Hypothalamic Glutamate/GABA Cotransmission Modulates Hippocampal Circuits and Supports Long-Term Potentiation

Subcortical input engages in cortico-hippocampal information processing. Neurons of the hypothalamic supramammillary nucleus (SuM) innervate the dentate gyrus (DG) by coreleasing two contrasting fast neurotransmitters, glutamate and GABA, and thereby support spatial navigation and contextual memory. However, the synaptic mechanisms by which SuM neurons regulate the DG activity and synaptic plasticity are not well understood. The DG comprises excitatory granule cells (GCs) as well as inhibitory interneurons (INs). Combining optogenetic, electrophysiological, and pharmacological approaches, we demonstrate that the SuM input differentially regulates the activities of different DG neurons in mice of either sex via distinct synaptic mechanisms. Although SuM activation results in synaptic excitation and inhibition in all postsynaptic cells, the ratio of these two components is variable and cell type-dependent. Specifically, dendrite-targeting INs receive predominantly synaptic excitation, whereas soma-targeting INs and GCs receive primarily synaptic inhibition. Although SuM excitation alone is insufficient to excite GCs, it enhances the GC spiking precision and reduces the latencies in response to excitatory drives. Furthermore, SuM excitation enhances the GC spiking in response to the cortical input, thereby promoting induction of long-term potentiation at cortical-GC synapses. Collectively, these findings provide physiological significance of the cotransmission of glutamate/GABA by SuM neurons in the DG network. SIGNIFICANCE STATEMENT The cortical-hippocampal pathways transfer mnemonic information during memory acquisition and retrieval, whereas subcortical input engages in modulation of communication between the cortex and hippocampus. The supramammillary nucleus (SuM) neurons of the hypothalamus innervate the dentate gyrus (DG) by coreleasing glutamate and GABA onto granule cells (GCs) and interneurons and support memories. However, how the SuM input regulates the activity of various DG cell types and thereby contributes to synaptic plasticity remains unexplored. Combining optogenetic and electrophysiological approaches, we demonstrate that the SuM input differentially regulates DG cell dynamics and consequently enhances GC excitability as well as synaptic plasticity at cortical input-GC synapses. Our findings highlight a significant role of glutamate/GABA cotransmission in regulating the input-output dynamics of DG circuits.

SuM neurons form synapses with the perisomatic region of the GC, and their axonal terminals coexpress the vesicular glutamate transporter Type II (VGluT2) and vesicular GABA transporter (VGAT) (Boulland et al., 2009;Soussi et al., 2010;Root et al., 2018). Notably, VGluT2 and VGAT are segregated to distinct synaptic vesicles at the SuM terminals in the DG (Boulland et al., 2009;Root et al., 2018). The segregated localization of neurotransmitter vesicles in the same terminals suggests differential cotransmission of glutamate and GABA at the SuM-DG synapses (Dugué et al., 2005;Somogyi, 2006;Vaaga et al., 2014). Consistent with this observation, SuM terminals in the DG simultaneously release both glutamate and GABA to GCs and GABAergic interneurons (INs) (Pedersen et al., 2017;Hashimotodani et al., 2018;Billwiller et al., 2020). Of note, the ratio of the glutamate-and GABA-mediated components recorded in INs varied from 0.34 to 7.7 (Hashimotodani et al., 2018). Given the diverse types of INs present in the DG, glutamate and GABA are likely to be differentially cotransmitted in an IN subtype-specific manner Hosp et al., 2014;Liu et al., 2014;Hsu et al., 2016;Booker and Vida, 2018). Differential recruitment of distinct IN subtypes can powerfully modulate the input and output logic of DG (Miles et al., 1996;Lee et al., 2016). However, whether the SuM input differentially recruits distinct IN subtypes in the DG remains unknown. Moreover, GABA, which is cotransmitted with glutamate by the SuM, is known to exert shunting inhibitory effects on GCs and thereby could bidirectionally control action potential firing in GCs (Chiang et al., 2012;Heigele et al., 2016). Yet it is unclear how the SuM input regulates the input-output dynamics of DG circuits.
Here, combining electrophysiological and optogenetic approaches, we demonstrate that SuM input differentially regulates the activity of DG neurons. Optogenetic activation of SuM input was able to excite dendrite-targeting INs (D-INs), but was not sufficient to activate soma-targeting INs (S-INs) and GCs. Consistent with these observations, GCs and S-INs received predominantly synaptic inhibition, whereas, D-INs received predominantly synaptic excitation. As a consequence, activation of the SuM input enhances the temporal precision of GC firing and shortened spike latencies in D-INs. Moreover, coactivation of the SuM input with the cortical input enhanced the responses of GCs to the cortical input. Finally, repeated coactivation of the SuM and cortical inputs resulted in enhanced LTP at the cortical-GC synapses.

Materials and Methods
Animals. We used the VGluT2-Cre driver line (Slc17a6 tm2(cre)Lowl /J, stock #016963), VGAT-Cre driver line (Slc32al tm2(cre)Lowl /J, stock #028862), and Gad2-Cre driver line (Gad2 tm2(cre)Zjh /J, stock #010802) obtained from The Jackson Laboratory, and WT mice with C57BL/6J genetic background obtained from National Laboratory Animal Center. Both male and female mice (3-5 months old) were used for the electrophysiological experiments. The mice were housed in a room with a reverse 12 h light/12 h dark cycle and were provided with food and water ad libitum. The protocols and procedures for the animal experiments were in accordance with the national and institutional guidelines and were approved by the Animal Care and Use Committee of National Yang Ming Chiao Tung University.
