Suppression of hypothalamic oestrogenic signal sustains hyperprolactinemia and metabolic adaptation in lactating mice

Abstract

17β-oestradiol (E2) inhibits overeating and promotes brown adipose tissue (BAT) thermogenesis, whereas prolactin (PRL) does the opposite. During lactation, the simultaneous decline in E2 and surge in PRL contribute to maternal metabolic adaptations, including hyperphagia and suppressed BAT thermogenesis. However, the underlying neuroendocrine mechanisms remain unclear. Here, we find that oestrogen receptor alpha (ERα)-expressing neurons in the medial basal hypothalamus (MBH), specifically the arcuate nucleus of the hypothalamus and the ventrolateral subdivision of the ventromedial hypothalamus (vlVMH), are suppressed during lactation. Deletion of ERα from MBH neurons in virgin female mice induces metabolic phenotypes characteristic of lactation, including hyperprolactinemia, hyperphagia and suppressed BAT thermogenesis. By contrast, activation of ERα^vlVMH neurons in lactating mice attenuates these phenotypes. Overall, our study reveals an inhibitory effect of E2–ERα^vlVMH signalling on PRL production, which is suppressed during lactation to sustain hyperprolactinemia and metabolic adaptations.

Lactating females experience hormonal and metabolic adaptations to the extra nutritional demands of nursing. In addition to the postpartum drop of E2 (refs. 1–3), PRL and food intake are dramatically increased, defined as lactational hyperprolactinemia^1,4 and hyperphagia⁵. Meanwhile, BAT thermogenesis^6,7 is suppressed, partially because of the inhibitory effect of the heat induced by milk production⁷. PRL and E2 are both reproductive and metabolic hormones^8–11. On the one hand, in addition to the classic roles of PRL in many maternal processes^12,13, PRL inhibits BAT thermogenesis^6,14 and promotes food intake in virgin animals^15,16, and lactational hyperprolactinemia promotes lactational hyperphagia⁵. On the other hand, E2, primarily acting through ERα, regulates energy homeostasis by suppressing overeating and increasing energy expenditure in virgin animals^17–21. Despite the opposite metabolic roles and changes of E2 and PRL during lactation, it is unclear whether and how E2 coordinates with PRL to regulate maternal metabolic adaptations.

PRL, produced primarily in the pituitary, is inhibited by dopamine (DA)^22–27 tonically and by PRL through negative feedback^22,25,28. PRL binds to PRL receptor (PRLR) in tuberoinfundibular dopaminergic (TIDA) neurons located in the arcuate nucleus of the hypothalamus (ARH) to trigger phosphorylation of signal transducer and activator of transcription 5 (STAT5), which increases the expression of tyrosine hydroxylase and therefore promotes DA release^22,29. DA then binds to DA receptor D2 (D2R) in the pituitary lactotroph cells to inhibit the release and expression of PRL. Deletion of either D2R³⁰, PRLR^25,26,31 or PRL–PRLR responder STAT5 (refs. 29,32) causes hyperprolactinemia. This negative feedback is modified during pregnancy and lactation by switching the neurotransmitter from DA to enkephalin, which in turn promotes PRL production²⁷. Lactating dams also experience suckling-induced transient PRL surges³³, which requires reduced dopaminergic tone³⁴ and suckling-induced oxytocin surge³⁵.

It is well known that E2 can promote PRL production, and PRL levels are significantly decreased in ovariectomized (OVX) female animals and increased by E2 supplement³⁶. E2 promotes lactotroph proliferation and reduces the inhibitory effect of DA through D2R on PRL^37–39. Furthermore, PRL expression is significantly reduced in the pituitary of mice with deletion of ERα but not ERβ^40,41. G protein-coupled oestrogen receptor is another oestrogen receptor that is highly expressed in the lactotroph cells and mediates the excitatory effect of E2 (ref. 42). This positive regulation is believed to account for the PRL surge at the proestrus when E2 is high in cycling females⁴³ but it does not explain the co-existence of high PRL and E2 decline during lactation¹. Interestingly, a few in vitro studies report that E2 also inhibits lactotroph proliferation⁴⁴ or induces apoptosis of lactotroph cells^45,46 under certain conditions. Furthermore, E2 has been reported to reduce PRL levels in OVX rats but only in the condition of pseudopregnancy^47,48. Therefore, it is tempting to speculate that similar inhibitory effects of E2 may also exist during lactation to regulate lactational hyperprolactinemia.

In addition to the pituitary, ERα, together with PRLR, are highly co-expressed in the neurons of the MBH, including the vlVMH⁴⁹ and TIDA neurons⁵⁰ in the ARH. Both ERα^ARH and ERα^vlVMH neurons respond to PRL surge to increase pSTAT5 (refs. 49,51). E2 decreases the synthesis of DA and the activity of DA neurons in the ARH, which in turn increases PRL levels indirectly^52–57, a stimulatory role of E2–ERα^ARH signalling potentially by inhibiting the PRLR negative feedback loop. However, the role of ERα^vlVMH neurons in PRL regulation has never been documented.

ERα^vlVMH neurons are excited during aggression and mating, and activation of the whole ERα^vlVMH population induces female aggression behaviours^58,59, whereas inhibition of these neurons reduces maternal aggression⁵⁹. Recently, another report has shown that a subpopulation of ERα^vlVMH neurons that do not express Npy2r (Npy2r⁻/ERα^+vlVMH) are highly active during mating behaviours, and activation of these neurons induces mating behaviours and reduces maternal aggression behaviours⁶⁰. Interestingly, loss of PRLR from the VMH significantly increases maternal aggression behaviours towards male mice without impairing maternal nursing behaviours⁶¹. These results support that ERα^vlVMH neurons regulate mating and maternal aggression by distinct subpopulations and are dependent on innate physical status, which may involve PRL signalling. Notably, ERα^vlVMH neurons also mediate the metabolic effects of E2. Activation of ERα^vlVMH neurons increases both thermogenesis and physical activities²¹. Loss of ERα from VMH neurons not only abolishes female sexual behaviours⁶² and increases aggressiveness towards juveniles⁶³, but also reduces BAT thermogenesis and energy expenditure, resulting in obesity^17–21. In addition, loss of ERα from the ARH leads to both infertility and hyperphagia¹⁷. However, it is unclear whether these neurons also regulate metabolic adaptations during lactation and whether their function involves PRL signalling.

Different from the stimulatory effects of E2 on PRL, we identified an inhibitory effect of E2 on PRL levels indirectly through ERα^vlVMH neurons. Importantly, we found that ERα neurons in the MBH were suppressed during lactation. We reason that lactational suppression of ERα^vlVMH neurons is required to allow lactational adaptations including hyperprolactinemia, hyperphagia and suppressed BAT thermogenesis. To confirm this hypothesis, we specifically deleted ERα from MBH or vlVMH neurons in virgin female mice or specifically activated ERα^vlVMH neurons in lactating dams and then characterized their impacts on feeding and BAT thermogenesis as well as PRL levels. Finally, we explored the molecular mechanism for the lactational suppression of ERα^vlVMH neurons and the circuitry mechanism for the regulatory role of ERα^vlVMH neurons in PRL regulation.

Results

Lactational metabolic adaptations and hyperprolactinemia

We used mouse models to study neuroendocrine mechanisms for lactational metabolic adaptations. Consistent with previous reports^4,6,64, we found that the weight of BAT trended to decrease during lactation (Fig. 1a). Importantly, the expression of UCP1 protein, a marker of BAT thermogenesis, was significantly reduced during lactation (Fig. 1b). Consistently, we found a reduction of BAT tissue, largely caused by the expansion of mammary glands during lactation (Fig. 1c). Meanwhile, lactating mice showed dramatic increases in food intake without increases in body weight (Fig. 1d). These results confirmed lactational suppression of BAT function and lactational hyperphagia in mice. We also confirmed hyperprolactinemia in mice (Extended Data Fig. 1a), which is believed to contribute to these metabolic adaptations^5,6,12,14. Interestingly, the number of pups nursed by the dams was positively correlated with the amount of food intake and the levels of PRL in dams but not with the BAT weight or UCP1 expression (Extended Data Fig. 1b–e).

Fig. 1 |. Lactational metabolic adaptations.

a–c, The weight of BAT (a), the expression of UCP1 in BAT by western blot (b) and H&E staining of BAT (c) in virgin and lactating (PPD11) female mice (n = 5). Boxed areas are magnified as indicated. d, Body weight (top panel, virgin, n = 10; first week during lactation, n = 65; second week during lactation, n = 58; dams without pups, n = 4) and daily food intake (bottom panel, virgin, n = 9; first week during lactation, n = 63; second week during lactation, n = 58; dams without pups, n = 5) of virgin and lactating female mice. Data are presented as mean ± s.e.m. and individual data points; in b, two-tailed unpaired t-tests; d, * or #, groups versus virgin or first week during lactation in ordinary one-way ANOVA followed by Dunnett’s multiple comparisons test. P values are shown.

