The neural mechanisms supporting the rise and fall of maternal aggression Authors: Takashi Yamaguchi, Rongzhen Yan, Mashrur Khan, Sota Kuno, Kanishk Tewatia, Takuya Osakada, Srinivas Parthasarathy, Michael E Pacold, Nirao M Shah, Dayu Lin Published: 2026 ABSTRACT AbstractMaternal aggression enables lactating females to protect their vulnerable young1,2, yet its rapid emergence after birth and swift decline when pups are absent remain poorly understood. Our study reveals the critical role of the pathway from posterior amygdala cells expressing oestrogen receptor alpha (PAEsr1) to the ventrolateral part of ventromedial hypothalamus cells expressing neuropeptide Y receptor 2 (VMHvlNpy2r) in the rise and fall of maternal aggression. Projection-specific manipulations and recordings show that PAEsr1 cells projecting to the VMHvl are naturally active during attack and are required for maternal aggression. During lactation, PA-to-VMHvlNpy2r synapses potentiate and VMHvlNpy2r cell excitability increases, enabling heightened aggression. PAEsr1 neurons express abundant oxytocin receptors, allowing oxytocin to boost PA output; after pup removal, declining oxytocin levels reduce PA drive and dampen maternal aggression, a deficit restored by pup reunion or optogenetic elevation of oxytocin. These findings reveal multiple forms of plasticity in a defined PAEsr1–VMHvlNpy2r circuit that collectively implement the adaptive, need-based control of maternal aggression.Subject terms: Social behaviour, Neural circuits, Synaptic plasticity Main After birth, young mammals are vulnerable and require protection. To ensure offspring survival, mothers of many species vigorously attack intruders, a behaviour that is known as maternal aggression1,2. As its name implies, maternal aggression is largely confined to the maternal period. In mice and rats, non-lactating females rarely attack conspecific intruders; aggression rises sharply 1 day after parturition, peaks during the first postpartum week, then gradually declines and disappears by the end of lactation3–5. This pronounced behavioural transition is thought to reflect changes in aggression circuitry, and—given its autonomous nature—pregnancy- and lactation-associated hormones have long been considered key contributors6. In the late 1970s, it was shown that oestrogen and progesterone regimens mimicking pregnancy, together with nipple development and pup suckling, could induce aggression in virgin female mice, whereas sex hormones alone were ineffective6,7. Preventing pup suckling through thelectomy (nipple removal) similarly blocked maternal aggression in mouse mothers8. These findings indicate that pregnancy-related hormonal changes, combined with postpartum nursing, are required to initiate maternal aggression. However, how female aggression circuits are reorganized during pregnancy and how pup suckling promotes maternal aggression remain unclear nearly five decades later. Esr1-expressing cells in the VMHvl (VMHvlEsr1) are essential for both male aggression and female aggression9–12. In female mice, VMHvlEsr1 cells comprise two functionally and anatomically distinct subpopulations: medially biased Npy2r-expressing cells (VMHvlNpy2r) are essential for female aggression13, whereas laterally biased cholecystokinin A receptor (encoded by Cckar) cells (VMHvlCckar cells), which are largely Npy2r-negative but Esr1-positive (VMHvlNpy2r(−)Esr1(+)), mediate female sexual receptivity13–15. During lactation, VMHvlNpy2r cells, but not VMHvlNpy2r(−)Esr1(+) or VMHvlCckar cells, exhibit elevated in vivo responses to intruders compared with virgin female mice13,15, although the cellular or synaptic mechanisms underlying this enhancement remain unclear. Beyond the VMHvl, the female aggression circuit is largely uncharacterized. We previously found that PAEsr1 cells, which are predominantly glutamatergic, provide strong excitatory input to the VMHvl and drive male aggression16. Inhibiting PAEsr1 cells projecting to the VMHvl (PAEsr1→VMHvl) impairs male aggression, whereas activating them promotes attack16,17. Moreover, the PAEsr1–VMHvl pathway is plastic: repeated victories potentiate PAEsr1-to-VMHvl connections and ultimately enhance aggression18,19. Given the critical role of the PAEsr1–VMHvl pathway in male aggression and experience-dependent plasticity, we examined whether this circuit also contributes to maternal aggression and whether circuit plasticity supports the marked escalation of aggression during lactation. We found that the PAEsr1–VMHvl circuit is essential for maternal aggression, but operates through mechanisms distinct from those in male mice. Specifically, oxytocin-dependent modulation enables rapid, need-based adjustment of maternal aggression to protect offspring. Maternal aggression needs PAEsr1→VMHvl cells To determine whether PAEsr1→VMHvl cells are necessary for maternal aggression, we virally expressed hM4Di in PAEsr1→VMHvl cells by injecting a retrograde HSV expressing Cre-dependent Flpo into the VMHvl and AAV-Ef1α-fDIO-hM4Di–mCherry (control: mCherry) into the PA of virgin C57BL/6N (C57) Esr1-2A-cre × Swiss–Webster (SW) F1 hybrid female mice (Fig. 1a). We used C57 × SW F1 mice because they display more robust and reliable maternal aggression compared with C57 mice, while maintaining genetic uniformity (Supplementary Note 1). Histological analysis confirmed that the vast majority of hM4Di–mCherry-expressing cells in the PA were ESR1 positive (96.81 ± 0.03%, n = 3 mice; Fig. 1b). hM4Di–mCherry-positive cells comprised around 10% of all PA cells (Extended Data Fig. 1c), similar to the percentage of PA cells that express Esr1 and retrogradely labelled with the retrograde tracer, cholera toxin B subunit (CTB), injected into the VMHvl, suggesting that most VMHvl-projecting PAEsr1 cells were likely targeted (Extended Data Fig. 1a–c). The PAEsr1–VMHvl pathway bidirectionally modulates maternal aggression. a, Experimental schematics and test schedule. b, Histological images showing hM4Di expression and ESR1 staining in the PA. Bottom, magnified views of the areas indicated by boxes. c,d, Attack latency increased (c) and attack duration decreased (d) after CNO injection in hM4Di mice, but not mCherry mice. For mice showing no attack, the latency was set to 600 s. e, The juvenile investigation duration was unchanged after CNO or saline injections. f, Experimental schematic and representative images showing ChR2–eYFP (green) and ESR1 (red) expression in the PA and VMHvl. Blue, DAPI. Brain illustrations were adapted from the Allen Brain Reference Atlas (https://atlas.brain-map.org). g, Experimental timeline. h,m, Behaviours toward juvenile intruders with or without light stimulation in virgin (h) and lactating (m) female mice. i,n, The total attack duration of ChR2 and GFP virgin (i) and lactating (n) female mice. j–l, There was no change in attack probability (j), latency to attack from light onset (k) or attack duration per trial (l) during the light-on versus light-off periods in both ChR2 and GFP virgin female mice. o–q, In lactating ChR2 mice, but not GFP mice, the attack probability increased (o), the latency to attack decreased (p) and the average attack duration per trial increased (q). Data are mean ± s.e.m. The numbers in parentheses represent the number of animals. Statistical significance was determined using mixed-effects analysis with repeated measures followed by Bonferroni’s multiple-comparison test (c–e, j–l and o–q), Mann–Whitney U-tests (i) and unpaired t-tests (n). All P values <0.05 are indicated. All tests were two-tailed. Detailed statistics are provided in Supplementary Table 1. Scale bars, 200 μm (b (top) and f) and 20 μm (b (bottom)). Source data Comparison of the retrograde labelling efficiency of HSV versus CTB. a, (Left) Experimental schematics and representative images showing Esr1 expression (green) and Alexa Fluor 555 Conjugated CTB (red) injected into the VMHvl. 3 V, third ventricle. (Right) Schematics showing the viral strategy to label PA cells projecting to the VMHvl. Brain illustrations are adapted from the Allen Brain Reference Atlas (https://atlas.brain-map.org). b, Representative images showing Esr1 expression (green) and CTB retrogradely labelled from the VMHvl (red) in the PA. Panels (b2) and (b3) show enlarged views of the boxed area in (b1). c, Percentage of CTB and Esr1 double-positive, hM4Di-mCherry-expressing, and GCaMP6f-expressing cells among all PA cells, based on DAPI staining. Bars and error bars represent mean ± s.e.m., and numbers in parentheses indicate the number of animals. Statistics: Ordinary one-way ANOVA followed by Tukey’s multiple-comparison test. All tests are two-sided. All p values > 0.05. See Supplementary Table 1 for more detailed statistics. Source data After virus injection, each female was paired with a male until visibly pregnant. Maternal aggression was then assessed between postpartum day 3 (PPD3) and PPD5, 30 min after intraperitoneal (i.p.) injection of saline or clozapine N-oxide (CNO) on separate, counterbalanced days (Fig. 1a). After saline injection, all of the female mice attacked the juveniles rapidly and repeatedly (Fig. 1c,d). By contrast, after CNO injection, all hM4Di female mice significantly decreased attacks during the 10-min test, whereas all of the mCherry controls remained highly aggressive (Fig. 1c,d). The decreased aggression was not due to reduced social interest, as investigation duration was unaffected by PAEsr1→VMHvl cell inactivation. (Fig. 1e). In a separate experiment, inhibiting PAEsr1 cells broadly also significantly reduced maternal aggression (Extended Data Fig. 2a–d). By contrast, maternal care was unaffected by inhibition of either PAEsr1→VMHvl or total PAEsr1 populations in mothers (Extended Data Fig. 2e,m,n). Chemogenetic inhibition of PAEsr1 and PAEsr1→VMHvl cells impairs maternal aggression without affecting maternal care or sexual receptivity. a, Experimental schematic and timeline to examine behavioural changes after inhibiting PAEsr1 cells. Brain illustrations are adapted from the Allen Brain Reference Atlas (https://atlas.brain-map.org). b-d, Attack latency (b), attack duration (c), and time spent investigating juvenile intruders (d) after saline or CNO injections in lactating females that expressed hM4Di in PAEsr1 cells. e, The latency to retrieve the first pup and all 5 pups after saline or CNO injections in lactating females expressing hM4Di in PAEsr1 cells. f, g, Male investigation duration (f) and lordosis quotient (g) after saline or CNO injections in virgin oestrus females expressing hM4Di in PAEsr1 cells. h-l, Latency to shallow thrust (h), duration of shallow thrust (i), latency to deep thrust (j), duration of deep thrust (k), and deep thrust success rate (l) when sexually experienced males mated with oestrus females expressing hM4Di in PAEsr1 cells and receiving either CNO or saline injections. m, Experimental schematic or timeline to examine changes in maternal and sexual behaviours after inhibiting VMHvl projecting PAEsr1 cells. n, Latency to retrieve the first pup and all 5 pups after saline or CNO injection in lactating females expressing hM4Di in PAEsr1→VMHvl cells. The latency was set to 600 s if no retrieval occurred. o, p, Male investigation duration (o) and lordosis quotient (p) after saline or CNO injection in virgin oestrus females expressing hM4Di in PAEsr1→VMHvl cells.q-u, Latency to shallow thrust (q), shallow thrust duration (r), latency to deep thrust (s), deep thrust duration (t), and deep thrust success rate (u) when sexually experienced males mated with oestrus females expressing hM4Di in PAEsr1→VMHvl cells, receiving either CNO or saline injection. All bars and error bars represent mean ± s.e.m. Numbers in parentheses indicate the number of animals. Statistical significance was determined using (b-d,e: 1st pup, h, i k) Wilcoxon matched-pairs signed-rank test, (e: all pup, f, g, j, l) paired-t-test and (n-u) Mixed-effects analysis with repeated measures followed by Bonferroni’s multiple-comparison test. All tests are two-sided. All p values < 0.05 are indicated. See Supplementary Table 1 for more detailed statistics. Source data As VMHvl is also critical for female sexual behaviours20, we also examined whether PAEsr1→VMHvl cells are necessary for female sexual receptivity, assessed by rejection probability and lordosis15,21. Female sexual