Somatostatin-expressing interneurons modulate neocortical network through GABAb receptors in a synapse-specific manner

The firing activity of somatostatin-expressing inhibitory neurons (SST-INs) can suppress network activity via both GABAa and GABAb receptors (Rs). Although SST-INs do not receive GABAaR input from other SST-INs, it is possible that SST-IN-released GABA could suppress the activity of SST-INs themselves via GABAbRs, providing a negative feedback loop. Here we characterized the influence of GABAbR modulation on SST-IN activity in layer 2/3 of the somatosensory cortex in mice. We compared this to the effects of GABAbR activation on parvalbumin-expressing interneurons (PV-INs). Using in vitro whole-cell patch clamp recordings, pharmacological and optogenetic manipulations, we found that the firing activity of SST-INs suppresses excitatory drive to themselves via presynaptic GABAbRs. Postsynaptic GABAbRs did not influence SST-IN spontaneous activity or intrinsic excitability. Although GABAbRs at pre- and postsynaptic inputs to PV-INs are modestly activated during cortical network activity in vitro, the spontaneous firing of SST-INs was not the source of GABA driving this GABAbR activation. Thus, SST-IN firing regulates excitatory synaptic strength through presynaptic GABAbRs at connections between pyramidal neurons (Pyr-Pyr) and synapses between pyramidal neurons and SST-INs (Pyr-SST), but not Pyr-PV and PV-Pyr synapses. Our study indicates that two main types of neocortical inhibitory interneurons are differentially modulated by SST-IN-mediated GABA release.

The compound functions of the neocortex depend on neuronal microcircuits of highly interconnected glutamatergic (excitatory) neurons and GABAergic (inhibitory) interneurons. The GABAb receptors (GABAbRs) are G-coupled metabotropic receptors that are expressed pre-and postsynaptically, where they inhibit neuronal activity via the modulation of the calcium and potassium channels, respectively. GABAbRs can be found on both excitatory and inhibitory cells, however, the distribution and function of GABAbRs on specific neuronal populations and their inputs and outputs are not well-characterized, especially for molecularly and anatomically diverse inhibitory neuron subtypes that have critical roles in shaping network output. Among these interneurons, somatostatin-expressing and parvalbumin-expressing interneurons (SST-INs and PV-INs, respectively) compromise more than half of the inhibitory interneurons in the mouse neocortex 1, 2 . It has been believed that PV-INs target mainly soma, proximal parts of dendrites or the axon initial segment 3 , whereas SST-INs are considered to innervate primarily (but not exclusively) distal parts of pyramidal (Pyr) neurons 4 . Interneurons show high basal firing activity that can be regulated by brain state, behavioral tasks, and also during learning [5][6][7] . The activity of neocortical PV-and SST-INs is regulated in a characteristic and often opposing manner 8 . For example, in the mouse barrel cortex, L2/3 SST-INs are spontaneously active during quiet wakefulness, whereas their activity is reduced during both passive and active whisker movements 8,9 . In contrast, PV-INs fire spontaneously during the quiet wakeful state and are profoundly activated by whisker sensing 8 . How brain state regulates the activity of these two interneurons is the subject of intense investigations.
SST-INs both directly and indirectly control neocortical network activity. Prior studies indicate that in vivo optogenetic silencing of SST-IN firing paradoxically increases the activity of neighboring Pyr neurons 8 . Our previous in vitro study identified a synaptic mechanism that may underlie this phenomenon, showing that SST-IN spontaneous activity strongly silences excitatory synaptic transmission between L2/3 Pyr neurons in mouse somatosensory cortex 10 . The effect is mediated by the presynaptic GABAbR activation, which reduces www.nature.com/scientificreports/ without a negative feedback loop. To these ends, it is significant that SST neurons do not synapse onto each other, like other interneurons subtypes in the cortical network 23 . , although synaptic transmission from these connections is generally weak 18,[24][25][26] . Excitatory connections between Pyr neurons can be silenced by presynaptic GABAbRs, and the spontaneous firing of SST-INs is sufficient to suppress Pyr to Pyr (Pyr-Pyr) communication 10 .