Lactational suppression of ERα^MBH neurons

Many hypothalamic nuclei and neuron populations have been implicated in the regulation of these lactational adaptations⁵. E2 has been reported to inhibit overeating and promote BAT thermogenesis in virgin female animals, and these metabolic effects are primarily mediated by ERα neurons located in the MBH, including the ARH and the vlVMH^17–21. We next used slice electrophysiology to record identified ERα neurons in the hypothalamic nuclei of ERα-ZsGreen mice⁶⁵ (Fig. 2a). We compared their spontaneous firing activities and responsiveness to rapid E2–ERα excitation among age-matched virgin, pregnant and lactating female mice. In virgin female mice, 89% of ERα^vlVMH neurons were active and displayed spontaneous firing activity, while 11% were silent ERα^vlVMH neurons without spontaneous firing activity. Strikingly, the population of silent ERα^vlVMH neurons gradually increased from late pregnancy (gestation day (G)18), with around half of ERα^vlVMH neurons being silent on postpartum day 5 (PPD5) (Fig. 2b,c,e). Even within those active ERα^vlVMH neurons with spontaneous firing, both firing frequency and resting membrane potential were significantly and continuously decreased after delivery (Fig. 2f,g). Interestingly, this suppression persisted long-term postpartum, with ERα^vlVMH neurons on PPD5 displaying the lowest neuronal activity (Extended Data Fig. 2a). To evaluate the responsiveness of ERα^vlVMH neurons to E2, we also compared their responsiveness to propylpyrazoletriol (PPT), an ERα agonist⁶⁶. Consistent with our previous report⁶⁶, 100% of ERα^vlVMH neurons from virgin female mice were activated by PPT. However, nearly 40% of ERα^vlVMH neurons from lactating dams were no longer responsive to PPT (Fig. 2d,h and Extended Data Fig. 2b). Furthermore, for those PPT-excited ERα^vlVMH neurons, the PPT-induced increases in firing frequency and resting membrane potential were significantly attenuated during late pregnancy and lactation (Fig. 2i,j and Extended Data Fig. 2g). ERα neurons in the ARH also displayed a reduction in spontaneous firing activity and PPT responsiveness, but the resting membrane potential and the silent and firing populations did not change during lactation (Fig. 2k and Extended Data Fig. 2c). ERα neurons in the preoptic area (POA) only displayed reduced spontaneous firing activity during lactation (Fig. 2l and Extended Data Fig. 2), and ERα neurons in the paraventricular nucleus of the hypothalamus (PVN) only displayed a nonsignificant increased trend of the silent population (Fig. 2m and Extended Data Fig. 2e).

Fig. 2 |. Lactational suppression of ERα^MBH neurons.

a, Bright-field illumination (top) and fluorescence for ZsGreen (bottom) of a recorded ERα neuron in a brain slice of an ERα-ZsGreen mouse. b, vlVMH region in the brain section. c, Representative firing trace of a spontaneously firing ERα neuron from a virgin female mouse (top) and a silent ERα neuron from a lactating dam on PPD5 (bottom). d, Representative traces of an ERα neuron excited by PPT from a virgin female mouse (top) or an ERα neuron irresponsive to PPT from a lactating dam on PPD5 (bottom). e, Percentage of spontaneous firing and silent ERα^vlVMH neurons from virgin (n = 27), G18 (n = 21), PPD2 (n = 24) and PPD5 (n = 23) mice. f,g, Spontaneous firing frequency (f) and resting membrane potential (g) of ERα^vlVMH neurons from virgin (f, n = 25; g, n = 27), G18 (f, n = 16; g, n = 21), PPD2 (f, n = 14; g, n = 24) and PPD5 (f, n = 11; g, n = 23) mice. h, Percentage of ERα^vlVMH neurons activated by PPT (virgin, n = 27; G18, n = 21; PPD2, n = 24; PPD5, n = 23). PPT-induced activation was defined as either >20% increases in firing rate or by >2 mV depolarization. i,j, PPT-induced increases of firing rate (i) and depolarization in PPT-responsive ERα^vlVMH neurons (j) from virgin (i, n = 25; j, n = 27), G18 (i, n = 14; j, n = 19), PPD2 (i, n = 12; j, n = 20) and PPD5 (i, n = 9; j, n = 14) mice. Neurons from three mice in each group were used in e–j. k–m, Percentage of spontaneous firing and silent subpopulation, firing frequency, resting membrane potential and PPT-induced response of firing rate and depolarization in ERα^ARH neurons (k, firing frequency: virgin, n = 10; PPD5, n = 9; resting membrane: virgin, n = 14; PPD5, n = 13), ERα^POA neurons (l, firing frequency: n = 10; resting membrane: virgin, n = 12; PPD5, n = 13) and ERα^PVN neurons (m, firing frequency: virgin, n = 9; PPD5, n = 8; resting membrane: virgin, n = 11; PPD5, n = 15) from virgin and PPD5 mice. Neurons from three mice in each group. Data are presented as part of whole bar graphs (e, h, k, l and m) or mean ± s.e.m. and individual data points (f, g, i, j, k, l and m). In e, h and m, groups versus virgin in two-sided χ² test; in f, g, i and j, groups versus virgin in ordinary one-way ANOVA followed by Dunnett’s multiple comparisons test; in k, l and m, two-tailed unpaired t-tests. P values are shown. ΔRM, changes of resting membrane potential; ΔFiring, changes of firing frequency. The brain section pictures in b, k, l and m were created in BioRender.com.

Together, these results showed that lactating dams displayed hyperprolactinemia, hyperphagia, suppressed BAT thermogenesis and suppression of hypothalamic ERα neurons. The suppression of firing activity and E2 responsiveness were more profound in ERα^vlVMH neurons than in ERα^ARH neurons, ERα^POA neurons and ERα^PVN neurons (Fig. 2e–m and Extended Data Fig. 2f). To further explore whether other oestrogen receptors also mediate oestrogenic effects in the vlVMH, we performed secondary analysis of published single-cell RNA sequencing (scRNA-seq) data^60,67. Two scRNA-seq datasets both indicated high expression of ERα but very low ERβ and almost undetectable G protein-coupled oestrogen receptor in the VMH (Extended Data Fig. 2h), in agreement with the essential role of E2–ERα signalling in energy balance and maternal behaviours. Therefore, despite lactating dams only displaying a modest reduction of serum E2 levels on PPD5 (Extended Data Fig. 2i)^68,69, our results indicated that E2 actions were suppressed in ERα^MBH neurons, with more profound silence and E2 irresponsiveness in ERα^vlVMH neurons during lactation.

ERα^MBH-knockout mice have lower BAT thermogenesis and hyperphagia

To determine whether the suppression of E2–ERα signalling in MBH neurons contributes to lactational metabolic adaptations, we used an adeno-associated virus (AAV) strategy to delete ERα from MBH neurons in virgin female mice (Fig. 3a). ERα was deleted from the entire vlVMH but only part of ARH (Fig. 3b and Extended Data Fig. 3a). These mice were defined as ERα^MBH-knockout (KO) mice. Regardless how much ARH was impacted, all virgin ERα^MBH-KO female mice showed significant body weight gain (Fig. 3c and Extended Data Fig. 3b) without increasing food intake during the first month after virus injection (Extended Data Fig. 3c). These ERα^MBH-KO female mice also displayed BAT whitening with the development of mammary gland-like patches in the BAT and lower BAT temperature (Fig. 3d,e). Furthermore, we found that the serum PRL level was comparable at week 2 after the virus injection but significantly increased at week 5 (Fig. 3f). Consistently, food intake increased only after 5 weeks post injection (Fig. 3g), coincident with the increase of PRL. Interestingly, feeding efficiency was significantly increased at 2 weeks after virus injection (Extended Data Fig. 3d), coincident with higher body weight but before the increases in food intake and PRL. Previous reports have shown that ERα^vlVMH neurons activate BAT thermogenesis^19,21 and female ERα^vlVMH-KO mice display impaired BAT thermogenesis and obesity^17,18,20, suggesting reduced energy expenditure. To confirm this idea, we measured the energy expenditure of a cohort of control and ERα^vlVMH-KO mice just 2 weeks after virus injection, before the increase in food intake (Extended Data Fig. 3e,f). As expected, energy expenditure and O₂ consumption were significantly lower in heavier ERα^vlVMH-KO mice, using body weight as a covariant^66,70 (Fig. 3h–j and Extended Data Fig. 3g–i). The higher the body weight was, the lower the energy expenditure was in the ERα^vlVMH-KO mice, just opposite from the control mice. The predicted energy expenditure, O₂ consumption and CO₂ production were significantly lower in ERα^vlVMH-KO mice at a presumed body weight^66,70 (Fig. 3h–j and Extended Data Fig. 3g–l). Together, these results suggest that the body weight gain in ERα^MBH-KO female mice is primarily attributed to reduced energy expenditure, and the increased food intake may be secondary to either the increased PRL level or the increased body weight. We further generated an additional group of control mice, with wild-type (WT) mice receiving the same virus injection, and these mice did not display any changes in body weight, food intake or PRL levels (Extended Data Fig. 3m–o).

Fig. 3 |. Deletion of ERα from MBH neurons suppresses BAT thermogenesis and promotes feeding.

a, Schematic design of virus injections in ERα^fl/fl mice to delete ERα in MBH neurons including the ARH and vlVMH, and phenotyping studies spanning 17 weeks. Created in BioRender.com. FI, food intake; TSH, thyroid-stimulating hormone. b, Immunostaining of ERα in virgin female control and ERα^MBH-KO mice. ERα was deleted from the entire vlVMH and part of the ARH. 3V, third ventricle. c–e, Body weight (c), H&E staining of BAT (d, 4 months after surgery) and BAT temperature (e, 4 weeks after virus injection) in virgin female control (c, n = 6; e, n = 7) and ERα^MBH-KO mice (c, n = 7; e, n = 8). Boxed areas in d are magnified as indicated. Arrows indicate the mammary gland-like patches. f,g, Serum PRL level (f) and food intake (g) of female control (n = 6) and ERα^MBHKO mice (n = 7) at 2 weeks and 5 weeks after virus injection, respectively. h, Temporal levels of energy expenditure measured by the TSE PhenoMaster during the dark cycle and light cycle over 3 days. i, ANCOVA analysis of energy expenditure indicated significant difference between control and ERα^MBH-KO mice using body weight as a covariant^66,70. j, Predicted energy expenditure (EE) was lower in ERα^MBH-KO mice at a presumed body weight^66,70. In h–j, control, n = 8; ERα^MBH-KO, n = 6. Data are mean ± s.e.m. and/or individual data points. In c, two-tailed unpaired t-tests; two-way ANOVA followed by Šídák’s multiple comparison test with Geisser–Greenhouse correction. In e and j, two-tailed unpaired t-tests; f, ordinary one-way ANOVA followed by Dunnett’s multiple comparisons test; g, two-tailed unpaired t-tests (control vs. ERα^MBH-KO) and two-tailed paired t-tests (ERα^MBH-KO vs. ERα^MBH-KO); i, ANCOVA analysis by CalR. P values are shown.

