INTRODUCTION
Social memory, which helps humans and animals to remember the same conspecifics, is a fundamental cognitive function for survival and reproduction
1,2. Social memory deficits are often associated with various neuropsychiatric and neurodegenerative disorders, e.g., autism, prosopagnosia and Alzheimer's disease
3,4. Revealing the underlying neural mechanisms responsible for the formation and maintenance of social memory is important for the prevention and treatment of these disorders.
Current models suggest that dorsal CA2 (dCA2) and ventral CA1 (vCA1) in the hippocampal subcircuits constitute key hubs for social memory processing: dCA2 orchestrates social memory formation, consolidation, and recall
5-9, while the vCA1–nucleus accumbens shell pathway mediates social memory storage
10. Extrahippocampal inputs to dCA2 from the medial septum
11, lateral entorhinal cortex (LEC)
12 and supramammillary nucleus (SuM)
13 further regulate these processes. Notably, the hypothalamic SuM acts as a novelty detector, routing social novelty signals to dCA2
13. While silencing of SuM–dCA2 projections spares social memory recall
13, their selective activation during rapid-eye-movement (REM) sleep is required for social memory consolidation
14.
The hippocampal formation is classically segregated into four subregions — dentate gyrus (DG), CA3, CA2, and CA1 — each specializing in distinct roles of episodic memory processing
15. Emerging evidence has revealed that the fasciola cinereum (FC), a previously overlooked hippocampal subregion, is a critical hub involved in seizure propagation
16 and the mediation of visual contextual memory acquisition
17. FC exhibits a unique cytoarchitectural structure, and neurons in FC show a unique genetic profile, morphology, connectome architecture, and electrophysiological properties, which distinguish FC from adjacent hippocampal subfields
17,18. Intriguingly, FC receives inputs from social behavior-related regions including the LEC and SuM
16-18. Thus, we hypothesize that the SuM would route social novelty signals to FC and support social memory function within this region.
In this study, we demonstrate that the hippocampal FC, receiving monosynaptic inputs from the hypothalamic SuM, serves as a previously unrecognized critical hub for social memory. Social experience drives input-specific potentiation at SuM–FC excitatory synapses and enhances postsynaptic excitability in FC neurons, thereby converting social novelty signals into enduring memory traces.
RESULTS
SuM neurons provide monosynaptic inputs to FC neurons
FC has been reported to receive inputs from LEC, SuM and perirhinal cortex
17. Among them, SuM was suggested to be a novelty detector and route novelty signals to CA2 and DG
13. To investigate the distribution pattern of SuM axon terminals in FC, we labeled the SuM neurons by injecting an adeno-associated virus (AAV) that drove the expression of enhanced green fluorescent protein (eGFP) into the SuM (Fig. 1a inset). Robust signals of eGFP in the cell bodies of SuM neurons and the axonal terminals within DG, CA2 and FC regions were observed (Fig. 1b, c). The axonal terminals from the SuM were mainly distributed around the cell bodies of FC neurons (Fig. 1c), revealed by immunohistochemical staining of PCP4, a marker for FC neurons
16.
To compare the spatial distribution of FC-projecting neurons with CA2- and DG-projecting neurons within SuM, we unilaterally injected retrograde tracers into FC (cholera toxin subunit B conjugated with Alexa 488 [CTB488]), CA2 (CTB555) and DG (CTB647; Fig. 1d). Serial coronal sections demonstrated that CTB-labeled neurons were mainly located in the lateral SuM (Fig. 1e, Supplementary Fig. S1a–c). The number of FC-projecting neurons was 74 ± 5 on the ipsilateral side, and 43 ± 6 on the contralateral side (Supplementary Fig. S1d,
n = 9 mice). Notably, FC-projecting neurons exhibited minimal overlap with those innervating DG (Fig. 1f left; 0.3% ± 0.2%) or CA2 (Fig. 1f right; 1% ± 0.3%). Although tracer diffusion to adjacent hippocampal subfields (CA1/CA3) may occur, our anterograde data (Fig. 1b) reveal a distinct projection pattern: the SuM projection to the distalmost CA1 is weak, while its projection to CA2 is more robust and typical than that to CA3
19,20. Consequently, our retrograde labeling strategy primarily captures the SuM–FC and SuM–CA2 pathways. Together, these retrograde and anterograde tracing results suggest that FC receives inputs from SuM, and FC-projecting SuM neurons form a unique subpopulation that is distinct from CA2- or DG-projecting populations.
To determine the functional connections of SuM–FC projections, we expressed channelrhodopsin in SuM–FC projecting neurons by injecting AAV-Syn-ChR2-mCherry into the SuM and performed electrophysiological recordings in FC neurons in brain slices while activating ChR2 with blue light pulses (Fig. 1g). Post-hoc histology confirmed viral expression in SuM–FC terminals (Fig. 1h). Furthermore, biocytin labeling of the recorded neurons (Fig. 1i) revealed a typical morphological feature of FC neurons: rounded somata and an absence of basal dendrites, which distinguished them from adjacent CA1 pyramidal cells
17,18. Light stimulation of SuM–FC terminals induced both excitatory postsynaptic currents (EPSCs) and inhibitory postsynaptic currents (IPSCs) in FC neurons (Fig. 1j). We observed that 63% of FC neurons (
n = 95 of 151 neurons) generated EPSC responses following light stimulation, while 42% of neurons (
n = 63 of 151 neurons) produced both EPSCs and IPSCs (Fig. 1k). The response latency of IPSCs was significantly longer than that of EPSCs (Fig. 1l). All light-evoked EPSCs were fully blocked following the application of sodium channel blocker tetrodotoxin (TTX) and recovered by application of potassium channel blocker 4-aminopyridine (4-AP) and TTX (Fig. 1m–o,
n = 27 neurons). The AMPA receptor antagonist, DNQX, and NMDA receptor antagonist, APV, completely abolished light-evoked EPSCs, confirming that SuM–FC excitatory transmission is mediated by glutamatergic synapses (Fig. 1p, q,
n = 17 neurons). However, only ~55% of the light-evoked IPSCs recovered following the application of TTX and 4-AP (Fig. 1r–t,
n = 11 of 20 neurons), suggesting a coexistence of monosynaptic and polysynaptic inhibitory connections. Furthermore, co-application of GABA
A receptor and GABA
B receptor antagonists SR95531 and CGP55845 completely blocked the light-evoked IPSCs, suggesting GABAergic release in SuM–FC projections (Fig. 1u, v,
n = 14 neurons).