Stereotaxic injection. For the retrograde tracer and virus injections, the mice were anesthetized with 4% isoflurane (v/v; Halocarbon Laboratories) in a 100% oxygen-containing induction chamber. The scalp was shaved, and the mice were transferred to a stereotaxic frame (IVM-3000; Scientifica) for the surgery. The mouth and nose of each mouse were covered using an anesthetizing mask that was supplied with ;1.5% isoflurane and had an airflow rate of 4 ml/min. To maintain the body temperature of the mice at 34°C-36°C, a biological temperature controller pad (Physitemp Instruments, or TMP-5b, Supertech Instruments) remained placed under the body of each mouse throughout the surgical procedure. The head was fixed using two ear bars; 75% ethanol was applied to the scalp to sterilize the surgical area, and an ophthalmic gel was applied to the eyes to avoid dryness. An analgesic (ketorolac, 6 mg/kg) was administered intraperitoneally. For the delivery of the tracer, unilateral or bilateral craniotomy was performed at the AP and ML coordinates of the dorsal DG (AP: À1.80 mm, ML: 61.30 mm). Then the tracer was delivered into the DG at the DV coordinate (DV: À2.20 and À2.0 mm). To target the SuM neurons, unilateral or bilateral craniotomy was performed over the SuM (AP: À2.85 mm, ML: 60.15 mm). Then viral vectors were delivered into the SuM at DV, À4.86 mm. The viral vectors (0.2-0.4 ml) and red retrobeads (0.2 ml) (LumaFlour) were delivered to the SuM and DG, respectively, using a 10 ml NanoFil syringe (World Precision Instruments) and a 34-G beveled metal needle. The injection volume (0.2-0.4 ml) and flow rate (0.1 ml/min) were controlled using a nanopump controller (KD Scientific). Subsequently, the needle was raised 0.1 mm above the site of injection for an additional 10 min to minimize the upward flow of the viral solution. Finally, the needle was gradually withdrawn. After the injection was performed, the incision was sutured, and the mice were transferred to the cage for recovery.
Preparation of brain slices. Acute brain slices containing the hippocampal and SuM sections were prepared 1 week after the retrograde tracer injection or at least 3 weeks after the viral injection. Transverse brain slices were used for whole-cell patch-clamp recording of the DG neurons, while coronal brain slices were used for recording of retrobeadpositive SuM neurons. The mice were anesthetized using isoflurane and decapitated rapidly. The brains were quickly removed and transferred to an ice-cold oxygenated (95% O 2 and 5% CO 2 ) sucrose solution containing the following (in mM): 87 NaCl, 25 NaHCO 3 , 1.25 NaH 2 PO 4 , 2.5 KCl, 10 glucose, 75 sucrose, 0.5 CaCl 2 , and 7 MgCl 2 . Next, 300-mm-thick slices were cut using a vibratome (DTK-1000; Dosaka). After sectioning, the slices were recovered at 34°C for 25 min in a holding chamber filled with an oxygenated sucrose solution, then transferred to room temperature (25 6 2°C) for additional experiments.
Electrophysiology and optical stimulation. For the recordings, individual slices were transferred to a submerged chamber and were continuously perfused with oxygenated ACSF containing the following (in mM): 125 NaCl, 25 NaHCO 3 , 1.25 NaH 2 PO 4 , 2.5 KCl, 25 glucose, 2 CaCl 2 , and 1 MgCl 2 . The ChR2-eYFP expression pattern was confirmed using fluorescence, and the neurons in the DG were selected visually for recording under an infrared differential interference contrast microscope (IR-DIC, BX51WI, Olympus). The axonal terminals that expressed ChR2 were stimulated with 470 nm light transmitted through the objective from an LED source (LED4D162, driven by DC4104, Thorlabs).
Whole-cell patch-clamp recordings were performed using a Multiclamp 700B amplifier (Molecular Devices). The recording electrode pipettes (4-7 MV) pulled from borosilicate glass tubing (outer diameter, 1.5 mm; inner diameter, 0.86 mm; Harvard Apparatus) were filled with a high Clinternal solution, containing the following (in mM): 15 K-gluconate, 140 KCl, 0.1 EGTA, 2 MgCl 2 , 4 Na 2 ATP, 10 HEPES, 0.5 Na 3 GTP, and 0.4% biocytin (w/v, Invitrogen). In certain set experiments for the determination of spike-timing precision and spike phase, a low Clinternal solution containing the following (in mM): 136.8 K-gluconate, 7.2 KCl, 0.2 EGTA, 4 MgATP, 10 HEPES, 0.5 Na 3 GTP, 7 Na 2 -phosphoreatine (pH 7.3 with KOH) and 0.4% biocytin was used. The pipette capacitance was compensated in the cell-attached mode. To measure the EPSC and the IPSC, whole-cell recording was performed using a high Clinternal solution (E GABA = ;0 mV, E AMPA = ;0 mV), and the EPSCs and IPSCs were isolated using a pharmacological approach. Bath application of SR95531 (1 mM) and CGP55845 (1 mM) was used to block GABA A and GABA B receptors, respectively, while an ionotropic glutamate receptor blocker, kynurenic acid (Kyn, 2 mM), was used to block ionotropic glutamatergic transmission. The GABAergic component (IPSC) traces were obtained by digital subtraction of traces recorded after bath application of SR, CGP from the baseline traces recorded in the presence of ACSF. The glutamatergic component (EPSC) traces were obtained by digital subtraction of traces recorded in the presence of SR, CGP, and Kyn from the traces recorded in the presence of SR and CGP.
