Long-term potentiation at pyramidal cell to somatostatin interneuron synapses controls hippocampal network plasticity and memory

Summary Hippocampal somatostatin (SOM) cells are dendrite-projecting inhibitory interneurons. CA1 SOM cells receive major excitatory inputs from pyramidal cells (PC-SOM synapses) which show mGluR1a- and mTORC1-mediated long-term potentiation (LTP). PC-SOM synapse LTP contributes to CA1 network metaplasticity and memory consolidation, but whether it is sufficient to regulate these processes remains unknown. Here we used optogenetic stimulation of CA1 pyramidal cells and whole-cell recordings in slices to show that optogenetic theta-burst stimulation (TBSopto) produces LTP at PC-SOM synapses. At the network level, we found that TBSopto differentially regulates metaplasticity of pyramidal cell inputs: enhancing LTP at Schaffer collateral synapses and depressing LTP at temporo-ammonic synapses. At the behavioral level, we uncovered that in vivo TBSopto regulates learning-induced LTP at PC-SOM synapses, as well as contextual fear memory. Thus, LTP of PC-SOM synapses is a long-term feedback mechanism controlling pyramidal cell synaptic plasticity, sufficient to regulate memory consolidation.

LTP at PC-SOM synapses can be persistent, as repeated activation of mGluR1a induces LTP of excitatory synapses lasting hours and involving gene transcription and mRNA translation (Ran et al., 2009;Artinian et al., 2019). mTORC1 signaling controls translation during LTP at PC-SOM synapses (Ran et al., 2009;Artinian et al., 2019). Interestingly, mTORC1 activity in SOM cells is required for learning-induced LTP of PC-SOM synapses, regulation of CA1 network plasticity, and long-term consolidation of spatial and contextual fear memory (Artinian et al., 2019). However, whether LTP at PC-SOM synapses is sufficient for the regulation of CA1 network metaplasticity and hippocampus-dependent memory remains to be determined.
In the present study, we address this question by using optogenetics to induce long-term plasticity at PC-SOM excitatory synapses. We found that optogenetic stimulation of pyramidal cells induces LTP at PC-SOM synapses in slices. Furthermore, optogenetic induction of LTP at PC-SOM synapses bidirectionally controls CA1 network plasticity by facilitating SC-PC LTP and suppressing TA-PC LTP. Moreover, we uncover that optogenetic induction of LTP at PC-SOM synapses in vivo regulates learning-induced LTP at these synapses and controls contextual fear memory. Our findings suggest that LTP at PC-SOM synapses is sufficient to control the state of plasticity in the CA1 network and the consolidation of contextual fear memory, uncovering a long-term feedback control mechanism via PC-SOM synapses sufficient for the regulation of CA1 pyramidal cell metaplasticity and hippocampus-dependent memory.
Next, we examined the mechanisms involved in the optogenetically induced LTP at PC-SOM synapses. First, low-frequency optogenetic stimulation (LFS opto ; 20 Hz) that elicited EPSP summation but a lower rate of firing in SOM cells ( Figure 1C) did not induce lasting changes in light-evoked EPSP amplitude (103.0% of baseline at 20-30 min post-induction; Figures 1C, 1J, and S1B). Second, in recordings with no optogenetic tetanization, light-evoked EPSP amplitude was unchanged over the same time period (95.7% of baseline at 20-30 min post-induction; Figures 1D, 1J, and S1C). Third, to determine whether the optogenetically induced LTP at PC-SOM synapses was mGluR1a-dependent, TBS opto was given in the presence of the mGluR1a antagonist LY367385 (40 mM) ( Figure 1E). In these conditions, TBS opto failed to induce lasting changes in EPSP amplitude (111.0% of baseline at 20-30 min post-induction; Figures 1E, 1J and S1D). Fourth, to determine whether optogenetically induced LTP at PC-SOM synapses is dependent on mTORC1-mediated translation, we used SOM-Raptor-KO mice that are deficient in mTORC1 signaling and late-LTP in SOM cells (Artinian et al., 2019). In these mice, TBS opto failed to induce lasting changes in light-evoked EPSP amplitude (104.0% of baseline at 20-30 min post-induction; Figures 1F, 1J, and S1E). Fifth, we examined if optogenetically induced LTP requires transcription by pre-incubating hippocampal ll OPEN ACCESS slices with 2 mM of the irreversible transcription inhibitor actinomycin D for 15 min before recording (Yuan and Burrell, 2013;Younts et al., 2016). In these conditions, TBS opto induced a slow onset LTP of light-evoked EPSP amplitude (119.5% of baseline at 20-30 min post-induction; Figures 1G, 1J, and S1F) that was comparable to the LTP induced in control (Figures 1B and 1J) or in vehicle (DMSO; Figures S1G and S1H). Sixth, we determined if optogenetic stimulation elicited LTP at PC synapses onto parvalbumin-expressing interneurons (PC-Pvalb synapses) using the injection of AAV2/9-CaMKIIa-hChR2(E123T/T159C)-mCherry and AAV2/9-EF1a-DIO-EYFP in Pvalb-Cre mice ( Figure 1H). In these mice, TBS opto failed to induce lasting increases in light-evoked EPSPs at PC-Pvalb synapses (90.0% of baseline at 20-30 min post-induction; Figures 1I, 1J, and S1I). In all these experiments, cell input resistance was unchanged after TBS opto (Figure 1K). Together, these results suggest that optogenetic stimulation of CA1 pyramidal cells produces a frequency-dependent LTP at PC-SOM synapses that is dependent on mGluR1a activation and mTORC1mediated translation, but independent of transcription. Moreover, optogenetic stimulation of CA1 pyramidal cells does not induce LTP at PC-Pvalb synapses.