ERα^MBH-KO mice develop hyperprolactinemia and prolactinoma

Strikingly, 4 months after the deletion of ERα from the MBH in virgin female mice, we found a dramatic increase in serum PRL level (>30-fold) compared to the control mice, associated with a nonsignificant trending decrease in serum E2 level (Fig. 4a,b). Consistent with this hyperprolactinemia, we detected dramatic increases in pSTAT5, a well-known PRL signalling responder⁵¹, in several hypothalamic nuclei, including the ARH and the PVN (Extended Data Fig. 4a). The pituitary of ERα^MBH-KO virgin female mice developed prolactinoma as indicated by a drastically enlarged pituitary (Fig. 4c). Haematoxylin and eosin (H&E) staining showed that the pituitary of ERα^MBH-KO mice displayed visible cystic spaces, peliosis and atypical nuclei for lactotrophs, together with increased expression of PRL in the pituitary (Fig. 4d and Extended Data Fig. 4b,c), confirming the development of prolactinoma. Interestingly, the hypothalamus from a virgin female ERα^MBH-KO mouse (with high retention of ERα in the ARH) was heavily compressed by this long-term enlarged prolactinoma, especially in the posterior ARH, such that it was difficult to distinguish the boundary between the ARH and vlVMH in the caudal hypothalamus (Extended Data Fig. 4d).

Fig. 4 |. Deletion of ERα from MBH neurons promotes the development of hyperprolactinemia and reproductive phenotypes seen in lactation.

a–d, Serum PRL levels (a), serum E2 levels (b), photograph of the prolactinoma and compressed hypothalamus (c) and immunofluorescence of PRL in the pituitary (d) of virgin female control (a, n = 6; b, n = 8) and ERα^MBH-KO (a, n = 8; b, n = 13) mice after 2 months on a chow diet and 2 months HFD feeding. e, The appearance of mammary glands in female virgin control, lactating and virgin ERα^MBH-KO mice. f, Mammary ductal branches and buds in virgin control and virgin ERα^MBH-KO mice observed by whole-mount H&E staining. g, The oestrous cycles of control (n = 8) and ERα^MBH-KO (n = 6) mice. E, oestrus; M, metoestrus; D, dioestrus; P, proestrus. h, The fertility rate of control (n = 14) and ERα^MBH-KO (n = 6) mice. Data are mean ± s.e.m. and individual data points (a and b) or pie graphs (g and h). In a, two-tailed unpaired t-tests; h, two-side χ² test. P values are shown.

Together, our findings show that virgin ERα^MBH-KO female mice displayed suppressed BAT thermogenesis and increased feeding, associated with hyperprolactinemia. Obesity can disrupt hypothalamus–pituitary endocrine systems⁷¹, leading to both reproductive and metabolic symptoms. To further explore the neuroendocrine mechanisms downstream of ERα^MBH neurons and their metabolic contribution, we measured a series of pituitary hormones at 4 months after virus injection, including 2 months of chow diet feeding and 2 months of high-fat diet (HFD) feeding. We used HFD to boost the metabolic and reproductive phenotypes to screen out all potential changes before narrowing down the changed hormones to a specific time point. Although no differences were observed in serum growth hormone, insulin-like growth factor 1 and adrenocorticotropic hormone, ERα^vlVMH-KO mice showed a slight but significant increase in thyroid-stimulating hormone (Extended Data Fig. 4e–h). However, the thyroid-stimulating hormone levels were comparable between control and ERα^vlVMH-KO mice 6 weeks after virus injection, right after the increases in body weight and food intake (Extended Data Fig. 4i). These results excluded the contribution of these hormones to the PRL increase and metabolic changes in virgin ERα^MBH-KO female mice.

ERα^MBH-KO mice develop lactational reproductive phenotypes

Consistent with the high PRL levels, we noticed that the mammary glands of virgin female ERα^MBH-KO mice were similar to those of lactating female mice (Fig. 4e). The mammary ductal branches and buds were increased in the virgin female ERα^MBH-KO mice (Fig. 4f), and H&E staining showed patched mammary tissues within the inguinal white adipose tissue (iWAT) (Extended Data Fig. 4j). Furthermore, ERα^MBH-KO virgin female mice displayed impaired oestrous cycles after 3 months post injection, with all mice remaining in the metoestrus phase (Fig. 4g), despite other reproduction-related serum levels of progesterone, luteinizing hormone and follicle-stimulating hormone being comparable between control and ERα^vlVMH-KO virgin female mice (Extended Data Fig. 4k–m). We then generated another cohort of ERα^MBH-KO virgin female mice and tested their fertility at 7 weeks after the surgery, right after the increase of PRL levels and lactational appearance of mammary glands. Compared to the 78% (11 out of 14) pregnancy rate and 71% (10 out of 14) successful delivery rate in female control mice, only 33% (2 out of 6) of hyperprolactinemic ERα^MBH-KO female mice became pregnant (Fig. 4h). One of the only two pregnant ERα^MBH-KO female mice ended in miscarriage, as indicated by acute drop of body weight and the appearance of blood during late pregnancy. The other pregnant ERα^MBH-KO female mice lost all the pups within 3 days. Therefore, loss of ERα in MBH neurons causes virgin females to exhibit a series of reproductive phenotypes seen in lactation: mammary gland development, cessation of the oestrus cycle and decreased fertility, which could be secondary to hyperprolactinemia.

PRL regulatory role of ERα^vlVMH neurons

Post-hoc validation confirmed complete deletion of ERα from the vlVMH of all ERα^MBH-KO mice, but only partial deletion of ERα from the ARH. However, a few ERα^MBH-KO mice with very mild deletion of ERα from the ARH (Extended Data Figs. 3a and 4d) also displayed the metabolic phenotypes and hyperprolactinemia or prolactinoma. Thus, we speculate that vlVMH is the major area that contributes to these phenotypes. It is very difficult to restrain the AAV virus specifically in the vlVMH without touching the ARH or in the ARH without touching the vlVMH in both sides. A previous report used a similar AAV-Cre approach to delete ERα from the ARH and found a mild increase of PRL 12 weeks post injection⁷². To minimize the chance of contamination to the ARH, we generated mice with minimal deletion of ERα from the vlVMH by injecting a reduced volume (50 nl) of AAV-Cre virus to only one side of the vlVMH. As a result, ERα was deleted from only half the population of vlVMH neurons on only one side, preserving intact ERα in the ARH (Extended Data Fig. 5a,b) and the other side of the vlVMH. Interestingly, these one-quarter ERα^vlVMH-KO mice displayed protruding nipples like lactating dams and increased PRL levels without changes in body weight or food intake within 12 weeks (Extended Data Fig. 5c–f). These results suggest that ERα^vlVMH neurons at least regulate PRL levels, and the changes in metabolic balance could be secondary to prolonged PRL elevation.

ERα^vlVMH neurons inhibit lactational hyperprolactinemia

We showed that loss of ERα in vlVMH neurons led to hyperprolactinemia. To test whether the suppression of ERα^vlVMH neurons is required for lactational hyperprolactinemia, we used an excitatory DREADD strategy to specifically activate ERα^vlVMH neurons during lactation. We stereotaxically injected the Cre-dependent excitatory AAV-DIO-hM3Dq-mCherry virus into the vlVMH of ERα-Cre mice to express hM3Dq protein specifically in ERα^vlVMH neurons (ERα^vlVMH-hM3Dq); WT mice receiving the same virus were used as controls (Fig. 5a). The activation of ERα^vlVMH neurons in ERα^vlVMH-hM3Dq mice was validated by increased firing frequency and resting membrane potential following the application of clozapine N-oxide (CNO) and co-expression of mCherry and c-fos after CNO injection (Extended Data Fig. 6a–d).

Fig. 5 |. Activation of ERα^vlVMH neurons inhibits lactational hyperprolactinemia, maternal metabolic adaptations and nursing behaviours.

a, Schematic design for the generation of ERα^vlVMH-hM3Dq mice. b, Schematic for chemogenetic activation of ERα^vlVMH neurons followed by measurements of serum PRL levels and BAT temperatures on PPD5. Created in BioRender.com. c, Basal PRL levels and numbers of pups of control (n = 14) and ERα^vlVMH-hM3Dq (n = 20) lactating dams from PPD1 to PPD5. Black lines represent PRL levels excluding the dams that lost all pups, and orange lines represent the number of pups. d,e, Serum PRL levels before and 2 h after CNO injection (d) and BAT temperatures before and 1 h after CNO injection (e) in control (d, n = 12; e, n = 9) and ERα^vlVMH-hM3Dq (d, n = 11; e, n = 6) lactating dams on PPD5, excluding the dams that lost all pups. f, Daily food intake of control (n = 15) and ERα^vlVMH-hM3Dq (n = 21) lactating dams including the dams that lost all pups. g, Schematic of the maternal behaviour test. h–j, Time for the completion of pups’ retrieval (h), maternal care (i) and time spent in the nest (j) of control (n = 5) and ERα^vlVMH-hM3Dq (n = 4) lactating dams on PPD5. Data are presented as mean ± s.e.m. and/or individual data points. In c and f, two-way ANOVA followed by Šídák’s multiple comparisons test with Geisser–Greenhouse correction; in d and e, two-way repeated-measures ANOVA by multiple comparisons test with uncorrected Fisher’s least significant difference; in j, P < 0.05 in two-tailed unpaired t-tests. P values are shown.

After recovering from the surgery, these female mice were bred with WT experienced male breeders to produce lactating dams and litters. We activated ERα^vlVMH neurons by intraperitoneal (i.p.) injections of CNO twice a day from PPD1 to PPD11. The basal PRL levels (before CNO injection each day, excluding the dams who lost pups) were comparable on PPD1 but gradually lower in ERα^vlVMH-hM3Dq mice on PPD2 and PPD5 followed by decreased numbers of pups (Fig. 5b,c). Given that suckling can induce a PRL surge^33–35, and PRL level was correlated with the number of pups (Extended Data Fig. 1c), we excluded the lactating dam who lost all pups from the PRL analysis. On PPD5, 1 h after the pups were removed, dams received a CNO injection, and serum samples were collected before and 2 h after the CNO injection for the measurement of serum PRL levels (Fig. 5b). After 2 h of separation from pups, PRL levels were decreased in both groups following the cessation of suckling. However, ERα^vlVMH-hM3Dq mice with the activation of ERα^vlVMH neurons showed a further dramatic decrease in serum PRL levels compared to the control group (Fig. 5d). We repeated this experiment and found the same phenotype on PPD2 and PPD9 (Extended Data Fig. 6e,f). These results indicate that activation of ERα^vlVMH neurons downregulates PRL levels in lactating dams.