Cell-attached recording was performed with patch pipettes filled with a high Clinternal solution before whole-cell recording of current spikes in GCs and INs. A 5 Hz, 5 ms light pulse was applied with a 15 s intersweep interval, and 6 sweeps were recorded. The spike probability was determined as the percentage of spikes among 6 sweeps. In the dual recording experiments, the distance between the recorded pair was ,200 mm. Although the serial resistance was not compensated, it was monitored continuously during the recording process. The recordings with the serial resistance , 25 MV were analyzed. Fast-spiking phenotype of hippocampal INs or putative S-INs recorded at room temperature (21°C-24°C) were defined by their maximal firing rate . 65 Hz and coefficient of variation of , 0.2 in response to 1 s depolarizing current injection (Lien and Jonas, 2003). The coefficient of variation was determined from the spike train with the maximal firing rate. For local field potential (LFP) recordings, a monopolar electrode (tip diameter; ;10 mm) filled with ACSF was placed in the subiculum to stimulate the perforant path (PP) fibers. Trains of current pulses (10-500 mA, 0.1 ms) were applied every 15 s using a stimulus isolator (Isoflex, A.M.P.I.). The recording electrode (tip diameter, ;5 mm) filled with ACSF was placed in the granule cell layer (GCL) to monitor the population spike (pSpike) in response to PP stimulation. Additional experiments were performed at stimulus intensities that evoked 30%-50% of the maximum pSpike amplitude and paired with the 10 ms light pulse for activation of the SuM input.
For the spike-timing precision experiments, sinusoidal waveforms were created and customized using Clampfit 10.3 (Molecular Devices). To test the ability of the SuM input to enhance spike-timing precision and phase, theta frequency (5 Hz trains of 5 pulses) sinusoidal current pulses were delivered into the GCs and were paired with 5 Hz square photostimulation of the SuM input. The 5 ms photostimulation was delivered during the ascending phase (31°-39°) of the sinusoidal waveform. The current injected (peak to trough, 50-150 pA) was set to evoke a single action potential close to the peak of the sinusoidal waveform while the membrane potential of the GCs was held at ;À80 mV. Twenty sweeps were recorded at 15 s interval and superimposed to observe the precision of action potential generation. To determine the spike jitter and phase, the time point for the peak in each spike was converted to phase (angle) using the customized Python codes. The mean and the SD represented spike phase (latency) and spike jitter, respectively. All cells used for spike-timing precision experiments reliably generated EPSP in response to 5 Hz photostimulation of the SuM input. The signals were recorded using Multiclamp 700B amplifiers (Molecular Devices), filtered at 4 kHz, and sampled at 10 kHz using a digitizer (Digidata 1440A, Molecular Devices), which was controlled using pCLAMP version 10.3 (Molecular Devices).
Next, the slices were rinsed 3 times with PBS and incubated in cocktails of fluorescent secondary antibodies, AlexaFluor-488 anti-rabbit, AlexaFluor-594 anti-rabbit, and AlexaFluor-647 anti-mouse at room temperature for 2 h or overnight at 4°C. The procedures were performed under continuous shaking conditions. After rinsing 6 times with PBS, the sections were mounted using the mounting medium Vectashield with DAPI. Fluorescent images were taken using a confocal microscope (Leica SP5 module, Leica Microsystems) or (LSM 700, Carl Zeiss) using 20Â, 40Â, or 63Â objectives and analyzed using ImageJ (National Institutes of Health, 1.52t). Single-plane coronal sections with bead expression were imaged using a Research High-Class Stereo Microscope System (SZX16, Olympus). For colocalization analysis of ChR2-eYFPexpressing boutons with VGluT2 and VGAT, boutons along ChR2-eYFP expressing axons were identified in z-stack images, examined for colocalization, and counted using cell counter plugin in Fiji (a distribution of ImageJ software, National Institutes of Health, 1.53c) (Billwiller et al., 2020).
Data analysis and statistics. Data were analyzed using Clampfit 10.3 (Molecular Devices), Prism 6.0 (GraphPad Software), or customized Python codes. The synaptic latency was determined as the time elapsed from the light onset to the onset of the synaptic response (Hsu et al., 2016). The onset of the synaptic response was determined by the intersection of a line through the 20% and 80% points of the rising phase of the EPSC or IPSC and the baseline. To calibrate evoked IPSCs during successive 5 Hz photostimulation, the EPSC obtained after bath application of SR95531 (1 mM) and CGP55845 (1 mM) was digitally subtracted from the mixed postsynaptic current (baseline). To calculate the conductance, the EPSC and the IPSC amplitudes were divided by their respective driving forces. The input resistance was determined by the ratio of a steady-state (the last 100 ms of a 1 s pulse) voltage response versus the injected 1 s hyperpolarizing (10 pA) current pulse (Liu et al., 2014). The magnitude of LTP was calculated 30-40 min after LTP induction. Data are presented as mean 6 SEM. Error bars in figures also show SEMs. Statistical significance was tested using the unpaired t test, Mann-Whitney test, Wilcoxon signed-rank test, or two-way repeatedmeasures ANOVA followed by Bonferroni's post hoc tests.