These properties of optogenetically induced LTP at PC-SOM synapses are similar to those previously reported for Hebbian LTP at excitatory synapses onto SOM cells characterized using electrical stimulation (Perez et al., 2001;Croce et al., 2010;Honore et al., 2021). So, next we determined if TBS opto can induce LTP of EPSPs evoked in SOM cells by electrical stimulation ( Figure S2A). We found that TBS opto induced a similar LTP of electrically evoked EPSPs that was dependent on mGluR1a and mTORC1, required tetanization, and was absent in mice that received only the control CaMKIIa-mCherry injection without hChR2 ( Figures S2B-S2G), indicating similar mechanisms as at PC-SOM synapses activated optogenetically.
TBS opto differentially regulates LTP at Schaffer collateral and temporo-ammonic synapses CA1 SOM cells (notably OLM cells) differentially control transmission at Schaffer collateral and temporoammonic synapses of CA1 pyramidal cells (Leao et al., 2012). In consequence, Hebbian LTP at excitatory synapses onto SOM cells is associated with increased LTP at Schaffer collateral inputs and reduced LTP at temporo-ammonic synapses of CA1 pyramidal cells (Vasuta et al., 2015;Artinian et al., 2019;Sharma et al., 2020). So, next we studied if TBS opto -induced LTP at PC-SOM synapses was sufficient to regulate metaplasticity of the CA1 network using whole-field optogenetic stimulation and field potential recordings in slices with synaptic inhibition intact. First, we established that whole-field optogenetic stimulation also elicited LTP at PC-SOM synapses ( Figure 2A). Whole-field TBS opto elicited EPSP summation and burst firing of SOM cells and induced LTP of light-evoked EPSP amplitude (125.0% of baseline at 20-30 min post-induction; Figures 2B and S3E). LTP was absent in cells that received no optogenetic (B and C) Bottom left: time plots of light-evoked EPSP amplitude for all SOM cells, showing LTP at PC-SOM synapses following TBS opto (B; n = 17 cells) but not after LFS opto (C; n = 8 cells). Right: example of representative cells of average EPSPs before (pre) and 20-30 min after (post) TBS opto or LFS opto . Top: EPSP summation and cell firing during TBS opto (80 Hz) or LFS opto (20 Hz). Paired t-tests; ***p < 0.0001; ns, p > 0.05. See Figures S1A and S1B for the respective time plots of EPSP amplitude of representative cells in (B) and (C), respectively. (D-G) Left: time plots of light-evoked EPSP amplitude for all SOM cells, showing absence of LTP at PC-SOM synapses following no tetanization (D; n = 13 cells), TBS opto in the presence of the mGluR1a antagonist LY367385 (40 mM) (E; n = 10 cells), or TBS opto in SOM-Cre-Raptor-KO mice (F; n = 15 cells); but LTP after TBS opto in slices pre-incubated with the transcription inhibitor actinomycin D (2mM) (G; n = 6 cells). Right: representative example of average EPSPs before (pre) and after (post) each treatment condition. Paired t-tests; *p < 0.05; ns, p > 0.05. See Figures S1C, S1D, and S1F for the respective time plots of EPSP amplitude of representative cells in (D), (E), (F), and (G), respectively; and Figure S1G  Wilcoxon matched-pairs signed-rank test (TBS opto and LFS opto ), or paired t-test (other groups); ns, p > 0.05. Data are represented as mean G SEM here and below. *p < 0.05, **p < 0.01, ***p < 0.001; p > 0.5 is considered not significant (ns). Statistical tests and actual p values for each comparison here and below are given in Table S1. Next, we examined the long-lasting effects of whole-field TBS opto on plasticity at Schaffer collateral synapses (SC-PC synapses) and temporo-ammonic synapses (TA-PC synapses) onto CA1 PCs in slices with synaptic inhibition intact. In these experiments, we used whole-field optogenetic stimulation and fEPSP recordings so that we could obtain long-term stable recordings of synaptic responses, for a 15 min baseline period before TBS opto and a 25 min period after TBS opto , to assess the effects of TBS opto on basal transmission, and for another 35 min period after the tetanization of SC-PC or TA-PC synapses, to assess effects on LTP at these synapses ( Figure 2). First, we investigated the regulation of plasticity at SC-PC synapses ( Figure 2C and S3F). A weak electrical TBS (wTBS) of the SC pathway failed to induce LTP of fEPSP slope (107.0% of baseline at 25-35 min post-induction; Figure 2D and S3G). In contrast, when the wTBS was preceded 25 min earlier by TBS opto , LTP was induced at SC-PC synapses (125.7% of baseline at 25-35 min post-induction; Figures 2D, 2F, and S3H). Thus, TBS opto facilitated LTP at SC-PC synapses. TBS opto did not affect basal transmission at SC-PC synapses ( Figure 2G). In mice with control CaMKIIa-mCherry injection without hChR2, TBS opto failed to facilitate LTP at SC-PC synapses (104.0% of baseline at 25-35 min post-induction; Figures 2D and 2F), indicating that the LTP facilitation was not owing to unspecific effects of light stimulation. To determine if the facilitation of LTP at SC-PC synapses was due to plasticity at SOM-PC synapses induced by TBS opto , we examined if the facilitation was dependent on mGluR1a and on mTORC1 in SOM cells. When wTBS was preceded 25 min earlier by TBS opto in the presence of the mGluR1a antagonist LY367385 (40 mM), the facilitation of LTP at SC-PC synapses was absent (107.0% of baseline (C) Summary graph for all mice, showing reduced freezing in the long-term memory test in mice that received Arch activation in phase with shocks (n = 9 mice, dark green), relative to those that received light stimulation without Arch (n = 8 mice, gray) or Arch activation out of phase with shocks (shifted; n = 7 mice, light green). Mann-Whitney Rank-Sum test or paired t-test (pre vs post-conditioning), ***p < 0.001 (gray). One way ANOVA(Memory tests), Holm-Sidak pairwise comparisons, *p < 0.05, ***p < 0.001, ns p > 0.05.  Figures 2K and 2N). These results suggest that the TBS opto -induced depression of LTP at TA-PC synapses is the result of mGluR1a-and mTORC1-dependent plasticity at SOM-PC synapses. Thus, TBS opto -induced LTP at PC-SOM synapses appears sufficient to regulate metaplasticity of the CA1 network, upregulating LTP at SC-PC synapses and downregulating LTP at TA-PC synapses.