SuM–FC projections are required for social memory formation but not recall
The above results suggest that SuM provides monosynaptic connections to FC neurons. Given that these neurons share molecular markers with CA2 pyramidal neurons
16-18, both the SuM–FC pathway and its downstream targets may mediate social memory. To test this hypothesis, we examined social memory performance after silencing each of these two groups of neurons in the learning or recall phase. We expressed the inhibitory opsin eNpHR3.0 within SuM–FC projections by bilaterally injecting AAV-CaMKII-eNpHR3.0-mCherry into the SuM. Fiber ferrules were bilaterally implanted above FC to deliver inhibition light (wavelength: 594 nm) to the axon terminals of SuM neurons (Fig. 2a; Supplementary Fig. S2a, b). Viral expression and fiber locations were confirmed by post-hoc histology (Fig. 2b). Optrode recordings confirmed a significant silencing effect on spike firing by light activation of eNpHR3.0 (Fig. 2c, d).
First, we conducted the 5-trial social memory task to investigate the effect of silencing the SuM–FC axonal projections on social learning
5. The mouse was allowed to freely interact with an unfamiliar intruder A in 4 consecutive trials. The inhibition light was delivered in the first 3 trials. And in the last trial, the subject mouse was introduced to another unfamiliar intruder B without light delivery (Fig. 2e). We observed that social interaction time in the mCherry group (control) decreased gradually throughout the social learning process with intruder A and increased to the level of the 1
st trial when exposed to a novel intruder B in the 5
th trial (Fig. 2f, black,
n = 10). The social interaction time (Fig. 2g, black; RMs 2-way ANOVA test,
F4,36 = 13.9,
P < 0.001) and social discrimination index (Fig. 2h, black; RMs 2-way ANOVA test,
F3,27 = 14.6,
P < 0.001) changed significantly across the task. However, in the eNpHR3.0 group, there was no significant change in the social interaction time (Fig. 2f, g, orange; RMs 2-way ANOVA test,
F4,48 = 3.2,
P = 0.02) or discrimination index (Fig. 2h, orange; RMs 2-way ANOVA test,
F3,36 = 2.8,
P = 0.05) across the task (Fig. 2f, orange,
n = 13). Notably, the social interaction time of the 3
rd trial (eNpHR3.0, 29.1 ± 2.5 s; control, 15.8 ± 2.8;
P = 0.002) or 4
th trial (eNpHR3.0, 27.9 ± 2.5 s; control, 14.1 ± 3.0;
P = 0.002) in the eNpHR3.0 group was significantly greater than that in the control group. The discrimination index of the 3
rd trial (eNpHR3.0, 0.08 ± 0.06; control, 0.37 ± 0.05;
P = 0.002) or 4
th trial (eNpHR3.0, 0.11 ± 0.06; control, 0.46 ± 0.11;
P = 0.007) in the eNpHR3.0 group was significantly lower than that in the control group. In contrast, in the 5
th trial, these two groups did not significantly differ in either the social interaction time or discrimination index. In contrast, the SuM–DG projection was dispensable for social learning (Supplementary Fig. S2e–h), despite its established role in encoding contextual novelty
13.
Then we performed another social memory task, the three-chamber social task
5,21, to verify this result. The mouse was allowed to freely explore a chamber with an unfamiliar mouse under a cup versus a chamber with an empty cup for 10 min. 20 min after training, the subject mouse was put back into the apparatus and allowed for a 10-min free exploration in the chamber with the familiar mouse versus the chamber with a novel mouse. The inhibition light was delivered only in the training phase (Fig. 2i). The subject mice in both control and eNpHR3.0 groups showed significant preference for the chamber with the mouse during training (examples in Fig. 2j top, and summary in Fig. 2k; control: 105.4 ± 11.6 s versus 46.4 ± 4.8 s,
P < 0.001, paired
t-test,
n = 10 mice; eNpHR3.0: 123.0 ± 17.2 s versus 43.7 ± 7.4 s,
P = 0.01, paired
t-test,
n = 6 mice). There was no significant difference in social discrimination index between the control group and the eNpHR3.0 group during training (Fig. 2m left,
P = 0.36, unpaired
t-test). During testing, the eNpHR3.0 group spent more time interacting with the familiar mouse than the control group (Fig. 2l, control: 47.7 ± 7.1 s, eNpHR3.0: 93.9 ± 12.2 s,
P = 0.003, unpaired
t-test). Likewise, the social discrimination index in the eNpHR3.0 group was significantly lower than that in the control group (Fig. 2m, right,
P < 0.001, unpaired
t-test).
Next, we tested the effects of silencing SuM–FC projections on social memory recall. We used the 5-trial social task as described above, except that the inhibition light was delivered only in the 4th trial (Fig. 2n). We found that social interaction time in both the control group and the eNpHR3.0 group exhibited a significant decrease in the training process and increased in the last trial while interacting with a novel intruder (Fig. 2o, p; control: n = 7, eNpHR3.0: n = 10). We observed no significant difference in either the social interaction time (Fig. 2p, P = 0.72) or discrimination index (Fig. 2q, P= 0.69) in the light delivery trial, i.e. the 4th trial. For the light inhibition during testing phase in the three-chamber social test, we also found no significant difference in either the social interaction time (Fig. 2u, P = 0.23) or discrimination index (Fig. 2v, P = 0.36). The locomotion speed was not affected by optogenetic silencing (Supplementary Fig. S2i).
Finally, we tested the effects of silencing SuM−FC projections on contextual fear memory and object recognition memory. We conducted contextual fear conditioning
22 with optogenetic silencing during either the training phase or memory recall phase. On day 1, mice underwent a contextual fear learning paradigm and were tested in the same context 24 h later. No significant differences in freezing levels were observed between the control and silencing groups during training or recall (Supplementary Fig. S3). We next performed an object recognition task
23,24 with optogenetic silencing during the training phase. Mice were first allowed to freely explore an arena containing two identical novel objects for 10 min. After a 10-min interval, they were reintroduced to the arena for a 10-min test session, where one of the familiar objects was replaced with a novel object (Supplementary Fig. S4a, b). We observed no significant differences in the interaction time or discrimination index between the control and silencing groups (Supplementary Fig. S4c, e). Taken together, these results suggest that silencing of SuM–FC axonal projections impairs social memory formation but does not affect social memory recall.
FCSuM recipient neurons are required for both social memory formation and recall
Next, we investigated the contribution of FC
SuM recipient neurons in social memory. To specifically label FC
SuM recipient neurons, anterograde trans-synaptic AAV2/1-Cre
25 was injected into SuM to deliver Cre recombinase transsynaptically to postsynaptic neurons in FC. Concurrently, AAV-DIO-eNpHR3.0 was injected into FC to achieve Cre-dependent expression of eNpHR3.0 in FC
SuM recipient neurons. Fiber ferrules were bilaterally implanted above FC to deliver inhibition light to FC
SuM recipient neurons (Fig. 3a; Supplementary Fig. S2c, d). Virus expression and fiber locations were confirmed by post-hoc histology (Fig. 3b). Optrode recordings were utilized to confirm the silencing effect on spike firing by light activation of eNpHR3.0 in FC (Fig. 3c, d).