Results
Anatomical and physiological features of DG-projecting SuM neurons To identify and characterize the morphophysiological properties of DG-projecting SuM neurons, a retrograde tracer (red retrobeads) was injected into the bilateral DG in the hippocampus (Fig. 1A, left) of 3 mice. The injection sites were confirmed by post hoc serial coronal sections (Fig. 1A, middle). The beads were restricted to the GCL and hilus of the DG (Fig. 1A, right top). One week after the injection, the retrogradely labeled DG-projecting neurons were detected primarily in the lateral subdivision of the SuM (SuML) above the mammillothalamic tract (mt) (Fig.  1A, right bottom). Only few labeled cells were detected in the medial subdivision of the SuM (SuMM) (Fig. 1A, right bottom) as reported previously (Soussi et al., 2010). Notably, in mice injected unilaterally in the right DG ( Fig. 1B), the labeled DGprojecting SuM cells were mostly detected ipsilateral to the injection side (Fig. 1C,D; data from 12 slices, 3 mice). Next, we performed whole-cell recordings from labeled DG-projecting SuM neurons located in the SuML in brain slices prepared from both bilateral and unilateral DG-injected mice (Fig. 1E). These cells had large cell bodies (!20 mm in diameter; Fig. 1E), with a resting membrane potential of À58.0 6 1.7 mV (n = 11 cells from 5 mice) and an input resistance of 508.3 6 69.4 MX (n = 11 cells from 5 mice). They exhibited a bursting firing pattern (at holding potential of À70 mV) in response to small current injection (10-30 pA) and displayed an accommodating firing pattern in response to increased depolarizing current ( Fig. 1F; n = 11 cells from 5 mice). The biocytin-filled SuM cells exhibited axonal projection extending toward the dorsal brain areas with dendrites located within the mammillary region ( Fig. 1G; n = 5 cells from 4 mice).
Next, we used an optogenetic approach to investigate the function of SuM projections. A CaMKIIa-ChR2-eYFP virus was injected into the SuM of WT mice (Fig. 1H, top). The SuM neuron projections were observed to form a dense pattern in the supragranular layer of the GCL and CA2 pyramidal layer (Fig.  1H, bottom, from 3 mice). To confirm that the ChR2-expressing SuM neurons respond to light stimulation, we made whole-cell recording from these neurons (Fig. 1I). When the recorded neurons were illuminated with blue light pulses (470 nm, 5 ms at 5 Hz), they generated spikes in current clamp at À70 mV ( Fig.  1I, traces; n = 7 cells, 5 mice). Similarly, a light-evoked ChR2mediated inward current was recorded in voltage clamp in the presence of an ionotropic glutamate receptor antagonist, Kyn (2 mM) (Fig. 1I, traces). Consistent with previous studies (Boulland et al., 2009;Soussi et al., 2010;Hashimotodani et al., 2018;Root et al., 2018), the ChR2-eYFP-expressing axon terminals in the DG (Fig. 1J) coexpressed VGluT2 and VGAT (Fig. 1K). A total of 1381 putative boutons (from 9 slices, 2 mice) were identified along the ChR2-eYFP-expressing axons. Overall, 92 6 1.4% (85%-98%) of the boutons expressed VGluT2, 88 6 2.3% (82%-97%) expressed VGAT, while 84 6 2.3% (78%-94%) expressed both VGluT2 and VGAT, similar to previous reports ( SuM input preferentially excites dendrite-targeting INs Next, we examined SuM-DG synaptic transmission by recording field EPSPs (fEPSPs) along the somatodendritic axis of GCs ( Fig. 2A, top). The fEPSPs exhibited downward at the GCL (À0.10 6 0.01 mV; n = 7) and inner molecular layer (À0.06 6 0.01 mV; n = 7). The polarity of fEPSP reversed at the middle molecular layer (0.03 6 0.00 mV; n = 7) and exhibited upward at the outer molecular layer (0.03 6 0.00 mV; n = 7). This was consistent with the observation that SuM axons mainly innervated the somatic and proximal dendritic regions of GCs (Hashimotodani et al., 2018). Then, we tested whether activation of SuM terminals alone was sufficient to excite any DG neurons. To this end, we injected a CaMKIIa-ChR2-eYFP virus into the SuM of WT mice or EF1a-DIO-ChR2-eYFP virus into VGluT2-Cre mice. Next, cell-attached recordings were performed from various types of DG neurons, such as GCs, S-INs, and D-INs ( Fig. 2A, bottom), and followed by biocytin-filled whole-cell recordings for post hoc morphologically identification (Liu et al., 2014;Hsu et al., 2016;Lee et al., 2016). Dentate GCs receive coherent theta (4-10 Hz)-band EPSCs in vivo (Pernía-Andrade and Jonas, 2014), and the SuM synchronizes with the DG . Thus, we investigated the response of DG cells to SuM activation at a physiologically relevant frequency (e.g., 5 Hz). Upon photostimulation of SuM axons (5 Hz, 5 ms pulses), no spikes were evoked in all recorded GCs (  (Freund and Buzsáki, 1996;Hsu et al., 2016). According to their soma locations and the input layers where their axons innervate, there are at least four distinct subtypes, including the total molecular layer cells (TML cells), hilar PP-associated cells (HIPP cells), molecular layer PP-associated cells (MOPP cells), and hilar commissural-associational pathway-related cells (HICAP cells) (Fig.  2E). Based on the results of morphologic reconstructions, the spike probability of each subtype was plotted against the stimulus number (Fig. 2F). The five nonresponsive D-INs, including two HICAP, two HIPP, and one MOPP, were not included in the plots here. Collectively, the SuM input alone was sufficient to activate most D-INs, but not GCs and S-INs.
