Hippocampal inputs engage CCK+ interneurons to mediate endocannabinoid-modulated feed-forward inhibition in the prefrontal cortex

Connections from the ventral hippocampus (vHPC) to the prefrontal cortex (PFC) regulate cognition, emotion, and memory. These functions are also tightly controlled by inhibitory networks in the PFC, whose disruption is thought to contribute to mental health disorders. However, relatively little is known about how the vHPC engages different populations of interneurons in the PFC. Here we use slice physiology and optogenetics to study vHPC-evoked feed-forward inhibition in the mouse PFC. We first show that cholecystokinin (CCK+), parvalbumin (PV+), and somatostatin (SOM+) expressing interneurons are prominent in layer 5 (L5) of infralimbic PFC. We then show that vHPC inputs primarily activate CCK+ and PV+ interneurons, with weaker connections onto SOM+ interneurons. CCK+ interneurons make stronger synapses onto pyramidal tract (PT) cells over nearby intratelencephalic (IT) cells. However, CCK+ inputs undergo depolarization-induced suppression of inhibition (DSI) and CB1 receptor modulation only at IT cells. Moreover, vHPC-evoked feed-forward inhibition undergoes DSI only at IT cells, confirming a central role for CCK+ interneurons. Together, our findings show how vHPC directly engages multiple populations of inhibitory cells in deep layers of the infralimbic PFC, highlighting unexpected roles for both CCK+ interneurons and endocannabinoid modulation in hippocampal-prefrontal communication.


Introduction
The prefrontal cortex (PFC) controls cognitive and emotional behaviors (Euston et al., 2012;Miller and Cohen, 2001) and is disrupted in many neuropsychiatric disorders (Godsil et al., 2013;Sigurdsson and Duvarci, 2015). PFC activity is driven and maintained by long-range glutamatergic inputs from a variety of other brain regions (Hoover and Vertes, 2007;Miller and Cohen, 2001). Strong, unidirectional connections from the ventral hippocampus (vHPC) contribute to both working memory and threat conditioning in rodents (Jones and Wilson, 2005;Sotres-Bayon et al., 2012;Spellman et al., 2015). Dysfunction of vHPC to PFC connectivity is also implicated in schizophrenia, anxiety disorders, chronic stress disorders, and depression (Godsil et al., 2013;Sigurdsson and Duvarci, 2015). To understand these roles, it is necessary to establish how vHPC inputs engage local excitatory and inhibitory networks within the PFC.
The vHPC primarily projects to the ventral medial PFC in rodents, with axons most prominent in layer 5 (L5) of infralimbic (IL) PFC (Phillips et al., 2019). These excitatory inputs contact multiple populations of pyramidal cells and are much stronger at intratelencephalic (IT) cells than nearby pyramidal tract (PT) cells (Liu and Carter, 2018). vHPC inputs can drive the robust firing of IT cells (Liu and Carter, 2018), which may be important for maintaining activity during behavioral tasks (Padilla-Coreano et al., 2016;Spellman et al., 2015). However, they also evoke prominent feed-forward inhibition at pyramidal cells (Marek et al., 2018), and excitation and inhibition evolve with different dynamics (Liu and Carter, 2018). Here we focus on the mechanisms responsible for this inhibition by establishing which interneurons are engaged by vHPC inputs to the PFC.
While PV+ and SOM+ interneurons are prominent in the PFC, there is also an unusually high density of cholecystokinin-expressing (CCK+) interneurons (Whissell et al., 2015). In the hippocampus, these inhibitory cells help regulate the motivation and emotional state of animals (Armstrong and Soltesz, 2012;Freund, 2003;Freund and Katona, 2007). They also highly express cannabinoid type 1 (CB1) receptors on their axon terminals (Katona et al., 1999), and can be strongly modulated by endocannabinoids . For example, brief depolarization of postsynaptic pyramidal cells releases endocannabinoids that bind to CB1 receptors on CCK+ axon terminals and inhibit presynaptic GABA release, a process known as depolarization-induced suppression of inhibition (DSI) .