ERα^vlVMH neurons attenuate lactational metabolic adaptations

We then asked whether the suppression of ERα^vlVMH neurons also contributes to the suppression of BAT thermogenesis and hyperphagia during lactation. Body weight, food intake and BAT temperature were comparable between ERα^vlVMH-hM3Dq and control mice during the entire pregnancy period (Extended Data Fig. 6g–i). During lactation, baseline BAT temperatures (before CNO injection each day) were not changed in ERα^vlVMH-hM3Dq mice (Extended Data Fig. 6l). To minimize the inhibitory effect of increased heat production from milk production on thermogenesis, we separated pups from dams for 1 h before the CNO injection and then measured the BAT temperature before and 1 h after the CNO injection. CNO injection significantly increased the BAT temperature of ERα^vlVMH-hM3Dq dams (Fig. 5e) compared to that of the control dams, suggesting an acute excitatory effect of ERα^vlVMH neurons on BAT function in lactating dams.

Activation of ERα^vlVMH neurons significantly reduced food intake of lactating ERα^vlVMH-hM3Dq dams without changing their body weight (Fig. 5f and Extended Data Fig. 6j). This reduced energy intake, together with reduced PRL levels, may inhibit the energy supply and milk production. As a result, the survival of the pups of ERα^vlVMH-hM3Dq dams was reduced (Fig. 5c). Given that the nutritional demand of the pups through energy withdrawal in the form of milk is also a major driver for lactational hyperphagia, we analysed the food intake of lactating dams with or without those who lost pups. Activation of ERα^vlVMH neurons significantly reduced food intake of lactating ERα^vlVMH-hM3Dq dams regardless of whether they lost pups or not (Fig. 5f and Extended Data Fig. 6k). These results indicate that suppression of ERα^vlVMH neurons contributes to both lactational hyperphagia and suppression of BAT thermogenesis.

Interestingly, despite the reduced pup number, the body weight of surviving pups nursed by ERα^vlVMH-hM3Dq dams was not significantly decreased (Extended Data Fig. 6m). Considering the essential role of ERα^vlVMH neurons in maternal behaviours⁶⁰ and the maternal effect of PRL^12,13, we further analysed the maternal behaviours in ERα^vlVMH-hM3Dq dams. Compared to control dams, ERα^vlVMH-hM3Dq dams tended to spend more time retrieving all pups but less time exhibiting maternal care behaviours, including grooming, crouching and nest-building (Fig. 5g–i and Extended Data Fig. 6n–p). In addition, ERα^vlVMH-hM3Dq dams spent significantly less time in the nest with pups (Fig. 5j). These results indicate that activation of ERα^vlVMH neurons disturbs the good maternal behaviours in lactating dams, potentially through reducing PRL levels. Therefore, ERα^vlVMH-hM3Dq dams might have developed a strategy to maintain the wellbeing of the pups as a whole at the cost of individual pup loss, a potential compensatory mechanism for the shortage of milk supply.

Central E2 actions inhibit lactational hyperprolactinemia

To further confirm whether E2 can inhibit PRL during lactation through a central mechanism, we supplemented E2 to lactating dams from PPD1 (the day after delivery), either systemically through subcutaneous (s.c.) pellet implantation or centrally through brain infusion (Fig. 6a). Then we measured the serum PRL levels of the lactating dams on PPD5. Interestingly, the systemic supplementation of E2 during lactation did not change the PRL levels (Fig. 6b), whereas the central supplementation of E2 significantly decreased PRL levels (Fig. 6c). These results are consistent with our finding of the inhibitory effect of E2 on PRL regulation via ERα^vlVMH neurons.

Fig. 6 |. E2 acts on the central nervous system to inhibit lactational hyperprolactinemia.

a, Schematic of E2 supplement to lactating dams either systemically (0.5 or 2.4 μg d⁻¹) by s.c. pellet implantation (top) or centrally (0.1 μg d⁻¹) by an ALZET Brain Infusion Kit attached to an s.c. implanted osmotic pump (bottom). V, vehicle. After the pups were separated from the dams for 2 h on PPD5, serum samples were collected for the measurement of PRL levels. Created in BioRender.com. b,c, PRL levels of lactating dams with systemic (b, placebo, n = 13; E2, 0.5 μg d⁻¹, n = 5; E2, 2.4 μg d⁻¹, n = 9) or central (c, vehicle, n = 5; E2, 0.1 μg d⁻¹, n = 6) supplement of E2. d, Schematic of E2 supplement to virgin OVX mice centrally (0.1 μg d⁻¹). e, Serum PRL levels with central supplement of vehicle or E2 in virgin OVX mice (n = 5). Created in BioRender.com. Data are presented as mean ± s.e.m. and individual data points. In c and e, two-tailed unpaired t-tests. P values are shown.

We also monitored the dams and pups to exclude the possibility that the PRL level is secondary to other effects of the E2 supplement. The number of surviving pups was not changed with the central supplement of E2 but gradually reduced with the systemic supplement of E2 from PPD6 (Extended Data Fig. 7a,b). Notably, we measured PRL in dams with matched pups on PPD5 before the number of surviving pups showed a significant difference, thus avoiding the confounding effect of different pup numbers. Neither systemic nor central supplement of E2 changed the body weight of lactating dams (Extended Data Fig. 7c,d). However, systemic E2 supplement, but not central E2 supplement, significantly reduced the food intake of dams and body weight of the pups, indicating reduced milk production (Extended Data Fig. 7e–h). This inhibitory effect of systemic E2 on lactation is consistent with previous reports that E2 inhibits mammary gland function and milk production⁷³. Indeed, mammary glands from the dams supplemented with the E2 pellet displayed disrupted mammary ducts and a trend of augmented adipocyte morphology (Extended Data Fig. 7i,j). This effect should not have been caused by PRL levels, as PRL levels were not changed by the systemic supplement of E2 (Fig. 6b). Interestingly, the central E2 supplement did not reduce the food intake of dams or the body weight of the pups (Extended Data Fig. 7f,h). These results support that the inhibitory effect of E2 by central actions on PRL level regulation is independent of other effects of E2, including body weight, food intake and the suckling effect from pups.

It has been well established that systemic E2 supplement increases PRL levels in virgin OVX-E2 animals³⁶, conflicting with unchanged PRL levels with systemic supplement of E2 in lactating dams. To evaluate the central effect of E2 on PRL in OVX mice, we supplement vehicle (OVX-V^C) or E2 (OVX-E2^C) to the brain of virgin OVX female mice. As expected, OVX-V^C female mice displayed very low PRL levels. Meanwhile, central E2 supplement attenuated the drop of PRL levels in OVX-E2^C mice (Fig. 6d,e), displaying an excitatory effect on PRL. These results indicate that the central inhibitory effect of ERα^vlVMH neurons may only manifest under certain physiological conditions, like lactation, but not in virgin OVX mice.

Lactational modulation of the transcriptome in ERα^vlVMH neurons

To explore the molecular mechanism for the suppression of ERα^vlVMH neurons, we manually picked up ERα^vlVMH neurons after electrophysiology recording and processed them for scRNA-seq using patch–seq. We collect ten neurons per group for three major groups of ERα^vlVMH neurons: dioestrus virgin firing versus PPD5 firing (PPT-responsive vs. PPT-irresponsive) versus PPD5 silent (PPT-responsive vs. PPT-irresponsive) (Fig. 7a and Extended Data Fig. 8a). We detected an average of 21,400 genes and 9,218,705 counts per cell for a total of 30 single neuron samples. Principal component analysis (PCA) of the gene profiles showed that virgin ERα^vlVMH subpopulation was clearly segregated from the two PPD5 subpopulations in the first two PCs (Fig. 7b). Additional PCA analysis further segregated PPD5 firing versus PPD5 silent subpopulations as two distinct neuron populations (Fig. 7c). However, PPT responsiveness was not captured by the differential gene profiles within each subpopulation with a third PC (Fig. 7c).

Fig. 7 |. Lactational remodulation of gene profiles suppresses the firing activity and E2 responsiveness of ERα^vlVMH neurons.

a, Schematic design for patch–seq sample collection. Each individual ERα^vlVMH neuron from ERα-ZsGreen virgin or lactating (PPD5) mice was recorded and then collected into a tube for scRNA-seq. ERα^vlVMH neurons from virgin mice were active (firing) and responsive to PPT (green). ERα^vlVMH neurons from PPD5 were regrouped based on their firing activity and responsiveness to PPT, resulting in firing subpopulations that were PPT-responsive or PPT-irresponsive (pink) and silent subpopulations that were PPT-responsive or PPT-irresponsive (grey). Created in BioRender.com. b,c, PCA of the gene profiles using PC1 versus PC2 (b) and PC2 versus PC3 (c). d, Volcano plot indicating the differentially expressed genes between virgin firing ERα^vlVMH neurons and PPD5 silent ERα^vlVMH subpopulation. The genes belonging to each category of neurotransmitter receptors, hormone receptors or ion channels are coded with each colour and indicated by solid circles. e, GO analysis indicates selectively enriched gene category in either PPD5 silent ERα^vlVMH subpopulation or virgin firing ERα^vlVMH neurons. f–h, Heatmap for genes that encode ion channels (f), neurotransmitter receptors (g) and hormone receptors (h) in virgin firing ERα^vlVMH neurons (green), PPD5 silent ERα^vlVMH subpopulation (grey) and PPD5 firing ERα^vlVMH subpopulation (pink). Genes labelled in red are expected to encode excitatory ion channels or neurotransmitter receptors, and those in blue are expected to be inhibitory. Bold genes in f encode ion channels that directly activate (red) or inhibit (blue) neuron activity.