Intriguingly, unlike GCs and S-INs, the coapplication of GABA A and GABA B receptor blockers SR95531 (1 mM) and CGP55845 (1 mM) slightly reduced the postsynaptic current recorded in most D-INs (Fig. 3D, bottom). However, further bath application of Kyn completely blocked the remaining large current, indicating a dominant excitatory transmission at the SuM-D-IN synapses (Fig. 3D). The pharmacologically isolated EPSC and IPSC ( Fig. 2D; EPSC, red trace and IPSC, blue trace) exhibited similar synaptic latencies (Fig. 3E, D Mann-Whitney test). The plot of EPSG versus IPSG recorded from each cell revealed a clear shift toward excitatory conductance (Fig. 3G, violet circles), and the slope was .1 (Fig. 3G). In another set of experiments of VGluT2-Cre mouse virally injected with EF1a-DIO-ChR2-eYFP (Fig. 4A), the monosynaptic cotransmission of the glutamate and GABA was also pharmacologically verified by adding TTX, a voltage-dependent sodium channel blocker, and 4-aminopyridine (4-AP), a voltage-dependent potassium channel blocker (Fig. 4B, GC and Fig. 4E, D-IN). The light-evoked postsynaptic current was completely abolished by bath application of TTX (1 mM) and was reversed by subsequent addition of 4-AP (1 mM; in the presence of TTX). Consistent with a previous report (Hsu et al., 2016), synaptic latencies were significantly increased by 4-AP ( Fig. 4C; SuM-GC; synaptic latency, baseline, 2.24 6 0.11 ms ms; TTX & 4-AP, 4.01 6 0.28 ms; n = 9 cells; 5 mice; Fig. 4F, SuM-D-IN; synaptic latency, baseline, 2.67 6 0.21 ms; TTX & 4-AP, 3.66 6 0.17 ms; n = 6 cells; 4 mice). Analysis of the EPSG and IPSG further confirmed that GABAergic transmission dominated at the SuM-GC synapse (Fig. 4D), whereas glutamatergic transmission was predominant at the SuM-D-IN synapse (Fig. 4G). Moreover, the scatter plot of all EPSGs and IPSGs obtained from individual cells revealed a slop of 0.14 at the SuM-GC synapse and a slop of 1.40 at the SuM-D-IN synapses (Fig. 4H). Similar results were obtained from GCs recorded in VGAT-Cre and Gad2-Cre transgenic mice (Fig. 4I-L). In addition to GCs and INs, we also checked the functional connectivity between the SuM input and mossy cells (MCs), which are excitatory neurons located in the hilus and featured by prominent thorny excrescences at their proximal dendrites (Fig. 5A). We performed sequential wholecell recordings from GCs and MCs (Fig. 5A). Consistent with a recent report that MCs rarely receive synaptic input from SuM (Hashimotodani et al., 2018), only 1 of 5 MCs (4 mice) recorded received the discernible response to photostimulation of the SuM input (Fig. 5D), and the current was small (À42 pA; Fig. 5D). The summary plots of first EPSG and IPSG obtained from different cell types in the DG are shown in Figure 5C, D.
Finally, to exclude the possibility that the distinct synaptic properties observed here were because of variable viral expression from slices to slices, we performed another set of experiments in WT mice (Fig. 6A), where simultaneous dual recordings of GCs and D-INs were obtained from the same slices (Fig. 6B). We found that photostimulation of SuM input (5 ms, 470 nm, 5 Hz light pulses) in the DG evoked inward currents in both GCs and INs (Fig. 6C, black traces, 6 of 7 pairs recorded). Coapplication of SR95531 (1 mM) and CGP55845 (1 mM) blocked ;70.5 6 5.0% of current in GCs, only ;25.5 6 5.5% was blocked in D-INs, and Kyn (2 mM) completely blocked the remaining current in both GCs and D-INs (Fig. 6C). The synaptic strength was stronger at the SuM-D-INs synapses compared with that at the SuM-GC synapses (Fig. 6D). Consistent with this, analysis of the peak excitatory and inhibitory conductances (EPSG 1 and IPSG 1 ) in some cells revealed that inhibitory transmission dominated at the SuM-GC synapses (Fig. 6E, left, EPSG 1 ; 0.22 6 0.05 nS, IPSG 1 ; 0.52 6 0.10 nS; n = 5 cells; 4 mice; p , 0.05; U = 2.0; Mann-Whitney test), while excitatory transmission dominated at the SuM-D-IN synapses (Fig. 6E, right, EPSG 1 ; 1.24 6 0.26 nS, IPSG 1 ; 0.40 6 0.09 nS; n = 5 cells; 4 mice; p , 0.01; U = 0.0; Mann-Whitney test). Together, these results demonstrated that the ratio of excitatory and inhibitory components at SuM-DG synapses depends on the subtypes of target cells.
SuM input shortens spike latency and enhances spike-timing precision Cortical principal neurons fire with large variability in response to identical stimuli in vivo (Shadlen and Newsome, 1998;Fricker and Miles, 2001;Carandini, 2004). Well-timed inhibition from GABAergic transmission is known to promote precise spike timing, which is essential for hippocampal network oscillation and is thought to be critical for several cognitive functions (Bacci and  , 2006;Woodruff and Sah, 2007;Hou et al., 2016). Here, we explored how SuM-driven synaptic excitatory and inhibitory conductances regulate spike generation in GCs and D-INs using the low chloride internal solution [Cl -] i = 7.2 mM, which is close to the physiological intracellular chloride concentration (Chiang et al., 2012). To simulate in vivo membrane oscillations, GCs and D-INs were driven by injecting sinusoidal current steps at low theta (5 Hz) frequencies (Fig. 7). Under this condition, photostimulation of the SuM input at the ascending phase of each theta cycle slightly increased spike numbers in GCs (Fig. 7A,B; see Materials and Methods). Given that D-INs received predominantly synaptic excitation on SuM activation, we next examined the modulatory effect of SuM activation on spike generation in D-INs in response to the same oscillatory input. Compared with the light-off epoch, photostimulation of the SuM input remarkably increased spike numbers in D-INs in response to sinusoidal current injections (Fig. 7C,D). Next, we examined the latency and spike jitter in GCs and D-INs by injecting a constant suprathreshold sinusoidal current, which was near enough to generate single spikes near the peak of each theta cycle (GCs, Fig. 7E; D-INs, Fig. 7H). Superimposition of spike trains from GCs (Fig. 7E) showed that SuM stimulation shortened the spike latencies and decreased spike jitters (Fig.  7E, traces). Both reduction in spike latencies and jitters were only significant in first spike (Fig. 7F,G), which could be explained by strong synaptic depression at the SuM to GC synapses. Notably, superimposition of spike trains from D-INs showed that pairing the SuM input with the suprathreshold sinusoidal stimulation (baseline-to-peak current amplitude of 80 pA) greatly reduced spike latencies (Fig. 7I). In great contrast to GCs, photostimulation of the SuM input did not have a significant effect on spike jitters in D-INs (Fig. 7J). This result was consistent with our observation of high synaptic excitation and low synaptic inhibition at the SuM-D-IN synapses. Together, activation of SuM input differentially regulates spike generation in GCs and D-INs.