While CCK+ interneurons are prominent in the PFC and may play a role in different forms of inhibition, they remain relatively understudied. A major technical reason is the low-level expression of CCK in pyramidal cells (Taniguchi et al., 2011), which makes CCK+ interneurons challenging to specifically target. For example, expressing Cre-dependent reporters in CCK-Cre transgenics also label pyramidal cells in the cortex (Taniguchi et al., 2011). Fortunately, this challenge can be overcome with intersectional viruses using the Dlx enhancer, which restricts expression to interneurons (Dimidschstein et al., 2016). This approach allows identification of CCK+ interneurons, enabling targeted recordings and optogenetic access to study their connectivity and modulation in the PFC.
Here we examine vHPC-evoked inhibition at L5 pyramidal cells in IL PFC using slice physiology, optogenetics, and intersectional viral tools. We find vHPC inputs activate PV+, SOM+, and CCK+ interneurons, with different dynamics during repetitive activity. Inputs to PV+ and CCK+ interneurons are strong but depressing, while those onto SOM+ interneurons are weak but facilitating. CCK+ interneurons contact L5 pyramidal cells, with stronger connections onto PT cells than neighboring IT cells. However, endocannabinoid modulation via DSI and direct activation of CB1 receptors only occurs at synapses onto IT cells. Endocannabinoid modulation of vHPC-evoked feed-forward inhibition also occurs only at IT cells, highlighting a central role for CCK+ interneurons. Together, our findings show how the vHPC engages interneurons to inhibit the PFC, while revealing a novel property of cell-type-specific endocannabinoid modulation in this circuit.
To confirm the targeting of CCK+ interneurons, we next used whole-cell recordings followed by post-hoc reconstructions ( Figure 1G). We found both axons and dendrites in L5 of IL PFC, and intrinsic properties similar to reports in other parts of the brain ( Figure 1H; Rin = 130 ± 13 MW, Vm = À68 ± 1 mV, Sag = 4.1 ± 1.0%, Adaptation = 0.81 ± 0.04, Tau = 9.1 ± 0.6 ms; n = 12 cells, 4 animals) (Daw et al., 2009). These results confirm that our viral strategy can identify CCK+ interneurons, which are present in deeper layers of IL, and also show that PV+, SOM+, and CCK+ interneurons are positioned to receive vHPC inputs and may mediate feed-forward inhibition.
Having established the targeting of vHPC inputs, we next assessed their ability to drive action potentials (APs) at the three classes of interneurons. We used similar viral and labeling approaches but in this case conducted current-clamp recordings in the absence of TTX and 4-AP and at physiological Ca 2+ concentration (2 mM). We also kept light intensity constant within each set of experiments, at a power that evoked action potentials in at least one of the recorded pair of cells. We found that single vHPC inputs evoked APs in pairs of CCK+ and IT cells but the interneurons showed a higher probability of firing ( Figure 2D; AP probability: IT = 0.32 ± 0.15, CCK+ = 0.84 ± 0.11, p = 0.03; n = 9 pairs, 4 animals). Similarly, we observed that vHPC inputs also preferentially activate PV+ over IT cells ( Figure 2E; AP probability: IT = 0.06 ± 0.06, PV+ = 0.65 ± 0.17, p = 0.04; n = 8 pairs, 4 animals). By contrast, SOM+ interneurons remained unresponsive to vHPC inputs, even when using higher light intensities that were able to activate IT cells ( Figure 2F; AP probability: IT = 0.81 ± 0.12, SOM+ = 0.13 ± 0.06, p = 0.01; n = 7 pairs, 3 animals). These results established a hierarchy for activation, suggesting CCK+ and PV+ interneurons are engaged by vHPC inputs and mediate feed-forward inhibition.