These results suggest that lactation remodulated ERα^vlVMH neurons from virgin status into two distinct subpopulations, both with reduced PPT responsiveness. We compared the expression of ion channels with greater than twofold change between virgin firing and PPD5 silent ERα^vlVMH subpopulations, with a corrected P value of <0.001. Gene ontology (GO) analysis of the differentially expressed genes revealed involvement in the regulation of neuron activity in both virgin and PPD5. Interestingly, genes that regulate feeding behaviour were more enriched in virgin than in lactational ERα^vlVMH neurons (Fig. 7d,e and Extended Data Fig. 8b–e). In agreement with the reduced activity in the PPD5 ERα^vlVMH neurons, except Abcc9, all inhibitory ion channels were significantly increased in the silent ERα^vlVMH subpopulation, with a relatively modest change in the firing ERα^vlVMH subpopulation. Specifically, the potassium channels, primarily voltage-gated K⁺ channels (Kv channels, encoded by Kcnh3, Kcng1, Kcnv1, Kcnh7, Kcnf1, Kcns1, Kcna4, Kcng2, Kcng4, Kcnb2, Kcnq5 and Kcnb1), directly inhibit neuron activities (Fig. 7f). Conversely, a group of excitatory ion channels were decreased in the silent ERα^vlVMH subpopulation, with a relatively modest change in the firing ERα^vlVMH subpopulation. The sodium channels, primarily voltage-gated Na⁺ channels (encoded by Scn9a, Scn7a and Scn5a), can directly increase neuron activity. Other ion channels may regulate neuron activity in response to different stimulations; for example, transient receptor potential channels and voltage-gated calcium channels. The profile change of these inhibitory and excitatory ion channels may contribute to the silent or reduced activity of PPD5 ERα^vlVMH neurons. Interestingly, numerous excitatory ion channels, chiefly calcium channels (encoded by Cacna1s, Cacna1c and Cacna1d), were increased in ERα^vlVMH neurons (Fig. 7f).

We also compared the expression of neurotransmitter receptors with greater than twofold changes and a corrected P value of <0.001. Consistent with the silent or reduced activity of PPD5 ERα^vlVMH neurons, we observed increases of four inhibitory receptors, including two GABA receptor subunits (encoded by Gabrd and Gabra5) and two serotonin receptors (encoded by Htr1f and Htr5b), and decreases of six excitatory receptors, including two glutamate receptors (encoded by Grid2 and Grin2d) and two cholinergic receptors (encoded by Chrm5 and Chrna4) in PPD5 ERα^vlVMH neurons. However, we also observed decreases in a larger group of inhibitory (primarily GABA) receptors and increases in excitatory (primarily cholinergic) receptors as well as NMDA/AMPA receptors. These opposite changes in neurotransmitters, together with the increased excitatory ion channels, were potentially caused by a counterregulatory or feedback/compensation mechanism secondary to the reduced neuronal activities (Fig. 7g). We further found that PPD5 ERα^vlVMH neurons displayed profound changes of expression in a list of hormone receptors, with relatively milder changes in firing PPD5 ERα^vlVMH neurons (Fig. 7h). Interestingly, many of these hormone receptors are involved in the regulation of reproduction (for example, Crhr1) or metabolic balance (for example, Ttr1 and Cry). ERα was slightly decreased in both PPD5 ERα^vlVMH subpopulations, corresponding to the generally reduced PPT responsiveness of these neurons (Fig. 7h). Together, these results suggest that the pattern of neuron responses to different inhibitory and excitatory inputs, and different hormones, is significantly modified by lactation.

PCA data suggest that PPT responsiveness was suppressed in general ERα^vlVMH neurons, ranging from no responsiveness to less responsiveness, regardless of whether they were firing or silent. We further compared all the genes potentially involved in the downstream signal pathways that mediate E2–ERα signalling. In addition to ERα, Adcy7 and Parkar2a were also decreased in ERα^vlVMH neurons (Extended Data Fig. 8f).

We also compared our patch–seq data with published scRNA-seq data comparing transcriptomes of individual VMH neurons between virgin and lactating female mice⁶⁰. Consistent with our data, several inhibitory potassium ion channels, including Kcnf1, Kcnb1, Kcnv1 and Lrrc55, were increased in ERα^VMH neurons during lactation. Moreover, they were more enriched in lactating ERα^VMH neurons than in virgin ERα^VMH or lactating non-ERα^VMH neurons (Extended Data Fig. 8g). In addition, neurotransmitter receptors Gria3, Grin2b and Htr1f, and hormone receptors Cckar and Npr3, as well as ERα downstream mediator gene Adcy1 were changed in the same directions as shown in the patch–seq data (Supplementary Table 1). Except for Adcy7 and Grin2b, all these genes were more enriched in lactating ERα^VMH neurons than in lactating non-ERα^VMH neurons. These cross-validation data added to the confidence that these gene profile changes contribute together to the lactational modification of ERα^vlVMH neurons, and these modifications are more specific to ERα^vlVMH neurons. Interestingly, despite the reduced expression of ERα in each neuron (Fig. 7h), the percentage of ERα neurons in the whole VMH showed a slightly increased trend during lactation (Extended Data Fig. 8h).

One remaining question is how the gene profiles of ERα^vlVMH neurons are modulated. The suppression of ERα^vlVMH neurons starts from late pregnancy; therefore, it is reasonable that the gene profile should be modulated before delivery (that is, before E2 drops). We reanalysed published RNA-seq data comparing transcriptomes of the VMH between OVX-V and OVX-E mice¹⁸, and none of the patch–seq-identified inhibitory channels displayed any change between OVX-V and OVX-E mice (Extended Data Fig. 8i). By contrast, Kcng2 was even higher in the OVX-V mice. These results do not support the E2 decline during lactation as a drive for the lactational suppression of ERα^vlVMH neurons. Neither lactation⁶⁰ (Extended Data Fig. 9a,b) nor OVX-V¹⁸ (Extended Data Fig. 9c) changed the expression of PRLR in the VMH. Unexpectedly, our patch–seq data indicate that the expression of Prlr and Stat5b was mildly decreased in silent PPD5 ERα^vlVMH subpopulation but more significantly decreased in firing PPD5 ERα^vlVMH subpopulation, suggesting more complicated interactions between E2 and PRL signalling within ERα^vlVMH subpopulations. Progesterone and androgens display changes similar to E2 during pregnancy and lactation^1–3, and both Pgr and Ar were mildly reduced in lactating ERα^vlVMH neurons (Extended Data Fig. 9d). Therefore, we cannot rule out the impacts of PRL, progesterone or androgens during pregnancy and lactation on the gene profile changes of ERα^vlVMH neurons.

ERα^vlVMH neurons project to DA^ARH neurons to regulate PRL

Our results support the essential role of ERα^vlVMH neurons in PRL regulation. A previous retrograde study indicated that the anterior pituitary does not receive direct innervation from the VMH⁷⁴, excluding the direct regulatory role of ERα^vlVMH neurons. Given that DA can tonically inhibit PRL production^22–27 and ERα^vlVMH neurons project to the ARH^59,75, we tested whether ERα^vlVMH neurons project to DA^ARH neurons and regulate PRL through a DA signal. We injected the AAV-DIO-ChR2-EYFP virus into the vlVMH of ERα-Cre mice to express ChR2–EYFP specifically in a small population of ERα^vlVMH neurons. We found a clear track along the axons of ERα^vlVMH neurons and abundant fibres in the ARH (Fig. 8a), confirming the projection of ERα^vlVMH neurons to the ARH. Then, we injected an anterograde tracing H129-ΔTK-TT-dTomato virus into the vlVMH of ERα-Cre mice. A total of 31% of dTomato-labelled ARH neurons co-expressed tyrosine hydroxylase, a key enzyme for DA synthesis and a marker for DA neurons, in the ARH (Fig. 8b). Furthermore, we used DAT-Flpo mice to target a subset of DA neurons⁷⁶. We combined DAT-Flpo mice and a rabies retrograde tracing strategy to specifically express rabies-mCherry in the DA^ARH neurons, and 57.9% of mCherry-labelled vlVMH neurons co-expressed ERα (Fig. 8c). These results supported the monosynaptic projection of ERα^vlVMH → DA^ARH circuit. Importantly, tyrosine hydroxylase was significantly reduced in the ARH of ERα^MBH-KO virgin female mice (Fig. 8d,e). We treated both control and ERα^MBH-KO virgin female mice with DA receptor D2R agonist bromocriptine (daily i.p., 10 mg kg⁻¹) for 15 days. Both acute (0.5 h; Fig. 8f) and chronic (15 days; Fig. 8g) treatment of the D2R agonist significantly reduced serum PRL levels in ERα^MBH-KO virgin female mice without affecting the control mice. However, PRL levels bounced back 5 days after bromocriptine withdrawal (Fig. 8g). These results suggested that ERα^vlVMH neurons inhibit PRL, at least partially, via DA signals.

Fig. 8 |. ERα^vlVMH neurons project to DA^ARH neurons to regulate PRL levels.

a, ERα^vlVMH neuron fibres labelled by ChR2-GFP project to the ARH in ERα-Cre mice that received the injection of AAV-DIO-ChR2-GFP into the vlVMH. Yellow arrows indicate the fibre tract extending from the vlVMH to the ARH. b, Anterograde tracing strategy, labelling DA^ARH neurons with dTomato by injection of Cre-dependent H129-ΔTK-TT-dTomato virus into the vlVMH of an ERα-Cre mouse. Created in BioRender.com. DA neurons were labelled by immunostaining of the tyrosine hydroxylase (TH) antibody. The pie graph indicates the percentage of co-stained neurons out of all DA^ARH neurons. c, Retrograde tracing strategy, labelling ERα^vlVMH neurons with mCherry by injections of pAAV-EF1a-fDIO-Cre virus together with AAV-DIO-GTB virus followed by EnvA G-deleted rabies-mCherry virus into the ARH of DAT-Flpo mice. Created in BioRender.com. ERα neurons were labelled by immunostaining with ERα antibody. Pie graph indicates the percentage of co-stained neurons out of all retrograded vlVMH neurons. d, Immunofluorescence of TH in the ARH of virgin control and ERα^MBH-KO mice. e, Quantification of TH signal in the ARH of virgin control (n = 6) and ERα^MBH-KO mice (n = 7). f, Serum PRL levels in virgin control (n = 4) and ERα^MBH-KO (n = 6) mice before and 0.5 h after the injection of bromocriptine. g, Serum PRL levels in virgin control and ERα^MBH-KO mice that received daily injection of bromocriptine for 2 weeks followed by withdrawal of treatment for 5 days (control, n = 4; ERα^MBH-KO, n = 6). Data are presented as mean ± s.e.m. and/or individual data points. In e, two-tailed unpaired t-tests; f and g, two-tailed paired t-tests. P values are shown.