First, we conducted the 5-trial social memory task to investigate the effect of silencing FCSuM recipient neurons on social learning, with inhibition light in the first 3 trials. We observed that the social interaction time (Fig. 3f, g, orange; RMs 2-way ANOVA test, F4,32 = 2.7, P = 0.05) or social discrimination index (Fig. 3h, orange; RMs 2-way ANOVA test, F3,24 = 1.9, P = 0.2) showed no significant change. Moreover, the social interaction time of the 3rd trial (eNpHR3.0, 24.3 ± 2.5 s; control, 8.8 ± 1.9 s; P = 0.006) and 4th trial (eNpHR3.0, 23.7 ± 3.4 s; control, 5.5 ± 0.9; P = 0.001) in the eNpHR3.0 group was significantly longer than that in the control group. And the discrimination index of the 3rd trial (eNpHR3.0, 0.12 ± 0.05; control, 0.39 ± 0.05; P < 0.001) or 4th trial (eNpHR3.0, 0.15 ± 0.06; control, 0.52 ± 0.04; P < 0.001) in the eNpHR3.0 group was significantly lower than that in the control group. Next, during the three-chamber social task with light inhibition in the training phase (Fig. 3i), we found a significantly longer time of interaction with the familiar mouse in the eNpHR3.0 group than that in the control group (Fig. 3j, l; control: 51.4 ± 7.8 s, eNpHR3.0: 80.0 ± 6.8 s, P = 0.03, unpaired t-test). Likewise, the social discrimination index in the eNpHR3.0 group was significantly lower than in the control group (Fig. 3m, P < 0.001, unpaired t-test).
Next, we measured the social memory after silencing FCSuM recipient neurons during social memory recall. In the 5-trial social task, the social time in the 4th trial, i.e. the inhibition trial, was significantly longer in the eNpHR3.0 mice than in the control mice (Fig. 3o, p; eNpHR3.0, 15.2 ± 1.6 s; control, 8.5 ± 1.1 s; P = 0.008). The discrimination index of eNpHR3.0 was also lower than that of control group (Fig. 3q; eNpHR3.0, 0.17 ± 0.04, control, 0.49 ± 0.04, P < 0.001). In addition, when inhibition light was given in the recall phase of the three-chamber social test, the social time with the familiar mouse was much longer in the eNpHR3.0 mice than in the control mice (Fig. 3s, u, P = 0.009). And the social discrimination index in the eNpHR3.0 group was much lower than in the control group (Fig. 3v, P = 0.02). Thus, optogenetic silencing of FCSuM recipient neurons impaired both the formation and recall of social memory.
We then examined the effect of activating FCSuM recipient neurons on social memory using the 5-trial social memory task (Supplementary Fig. S5a), with photostimulation applied either during the first three trials (training activation group) or during the fourth trial (testing activation group). We observed that activation of FCSuM recipient neurons during training did not affect social learning (3rd trial: training activation, 11.5 ± 2.2 s; control, 13.4 ± 2.7 s; P > 0.05). However, after the activation light was removed in the 4th trial, the social interaction time in the training activation group was significantly longer than that in the control group (Supplementary Fig. S5b, c; training activation, 23.3 ± 3.9 s; control, 9.1 ± 2.1 s; P < 0.01). Accordingly, the discrimination index in the training activation group was significantly lower than that in the control group in the 4th trial (Supplementary Fig. S5d; training activation, 0.05 ± 0.07; control, 0.47 ± 0.07; P < 0.001). In the testing activation group, we also observed a significantly longer interaction time in the 4th trial compared to the control (Supplementary Fig. S5b, c; yellow, 24.1 ± 2.5 s; P < 0.01). The discrimination index in the testing activation group was also significantly lower than that in the control group in the 4th trial (Supplementary Fig. S5d; 0.05 ± 0.05, P < 0.001). Together, these results indicate that optogenetic activation of FCSuM recipient neurons impairs the recall of social memory.
However, neither optogenetic silencing of FC
SuM recipient neurons (Supplementary Fig. S6a–h) nor chemogenetic silencing of FC Amigo2-expressing neurons (Amigo2 is a marker for FC and CA2 in the hippocampus
18; Supplementary Fig. S6i–l) impaired contextual fear memory. In addition, chemogenetic silencing of FC Amigo2-expressing neurons had no effect on object recognition memory (Supplementary Fig. S7).
SuM–FC projection acts as a novelty detector, driving experience-dependent plasticity in downstream FC neurons
SuM neurons have been reported to transmit social and spatial novelty signals to the hippocampus
13. Here, we investigated whether the SuM
FC projecting neurons could route social novelty signal to FC neurons, and the specific neuronal dynamics underlying this process. We utilized the optrode recordings to investigate neuronal spike activity at the single-cell level across the social learning process. We labeled the SuM
FC projecting neurons with channelrhodopsin-2 (ChR2) by injecting a Cre-dependent retrograde AAV (retroAAV-Cre) into FC, and an AAV vector carrying DIO-ChR2-mCherry into SuM (Fig. 4a). We then implanted an optrode into SuM to deliver the stimulation light and record the activity from single neurons (Fig. 4b; see tetrode locations in Supplementary Fig. S8a). We identified the neurons as SuM
FC projecting if they fired spikes reliably to light stimuli (Fig. 4e, f; latency, median/25
th–75
th percentiles: 3.12/2.03–4.58 ms; success rate > 87%,
n = 13 neurons).
We recorded the spike activity of SuM
FC projecting neurons during the 5-trial social learning task (Fig. 4g). For the spike activity of SuM
FC projecting neurons, we observed a significant decrease in the firing rate of these neurons during social interaction, as the subject mice became familiarized to the strangers (Fig. 4k, orange,
n = 13 neurons from 10 mice; RMs 1-way ANOVA test with LSD post-hoc comparison). In the 3
rd trial, the firing rate decreased to 72 ± 5% of the 1
st trial (
P < 0.001), and in the 4
th trial decreased to 63 ± 7% (
P < 0.001). In the last trial interaction with a novel mouse, the firing rate increased to a similar level as in the 1
st trial (94 ± 6%,
P = 0.3). Additionally, the normalized firing rate of SuM
FC projecting neurons showed a positive correlation with the social interaction time (Fig. 4l,
n = 13 neurons from 10 mice, Pearson correlation test,
P = 0.001,
r = 0.39). In addition to the rate of single spike firing, we analyzed burst firing and theta modulation in these neurons during the task
26. While the rate of burst firing remained unchanged, we observed a significant decrease in theta modulation during interactions with a familiar conspecific compared to a novel one (Supplementary Fig. S8c–f).
Next, we tested the spike activity of the downstream FC neurons during social learning. We conducted the same optrode recording approach described above to record the activity of FCSuM recipient neurons. To express ChR2 in FCSuM recipient neurons, we injected AAV2/1-Cre into SuM and AAV-DIO-ChR2-mCherry into FC (Fig. 4c). After one month of virus expression, an optrode was implanted into FC for identification and recording (Fig. 4d; Supplementary Fig. S8b). The units that responded reliably to light stimulation were identified as FCSuM recipient neurons (Fig. 4h, i; latency, median/25th-75th percentiles: 2.30/1.56–2.95 ms; success rate > 85%, n = 20 neurons).
