The main observations of the study were that (i) silencing SOM-INs during training results in impairment of object location memory, indicating that activity of SOM-INs supports object location learning; (ii) TBSopto given 30 min prior to training results in facilitation of object location memory, suggesting that induction of synaptic plasticity at PC-SOM synapses facilitates object location learning; and (iii) TBSopto given 30 min prior to training did not result in facilitating object location memory in SOM-Cre-Raptor-KO mice, suggesting that facilitation of object location learning by TBSopto in SOM-Cre-EYFP mice was due to induction of mTORC1-mediated synaptic plasticity at PC-SOM synapses. Thus, hippocampal somatostatin interneuron activity is required for object location learning, a hippocampus-dependent form of novelty motivated spatial learning that is facilitated by plasticity at PC-SOM synapses.
SOM cells activity is necessary for encoding object location memory
Our results indicate that SOM-IN activity is necessary during object location learning. SOM-IN activity is also required for linking context to fear. In contextual fear conditioning, inhibition of SOM-INs during the application of the conditioning stimulus (shocks) reduces contextual fear memory, but inhibition during the context exploration does not [2, 3]. In a passive avoidance task, in which mice avoid an attractive context that is associated with an aversive stimulus, inhibition of the OLMa2 sub-population of SOM-INs during training also reduces fear memory . Our results show that SOM-IN activity is not only necessary for associating fear to context, but also for encoding object relations in space in a novelty-driven task that does not involve aversive stimuli. Although we did not provide electrophysiological confirmation of optogenetic inhibition of SOM-INs by Arch activation in the present work, we showed in a previous report with whole-cell recording in slices that optogenetic activation of Arch hyperpolarizes SOM-INs . A paradoxical release of neurotransmitters from Arch-expressing terminals has been reported with prolonged stimulation with yellow light , which could potentially have interfered with our behavioral experiments. Additional experiments using the light-gated chloride pump halorhodopsin (eNpHR3.0) which lacks such paradoxical effects would be useful to confirm our results. However, we have previously shown that sustained Arch activation does not affect basal transmission at Schaffer collateral synapses in hippocampal slices . In addition, we found previously that in open-field control experiments, 5 min continuous yellow light activation or Arch inhibition of SOM-INs did not affect mice exploration, anxiety and locomotion , suggesting that the Arch-induced impairment of object location learning was due to SOM-IN silencing. Thus, our findings of an essential role of SOM-IN activity in object location learning, specifically in a mismatch novelty paradigm, extend the necessary role of SOM-IN activity in encoding a wide range of aspects of the context and its relation to salient events in mice.
Interestingly, SOM-IN activity after contextual fear conditioning supports memory consolidation . If the same timing to plasticity is involved in non-aversive behavior—such as the object location task—it would be useful to determine if, using the protocols developed in the present study, activation or inhibition of SOM-IN function, either before or after the object location learning phase, affects consolidation of this memory circuit. Also, the role of SOM-INs vary along the dorso-ventral axis of the hippocampus. Inhibiting OLMa2 cells of intermediate CA1 during passive avoidance training had no effect on fear memory. Furthermore, an opposite role was found in the object recognition task, another novelty motivated memory task where the animal learns the features of objects to differentiate familiar and novel objects. Inhibition of dorsal OLMa2 cells during training has no effect on novel object recognition, while inhibition of intermediate OLMa2 cells facilitates novel object recognition . In this context, it will be interesting to examine the role of SOM-INs in object location memory according to their position along the dorso-ventral axis.