Discussion

We identified the reduction of E2–ERα^vlVMH signalling during lactation as an important contributor to lactational hyperprolactinemia and metabolic adaptations including hyperphagia and suppressed BAT thermogenesis. Importantly, we revealed an inhibitory effect of E2 on PRL levels via ERα^vlVMH neurons, partially mediated by DA^ARH neurons. This inhibitory effect needs to be suppressed during lactation to maintain high PRL levels, which may synchronically contribute to lactational hyperphagia and suppression of BAT thermogenesis (Extended Data Fig. 9e).

E2 acts on ERα^MBH neurons to suppress feeding and activate thermogenesis in virgin female mice^17–21. Despite a modest drop in serum E2 level in lactating mice, we found that ERα^vlVMH neurons are suppressed not only in neuronal activity but also in their responsiveness to E2. Importantly, loss of ERα from the MBH in virgin female mice increased feeding and suppressed BAT thermogenesis, whereas activation of ERα^vlVMH neurons in lactating dams did the opposite. These results together support that the suppression of E2–ERα^vlVMH signalling is required to maintain lactational metabolic adaptations.

Our patch–seq data support that the suppression of ERα^vlVMH neuron activities results from a combination of increases in numerous inhibitory potassium channels and decreases in several excitatory sodium channels. In addition, the profile of receptors for different neurotransmitters and hormones is also altered during lactation, suggesting remodelled responsiveness of ERα^vlVMH neurons to the presynaptic inputs and hormone stimulations. The suppression of E2–ERα^vlVMH signalling starts from late pregnancy but is gradually enhanced during lactation and eventually persists through the entire lactation period. Late-pregnancy animals also display hyperphagia and suppressed BAT function⁷⁷, but with high E2 levels. We speculate that the suppression of E2–ERα^vlVMH signalling contributes more to metabolic adaptations during lactation than late pregnancy.

Our results support an inhibitory effect of E2 on PRL homeostasis via central ERα^vlVMH neurons, opposite to the excitatory effects of E2 on PRL via the pituitary^40,41 and the ARH^37–39. ERα is also expressed in TIDA neurons⁵⁰. Acute supplementation of E2 to OVX rats decreases the expression and activity of tyrosine hydroxylase^52,53, and chronic E2 treatment reduces DA synthesis and activity of TIDA neurons^54–57. The negative association between E2 signalling and DA favours an excitatory effect of E2 on PRL production via TIDA neurons, which needs to be confirmed by direct evidence. Even if E2–ERα signalling in TIDA neurons promotes PRL levels, this excitatory effect should be minor and masked by the inhibitory effect during lactation, as central E2 supplementation decreases PRL levels, potentially via ERα neurons in other brain regions (like the vlVMH), during lactation. ERα is also highly expressed in other brain regions, including POA and PVN⁶⁵. It would be worth exploring whether the inhibitory effect of E2 via central actions on PRL regulation is also mediated by these ERα neuron populations.

Contrary to the excitatory E2 effects on PRL, we showed that central E2 supplementation reduces PRL levels in lactating dams but not in OVX mice. Three decades ago, two reports showed that E2 supplement reduces PRL levels in OVX rats in the condition of pseudopregnancy^47,48. These results argue that the inhibitory effect of E2 on PRL may only be triggered under certain physiological conditions, like pseudopregnancy or lactation, to counterregulate high PRL levels. E2 supplemented by s.c. implantation enters the circulation and then reaches the brain⁷⁸. Thus, the net E2 effect on PRL can result from the balance between the excitatory effect via the pituitary or the ARH versus the inhibitory effect via ERα^vlVMH neurons. The central oestrogenic effect on PRL is excitatory in virgin females but turns inhibitory during lactation, which counteracts the excitatory effects in lactating dams with systemic E2 supplements. The high PRL levels in ERα^MBH-KO and ERα^vlVMH-KO mice suggest that the inhibitory effect of ERα^vlVMH neurons on PRL does exist in cycling female mice, but it is not known when (for example, during the proestrus surge) this mechanism is turned on. This bidirectional regulatory mechanism is essential to maintain PRL within a narrow physiological range, which may have roles in different female physiological stages with dynamic PRL levels, including puberty, oestrous cycles, pregnancy and lactation.

ERα^vlVMH neurons project to many brain regions, including the POA, the lateral hypothalamus, the PVN and the ARH, which are all important for feeding and energy homeostasis^59,75. The PVN contains oxytocin neurons that can increase PRL levels³⁵, the lateral hypothalamus contains orexin neurons that inhibit PRL levels⁷⁹ and the ARH contains TIDA neurons that mediate the typical negative feedback of PRL²². The inhibitory role of E2–ERα^vlVMH signalling on PRL levels resembles a negative feedback mechanism. Mice with deletion of either D2R³⁰ or PRLR^25,31 develop hyperprolactinemia, which is recapitulated in virgin female ERα^vlVMH-KO mice. Importantly, this PRL–PRLR–DA signalling pathway is modulated during late pregnancy by switching DA to enkephalin, and as a result, the negative feedback is switched to positive feedback²⁷. This phase switch of TIDA neurons probably depends on basal PRL levels. Our study supports the ERα^vlVMH → DA^ARH circuit. However, it remains to be investigated how this circuit regulates PRL levels during lactation and how it is impacted by the switch between DA and enkephalin.

Our results indicate that the DA antagonist bromocriptine only partially attenuated but did not normalize PRL levels in ERα^MBH-KO mice. Notably, ERα neurons in the VMH highly co-express PRLR⁴⁹. It is also possible that PRL acts on the PRLR in the ERα^vlVMH neurons to reduce PRL levels, representing a DA-independent negative feedback. This mechanism may require E2–ERα signalling, as E2 can increase the induction of pSTAT5 in the VMH⁴⁹. Despite unchanged expression of PRLR in the whole VMH by E2 or location, the PRL–PRLR–STAT5 signalling in ERα^vlVMH neurons was reduced by lactation but enhanced in the silent lactating ERα^vlVMH neuron subpopulation. However, a recent study reported that female mice with deletion of PRLR in the VMH are still fertile without any delay in the latency of pregnancy or maternal behaviours⁶¹. If this PRLR-mediated DA-independent negative feedback does exist in ERα^vlVMH neurons, loss of PRLR from the VMH should increase PRL level to reduce fertility unless high PRL has to act on PRLR in the VMH to suppress fertility. It is worthwhile to test the possibility of this DA-independent negative feedback mechanism via ERα^vlVMH neurons.

Interestingly, hyperprolactinemia-associated phenotypes, including obesity and infertility, have been reported in many different mouse models with loss of ERα from the VMH, but high PRL has never been reported^{17–20,62,63,72}. Before the measurement of PRL levels or terminal prolactinoma, the only sign of hyperprolactinemia is protruding nipples, which is too subtle to be noticed. In our studies, various sizes of prolactinomas were developed either under a HFD for 2 months following 2 months of a chow diet, or by prolonged impact by deletion of ERα for more than 5 months on a chow diet. However, previous ERα^vlVMH-KO mouse models using a short hairpin RNA or AAV-Cre strategy are either terminated 4 days to 3 months on a chow diet post surgery^{18–20,62,72} or together with OVX surgery^20,63, which results in low PRL levels caused by the loss of the excitatory effect of E2. The embryonic deletion of ERα from SF-1 neurons in the VMH¹⁷ only targets half the population of the ERα^vlVMH neurons, and the PRL regulatory effect may also be attenuated by developmental compensation or ERα deletion from SF-1 cells in other regions.

Lactational hyperphagia can be attenuated by blocking PRL production⁵ and be rescued by central administration of PRL⁸⁰, supporting a central mechanism of lactational hyperprolactinemia on hyperphagia. It is plausible that the increase in food intake in virgin female ERα^MBH-KO mice is secondary to hyperprolactinemia. Echoing this possibility, the PRL–PRLR downstream signalling pSTAT5 is significantly increased in the ARH and the PVH of hyperprolactinemic ERα^vlVMH-KO mice. It is worth examining whether the decrease in E2–ERα^vlVMH signalling maintains lactational hyperphagia through excessive PRL–PRLR signalling in the hypothalamus.

Central action of E2 activates BAT thermogenesis via ERα^vlVMH neurons^19,21. Consistently, ERα^MBH-KO mice display suppressed BAT function including decreased UCP1 in the BAT¹⁸ and lower BAT temperature. BAT thermogenesis can be suppressed by PRL and be rescued by inhibiting PRL production in virgin female rats⁶. Consistently, patients with hyperprolactinemia display metabolic syndromes including obesity⁸¹, and PRLR-KO mice display increased energy expenditure and enhanced BAT function⁸². However, our current study could not distinguish whether suppressed BAT thermogenesis in virgin ERα^MBH-KO mice and/or lactating female mice is mediated by the suppression of ERα^vlVMH neurons per se independent of PRL or by the excessive PRL–PRLR signalling in the BAT.

It is surprising that central E2 supplement reduces PRL levels but does not reduce the food intake of dams, while DREADD activation of ERα^vlVMH neurons does both. One explanation is that because of the reduced E2 responsiveness of ERα^vlVMH neurons during lactation, central E2 supplementation is strong enough to increase the ERα^vlVMH neuron activity required for inhibiting PRL but not strong enough to inhibit feeding behaviour. However, ERα^vlVMH neurons in the ERα^vlVMH-hM3Dq dams are artificially forced to fire despite lower E2 responsiveness. It would be interesting to tease out the difference in sensitivity between PRL control and feeding by the same ERα^vlVMH neurons. Another possibility is that chronic continuous brain infusion of E2 reduced PRL to a constant level, and this stable status may give lactating dams a chance to develop compensatory mechanisms through other ways to maintain feeding. By contrast, ERα^vlVMH-hM3Dq dams experience PRL collapse twice a day with more profound PRL drops than the central E2 supplement effect. As a result, no compensatory mechanisms can develop with this very dynamic change in PRL daily.