We recorded 20 FCSuM recipient neurons from 10 mice across the 5-trial social learning task (Fig. 4j). Mice exhibited a significant decrease in interaction with intruder A after the 4-trial training (Fig. 4m, black, n = 14 trials from 10 mice; RMs 1-way ANOVA test with LSD post-hoc comparison, trial 1 versus trial 4, P < 0.001). For the activity of FCSuM recipient neurons, we observed a significant increase in the firing rate of these neurons during social learning (Fig. 4m, purple; n = 16 of 20 neurons), which was different from the SuMFC projecting neurons. In the 4th trial, the firing rate increased 11.8 ± 0.2-fold relative to the 1st trial (P = 0.004). In the last trial, following interaction with a novel mouse, the firing rate decreased to a similar level as the 1st trial (98 ± 8%, P = 0.8). The normalized firing rate of FCSuM recipient neurons showed a negative correlation with the social interaction time (Fig. 4n, n = 16 neurons from 10 mice, Pearson correlation test, P = 0.007, r = ‒0.30). In contrast, the firing rate of both SuMFC projecting and FCSuM recipient neurons remained unchanged in the object recognition test (Supplementary Fig. S9). These results suggest that SuM–FC projections serve as a social novelty detector, while the downstream FC neurons act as a social familiarity detector.
To determine whether social learning alters the intrinsic properties of FCSuM recipient neurons, we labeled these cells by injecting AAV2/1-Cre into the SuM and Cre-dependent AAV-DIO-EGFP into FC (Fig. 5a, b). Whole-cell recordings in brain slices revealed that FCSuM recipient neurons in trained mice generated more action potentials (APs) with the same current injection intensity (Fig. 5c). The firing rates for nearly all current steps (Fig. 5d) and the resting membrane potential (Fig. 5e, P = 0.03) were significantly increased, while the rheobase (Fig. 5f, P = 0.005) decreased after social learning. These results indicate that the intrinsic excitability in FCSuM recipient neurons significantly increases after social learning.
Plasticity at SuM–FC synapses and its requirement for social memory
To investigate whether social learning induces synaptic plasticity at SuM–FC synapses, we performed in vitro whole-cell recordings in FC neurons in brain slices after mice had undergone the social learning task. Spontaneous EPSCs showed significant increases in both frequency and amplitude following social learning (Fig. 6a–c). Similarly, miniature EPSCs also exhibited elevated amplitude and frequency (Fig. 6d–f) in the presence of TTX. However, the frequency and amplitude of spontaneous or miniature IPSCs remained unchanged between naive and trained mice (Fig. 6g–l).
Furthermore, we examined the SuM–FC synaptic connection by expressing ChR2 within SuM–FC projections and conducting
in vitro whole-cell recordings in FC neurons. We compared light-evoked synaptic responses before and after social learning and found a significant increase in the amplitude of light-evoked EPSCs (Fig. 6m, n) without alterations in response latency (Fig. 6o). The light-evoked IPSCs and paired-pulse ratio of EPSCs or IPSCs remained unchanged (Fig. 6p–u). We also measured the changes in α-amino-3-hydroxy-5-methylisoxazole-4-propionate (AMPA) and N-methyl-D-aspartate (NMDA) currents. The AMPA/NMDA ratio in FC neurons was significantly enhanced following social learning (Fig. 6v, w), suggestive of an increase in postsynaptic AMPA receptor insertion
11,27,28. Together, these results indicate that social learning significantly strengthens SuM–FC excitatory transmission, while inhibitory synapses remain unaltered.
If enhanced synaptic strength at SuM–FC excitatory synapses is indeed required for social memory formation, then suppressing this potentiation should impair social memory. Therefore, we tested if long-term depression (LTD) at SuM–FC synapses impairs social memory. First, we confirmed that low-frequency (1 Hz for 5 min) optogenetic stimulation
27,29 of SuM inputs induced LTD at synapses in FC neurons using
in vitro whole-cell recordings (Fig. 7a–d,
P = 0.004). Then we applied the same stimulation protocol
in vivo after social learning in the 5-trial social learning task (LTD group; Fig. 7e, f). We observed that the social time in the 4
th trial was significantly longer in the LTD mice than in the control mice (Fig. 7g, h: LTD, 22.8 ± 2.2 s; control, 9.1 ± 0.9 s; Supplementary Fig. S10c: 5-Hz control, 7.5 ± 1.5 s;
P < 0.001). Likewise, the discrimination index of the LTD group was lower than that of the control group (Fig. 7i: LTD, 0.08 ± 0.06; control, 0.57 ± 0.05; Supplementary Fig. S10d: 5-Hz control, 0.56 ± 0.06;
P < 0.001). For the three-chamber test (Fig. 7j), LTD mice spent more time interacting with the familiar mouse (Fig. 7k, l,
P = 0.02), and the social discrimination index in the LTD group was significantly lower than that of the control group (Fig. 7m,
P = 0.007). Thus, LTD at SuM–FC synapses after social learning significantly impairs the expression of social memory. This behavioral effect of offline LTD after learning does not contrast with that of SuM–FC silencing during memory recall (Fig. 2n–v), as these two manipulations target distinct neural processes: the former directly erodes established synaptic plasticity in FC neurons, while the latter disrupts ongoing SuM input signals to FC.
To assess the possible effect of synaptic potentiation at the SuM–FC synapse on social memory, we employed a direct social memory test in mice following long-term potentiation (LTP) induction protocol
27,29. Experimental mice, which had been injected with AAV-ChR2 in SuM and implanted with optical fibers above the bilateral FC, were first allowed to explore a cage previously inhabited by a novel mouse (Mouse A) for 3 min. During this exploration, they received blue light stimulation delivered in four cycles (20 s on / 20 s off). The stimulation consisted of 20 Hz light pulses with a 5 ms pulse width
27. Five minutes after this familiarization phase, the social memory test was conducted. During the test, the subject mouse sequentially interacted with the Mouse A and a novel mouse (Mouse B), each for 1 min, with a 5-min interval between sessions (Supplementary Fig. S10e). We found that the LTP group mice spent significantly less time in social interaction with Mouse A than the control group (Supplementary Fig. S10f, g; LTP, 22.8 ± 4.2 s; control, 38.7 ± 5.1 s;
P = 0.04). In contrast, both groups spent a similar time in social interaction with Mouse B (Supplementary Fig. S10g; LTP, 40.5 ± 5.5 s; control, 37.2 ± 3.7 s;
P = 0.6). And the discrimination index in LTP group increased significantly as compared to the control group (Supplementary Fig. S10h; LTP, 0.29 ± 0.05; control, ‒0.01 ± 0.03;
P < 0.001). Thus, LTP at SuM–FC synapses promotes the formation of social memory.