LTP at PC-SOM synapses facilitates object location memory
Our results indicate that object location memory is facilitated by optogenetic induction of long-term potentiation of SOM interneuron excitatory afferents. It is noteworthy that control mice in our three experimental groups (Figs. 1D, 2D, 3D) did not show similar levels of object location learning. The reason for this discrepancy is unclear but may be due to differences in experimental protocols. For example, the timeline of surgical procedures differed in the Arch and hChR2 groups. Viral injection and cannulation were performed in two separate sessions separated by a one-week recovery period for Arch experiments, while viral injection and cannulation were carried out in one session for hChR2 experiments. However, the three groups of control mice showed normal anxiety and locomotion in the open-field test (Fig. 1E–G, 2E–G, 3E–G). Alternatively, the TBSopto control experiments included an additional session with connection to the optic probe without optogenetic stimulation in home cage, and 30 min later the training session (Fig. 1B versus 2B), which may have negatively affected the object location learning during the training session. However, it is important to note that both control groups in SOM-Cre-EYFP and SOM-Cre-Raptor-KO mice showed similar behavior in the object location learning task, i.e. sub-threshold for learning, validating the comparison of the effects of TBSopto between these mice. The results that facilitation of object location learning was absent in SOM-Cre-Raptor-KO mice indicates that facilitation of object location learning by TBSopto requires mTORC1 function in SOM-INs, suggesting that LTP at PC-SOM synapse positively regulates object location learning.
Previous work indicates that TBSopto given before contextual fear conditioning has the opposite effect and negatively regulates contextual fear memory . In addition, optogenetic induction of PC-SOM synapse LTP prior to contextual fear conditioning, leads to a reduction of subsequent contextual fear conditioning-induced LTP at PC-SOM synapses . In slices, TBSopto given 15 min prior to chemically inducing PC-SOM synapses LTP blocks the protein synthesis normally produced by the chemical LTP induction , strengthening the idea that TBSopto interacts with later LTP induction. Interestingly, passive avoidance learning induces long-term potentiation in a subset of pyramidal neurons and this learning-induced potentiation occludes subsequent high frequency stimulation-induced LTP . Thus, an analogous situation may occur in SOM-INs, with induction of LTP by TBSopto at PC-SOM synapses preventing and even reducing LTP induced subsequently during contextual fear conditioning, leading to a deficit in contextual fear memory .
Hence the question of why TBSopto-induced LTP has different effects on contextual fear and object location memory, when SOM-IN activity is required for both contextual fear and object location learning? An explanation could be that depotentiation of PC-SOM synapses may be required for object location learning. This may be analogous to plasticity at pyramidal cell synapses, where object location learning generally induces long-term depression at Schaffer collateral synapses onto pyramidal cells . LTD of SC-PC synapses may passively propagate to PC-SOM synapses  and negatively affect metaplasticity of Schaffer collateral synapses of the CA1 network during learning . Thus, it would be interesting to examine how long-term depression at PC-SOM synapses regulates plasticity of CA3 and entorhinal inputs to pyramidal cells, and how it is influenced by prior induction of TBSopto.
However in pyramidal cells, the saturation of hippocampal LTP impairs spatial learning in the Morris water maze task . Also, in mice with cell-specific impairment of mTORC1 activity in SOM-INs, basal synaptic transmission is normal, but LTP is impaired at PC-SOM synapses, and spatial memory in the Barnes maze is deficient. Conversely, in mice with cell-specific facilitation of mTORC1 activity in SOM-INs, LTP at PC-SOM synapses and spatial memory in the Barnes maze are increased . Thus, LTP at PC-SOM synapses appears to be necessary for spatial learning. Consequently, a more plausible explanation for the opposite effect of TBSopto on contextual fear and object location memory, may be that in the present control conditions object location learning was subthreshold for the consolidation of memory, and therefore, long-term plasticity at PC-SOM synapses was likely not induced by learning in these control training conditions. In such “weak” training conditions, prior induction of PC-SOM synapse LTP by TBSopto may not lead later to occlusion of learning-induced long-term plasticity at PC-SOM synapses but to facilitation, resulting in memory formation. Hence, induction of PC-SOM synaptic plasticity by contextual fear and spatial learning may differ, resulting in different sensitivity to occlusion. It will be important in future experiments to determine the plasticity mechanisms induced by spatial learning at PC-SOM synapses. Given that we observed TBSopto-induced facilitation of object location learning in "weak” training conditions, it would be important to test if TBSopto also induces a facilitation of object location learning in “stronger” training conditions (i.e. that induce learning), to clarify the role of PC-SOM plasticity in spatial learning. Since learning of novel object location involves an interplay of LTD, LTP, and metaplasticity at CA1 PC synapses [23, 25], it would also be pertinent to examine how LTP at PC-SOM synapses modulates CA1 network metaplasticity induced by novel object location. Whether synaptic plasticity of PC-SOM synapses also regulates other novelty mismatch learning paradigms  is another interesting question to address.