It has been reported that deletion of ERα from proopiomelanocortin neurons in the ARH inhibits oestrogen negative feedback, resulting in increases of serum E2 and progesterone levels¹⁷. The increased E2 may be the cause of the mild increase of PRL levels in female mice with ERα deleted from the ARH⁷². Despite that hyperprolactinemia inhibits the secretion of E2, progesterone, luteinizing hormone and follicle-stimulating hormone^83,84, these hormones are not significantly changed in ERα^MBH-KO mice. Therefore, ERα^vlVMH neurons, unlike ERα^ARH neurons, are not involved in an E2 negative feedback loop, and the hyperprolactinemia in female ERα^MBH-KO mice is not a result of the excitatory effect of E2 on PRL^40,41 or the inhibitory effect of progesterone on PRL⁸⁵. Patients with hyperprolactinemia and ERα^MBH-KO mice have similar reproductive symptoms⁸³. Although the unchanged gonadal hormone levels cannot explain the disrupted reproductive function in ERα^MBH-KO female mice, we cannot exclude the possibility that the dynamic surges of these hormones are changed and contribute to the infertility phenotype. Other possibilities include PRL-dependent effects on the ovary and uterus⁸⁶ or effects of E2–ERα^vlVMH signalling on social behaviours^59,60,87 that may affect reproductive behaviours potentially independent of PRL.

As discussed, we do not have direct evidence to support whether ERα^vlVMH neurons regulate food intake and BAT thermogenesis by regulating PRL levels. Future studies are warranted to determine whether and where PRL acts to regulate food intake (PRLR in certain brain regions) and BAT thermogenesis (PRLR in the BAT), and whether hyperprolactinemia mediates the metabolic effects in virgin ERα^vlVMH-KO mice. Questions remain regarding when and how the gene profiles of ERα^vlVMH neurons are modulated, and whether the modulation of ERα^vlVMH neurons is contributed by changes in E2, PRL, progesterone or even androgen.

We acknowledge that DAT mice used in this study can only target a subset of DA neurons⁷⁶. Future studies are needed to determine how ERα^vlVMH neurons regulate PRL via DA neurons or DA-independent mechanisms along with when and how the inhibitory effect of E2–ERα^vlVMH on PRL is turned on in virgin female mice.

Methods

Mice

The care of animals and all procedures were approved by the Institutional Animal Care and Use Committee of Baylor College of Medicine and Pennington Biomedical Research Center. Mice were housed in a temperature-controlled room at 22–24°C and ~30–70% humidity using a 12 h light, 12 h dark cycle. The virgin mice were fed regular chow (5V5R, Select Rodent 50 IF/6F, PicoLab; kcal%: 21% protein, 15% fat, 64% carbohydrates) or a HFD (60 kcal% fat, D12492i, Research Diets). The lactating mice were fed a breeding diet (5V5M, Select Mouse 50 IF/9F, PicoLab; kcal%: 20% protein, 22% fat, 58% carbohydrates). Water was provided ad libitum.

Deletion of ERα from MBH neurons

Female ERα^fl/fl mice⁸⁸ (10 weeks old) were anaesthetised by isoflurane and received bilateral stereotaxic injections of AAV8-Cre-GFP or AAV8-GFP virus (200 nl, 5 × 10¹² GC per ml, UNC Gene Therapy Center) into the vlVMH (ML ±0.75 mm, AP −1.60 mm, DV −5.85 mm).

In the first cohort of female mice, body weight and food intake were measured every 4 days. Serum samples from the tail were collected 2, 5 and 6 weeks after surgery to measure serum hormone levels. At 5 weeks after virus injection, these female mice were paired with experienced WT C57BL/6J male breeders for 1 week, then female mice were separated for the measurement of body weight and food intake. If a female mouse with continuously increasing body weight experienced a sudden drop in body weight and/or obvious blood traces, they were recorded as having a spontaneous miscarriage. Then, 3 weeks after the weaning of pups, these mice received daily i.p. injections of vehicle or bromocriptine (10 mg kg⁻¹) for 15 days continuously. Tail serum samples were collected before and 0.5 h after the injection on days 1, 8 and 15 of treatment and 5 days after the last injection for PRL measurement. A cohort of WT control female mice received the same virus injections and experienced the same procedures to exclude the side effects of these viruses.

A second cohort of mice was put into metabolic cages (TSE PhenoMaster) for 10 days to measure energy expenditure 2 weeks after surgery, before a change in food intake. Energy expenditure was analysed using CalR analysis^66,70 with body weight as a covariate. Predicted energy expenditure was calculated with the presumed body weight at 27 g.

A third cohort of mice was fed with a chow diet for 2 months and then switched to HFD for another 2 months to boost the reproductive and metabolic phenotypes. Then, 1 month after HFD, vaginal cells were flushed and observed under a light microscope to estimate oestrous phases based on the proportion of leucocytes, cornified epithelial cells and nucleated epithelial cells. Terminal serum samples were collected to measure a series of reproductive and metabolic hormones. Then all the mice were perfused with 10% formalin and brains were dissected. The iWAT tissues containing the fourth mammary glands were dissected and stained as whole mounts as previously described⁸⁹. Pituitaries were processed for H&E staining and immunohistochemistry staining.

A fourth cohort of female mice was implanted with a telemetry probe (IPTT-300, BioMedic Data Systems) underneath the skin above the interscapular BAT to measure BAT temperature 4 weeks after surgery. Terminal iWAT and BAT were dissected for H&E staining.

A fifth cohort of female mice only received 50 nl of virus to one side of the vlVMH, such that the virus infection would be restricted to only partial vlVMH. Mice with virus that spilled out of the vlVMH were excluded from the analysis. Food intake and body weight were monitored every 4 days. Serum samples were collected 6 and 12 weeks after surgery to measure serum PRL levels.

DREADD manipulation of ERα^vlVMH neurons

ERα-Cre transgenic mice⁵⁸ and their WT control littermates (10 weeks old) were anaesthetised by isoflurane and received bilateral stereotaxic injections of AAV-DIO-hM3Dq(Gq)-mCherry virus (200 nl, 5 × 10¹² GC per ml; Addgene, no. 44361-AAV8) into the vlVMH (ML ±0.75 mm, AP −1.60 mm, DV −5.85 mm) to express hM3Dq protein specifically in ERα^vlVMH neurons (ERα^vlVMH-hM3Dq). The mice were also implanted with a temperature probe (IPTT-300, BioMedic Data Systems) underneath the skin above the BAT to monitor BAT temperature. The mice were fed with a chow diet and recovered for at least 2 weeks before any other procedures.

ERα^vlVMH-hM3Dq and control mice were paired with reproductively experienced WT male mice to generate lactating dams. The delivery day was defined as PPD0. The lactating dams received daily i.p. injections of CNO (3 mg kg⁻¹, Cayman Chemical; twice a day at 09:00 h and 17:00 h) from PPD1 to PPD11. On PPD5, dams and pups were separated at 08:00 h and CNO was injected at 09:00 h. Serum samples were collected before and 2 h after the CNO injection. To confirm lactational hyperprolactinemia, serum samples were collected from a batch of control mice before breeding and on PPD5 (before CNO injection) for the measurements of PRL levels. BAT temperature was measured before and 1 h after the CNO injection. After serum collection, lactating dams were sent back to the home cages with their pups. After 1 h of habituation, maternal behaviours were tested by placing three random biological pups (from each tested dam) in one of three corners away from the nest, then introducing their dams in the nest area in the home cage. Maternal behaviour was recorded for 15 min and analysed using the following criteria: retrieval time (time spent moving all pups to the nest site), crouching time (time spent in a nursing-like posture over pups), grooming time (time spent sniffing and licking a pup), nest-building time (time spent collecting and arranging nesting material and making a nest) and time in the nest. The time that lactating dams spent on grooming, crouching or nest-building was reported as ‘maternal care’⁹⁰.

Each day, we also recorded the body weight and food intake of dams and the body weight and number of pups. Dams that lost all pups were excluded from PRL or food intake analysis as indicated. At the end of each experiment, all mice were perfused to post-hoc validate the injection accuracy. iWAT and BAT were dissected for H&E staining.

Systemic or central supplement of E2 to lactating dams

WT C57BL/6J female mice (10 weeks old) were paired with reproductively experienced WT male mice to generate lactating dams. Male and female mice were put together before the dark cycle, and female mice were checked for the presence of a vaginal plug early the following morning. Female mice with a vaginal plug (gestation day 0, G0) were singly caged. On PPD1, these lactating dams received s.c. implantations of pellets containing E2 (0.5 μg d⁻¹ or 2.4 μg d⁻¹ lasting for 21 days) or empty placebo pellets (V). These pellets were purchased from Innovative Research of America (E-121 for E2 and C-111 for V).

WT female mice (10 weeks old) were implanted with an intracerebroventricular cannula (ALZET Brain Infusion Kit 3) attached to an osmotic pump (no. 0009922, Model-1004, ALZET) (s.c.) filled with artificial cerebrospinal fluid (ACSF; 0.11 μl h⁻¹). After 1 week of recovery, these mice were paired with experienced WT male mice before the dark cycle, and the vaginal plug was checked early the next morning. On G16, the osmotic pump with ACSF was replaced with a new osmotic pump filled with E2 or vehicle. We estimate that it would take 4 days to clear the remaining ACSF in the connecting tube. Therefore, E2 (0.11 μl h⁻¹, 0.1 μg d⁻¹; E8875, Sigma-Aldrich)^91,92 or vehicle (10% dimethylsulfoxide in ACSF) would begin on PPD1 and continue for 28 days. Another cohort of virgin female WT mice were implanted with an intracerebroventricular cannula attached to an osmotic pump filled with E2 or vehicle. These female mice all received OVX surgery at the same time, and serum samples were collected 1 week after surgery.

Body weight and food intake of the dams, and body weight and the number of pups were measured daily between PPD0 and PPD11. On PPD5, the pups were separated from the dams for 2 h and then serum samples were collected from the tail blood of lactating dams to measure serum PRL levels.