We mapped the inputs and outputs of FC using retrograde and anterograde viral tracing in Amigo2-Cre mice. To identify inputs, we injected a mixture of AAV-DIO-RVG and AAV-DIO-EGFP-TVA into the FC, followed by RV-ENVA-ΔG-DsRed into the same site three weeks later (Supplementary Fig. S11a, b). This approach labeled presynaptic neurons in the medial septum (MS), hippocampal subfields CA1, CA2, and CA3C, SuM, and the perirhinal cortex (PRh) (Supplementary Fig. S11c–e). To map outputs, we injected AAV-DIO-EGFP into the FC (Supplementary Fig. S11f, g) and found GFP-positive axonal terminals in the dorsal lateral septum (dLS), MS, CA2, dorsal CA1 (dCA1), and dorsal subiculum (dSub) (Supplementary Fig. S11h–j).
To further determine the effect of silencing FCSuM recipient neurons on the activity of CA2 neurons, we performed Ca2+ recording in CA2 together with optogenetic inhibition of FCSuM recipient neurons. We anterogradely labeled FCSuM recipient neurons with eNpHR3.0 and simultaneously labeled CA2SuM recipient neurons with a Ca2+ indicator GCaMP6f. Optical fibers was implanted bilaterally above FC to enable photoinhibition, and Ca2+ signals were recorded in CA2 (Supplementary Fig. S12a–c). We found that 594-nm light stimulation in FCSuM recipient neurons significantly reduced Ca2+ signals in CA2 compared to the control condition without light stimulation (Supplementary Fig. S12d, e; RMs 1-way ANOVA with LSD post-hoc test, F2,10 = 52.7, P < 0.001, n = 6 mice). These results indicate that inhibition of FCSuM recipient neurons effectively suppresses the activity of CA2 neurons.
DISCUSSION
While the dCA2 and vCA1 have been established as key hippocampal loci for social memory
11-14, our study reveals the hippocampal subregion FC as an additional component in the neural circuit map of social cognition. Multiple lines of evidence here converge to establish the essential role of FC and its SuM inputs for social memory: (1) SuM provides direct inputs to FC, which offers an anatomical basis for routing social novelty signals from the hypothalamus to hippocampal circuits during memory encoding; (2) the increased activity of SuM
FC projecting neurons during social novelty representation indicates the necessity of SuM−FC inputs for memory encoding, consistent with the result that silencing the SuM projections impairs only social memory formation but not recall; (3) the elevated activity in the downstream FC neurons during interaction with a familiar mouse demonstrates the importance of FC for memory storage, in line with the result that silencing FC
SuM recipient neurons impairs both social memory formation and recall; (4) the elevated activity in FC neurons may be attributed to experience-dependent plasticity. This plasticity encompasses potentiation of both intrinsic excitability and glutamatergic transmission, as demonstrated by
in vitro electrophysiology, and its suppression significantly impairs social memory (Fig. 7n).
The results of our current study, along with previous work, suggest that the SuM acts as a detector for social novelty
13,30. Like SuM–CA2 projections, where inhibition impairs only memory formation but spares memory recall
13, SuM–FC circuitry emerges as a parallel conduit for social novelty signals transmitted into the hippocampal formation. Therefore, the current study’s finding of a dynamic withdrawal of SuM–FC activity upon memory formation mirrors the "encoder-to-storage" transition described in SuM-CA2 pathways during social learning
13.
Our
in vitro electrophysiology reveals unique synaptic integration properties in FC neurons receiving input from the SuM. In all recorded FC neurons with IPSCs, EPSCs were consistently present and monosynaptic, whereas IPSCs were either mono- or polysynaptic. The lack of FC neurons receiving pure IPSCs (without EPSCs) indicates that GABAergic input from SuM to FC originates exclusively from glutamate/GABA co-releasing SuM neurons — a distinctive feature shared with the dentate gyrus
13,31-33 but absent in CA2
13,34. This suggests that FC may perform composite computation of hypothalamic signals. Surprisingly, social experience selectively potentiates excitatory transmission, possibly through an increase in postsynaptic AMPAR insertion, but we did not find evidence of specific GABAergic modulation. This dominance of glutamatergic signaling suggests an FC subregion-specific plasticity rule within hippocampal microcircuits. The unchanged inhibition following social learning may serve to maintain temporal precision of SuM–FC signaling and prevent seizure susceptibility, given the previously established role of FC in epilepsy
16.
We note that our observations that "silencing SuM–FC projections impaired memory formation, but not recall" (Fig. 2) and "depression at SuM–FC synapses abolished established social memory" (Fig. 7) are not contradictory. Optogenetic silencing of SuM–FC projections blocks presynaptic neurotransmitter release during the training phase, thereby preventing the encoding of social information without affecting previously stored memories. In contrast, LTD at SuM–FC synapses acts postsynaptically, suppressing synaptic structures that support already formed memories. Thus, the former disrupts the formation of memory by interfering with signal transmission at the time of learning, whereas the latter impairs the expression of existing memories by remodeling established synaptic connections.
Our manipulation of FC neurons in the contextual fear conditioning task (Supplementary Fig. 6) yielded results that differ from the previous study suggesting the importance of FC in visual contextual memory acquisition
17, a discrepancy that may arise from two differences: 1) behavioral paradigms: they used the visual-guide T-maze task requiring complex visual discrimination, whereas we used the contextual fear conditioning task relying on the integration of multimodal environmental cues; 2) loss-of-function strategies: they used colchicine to induce permanent and non-specific cell death, whereas we used AAV-based cell type- and pathway-specific optogenetic and chemogenetic silencing of FC neurons. In addition, chemogenetic inhibition of FC Amigo2
+ neurons (Supplementary Fig. 6l) appears to show a non-significant trend toward enhanced contextual fear. Although this effect did not reach statistical significance and was markedly weaker than the observed social memory impairment, it does not rule out the possibility of FC encoding a more general salience/arousal signal for social memory processing in a specific context.
Accumulating evidence indicates that multiple inputs to CA2 regulate memory formation but not recall. For instance, activation of vasopressin input from the paraventricular nucleus enhances social memory formation
35, and silencing of glutamatergic input from the medial septum reduces social memory formation but not recall
11. Whether the SuM–FC pathway coordinates with other inputs (
e.g., the entorhinal cortex) to integrate multisensory social cues remains unclear. The rapid plasticity we observed immediately after social training raises the question of whether short-term and long-term plasticity involve distinct mechanisms. We speculate that the rapid AMPAR-mediated potentiation represents an early, labile form supporting short-term recognition, whereas overnight consolidation, as we described previously in the SuM–CA2 pathway
14, may engage transcriptional and structural changes for long-term storage. Future investigations into whether SuM–FC projections could mediate complementary memory processes across sleep-wake cycles would be of interest.