Optogenetic theta-burst stimulation interferes with contextual fear memory consolidation
At the behavioral level, mTORC1 activity in SOM cells, that is required for learning-induced LTP of PC-SOM synapses, contributes to long-term consolidation of spatial and contextual fear memory (Artinian et al., 2019). So, next we examined if LTP at PC-SOM synapses is sufficient to regulate hippocampus-dependent memory. First, we verified that, as previously reported (Lovett-Barron et al., 2014), the activity of CA1 SOM cells is necessary for long-term contextual fear memory using optogenetic silencing of CA1 SOM cells with archaerhodopsin (Arch) (Vasuta et al., 2015) during contextual fear conditioning ( Figures 3A and 3B). Optogenetic activation of Arch in phase with the presentation of shocks during conditioning, resulted in reduced freezing during the long-term contextual fear memory test, relative to light stimulation in absence of Arch (51.4% reduction; Figure 3C) or to the activation of Arch out of phase (shifted) with the presentation of shocks (64.9% reduction; Figure 3C). In the open field test ( Figure S4A), silencing of CA1 SOM cells with light stimulation of Arch did not affect anxiety level (time spent in periphery or center, and ratio of time in center/periphery; Figure S4B) or locomotion (total distance traveled and zone transitions; Figure S4C) relative to mice receiving light stimulation without Arch. These results indicate that silencing SOM cells during contextual fear conditioning impairs long-term contextual fear memory, confirming that SOM cell activity is necessary during long-term contextual fear learning (Lovett- Barron et al., 2014).
Next, we determined if LTP at PC-SOM synapses is sufficient to regulate hippocampus-dependent memory. We used CA1 injection of AAV2/9-CaMKIIa-hChR2(E123T/T159C)-mCherry in SOM-Cre-EYFP mice and the same TBS opto induction protocol that elicits LTP at PC-SOM synapses in slices. We delivered the optogenetic stimulation in vivo and investigated its effect on contextual fear memory ( Figures 3D and 3E). TBS opto given in CA1 30 min (E) Experimental protocol of TBS opto given 30 min before contextual fear conditioning.
(F) Left: Summary graph for all SOM-Cre-EYFP mice, showing reduced freezing in the long-term memory test in mice that received TBS opto (n = 8 mice, blue), relative to mice that received no TBS opto (n = 7 mice, violet) or TBS opto without hChR2 (n = 8 mice, gray). Mann-Whitney Rank-Sum test (pre vs post-conditioning), ***p < 0.001 (gray). Kruskal-Wallis One-way ANOVA on Ranks (Memory tests), Dunn's pairwise comparisons test, ***p = 0.001, ns p > 0.05. Right: Similar data presentation for SOM-Cre-Raptor-KO mice, showing significant freezing in the memory test relative to pre-conditioning, and no difference in freezing in the long-term memory test in mice that received TBS opto (n = 6 mice, blue) or no TBS opto (n = 7 mice, violet). Mann-Whitney Rank-Sum test (pre vs post-conditioning), **p < 0.005 (gray iScience Article before conditioning resulted in reduced freezing during the long-term contextual fear memory test, relative to mice receiving no TBS opto before conditioning (73.7% reduction; Figure 3F) or TBS opto in the absence of hChR2 (71.5% reduction; Figure 3F). Mice in the three groups showed similar normal anxiety level (time spent in periphery or center, and ratio of time in center/periphery; Figure S4E) and locomotion (total distance traveled and zone transitions; Figure S4F) in the open field test. These results indicate that the TBS opto induction protocol that elicit LTP at PC-SOM synapses is sufficient to regulate long-term contextual fear memory in vivo.
As LTP induced by TBS opto at PC-SOM synapses, and its regulation of CA1 PC metaplasticity, are blocked by conditional knock-out of Rptor in SOM cells (Figures 1 and 2), we tested next if the reduction of fear memory by TBS opto was owing to mTORC1 signaling in SOM cells. TBS opto in CA1 given 30 min before conditioning in SOM-Cre-Raptor-KO mice did not affect freezing during the long-term contextual fear memory test, relative to mice not receiving TBS opto before conditioning ( Figure 3F). Both mice groups showed increased freezing in the memory tests demonstrating significant contextual fear learning ( Figure 3F) and behaved similarly in the open field test ( Figures S4D-S4F) indicating normal anxiety level and locomotion. These results indicate that the TBS opto -induced impairment of fear memory was dependent on mTORC1 signaling in SOM interneurons and not owing to other non-specific effects of optogenetic activation of CA1 PCs, suggesting that TBS optoinduced LTP at PC-SOM synapses is sufficient for the regulation of hippocampus-dependent memory.
Prior induction of long-term potentiation by optogenetic theta-burst stimulation results in subsequent theta-burst stimulation-and learning-induced depotentiation As the behavior experiments indicate that TBS opto impairs contextual fear memory consolidation when given before contextual fear conditioning, we examined the interaction between optogenetically and electrically induced LTP of electrically evoked EPSPs in SOM cells in slices ( Figure 4A).
After a baseline period, the first episode of electrical TBS elicited LTP of EPSP amplitude (134.0% of baseline at 10-15 min post-induction TBS#1), and a second TBS episode induced a further increase in LTP (164.0% of baseline at 30-35 min post-induction TBS#1; Figure 4B). In contrast, the first episode of TBS opto elicited LTP of EPSP amplitude (140.0% of baseline at 10-15 min post-induction of TBS opto #1), and a second TBS opto episode did not induce a further increase in LTP (142.0% of baseline at 30-35 min post-induction TBS opto #1; Figure 4C). Interestingly, after the first episode of TBS opto that elicited LTP of EPSP amplitude ( Figure 4E). Finally, a single TBS episode given at the later time point induced iScience Article LTP (129.0% of baseline; Figure 4F), suggesting that the depotentiation induced by TBS was not owing to a response run-down due to recording time, or dialysis of cell content. Thus, prior induction of LTP by TBS opto engages some mechanisms resulting in depotentiation when TBS is given subsequently.
Contextual fear conditioning induces a persistent LTP of PC-SOM synapses, that can be recorded ex vivo 24h after conditioning (Artinian et al., 2019). Therefore, we used ex vivo whole-cell recordings in slices 24 h after conditioning, to examine if in vivo TBS opto given 30 min before contextual fear conditioning interferes with learning-induced LTP at excitatory synapses of SOM cells ( Figure 4G). SOM cells from mice that received TBS opto before contextual fear conditioning exhibited decreases in spontaneous EPSC frequency and amplitude, relative to cells from mice with contextual fear conditioning only (56.1% decrease in frequency, 31.9% decrease in amplitude; Figure 4H). Similarly, the input-output function of electrically evoked EPSCs was decreased in SOM cells from mice that received TBS opto before contextual fear conditioning, relative to cells from mice with contextual fear conditioning only (64.1% decrease in the slope of EPSC input-output function; Figure 4I). These results suggest that in vivo TBS opto given 30 min before contextual fear conditioning interferes with learning-induced LTP atPC-SOM synapses, consistent with our findings that the induction of Hebbian LTP results in depotentiation at PC-SOM synapses after prior induction of LTP by TBS opto in slices. Overall, our findings indicate that LTP at PC-SOM synapses is sufficient to regulate CA1 pyramidal cell metaplasticity and hippocampus-dependent memory.