receptivity did not differ between saline- and CNO-treated days (Extended Data Fig. 2o,p). Male mice achieved intromission with comparable latencies and durations under both conditions (Extended Data Fig. 2q–u). Inhibiting PAEsr1 cells yielded similar results, indicating their dispensable role in female sexual behaviours (Extended Data Fig. 2f–l). Thus, PAEsr1→VMHvl cells are specifically required for maternal aggression, but not other female social behaviours. PAEsr1–VMHvl drives maternal aggression To test whether the PAEsr1→VMHvl projection is sufficient to drive attack, we bilaterally injected AAV-Ef1α-FLEX-ChR2-eYFP into the PA and implanted optic fibres above the VMHvl in C57 Esr1-2A-cre × SW F1 hybrid female mice (Fig. 1f). Then, 3 weeks later, we first tested stimulation-induced behaviours in dioestrous virgin females by delivering light pulses (470 nm, 20 Hz, 20 ms, 1.0–5.0 mW) for 60 s every 2–3 min after introducing a juvenile intruder (Fig. 1g). None of the dioestrous females showed time-locked attacks during PAEsr1-to-VMHvl activation, although ChR2 animals attacked more than GFP animals overall during the test session (Fig. 1h–l). After the aggression test, the female mice were placed into the home cage of a sexually experienced male mouse; light stimulation did not increase female sexual receptivity (Extended Data Fig. 3a,b). Both GFP and ChR2 female mice consistently rejected male mice regardless of stimulation, and intromission never occurred (Extended Data Fig. 3b–d). Optogenetic activation of PAEsr1 to VMHvl projection does not alter female sexual receptivity. a, Experimental schematic and timeline. Brain illustrations are adapted from the Allen Brain Reference Atlas (https://atlas.brain-map.org). b, Representative raster plots showing behaviours of virgin dioestrus females expressing ChR2 in PAEsr1 cells during interaction with a male intruder when blue light was delivered to the VMHvl. Top: light-on period (cyan); bottom: behaviours. c, d, Time females spent rejecting the male (c), and duration of male mounting (d) during each light-on and light-off period. All bars and error bars represent mean ± s.e.m. Numbers in parentheses indicate the number of animals. Statistical significance was determined using (c, d) Mixed-effects analysis with repeated measures followed by Bonferroni’s multiple-comparison test. All tests are two-sided. All p values < 0.05 are indicated. See Supplementary Table 1 for more detailed statistics. Source data Afterwards, the test females were paired with males and terminal stimulation-evoked attack was re-examined during PPD3–PPD7, when maternal aggression is naturally expressed. After light stimulation, ChR2 animals, but not GFP animals, had a significantly increased probability of attacking the juvenile intruder (Fig. 1m–q). Light-on trials were associated with a reduced attack latency and an increased attack duration compared with the light-off trials (Fig. 1p,q). Together, these results indicate that PAEsr1 cells provide a critical excitatory input to the VMHvl for driving female aggression, with greater efficacy during lactation than in the virgin state. VMHvlNpy2r plasticity during lactation To address why PA inputs to the VMHvl elicit time-locked attacks in mothers but not in virgin females, we investigated VMHvl cell responses to PA inputs using in vitro patch-clamp recordings (Fig. 2a). We focused on VMHvlNpy2r cells given their established role in maternal aggression13. Indeed, direct optogenetic activation of VMHvlNpy2r cells reliably induced attacks in virgin females, suggesting their deterministic role in initiating female aggression (Extended Data Fig. 4). We next performed current-clamp recordings of VMHvlNpy2r cells and compared their spiking probability during PA-VMHvl terminal stimulation in virgin versus lactating females (PPD3–PPD5) (Fig. 2b). While PA terminal activation (1 ms, 20 Hz, 473 nm light pulses) induced action potentials in less than 15% of VMHvlNpy2r cells in virgin dioestrous or oestrous female mice, nearly half (11 out of 23) spiked in lactating females (Fig. 2b,c). This enhanced spiking could reflect changes in synaptic strength or intrinsic excitability. Changes in VMHvlNpy2r responses to intruders and physiology properties in lactating females. a, Experimental schematic, histology and test timeline. b, Representative oEPSP and action potential (AP) traces evoked by 1-ms light pulses (cyan) in dioestrous (black), oestrous (blue) and lactating (red) female mice. c, The percentage of VMHvlNpy2r cells exhibiting light-induced spiking. d, Example oEPSC and oIPSC traces evoked by 1-ms light pulses. e, The distribution of light-evoked synaptic response types. f,g, The latency (f) and amplitude (g) of oEPSCs and oIPSCs. h, Representative paired oEPSCs evoked at 100-ms intervals in dioestrous and lactating female mice. i, Paired-pulse ratios (PPRs). j, Experimental schematic, test schedule and representative histology for VMHvlNpy2r cell recordings. Brain illustrations were adapted from the Allen Brain Reference Atlas (https://atlas.brain-map.org). k, Example spontaneous EPSC (sEPSC) and IPSC (sIPSC) traces across reproductive states. l–n, The sEPSC and sIPSC amplitude (l) and frequency (m) and the excitatory/inhibitory ratio (n). o, Representative current-clamp responses to 50 and 150 pA injections. p–t, F–I curves (p), the maximum number of APs (q), input resistance (r), RMP (s) and rheobase (t) of VMHvlNpy2r cells. Data are mean ± s.e.m. The numbers in parentheses denote recorded cells. Statistical analysis was performed using Fisher’s exact tests or χ2 tests with Benjamini–Hochberg FDR correction (c), χ2 tests with FDR correction (e), mixed-effects analysis with Bonferroni correction (f, g and p), two-way repeated-measures ANOVA with Bonferroni correction (i) and Kruskal–Wallis test with Dunn’s multiple-comparison correction (l–n and q–t). Cells in c and e–g were from 5 dioestrous, 3 oestrous and 6 lactating female mice; in i, from 5 dioestrous and 6 lactating female mice; and in l–n and p–t, from 5 dioestrous, 4 oestrous and 6 lactating female mice. All P and q values <0.05 are indicated except for in p; *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.001. All tests were two-tailed. Detailed statistics are provided in Supplementary Table 1. Scale bars, 200 μm (a and j). Source data Optogenetic activation of VMHvlNpy2r cells induces time-locked attack in virgin females. a, Experimental schematic and representative images showing ChR2 (green) and Fos (red) expression in VMHvl. Right panel: enlarged view of the boxed area. Brain illustrations are adapted from the Allen Brain Reference Atlas (https://atlas.brain-map.org). b, Experimental timeline. c, Representative raster plots showing behaviours of virgin females expressing ChR2 in VMHvlNpy2r cells during interaction with a juvenile intruder, with blue or sham light delivered to VMHvl. Top: sham (black) or light (cyan) delivery periods; bottom: behaviours. d-f, Percentage of trials in which animals attacked (d), attack latency per trial (e), and average attack duration per trial (f) toward juvenile intruders during light-on and sham trials. Bars and error bars represent mean ± s.e.m.; numbers in parentheses indicate the number of animals. Statistical significance was determined using Wilcoxon matched-pairs signed-rank test. All tests are two-sided. All p values < 0.05 are indicated. See Supplementary Table 1 for more detailed statistics. Source data To examine the former, we performed voltage-clamp recordings of VMHvlNpy2r cells in virgins and mothers (PPD3–PPD5) while optogenetically activating PA terminals to evoke excitatory postsynaptic currents (oEPSCs) and inhibitory postsynaptic currents (oIPSCs) (Fig. 2d,e). As in male mice16, the oEPSC latency was short (mean ± s.e.m., 3.49 ± 0.38 ms), indicating a monosynaptic connection (Fig. 2f), whereas the oIPSC latency was longer (mean ± s.e.m., 12.47 ± 1.31 ms), consistent with a polysynaptic connection (Fig. 2f). During dioestrus, 65% (13 out of 20) of cells exhibited oEPSCs, whereas only 12% (2 out of 16) did so during oestrus (Fig. 2e). The oEPSC amplitude was also reduced in oestrous female mice compared with in dioestrous and lactating female mice (Fig. 2g). This difference is unlikely presynaptic, as paired-pulse ratios were comparable across reproductive states (Fig. 2h,i). The probability of evoking oIPSCs showed a similar reproductive-state dependence, consistent with their secondary nature (Fig. 2e,g). Moreover, for spontaneous excitatory postsynaptic currents (sEPSCs), but not for spontaneous inhibitory postsynaptic currents (sIPSCs), the frequency of VMHvlNpy2r cells was the lowest during oestrus (Fig. 2j–n). Together, these findings indicate that PA–VMHvlNpy2r connectivity is strongest during lactation and weakest during oestrus. Next, we examined the intrinsic properties of VMHvlNpy2r cells using current-clamp recordings. Although rheobase, resting membrane potential (RMP) and input resistance did not differ across reproductive states, VMHvlNpy2r cells in lactating females fired more spikes across current steps and reached a higher maximal firing rate compared with in dioestrous females (Fig. 2o–t). VMHvlNpy2r cells in oestrous females exhibited intermediate excitability (Fig. 2o–q). Collectively, these results suggest that enhanced PA-driven spiking of VMHvlNpy2r cells during lactation arises from both strengthened synaptic input and increased intrinsic excitability. The increased VMHvlNpy2r cell responses are also evident in vivo. As previously reported13, Ca2+ responses of VMHvlNpy2r cells to juvenile intruders increased significantly during lactation (Extended Data Fig. 5). In lactating female mice, the peak Ca2+ response of VMHvlNpy2r cells during intruder introduction was around threefold of that in virgin females (Extended Data Fig. 5c,d,g). During social investigation, VMHvlNpy2r cell activity also consistently increased, although the response magnitude did not differ between virgin and lactating females (Extended Data Fig. 5e,g). Females reliably attacked the intruder only during lactation (Extended Data Fig. 5b). During attack, VMHvlNpy2r cells exhibited reliable, moderate activity increases (Extended Data Fig. 5f,g). Between dioestrus and oestrus, VMHvlNpy2r cell Ca2+ responses to intruders were comparable, consistent with low aggression in virgin female mice regardless of oestrous stage11 (Extended Data Fig. 5b-g). Increased VMHvlNpy2r responses to juvenile intruders in lactating females. a, Experimental schematic, test schedule, and representative image showing GCaMP6f (green) expression in VMHvlNpy2r cells. Blue, Nissl. Brain illustrations are adapted from the Allen Brain Reference Atlas (https://atlas.brain-map.org). b, Attack duration of females toward juvenile intruders. c, Representative Ca2+ traces of VMHvlNpy2r cells during interaction with a juvenile intruder in dioestrus (c1), oestrus (c2), and lactating (c3) females. Coloured shading indicates behavioural episodes. d-f, Peri-event time histograms of the Ca2+ signal of VMHvlNpy2r cells aligned to intruder introduction (d), investigation (e) and attack onset (f) across dioestrus, oestrus or lactating females. g, Peak Ca2+ signal of VMHvlNpy2r cells during intruder introduction, investigation, and attack. All bars and error bars and lines and shades represent mean ± s.e.m. Numbers in parentheses indicate the number of animals (b, d-g). Statistical significance was determined using (b) Friedman test with Dunn’s multiple comparisons and (g) One-way ANOVA with repeated measures followed by Tukey’s multiple-comparisons test. All tests are two-sided. All p values < 0.05 are indicated. See Supplementary Table 1 for more detailed statistics. Source data Limited PA-to-VMHvlNpy2r(−)Esr1(+) input PAEsr1 cells send dense fibres to both aggression-related VMHvlm and mating-related VMHvll (Extended Data Fig. 6a,b). However, we observed no sexual behaviour change after either inhibiting PAEsr1→VMHvl cells or activating PAEsr1-to-VMHvl projection (Extended Data Figs. 2p–u and 3). To understand this dissociation, we investigated the connection from PA to VMHvlNpy2r(−)Esr1(+) cells, a population known to bidirectionally control female sexual receptivity13,15. Specifically, we injected AAV-hSyn-Coff/Fon-mCherry into the VMHvl of Npy2r-ires-cre × Esr1-2A-Flpo mice to label VMHvlNpy2r(−)Esr1(+) cells and AAV-hSyn-Chronos-GFP into the PA to enable optogenetic control of PA-to-VMHvl terminals in brain slices (Extended Data Fig. 6c). Histological analysis confirmed that the vast majority of mCherry-expressing cells were Npy2r negative and Esr1 positive (94.9 ± 1.6% cells from 3 animals; Extended Data Fig. 6c,d). Notably, none of the 18 recorded VMHvlNpy2r(−)Esr1(+) cells in oestrous females fired a single spike after PA terminal stimulation, and only 3 out of 15 (20%) cells showed light-evoked EPSCs despite strong Chronos expression in the PA (Extended Data Fig. 6e–h,j,k,m–o). We focused on oestrous females because PA input would be most likely to influence mating-related VMHvl output during oestrus, when these cells are maximally excitable15. Consistent with this, PA input exerted even less influence in sexually inactive lactating females, with only 3 out of 26 (12%) VMHvlNpy2r(−)Esr1(+) cells showing light-evoked EPSCs (Extended Data Fig. 6k,m–o). This contrasts sharply with the 79% of aggression-related VMHvlNpy2r cells that exhibited oEPSCs during PA terminal stimulation in lactating females (Fig. 2e). Accordingly, VMHvlNpy2r(−)Esr1(+) cells exhibited lower spontaneous sEPSC frequencies during both oestrus and lactation compared with VMHvlNpy2r cells (Fig. 2k,m and Extended Data Fig. 6i,l,p,q). PA to VMHvlNpy2r-Esr1+ cell connectivity in virgin and lactating females. a, Representative images showing terminals from PAEsr1 cells expressing ChR2-EYFP in the medial (VMHvlm) and lateral (VMHvll) subdivisions of the VMHvl in dioestrus and lactating females. b, Normalized PAEsr1 terminal density in VMHvlm and VMHvll of dioestrus and lactating females. Fraw: average pixel intensity in fibre-containing subregions; Fbackground: dorsal medial VMH lacking terminals; Fsignal = Fraw – Fbackground. Fsignal was normalized to the maximum per animal and used to calculate average terminal field density across animals. c, Experimental schematic for optogenetic-assisted circuit mapping. Top: Chronos-GFP expression in PA. Bottom: mCherry (red), Npy2r (green), and Esr1 (cyan) mRNA in VMHvl and their overlap. d, Percentage of Npy2r+ and Npy2r-Esr1+ cells expressing mCherry. Bars and error bars represent mean ± s.e.m. Numbers in parentheses indicate the number of animals. e, Experimental timeline. f, Example voltage clamp recording trace of a VMHvlNpy2r-Esr1+ cell in response to 5-ms blue-light pulses (blue vertical lines). g, The percentage of VMHvlNpy2r-Esr1+ cells that showed PA terminal stimulation-evoked spiking in oestrus females. (n = 18 cells from 3 females).h, i, Example voltage-clamp recording traces of VMHvlNpy2r-Esr1+ cells in oestrus females following 5-ms blue-light pulses (cyan) (h) and spontaneous activity (i). j, Pie charts showing light-evoked synaptic response patterns of VMHvlNpy2r-Esr1+ cells in oestrus (n = 3) females. k, l, Example voltage-clamp recording traces of VMHvlNpy2r-Esr1+ cells in lactating females following 5-ms blue-light pulses (cyan) (k) and spontaneous activity (l). m, Pie charts showing light-evoked synaptic response patterns of VMHvlNpy2r-Esr1+ cells in lactating (n = 6) females. n, o, Amplitude (n) and latency (o) of oEPSCs and oIPSCs of VMHvlNpy2r-Esr1+ cells in oestrus (n = 3) and lactating (n = 6) females. p, q, Amplitude (p) and frequency (q) of sEPSCs and sIPSCs of VMHvlNpy2r-Esr1+ cells in oestrus (n = 3) and lactating (n = 6) females. Only 6/15 recorded cells in oestrus females showed sEPSC. All bars and error bars represent mean ± s.e.m. Numbers in parentheses indicate the number of subject animals (d) or recorded cells (g, n-q). Statistical significance was determined using (b, n, o) Mixed-effects analysis with repeated measure followed by Bonferroni’s multiple-comparison test, (d) unpaired t-test, and (p, q) Mann-Whiteney test. All tests are two-sided. All p values < 0.05 are indicated. See Supplementary Table 1 for more detailed statistics. Source data These results suggest that, first, PA input preferentially targets VMHvlNpy2r cells over VMHvlNpy2r(−)Esr1(+) cells, especially during lactation; second, PA inputs could not drive VMHvlNpy2r(−)Esr1(+) cells to spike even during oestrus, likely explaining the lack of female sexual behaviour change during PAEsr1→VMHvl manipulation. Higher PAEsr1→VMHvl response in mothers We next examined whether PA cells themselves show heightened responses to aggression-related cues during lactation, thereby contributing to elevated maternal aggression. We recorded PAEsr1→VMHvl cell Ca2+ activity using fibre photometry (Fig. 3a). Specifically, we injected HSV-hEf1α-LS1L-Flpo into the VMHvl and AAV-Ef1α-fDIO-GCaMP6s into the PA of C57 Esr1-2A-cre × SW F1 hybrid female mice (Fig. 3a). Histology confirmed that GCaMP6s expression was largely restricted to ESR1+ cells (92.6% cells, n = 2 mice; Fig. 3b). As the PA GCaMP6 signal tends to shift over time, probably due to high expression, we recorded PAEsr1→VMHvl cell activity in separate virgin and lactating cohorts with similar virus incubation periods (Fig. 3c). PAEsr1→VMHvl cells showed their strongest activation during intruder introduction, with maximal responses during lactation (Fig. 3e–g). During social investigation, the cells responded moderately, with a trend of higher responses in lactating females (Fig. 3e,h). Moreover, PAEsr1→VMHvl cells consistently increased activity during attack, which occurred exclusively in mothers (Fig. 3d–f,i). Changes in PAEsr1→VMHvl responses to intruders and physiology properties in lactating female mice. a, Experimental schematic. b, Representative images showing GCaMP6s (green) and ESR1 (red) expression in the PA. Bottom, magnified views. c, Experimental timeline. d, The mean attack duration toward juvenile intruders. e, Peak Ca2+ responses of PAEsr1→VMHvl cells during intruder introduction, investigation and attack. f, Representative Ca2+ traces during interactions with juvenile intruders in dioestrous (left), oestrous (middle) and lactating (right) female mice. The shaded regions indicate behavioural epochs. g–i, Peri-event time histograms of the Ca2+ signals aligned to intruder introduction (g), investigation (h) and attack (i). j, Experimental schematic and representative images showing ESR1 expression (green) and VMHvl-injected retrobeads (red). Left, magnified view. Brain illustrations were adapted from the Allen Brain Reference Atlas (https://atlas.brain-map.org). k, Experimental timeline. l, Representative voltage-clamp recording traces of PAEsr1→VMHvl cells in dioestrous and lactating female mice. m–o, The sEPSC and sIPSC amplitude (m) and frequency (n) and excitatory/inhibitory (E/I) ratio (o) of PAEsr1→VMHvl cells in dioestrous and lactating female mice. p, Example current-clamp responses to 50 and 150 pA current injections. q–u, F–I curves (q), the maximum number of action potentials (r), rheobase (s), RMP (t) and input resistance (u) of PAEsr1→VMHvl cells in dioestrous (black) and lactating (red) female mice. Data are mean ± s.e.m. The numbers in parentheses denote mice (d, e and g–i) or recorded cells (m–o and q–u). Statistical analysis was performed using Kruskal–Wallis tests with Dunn’s correction (d and e (investigation)), one-way ANOVA with Tukey’s correction (e (introduction)), unpaired t-tests (m (sEPSC), o (amplitude) and r–t), Mann–Whitney U-tests (m (sIPSC), n, o (frequency) and u) and mixed-effects analysis with repeated measures and Bonferroni correction (q). Cells in m–o and q–u were recorded from 4 dioestrous and 4 lactating female mice. All P values <0.05 are indicated. All tests were two-tailed. Detailed statistics are provided in Supplementary Table 1. Scale bars, 200 μm (b (top) and j (middle and right)) and 20 μm (b (bottom) and j (left)). Source data To confirm that enhanced responses were not driven by behavioural differences, we recorded PAEsr1→VMHvl cells in head-fixed awake female mice exposed to social stimuli (Extended Data Fig. 7a). Consistent with free-moving recordings, PAEsr1→VMHvl cells showed significantly higher responses to juveniles, adult males and adult females in lactating females than in dioestrous or oestrous virgins (Extended Data Fig. 7b–e), while minimal responses were observed to objects (Extended Data Fig. 7b–e). These in vivo data indicate that PAEsr1→VMHvl cells also undergo changes during lactation, contributing to increased aggression circuit output. Female PAEsr1→VMHvl cells show enhanced responses to social cues during lactation. a, Schematic of head-fixed recording setup and experimental timeline. b-d, Peri-event time histograms (top) and heatmaps (bottom) of PAEsr1→VMHvl cell Ca2+ signals aligned to the onset of adult male, adult female, juvenile, and object presentations in females during dioestrus (b), oestrus (c), and lactation (d). e, Peak ΔF/F during presentation of various stimuli in females during dioestrus, oestrus, and lactation. All data are presented as the mean ± s.e.m. Numbers in parentheses indicate the number of recording mice. Statistical significance was determined using Mixed-effects analysis with repeated-measure followed by Bonferroni’s multiple-comparison test. All tests are two-sided. All p values < 0.05 are indicated. See Supplementary Table 1 for more detailed statistics. Source data Next, we performed in vitro patch-clamp recordings of PAEsr1→VMHvl cells in brain slices from dioestrous and lactating Esr1-zsGreen female mice to examine the cellular basis of these changes. We injected retrobeads into the VMHvl and recorded retrobead-labelled PA cells 7–10 days later (Fig. 3j,k). Both sEPSCs and sIPSCs of PAEsr1→VMHvl cells increased moderately during lactation (Fig. 3l–n). Specifically, the sEPSC amplitude and sIPSC frequency were significantly elevated, while the sEPSC frequency and sIPSC amplitude showed increasing trends in lactating females (Fig. 3m,n). Despite these changes, the excitation/inhibition ratio remained unchanged (Fig. 3o). Furthermore, in contrast to VMHvlNpy2r cells, frequency–current (F–I) curves revealed no change in PAEsr1→VMHvl cell excitability during lactation compared with in the virgin state (Fig. 3p–r). Rheobase, RMP and input resistance were also unchanged (Fig. 3s–u). These data suggest that PAEsr1→VMHvl and VMHvlNpy2r cells undergo distinct physiological adaptations during lactation. Oxytocin modulates aggression through PA Given a lack of increase in excitability or excitatory/inhibitory ratio of PA cells during lactation, we investigated whether additional mechanisms exist to support increased PAEsr1→VMHvl responses to intruders during lactation. Previous studies22,23 and gene expression database (https://mouse.brain-map.org)24 indicate high oxytocin receptor (Oxtr) expression in the PA. Given the elevated oxytocin levels during lactation and the typical excitatory Gq-mediated effects of OXTR25, we hypothesized that OXTR signalling may facilitate PA activity during lactation. To test this, we first examined the Oxtr expression in the PA using RNAscope-based in situ hybridization26 and found that almost all PAEsr1 cells express Oxtr regardless of the reproductive state (Fig. 4a–c). Notably, Oxtr expression levels in PAEsr1 cells were significantly higher in lactating females, suggesting enhanced oxytocin sensitivity (Fig. 4a,c). We next examined the effect of oxytocin on PAEsr1 cell activity in lactating females using in vitro current-clamp recording (Fig. 4d). Application of TGOT, a selective OXTR agonist, significantly increased input resistance without altering the RMP (Fig. 4e,f). In the presence of TGOT, PAEsr1 cells fired more action potentials in response to 50 or 100 pA current injection (Fig. 4g–i), indicating that oxytocin enhances PAEsr1 input–output gain. OXTR promotes maternal aggression by enhancing PA activity. a, Representative images showing Oxtr (magenta) and Esr1 (green) mRNA in the PA of dioestrous and lactating female mice. Right, magnified views of the areas indicated by boxes. b, The percentage of PAEsr1 cells expressing Oxtr. c, The number of Esr1 and Oxtr mRNA puncta per expressing cell in the PA of dioestrous and lactating female mice. d, Experimental schematics and timeline. e,f, Input resistance (e) and RMP (f) of PAEsr1 cells in lactating female mice before and after application of TGOT—a selective OXTR agonist. g, Example current-clamp recording traces in