Short-term dynamics differ between populations of interneurons
Repetitive activity at vHPC to PFC connections is functionally important and depends on stimulus frequency (Liu and Carter, 2018;Siapas et al., 2005). At the synaptic level, repetitive activity engages short-term plasticity to change the strength of individual connections (Zucker and Regehr, 2002). We next examined the response to repetitive vHPC inputs by stimulating with brief trains (5 pulses at 20 Hz) in the absence of TTX and 4-AP (Liu and Carter, 2018;McGarry and Carter, 2016). We observed that EPSCs at CCK+ and PV+ interneurons strongly depress over the course of stimulus trains ( Figure 3A & B; EPSC 2 / EPSC 1 : CCK+ = 0.82 ± 0.06; n = 6 cells, 3 animals; PV+ = 0.88 ± 0.06; n = 7 cells, 3 animals). By contrast, the EPSCs at SOM+ interneurons initially facilitated during trains ( Figure 3A & B; EPSC 2 / EPSC 1 = 1.78 ± 0.23; n = 8 cells, 3 animals). These results indicate that vHPC engages all three cell types, with connections at PV+ and CCK+ interneurons strong but depressing, and those at SOM+ interneurons weak but facilitating. The short-term dynamics of vHPC inputs suggested differential engagement of PV+, SOM+, and CCK+ interneurons during repetitive activity. We observed that stimulus trains of vHPC inputs generated decreasing AP probabilities at CCK+ and PV+ interneurons ( Figure   0.70 ± 0.13; n = 6 cells, 3 animals;). By contrast, although SOM+ interneurons did not fire with single pulses of vHPC inputs, they were activated during trains. Importantly, these responses were not due to recurrent network activity, as stimulation intensity was chosen here to ensure that IT cells remained quiescent (Figure 3C & D; stimulus 1 and 2 AP probabilities: IT = 0 and 0, SOM+ = 0 and 0.57 ± 0.14; n = 6 pairs, 4 animals). These findings suggest that vHPC differentially engages three interneuron types with distinct dynamics during repetitive activity, with strong and depressing PV+ and CCK+ activity contributing early, and weaker but facilitating SOM+ activity contributing later.

CCK+ interneurons make connections onto L5 pyramidal cells
Our results indicate that vHPC inputs strongly engage CCK+ interneurons in the PFC, suggesting a role in feed-forward inhibition. However, the connections made by CCK+ interneurons onto different pyramidal cell subtypes are not well established in the PFC. To study CCK+ output, we developed a new virus (AAV-Dlx-Flex-ChR2-mCherry) to express ChR2 in a Cre-dependent manner under the Dlx enhancer (see Materials and methods). We injected this virus into the PFC of CCK-Cre mice to selectively express ChR2 in CCK+ interneurons ( Figure 4A). Whole-cell current-clamp recordings showed that labeled cells could be rapidly and reliably activated by brief pulses of blue light, with a single AP elicited for each pulse during the train ( Figure 4A; stimulus 1 to 5, AP number: 1.0 ± 0.0, n = 6 cells, 3 animals). To study inhibitory connections, we then recorded CCK+-evoked currents from unlabeled L5 pyramidal cells in IL PFC ( Figure 4B). To detect both EPSCs or IPSCs, we used a low chloride internal and held at À50 mV, such that inward currents were EPSCs and outward currents were IPSCs (Glickfeld and Scanziani, 2006). We observed robust CCK+-evoked IPSCs, which were unaffected by blockers of AMPAR (10 mM NBQX) and NMDAR (10 mM CPP) but abolished by blockers of GABA A R (10 mM gabazine) ( Figure 4B; ACSF = 97 ± 21 pA, NBQX+CPP = 97 ± 21 pA, gabazine = 0.4 ± 0.2 pA; ACSF versus NBQX + CPP, p = 0.69; NBQX + CPP versus gabazine, p = 0.01; n = 7 cells, 3 animals). These findings indicate that CCK+ interneurons make inhibitory connections onto neighboring L5 pyramidal cells and that our viral strategy avoids contamination from excitatory contacts due to inadvertent activation of CCK+ pyramidal cells.
Previous studies indicate that inhibitory inputs from PV+ and SOM+ interneurons are strongly biased onto PT cells over nearby IT cells . To test if similar biases occur for CCK+ interneurons, we labeled PT and IT cells by injecting retrograde tracers into periaqueductal gray (PAG) and cPFC, respectively ( Figure 4C). Recording from pairs of pyramidal cells, we found that CCK-evoked IPSCs were larger onto PT cells than IT cells ( Figure 4D; IPSC IT = 113 ± 59 pA, PT = 245 ± 79 pA, p = 0.01; n = 7 pairs, 4 animals). These findings indicate that CCK+ interneurons make cell-type specific connections, preferentially targeting PT cells in L5 of the IL PFC.