Both CA2 and FC receive convergent inputs from key structures implicated in social memory, including the MS, LEC, and SuM
17,36,37. However, their intrahippocampal afferent sources diverge: CA2 receives prominent inputs primarily from CA3a and DG
36, while FC integrates information from dCA1, CA3c, and the perirhinal cortex (PRh) (Supplementary Fig. S11e). This divergence also extends to their intrahippocampal efferent projections: FC projects to dCA1, CA2, DG, and the dorsal subiculum (dSub), whereas CA2 primarily targets CA3, dCA1 and vCA1 (Supplementary Fig. S11j)
18,38. Critically, direct reciprocal projections exist between CA2 and FC
18, establishing a bidirectional communication loop. This connectivity pattern suggests a specialized functional hierarchy for social memory processing. Converging evidence — specifically, the FC–CA2 connectivity, the lack of direct FC to vCA1 projections, and the previously-known CA2 to vCA1 pathway — led us to propose a working model in which social information flows along an FC→CA2→vCA1 pathway. Within this hierarchy, FC receives and processes extrahippocampal and intrahippocampal social information. CA2 then integrates inputs from both FC and other intrahippocampal/extrahippocampal sources before projecting to vCA1 (Supplementary Fig. S11k). Therefore, the FC→CA2→vCA1 pathway suggests that FC may introduce an additional computational layer for social information originating from dCA1, CA3c, and PRh, which do not project to dCA2
39-41. It is likely that CA2 acts as a pivotal intermediary node, receiving this integrated social information from FC and facilitating its transfer to vCA1 for social memory storage.
It is now well-established that the FC represents a unique subfield of the hippocampus. Despite sharing markers like PCP4 with CA2, some differences have been identified between these two regions. First, genes including
Mef2c,
Zfp536,
Notch2,
Fam163b,
Rgma, and
Igfbp4 are expressed at higher levels in FC, while
Cpne7,
Ttr,
Cpne6, and
Scn1a show preferential expression in CA2
17,42. Second, FC neurons are characterized by a pronounced predominance of apical dendrites and a remarkably sparse distribution of basal dendrites, a morphological pattern that distinctly sets them apart from neurons in the CA1, CA2, and CA3 subfields
17,18. Third, we found here that the upstream SuM neurons projecting to FC and CA2 are largely non-overlapping (Fig. 1c). Fourth, in terms of functional characteristics, network activity in FC is dominated by theta rhythm and does not exhibit sharp-wave ripples
17, which is in contrast to CA2 that actively contributes to sharp-wave ripple events
6,43. Fifth, at the cellular level, FC neurons rarely display burst firing
18, further distinguishing them from CA2 neurons that are notably prone to burst discharges
44. Our study revealed that a single social interaction enhances the excitability of FC
SuM recipient neurons, nearly doubling their average firing rate during interaction with a familiar conspecific (Figs. 4m). This suggests that FC neurons encode social experience probably through population-level activity. This role appears complementary to the social coding functions of the CA2 region, where population activities signal social novelty and sparse activity encodes social identity
45-47. Whether this functional profile extends to female mice remains to be determined.
In conclusion, we redefine the hippocampal social memory circuitry by identifying FC as a new node where novelty signals from the hypothalamic SuM region are transformed into enduring memories through experience-dependent plasticity. This mechanism bridges the gap between dynamic social interactions and stable memory traces, providing a neurophysiological framework to investigate disorders of social memory, such as those occurring in autism spectrum disorders and other neuropsychiatric conditions.
MATERIALS AND METHODS
Animals
Adult male C57BL/6J mice aged from 3 to 5 months were used for tracing, recording and behavioral experiments. Amigo2-Cre male mice were used for tracing and behavioral experiments. Mice were kept in groups of 4 to 5, except for those implanted with an optical fiber or optrode, which were housed singly. Mice were provided with free access to food and water, and they were housed under a 12/12-h light/dark cycle (lights on at 7:00 am). All the experimental procedures were carried out in accordance with the institutional animal welfare guidelines and approved by the Animal Care and Use Committee of the Third Military Medical University.
AAV, rabies virus, and CTB
The following AAV constructs were used in this study: AAV2/9-EF1α-EGFP (titer: 1.49 × 1013 viral particles/mL), AAV2/9-Syn-hChR2-mCherry (titer: 1.72 × 1013 viral particles/mL), AAV2/2Retro Plus-Syn-Cre (titer: 1.61 × 1013 viral particles/mL), AAV2/9-DIO-hChR2-mCherry (titer: 3.67 × 1012 viral particles/mL), AAV2/9-CaMKII-eNpHR3.0-mCherry (titer: 1.49 × 1013 viral particles/mL), AAV2/9-Syn-mCherry (titer: 6.22 × 1012 viral particles/mL), AAV2/9-DIO-eNpHR3.0-mCherry (titer: 3.00 × 1012 viral particles/mL), AAV2/9-DIO-mCherry (titer: 2.46 × 1012 viral particles/mL), scAAV2/1-hSyn-Cre (titer: 1.06 × 1013 viral particles/mL), AAV2/5-DIO-EGFP (titer: 3.44 × 1012 viral particles/mL), AAV2/5-Ef1α-DIO-EGFP-2A-VTA (titer: 2.00 × 1012 viral particles/mL), and AAV2/5-Ef1-1α-DIO-RVG (titer: 2.00 × 1012 viral particles/mL). All AAV constructs were purchased from Taitool Bioscience Co., Ltd. (Shanghai, China), BrainVTA Co., Ltd. (Wuhan China) or Obio Biotechnology Co., Ltd. (Shanghai, China). For rabies virus, the RV-ENVA-ΔG-DsRed (titer: 2.00 × 108 viral particles/mL) was purchased from BrainVTA Co., Ltd. (Wuhan China). CTB488 (0.2%, c34775, ThermoFisher), CTB555 (0.2%, c34776, ThermoFisher), and CTB647 (0.2%, c34778, ThermoFisher) were used to identify SuM neurons that projected to FC, CA2 and DG.
Surgical procedures
For all surgeries, mice were anesthetized with isoflurane in oxygen (induced at 3% and maintained at 1–2%) and then placed into a stereotaxic frame with a heating pad (37.5–38 °C) to maintain body temperature. Following surgery, the mice were kept on the heating pad until awake and then returned to the home cage for full recovery. To reduce inflammation, mice were given a single dose of dexamethasone sodium phosphate (1 mg/mL, 0.1 mL/10 g) and ceftriaxone sodium (50 mg/mL, 0.1 mL/10 g) by intraperitoneal injection daily for 3 days after surgery.
