Disinhibition-assisted long-term potentiation in the prefrontal-amygdala pathway via suppression of somatostatin-expressing interneurons

Abstract. Significance: Natural brain adaptations often involve changes in synaptic strength. The artificial manipulations can help investigate the role of synaptic strength in a specific brain circuit not only in various physiological phenomena like correlated neuronal firing and oscillations but also in behaviors. High- and low-frequency stimulation at presynaptic sites has been used widely to induce long-term potentiation (LTP) and depression. This approach is effective in many brain areas but not in the basolateral amygdala (BLA) because the robust local GABAergic tone inside BLA restricts synaptic plasticity. Aim: We aimed at identifying the subclass of GABAergic neurons that gate LTP in the BLA afferents from the dorsomedial prefrontal cortex (dmPFC). Approach: Chemogenetic or optogenetic suppression of specific GABAergic neurons in BLA was combined with high-frequency stimulation of the BLA afferents as a method for LTP induction. Results: Chemogenetic suppression of somatostatin-positive interneurons (Sst-INs) enabled the ex vivo LTP by high-frequency stimulation of the afferent but the suppression of parvalbumin-positive interneurons (PV-INs) did not. Moreover, optogenetic suppression of Sst-INs with Arch also enabled LTP of the dmPFC-BLA synapses, both ex vivo and in vivo. Conclusions: These findings reveal that Sst-INs but not PV-INs gate LTP in the dmPFC-BLA pathway and provide a method for artificial synaptic facilitation in BLA.

1 Introduction low-frequency presynaptic stimulation, respectively. 3 With optogenetic stimulation, this simple approach proved successful in several circuits ex vivo and in vivo. The 20 Hz and theta-burst stimulation produced long-term potentiation (LTP) in the recurrent synapses of the hippocampal area CA3 4 and corticostriatal synapses, 5 respectively. In contrast, the low-frequency stimulation produced long-term depression (LTD) in the inputs to the nucleus accumbens and basolateral amygdala (BLA) from the infralimbic cortex. 6,7 However, not all synapses follow the frequency rule. For example, the high-frequency stimulation generated LTD in inputs from the BLA to the dorsomedial prefrontal cortex (dmPFC) 8 and failed to produce LTP in the prefrontal-amygdala synapses. 9 Evidence accumulates that the strength of remote synaptic inputs to the BLA is a critical determinant of fear behaviors and readily changes by emotional experiences. Specifically, auditory fear conditioning, in which the animal experiences a neutral conditioned stimulus followed by electrical footshocks as the unconditioned stimulus (US), facilitates inputs in the lateral subdivision of BLA from the sensory cortex and sensory thalamus. 10,11 Conversely, optogenetically induced LTD and LTP in the same pathway deactivate and reactivate fear memories, respectively. 12 Besides the sensory inputs, BLA receives abundant inputs from the dmPFC, 13 and the properties of the dmPFC-BLA reciprocal circuit have been correlated with distinct emotional states. Specifically, the fear extinction training has been found to depress the prefrontalamygdala synapses and to strengthen the reciprocal amygdala-prefrontal synapses. 14,15 The pattern of neuronal synchronization between dmPFC and BLA has been found to determine whether the animal expresses defensive behaviors or feels safe, 16,17 suggesting that the strength of synaptic connections in the dmPFC-BLA reciprocal circuit regulates the emotional state. Testing this prediction would require artificial facilitation or depression of the prefrontalamygdala synapses. However, there are no reports of artificially induced LTP in the prefrontal-amygdala synapses, likely because the local GABAergic neurons provide potent feedforward inhibition and gate plasticity in the remote glutamatergic inputs. 18,19 Meanwhile, the robustness of the "natural" behavior-driven plasticity is explained by the US actions to disinhibit the BLA by attenuating GABAergic transmission via multiple mechanisms, including the secretion of neuromodulators. [19][20][21] Here, to achieve reliable artificial LTP inductions in the BLA input from dmPFC, we tested the effects of chemogenetic/optogenetic suppression of GABAergic transmission during the LTP induction by high-frequency stimulation. The two major classes of GABAergic neurons-the parvalbumin-positive interneurons (PV-INs) and somatostatin-positive interneurons (Sst-INs) 22 were suppressed individually, which revealed that the Sst-INs gate the artificially induced LTP.