Related studies on the hippocampus show that pyramidal cell innervation by CCK+ interneurons can vary along the somato-dendritic axis (Lee et al., 2010). We next used subcellular channelrhodopsin assisted circuit mapping (sCRACM) to study the subcellular targeting of CCK+ interneurons onto defined projection neurons in the PFC (Petreanu et al., 2009). We expressed ChR2 in CCK+ interneurons and recorded from retrogradely labeled IT and PT cells in the presence of 1 mM TTX and 10 mM 4-AP to enable terminal activation and isolate monosynaptic inputs (Little and Carter, 2012). We activated CCK+ inputs across the entire somato-dendritic axis using a grid (10 Â 10) of pseudorandomly delivered (1 Hz . Therefore, in contrast to the strong dendritic innervation by SOM+, 5HT3aR+, and NDNF+ cells (Anastasiades et al., 2020;Marlin and Carter, 2014), these results indicate that CCK+ inputs are primarily near the soma of IT and PT cells.

Expression specificity of CB1Rs
Throughout the brain, CCK+ interneurons are distinguished from PV+ cells by enrichment of the cannabinoid type one receptor (CB1R) on presynaptic terminals (Bodor et al., 2005;Dudok et al., 2015;Katona et al., 1999). To examine the expression of CB1R in the IL PFC, we next used immunocytochemistry and detected CB1R+ puncta located at PV+ and CCK+ axons. We found CB1R+ puncta showed higher co-localization with CCK+ axons (based on GFP+ expression) than with PV+ axons ( Having already established that CCK+ outputs preferentially target PT cells, we next examined if CB1R+ expression is cell-type specific. We labeled IT and PT cells by injecting retrogradely transported AAVrg-TdTomato into the cPFC and AAVrg-GFP into the PAG, respectively. We then used immunocytochemistry to examine CB1R+ puncta surrounding the cell bodies of neighboring IT and PT cells ( Figure 4H). While we observed CB1R+ puncta around the cell bodies of both cell types ( Figure 4I), their density was much greater at IT cells ( Figure 4J; IT = 9.3 ± 0.5 puncta/cell, PT = 5.7 ± 0.5 puncta/cell, p<0.0001; n = 4 animals, 247 IT cells, 207 PT cells). These results indicate that presynaptic CB1Rs are more prominent at perisomatic connections onto IT cells, suggesting that endocannabinoid modulation may be more extensive at those projection neurons.
To test if equivalent CB1R modulation also occurs under more physiological conditions, we performed related current-clamp recordings. We held IT cells at À50 mV to create a driving force and before APs = 11.3 ± 1.5 mV, after APs = 7.7 ± 1.1 mV, recovery = 11.8 ± 1.4 mV; before versus. after APs, p = 0.015; n = 7 cells, 4 animals). However, an additional hyperpolarization also occurred after the APs, which could reduce the driving force for any IPSPs. To account for this possibility, we also recorded CCK+-evoked IPSPs at the matched hyperpolarized membrane potential for each cell and observed intermediate responses (

Discussion
We have explored several new features related to the organization and modulation of connections from vHPC to PFC ( Figure 7F). First, we found vHPC contacts and strongly activates CB1R-expressing CCK+ interneurons in L5 of IL PFC. Second, we showed that CCK+ interneurons contact nearby pyramidal cells, suggesting they participate in feed-forward inhibition. Third, we found that CCK+ connections undergo CB1R-mediated modulation and DSI, which is selective for IT and not PT cells. Fourth, endocannabinoids also modulate vHPC-evoked inhibition, which also undergoes DSI selectively at IT cells. Together, our results reveal a central role for CCK+ interneurons and endocannabinoid modulation in communication between vHPC and PFC.