DISCUSSION
Long-term synaptic plasticity at excitatory input synapses onto specific inhibitory interneurons is an intriguing feature of hippocampal synaptic networks but its role in hippocampal memory function has remained largely undetermined (Kullmann et al., 2012;Honore et al., 2021;Bartos et al., 2011;McBain and Kauer, 2009). Brain-wide manipulation of mTORC1 activity specifically in SOM cells indicated that LTP at afferent synapses of hippocampal SOM cells contributes to the regulation of CA1 network plasticity and long-term consolidation of spatial and contextual fear memory (Artinian et al., 2019). Here we used hippocampal CA1-specific optogenetic approaches to establish that LTP at PC-SOM synapses is sufficient to differentially control metaplasticity of SC-PC and TA-PC synapses, and regulate contextual fear memory, uncovering a long-term feedback mechanism controlling pyramidal cell synaptic plasticity and memory consolidation.

Long-term potentiation induced by optogenetic theta-burst stimulation
TBS opto was sufficient to induce LTP of synaptic responses elicited by optogenetic activation or local electrical stimulation of PC-SOM synapses. Optogenetically induced LTP at PC-SOM synapses shared key features with Hebbian LTP induced by electrical tetanization: slow onset potentiation, frequency-dependence, mGluR1a-mediated, and SOM versus Pvalb cell-specificity (Perez et al., 2001;Lapointe et al., 2004;Croce et al., 2010;Vasuta et al., 2015). As TBS opto restricts activation to CA1 PCs, without affecting other CA1 afferent fibers potentially activated with tetanization by electrical stimulation (Nabavi et al., 2014; Nicholson and Kullmann, 2021;Cardin et al., 2010) , 1989). Our findings establish that the activation of these other projection systems is not required for LTP at PC-SOM synapses. This is in contrast to another form of plasticity at excitatory synapses onto putative-SOM cells in CA1 stratum oriens, anti-Hebbian LTP (Lamsa et al., 2007;Nicholson and Kullmann, 2021), which can be induced by optogenetic activation of cholinergic afferents but not of CA1 PCs, and, thus, is owing to cholinergic heterosynaptic LTP at PC-oriens interneurons excitatory synapses (Nicholson and Kullmann, 2021). Our finding that TBS opto of PC-SOM synapses is sufficient for the induction of LTP is consistent with previous work that Hebbian LTP is afferent pathway-specific, occurring at CA1 PC-SOM synapses but not at CA3 PC-SOM synapses (Croce et al., 2010).
Interestingly, we uncovered that TBS opto -induced LTP at PC-SOM synapses was mTORC1-dependent and transcription-independent. A more persistent (lasting many hours) form of mGluR1a-mediated LTP at PC-SOM synapses is both transcription-and translation-dependent (Ran et al., 2009;Artinian et al., 2019). As mTORC1 is a key signaling control mechanism of translation ( iScience Article translation-dependent, transcription-independent LTD at SC-PC synapses (Huber et al., 2000;Banko et al., 2006). Such a mTORC1-and protein synthesis-dependent mechanism is consistent with the slow onset gradual development over minutes of LTP at PC-SOM synapses (Huber et al., 2000). SOM interneurons express both mGluR1 and mGluR5 (van Hooft et al., 2000;Topolnik et al., 2006). Previous work indicated that mGluR1a and mGluR5 signal via distinct pathways in SOM interneurons, notably OLM cells (Topolnik et al., 2006). Activation of mGluR1a leading to Src/ERK-dependent Ca 2+ entry via TRP channels and intracellular Ca 2+ release is necessary for LTP induction at SOM interneuron excitatory synapses, whereas mGluR5 activation and intracellular Ca 2+ release are not involved in LTP induction (Topolnik et al., 2006). Here we examined the role of mGluR1a in TBS opto -induced LTP at PC-SOM synapses; however, additional work will be necessary to clarify if mGluR5 is implicated in this plasticity.