Animals
All mice were either wild type or transgenic males on the 129SvEv/C57BL/6N F1 hybrid background. The experiments were limited to the male mice, to avoid potential sex-dependent LTP variability, which has been reported both in the amygdala 23,24 and hippocampus. 25,26 To obtain the mice-expressing hM4Di 27 in Sst-INs or PV-INs, homozygous R26-LSL-Gi-DREADD males (JAX Stock No: 026219) on C57BL/6N background were crossed with homozygous interneuron-specific Cre driver females on 129SvEv background-either the Sst-IN-specific Cre driver (Sst-Cre), Sst tm2.1ðcreÞZjh28 (JAX: 013044) or the PV-IN-specific Cre driver (PV-Cre), Pvalb tm1ðcreÞArbr29 (JAX: 008069). The expression specificity of floxed reporters in the BLA of these driver lines has been validated using immunostaining and wholecell recordings by the Luthi lab. 22 To obtain heterozygous Sst-Cre mice, wild type C57BL/6N males were crossed with the 129SvEv homozygous Sst tm2.1ðcreÞZjh females. All breeding were the trios of one male and two females on the C57BL/6N and 129SvEv backgrounds, respectively. Male pups were weaned at p21-p25 and housed three to five littermates per cage. All experiments were approved by Virginia Tech IACUC and followed the NIH Guide for the Care and Use of Laboratory Animals.

Surgery-Viral Injection and Optrode Implantation
Adenoassociated viruses (AAVs) for expressing Chronos or Cre-activated Arch were generated from pAAV-Syn-Chronos-GFP 30 (Addgene #59170) or pAAV-FLEX-Arch-GFP (Addgene #22222), respectively, gifts from Edward Boyden. The viruses (pseudotype 5 for Chronos and pseudotype 1 for Arch) were prepared by the University of North Carolina Vector Core (Chapel Hill, North Carolina). At p28, the heterozygous Sst-Cre male mice were anesthetized by intramuscular injection of ketamine/xylazine/acepromazine, 100∕5.4∕1 mg∕kg, as routinely done for electrode implantations in the Buzsaki lab, 31 in the volume not exceeding 0.05 ml, placed in a stereotaxic apparatus (David Kopf, Tujunga, California) and underwent minimum craniotomy (∼0.5 mm diameter). For the dmPFC virus injection, the dura mater was preserved. A heater-pulled short-taper glass pipette (shaft: 0.6∕0.4 mm external/internal diameter, beveled tip: 50 μm, diameter, Drummond, Broomall, Pennsylvania) filled with the virus solution (10 12 viral particles∕ml) was slowly lowered to the target (1.3 mm anterior, 0.4 mm lateral from bregma, and 1.3 mm ventral from brain surface). The solution (0.5 μl) was injected bilaterally at the rate of 0.2 μl∕ min using a syringe pump connected to the pipette through plastic tubing filled with water as described. 32 For the BLA virus injections, the dura mater was removed to allow straight penetration by a less rigid long-taper pipette. The virus solution (0.4 μl, 10 12 particles∕ml) was injected bilaterally at the rate of 0.1 μl∕ min at the coordinates (1.2 mm posterior, 3.2 mm lateral from bregma, and 4.2 mm ventral from brain surface). Custom-made optrodes were fabricated with a miniature LED [C460EZ500, CREE, peak wavelength 463.3 nm, bandwidth 20 nm full-width half maximum (FWHM)], coupled to an optical fiber (0.66 NA, 0.2-mm core diameter, Prizmatix) with UV-cured index-matched glue (Norland Optical Adhesive 85, Norland) and two 33-μm tungsten wires (polyimide-insulated, California Fine Wire) extended 0.5 mm beyond the optical fiber. The optrodes for in vivo recording were implanted in BLA at p60, as described. 33 For postoperation analgesia, ketoprofen (5 mg∕kg) was administered subcutaneously.