The PFC possesses a rich variety of GABAergic interneurons, which are known to have unique roles in goal-directed behaviors (Abbas et al., 2018;Courtin et al., 2014;Kepecs and Fishell, 2014;Kvitsiani et al., 2013). Interestingly, the PFC has fewer PV+ interneurons and more CCK+ interneurons compared to other cortices (Kim et al., 2017;Whissell et al., 2015). Our results indicate that CCK+ interneurons are abundant in L5 of IL PFC, which we previously showed receives the strongest connections from vHPC (Liu and Carter, 2018). We found these cells are distinct from PV+ and SOM+ interneurons, with different morphological and physiological properties. We also found that they are enriched in presynaptic CB1Rs, as in other parts of cortex, hippocampus, and amygdala (Armstrong and Soltesz, 2012;Bodor et al., 2005;Vogel et al., 2016). The abundance of CCK+ interneurons in the ventral mPFC is consistent with their role in cognitive and emotional behaviors (Freund, 2003;Freund and Katona, 2007). In the future, it will be interesting to characterize CCK+ interneurons in other layers and subregions of the PFC.
One of our key results is that vHPC inputs densely contact and strongly activate CCK+ interneurons in L5 of IL PFC. vHPC inputs are strong and depressing onto CCK+ interneurons, in contrast to the facilitating inputs onto IT cells (Liu and Carter, 2018). Importantly, vHPC-evoked firing of CCK+ interneurons occurs without activation of IT cells, indicating polysynaptic recurrent activity is not required for activation of CCK+ interneurons, which are therefore likely to mediate feed-forward inhibition. With higher intensity of input stimulation, CCK+ interneurons may also be activated by local inputs, as observed in the hippocampus (Glickfeld and Scanziani, 2006), and similar to SOM+ interneurons in the cortex (Silberberg and Markram, 2007). Indeed, in the hippocampus, CCK+ interneurons have been shown to participate in both feed-forward and feed-back networks (Basu et al., 2013;Glickfeld and Scanziani, 2006). Our findings show that CCK+ interneurons have an important and underappreciated role in hippocampal evoked feed-forward inhibition, and it will be important to assess whether this generalizes to other inputs to the PFC.
While we focused on CCK+ interneurons, we also confirmed that vHPC inputs engage other interneurons in L5 of IL PFC. The activation of PV+ interneurons is strong, consistent with their role in feed-forward inhibition in PFC and elsewhere Cruikshank et al., 2007;Delevich et al., 2015;Gabernet et al., 2005;McGarry and Carter, 2016). Previous results have indicated that vHPC engages PV+ interneurons in superficial layers of the medial PFC (Marek et al., 2018). By contrast, we find particularly strong connections in L5, where our previous study shows vHPC inputs are most dense (Liu and Carter, 2018). In the future, it will also be interesting to address the relative contribution of PV+ and CCK+ interneurons in feed-forward inhibition, which may be accomplished by novel genetic tools that allow for the labeling of CCK+ and PV+ Before depol. After depol. Recovery interneurons in the same animal (Fenno et al., 2014). SOM+ interneuron activation is weak, but increases with repetitive activity, similar to BLA inputs to superficial PFC (McGarry and Carter, 2016). The engagement of SOM+ interneurons is consistent with a role in oscillations linking the vHPC and PFC and involvement in working memory (Abbas et al., 2018). Interestingly, the activation of SOM+ interneurons by long-range inputs also occurs in the granular sensory cortex, where it also builds during stimulus trains, suggesting this is a general property (Tan et al., 2008).
Our intersectional viral approach enabled cell-type specific identification and optogenetic activation of cortical CCK+ interneurons. We found that CCK+ interneurons make robust inhibitory connections onto neighboring pyramidal cells in L5 of IL PFC. Importantly, these connections are much stronger onto PT cells, similar to our previous findings for PV+, SOM+, and NDNF+ interneurons (Anastasiades et al., 2020;Anastasiades et al., 2018). Biased inhibition is thus a general property of inhibitory connections in the PFC, although strength also depends on intrinsic properties . In the hippocampus, CCK+ interneurons also make unique connections onto different pyramidal cell populations defined by sublayer (Valero et al., 2015). While hippocampal CCK+ innervations vary along the somato-dendritic axis of pyramidal cells (Bodor et al., 2005;Lee et al., 2010), our sCRACM results suggest CCK+ interneurons target the soma of both IT and PT cells in L5 of IL PFC. This subcellular targeting could explain why we observed little asynchronous CCK+ release, which is stronger at dendritic synapses compared to somatic synapses at CCK+ connections in the hippocampus (Daw et al., 2009;Lee et al., 2010).