response to 50 and 100 pA current injections before and after TGOT application. h,i, The number of spikes evoked by 50 pA (h) and 100 pA (i) current injections before and after TGOT perfusion. j, Experimental schematics and timeline. k–m, The latency to attack (k), attack duration (l) and intruder investigation duration (m) after injecting saline or L-371,257 hydrochloride, a potent OXTRA. n, Experimental schematics and timeline. o–q, The latency to attack (o), attack duration (p) and intruder investigation duration (q) in GFP- and Cre-injected mice. The attack latency was set to 600 s if no attack occurred. Data are mean ± s.e.m. The numbers in parentheses indicate the number of mice (b, c, k–m and o–q) and recorded cells (e, f, h and i). Statistical analysis was performed using unpaired t-tests (b, c, o and p), Wilcoxon matched-pairs signed rank tests (e, k and m), paired t-tests (f, h, i and l) and Mann–Whitney U-tests (q). Cells in e, f, h and i were recorded from four lactating female mice. All P values <0.05 are indicated. All tests were two-tailed. Detailed statistics are provided in Supplementary Table 1. Brain illustrations in d, j and n were adapted from the Allen Brain Reference Atlas (https://atlas.brain-map.org). Scale bars, 200 μm (a, main images) and 20 μm (a, insets). Source data To assess the functional role of PA OXTR signalling in maternal aggression, we injected saline or an OXTR antagonist (OXTRA; L-371,257, 100 µM, 250 nl per side) into the PA of lactating female mice 30 min before intruder introduction (Fig. 4j). After OXTRA injection, C57 × SW F1 hybrid females attacked the intruder less frequently and with longer latency compared with saline injection, while investigation duration remained unchanged (Fig. 4k–m). Similar reductions in aggression were observed in SW mothers after PA OXTRA injection (Extended Data Fig. 8). Antagonizing OXTR in the PA impairs maternal aggression in SW females. a, Experimental schematic and timeline. Brain illustrations are adapted from the Allen Brain Reference Atlas (https://atlas.brain-map.org). b, c, d, Latency to attack (b), attack duration (c), and juvenile investigation duration (d) following injection of saline or L-371,257 hydrochloride, a potent OXTR antagonist (OXTRA), into the PA of SW females. OXTRA reduced attack duration but did not affect attack latency. All bars and error bars represent mean ± s.e.m. Numbers in parentheses indicate the number of subject animals. Statistical significance was determined using (b, c) paired t-test and (d) Wilcoxon matched-pairs signed rank test. All tests are two-sided. All p values < 0.05 are indicated. See Supplementary Table 1 for more detailed statistics. Source data We further deleted Oxtr in the PA by bilaterally injecting cre-GFP virus into Oxtrflox/flox female mice on the SW and C57 mixed background (OxtrPA-KO)27 (Fig. 4n). Controls received GFP virus (OxtrPA-GFP). Compared with the controls, OxtrPA-KO female mice showed a reduced attack duration and prolonged latency, with no change in investigation time (Fig. 4o–q). Together, these manipulations demonstrate that PA OXTR signalling promotes maternal aggression. Oxtr is also highly expressed in the VMHvl and is upregulated during lactation (Extended Data Fig. 9a,b). By contrast, Esr1 expression is unchanged, whereas Npy2r expression decreases (Extended Data Fig. 9a,b). However, Oxtr is not enriched in aggression-related VMHvl cells: only 28% of Npy2r-positive cells express Oxtr in oestrous female mice, and this proportion decreases to 18% during lactation (Extended Data Fig. 9c,d). In contrast to effects of PA OXTR manipulation, neither OXTRA injection nor conditional Oxtr deletion in the VMHvl significantly altered maternal aggression, indicating that OXTR modulates maternal aggression primarily through the PA rather than the VMHvl (Extended Data Fig. 9e-l). VMHvl OXTR signalling is not essential for maternal aggression. a, Representative images showing Oxtr (magenta) and Npy2r (green) mRNA in the VMHvl of oestrus and lactating females. Right panels show enlarged views of the boxed areas. b, Number of Oxtr, Npy2r and Esr1 mRNA puncta per corresponding Oxtr-, Npy2r- and Esr1-expressing cell in the VMHvl of oestrus and lactating females. c, Venn diagrams showing the percentage of Npy2r+, Oxtr+ and Oxtr+Npy2r+ cells in the VMHvl in virgin (top) and lactating (bottom) females. d, Percentages of Oxtr+Npy2r+ cells within Npy2r+ and Oxtr+ populations. e, Schematic and timeline of experiments testing the impact of VMHvl OXTR antagonism on maternal aggression. f-h, Attack latency (f), attack duration (g), and juvenile investigation duration (h) did not differ after OXTRA (red) and saline (black) injections. The attack latency was set to 600 s if no attack occurred. i, Schematic and timeline of experiments testing the impact of VMHvl Oxtr knockout on maternal aggression. j-l, Conditional Oxtr knockout in the VMHvl did not change attack latency (j), attack duration (k), and juvenile investigation duration (l). Bars and error bars represent mean ± s.e.m. Numbers in parentheses indicate the number of subject mice. Statistical significance was determined using (b, d: Npy2r+Oxtr+/Npy2r+, k) unpaired t-test, (d: Npy2r+Oxtr+/Oxtr+, j, l) Mann-Whitney test, (f) Wilcoxon matched-pairs signed rank test, and (g, h) paired t-test. All tests are two-sided. All p values < 0.05 are indicated. See Supplementary Table 1 for more detailed statistics. Brain illustrations are adapted from the Allen Brain Reference Atlas (https://atlas.brain-map.org). Source data Oxytocin links pup to maternal attack As maternal aggression serves to protect the young, females adjust this energetically costly behaviour to pup needs. During the first postpartum week, when pups are most vulnerable to infanticide, maternal aggression peaks1,2,5 (Supplementary Note 1). In scenarios in which pups no longer exist, for example, due to predation or illness, maternal aggression rapidly declines2,5. Consistent with previous studies28,29, we found that female aggression significantly decreased 24 h after pup removal and was restored 24 h after reunion (Fig. 5a,c,d). Serum oxytocin levels closely tracked maternal aggression changes, decreasing after pup separation and increasing after reunion (Fig. 5b). This is not surprising given that oxytocin is released during nursing, which requires pup presence30–32. Pup-dependent oxytocin release maintains maternal aggression. a, Experimental schematic testing the effect of pup separation (sep.) on maternal aggression. Blood collection and RI tests were performed in separate animals. b, The serum oxytocin levels in virgin females (black), PPD3 females (red), PPD4 females 24 h after pup separation (blue) and PPD5 females 24 h after pup reunion (reun.; magenta). c,d, The attack latency (c) and attack duration (d) in PPD3–PPD5 female mice. For separation group, pups were removed after PPD3 testing and returned after PPD4 testing. The latency was set to 600 s if no attack occurred. e, Schematic and histology showing ChR2 expression and OXT staining. Right, enlarged view. f, Experiment timeline testing the effect of PVNOxt activation on maternal aggression after pup separation. g, Serum oxytocin levels in PPD4 female mice after 24 h pup separation before (blue) and after (magenta) PVNOxt stimulation. h,i, The attack latency (h) and attack duration (i) in PPD3 and PPD4 female mice after 24 h pup separation before and after PVNOxt light delivery in the ChR2 and GFP groups. j, Experimental schematic for testing the effect of PA OXTR antagonism in PVNOxt-stimulation-induced recovery of maternal aggression. k,l, The attack latency (k) and duration (l) in PPD3 and PPD4 female mice after 24 h pup separation, before and after PVNOxt stimulation in the PA OXTRA- and saline-injected groups. m, Experimental schematic for examining the effect of PVNOxt stimulation on PAVMHvl Ca2+ responses to juvenile intruders after pup separation. n, The peak Ca2+ signals of PAVMHvl cells during intruder introduction, investigation and attack. o, Representative Ca2+ traces of PAVMHvl cells during juvenile intruder interaction in PPD3 female mice (left) and PPD4 female mice after pup separation before (middle) and after (right) 1 h PVNOxt cell activation. The shaded regions denote behavioural epochs. The dashed red line indicates ΔF/F = 0.6. p–r, Peri-event time histograms of the Ca2+ signal of PAVMHvl cells aligned to intruder introduction (p), investigation (q) and attack (r). s, Representative RNAscope images showing Oxtr (magenta) and Esr1 (green) expression in the PA of a pup-separated lactating female. Right, magnified view. t,u, The percentage of PAEsr1 cells expressing Oxtr (t) and the number of Oxtr puncta per expressing cell (u) in lactating female mice after 24 h pup separation. The dashed lines indicate average values in virgin (black) and lactating (red) female mice. v, Model summarizing circuit plasticity underlying maternal aggression. During lactation, strengthened PA-to-VMHvlNpy2r connectivity and increased VMHvlNpy2r excitability enhance aggression. Pup-dependent oxytocin release further amplifies PAEsr1 output, enabling rapid and reversible adjustment of aggression according to pup-protection needs. Data are mean ± s.e.m. The numbers indicate mice. Statistical analysis was performed using one-way ANOVA with Tukey correction (b), mixed-effects analysis with repeated measures and Bonferroni correction (c,d,h,i,k and l), paired t-tests (g), repeated-measures one-way ANOVA with Tukey correction (introduction, investigation) or Friedman test with Dunn correction (attack) (n), and unpaired t-tests (t and u). All tests were two-sided. All P values <0.05 are shown. Detailed statistics are provided in Supplementary Table 1. Brain illustrations in e, j and m were adapted from the Allen Brain Reference Atlas (https://atlas.brain-map.org). Scale bars, 200 μm (e and s (left)) and 20 μm (e and s (right)). Source data To test whether decreased oxytocin underlies the drop in maternal aggression after pup separation, we optogenetically activated oxytocin neurons in the paraventricular hypothalamic nucleus (PVNOxt), which are the primary source of oxytocin and are highly activated during pup interaction30,31,33. We virally expressed ChR2 (GFP control) in PVN cells of C57 Oxt-ires-cre × SW F1 hybrid virgin female mice, which were later paired and became mothers (Fig. 5e). On PPD3, we assessed aggression using a juvenile intruder, removed pups for 24 h and retested aggression (Fig. 5f). As expected, both GFP and ChR2 females attacked the intruder before separation and showed markedly reduced aggression afterwards (Fig. 5h,i). We then delivered light stimulation (5 mW, 20 Hz, 20 ms; 20 s on, 40 s off) for 1 h, which significantly elevated serum oxytocin (Fig. 5g). Notably, aggression in ChR2 female mice recovered to preseparation levels, whereas GFP female mice remained largely non-aggressive (Fig. 5h,i). Given the critical role of PA OXTR signalling in promoting maternal aggression, we next examined whether PVNOxt-induced recovery depends on PA OXTR (Fig. 5j). OXTRA, but not saline, injected into the PA fully blocked the aggression recovery after PVNOxt stimulation (Fig. 5k,l). Consistent with the behavioural results, Ca2+ responses of VMHvl-projecting PA cells to juvenile intruders were significantly reduced after 24 h of pup separation (Fig. 5m–r). One hour of PVNOxt stimulation restored these responses to pre-separation levels (Fig. 5m–r). These findings indicate that oxytocin release from PVNOxt neurons activates PA OXTR signalling, enhancing PA output and facilitating maternal aggression. Importantly, the aggression circuit does not revert to the virgin state after 24 h of pup separation. PA Oxtr expression remained elevated 1 day after separation (Fig. 5s–u), indicating sustained oxytocin sensitivity. Moreover, optogenetic activation of PAEsr1→VMHvl terminals continued to elicit time-locked attacks, suggesting that PA–VMHvl connectivity and VMHvlNpy2r excitability remain potentiated (Extended Data Fig. 10). This circuit stability allows rapid reinstatement of maternal aggression after pup return. Together, these results suggest that oxytocin conveys information about pup presence and developmental stage to dynamically gate maternal aggression, ensuring robust expression only when