to regulate synaptic plasticity (Kanter and Haberly, 1993; Kanter et al., 1996) and stabilize neural activity (Luna and Schoppa, 2008; Poo and Isaacson, 2011; Bolding and Franks, 2018). Thus, during learning, there must be an interplay between inhibitory and disinhibitory circuitry to selectively gate the enhancement of recurrent excitation while maintaining network stability.

Intracortical excitatory synapses in piriform cortex can be strengthened by both homosynaptic and heterosynaptic forms of NMDAR-dependent LTP. From a circuit perspective, homosynaptic and heterosynaptic plasticity differs. Homosynaptic LTP is induced by strong, high-frequency tetanic stimulation of the intracortical fiber tract alone (Kanter and Haberly, 1990). Conversely, heterosynaptic associative LTP requires activation of the both the afferent fiber tract and the intracortical tract (Stripling et al., 1988; Kanter and Haberly, 1993; Poo and Isaacson, 2007). Thus, associative plasticity more directly links LTP at intracortical synapses to incoming afferent olfactory input. Moreover, associative LTP requires dendritic disinhibition by GABA_A receptor antagonists (Kanter and Haberly, 1993; Kanter et al., 1996), whereas homosynaptic plasticity does not (Kanter and Haberly, 1990). Therefore, a circuit mechanism that mediates dendritic disinhibition during odor stimulation could gate associative LTP and ensemble formation and promote odor learning.

We have previously shown that SST-INs provide substantial dendritic inhibition to PNs in APC (Large et al., 2016a). In the present study, we show that transient inactivation of SST-INs during induction protocol is sufficient to promote associative LTP. We sought a circuit mechanism that could inhibit SST-INs and ultimately gate associative LTP. Since VIP-INs inhibit SST-INs in other cortical areas (Lee et al., 2013; Pfeffer et al., 2013; Karnani et al., 2016), we investigated the connectivity of VIP-INs in APC. We found that VIP-INs strongly inhibit pSST-INs but weakly inhibit pPV-INs and PNs. Moreover, optogenetic activation of VIP-INs decreases pSST-IN spike responses during TBS of the afferent pathway. Finally, activation of VIP-INs specifically during the induction protocol is sufficient to gate associative LTP. Based on these findings, we propose that a VIP-SST-PN disinhibitory circuit gates associative LTP at intracortical excitatory synapses in APC. Our study complements a recent study in somatosensory cortex that used chemogenetics to inhibit VIP-INs or SST-INs and prevent or promote LTP induction, respectively (Williams and Holtmaat, 2019). Together, these two studies support that a common VIP-SST-PN disinhibitory circuit motif regulates synaptic plasticity across cortical areas.

Three additional circuit motifs are potentially recruited with VIP-IN activation that are worthy of discussion. We find that optogenetic inactivation of PV-INs is also sufficient to promote LTP. Therefore, an additional possibility is that VIP-INs gate LTP through a VIP-PV-PN disinhibitory circuit. However, we predict that this circuit motif contributes minimally to associative LTP induction. We find that VIP-INs only weakly inhibit pPV-INs, which is unlikely to prevent PV-IN spiking during TBS or strong network recruitment. Moreover, activating VIP-INs could actually enhance PV-IN spike activity through a VIP-SST-PV circuit. SST-INs strongly inhibit PV-INs (Pfeffer et al., 2013; Xu et al., 2013; Large et al., 2016a), and inactivation of SST-INs in APC increases pPV-IN activity (Sturgill and Isaacson, 2015). Thus, VIP-mediated inhibition of SST-INs could transiently disinhibit PV-INs and increase somatic inhibition. Enhanced somatic inhibition may compensate for a loss of SST-mediated inhibition at the dendrite and/or soma. Finally, VIP-INs also inhibit PNs directly, possibly through basket-like synapses on the soma (Suzuki and Bekkers, 2010a). As with VIP-PV synapses, this third VIP-PN circuit motif is also weak; and its role is unknown. Importantly, recruitment of these additional circuits does not impair LTP induction with VIP-IN activation, although they may be the reason that VIP-IN activation does not change PN FRs during LTP induction. The interplay of these additional circuits suggests that VIP-INs mediate a shift in inhibition from the dendrite to soma that may stabilize neural and network activity while opening a dendritic window for plasticity.

Interestingly, in the absence of optogenetic activation, VIP-INs are not recruited by pairing afferent and intracortical stimulation; this may underlie the inability induce associative LTP in slices with inhibition intact (Fig. 1). At present, it is not known how VIP-INs in APC are recruited. In other cortices, VIP-IN activity is enhanced by arousal, locomotion, task engagement, or reward (Lee et al., 2013; Pi et al., 2013; Fu et al., 2014; Jackson et al., 2016). VIP-INs receive additional excitatory drive from other cortical areas (Lee et al., 2013; Williams and Holtmaat, 2019) as well as neuromodulatory inputs (Porter et al., 1999; Lee et al., 2010; Alitto and Dan, 2012; Kuchibhotla et al., 2017; Prönneke et al., 2019). Direct excitatory inputs to VIP-INs would have to be well timed with respect to sensory stimulation. By using optogenetics to activate VIP-INs during the brief time window when afferent and intracortical pathways are coactivated, we show that transient dendritic disinhibition is sufficient to gate the cascade of intracellular mechanisms underlying synaptic enhancement. Conversely, neuromodulation could have a more sustained influence on the excitability of VIP-INs such that afferent and intracortical drive is sufficient to drive VIP-INs with appropriate timing. The APC receives excitatory inputs from higher cortices, including orbitofrontal cortex (Illig, 2005) as well as neuromodulatory centers (Zaborszky et al., 1986; Linster et al., 1999). Both of these pathways have been implicated in olfactory learning and plasticity (Patil et al., 1998; Patil and Hasselmo, 1999; Linster et al., 2003; Li et al., 2006; Chapuis and Wilson, 2013; Cohen et al., 2015; Strauch and Manahan-Vaughan, 2018). Future studies are needed to evaluate the links between descending excitation and/or neuromodulation, the recruitment of VIP-INs, and the gating of synaptic plasticity during olfactory learning.

Finally, many previous studies have focused on the influence of disinhibitory circuits on increasing PN FRs or decreasing IN FRs in vivo (Pi et al., 2013; Fu et al., 2014; Karnani et al., 2016). We found that VIP activation had a robust effect on SST-IN FRs during the LTP induction protocol. However, PN FRs were not changed by activation of VIP-INs and only modestly affected by inactivation of SST-INs or PV-INs. This suggests that substantial changes in PN FRs are not required for LTP induction as long as the dendrite is disinhibited. Our findings highlight a role for the VIP-SST-PN disinhibitory circuit in gating the subthreshold synaptic and cellular interactions that give rise to long-term plasticity independent of changes in PN FRs. Since we designed our stimuli to minimize the influence of spiking activity on LTP induction, we cannot entirely rule out a role for backpropagating spikes in promoting LTP (Johenning et al., 2009). In the intact system, it is conceivable that both dendritic disinhibition and backpropagation influence LTP induction. Alternatively, additional circuit motifs whereby VIP-INs disinhibit PV-INs or inhibit PNs directly could simultaneously support inhibitory stabilization at the soma. It is difficult to delineate the roles for individual circuit motifs that are nested in complex neural networks. The challenge going forward is to determine how multiple circuit mechanisms work in concert during odor processing to promote olfactory learning.