Another key finding was that endocannabinoid modulation of CCK+ connections depends on the postsynaptic target. Due to their distinct postsynaptic receptor expression profiles, IT and PT cells are known to respond differently to many neuromodulators, including dopamine, acetylcholine, and serotonin (Anastasiades et al., 2019;Baker et al., 2018;Dembrow and Johnston, 2014;Shepherd, 2013;Stephens et al., 2018). We observed robust endocannabinoid-mediated DSI at IT cells but not at nearby PT cells, despite the latter receiving stronger connections. Our result suggests the endocannabinoid system also shows marked cell-type specificity, with selective modulation on cortico-cortical networks. In the future, it will be important to assess the functional impact of this specificity on local processing in the PFC, including processing within and between hemispheres. It will also be important to establish if this specificity extends to other layers of the PFC, as IT cells are distributed from superficial L2 to deep L6 (Anastasiades et al., 2019).
In principle, selective DSI at CCK+ connections onto IT cells could reflect differences in the postsynaptic release or presynaptic detection of endocannabinoids. Our results are consistent with the latter explanation. First, our immunocytochemistry shows more CB1R puncta around IT cells, suggesting these receptors are selectively localized. This result was particularly surprising because we also found CCK+ interneurons make stronger connections onto PT cells. Second, direct activation of CB1Rs with WIN reduces CCK+-evoked IPSCs only at IT cells, with no effect at neighboring PT cells. This experiment bypasses the postsynaptic release of endocannabinoids, suggesting a presynaptic mechanism accounts for the specificity of DSI. The increase in PPR is also consistent with presynaptic modulation by CB1R, indicating reduced release probability . By contrast, a recent study of differential DSI at CCK+ connections onto projection neurons in the amygdala suggested a post-synaptic mechanism, due to selective expression of endocannabinoid-synthesizing enzyme DGLa in pyramidal cells (Vogel et al., 2016). These findings indicate that different mechanisms leading to the modulation specificity may occur in different brain regions, underscoring the complexity of endocannabinoid systems.
The absence of WIN modulation at CCK+ connections onto PT cells, despite the low level of CB1R presence around their cell bodies, was surprising. Previous studies indicate that the nanoscale organization of CB1Rs relative to voltage-gated calcium channels influences the effectiveness of presynaptic CB1Rs (Dudok et al., 2015). This explains the weaker endocannabinoid modulation of dendritic-targeting than somatic-targeting CCK+ inputs in the hippocampus (Lee et al., 2010). Although we showed CCK+ interneurons restrict their targeting to the soma of both IT and PT cells, different nanoscale CB1R organizations could still occur at boutons, making the CB1R at CCK+ inputs onto PT neurons non-functional. Alternatively, some PT-targeting CB1R+ boutons may come from other interneuron subtypes, as previous studies showed that a small percentage of non-CCK+ cells also express CB1R (Bodor et al., 2005;Marsicano and Lutz, 1999).
A related question is whether CB1R+ and CB1R-connections arise from the same CCK+ interneurons. One possibility is that two different types of CCK+ interneurons exist: one that projects to IT cells and is sensitive to CB1R modulation and another that projects to PT cells and is insensitive to CB1R modulation. Indeed, several subtypes of CCK+ interneurons are found across the brain, which differ in their subcellular targeting of postsynaptic cells, the size of their soma, and the expression of molecular markers such as VIP and VGLUT3 (Bodor et al., 2005;del Pino et al., 2017;Lee et al., 2010;Omiya et al., 2015;Pelkey et al., 2020;Somogyi et al., 2004). However, another possibility is that the same CCK+ interneuron projects to both projection neurons but the contacts at PT cells lack CB1R modulation (Bodor et al., 2005;Dudok et al., 2015;Lee et al., 2010). Distinguishing between these possibilities is challenging, but may be accomplished in the future using triple recordings from CCK+ interneurons and pyramidal cells (Reyes et al., 1998), facilitated by soma-restricted optogenetic tools (Collins et al., 2018;Mardinly et al., 2018), which can be combined with our Dlx viral approach to specifically target CCK+ interneurons.