Here, we decided to examine whether spontaneous activity of SST-INs could also regulate excitatory drive onto SST-INs themselves. This is important, because it would indicate that when SST-IN activity is high (perhaps due to control by neuromodulators) they are unresponsive to sensory drive or local computations. First we determined whether L2/3 Pyr-SST synapses could be regulated by GABAbRs. Using dual patch-clamp recording, we established synaptically connected Pyr-SST pairs in L2/3 of the somatosensory cortex in a mouse brain slice in mACSF ( Fig. 3G-I). Bath application of the GABAbR blocker CGP reduced failure rates of Pyr-SST EPSPs by 26% ( Fig. 3I; from 0.73 ± 0.18 to 0.54 ± 0.29; n = 7, *p = 0.045, paired t-test) but did not change EPSP   (Fig. 4) using transgenic expression of the hyperpolarizing pump (Arch) in SST-Cre mice. Using paired whole-cell recordings of synaptically connected Pyr-SST, we tested whether acute silencing of SST-IN spontaneous activity might enhance EPSP strength and reliability of these connections. Illumination of the brain slice with yellow-green light (LED) for 1-1.5 s fully suppressed SST-IN firing, hyperpolarizing their membrane potentials by about 10-20 mV. During the light ON period, Pyr-SST EPSP failure rates were significantly reduced compared to OFF trials ( Fig. 4D; 0.89 ± 0.11 mV in OFF and 0.77 ± 0.23 mV in ON, n = 13, *p = 0.022, paired t-test) and EPSP amplitude was not changed (  To confirm that this effect was due to GABAbRs, we analyzed the effect of SST-IN silencing when GABAbRs were blocked by CGP (Fig. 4E,F). Under these conditions, SST cell silencing did not change EPSP failure rates Using paired whole-cell recordings of connected SST and Pyr neurons (SST-Pyr), we compared the effects of GABAbR agonist (baclofen) on IPSC. Bath application of baclofen decreased the first IPSC amplitude by 84% (Fig. 5A,B; control 79.94 ± 46.49 pA vs. baclofen 13.09 ± 5.35 pA; n = 5, *p = 0.031, paired t-test) but had no effect on failure rates ( Fig. 5C; 0.00 ± 0.00 in control to 0.04 ± 0.05 in baclofen, n.s. p = 0.5, Wilcoxon test). Wash-on of CGP reversed the effects on IPSC amplitude to control values ( Fig. 5B; 72.87 ± 29.12 pA in CGP, n.s. p = 0.654 ctr vs. CGP, paired t-test). There was no effect on failure rates, since synaptic efficacy was high at these connections ( Fig. 5C; CGP 0.0 ± 0.0, n.s. p = 0.5, ctrl vs. CGP, Wilcoxon test). If there are presynaptic GABAbRs at SST-IN terminals, we predicted that the GABAbR agonist baclofen would increase the paired-pulse ratio of the second response to the first (PPR) at SST-Pyr synapses. In superficial layers of the neocortex, SST-Pyr synapses are typically depressing, but this was shifted to a mild facilitation after the application of baclofen ( Fig. 5D; control 0.51 ± 0.20 versus baclofen 1.10 ± 0.15; *p = 0.017, paired t-test), indicating a presynaptic locus of drug action. Subsequent bath application of the GABAbR antagonist CGP reversed this facilitation back to control levels ( Fig. 5D; 0.56 ± 0.11 in CGP, n.s. p = 0.661, paired t-test), suggesting that GABAbR activation was negligible under our recording conditions and consistent with the results obtained for SST-INs in the hippocampus 29 .
Because the agonist was effective at reducing neurotransmitter release at SST-Pyr synapses, we asked whether presynaptic GABAbRs might sometimes be activated by the synapse's own GABA release. In this case, the amplitude of the IPSC might be suppressed when neocortical SST neurons fire at some regular frequency, as has been described in vivo during the quiet resting state 8,20 and also in vitro, in mACSF 10, 15, 18 . We thus examined the   www.nature.com/scientificreports/ IPSC of connected SST and Pyr neurons in mACSF ( Fig. 5E-H). Under baseline conditions, the evoked IPSC had a relatively low failure rate (0.18 ± 0.21, n = 12, Fig. 5G) indicating that GABAaR-dependent inhibition from SST-INs is strong even when spontaneous network activity is high. Bath application of CGP did not alter IPSC amplitude ( Fig. 5F Importantly (and in contrast to Pyr neurons), we did not observe any changes in resting membrane potential nor input resistance of SST-INs when GABAbRs were either activated or blocked pharmacologically (Table 1). These data indicate that GABAbRs in neocortical SST-INs are unlikely to act through potassium channels, similarly what has been observed for SST-INs in the hippocampal network 30 .