Huguenard
SuM input enhances GC excitability, thereby supporting LTP Subcortical inputs modulate GC responses to cortical inputs in vivo (Nakanishi et al., 2001;Li et al., 2020). In the DG circuits, the equilibrium potential of GABAergic conductance (E GABA ) is ;À72 mV (Chiang et al., 2012), which is more depolarized than the resting potential of GCs (ranging from À80 to À90 mV). Thus, GABA, which is cotransmitted with glutamate by the SuM, could exert either the "shunting inhibitory" or "depolarizing (or excitatory)" effect on GCs. Our previous studies (Chiang et al., 2012;Hsu et al., 2016) report that GABA could promote action potential generation in GCs. Next, we investigated the functional relevance of glutamate/GABA cotransmission on GC responses to the excitatory PP input. We performed LFP recordings in the GCL in response to photostimulation of the SuM input and/or electrical stimulation of the PP input (Fig. 8A). The evoked response consisted of the fEPSP and pSpike, a proxy of synaptic strength and GC activity, respectively. Photostimulation of the SuM input evoked the fEPSP but did not generate the pSpike (Fig. 8B, black trace), whereas electrical stimulation of the PP generated a compound response, which consisted of the fEPSP followed by the pSpike (Fig. 8B, gray area trace). Notably, paired activation of the PP and SuM inputs significantly increased the pSpike area (Fig. 8B, blue area trace), indicating an increase in GC spike numbers. The summated trace obtained by digital summation of SuM-fEPSP and PP response was shown in the red trace (Fig. 8B, arithmetic sum). Finally, we overlaid all traces and revealed that the SuM-fEPSP emerged before the onset of pSpikes (Fig. 8B, overlay). In sum, the pSpike area induced by coactivation of SuM and PP inputs was significantly larger than that of summated trace (Fig. 8C, left). Notably, there was no significant change in the relative slope of fEPSP (Fig. 8C, right). Further analysis of successive GC responses to either PP activation alone or coactivation of PP and SuM during the 5 Hz trains (Fig. 8D, top traces) showed significant increases in the pSpike area (Fig. 8D, bottom left plot), but not in the fEPSP slope (Fig.  8D, bottom right plot). The lack of changes in the fEPSP slope during coactivation of PP and SuM supports the anatomic finding that SuM axons preferentially innervate the proximal part of GC dendrites.
We hypothesize that the excitatory effect of SuM activation on GCs could enhance LTP induction. To test this hypothesis, we stimulated the cortical input to GCs using a weak protocol (e. g., 20 Hz train stimulation) without and with SuM activation (Fig. 8E). After train stimulation, we measured the changes in the synaptic responses. For the SuM 1 PP protocol, the electrical stimulation of the PP and photostimulation of the SuM input were timed to occur simultaneously (Dt = 0 ms; Fig. 8E, left). The pSpikes were monitored after induction of LTP (Fig. 8E). Notably, 20 Hz PP stimulation alone could not induce LTP (black circles); however, pairing it with photostimulation of the SuM input (20 Hz, 4 trains, 470 nm, 10 ms) resulted in an increase in pSpike and fEPSP slope (Fig. 8F,G). Collectively, the SuM input enhanced GC responses to cortical inputs, thereby facilitating induction of LTP at the PP-GC synapses.

Discussion
Glutamate and GABA are packed in distinct vesicles at the SuM terminals (Boulland et al., 2009;Root et al., 2018). Therefore, the loading, release, and recycling of these two neurotransmitters at the SuM terminals are likely to be regulated differentially. In this study, we demonstrated that glutamate/GABA coreleasing SuM neurons establish synapses with GCs and various subtypes of GABAergic INs in the DG. Notably, the synaptic excitation and The target cell-dependent excitation and inhibition at the SuM-DG synapses may be important for precise processing of neural information (Liu, 2004;Turrigiano and Nelson, 2004). We demonstrated a dominant inhibitory transmission at the SuM-S-IN synapses (Fig. 3C), which might be responsible for weak disynaptic somatic inhibition in GCs (Hashimotodani et al., 2018). Feedforward inhibition is believed to enhance spike timing precision by curtailing EPSPs (Pouille and Scanziani, 2001). The reduced disynaptic feedforward inhibition appears to be compensated by cotransmission of GABA along with glutamate at SuM-GC synapses. The imbalance of synaptic excitation and inhibition has been associated with neurologic disorders, including epilepsy, autism spectrum disorders, schizophrenia, addiction, depression, and social dysfunction (Yizhar et al., 2011;Shabel et al., 2014;Meye et al., 2016). Consistent with this notion, the SuM fibers in the supragranular layer extend aberrant axonal sprouting to the inner molecular layer and are mostly VGluT2 1 in an epileptic rat model (Soussi et al., 2015).