Network regulation
Our finding that TBS opto up-regulated LTP at SC-PC synapses and down-regulated LTP at TA-PC synapses in a mGluR1a-and SOM cell-mTORC1-dependent manner, is consistent with a differential gating of plasticity at PC input pathways by SOM cells (Leao et al., 2012) that is controlled over long-term periods by the plasticity of SOM cell input synapses (Vasuta et al., 2015;Artinian et al., 2019;Sharma et al., 2020;Leao et al., 2012). It is noteworthy that TBS opto did not affect basal transmission at SC-and TA-PC synapses, but regulated theta-burst-induced plasticity at these synapses. This modulation of plasticity reflects the property of PC-SOM synapses that show facilitation and postsynaptic firing with repeated stimulation, resulting in effective recruitment with repetitive activation of PCs (Croce et al., 2010;Honore et al., 2021;Pouille and Scanziani, 2004).
What are the network mechanisms responsible for the differential modulation of LTP of CA3 and entorhinal inputs of pyramidal cells? Previous work indicated that somatostatin-expressing oriens-lacunosum-moleculare (OLM) interneurons differentially modulate CA3 and entorhinal inputs to CA1 pyramidal cells (Leao et al., 2012). OLM inhibitory interneurons suppress entorhinal inputs via postsynaptic inhibition of PC distal dendrites in stratum lacunosum-moleculare (Leao et al., 2012). In addition, OLM inhibitory interneurons facilitate CA3 inputs by the disinhibition of the more proximal dendrites of pyramidal cells, i.e. OLM interneurons inhibit other inhibitory interneurons in stratum radiatum which themselves inhibit local pyramidal cell dendrites (Leao et al., 2012). Moreover, OLM interneurons similarly modulated LTP in these pathways, depressing LTP of entorhinal inputs and facilitating LTP of CA3 inputs (Leao et al., 2012). Our present findings are consistent with these mechanisms of regulation of CA1 network plasticity. In addition, previous work showed that a consequence of LTP at PC-SOM input synapses is to enhance synaptically evoked firing of SOM interneurons (Croce et al., 2010). Thus, our present results and previous reports (Leao et al., 2012;Croce et al., 2010) are consistent with network mechanisms whereby LTP at input synapses of SOM INs increases their firing output, which enhances LTP of the CA3 input pathway via an increased disinhibitory action in proximal dendrites of pyramidal cells (Artinian et al., 2019;Vasuta et al., 2015), and suppresses LTP of the entorhinal input pathway via an increased inhibitory action in distal dendrites of pyramidal cells (Sharma et al., 2020).
Our findings highlight the different roles of long-term synaptic plasticity at excitatory synapses onto PC versus SOM interneurons. It has been proposed that LTP at PC excitatory synapses serves to generate enduring changes at synapses of engram cells that encode memories (Nabavi et al., 2014;Choi et al., 2018). In contrast, our results suggest that LTP at SOM cell excitatory synapses may serve to regulate the state of plasticity (or metaplasticity) of PC input synapses (Honore et al., 2021). These distinct roles suggest that hippocampal-dependent encoding of memories, via LTP in PCs, can still occur, albeit in a reduced manner, in the absence of LTP at SOM cell synapses, as for example in mice with conditional knockout of Rptor in SOM cells (Artinian et al., 2019). However, LTP at PC-SOM synapses controls the efficiency of hippocampal-dependent encoding memories by PCs, providing additional versatility to CA1 network plasticity. Our findings provide a link between LTP at PC-SOM synapses, regulation of metaplasticity at SC-PC and TA-PC synapses, and modulation of contextual fear memory. Contextual fear learning was previously shown to induce LTP at PC-SOM synapses (Artinian et al., 2019). Thus, during contextual fear learning, LTP is induced at PC-SOM synapses, and this may cause an upregulation of the CA1 network plasticity changes, such as LTP at SC-PC synapses, that encode memory. Thus, impairment in learning-induced plasticity at PC-SOM synapses caused by TBS opto in vivo (Figure 4) may result in a loss of upregulation of metaplasticity in the CA1 network during learning (Figure 2), and a deficit in memory consolidation (Figure 3). Such a link would be strengthened by a demonstration that contextual fear learning induces long-term changes ll OPEN ACCESS iScience Article at SC-PC and TA-PC synapses, and that these changes are modulated by in vivo manipulation of LTP at PC-SOM synapses. However, given the spatially restricted synaptic plasticity changes in pyramidal cells (Whitlock et al., 2006) and the sparse coding of synaptic changes in hippocampal engram cells (Choi et al., 2018), such experiments require different approaches than those used in the present study.
The gating of PC input pathways has been well established for OLM cells, a subtype of SOM cells (Leao et al., 2012). However, SOM cells also include other types of interneurons: bistratified cells, and other cells with both local projections and distal projections to septum, subiculum, CA3, dentate gyrus, or entorhinal cortex (Honore et al., 2021;Pelkey et al., 2017;. Although excitatory synapses onto bistratified cells and OLM cells show Hebbian LTP (Perez et al., 2001;Croce et al., 2010), it remains to be determined if SOM projection cells also do. Thus, it will be interesting to build on our findings and determine how plasticity at input synapses of other CA1 SOM cell types may participate in network regulation, possibly even controlling more distant hippocampal-related pathways.
Learning-induced changes at the input synapses of SOM cells may only be part of a more integrated response of SOM cells during learning. An increase in the intrinsic excitability of CA1 SOM cells, owing to a reduced Ca 2+ -dependent K + conductance, was found after learning a hippocampus-dependent trace eyeblink conditioning task (McKay et al., 2013). Although it remains to be determined if SOM cell inhibitory synapses demonstrate learning-induced changes (Chiu et al., 2018;Udakis et al., 2020), mTORC1-dependent axonal sprouting by CA1 SOM cells takes place in the CACNA1A mouse model of epilepsy which is characterized by impaired synaptic inhibition by Pvalb interneurons (Jiang et al., 2018). These previous studies and our findings indicate that multiple plasticity mechanisms occur in SOM cells at the level of their input function, intrinsic excitability, and perhaps even output synapses, suggesting that an integrated gainof-function response may occur in SOM cells during learning.