The reduction in PV-IN excitability was associated with significant GABAbR modulation of Vrest. Resting membrane potential was hyperpolarized by ~ 2 mV by baclofen and depolarized by ~ 2 mV with CGP, and input resistance was decreased by baclofen and increased with CGP (Table 1). In contrast, both parameters in SST-INs showed no significant difference between control and drug applications (Table 1). These data suggest that postsynaptic GABAbRs are coupled to potassium channels in PV-but not SST-INs 30 .

GABAbR-mediated suppression of PV-IN output onto Pyr neurons. GABAbRs on PV-INs might
primarily control the intrinsic membrane properties of these cells, or they also can directly regulate synaptic release. To determine whether L2/3 PV-IN terminals have presynaptic GABAbRs, we examined the effects of GABAbR antagonist on PV-IN-mediated IPSCs in Pyr neurons using paired whole-cell recordings (Fig. 7). PV-Pyr connections are extremely abundant in L2/3 1, 28 , and we also observed that the probability of PV-Pyr connections reached 82% (9 connected pairs out of 11 tested). When network activity was high (in mACSF), we observed that IPSC failure rates were very low (0.05 ± 0.08 pA, n = 8, Fig. 7C), indicating that the efficacy of PV-Pyr synapses was very high even in the presence of high spontaneous firing of SST-INs. Bath application of the GABAbR antagonist CGP increased IPSC amplitude by 48% (Fig. 7A,B; control 51.09 ± 37.44 pA vs. CGP 69.70 ± 51.36 pA, n = 8, *p = 0.035, paired t-test). Because failure rates were already negligible, CGP had no effect ( Fig. 7C; 0.00 ± 0.00 in CGP, n.s. p = 0.250, Wilcoxon test). PPR decreased from 0.89 ± 0.11 in control to 0.76 ± 0.19 in CGP ( Fig. 7D; n = 8, *p = 0.043, paired t-test). The effects of CGP on IPSC amplitude and PPR indicate that fast inhibition from PV-INs is modulated by tonic activity of presynaptic GABAbRs under spontaneous network activity in acute brain slices. www.nature.com/scientificreports/ Using dual patch-clamp recordings, we tested whether acute silencing of SST-INs could change EPSP strength and reliability at both Pyr-PV and PV-Pyr synapses. SST-INs were silenced for 1 s before activation of Pyr-PV synapses (Fig. 8A,B). Surprisingly, SST-IN silencing had no effect on EPSP amplitude at Pyr-PV connections ( Both light and electron microscopic analysis of GABAbR immunoreactivity shows that these receptors are expressed in cell bodies, dendrites and terminals of most types of neurons in the neocortex and the hippocampus; however, the net effect of GABAbRs will depend upon the specific ion channels that they are linked to, leading to different effects on network function and connectivity. Canonically, postsynaptic GABAbRs activate G-proteincoupled inwardly-rectifying potassium channels (such as Kir-3), leading to hyperpolarization of the neuronal membrane and thus decreasing neuronal activity [33][34][35] . However, consistent with the analysis of hippocampal SST-INs 30 , we observed a moderate but not significant alteration in membrane potential and input resistance of www.nature.com/scientificreports/ L2/3 SST-INs after GABAbR agonist and antagonist administration, suggesting that neocortical SST-INs possess postsynaptic GABAbRs which are not strongly coupled to Kir channels. It is possible that GABAbRs might be co-clustered with and inhibit L-type channels as observed in hippocampal SST-INs 30 . In contrast to SST-INs, we found significant changes in the membrane potential and input resistance in L2/3 PV-INs and Pyr after the application of GABAbR agonists and antagonists, indicating that postsynaptic GABAbRs activate potassium channels on these neurons. Thus, the cell-type specific differences in pharmacological effects on resting membrane potential and input resistance between these cell types are due to the localization of Kir channels that are sensitive to GABAbR modulation. Presynaptic GABAbRs, which can suppress the release of both glutamate and GABA, are widely distributed across the brain. However, it is unclear whether specific interneuron subtypes possess GABAbRs that modulate synaptic release. Immunoreactivity for presynaptic GABAbRs has been observed on hippocampal interneurons expressing SST, neuropeptide Y, calretinin, calbindin, or cholecystokinin (CCK) but not on PV-INs 36 . Other studies that used electrophysiological