Our results of facilitation of object location memory by TBSopto suggests that LTP at PC-SOM synapse regulates hippocampal spatial memory. Because TBSopto was given in vivo, it may have stimulated other synaptic targets of CA1 pyramidal cells, such as subicular neurons, potentially resulting in plasticity at these other output synapses and influencing hippocampal memory [39,40,41]. To address this possibility, the effect of TBSopto was examined in mice with conditional knock-out of Rptor in SOM-INs in which mTORC1-mediated LTP at PC-SOM synapses is blocked . The finding that facilitation of object location memory is absent in these mice, suggests that facilitation of object location memory was due to TBSopto-induced plasticity in SOM-INs, and not to actions via other synaptic targets of CA1 pyramidal cells. The finding is also consistent with previous work indicating that, in parvalbumin expressing interneurons, which are another synaptic target of CA1 pyramidal cells, TBSopto does not induce LTP at PC-parvalbumin cell synapses . Interestingly, a subset of the subicular neurons targeted by CA1 pyramidal cells, project back to CA1 pyramidal neurons. Moreover, this feedback from the subiculum is important for formation of object location memory and regulation of pyramidal cell place fields, but not for object recognition . Furthermore, these subicular neurons that feedback to CA1 are inhibited by CA1 SOM-INs , likely SOM projection cells . It will be interesting to determine if the synaptic inputs of these SOM projection cells are potentiated by TBSopto and if so, how this regulates the subicular-CA1 network and hippocampal memory. Finally, another useful point in future experiments may be to relate the degree of pyramidal cell recruitment, perhaps visualized with immediate early gene expression, with the TBSopto-induced changes in behavioral performance. Such experiments would provide important information about network changes associated with PC-SOM synaptic plasticity and learning.
Our findings are consistent with previous demonstrations that the optogenetic LTP induction protocol (TBSopto) in vitro elicits specific long-term potentiation at PC-SOM synapses during whole-cell recordings from SOM-INs in slices, and in vivo regulates contextual fear learning-induced LTP at PC-SOM synapses . However, a caveat of our work is that we did not provide direct proof that the TBSopto induces LTP in vivo. Hence, it would be important to develop an in vivo assay for LTP at PC-SOM synapses to confirm directly the optogenetic induction of LTP in vivo . In addition, it is unlikely that the effects of TBSopto in the present study were due to light-generated heat effect since previous whole-cell experiments in slices have shown that the same TBSopto given to control slices from mice with pyramidal cell expression of mCherry (and no hChR2) did not affect transmission at PC-SOM synapses, and neither did TBSopto given to hChR2-expressing pyramidal cells in the presence of a mGluR1a antagonist, or in mice with a conditional knockout of Rptor, an essential component of mTORC1, in SOM-INs . Moreover, in previous behavioral experiments, TBSopto given to mice expressing mCherry (and no hChR2) in pyramidal cells showed normal contextual fear learning compared to unstimulated control mice . Finally, in the present study, TBSopto did not facilitate object location learning in mice with a conditional knockout of Rptor in SOM-INs (Fig. 3), suggesting that TBSopto facilitation of object location memory in control mice was unlikely to be due to light-generated heat effects.
A second caveat of our study is the use of mice with global conditional knock-out of Rptor in SOM cells that will impair mTORC1 function in SOM cells in other non-hippocampal brain regions and may result in non-specific behavioral changes. However previous work has shown that object location memory is critically dependent on dorsal CA1 hippocampus, compared to other brain regions [26,27,28]. In addition, these mice display normal exploratory behavior, anxiety level and locomotion in the open field test (Fig. 3), intact sensorimotor gating measured during fear conditioning, and intact non-hippocampal memory such as auditory-cued fear memory [3, 19]. Hence, the effect of the global conditional deletion of Rptor in SOM cells is most likely due to interfering with hippocampal SOM-IN mTORC1 function rather than non-hippocampal effects. However, regional- and cell-specific conditional deletion of Rptor would be useful for confirmation.