needed. Optogenetic activation of PAEsr1→VMHvl cells is effective in promoting attack in lactating females following 24-hour pup separation. a, Experimental schematics. Brain illustrations are adapted from the Allen Brain Reference Atlas (https://atlas.brain-map.org). b, Experimental timeline used to examine the effect of PAEsr1→VMHvl stimulation on maternal aggression towards juvenile intruders following pup separation. c, Total attack duration of ChR2 and GFP lactating females. d-f, Following 24-hour pup separation, light stimulation in lactating ChR2 animals, but not GFP animals, (d) increased attack probability, (e) decreased attack latency, (f) and increased the average attack duration per trial. Bars and error bars represent mean ± s.e.m. Numbers in parentheses indicate the number of animals. Statistical significance was determined using (c) Mann-Whitney test, and (d, e, and f) mixed-effects analysis with repeated-measure followed by Bonferroni’s multiple-comparison test. All tests are two-sided. All p values < 0.05 are indicated. See Supplementary Table 1 for more detailed statistics. Source data Discussion Female aggression increases abruptly after childbirth in mammals to protect offspring, which requires enhanced output from the neural circuits of aggression. We identified PAEsr1 cells as an essential upstream population of the VMHvlNpy2r cells in the promotion of aggression. During lactation, the PAEsr1–VMHvlNpy2r circuit undergoes synaptic and cellular adaptations that enhance VMHvlNpy2r responses to intruders, driving aggression. Moreover, PAEsr1 cells express abundant Oxtr, enabling oxytocin to further amplify PA output. As oxytocin release is tightly coupled to pup suckling, and therefore to pup presence and developmental stage, PAEsr1 activity can be dynamically regulated by offspring status, ensuring that maternal aggression is expressed only when needed to protect the young (Fig. 5v). Plasticity of the PA-to-VMHvl pathway The VMHvl is well established as a key region for social behaviours. In female mice, two VMHvl subpopulations marked by Npy2r and Cckar selectively govern aggressive and sexual behaviours, respectively13–15. Our study of VMHvlNpy2r cells, together with our recent work on VMHvlCckar cells15, revealed their differential connections with PA and distinct reproductive state-dependent plasticity. In virgin female mice, sexual receptivity is high during oestrus and low during dioestrus. Accordingly, VMHvlCckar cells exhibit greater excitability and reduced inhibitory synaptic input during oestrus compared with dioestrus15. In most laboratory mouse strains, aggression in virgin female mice remains low across oestrous stages, yet we identified oestrous-state-dependent physiological differences in the aggression circuit. During dioestrus, PA-VMHvlNpy2r connection probability is relatively high, but VMHvlNpy2r cell excitability is low. By contrast, during oestrus, VMHvlNpy2r cell excitability is relatively high, but PA-VMHvlNpy2r connections are nearly absent. The mechanisms driving these cycle-dependent synaptic and cellular changes remain unclear, but either reduced excitability or weak synaptic input would constrain aggression circuit activation in virgin females. These oestrous-cycle-dependent circuit changes suggest that, if the aggression circuit were more active, virgin females might exhibit cycle-dependent variation in aggression. Consistent with this, female California mice and rats display higher aggression during dioestrus than during proestrus–oestrus34,35. Notably, the PA provides almost no input to mating-related VMHvlNpy2r(−)Esr1(+) cells during oestrus, despite the dense PA terminals in the VMHvll. Behaviourally, manipulating the PA-to-VMHvl pathway does not alter female sexual behaviours, indicating that VMHvl-mediated female sexual receptivity operates independently of PA inputs. Consistent with this, VMHvlCckar and VMHvlNpy2r(−)Esr1(+) cells show low sEPSC frequencies, especially compared with aggression-related VMHvlNpy2r cells (Fig. 2m and Extended Data Fig. 6q). This weak reliance on excitatory input suggests that sexual receptivity is governed primarily by hormonal state rather than external drive. Indeed, in female rats that are ‘in heat’, minimal stimulation, such as a light touch on the back, is sufficient to elicit lordosis36. During lactation, the excitability of VMHvlCckar cells is low (maximum firing rate of around 5 Hz)15, while VMHvlNpy2r cell excitability becomes high (maximum firing rate of 10 Hz). Moreover, PA-to-VMHvlNpy2r cell connectivity is strengthened, with more VMHvlNpy2r cells receiving PA input, while PA becomes largely disconnected from VMHvlNpy2r(−)Esr1(+) cells. Consequently, the PA can effectively drive VMHvlNpy2r cells without engaging mating-related VMHvl cells. This segregation is probably important, as mating-related VMHvl cells appear to antagonize aggression-related VMHvl cells15. Indeed, during lactation, activation of VMHvlCckar cells suppresses maternal aggression despite unchanged sexual receptivity15. After 24 h of pup separation, natural maternal aggression was reduced; however, PAEsr1-to-VMHvl terminal stimulation still effectively promoted attack, indicating that VMHvlNpy2r cell excitability and PAEsr1-to-VMHvl synaptic strength probably remain high. These sustained circuit properties probably enable rapid reinstatement of aggression after pup return. We speculate that PAEsr1–VMHvl circuit remodelling is induced by pregnancy-associated hormone changes and is largely independent of oxytocin levels during lactation. The molecular mechanisms underlying enhanced VMHvlNPY2r excitability and strengthened PA-to-VMHvl connectivity remain important topics for future investigation. Comparing male and female aggression circuits The PAEsr1–VMHvlEsr1 circuit is also essential for male aggression, and direct comparison across sexes provides insights into sex differences in aggressive behaviour. In males, most (around 80%) VMHvlEsr1 cells receive excitatory PA input even without prior fighting experience16. By contrast, PA input to VMHvlNpy2r cells in virgin females is sparse and oestrous-state dependent, reaching around 10% in oestrus and around 65% in dioestrus. During lactation, this proportion rises to about 80%, comparable to naive males. Beyond synaptic connectivity, intrinsic excitability also differs between sexes. The rheobase of VMHvlNpy2r cells in virgin females is higher than that of VMHvlEsr1 cells in naive males19, meaning that female aggression-related VMHvl cells require stronger excitatory drive to reach spiking threshold. Together with stronger PA inputs, lower firing thresholds and a larger number of aggression-related VMHvl cells in males (note that only about half of female VMHvlEsr1 cells contribute to aggression11), these features probably make attack behaviour more readily initiated in males than in females. Oxytocin dynamically signals the need to protect pups Maternal aggression is unique in that its primary function is to protect offspring. Accordingly, its intensity must closely track pup needs, ensuring females avoid unnecessary and potentially dangerous conflicts. Here we identify oxytocin as a key mediator linking pup needs to aggression-circuit output. High oxytocin levels maintained by suckling are essential for sustaining robust maternal aggression. During lactation, Oxtr expression markedly increases in the VMHvl, PA and many other limbic regions, probably driven by ESR1 activation after the pregnancy-associated oestrogen surge37,38. Notably, PA, rather than VMHvl, emerges as the critical site where oxytocin enhances aggression-circuit output. In the PA, Oxtr is broadly and strongly expressed in nearly all PAEsr1 cells, allowing oxytocin to increase the input–output gain of the cells by elevating input resistance, potentially through the closure of certain potassium channels39. As this effect builds slowly, it may account for the slow decline in maternal aggression after pup separation and the moderate suppression observed after acute PA OXTR antagonism. In contrast to oxytocin’s role in synaptic plasticity, which can induce long-lasting circuit changes, the effect of oxytocin on PA neurons appears to require its continued presence. This feature probably allows oxytocin to couple PA output to pup needs with high temporal fidelity. Notably, after Oxtr knockout in the PA, maternal aggression is heavily suppressed but not eliminated. The remaining aggression probably reflects an amplifying rather than an absolute gating role of oxytocin: even in the absence of oxytocin, PA input can still drive VMHvl activity to some degree. Oxtr is also abundantly expressed in the male PA, raising the possibility that oxytocin modulates aggression in fathers. This seems plausible in biparental species such as California mice, in which males display heightened paternal aggression40. Although males do not nurse, oxytocin can be released during infant interactions41. Supporting this idea, excitatory input from the lateral hypothalamus to PVNOxt neurons is strengthened in fathers compared with in virgin males42, providing a potential mechanism for increased oxytocin release during fatherhood. However, a recent study in wild mice reported a sexually dimorphic role of oxytocin in aggression: it promotes interfemale aggression while having no detectable effect on intermale aggression43. Whether male house mice (Mus musculus) display parenthood-associated aggression and whether PA oxytocin signalling contributes to such behaviours remain open questions. Oxytocin in the PA may arise from sparse direct projections of PVNOxt neurons44, but PA neurons may also detect oxytocin in the cerebrospinal fluid, given their proximity to the lateral ventricle. Indeed, CSF oxytocin levels increase during suckling45. As maternal aggression does not fluctuate on rapid timescales, the slower dynamics of CSF oxytocin should suffice to convey pup need. Importantly, this mode of regulation is reversible: oxytocin levels rise rapidly after pup reunion, and maternal aggression is promptly restored. In the VMHvl, OXTR signalling may have functions unrelated to female aggression. Our recent work showed that oxytocin originating from the retrochiasmatic supraoptic nucleus promotes social avoidance learning after defeat by enhancing synaptic potentiation of anterior VMHvlOXTR cells in both male and female mice46. Nonetheless, VMHvl oxytocin may still contribute to aggression in males, as simultaneous mutation of oxytocin and vasopressin receptors in VMHvlEsr1 cells suppresses male aggression47. Our study reveals the neural mechanisms underlying the distinctive dynamics of female mouse aggression during motherhood. Synaptic and cellular changes, probably triggered by the surge of sex hormones during pregnancy and parturition, prime the aggression circuit for rapid engagement, while oxytocin provides an additional layer of control to ensure aggression is deployed appropriately to protect the young. Together, these mechanisms allow the mother to attack intruders vigorously on first encounter while rapidly and reversibly adjusting her aggression level in response to her offspring’s needs. Methods Mice All procedures were approved by the New York University Langone Medical Center Institutional Animal Care and Use Committee in compliance with the US National Institutes of Health Guidelines for the Care and Use of Laboratory Animals. Adult mice 12–38 weeks old were used for all of the studies. Mice were housed at 18–23 °C with 40–60% humidity under a 12 h–12 h light–dark cycle (dark cycle, 22:00 to 10:00), with food and water available ad libitum. For C57 and SW mixed-background mice, we bred the F1 mixed-background mice by crossing a male C57BL/6N (Charles River, 027), C57 Esr1-2A-cre (Jackson, 017911), Esr1-2A-Flpo (Jackson, 037009), Npy2r-ires-cre (Jackson, 029285), or Oxt-ires-cre (Jackson, 024234) with a female SW (Taconic, SW-F) mouse. For mixed-background Npy2r-ires-cre/Esr1-2A-Flpo double transgenic female mice, we bred the F3 mixed-background mice by crossing F2 mixed-background homozygous Npy2r-ires-cre female mice with C57 homozygous Esr1-2A-Flpo male mice. For hybrid Oxtrflox/flox female mice (Jackson, 008471), we bred the F2 mixed-background mice by crossing F1 SW and C57 mixed background Oxtrflox/+ mice. These mixed-background breeding strategies