recordings and freeze-fracture replica-based quantitative immunogold electron microscopy have revealed the presence of presynaptic GABAbRs on hippocampal CCK and PV-INs 14 . Fast, GABAaR inhibition from PV to Pyr neurons is weakly reduced by presynaptic GABAbRs in the hippocampus 14,37,38 ; however, these effects may be region-specific since GABAa-mediated inhibition from L5 fast-spiking (presumably PV) interneurons was strongly modulated by GABAbRs 39 . In developing gerbil auditory cortex, GABAaR-mediated inhibition from L2/3 FS (presumably PV-INs), but not from LTS (presumably SST-INs) is controlled by presynaptic GABAbRs 40 . Our study showed that despite the fact that both SST-and PV-INs have presynaptic GABAbRs (autoreceptors), only the fast inhibition from PV-INs is tonically suppressed through GABAbRs, whereas fast inhibition mediated by SST-INs appears insensitive to GABAbR tonic modulation. Thus, mechanisms for presynaptic GABAbR modulation are region-specific, differing between hippocampal and neocortical interneurons.

Tonic GABAbR activity in PV-
In vitro recordings show that SST-INs can exhibit high levels of spontaneous activity that is independent of glutamatergic input or electrical stimulation 15 . Indeed, we find that local synaptic drive to SST-INs under our in vitro recording conditions is markedly low, consistent with ours and others' previous work rev.; 23,41 . Here we show that optogenetic silencing of SST-INs in these active network states enhanced synaptic input from Pyr to SST-INs, an effect that was dependent upon GABAbRs. These data suggest that SST-IN inhibition may exist in two different regimes: one, where spontaneous SST-IN activity is temporally imprecise and has a more global effect on reducing functional coupling within the network, and the second where spontaneous activity of SST-INs is low and can be temporally synchronized to provide fast, feedback inhibition in the local network. Spontaneous activity of SST-INs thus reduces network activity in two separate ways: first, via direct and fast GABAaR-dependent inhibition of downstream targets and second, via slow and indirect suppression of local synaptic transmission.
The conditions that enable spontaneous activity of SST-INs, where they are a significant source of ambient GABA that silences not only local excitatory transmission between Pyr neurons 10 but also excitatory synapses onto SST-INs, will be of great interest to identify. It remains unknown under what brain states presynaptic GABAbR are activated in SST-IN terminals in the neocortex; it is reasonable to hypothesize that this might happen during the states in which SST-IN activity is higher than during the quiet state.
Notably, GABAbR-mediated inhibition by SST-INs is not global, as synaptic efficacy between Pyr neurons and PV-INs was not changed with optogenetic suppression of SST-IN activity. The second important finding of our study shows that fast synaptic inhibition mediated by SST-INs on to local Pyr cells is not suppressed by the tonic activity of presynaptic GABAbRs. These data indicate that SST-INs might provide effective inhibition to the local network under high network activity conditions, in contrast to PV-INs which fast inhibition can be suppressed by tonic activity of GABAbRs. Future studies will reveal the specific cellular source of GABA responsible for activating GABAbR on SST-and PV-IN terminals.
Inhibitory interneurons play an essential role in controlling cortical activity at various temporal and spatial scales 42 . SST-INs preferentially inhibit distal parts of Pyr neurons whereas PV-INs synapse on the soma and proximal parts of the target neurons 3, 4, 43 , regulating inputs and outputs of excitatory neurons, respectively. Here, we provide a functional comparison for the sensitivity of two main neocortical interneurons in the activation of GABAbRs. Under conditions of high GABA release, the differential sensitivity of SST-and PV-INs to GABAbR modulation will favor fast (GABAaRs) inhibition mediated by SST-over PV-INs. Our data show that the weak excitatory synaptic input onto SST-INs from local Pyr cells can be further reduced by the spontaneous activity of SST-INs. However, SST-INs can still generate powerful GABAa-mediated inhibition thanks to their intrinsic spontaneous activity that is insensitive to GABAbRs.