Ultimately, the ability of vHPC inputs to engage CCK+ interneurons, which in turn contact pyramidal cells, implicates a key role in feed-forward inhibition. Consistent with this idea, we observed prominent CB1R-mediated DSI of vHPC-evoked inhibition only at IT cells and not neighboring PT cells. Because there is no change in excitation, this modulation will selectively increase the excitation/inhibition (E/I) ratio at IT cells. In the intact brain, this could allow the vHPC to more effectively activate IT cells compared to neighboring PT cells. For example, when IT cells are highly active, increased endocannabinoid tone could promote local processing within the PFC. Disinhibition at IT cells could also allow for stronger responses to excitatory inputs, a potential mechanism for increased synchrony between hippocampus and PFC during working memory tasks and anxiety-like behavior (Adhikari et al., 2010;Fujisawa and Buzsáki, 2011;O'Neill et al., 2013). Furthermore, altered E/I balance could potentially shift the output of the PFC toward other intratelencephalic targets throughout the brain, including other parts of the cortex, striatum, amygdala, and claustrum (Anastasiades et al., 2019;Harris and Shepherd, 2015).
Lastly, our results have implications for the functional properties of hippocampal-prefrontal networks in health and disease (Euston et al., 2012;Peters et al., 2010;Sierra-Mercado et al., 2011;Sotres-Bayon et al., 2012). Previous studies have shown that endocannabinoid signaling can strongly influence both cognition and emotion (Mechoulam and Parker, 2013). Altering endocannabinoid levels also affects executive function, working memory, stress, anxiety, and threat learning (Lin et al., 2009;Lutz et al., 2015;Marcus et al., 2020;Marsicano et al., 2002;Volk and Lewis, 2016). Our results indicate endocannabinoids influence communication between vHPC to PFC by selectively modulating connections from CCK+ interneurons to IT cells. In the future, it will be particularly interesting to explicitly assess the role of this microcircuit and the impact of endocannabinoid modulation on PFC function and dysfunction, including threat learning and anxiety disorders (Peters et al., 2010;Sierra-Mercado et al., 2011;Sotres-Bayon et al., 2012).

Stereotaxic injections
Mice aged 4-6 weeks were deeply anesthetized with either isoflurane or a mixture of ketamine and xylazine, then head-fixed in a stereotaxic (Kopf Instruments). A small craniotomy was made over the injection site, using these coordinates relative to Bregma: PFC = ±0.4, -2.3, +2.1 mm, PAG = À0.6, both À2.5 and À3, -4.0 mm, vHPC = À3.3, both À3.6 and À4.2, -3 mm (mediolateral, dorsoventral, and rostrocaudal axes). For retrograde labeling, pipettes were filled with red retrogradely transported fluorescent beads (Lumafluor), Cholera Toxin Subunit B (CTB) conjugated to Alexa 647 (Life Technologies), or viruses. Borosilicate pipettes with 5 to 10 mm diameter tips were back-filled with dye and/or virus, and a volume of 130-550 nl was pressure-injected using a Nanoject III (Drummond) every 30 s. The pipette was left in place for an additional 5 min, allowing time to diffuse away from the pipette tip, before being slowly retracted from the brain. For both retrograde and viral labeling, animals were housed for 2-3 weeks before slicing.