enabled cell-type-specific manipulations in aggressive lactating female mice. C57 Esr1-ZsGreen mice were generated and provided by the Y. Xu lab originally and then bred in-house. Stimulus animals were adult (>8 weeks old) and juvenile (15 days old in one case, and 19–28 days old in all other cases) C57BL/6N female and male mice, originally purchased from Charles River and then bred in-house. Stimulus animals were group-housed. After surgery, all test virgin animals were singly housed, and all lactating animals were co-housed with their pups except during the pup-separation procedure. All experiments were performed during the dark cycle (10:00 to 16:00) of the animals. Viruses AAV2-CAG-FLEX-GFP (4.00 × 1012 viral genomes (vg) per ml), AAV2-hSyn-GFP (7.30 × 1012 vg per ml), AAV2-Ef1α-DIO-ChR2-eYFP (4.00 × 1012 vg per ml), AAV5-hSyn-Chronos-GFP (2.10 × 1013 vg per ml) and AAV2-Syn-Flex-ChrimsonR-tdTomato (6.00 × 1012 vg per ml) viruses were purchased from the University of North Carolina vector core. AAV1-CMV-Hl-eGFP-cre (2.63 × 1013 vg per ml), AAV2-Ef1α-DIO-hM4Di-mCherry (5.00 × 1012 vg per ml), AAV2-Ef1α-fDIO-mCherry (8.60 × 1012 vg per ml), AAV8-Ef1α-fDIO-GCaMP6s (1.80 × 1013 vg per ml), AAVrg-Ef1α-Flpo (7 × 1012 vg per ml) and AAV8-Ef1α-Coff/Fon-mCherry48 (2.00 × 1013 vg per ml) were purchased from Addgene. AAV2-CAG-FLEX-GCaMP6f-WPRE-SV40 (2.27 × 1013 vg per ml) was custom-made by the University of Pennsylvania vector core facility. AAVDJ-Ef1α-fDIO-hM4Di-mCherry (2.65 × 1013 vg per ml) was custom-made by Vigene Biosciences, and the plasmid was provided by B. Lim at USCD. HSV-hEf1α-LS1L-Flpo (1.00 × 109 vg per ml) was purchased from the Massachusetts General Hospital Neuroscience Center vector core. Drugs For chemogenetic inhibition, 5 mg per kg CNO (Sigma-Aldrich, C0832) dissolved in saline was administered i.p. To block OXTRs in the PA and VMHvl, 250 nl per side of 100 µM L-371,257 hydrochloride (Tocris, 2410) in saline was bilaterally injected through the implanted cannulae. Stereotaxic surgery Mice (12–20 weeks old) were anaesthetized with isoflurane (1–1.5%) and secured in a stereotaxic apparatus (Kopf Instruments, Model 1900). Viruses and tracers were delivered into the targeted brain regions as previously described16,49. In brief, injections were performed using glass capillaries and a nanoinjector (World Precision Instruments, Nanoliter) at rates of 10 nl min−1 for viruses and 5 nl min−1 for tracers. Only animals with correctly verified injection sites were included in the final analysis (n = 176 out of 194). For retrograde tracing experiments, 50 nl of Alexa-Fluor-555-conjugated CTB (Thermo Fisher Scientific, C22843) was unilaterally injected into the VMHvl (bregma coordinates: anteroposterior (AP) −1.45 mm; mediolateral (ML) ±0.680 mm; dorsoventral (DV) −5.680 mm). Brains were collected 10 days later. Only mice with confirmed targeting of the VMHvl were included in the final analysis (n = 5 out of 8). For pharmacogenetic inhibition of PAEsr1 cells, 120 nl of AAV2-hSyn-DIO-hM4Di-mCherry was bilaterally injected into the PA of Esr1-2A-cre mice (bregma coordinates: AP, −2.30 mm; ML, ±2.40 mm; DV, −4.80 mm). Only animals with confirmed bilateral targeting of the PA were included in the final analysis (n = 7 out of 8). For pharmacogenetic inhibition of PAEsr1→VMHvl cells, 130 nl of HSV-hEf1α-LS1L-Flpo was injected into the VMHvl while 120 nl of AAVDJ-Ef1α-fDIO-hM4Di-mCherry was injected into the PA bilaterally of Esr1-2A-cre mice. Control Esr1-2A-cre animals were bilaterally injected with 130 nl of HSV-hEf1α-LS1L-Flpo into the VMHvl and 100 nl of AAV2-Ef1α-fDIO-mCherry into the PA. Test animals that showed correct bilateral targeting in the VMHvl and virus expression in the PA were included in the final analysis (n = 7 out of 9). For optogenetic stimulation of the PA-VMHvl projection, 120 nl of AAV2-Ef1α-DIO-ChR2-eYFP was bilaterally injected into the PA of Esr1-2A-cre mice. Control Esr1-2A-cre mice received 100 nl of AAV2-CAG-FLEX-GFP bilaterally into the PA. During the same surgery, a 200-µm optic fibre assembly (Thorlabs, FT200EMT, CFLC230) was implanted 200 µm above each side of the VMHvl. Only animals with confirmed bilateral expression of ChR2–eYFP or GFP in the PA and correct bilateral fibre placements in the VMHvl were included in the final analysis (ChR2–eYFP: n = 14 out of 17; GFP: n = 11 out of 11). For optogenetic stimulation of VMHvlNpy2r cells, 120 nl of AAV2-Ef1α-DIO-ChR2-eYFP was bilaterally injected into the VMHvl of Npy2r-ires-cre mice. During the same surgery, a 200-µm optic fibre assembly (Thorlabs, FT200EMT, CFLC230) was implanted 200 µm above each side of the VMHvl. Only animals with confirmed bilateral expression of ChR2–eYFP and correct fibre placements were included in the final analysis (ChR2–eYFP: n = 3 out of 3). For optogenetic stimulation of PVNOxt cells, 120 nl of AAV2-Ef1α-DIO-ChR2-eYFP was bilaterally injected into the PVN of Oxt-ires-cre mice (bregma coordinates: AP, −0.80 mm; ML, ±0.250 mm; DV, −4.50 mm). Control Oxt-ires-cre mice received 100 nl of AAV2-CAG-FLEX-GFP bilaterally into the PVN. During the same surgery, a 200-µm optic fibre assembly (Thorlabs, FT200EMT, CFLC230) was implanted 200 µm above each side of the PVN. Only mice with confirmed bilateral expression of ChR2–eYFP or GFP and correct fibre placements were included in the final analysis (ChR2–eYFP: n = 5 out of 6; GFP: n = 6 out of 6). For combined pharmacological inhibition of PA OXTR and optogenetic stimulation of PVNOxt cells, 120 nl of AAV2-Ef1α-DIO-ChR2-eYFP was bilaterally injected into the PVN. During the same surgery, a 200-µm optic fibre assembly (Thorlabs, FT200EMT, CFLC230) was implanted 200 µm above the PVN, and cannulae (RWD Life Science, 62102) were implanted 500 µm above the PA bilaterally. Only animals with confirmed bilateral expression of ChR2–eYFP in the PVN and correct placements of both fibre and cannulae were included in the final analysis (n = 7 out of 7). For fibre photometric recording of PAEsr1→VMHvl cells, 130 nl of HSV-hEf1α-LS1L-Flpo was injected into the VMHvl, and 120 nl of AAV8-Ef1α-fDIO-GCaMP6s was injected into the ipsilateral PA of Esr1-2A-cre mice. A custom-made optic-fibre assembly (Thorlabs, FR400URT, CF440) was then implanted around 150 µm above the dorsal boundary of the PA and secured with dental cement (C&B Metabond, S380). Recordings commenced 3 weeks after virus injection, during both virgin and lactating states. Only animals with confirmed unilateral VMHvl targeting, correct GCaMP6s expression and proper fibre placement in the PA were included in the final analysis (n = 16 out of 19). For fibre photometric recording of VMHvlNpy2r cells, 120 nl of AAV2-CAG-FLEX-GCaMP6f-WPRE-SV40 was injected into the VMHvl of Npy2r-ires-cre mice. A custom-made optic fibre assembly (Thorlabs, FR400URT, CF440) was then implanted ~150 µm above the dorsal boundary of the VMHvl and secured with dental cement (C&B Metabond, S380). Recordings began 3 weeks after virus injection. Only animals with confirmed GCaMP6f expression in the VMHvl were included in the final analysis (n = 6 out of 7). For combined fibre photometry recording of VMHvl-projecting PA cells and optogenetic stimulation of PVNOxt cells, 130 nl of AAVrg-Ef1α-Flpo was injected into the VMHvl, 120 nl of AAV8-Ef1α-fDIO-GCaMP6s into the ipsilateral PA and 120 nl of AAV2-Syn-FLEX-ChrimsonR-tdTomato was bilaterally injected into the PVN of Oxt-ires-cre mice. During the same surgery, a 400-µm optic fibre assembly (Thorlabs, FR400URT, CF440) was implanted unilaterally, around 200 µm above both the PVN and PA. Only animals with confirmed bilateral ChrimsonR-tdTomato expression in the PVN, GCaMP6 expression in the PA and correct fibre placements were included in the final analysis (n = 6 out of 8). For pharmacological inhibition of OXTR, cannulae (RWD Life Science, 62102, 62129) were bilaterally implanted 0.5 mm above the PA or VMHvl of WT mice. Animals included in the final analysis were as follows: PA, 10 out of 11 (C57 × SW F1 hybrid); PA, 3 out of 3 (SW); VMHvl, 7 out of 7 (C57 × SW F1 hybrid). For conditional Oxtr knockout, 200 nl of AAV1-CMV-HI-eGFP-cre was bilaterally injected into the PA or VMHvl. Control animals received 100 nl of AAV2-hSyn-GFP bilaterally in the same regions. Only animals with confirmed GFP expression in the PA or VMHvl were included in the final analysis (PA with eGFP-cre: n = 8 out of 9; PA with GFP: n = 7 out of 7; VMHvl with eGFP-cre: n = 6 out of 6; VMHvl with GFP: n = 5 out of 5). For slice recordings of VMHvl-projecting PAEsr1 cells, 30 red retrobeads (Lumafluor, R170) were bilaterally injected into the VMHvl of Esr1-zsGreen mice. Brains were collected for recording 7–10 days later, with females in either the virgin or lactating state. Only animals with confirmed VMHvl targeting were included in the final analysis (n = 4 out of 4). For slice recordings of VMHvlNpy2r cells, 100 nl of AAV2-CAG-FLEX-GFP was bilaterally injected into the VMHvl of Npy2r-ires-cre mice. Brains were collected for recording 3–4 weeks later, with females in either the virgin or lactating state. Only animals with confirmed VMHvl targeting were included in the final analysis (n = 15 out of 15). To probe responses of VMHvlNpy2r and VMHvlNpy2r(−)Esr1(+) cells to PA inputs, 200 nl of AAV5-hSyn-Chronos-GFP was bilaterally injected into the PA, and 200 nl of a 1:1 mixture of AAV2-CAG-FLEX-GFP and AAV8-Ef1α-Coff/Fon-mCherry was bilaterally injected into the VMHvl of Npy2r-ires-cre × Esr1-2A-Flpo mice. Brains were collected for recording 3–4 weeks later, with females in either the virgin or lactating state. Only animals with confirmed VMHvl targeting were included in the final analysis (n = 14 out of 14). To examine the effect of oxytocin on PAEsr1 cells, 120 nl of AAV2-Ef1α-fDIO-mCherry was bilaterally injected into the PA of Esr1-2A-Flpo mice. Brains were collected for recording 3–4 weeks later, with females in either the virgin or lactating state. Only animals with confirmed PA targeting were included in the final analysis (n = 4 out of 4). Behavioural analysis Animal behaviours in all experiments were video recorded from both the side and top of the cage using two synchronized cameras (Basler, acA640-100gm) and commercial acquisition software (StreamPix 5, Norpix) in a semi-dark room with infrared illumination at 25 fps. Behavioural annotation and tracking were performed frame by frame using custom MATLAB software (https://pdollar.github.io/toolbox/). For female sexual behaviours, definitions were as follows: lordosis—the female was on the ground, motionless or arched her back while the male was mounting or intromitting. The lordosis quotient was calculated as the ratio of lordosis events to male mounting events. The deep-thrust success rate was calculated as the ratio of mounts followed by intromission to total mount events. For male sexual behaviours: shallow thrust—the male placed his forelegs over the female’s back with hindlimbs on the ground, accompanied by shallow pelvic thrusts; deep thrust—a deep, rhythmic thrust following the shallow thrust. For aggression: attack was defined as a sequence of actions initiated by the resident toward the juvenile intruder, including lunges, bites, tumbling and rapid locomotion between such behaviours. Investigation was defined as nose contact with any part of the stimulus animal. Pup retrieval was defined as lifting a pup with the jaw until it was placed in or near the nest. All mouse behaviours were manually annotated frame by frame by experimenters who were either blinded or not blinded to group assignment. We compared annotations of 15 videos performed both blinded and non-blinded and found a high level of consistency. Female aggression test To assess female aggression, we performed the resident–intruder (RI) test during the dark cycle (22:00 to 04:00). A male or female juvenile intruder (19–28 days old), an adult C57BL/6N male or an adult C57BL/6N female was introduced into the female’s home cage (Sealsafe Plus, GM500, 39.1 × 19.9 × 16 cm) for 10 min. No differences were observed in female aggression toward male versus female juvenile intruders. To evaluate aggression across reproductive states, RI tests were conducted at approximately the same time the day before mating and on postpartum days 1, 3, 5, 7, 14 and 21. When multiple intruders were tested on the same