For virus injections, an incision was made along the midline of the skull, and a small craniotomy (0.5 × 0.5 mm) above the SuM or FC was performed on 2- to 3-month-old mice. A glass pipette with a tip diameter ranging from 10 to 20 μm was used to deliver the virus to the target brain region. To express EGFP, ChR2, eNpHR3.0, or mCherry in the SuM–FC projection, AAV-EGFP (30–50 nL), AAV-ChR2 (100–200 nL), AAV-eNpHR3.0 (100–200 nL) or AAV-mCherry (100–200 nL) was injected into the SuM (AP: ‒2.8 mm, ML: 0.5 mm, DV: 5.0 mm from dura). To express ChR2 in SuMFC projecting neurons, AAV2/2Retro-Cre was injected into FC (AP: ‒1.8 mm, ML: 0.5 mm, 10° angle toward the midline, DV: 1.75 mm, 100–200 nL), and AAV-DIO-ChR2 was injected into the SuM (100–200 nL). To express ChR2, eNpHR3.0, mCherry, or EGFP in FCSuM recipient neurons, scAAV2/1-Cre was injected into the SuM (100–200 nL), and ~100 nL of AAV-DIO-ChR2, AAV-DIO-eNpHR3.0, AAV-DIO-mCherry, or AAV-DIO-EGFP was injected into FC. For optrode recording and histological tracing experiments, the viruses were injected unilaterally, while for behavioral experiments the viruses were injected bilaterally. The viruses were allowed to express for at least one month prior to subsequent experiments.
For optrode recording surgery, mice expressing ChR2 in the SuMFC projecting or FCSuM recipient neurons were used. A custom-made optrode was inserted above SuM (AP ‒2.8 mm, ML 0.5 mm) to a depth of 4.7 mm, or above FC (AP: ‒1.8 mm, ML: 0.5 mm, 10° angle toward the midline) to a depth of 1.3 mm. The optrode was then secured to the skull using C&B Super-Bond (BearDayton). Common dental cement and super glue were used for further reinforcement. After a recovery of 4–5 days, the optrode was gradually advanced to the target depth.
For fiber ferrule implantation in optical silencing experiments, mice expressing eNpHR3.0-mCherry or mCherry in bilateral SuM–FC projections, or FCSuM recipient neurons were used. Optical fiber ferrules (200 μm diameter, NA 0.37) were implanted with the tips placed above the bilateral FC (AP: ‒1.8 mm, ML: 0.5 mm, 10° angle toward the midline, DV: 1.7 mm). The cannula was then secured to the skull with dental cement and a thin layer of black acrylic paint was applied evenly to the outermost layer to avoid light leakage. Mice were allowed to recover for at least 5 days prior to behavioral experiments. After behavioral experiments the locations of the fiber tips were verified by post-hoc histology.
Optrode recording
The custom-made optrode consisted of four tetrodes (California Fine Wire, CFW2002936), a 200 μm-diameter fiber (NA 0.37) attached with a laser diode (Osrma, PL450B, PL520B), and a microdrive were fabricated as previous studies
48. Electrophysiological signals were recorded during social interaction at a sample rate of 20 KHz on a RHD2000 USB board (Intan Technology, C3100). An electrical lesion was made by passing a 100 μA-current (duration, 10 s) through each tetrode to verify the recording location after recordings.
To identify SuMFC projecting or FCSuM recipient neurons, 200 optogenetic excitation light pulses (intensity, 10 mW; duration, 10 ms; interval 0.5 s) were delivered through the optical fiber to circuit-specific ChR2-expression mice. Units produced by light excitation with high response reliability (> 85% for all the units in our data), short spike latency (< 5 ms), and low response jitter (< 1.6 ms) were identified as ChR2-positive neurons.
Behavioral task procedure and analysis
All mice were handled for 5 min each day for 3 consecutive days prior to behavioral tasks. All behavior apparatuses were cleaned by 75% ethanol before and between trials to eliminate confounding olfactory factors. During the task, behavioral videos for all mice were recorded with an infrared camera at 25 Hz with a spatial resolution of 1920 × 1080 (Cannon, XA-25). The location of the test mouse was automatically tracked with STPoseNet video tracking tools
49.
Five-trial social memory test
The five trial social memory test was performed in a clean cage of the same size as the home cage (30 × 20 × 20 cm). The test mice were put into the cage for 5 min of free exploration. Subsequently, a novel male mouse aged between 8 and 10 weeks was introduced as intruder A in trial 1 to 4 for one min. On the 5th trial, another novel male intruder B (of the same age and housed together with A), was placed into the cage for one min of free interaction. The social interaction time that the test mice spent sniffing, allogrooming, and close following was reviewed by experiment-blinded analysis of the videos. Social preference indexes across the task were calculated using the equation:
Discrimination indextrial n = [(social time in trial 1) ‒ (social time in trial n)] / [(social time in trial 1) + (social time in trial n)], where n = 2, 3, or 4.
Three-chamber social novelty test
The three-chamber social novelty test was used as an additional social memory test and conducted as previously described
5,14. The three-chamber apparatus was made from an opaque square open-topped box (75 cm × 50 cm × 40 cm), which was divided into 3 equal-sized chambers by transparent plastic walls. Two 12-cm-diameter transparent plexiglass cylinders with 1-cm holes in the wall were put in the two end-chambers. The behavioral task involved three stages: habitation, training, and testing. For habitation, the test mouse was allowed to freely explore in the arena for 10 min. A novel male stimulus mouse was then put in either one of the cylinders, with the other cylinder empty. The test mouse was gently placed into middle chamber and allowed to freely explore for 10 min. The testing trial was carried out 0.5 h after training. Another novel stimulus mouse was put into the other cylinder, and the locations of the two cylinders were randomly switched. The test mouse was returned to the middle chamber for another 10-min free exploration. Social interaction was defined as the nose of the test mouse being within 2 cm from the outside of the cylinder. A social preference index was calculated using the equation:
Discrimination index = [(time for novel mouse) ‒ (time for familiar mouse)] / [(time for novel mouse) + (time for familiar mouse)].
Contextual fear conditioning
The mice were subjected to the contextual fear conditioning task as the final behavioral assessment to test contextual memory in accordance with a standard protocol
22. In the training session, a 260-s baseline was recorded, followed by three deliveries of the unconditioned stimulus (US; 2-s, 0.75-mA foot shock) at intervals of 180 s. Then, 24 h after the training, the mice were reintroduced into the training box for a 10-minute recording. The freezing level was determined based on the body movement in a time bin of 20 s
50. The freezing data was further grouped for statistical analysis.
Optogenetic silencing
For optogenetic silencing experiments, continuous orange (594 nm, Changchun New Industries) light was delivered to the bilateral FC area to inhibit either SuM terminals or FC neurons. The light intensity was adjusted to ~10 mW at the tip of each fiber ferrule. Light delivery was triggered by a transistor-transistor logic signal from a self-designed program written with the LabVIEW platform.
In vitro electrophysiological recording
Mice injected with AAV-ChR2-mCherry, or AAV2/1-Cre in SuM and AAV-DIO-EGFP in FC were deeply anesthetized and transcardially perfused with ice-cold, oxygenated (95% O2 and 5% CO2) cutting solution containing (in mM): 110 N-methyl-d-glucamine, 2.5 KCl, 1.2 NaH2PO4, 25 NaHCO3, 25 glucose, 10 MgSO4 and 0.5 CaCl2, adjusted to pH 7.3 with 6 M HCl, 310 mOsm. Brains were rapidly removed, and coronal slices (270 μm thick) containing the FC region were prepared using a vibratome (Leica, VT1200S). Slices were incubated in the oxygenated cutting solution at 35 °C for 15 min, and then in oxygenated artificial cerebrospinal fluid (ACSF in mM: 119 NaCl, 2.5 KCl, 1.2 NaH2PO3, 25 NaHCO3, 2 MgSO4, 2 CaCl2, 12.5 glucose, adjusted to pH 7.3) at 35 °C for 30 min. Then slices were transferred to a recording chamber and continuously perfused by oxygenated ACSF throughout the entire recording process.