Histology and fluorescence microscopy
Mice were anesthetized with a lethal dose of ketamine and xylazine, then perfused intracardially with 0.01 M phosphate-buffered saline (PBS) followed by 4% paraformaldehyde (

Electrophysiology
Whole-cell recordings were obtained from pyramidal neurons or interneurons located in layer 5 (L5) of infralimbic (IL) PFC. Neurons were identified by infrared-differential interference contrast or fluorescence, as previously described (Chalifoux and Carter, 2010). In the case of pyramidal cells, the projection target was established by the presence of retrobeads or Alexa-conjugated CTB, as previously described (Little and Carter, 2013). Pairs of adjacent cells were chosen for sequential recording, ensuring they received similar inputs (typically < 50 mm between cells). Borosilicate pipettes (2-5 MW) were filled with internal solutions. Three types of recording internal solutions were used. For current-clamp recordings (in mM): 135 K-gluconate, 7 KCl, 10 HEPES, 10 Na-phosphocreatine, 4 Mg 2 -ATP, and 0.4 Na-GTP, 290-295 mOsm, pH 7.3, with KOH. For voltage-clamp recordings (in mM): 135 Cs-gluconate, 10 HEPES, 10 Na-phosphocreatine, 4 Mg 2 -ATP, and 0.4 Na-GTP, 0.5 EGTA, 10 TEA-chloride, and 2 QX314, 290-295 mOsm, pH 7.3, with CsOH. For DSI experiments (in mM): 130 K-gluconate, 1.5 MgCl 2 , 10 HEPES, 1.1 EGTA, 10 phosphocreatine, 2 MgATP, 0.4 NaGTP. In some experiments studying cellular morphology, 5% biocytin was also included in the recording internal solution. After allowing biocytin to diffuse through the recorded cell for at least 30 min, slices were fixed with 4% PFA before staining with streptavidin conjugated to Alexa 647 (Invitrogen). Electrophysiology recordings were made with a Multiclamp 700B amplifier (Axon Instruments), filtered at 4 kHz for current-clamp, and 2 kHz for voltage-clamp, and sampled at 10 kHz. The initial series resistance was <20 MW, and recordings were ended if series resistance rose above 25 MW. In some experiments, 1 mM TTX was added to block action potentials, and 100 mM 4-AP and 4 mM external Ca 2+ to restore presynaptic release. In many experiments, 10 mM CPP was used to block NMDA receptors. In current-clamp experiments characterizing intrinsic properties, 10 mM NBQX, 10 mM CPP, and 10 mM gabazine were used to block excitation and inhibition. In some experiments, 10 mm AM-251 was used to block CB1 receptors or 1 mM WIN 55,212-2 was used to activate CB1 receptors. All chemicals were purchased from either Sigma or Tocris Bioscience.

Optogenetics
Channelrhodopsin-2 (ChR2) was expressed in presynaptic neurons and activated with a brief light pulse from a blue LED (473 nm) (Thorlabs). For wide-field illumination, light was delivered via a 10 Â 0.3 NA objective (Olympus) centered on the recorded cell. LED power was routinely calibrated at the back aperture of the objective. LED power and duration were adjusted to obtain reliable responses, with typical values of 0.4 to 10 mW and 2 ms, respectively. Subcellular targeting experiments were performed with a Polygon DMD device (Mightex) focused through a 10 Â 0.3 NA objective (Olympus) with a 75 mm pixel size. Pulses were delivered at 1 Hz using a pseudo-random 10 Â 10 grid pattern, yielding an effective mapping area of 750 mm Â 750 mm. Experiments used a 2 ms LED pulse yielding an effective power of 0.17 mW per pixel.

Data analysis
Electrophysiology and imaging data were acquired using National Instruments boards and MATLAB (MathWorks) (Pologruto et al., 2003). Off-line analysis was performed using Igor Pro (WaveMetrics). Intrinsic properties were determined as follows. Input resistance was calculated from the steadystate voltage during a À50 pA, 500 ms current step. Voltage sag ratio was calculated as (V sag À V ss ) / (V sag À V baseline ), where V sag is average over a 1 ms window around the minimum, V ss is average of last 50 ms, and V baseline is average of 50 ms preceding the current injection. The membrane time constant (tau) was measured using exponential fits to these hyperpolarizations. Adaptation was calculated as the ratio of the first and last inter-spike intervals, such that a value of 1 indicates no adaptation and values < 1 indicate lengthening of the inter-spike interval. For experiments with a single optogenetic stimulation, the PSC amplitude was measured as the average value across 1 ms around the peak subtracted by the average 100 ms baseline value prior to the stimulation. For experiments with a train of optogenetic stimulation, each PSC amplitude was measured as the average value in a 1 ms window around the peak, minus the average 2 ms baseline value before each stimulation. Most summary data are reported in the text and figures as arithmetic mean ± SEM. Ratios of responses at pairs of cells are reported as geometric mean in the text, and with ± 95% confidence interval (CI) in the figures, unless otherwise noted. Comparisons between unpaired data were performed using non-parametric Mann-Whitney test. Comparisons between data recorded in pairs were performed using non-parametric Wilcoxon test. Two-tailed p values < 0.05 were considered significant.