day, the juvenile intruder was always tested first, followed by adult female and adult male intruders, with at least a 2 min interval between tests. During the postpartum period, pups were removed 15 min before the RI test and returned immediately afterward. To examine the effect of pup separation on maternal aggression, all pups were removed from the home cage of lactating females on postpartum day 3 and cared for by a foster female. Then, 24 h after pup removal, a juvenile mouse was introduced into the test female’s cage for 10 min. After behavioural testing, the pups were returned to the mother’s cage, and maternal aggression towards a juvenile intruder was reassessed 24 h later. Fibre photometry Due to loss of Ca2+ signals in the PA after prolonged virus incubation in some animals, GCaMP6 activity of PAEsr1→VMHvl cells was recorded during virgin and lactating states in separate groups of females. For virgin females, recordings began 3–4 weeks after virus injection. Oestrous and dioestrous states were determined through vaginal cytology, and females were recorded during dioestrus and oestrus on separate days. During each recording session, a male or female juvenile (19–28 days old) was introduced into the cage for 10 min. For lactating females, animals were co-housed with males 3–4 days after surgery until pregnancy was visually confirmed. Cages were checked daily, and the day that pups were first observed was designated PPD1. Recordings in the lactating state were conducted on PPD3–PPD5, typically 4–5 weeks after virus injection. During these sessions, pups were temporarily removed, and a male or female juvenile (19–28 days old) was introduced for 10 min. For head-fixed recording, mice were habituated to head fixation for at least 3 days, 30 min per day. On the recording day, mice were head-fixed immediately after the RI test. An anaesthetized male, female or juvenile C57BL/6N mouse, or a 15 ml conical tube, was placed approximately 2 mm in front of the recording mouse’s nostrils by an experimenter. Each stimulus was presented four times for 5 s, with 20 s intervals between presentations and 1 min between different stimuli. For fibre photometry recording of VMHvlNpy2r cells, Ca2+ activity was monitored in the same females during both virgin and lactating states. Then, 3 weeks after virus injection, recordings were performed during dioestrus and oestrus. During each session, a male or female juvenile (19–28 days old) was introduced for 10 min. Subsequently, each female was paired with an adult male, and the male was removed once the female became visibly pregnant. Cell responses to a juvenile intruder were then recorded for 10 min on PPD3–PPD5. The fibre photometry setup was as previously described11,16. In brief, a 390 Hz sinusoidal blue LED light (30 µW; LED, M470F1; LED driver, LEDD1B; Thorlabs) was band-pass filtered (passing band: 472 ± 15 nm; FF02-472/30-25, Semrock) and delivered to the brain to excite GCaMP6. Emission light travelled back through the same optic fibre, was band-pass filtered (passing band: 535 ± 25 nm; FF01-535/505, Semrock), passed through an adjustable zoom lens (SM1NR01, Thorlabs; Edmund Optics, 62-561), detected by a Femtowatt Silicon Photoreceiver (Newport, 2151) and recorded using a real-time processor (RZ5, TDT). The envelope of the 390 Hz signals, reflecting GCaMP6 intensity, was extracted in real time using a custom TDT OpenEx program and low-pass filtered at 5 Hz. For data analysis, the MATLAB function msbackadj with a moving window of 25% of the total recording duration was first applied to obtain the instantaneous baseline signal. The instantaneous ΔF/F was calculated as: ΔF/F = (Fraw − Fbaseline)/Fbaseline. Peri-event time histograms were constructed by aligning ΔF/F signals to behaviour onset. The peak ΔF/F for each behavioural episode was defined as the maximum ΔF/F from 0 to 3 s after behaviour onset minus the average ΔF/F from −10 to −1 s before onset. The introduction peak ΔF/F for each stimulus was calculated as the maximum ΔF/F from 0 to 10 s after stimulus introduction minus the average ΔF/F from −15 to −1 s before onset. Pharmacogenetic inactivation To examine the effect of PAEsr1 or PAEsr1→ VMHvl inhibition on female sexual behaviours, the oestrous state of each female was first determined through vaginal cytology on the test day. If a female was in oestrus, a highly sexually experienced male was introduced into her home cage for around 5 min to assess sexual receptivity. The male typically attempted to mount immediately, and, if intromission occurred during this pretest, the female was considered sexually receptive and proceeded to the test. Females first received an i.p. injection of saline. Then, 30 min later, a different sexually experienced male was introduced into the female’s home cage for 10 min or until the male attempted 10 mounts. Next, 1 h later, the same female was i.p. injected with 5 mg per kg CNO (Sigma-Aldrich, C0832) and, 30 min later, a different sexually experienced male was introduced again to reassess sexual receptivity. Different males were used for each test to prevent the confounding effect of heightened sexual eagerness after previous encounters. After the mating tests, all females were paired with males until pregnancy was visually confirmed. Maternal care and aggression were then tested during PPD3–PPD5. For the maternal care test, all pups except for one were removed immediately after CNO or saline injection. Then, 30 min later, five pups were introduced far from the nest, and interactions were observed for 10 min. Next, all of the pups were removed, and 2 min later, a male or female juvenile intruder (19–28 days old) was introduced for a 10-minute RI test. Maternal care and aggression tests were repeated the next day using the same female mice 30 min after CNO or saline injection. CNO and saline administration was counterbalanced across animals. Optogenetic activation After 3 weeks of viral incubation, on test days, two 200-µm or 400-µm multimode optical fibres (Thorlabs, FT200EMT, FR400URT) were connected to the bilaterally or unilaterally implanted fibre assemblies using matching sleeves (Thorlabs, ADAL1). To activate the PAEsr1→VMHvl projection in virgin females, vaginal smears were examined 30 min before testing to identify dioestrous females. During the test, each female was introduced into the home cage of a sexually experienced male. After introduction, the male typically attempted to mount immediately. Once mounting occurred frequently and regularly, light pulses (5 mW, 20 Hz, 20 ms, 60 s on/60 s off) were delivered through the bilaterally implanted fibres using a laser (Shanghai Dream Laser) controlled by OpenEx (TDT). Light stimulation was repeated 5–6 times per session. For activating PAEsr1 to VMHvl terminals, a male or female juvenile intruder (15–28 days old) was introduced into the female’s home cage, and light pulses (5 mW or 0 mW, 20 Hz, 20 ms, 60 s on/60 s off, 5–6 times each) were delivered 2 min after intruder introduction. For activating VMHvlNpy2r cells in virgin females, 5 mW or 0 mW, 473 nm, 20 ms, 20 Hz light was delivered for 30 s, each for 4–5 times, 2–3 min after juvenile introduction. The order of sham (0 mW) and real (5 mW) stimulation was counterbalanced across animals. The light intensity at the fibre tip was measured during pulsing using an optical power meter (Thorlabs, PM100D). To examine the effect of PVNOxt cell activation on the serum oxytocin levels, PA cell activity and maternal aggression, baseline aggression was first assessed 24 h after pup separation. A male or female juvenile intruder (19–28 days old) was introduced into the test female’s home cage for 10 min. Subsequently, blue or yellow light (5 mW, 20 Hz, 20 ms pulses, 20 s on/40 s off) was delivered through the implanted optic fibres for 1 h. Immediately after light stimulation, a different juvenile intruder was introduced into the female mouse’s cage for 10 min, during which maternal behaviours and, in some cases, PA cell responses, were monitored. Oxtr region-specific knockout To knock out Oxtr in the PA and VMHvl, we bilaterally injected 120 nl of AAV1 expressing GFP-cre (control: GFP) into the PA or VMHvl of Oxtrflox/flox female mice. We have previously shown that this approach effectively knocks out Oxtr in the VMHvl (ref. 46). Then, 3–5 days after surgery, each female was co-housed with a male until pregnancy was visually confirmed. Between PPD3 and PPD5, maternal aggression was assessed by introducing a juvenile intruder into the female’s home cage for 10 min. In vivo OXTR antagonist application To block OXTR in the PA and VMHvl, cannulae (RWD Life Science, 62102, 62129) were bilaterally implanted 0.5 mm above the PA or VMHvl of female mice. Then, 3–5 days after surgery, each female was co-housed with a male until pregnancy was visually confirmed. Between PPD3 and PPD5, 100 nl of 100 µM L-371,257 hydrochloride (Tocris, 2641) was injected into each side of the PA or VMHvl through the cannula using a Hamilton syringe (65457-02) while the test animal was head-fixed on a running wheel. After injection, females were returned to their home cage. Then, 30 min later, a juvenile intruder was introduced, and the female’s behaviour was observed for 10 min. Control animals received saline injections and underwent the same behavioural testing. Measure oxytocin Oxytocin levels in mouse tail blood were measured using a previously described method50. Mice were anaesthetized with 1.5% isoflurane and placed in a stereotaxic apparatus. Tail blood was collected and centrifuged at 4 °C, 1,500 rcf for 10 min to isolate serum, which was stored at −80 °C until all samples were collected. Serum oxytocin levels were then measured using an ELISA kit (Arbor Assays, K048-H5) according to the manufacturer’s instructions. Blood samples were collected only once per animal to avoid repeated sampling-induced alterations in maternal behaviour or oxytocin levels. In vitro electrophysiological recording To prepare brain slices for patch-clamp recording, mice were anaesthetized with isoflurane, and brains were quickly removed and then immersed in ice-cold cutting solution for 1–2 min (110 mM choline chloride, 25 mM NaHCO3, 2.5 mM KCl, 7 mM MgCl2, 0.5 mM CaCl2, 1.25 mM NaH2PO4, 25 mM glucose, 11.6 mM ascorbic acid and 3.1 mM pyruvic acid). PA or VMHvl coronal sections (275 μm) were cut using the Leica VT1200s vibratome, collected in oxygenated (95% O2 and 5% CO2) and preheated (32–34 °C) ACSF solution (125 mM NaCl, 2.5 mM KCl, 1.25 mM NaH2PO4, 25 mM NaHCO3, 1 mM MgCl2, 2 mM CaCl2 and 11 mM glucose) and incubated for 30 min. The sections were then transferred to room temperature and continuously oxygenated until use. Current and voltage whole-cell patch-clamp recordings were performed using micropipettes filled with intracellular solution containing 145 mM potassium gluconate, 2 mM MgCl2, 2 mM Na2ATP, 10 mM HEPES, 0.2 mM EGTA (286 mOsm, pH 7.2) or 135 mM CsMeSO3, 10 mM HEPES, 1 mM EGTA, 3.3 mM QX-314 (chloride salt), 4 mM Mg-ATP, 0.3 mM Na-GTP and 8 mM sodium phosphocreatine (pH 7.3 adjusted with CsOH). The signals were recorded using the MultiClamp 700B amplifier (Molecular Devices) and Clampex 11.0 software (Axon Instruments), and digitized at 20 kHz with Digidata 1550B (Axon Instruments). After recording, data were analysed using Clampfit (Molecular Devices) or MATLAB (MathWorks). To characterize the physiological and synaptic properties of PAEsr1, VMHvlNpy2r and VMHvlNpy2r(−)Esr1(+) cells, we identified ZsGreen-, GFP- and mCherry-positive cells in the PA or VMHvl on slices from Esr1-ZsGreen mice and Npy2r-ires-cre × Esr1-2A-Flpo mice injected with floxed GFP and Coff/Fon mCherry viruses using an Olympus ×40 water-immersion objective coupled with GFP and TXRED filters. To investigate intrinsic excitability, cells were recorded in the current-clamp mode, and the number of action potentials was counted over 500-ms current steps. The current steps consisted of 18 sweeps from −20 pA to 150 pA at 10 pA per step. sEPSCs and sIPSCs were recorded in the voltage-clamp mode. The membrane voltage was held at −70 mV for sEPSC recordings and 0 mV for sIPSC recordings. To investigate the synaptic connectivity between the PA and VMHvl, Npy2r-ires-cre × Esr1-2A-Flpo female mice were injected with 200 nl of AAV5-hSyn-Chronos-GFP into the bilateral PA and 200 nl of a 