Whole-cell voltage-clamp recordings were performed in FC neurons. Visualization of FC neurons was achieved using an upright microscope (BX51WI, Olympus) equipped with infrared differential interference contrast (DIC) optics, enabling visualization through transmitted light or epifluorescence. The recordings were conducted with an EPC-10 amplifier (HEKA Elektronik). Recording micropipettes were fabricated from borosilicate glass capillaries (BF 150-86-10, Sutter Instruments) pulled using a micropipette puller (P-97, Sutter Instruments) with resistances ranging from 4 to 6 MΩ. Excitation of ChR2 expressed in SuM axons was achieved by light from a 473-nm laser (MBL-III-473, Changchun New Industries, 5 ms duration, 10 mW at fiber tip, ~10 mW/mm
2 at slice surface) delivered through an optical fiber
27,29. All signals were digitized at a rate of 20 kHz and low-pass-filtered at 2 kHz using the amplifier circuitry. Whole-cell recordings were excluded if series resistance exceeded 20 MΩ. Additionally, we added biocytin (4 mg/mL) to the intrapipette solution to allow morphological identification of the recorded neurons.
For postsynaptic current recordings, the micropipette was filled with an intrapipette solution containing (in mM): 130 cesium methane sulfonate, 5 KCl, 10 HEPES, 2 NaCl, 5 Mg-ATP, 0.4 Na2GTP, 5 EGTA, and 10 disodium phosphocreatine, with a pH of 7.3 and an osmolarity of 300 mOsm. EPSCs and IPSCs were recorded at ‒70 mV and +10 mV respectively. TTX (1 μM, Sigma) was applied to prevent action potential generation, and 4-AP (100 μM, Sigma) was added to block KV1 potassium channels. NBQX (10 μM, Sigma) and APV (50 μM, Sigma) were bath-applied to block AMPA and NMDA receptors, respectively. SR95531 (1 μM, Sigma) and CGP55845 (2 μM, Sigma) were bath-applied to block GABAA and GABAB receptors. To isolate AMPA receptor- and NMDA receptor-mediated currents, we bath-applied 1 μM SR95531 while recording under two distinct voltage-clamp conditions: AMPAR currents held at ‒70 mV and NMDAR currents held at +40 mV. Paired-Pulse Ratio was recorded with two pulses delivered with a 100 ms interval.
Current-clamp recordings were performed to assess the intrinsic excitability of FCSuM recipient neurons. The micropipette was filled with an intrapipette solution containing (in mM): 125 K-gluconate, 20 KCl, 10 HEPES, 4 ATP, 1 EGTA, and 2 MgCl2, with a pH of 7.3 and an osmolarity of 300 mOsm. Cells were subjected to a series of 0.5-s current steps ranging from ‒100 pA to +200 pA (20 pA increments) while monitoring voltage responses. Key intrinsic properties were characterized, including rheobase, resting membrane potential (RMP), input resistance, and action potential (AP) threshold. The rheobase was the minimum injected current required to elicit the first AP. Resting membrane potential (RMP) was measured immediately after membrane rupture (0 pA injecting current). Input resistance was calculated as the slope of the linear fit to the voltage-current (I-V) curve generated by current injections ranging from ‒100 pA to ‒20 pA. AP threshold was defined as the membrane potential where dV/dt reached 5% of the maximal AP upstroke slope.
We probed the light-evoked EPSCs with a 0.1 Hz, 5 ms, 470 nm light pulse delivered through the fiber for 20 trials before LTD induction. For LTD induction, low-frequency stimulation (1 Hz, 5 ms, 300 pulses) was applied. Then we probed light-evoked EPSCs for 20 min (0.1 Hz, 5 ms) after stimulation.
Histology
All tested mice were perfused with 4% paraformaldehyde (PFA) in PBS, placed in 15% sucrose in PBS for 24 h, and then the brains were sectioned into 50 μm slices using a cryostat (NX50, ThermoFisher). Then the brain slices were mounted onto glass slides and imaged with a confocal (Zeiss, LSM 700) or wide field (Olympus, BX51) microscope after staining with DAPI or other antibodies. For PCP4 immunohistochemical staining, we processed the brains of mice injected with AAV2/9-EF1α-EGFP in the SuM. The following primary antibodies were used: rabbit anti-PCP4 (1:500; Sigma, HPA005792), chicken anti-GFP (1:200; Abcam, ab13970).
Statistical analyses
The raw extracellular electrophysiological data were preprocessed as described previously
48 to extract the spikes. Briefly, all the events that exceeded an amplitude threshold of four standard deviations above the background were saved for subsequent spike sorting analysis. All detected events for each tetrode were sorted in the toolbox MClust based on the features of waveforms
28. The burst properties of SuM
FC projecting neurons were analyzed based on interspike intervals (ISIs). Bursts were defined as events with ISIs between 3 and 10 ms, and their burstiness was quantified by a burstscore (number of spikes within bursts / total spikes). Theta modulation of SuM
FC projecting neurons was assessed by calculating theta coherence
26. Spike-field coherence was first determined for each unit during social behavior. To ensure comparability across units, we randomly selected an equal number of spikes per unit (matching the minimum spike count) and averaged 300-ms epochs of raw SuM LFP surrounding each selected spike. The power spectral density of this averaged signal was normalized to its mean power to derive a coherence value (0–100%). Theta coherence was defined as the maximum coherence value within the 6–10 Hz band.
Statistical tests were performed in MATLAB and SPSS software (Supplementary Table S1). Data were first tested for normality and equal variance between groups. When both were achieved, parametric tests were used. When normality test or equal variance test failed, non-parametric tests were used. For parametric tests, paired and unpaired t-tests, RMs 1-way ANOVA with LSD post-hoc comparison, 1-way ANOVA with LSD post-hoc comparison and RMs 2-way ANOVA with Sidak’s post-hoc comparison were used. For non-parametric tests, Wilcoxon signed-rank test or Wilcoxon rank-sum test were used. All tests were two-tailed. For data presentation, boxplots were used where the center line represents the median, the box represents the 25–75% interquartile range (IQR), and the whiskers extend to the most extreme data points (not considering outliers). For statistical testing, all data points were included without exclusion.
DATA AVAILABILITY
All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data and materials in this study are available from the corresponding author upon request.
The Author(s) 2026. Published by Higher Education Press. This is an Open Access article distributed under the terms of the CC BY license (https://creativecommons.org/licenses/by/4.0/).