LTP
In both the whole-cell and LFP recording, the strength of test pulses (1 ms duration) was adjusted to elicit responses at 30% to 40% of the maximum. In the whole-cell recordings, test pulses were given every 30 s. LTP was induced by six 2-s trains of 50 Hz and 1 ms pulses. The trains were given at the 10-s interval [ Fig. 2(a)]. In the LFP recordings, test pulses were given every 20 s. LTP was induced using the "spaced protocol." It included pairs of 1-s trains of 50 Hz and 1 ms pulses, separated by 10 s. The pairs were repeated five times at the 3-min interval [ Fig. 3(a)]. This protocol is the same as in a published study on LTP in BLA, 36 except the stimulation frequency was decreased from 100 to 50 Hz to allow reliable activation of Chronos. 30 In some experiments, continuous yellow light was given during the trains of the blue light pulses. The yellow light strength was set below the levels that trigger the release of glutamate from the axonal terminals expressing Chronos (data not shown).

In Vivo Recordings
The subject animals, bilaterally injected with the AAV-Chronos in the dmPFC, AAV-Arch in BLA, and bilaterally implanted with the optrodes in the BLA, were housed with the littermates until the experiment. Using the RHA2000-Series Amplifier USB Evaluation Board (RHA2000-EVAL, Intan Technologies), the local field potentials (LFPs) were recorded from BLA of the subject mouse in the home cage without the lid, from where the cagemates were removed temporarily for the duration of the recording. Mice were habituated to the recording environment by connecting to the recording system for 2 to 3 h per day during 2 to 3 consecutive days. Field excitatory postsynaptic potentials (fEPSPs) were elicited in BLA by blue light stimulation of dmPFC terminals expressing Chronos. The strength of the test pulses (1 ms, 2 to 3 mW at the tip of optrode) was adjusted to obtain the fEPSP slope at 30% to 40% of the maximum. The LED driver (PlexBright LD-1, Plexon) was analog-modulated by DAQ (Analog Shield, Digilent). The LED driving current was routed to the optrodes in either hemisphere by electrical relays (Arduino 4 relays shield, Arduino). Arduino with a custom Arduino sketch controlled both the DAQ and the relays to give the light stimulation on each side every 30 s, alternating the sides every 15 s. Once the baseline of evoked fEPSPs stabilized, LTP was induced by the same blue light stimulation protocol as in the LFP and LTP experiments ex vivo [ Fig. 3(a)], except that the protocol was repeated three times for every 1 h. The positions of the optrodes were confirmed by histological analysis.

Data Analysis
Data were processed using custom scripts written in MATLAB (MathWorks) and Clampfit software (Molecular Devices). Statistical analyses were performed using GraphPad Prism 5 (GraphPad Software, La Jolla, California). Normality was tested using the Shapiro-Wilk test. Datasets with normal distribution were compared using the one-sample t-test. The datasets with non-normal distribution were analyzed using the Mann-Whitney test and the Wilcoxon's signed rank test. The difference was deemed significant with p < 0.05.

DREADD-hM4D(Gi) Suppresses GABA Release from BLA Interneurons
The efficiency of DREADD suppression was tested by double-patch recording from connected pairs of an IN-expressing hM4Di-Citrine identified by the fluorescence and a putative principal neuron (PN). The brief depolarizing current was injected in the IN to trigger single AP. It resulted in the inhibitory postsynaptic current (IPSC) in the connected PN, including clozapine-N-oxide (CNO) (1 μM) in the bath did not prevent APs but diminished IPSCs (Fig. 1). The DREADD suppression of presynaptic GABA release despite the presence of AP was consistent with published findings. 37