Optogenetic theta-burst stimulation interaction with Hebbian and learning-induced longterm potentiation
Our results that TBS opto given in vivo before contextual fear conditioning affects long-term contextual fear memory suggests that LTP at PC-SOM synapses regulates hippocampal memory. In these experiments, TBS opto was given in vivo. Thus, it is conceivable that optogenetic activation of PCs in the CA1 region also activated the other major synaptic target of pyramidal cells, subicular pyramidal neurons (Cenquizca and Swanson, 2007;Taube, 1993), inducing long-term plasticity at these synapses (O'Mara et al., 2000;Huang and Kandel, 2005) and influencing hippocampal memory function. To examine this possibility, we examined the effects of TBS opto in mice with a cell-specific conditional knockout of Rptor in somatostatin interneurons, in which LTP at PC-SOM synapses is blocked. We found that the regulation of long-term contextual fear memory by TBS opto was absent in these mice (Figure 3), suggesting that the regulation of hippocampal memory by TBS opto was owing to changes at PC-SOM synapses and not at other efferent targets of CA1 PCs, like subicular neurons. However, it would be interesting in future experiments to design an induction protocol applicable in vivo that selectively activates CA1 PC-SOM synaptic plasticity and, thus, rule out actions via other synaptic targets of CA1 PCs. Nonetheless, our results are consistent with our other observations that PC-SOM synapse efficacy was reduced, like fear memory, by TBS opto given before learning (Figure 4), as well as with our previous findings that learning-induced LTP at PC-SOM synapses is necessary for long-term contextual fear memory (Artinian et al., 2019). It is noteworthy that TBS opto in slices produces LTP at PC-SOM synapses and not at PC-PV synapses, indicating some CA1 PC target cellspecificity in TBS opto -induced long-term plasticity.
Our results indicate that LTP at PC-SOM synapses induced by TBS opto share similar mechanisms as Hebbian LTP induced by electrical TBS and LTP induced by contextual fear learning at excitatory synapses onto SOM cells. As reported for LTP at hippocampal synapses ( Bliss and Lomo, 1973), LTP induced by the first episode of TBS opto occluded further LTP by a second TBS opto . However, LTP induced by a first episode of electrical TBS did not, and more LTP was induced by the second episode of TBS. In contrast, when electrical TBS was given after LTP induced by a first TBS opto episode, depotentiation occurred at PC-SOM synapses. Similarly, in vivo, when contextual fear conditioning was given 30 min after TBS opto , the efficacy of PC-SOM synapses was reduced during ex vivo recordings 24 h later. These results indicate that although they share similar mechanisms, LTP induced by TBS opto , electrical TBS, and fear learning are not identical. This is not surprising given that electrical stimulation may activate additional intra-hippocampal fibers, and learning may implicate other systems in addition to CA1 PC-SOM synapses. These iScience Article additional mechanisms are likely implicated in the depotentiation of PC-SOM synapses. Multiple mechanisms regulate PC-SOM synapse function and plasticity, notably pre-and postsynaptic GABA B -receptormediated inhibition (Booker et al., , 2020, astrocyte-mediated regulation of glutamate uptake (Huang et al., 2004), and GABA A synaptic inhibition by vasoactive intestinal polypeptide inhibitory interneurons (Tyan et al., 2014). It will be important to determine if any of these control mechanisms are implicated in the modulation/depotentiation of PC-SOM synapses. Importantly, our findings that learning-induced potentiation is impaired by prior application of TBS opto in vivo may explain why contextual fear memory may be impaired by TBS opto , providing further support that LTP induced at PC-SOM synapses is sufficient to regulate CA1 network metaplasticity and the contextual fear memory.