Because Pyr neurons receive not only local inputs from neighboring Pyr neurons within L2/3 but also from other cortical layers 24,25,28,44 as well as other brain areas [45][46][47] , GABAbR-mediated silencing of local inputs might shift the balance of control of L2/3 Pyr neurons from local to long-range inputs. Consistent with this, prior studies have shown that thalamocortical inputs to L4 excitatory neurons are resistant to GABAbR modulation 48 but see 39 . Thus, our data suggest that when SST-IN activity is high, local circuit amplification of incoming sensory information will be weak.
Pharmacological experiments for induction of synaptic plasticity in acute brain slices have yielded mixed results about the role of GABAbRs. For example, in the hippocampus, GABAbRs prevent the induction of longterm potentiation (LTP) in SST-INs 30 but can facilitate the induction of LTP in CA1 Pyr cells 49,50 , indicating that the effects of these receptors may be synapse-specific and also implicating these receptors in synaptic plasticity during learning and memory. In layer 4 (L4) of primary visual cortex of rats, the strengthening of synapses from FS cells (presumably PV-IN) to Pyr neurons requires the activity of postsynaptic GABAbRs, although LTP between L4 excitatory cells was not affected by GABAbR antagonists 51 . Critically, these pharmacological manipulations may not precisely recapitulate the tonic activation of GABAbRs under more naturalistic activity conditions. It remains unknown whether LTP at other excitatory synapses is sensitive to GABAbR activation, and www.nature.com/scientificreports/ which interneurons may regulate plasticity at input-and target-specific synapses. Interestingly, recent studies have reported that dually innervated dendritic spines on hippocampal pyramidal neurons are resistant to structural plasticity due to tonic inhibition of NMDA receptors through GABAbRs and SST-INs 52 . Taken together, our data demonstrate that GABAbRs modulate neocortical networks in a neuron-and synapse-specific manner ( Supplementary Fig. 2). The differential sensitivity of specific synapses and neurons to GABAbR-modulation may fine-tune the balance of excitation and inhibition in a compartment-specific manner, providing tight control of information flow to Pyr neurons in superficial layers.
Whole-cell recording. To enable spontaneous firing of SST-INs and tonic activation of GABAbRs, recordings were performed in modified ACSF (mACSF) solution composed of (in mM): 119 NaCl, 3.5 KCl, 0.5 MgSO 4 , 1 CaCl 2 , 1 NaH 2 PO 4 , 26.2 NaHCO 3 , 11 glucose equilibrated with 95/5% O 2 /CO 2 , as described before 10,18 . To prevent the network from spontaneous firing, a part of experiments were performed in regular ACSF (rACSF) where the following ingredients were changed to 2. Because of the difficulty in identifying connected pairs, the majority of recordings were carried out at room temperature (24 °C) to enable longer recording periods, since prolonged incubation at warmer temperatures degrades cell health and diminished recording quality.
Electrophysiological data were acquired by Multiclamp 700A or Multiclamp 700B (Molecular Devices) and digitized with a National Instruments acquisition interface or Digidata 1550B (Molecular Devices). The data were filtered at 3 kHz, digitized at 10-20 kHz and collected by Igor Pro 6.0 (Wavemetrics) or pClamp (Molecular Devices). Series and input resistances were analyzed online, recordings were discarded when access or series resistances were unstable more than 30%.
Neuron classification. Neurons were classified as Pyr neurons according to Pyr-like soma shape, the presence of an apical dendrite and spines visible after Alexa filling reconstruction as well as according to regular spiking in response to 500 ms suprathreshold intracellular current injection. SST-IN were identified using fluo- www.nature.com/scientificreports/ rescent reporter gene expression in Sst-Cre::Ai14, whereas PV-INs were visualized by the fluorescent marker genetically encoded in Pvalb-T2A-Cre-D::Ai14 or PValb-tdTomato-miGAD67 mice. Additionally, the firing responses to the somata current injection were analyzed. SST-INs responded to current steps with low threshold spiking (LTS) firing with a spike-rate adaptation, and the AHP after the first action potential was more negative than the last AHP during the current step 15 . PV-INs were identified by fast-spiking (FS) firing with a nonadapting firing pattern and AHP magnitudes that were uniform throughout a given current step.

Connectivity analysis.