A proposed modulatory role of SuM in the DG network Here, we proposed a network mechanism by which the SuM input modulates the input-output logic of the DG network (Fig. 9). As shown by our experimental data, SuM neurons corelease glutamate and GABA. According to our study, S-INs receive greater synaptic inhibition than excitation (E , I), whereas D-INs receive stronger synaptic excitation than inhibition (E . I). Moreover, only D-INs generate spikes in response Figure 8. SuM input promotes GC responses to cortical input, thereby enhancing LTP at the PP-GC synapses. A, Experimental schematic showing a stimulation electrode (stim.) placed in the subiculum to electrically activate the PP fibers, a field-recording electrode in the GCL to monitor LFP and pSpike, and a blue light for photostimulation of the SuM axon terminals in the GCL. B, Representative traces of SuM-mediated fEPSP (black trace) after photostimulation, PP-mediated pSpike (filled area in gray) on electrical stimulation, and a pSpike (filled area in light blue) after the coactivation (Dt = 0 ms) of the SuM and PP. Red represents the arithmetic sum of fEPSP and pSpike. The traces of pSpikes were superimposed and aligned with fEPSP. C, Left bar graph, Summary plots of the pSpike areas evoked by SuM1PP coactivation (light blue) and arithmetic sum of SuM-evoked fEPSP and PP-evoked pSpike (light red). Areas were normalized to pSpike area evoked by the PP alone. SuM1PP coactivation, 1.43 6 0.16; SuM1PP arithmetic sum, 1.17 6 0.05; n = 6; p = 0.0313. Right bar graph, Summary plots of relative fEPSP slope, SuM1PP coactivation, 1.01 6 0.02; SuM1PP arithmetic sum, 0.97 6 0.01; n = 6; Wilcoxon signed-rank test. D, Top, Representative traces of pSpike responses to PP stimulation alone (black traces) and SuM1PP (blue traces) during a 5 Hz train. Bottom left, Summary of the effect of SuM activation on PP-evoked pSpikes versus stimulus number. PP, n = 6; PP 1 SuM, n = 6; p , 0.05; two-way ANOVA with Bonferroni post hoc test. Right, fEPSP slope before and after photostimulation of the SuM input. PP, n = 6; SuM1PP, n = 6; two-way ANOVA with Bonferroni post hoc test. E, Left, Representative traces of baseline pSpikes in response to stimulation of PP alone. Middle, LTP induction protocol consisting of four trains of 20 Hz electrical stimulation of the PP alone at 15 s intertrain interval (top) or coactivation of the PP and 20 Hz, 4 trains, 10 ms photostimulation of the SuM input (bottom). Right, Sample traces of pSpikes after LTP induction. F, Time course of the normalized pSpike area recorded from the GCL in response to 20 Hz, 4 trains stimulation of PP inputs alone (black circles), or coactivation of the PP input stimulation and 20 Hz photostimulation of the SuM input (blue circles). PP alone, 104.8 6 8.59%; n = 6; SuM1PP, 167.6 6 5.30%; n = 6; p = 0.0009; paired t test. G, Time course of the normalized fEPSP slope of pSpikes recorded from the GCL in response to 20 Hz, 4 train stimulation of PP inputs alone (black circles), or coactivation of PP input stimulation with 20 Hz photostimulation of the SuM input (blue circles). PP alone, 110.6 6 2.20%; n = 6; SuM1PP, 128.5 6 5.19%; n = 6; p = 0.0598; paired t test. Error bars indicate mean 6 SEM. *p , 0.05.
to SuM activation (Fig. 9A), whereas S-INs respond with biphasic subthreshold potential changes (fast EPSP and slow IPSP). Our previous studies demonstrated that single action potential generation in D-INs hardly triggers synaptic release onto GCs (Liu et al., 2014) and is therefore ineffective in modulating the GC output . Thus, SuM activation alone primarily causes small excitatory (red) and large inhibitory (blue) conductance changes around the somata of GCs (Fig. 9A). As shown by our previous study (Chiang et al., 2012), GABA is depolarizing as the E GABA (;À72 mV) . resting membrane potential in GCs and could promote spike generation in GCs in response to the cortical input. The summation of the glutamate-and GABAmediated conductances therefore results in subthreshold postsynaptic depolarization in GCs (Fig. 9A). In great contrast to the SuM input, the PP input alone is sufficient to evoke spikes in S-INs (Liu et al., 2014;Lee et al., 2016). Accordingly, we propose that coactivation of SuM and PP inputs can trigger action potentials in both D-INs and S-INs (Fig. 9B) glutamatergic, monosynaptic glutamatergic-GABAergic, and disynaptic somatic GABAergic conductance changes in GCs, respectively (Fig. 9B). In line with our experimental data, the synaptic summation of these inputs results in action potential generation in GCs (Fig. 9B). During 20 Hz coactivation of the PP and SuM inputs, both D-INs and S-INs generate repetitive spikes (Fig. 9C). Notably, D-INs dramatically increase their synaptic output while they fire at burst frequency .20 Hz (Liu et al., 2014). Accordingly, activation of the PP, SuM, S-INs, and D-INs results in monosynaptic glutamatergic, monosynaptic glutamatergic-GABAergic, disynaptic somatic, and disynaptic dendritic GABAergic conductance changes in GCs, respectively (Fig. 9C). Overall, the synaptic summation of these inputs at 20 Hz results in multiple action potentials in GCs (Fig. 9C), which is supported by our experimental data (Fig. 8D). The enhanced spike generation in GCs during LTP induction is believed to be essential during the induction of Hebbian LTP.