Chemogenetic Suppression of Sst-INs Enables LTP Induction Ex Vivo
For faithful activation of dmPFC axons at high frequency, a fast opsin Chronos 30 was expressed in dmPFC. The 50-Hz trains of light pulses were used for LTP induction [ Fig. 2(a)]. First, we examined the effect of DREADD suppression of Sst-INs on LTP by whole-cell recording from PNs in BLA slices expressing hM4Di in Sst-INs. In the absence of CNO, the 50-Hz stimulation of dmPFC axons caused a brief post-tetanic potentiation of the excitatory postsynaptic currents (EPSCs), followed by a rapid EPSC decline with a tendency toward depression at the 25 to 30 min after the induction (p ¼ 0.068, one-sample t-test). In the presence of CNO, the stimulation caused increases in EPSCs lasting for at least 30 min (Fig. 2), suggesting that suppression of Sst-INs enables LTP induction.
Repeated neuronal stimulation over extended time intervals, or the spaced LTP protocols, induces LTP more effectively than the shorter protocols, which is consistent with greater efficiency of spaced over massed training. 38 However, our attempts to induce LTP with a spaced protocol during the whole-cell recording have failed (data not shown), presumably because the dialysis of the intracellular content limits the time between obtaining the whole-cell configuration and effective LTP induction. To overcome this limitation, in the following experiments, LTP was tested by recording LFPs and using the spaced LTP protocol [ Fig. 3(a)]. We run the CNO For CNO control, we recorded from slices that did not express hM4Di. There was no significant LTP in the absence or presence of CNO, but there was a tendency toward LTP with CNO [p ¼ 0.09, one-sample t-test, Fig. 3(b)]. In slices with hM4Di in PV-INs, there was no significant LTP in the absence or presence of CNO but a tendency toward LTP in the absence of CNO [p ¼ 0.09, one-sample t-test, Fig. 3(c)]. In slices expressing hM4Di in Sst-INs, there was a significant LTP in the presence of CNO and no LTP in the absence of CNO [ Fig. 3(d)]. These data indicate that (a) CNO in the absence of hM4Di has a minor effect if any on LTP induction, (b) suppression of PV-INs does not aid LTP induction but may rather impede it, and (c) suppression of Sst-INs enables LTP.  Fig. S1).

Arch in Sst-INs Enables LTP Induction Ex Vivo and In Vivo
LTP induction in the dmPFC-BLA input was tested by giving trains of blue light pulses alone or combined with the continuous yellow light of different intensities [ Fig. 4(d)]. The yellow light by itself did not cause the release of glutamate from the dmPFC axonal terminals in BLA (data not shown). Unexpectedly, the trains of blue light in the absence of yellow light induced a significant LTP [ Fig. 4(e), left, black-filled circles]. Combining the trains of blue light and the yellow light of low intensity (0.15 mW∕mm 2 ) produced more significant LTP and with a tendency to be higher than with the blue light alone [ Fig. 4(e), left, orange-filled circles]. Increasing the yellow irradiance to 0.24 mW∕mm 2 prevented LTP induction [ Fig. 4(e), left, red-filled inverted triangles]. In slices without Arch, the low-intensity or the high-intensity yellow light did not enable LTP induction by the pulses of blue light [ Fig. 4(e), middle]. Together, these data indicate that Arch enables LTP induction by the trains of blue light, and it occurs even in the absence of yellow light, suggesting that the blue light inhibits Sst-INs expressing Arch. Consistently, whole-cell recordings from an Sst-IN with Arch revealed hyperpolarizing currents elicited by blue light (Supplemental Fig. S2). We also found that continuous yellow light at 0.24 mW∕mm 2 inhibited EPSCs evoked in BLA neurons by stimulating Chronos-expressing dmPFC axons with pulses of blue light (Supplemental Fig. S3). It may result from the desensitization of Chronos by yellow light and would explain the failure of LTP induction in the presence of a stronger yellow light.  Insets represent examples of averaged fEPSPs before (1) and after (2) LTP induction as indicated by horizontal grey and black bars. Scales: 0.2 mV, 10 ms. Right: Summary data for relative fEPSP slopes averaged during (2) for each slice. n ¼ 14 slices from 4 mice (Arch-no yellow), n ¼ 17 slices from 10 mice (Arch-low yellow), n ¼ 5 slices from 3 mice (Arch-high yellow), n ¼ 16 slices from 4 mice (no Arch-no yellow), n ¼ 10 slices from 5 mice (no Arch-low yellow), and n ¼ 8 slices from 7 mice (no Arch-high yellow). **p < 0.01, ***p < 0.001, compared to 1, which is the averaged relative fEPSP slope during the baseline indicated by gray bars (1), Wilcoxon's signed rank test. Boxes and the thick bars inside represent SEM and means.
To test LTP induction in vivo, mice expressing Arch in the BLA Sst-INs and Chronos in dmPFC were implanted with optrodes, whose two electrodes were positioned in BLA and an optical fiber above BLA