1:1 mixture of AAV2-CAG-FLEX-GFP and AAV8-Ef1α-Coff/Fon-mCherry into the bilateral VMHvl. Then, 3 weeks after virus injection and during recording, we voltage-clamp recorded Npy2r-expressing (GFP+) or Npy2r−Esr1+ (mCherry+) VMHvl cells and then delivered 1-ms pulses of full-field illumination using a blue LED light (pE-300white; CoolLED) at 10-s intervals to activate Chronos-expressing PA terminals. oEPSCs and oIPSCs were recorded when holding the membrane potential of recorded neurons at −70 and 0 mV, respectively (corrected for a ~6 mV liquid junction potential). oEPSPs and APs were recorded under the current-clamp mode. To examine the effect of oxytocin on the cell properties of PAEsr1 cells, we injected 120 nl of AAV2-Ef1α-fDIO-mCherry into the PA of Esr1-2A-Flpo female mice. After 3 weeks of virus incubation, we obtained brain slices and performed current-clamp recordings of PAEsr1 (mCherry+) cells. Cell excitability was measured by injecting 500-ms current steps before and after 250 nM TGOT perfusion. The physiological properties of the cells were characterized −10–0 min before and 10–20 min after TGOT perfusion. Immunohistochemistry Immunofluorescence staining was performed as previously described16. In brief, mice were perfused transcardially with 1× PBS followed by 4% paraformaldehyde (PFA) in 0.1 M PBS. Brains were extracted, post-fixed in 4% PFA for 3-4 h at 4 °C followed by 48 h in 20% sucrose, embedded in OCT mounting medium, frozen on dry ice, cut to 50-μm-thick sections using a cryostat (Leica Biosystems) and collected in PBS using a 12-well plate. For antibody staining, brain sections were washed with PBS three times and blocked in PBS-T (0.3% Triton X-100 in 1× PBS) with 10% normal donkey serum for 1 h at room temperature. The sections were then incubated in primary antibody diluted in blocking solution at 4 °C for 24–72 h. We stained for ESR1 (rabbit anti-ESR1; 1:1,000; Invitrogen, PA1-309, YG378288) or FOS (guinea pig anti-FOS; 1:1,000, Synaptic Systems, 226 308). The sections were then washed with PBS-T three times and incubated in secondary antibodies diluted in blocking solution (donkey anti-rabbit or guinea pig, Alexa Fluor 488 or Cy3, 1:1,000; Jackson ImmunoResearch, 711-545-152, 711-165-152 and 706-165-148) with DAPI (1:20,000, Thermo Fisher Scientific, D1306) or NeuroTrace 435/455 Blue Fluorescent Nissl Stain (1:200, Thermo Fisher Scientific, N21479) for 2–3 h at room temperature. The sections were then washed three times with PBS, mounted onto Superfrost slides (Thermo Fisher Scientific, 12-550-15), and coverslipped for imaging on a confocal microscope (Zeiss LSM 700 microscope) and/or a virtual slide scanner (Olympus, VS120). Fluorescence in situ hybridization Extracted brains were frozen on dry ice, and 18-µm coronal sections were prepared using a cryostat (Leica Biosystems). RNAscope labelling was performed for all genes using Esr1 (478201-C1), Npy2r (315951-C2) and Oxtr (412171-C3) probes (Advanced Cell Diagnostics) according to the manufacturer’s instructions26. Dual in situ hybridization and immunohistochemistry Brain sections were treated with protease-free PretreatPro reagent according to the manufacturer’s protocol. After RNA in situ hybridization, the sections were blocked in PBS-T (0.3% Triton X-100 in 1× PBS) containing 10% normal donkey serum and 1% BSA for 30 min at room temperature. The sections were then incubated with mCherry primary antibody (rabbit anti-DsRed, 1:1,000, Takara, 632496) diluted in PBS-T with 1% BSA for 1 h at room temperature. After two washes in PBS-T, sections were incubated with secondary antibodies (donkey anti-rabbit, Alexa Fluor 594, 1:500; Thermo Fisher Scientific, A21207) diluted in PBS-T with 1% BSA. Cell counting and axon-terminal quantification To analyse labelled cells, ×20 confocal or epifluorescence images were acquired, and cells were counted manually using ImageJ. For DAPI-, Nissl-, CTB- and ESR1-positive cells, ×20 fluorescence confocal images were acquired. For GCaMP6- and mCherry-positive cells in pharmacogenetic and fibre photometry experiments, ×20 fluorescence images were obtained using a virtual slide scanner (Olympus, VS120). For Esr1-, Npy2r- and Oxtr-positive cells, ×20 fluorescence confocal images were acquired. To quantify single-cell expression of Esr1, Npy2r and Oxtr, puncta in ten individual cells from a single section were manually counted by a blinded experimenter using the ImageJ Cell Counter plugin. To quantify the PAEsr1 axon density in the VMHvl, a 200 × 200 μm2 boxed area containing fibres was selected in each subregion, and the average pixel intensity (Fraw) was measured using ImageJ and Photoshop. A same-size area in the dorsomedial VMH (VMHdm) lacking fibre terminals was used to determine background intensity (Fbackground). Fsignal was then calculated as Fraw − Fbackground. For each animal, Fsignal was normalized to the maximum Fsignal across all analysed regions. The normalized Fsignal values were then used to calculate the average terminal field intensity across animals. Statistics Sample sizes were not predetermined statistically but are comparable to previously reported studies16,49. All experiments were conducted using 2 to 11 cohorts of animals. For functional and in vivo recording experiments, histology images were collected from all animals. For slice recording experiments, histology images were only collected from a subset of animals, although the correct virus expression was always confirmed during the recording on the basis of the fluorescence protein expression. Histology images are representative and qualitatively similar to images from other animals. The results were reproducible across cohorts and combined for final analysis. All statistical analyses were performed using MATLAB 2020a (MathWorks) or Prism 10 (GraphPad). All datasets were tested for normality with the Shapiro–Wilk test, except for repeated-measures two-way ANOVA, mixed-effects analysis, Cochran’s Q test, McNemar test, χ2 test and Fisher’s exact test. For datasets that passed the normality test, parametric tests were applied, including unpaired t-tests, paired t-tests, and repeated-measures or ordinary one-way ANOVA with Geisser–Greenhouse correction followed by Tukey’s multiple-comparison post hoc test. For datasets that did not meet normality criteria, nonparametric tests were used, including the Mann–Whitney U-test, Wilcoxon matched-pairs signed-rank test, Friedman test with Dunn’s multiple-comparison post hoc test and Kruskal–Wallis test with Dunn’s multiple-comparison test. Repeated-measures and ordinary two-way ANOVA with Bonferroni’s multiple-comparison post hoc test were used for analyses involving two variables. Mixed-effects models were used to analyse repeated-measures data with missing values. Post hoc Bonferroni’s tests were performed after repeated-measures two-way ANOVA or mixed-effects analysis. Categorical data were analysed using Cochran’s Q test, χ2 test, McNemar test or Fisher’s exact test. When multiple pairwise comparisons were performed, P values were adjusted using the FDR (FDR = 0.05). Details of each statistical test, including exact P values, F values, t values, d.f. and cohort numbers, are provided in Supplementary Table 1. All error bars or error shades represent ±s.e.m. All P and q values equal to or smaller than 0.05 are indicated in the figures. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Online content Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-026-10354-5. Supplementary information Comparison of maternal aggression across C57, SW and C57 × SW F1 hybrid mice. Includes Supplementary Figs. 1 and 2. Reporting Summary Detailed statistics associated with all Figures, Extended Data Figures and Supplementary Figures. Peer Review File Source data Source Data Fig. 1 Source Data Fig. 2 Source Data Fig. 3 Source Data Fig. 4 Source Data Fig. 5 Source Data Extended Data Fig. 1 Source Data Extended Data Fig. 2 Source Data Extended Data Fig. 3 Source Data Extended Data Fig. 4 Source Data Extended Data Fig. 5 Source Data Extended Data Fig. 6 Source Data Extended Data Fig. 7 Source Data Extended Data Fig. 8 Source Data Extended Data Fig. 9 Source Data Extended Data Fig. 10 Acknowledgements We thank all of the members of the Lin laboratory for discussions; Y. Jiang for assistance with genotyping; P. Dua for help with behavioural annotation and genotyping; L. Mei for generating brain atlas images; X. Yong for sharing the Esr1-zsGreen mouse line; S. Liberles for the Npy2r-ires-cre mouse; K. Deisseroth and the INTRSECT 2.0 Project for the pAAV-Ef1a-Coff/Fon-mCherry plasmid; and B. Lim for the AAV-Ef1α-fDIO-hM4Di-mCherry plasmid. This research was supported by US National Institites of Health grants R01HD116127, R01MH101377 and U19NS107616 (D.L.), R01NS049488, R01HD109519 and R01HD108363 (N.M.S.), and R35GM147119 (M.E.P.); the Brain Research Foundation (D.L.); theVulnerable Brain Project (D.L.); an Irma T. Hirschl Career Scientist Award (M.E.P); a MIND Prize from the Pershing Square Foundation (M.E.P.); the Sumitomo Foundation (T.Y.); the Research Foundation for Opto-Science and Technology (T.Y.); the Brain Science Foundation (T.Y.); the Brain and Behaviour Research Foundation (R.Y.); and a Human Frontier Science Program postdoctoral fellowship (S.P.). Extended data figures and tables Author contributions D.L. and T.Y. conceived the project, designed the experiments and wrote the paper. D.L. supervised the project. T.Y. performed all functional manipulations, in vivo recordings and histological experiments, and analysed the corresponding data. R.Y. conducted all in vitro slice recordings, analysed the data and wrote the related results. M.K. and K.T. assisted with behavioural testing and annotation. S.K. and M.E.P. contributed to histological analyses. T.O. performed preliminary experiments on the role of VMHvl OXTR in maternal aggression. S.P. and N.M.S. generated the Esr1-2A-Flpo knock-in mice. Peer review Peer review information Nature thanks Oliver Bosch, Yoh Isogai and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available. Data availability Behavioural annotations, fibre photometry, slice electrophysiology and raw representative histology images can be downloaded from Zenodo (10.5281/zenodo.18394527)51. Behaviour videos and additional histology images are available from the corresponding authors on reasonable request. They have not been deposited in a public database owing to their large size and size limitation of online depositories. Source data are provided with this paper. Code availability MATLAB codes used in this study can be downloaded from Zenodo (10.5281/zenodo.18394527)51. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. These authors contributed equally: Takashi Yamaguchi, Rongzhen Yan Contributor Information Takashi Yamaguchi, Email: Takashi.Yamaguchi@nyulangone.org. Dayu Lin, Email: Dayu.Lin@nyulangone.org. Extended data is available for this paper at 10.1038/s41586-026-10354-5. Supplementary information The online version contains supplementary material available at 10.1038/s41586-026-10354-5. References Associated Data Supplementary Materials Comparison of maternal aggression across C57, SW and C57 × SW F1 hybrid mice. Includes Supplementary Figs. 1 and 2. Reporting Summary Detailed statistics associated with all Figures, Extended Data Figures and Supplementary Figures. Peer Review File Source Data Fig. 1 Source Data Fig. 2 Source Data Fig. 3 Source Data Fig. 4 Source Data Fig. 5 Source Data Extended Data Fig. 1 Source Data Extended Data Fig. 2 Source Data Extended Data Fig. 3 Source Data Extended Data Fig. 4 Source Data Extended Data Fig. 5 Source Data Extended Data Fig. 6 Source Data Extended Data Fig. 7 Source Data Extended Data Fig. 8 Source Data Extended Data Fig. 9 Source Data Extended Data Fig. 10 Data Availability Statement Behavioural annotations, fibre photometry, slice electrophysiology and raw representative histology images can be downloaded from Zenodo (10.5281/zenodo.18394527)51. Behaviour videos and additional histology images are available from the corresponding authors on reasonable request. They have not been deposited in a public database owing to their large size and size limitation of online depositories. Source data are provided with this paper. MATLAB codes used in this study can be downloaded from Zenodo (10.5281/zenodo.18394527)51.