After LTP induction, the pSpike was greatly enhanced (Fig.  8F), whereas the fEPSP was modestly enhanced (Fig. 8G). Although several potential mechanisms could account for these changes, a parsimonious explanation is the formation of Hebbian LTP (Fig. 9D). Specifically, activity-dependent Hebbian LTP is accompanied by synaptic potentiation or a long-lasting increase in GC excitability as demonstrated by enhanced EPSPspike (E-S) coupling (Fig. 8F). Alternatively, the enhancement of E-S coupling after LTP induction could be mediated through network mechanisms. Given that the fEPSP at the PP-GC synapse was modestly increased (Fig. 8G), we proposed that the D-IN-GC synapse may undergo weak LTD (iLTD), resulting a slight increase in the fEPSP (Fig. 8G) after LTP induction. In contrast, the S-IN-GC synapse undergoes strong iLTD, resulting in a large decrease in somatic inhibition and therefore a large increase in the pSpike (Fig. 8F). The future work is to investigate the changes in the synaptic efficacy at individual synapse in the DG circuits after LTP induction.
Cortical and subcortical afferents differentially recruit distinct types of DG INs Extrinsic inputs differentially activate subtypes of GABAergic INs in the DG and play important roles in gating information transmission to the hippocampus (Hefft and Jonas, 2005;Ewell and Jones, 2010;Armstrong et al., 2011;Chiang et al., 2012;Liu et al., 2014;Hsu et al., 2016;Lee et al., 2016). We recently demonstrated that the commissural fibers of hilar MCs provide a strong excitatory drive to the S-INs, and D-INs, including ML cells and TML cells, while the medial PP provides strong excitatory input to the S-INs (Hsu et al., 2016). In contrast, HIPP and HICAP cells receive weak excitatory inputs from the PP and are weakly recruited by the commissural fibers of hilar MCs (Hsu et al., 2016). This study revealed that activation of the SuM input alone can reliably recruit HIPP and HICAP cells. We have shown that both HIPP and HICAP cells dynamically regulate dendritic excitability of GCs (Liu et al., 2014). They weakly inhibit GCs when they fire sparsely, whereas they inhibit GCs robustly in the burst spiking mode (Liu et al., 2014). Overall, cortical and subcortical inputs may engage in hippocampal-dependent functions, such as cognition and affective behaviors through differential recruitment of distinct types of DG INs.

SuM input differentially regulate inhibitory circuits
Although INs primarily innervate principal neurons, a growing body of evidence shows that DG INs connect and inhibit each other (Wang and Buzsáki, 1996;Bartos et al., 2007;Liu et al., 2014). Here, we show that the SuM input robustly recruits HIPP, TML, MOPP, and HICAP cells in the DG. These types of D-INs, especially HIPP and HICAP, are known to form synaptic connections with fast-spiking basket cells (BCs) (12.8% connectivity at HIPP-BC synapses and 16.3% connectivity at HICAP-BC synapses) and effectively inhibit spike generation and reduce spike jitters in BCs (Acsády et al., 2000;Savanthrapadian et al., 2014). Therefore, their direct or indirect activation could cause somatic disinhibition in GCs and result in increased GC excitability. The DG ensembles are highly sensitive to the change of contextual cues (Danielson et al., 2016;Pignatelli et al., 2019). Somatostatinexpressing cells, including HIPP and TML cells, control the size of memory ensembles (Stefanelli et al., 2016). Therefore, activation of HIPP cells by the SuM input could regulate the size and specificity of memory engram.
GABAergic INs are believed to generate and maintain hippocampal theta activity (Freund and Buzsáki, 1996;Fricker and Miles, 2001;McBain and Fisahn, 2001;Freund, 2003;Ito et al., 2018). Given that the SuM plays an essential role in the generation and regulation of hippocampal theta activity, it would be interesting to determine the process by which D-INs are selectively recruited by SuM neurons in vivo. It will be more physiologically relevant to determine the process by which target cell-specific cotransmission of glutamate and GABA at the SuM-DG synapses contributes to brain computation in different behavioral states. The high excitation/low inhibition (E . I) at the SuM-D-IN synapses can promote dendritic inhibition, whereas the low excitation/high inhibition (E , I) at the SuM-GC synapses may help maintain minimal excitatory drive to GCs on one hand, and ensure high spiking precision on the other hand. The differential cotransmission of these two contrasting neurotransmitters at these two synapses may be crucial to the sparsity of GC activation, which plays a central role in pattern separation.
Correct representation of sensory information relies on the precise temporal firing of neurons (Reich et al., 1997;Kara et al., 2000;Reinagel and Reid, 2002). Here, we demonstrated that SuM-mediated glutamate-GABA cotransmission promotes spike-timing fidelity and reduces action potential latency in GCs. This could be essential for ensuring the temporal precision of cognition and fidelity in separating barrage of sensory information into distinct outputs, as described in pattern separation. Moreover, the interaction among coincident inputs gives rise to associative plasticity and long-term regulation of information flow. Consistent with this view, pairing the SuM input with the PP enhances the responses of GCs to cortical inputs, and also promotes long-lasting increase in the excitability of GCs. During LTP induction (Fig. 8C), spikes are reliably generated in GCs. After the LTP induction, the PP-GC synapse is strengthened, and there is a long-lasting increase in the excitability of GCs. In addition to synaptic summation, the observed net enhancement of GCs activity could be explained by IN network functions as illustrated in our proposed models (Fig. 9). Given that fast-spiking BCs in the DG provide powerful inhibition onto GCs, suppression of their activities increases the response of GCs to the cortical input . Notably, dendritic inhibition driven by HIPP cells can reduce spike generation in BCs (Savanthrapadian et al., 2014). Our study showed that activation of the SuM input reliably excites HIPP and TML cells, which could suppress BCs activities, leading to somatic disinhibition of GCs and enhanced spike generation.