Discussion
This study was aimed at developing an efficient protocol for the artificial induction of one form of neuronal plasticity, LTP, in the dmPFC-BLA synapses. This study has three findings: Each data point represents a single independently tested amygdala (n ¼ 6 amygdalae from three mice). Right: summary data for the relative fEPSP slope and amplitude facilitation during the 3 h after the end of LTP induction, identified by the horizontal black bar (2). *p < 0.05, compared to 1, which is the averaged relative fEPSP slope/amplitude during the baseline indicated by gray bars (1), Wilcoxon's signed rank test.
(3) blue light alone is sufficient for the optogenetically assisted facilitation because the wavelength partially activates Arch expressed in Sst-INs.
The several hour duration of the LTP obtained in vivo implies a protein synthesis-dependent mechanism; 39 however, a definite conclusion would require testing the effect of protein synthesis inhibitors.
The finding that suppression of Sst-INs, but not PV-INs, enables LTP induction in BLA input from dmPFC, suggests that Sst-INs are distinct groups of GABAergic neurons, specializing in gating synaptic plasticity in remote inputs to BLA.
Like in the cortex, the Sst-INs axonal terminals in BLA target mostly the distal dendrites of PNs, whereas the PV-INs preferentially target somas and proximal dendrites. [40][41][42][43] Then, a straightforward interpretation of our findings is that dendritic but not somatic disinhibition favors LTP in BLA. Nevertheless, the finding by Luthi's lab that auditory cue activates PV-INs to suppress Sst-INs, and then aversive footshock suppresses PV-Ins, and Sst-INs altogether 22 suggest that both dendritic and somatic disinhibition are involved in fear learning. Meanwhile, our findings indicate that dendritic disinhibition alone is sufficient for the facilitation of BLA inputs. However, because Sst-INs also inhibit other INs, including PV-INs, some of which form reciprocal connections, 43-45 more complex mechanisms than simple dendritic disinhibition may contribute to the LTP after the suppression of Sst-INs.
A similar role of Sst-INs was reported in the somatosensory cortex, where Sst-INs gate LTP in the lemniscal sensory pathway and their suppression by PV-and VIP-INs "opens that gate" and allows LTP induction. 46 The LTP gating by Sst-INs, however, is not a universal phenomenon throughout the brain. For example, in the hippocampus, the Sst-INs located in the oriens/alveus region of the area CA1 rather enhance LTP in the Schaffer collateral pathway by inhibiting GABAergic neurons in the stratum radiatum, thereby disinhibiting the CA1 principal cells. 47 The PV-INs, in turn, appear to gate the hippocampal LTP, based on the finding of a stronger LTP in the model mice for the presymptomatic amyotrophic lateral sclerosis (ALS) and Alzheimer, in which a mutated NRG1 receptor Erb4 causes deficiencies of the PV-INs. 48 Another technical aspect is the choice between chemogenetic and optogenetic suppression. While DREADD is highly effective in suppressing GABA release from INs even when they fire APs, the drawbacks for in vivo experiments are the long washout times and the off target-effects of chemogenetic ligands. 51 In addition, in hM4Di transgenic mice, CNO would suppress all Sst-INs, including those outside the target region, and cause nonspecific effects, which can be avoided by employing viral vectors for region-specific expression of hM4Di.
The optogenetic suppression of INs circumvents these problems, but using two different wavelengths of light in vivo is more expensive and technically demanding. An additional drawback of the two-light design is that stronger yellow light interferes with LTP induction, apparently by desensitizing Chronos, as seen with the ReaChR, 52 even at the levels that do not activate Chronos to release neurotransmitter. This nonspecific effect of the yellow light on a blue-light activated opsin is an artifact that needs to be considered along with other optogenetics artifacts carefully summarized in the recent reviews. 53,54 Fortunately, the problem can be avoided by using blue light alone, which is sufficient for LTP induction, both ex vivo and in vivo, with Arch expressed in Sst-INs. This is because the blue light (470 nm) still activates Arch, even though at the 35% efficiency when compared to the optimal wavelength (560 nm). 55 However, using the same light source for inhibition of INs and excitation of glutamatergic axons limits the freedom to deliver different stimuli to two neuronal populations. Perhaps, the singlecolor disinhibition assisted LTP can be improved further by replacing Arch with a blue-shifted inhibitory opsin, like a proton pump Mac, activated by the blue 470-nm light at about 60% of the maximum efficiency. 55

Disclosures
The authors have no conflicts of interest to disclose.