Synaptically connected neurons were identified by paired whole-cell patch-clamp recordings. The distance between cells was ≤ 300 μm. To evaluate Pyr to SST-IN connections (Pyr-SST) or Pyr to PV-IN (Pyr-PV), both cells were maintained in current clamp mode and EPSPs were recorded in SST-or PV-IN in response to the current injection to the presynaptic Pyr. Interneuron membrane potential was maintained at the hyperpolarized value about − 60 mV to block spontaneous spiking activity of the recorded cell and to visualize EPSPs. To evaluate inhibitory interneuron to Pyr connections, the membrane potential of an interneuron was maintained in current clamp, whereas the postsynaptic cell was kept in voltage clamp at the holding potential of 0 mV to record IPSCs in the response to the presynaptic spikes evoked by the current injection. Alternatively, IPSPs were recorded in current clamp mode at the depolarizing membrane potential of about − 55 mV. To assess connectivity, we analyzed responses from 20 trials delivered at 0.1 Hz of a 10-stimulus spike train (3-5 ms long, 1 nA at 20 Hz). The 10-spike train was critical for accurate assessment of synaptic connections, since weak and facilitating connections (such as at Pyr to SST-IN synapses) cannot be detected with single spikes 28 . Evoked postsynaptic responses were calculated using responses from 10 trials of the stimulus train. To precisely identify the onset of the synaptic response, analysis was focused on events that occurred during DOWN-states, since responses during UP-states were difficult to isolate from background activity in the sweep 54 . Because in vivo, L2/3 Pyr neurons rarely respond more than a single spike and almost never at frequencies exceeding 20 Hz 55 , we considered responses to the first presynaptic spike to be most representative of synaptic function as it might occur during normal sensory-evoked activity. Thus, amplitude and failure rate measurements are plotted for the EPSP (or IPSC/P) evoked by the first stimulus in the spike train. Spontaneous excitatory postsynaptic currents (sEPSCs) were recorded in voltage clamp at − 70 mV. Spontaneous firing was recorded at the resting membrane potentials without any correction. Spontaneous activity was analyzed within at least 3 min of current-clamp recordings in control ACSF and ACSF with agents.
Intrinsic excitability was accessed using square pulses of 500 ms of increasing amplitude up to maximal firing frequency (steps of 10 pA and 20 pA, for SST-and PV-INs, respectively). To control for potential effects of GABAbR agents on Vrest, the membrane potential of interneurons was maintained at − 65 mV across different pharmacological conditions. Pharmacology. The GABAb receptor antagonist (CGP 55845, 1 μM) and agonist (baclofen, 10 μM), as well as the AMPAR antagonist (DNQX, 20 μM) and NMDAR antagonist (APV, 50 μM) were bath applied for at least 10 min before data acquisition to assess drug effect for 20 trails of the 10-pulse train. All the pharmacological agents were purchased from Tocris.
Optical stimulation. For ArchT activation, photo stimulation was produced by a light-emitting diode (white LED with 590 nm filter set to maximum range, Prizmatix, Israel) and delivered through a 40 × waterimmersion objective. After establishing a connection, SST-IN silencing was initiated 1 s or 1.5 s prior to the 10 pulse presynaptic train and maintained to the end of the spike train. Because the light hyperpolarized the membrane potential of SST-INs, the light ON trials were collected before light OFF trials (20 repetitions at 0.1 Hz for both periods). The light OFF trails were collected at the membrane potential of SST-INs adjusted by the constant somata current injection to the same value as it was during light ON period. Data analysis. Population data are presented as mean ± SD. 1-2 cells or 1 synaptically connected pair were analyzed in an individual mouse. Statistical significance was defined as p < 0.05 using a two-tailed paired t-test or Wilcoxon test depending on the normality distribution. The normality distribution was tested with the Shapiro-Wilk test and equal variance was analyzed with Brown-Forsythe test. For intrinsic excitability, a plot of the relation between the number of action potentials and the intensity of the injected current (I-F curve) was created for every neuron in control ACSF and after the drug application. The effect of a drug on excitability was analyzed as the difference in the rheobase, the maximal firing frequency and the difference of the spiking frequency at the same current steps in the comparison to control ACSF.

Data availability
The data and material that support the findings of this study are available upon request to the corresponding author.