Limitations of the study
Some limitations should be considered when interpreting our results. First, although our interpretation of the results is concordant with optical stimulation resulting in LTP at PC-SOM synapses in vivo, this remains to be demonstrated directly. This would require recording synaptic activity of SOM INs in vivo and assessing the effect of optogenetic induction in vivo on this synaptic activity. Such experiments, however, require different experimental approaches than those used here. Instead, we addressed this issue using the approach shown in Figure 4, with optical stimulation in vivo, contextual fear conditioning, and ex vivo recordings. Our ex vivo results are consistent with optical stimulation in vivo eliciting LTP and with subsequent contextual fear conditioning resulting in depotentiation, instead of the late form of LTP that is normally elicited (Artinian et al., 2019). These results are analogous to the effect found in the slice experiments whereby optical stimulation in vitro elicits LTP and subsequent TBS with electrical stimulation results in depotentiation, instead of the LTP normally elicited. Thus, our results are consistent with the notion that optical stimulation elicits LTP at PC-SOM synapses both in vivo and in vitro. Previous work showed that a different induction protocol with repeated episodes of stimulation elicits a late, persistent form of LTP that lasts many hours to a day at SOM interneuron excitatory synapses (Artinian et al., 2019;Ran et al., 2009). Thus, it would be important to test if such a repeated optical stimulation protocol in vivo can produce persistent changes at PC-SOM synapses that can be detected with ex vivo recordings.
Second, a caveat of our study concerns our results that the regulation of long-term contextual fear memory by TBS opto was absent in mice with conditional knockout of Rptor in somatostatin cells, suggesting that the regulation of hippocampal memory by TBS opto was owing to changes at PC-SOM synapses. The lack of effect of TBS opto on contextual fear memory in the Raptor-KO model could be due to the low level of freezing response in these mice, indicating that these animals cannot learn any worse. In our experiments, we used a weak contextual fear conditioning protocol based on previous work showing that inactivating SOM interneurons impairs contextual fear memory (Lovett-Barron et al., 2014). It is important to note that with this contextual fear conditioning protocol, SOM-Cre-Raptor-KO mice show impaired fear memory relative to control mice; however, the SOM-Cre-Raptor-KO mice do display significant freezing responses in the memory probe test relative to pre-conditioning, indicating some level of contextual memory ( Figure 3F). Using a stronger contextual fear conditioning protocol, previous work showed also that although the SOM-Cre-Raptor-KO mice display a memory impairment, they still show significant contextual fear learning (Artinian et al., 2019). Thus, these findings suggest that a component of contextual fear learning requires mTORC1 and SOM interneurons, but another component does not. A similar conclusion was found for spatial learning in the Barnes maze (Artinian et al., 2019). Thus, repeating the present experiments with a stronger contextual fear conditioning protocol (Artinian et al., 2019) could help resolve the possibility of a floor effect in our experiments. However, the residual significant freezing responses in SOM-Cre-Raptor-KO mice in the contextual fear memory probe test suggest the absence of a floor effect, and, thus, are consistent with the effects of TBS opto on contextual fear memory being due to LTP at PC-SOM synapses.

STAR+METHODS
Detailed methods are provided in the online version of this paper and include the following:

Materials availability
This study did not generate new unique reagents.
Data and code availability d The published article includes all datasets generated or analyzed during this study. The raw electrophysiology data supporting the current study have not been deposited in a public repository because there is currently no standardized format or repository for such data, but they are available from the corresponding author on request.
d This paper does not report original code.
d Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Sst iresÀCre mice (Taniguchi et al., 2011) were crossed with Rosa26 lslÀEYFP reporter mice (Madisen et al., 2010) to generate Cre-dependent Enhanced Yellow Fluorescent Protein (EYFP) expression in SOM interneurons (SOM-Cre-EYFP mice) (Artinian et al., 2019). Sst iresÀCre ;Rosa26 lslÀEYFP mice were crossed with floxed Rptor mice (Sengupta et al., 2010) for cell-specific homozygous knock-out of Rptor in SOM cells (SOM-Cre-Raptor-KO mice) (Artinian et al., 2019). Pvalb ires-Cre mice (Hippenmeyer et al., 2005) were used for Credependent targeting of PV interneurons (Pvalb-Cre mice). Mice were housed 2-4 animals per cage, except for in vivo optogenetic studies in which mice were housed singly after cannula implantation. Food and water were given ad libitum. The mice were maintained on a 12 h light/dark cycle with all experiments performed during the light phase.

EXPERIMENTAL MODEL AND SUBJECT DETAILS
The in vitro electrophysiological slice experiments were performed on 6-7 weeks old male or female mice at the diestrus phase to reduce possible variability in responses. Behavioral studies and ex vivo electrophysiological slice experiments were carried out on 8-10 weeks old male mice.

Hippocampal slice preparation
Six to seven weeks old animals (10-18 days after viral injection) were anesthetized by isoflurane inhalation and the brain was rapidly removed and placed in cold (4 C iScience Article pyramidal cells (TBS opto ) was elicited by TBS consisting of three episodes (given at 30 s intervals) of five bursts (at 300 ms inter-burst interval) of four light pulses at 80 Hz. In some experiments, low frequency optogenetic stimulation (LFS opto ) was given as three episodes (given at 30 s intervals) of three bursts (at 500 ms inter-burst interval) of three light pulses at 20 Hz. Test EPSP amplitude was averaged in 2 min bins before and after LTP induction. EPSPs were characterized in one cell per slice, and responses were analyzed offline using Clampfit (pClamp 10.5/10.6; Molecular Devices) and GraphPad Prism 6.

Field potential recording
For experiments with field potential recordings, transverse hippocampal slices were prepared as described above, except oxygenated ACSF contained 1.3 mM MgSO 4 and 2.5 mM CaCl 2 . The slices were let to recover for at least 2 h at 32 C in ACSF, and for an additional 30 min at 27 -28 C while submerged in a recording chamber continuously perfused (2-2.5 mL/min) with ACSF. Extracellular field EPSPs were recorded with borosilicate glass pipettes (3-4MU) filled with ACSF using a differential extracellular amplifier (Microelectrode AC Amplifier Model 1800, A-M Systems), filtered at 2 kHz, digitized at 10 kHz (Digidata 1440A), and analyzed with pClamp10.5. For Schaffer collateral-evoked fEPSPs, the concentric bipolar tungsten stimulating electrode (FHC) and recording pipette were placed in stratum radiatum. For temporo-ammonic pathway evoked fEPSPs, the stimulation and recording electrodes were positioned in stratum lacunosum-moleculare. Stimulus (0.1 ms duration; 15 s À1 ) strength was adjusted to elicit 50% of maximal fEPSP and fEPSP slope was measured at 10-90% of fEPSP amplitude. CA1 pyramidal cell long-term potentiation was induced by weak theta burst stimulation (wTBS; two bursts of four pulses at 100 Hz, with 200 ms inter-burst interval) (Leao et al., 2012). Optogenetic induction of LTP in SOM interneurons was elicited as described above through a fiber optic positioned above the CA1 stratum oriens/alveus region of the slice.

Optogenetic stimulation in vivo
One week after viral injections, mice were anesthetised using ketamine and xylazine as described above, and fiber cannulas were stereotaxically implanted bilaterally (coordinates relative to bregma: À1.95 AP; G1.60 ML; and À1.15 DV) and sealed with dental cement (Metabond, Parkell Inc). Animals were housed singly after the optic fiber implantation and allowed to rest for a week before behavioral experimentation.

Behavioral experiments
For all behavioral experiments, the experimenter was blind to the experimental groups (opsin versus reporter) from the day habituation began until data analysis was completed.

Habituation
Mice were gently handled daily for 5 days (5min/day) to progressively habituate them to the experimenter and reduce the stress related to the experimental handling and optic fiber connection (without light stimulation).

Open field test
Mice were allowed to freely explore a square arena (45 3 45 cm; Pan-Lab) for 15 min for the Arch experiments and 5 min for the hChR2 experiments. The anxiety level was assessed by measuring the time spent and the distance traveled in the center (1/3 central zone) of the arena compared to the periphery (1/3 peripheral zone). Locomotion was evaluated by measuring the total distance traveled and the number of zone transitions (16 equal squared zones). Mice were first video-tracked at 25 frames/s and their movements subsequently analyzed using a position tracking system (Smart 3.0; PanLab). For the Arch experiments, mice received continuous light stimulation (520 nm) for the 5-10 min period.

Contextual fear conditioning
Mice were trained in conditioning chambers that were housed in sound-and light-isolated cubicles (Coulbourn Instruments). The chambers contained a stainless-steel grid floor, overhead LED lighting, camera and were supplied with background noise (60 dB) by an air extractor fan. The experimental protocol was based on Lovett- Baron  iScience Article context was rectangular with two transparent and two stainless steel walls, was cleaned with 70% ethanol solution before and after each trial, and ethanol solution was added under the grid floor as a contextual smell. After 2 min free exploration, the animals received a weak fear conditioning consisting of two presentations of unconditioned stimulus (2 s foot shock, 0.8mA; separated by 60 s). To test for long-term contextual fear memory, the mice were returned to the training context during a test period of 3 min, at 24 h after conditioning. The freezing behavior was assessed using FreezeFrame (Coulbourn Instruments).
For the Arch optogenetic experiments, mice received light stimulation (520 nm; 6 s pulses) in phase (starting 2 s before) or shifted (30 s before) relative to foot shocks during conditioning. For the TBS optogenetic experiments, mice were brought in the testing room on the training day and, in their home cage, given the optogenetic TBS protocol (473 nm; five bursts of four light pulses of 1ms duration at 80 Hz, given at 300 ms inter-burst interval, and repeated 3 times with 30 s interval) or no light stimulation, 30 min prior to contextual fear conditioning.

Histology
Following in vivo optogenetic experiments, mice were deeply anesthetized intraperitoneally with sodium pentobarbital (MTC Pharmaceuticals), perfused transcardially first with 0.1M PBS (PBS) and next with 4% paraformaldehyde in 0.1M PBS (PFA). Brains were isolated, postfixed in 4% PFA for 24 h and cryoprotected in 30% sucrose. Coronal brain sections (50 mm thick) were obtained using freezing microtome (SM200R; Leica) and mounted in Pro-Long Gold (Invitrogen). Sections were examined using an epifluorescence microscope (Eclipse E600; Nikon) and images were acquired with the Simple PCI software. Immunofluorescence was used to enhance GFP fluorescence associated with Arch. For these mice, brain sections were obtained as described above, permeabilized with 0.5% Triton X-100 in PBS for 15 min, and unspecific binding was blocked with 10% normal goat serum in 0.1% Triton X-100/PBS for 1 h. Rabbit polyclonal GFP (1/200) antibodies were incubated for 48 h at 4 C. Sections were subsequently incubated at room temperature with Alexa Fluor 594-conjugated goat anti-rabbit IgG (1/500; 90 min), mounted in Pro-Long Gold, and imaged as described above. For behavioral experiments, data were excluded if mice did not show virus expression restricted to CA1, and if optic fibers placement was outside the CA1 hippocampus.

QUANTIFICATION AND STATISTICAL ANALYSIS
No statistical methods were used to predetermine the sample size, but our sample sizes are similar to those used in the field. Statistical analysis was performed using SigmaPlot or GraphPad Prism 6 and 9. Data were tested for normality using the Shapiro-Wilk, D'Agostino and Pearson, Kolmogorov-Smirnov, and Brown-Forsythe tests. We used Student's t tests, one-or two-way ANOVA, and one-or two-way mixed repeated-measures ANOVA, with Tukey's pairwise comparison tests (with Bonferroni adjustments for multiple comparisons) or with Holm-Sidak pairwise comparison tests, when data passed normality and homoscedasticity assumptions. Wilcoxon matched pairs signed rank tests, Mann-Whitney Rank Sum tests and Kruskal-Wallis one-way ANOVA on Ranks with Dunn's pairwise comparisons, were used when data did not pass normality and homoscedasticity assumptions. All the tests were two-sided. All data in the  Table S1. Baseline EPSP amplitudes are reported for all groups in Table S2, with Dunn's multiple comparisons tests indicating no differences between SOM interneuron groups.