To measure serum E2 levels, terminal serum samples were collected from a cohort of WT mice without any treatments. These mice included late-pregnancy mice during peri-parturition around G18–G21 and lactating dams on PPD2, PPD5 and PPD9 as well as age-matched dioestrus virgin female mice. The weights of BAT from virgin mice and lactating dams on PP11 were recorded, and then BAT tissues were processed for western blot with anti-UCP1 antibody (ab10983, Abcam, 1:1,000). iWAT containing the fourth mammary glands were processed for H&E staining.

Immunohistochemistry, immunofluorescence and H&E staining

The mice brains were processed for frozen sectioning at 25 μm and collected into five separate series. One series of brain sections was subjected to post-hoc histological validation of virus injection accuracy. Two more series of brain sections were processed for immunohistochemistry or immunofluorescence staining with ERα (06–935, Millipore, 1:5,000), c-fos (226008, Synaptic Systems, 1:1,000), tyrosine hydroxylase (AB152, Millipore, 1:1,000), GFP (AB13970, Millipore, 1:2,000) and pSTAT5 antibodies (9359S, Cell Signalling Technology, 1:2,000). Tyrosine hydroxylase signals were quantified in the neuron bodies and fibres in the ARH using ImageJ (v.1.54g) software. Formalin-fixed pituitary, iWAT and BAT tissues were sent to the Comparative Pathology Laboratory at Baylor College of Medicine, where the tissues were embedded in paraffin, sectioned and subjected to H&E staining. Pituitary sections were subjected to immunofluorescence staining with antibody against PRL (A1618, Abclonal, 1:1,000).

Measurements of serum hormone levels

Growth hormone, insulin-like growth factor 1, adrenocorticotropic hormone, thyroid-stimulating hormone, follicle-stimulating hormone and luteinizing hormone were measured using ELISA kits from Abclonal; PRL was measured using an ELISA kit from Life Technologies. Serum E2 and progesterone were measured using liquid chromatography hyphenated with mass spectrometry in the Metabolomics Core at Baylor College of Medicine.

Electrophysiological recordings

ERα-ZsGreen virgin female mice were used to generate lactating dams as described above. Approximately 12–16-week-old ERα-ZsGreen virgin (in dioestrus), pregnant (G16–G18) or lactating female mice (PPD1, PPD2, PPD5, PPD10, PPD20) were used for electrophysiological recordings using a protocol described previously⁹³. In brief, the mice were transcardially perfused and the entire brains were sectioned with a Leica VT 1200 s (Leica Biosystems). Three brain slices containing the ARH and vlVMH were obtained, and ZsGreen-labelled neurons were visualized and patched for recording. To examine the acute neural responses to PPT, a selective ERα agonist⁶⁶, a current clamp was engaged to measure the neural firing rate before and after a 1 s puff of 100 nM PPT. In another experiment, ERα^vlVMH neurons expressing hM3Dq protein specifically in the vlVMH region (ERα^vlVMH-hM3Dq) were recorded under a current clamp. ERα^vlVMH-hM3Dq neurons were perfused by CNO (5 μM) for 2–4 min. Firing rate and membrane potential were analysed before and post CNO treatment²¹.

Patch–seq and data analysis

The scRNA-seq post-electrophysiology experiment was conducted as described previously⁹³. In brief, 11-week-old virgin female ERα-ZsGreen mice (n = 3) in dioestrus and PPD5 lactating ERα-ZsGreen dams (n = 3) were used for sample collections. We collected the cellular component of each ERα^vlVMH neuron after the recording. The samples were then transferred into a PCR tube. Samples were composed of one neuron each. cDNA libraries were constructed using the Lexogen QuantSeq 3′ mRNA-Seq V2 Library Prep Kit FWD with UDI, following a 5 min room temperature lysis using 0.2% Triton and RNAase inhibitor. Library integrity was confirmed using Agilent BioAnalyzer High Sensitivity DNA Assay. Libraries comprised ~275 bp and were pooled in equimolar amounts and sequenced on an Illumina NextSeq 1000 at 75 bp. Analyses and data visualization were performed in R (v.4.3.3) (R Core Team 2024) within RStudio (v.2023.09.0 Build 463) (RStudio Team 2023). Principal components were calculated using the ‘prcomp’ function from the stats package (v.4.3.3) and were visualized using the ‘autoplot’ function in ggplot2 (v.3.4.4). Differentially expressed genes were determined using DESeq2 (v.1.39.8). Volcano plots were made using ggplot2 and labelled using ggrepel (v.0.9.4). GO was performed using differentially expressed genes and gprofiler2 (v.0.2.1). Ontology terms were manually curated and visualized using ggplot2. Gene lists were downloaded from https://geneontology.org for GO:0034702 monoatomic ion channel complex (ion channels), GO:0098878 neurotransmitter receptor complex (neurotransmitter receptor) and GO:0051427 hormone receptor binding (hormone receptor). Lists were then used to create heatmaps after normalizing gene expression against the average expression of any particular gene across all conditions. Ion channels were filtered and classified as either excitatory or inhibitory by an expert neuroscientist and electrophysiologist, respectively. Ion channels appearing on the neurotransmitter receptor list were removed to reduce redundancy. Finally, heatmaps were visualized using ‘geom_tile’ from ggplot2.

Secondary analysis of RNA-seq data

scRNA-seq data were downloaded from the Gene Expression Omnibus (GSE193921 (ref. 60) and GSE143818 (ref. 67)). Data were preprocessed according to the methods published with the original data. In brief, cells with <600 detected genes, >50,000 unique molecular identifiers or >15% unique molecular identifiers attributed to mitochondria mRNA were discarded. Data were further filtered for possible doublets by discarding cells that co-expressed marker genes from two different cell types: Stmn2 for neurons; Cldn5 for endothelial cells; C1qc for microglia; Opalin for oligodendrocytes; Gja1 for astrocytes; Pdgfra for OPCs; and Mustn1 for mural cells. VMH neurons were identified by keeping only cells that expressed one of the following marker genes: Slc17a6, Adcyap1, Fezf1, Dlk1 or Nr5a1. Finally, cells from the control mice without intruders in the conditions of ‘lactating’ and ‘virgin’ were kept for downstream analyses. Analyses and data visualization were performed in R (v.4.3.3) (R Core Team 2024) within RStudio (v.2023.09.0 Build 463) (RStudio Team 2023). Data were normalized using the ‘SCTransform’ function in Seurat (v.4.3.0.90). Differential expressions were calculated using the ‘FindMarkers’ function.

RNA-seq data for the gene profiles of VMH between OVX-V and OVX-E mice were downloaded from the Gene Expression Omnibus (GSE181204 (ref. 18)). Sequencing counts of the interested genes in each mouse were extracted and normalized to housekeeping gene Hprt. The genes of inhibitory channels were further normalized against the average expression of each particular gene in all samples including OVX-V and OVX-E mice. The normalized gene expression data were used to create a heatmap.

Neurotracing studies

For anterograde tracing, female ERα-Cre mice were anaesthetised by isoflurane and received stereotaxic injections of AAV-DIO-ChR2-EYFP (100 nl, 4 × 10¹² GC per ml, UNC Gene Therapy Center) or H129-ΔTK-TT-dTomato (50 nl, 7.06 × 10⁸ GC per ml, The Center for Neuroanatomy with Neurotropic Viruses) into the right side of the vlVMH (ML 0.75 mm, AP −1.60 mm, DV −5.85 mm). The mice brains were dissected and processed for immunofluorescence staining of GFP or tyrosine hydroxylase 1 week (for ChR2) or 2 days (for H129) after surgery.

For retrograde tracing, DAT-Flpo transgenic mice were generated previously to target a subset of DA neurons⁷⁶. Female DAT-Flpo mice were anaesthetised by isoflurane and received stereotaxic injections of pAAV-EF1a-fDIO-Cre (AAV8) (100 nl, 1.05 × 10¹³ GC per ml, Addgene, no. 121675) and AAV-DIO-GTB (100 nl, 7.51 × 10¹² GC per ml, GT3 Core Facility of the Salk Institute) into the right side of the ARH (ML 0.25 mm, AP −1.45 mm, DV −5.9 mm). Then, 3 weeks later, the same mice received a stereotaxic injection of EnvA G-deleted rabies-mCherry (200 nl, 1.78 × 10⁹ GC per ml, GT3 Core Facility of the Salk Institute) into the same site of the ARH. The mice brains were dissected and processed for immunofluorescence staining of ERα 1 week after the last injection.

Statistics and reproducibility

Experiments were replicated at least three times successfully. Data, excluding RNA-seq analyses, were tested for normal distribution using the Kolmogorov–Smirnov normality test with Dallal–Wilkinson–Lillie for P values, and statistical tests were performed using GraphPad Prism software (v.9.4.0). Sample sizes represent biological replicates and are similar to those in our previous publications^66,94; however, no specific statistical calculation was used to predetermine sample size. The data are presented as mean ± s.e.m. and/or individual points. Methods of statistical analyses were chosen based on the design of each experiment and are indicated in figure legends. P values are shown in the figure panels. For the Cre-deletion injection, mice were drawn at random from littermates to receive functional or control virus injections. For virgin and lactating mice, all WT mice were littermates that were selected randomly and assigned to breeding or virgin groups. For supplementation with E2, virgin or lactating mice received E2 or control at random. For chemogenetic manipulation, mice were drawn at random from littermates with a roughly equal mix of Cre and non-Cre, and both mice were injected with the same Cre-dependent virus, thus obtaining control and experimental groups dependent on genotypes. Data collection and analysis were not performed blind to the conditions of the experiments. Some data points (for example, dams who lost pups) were excluded from certain analyses, including PRL and BAT temperature measurements, as indicated.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Extended Data

Extended Data Fig. 1 |. Lactational hyperphagia and hyperprolactinemia.

(a) Serum levels of PRL in virgin and lactating female mice (n = 6). (b-e) The correlation between the number of pups and food intake (b), PRL levels (c), or BAT size (d), or UCP1 (e) in the BAT in lactating dams (b, n = 44; c, n = 7; d, n = 5; e, n = 5). Data are presented as mean ± SEM and/or individual data points. a, two-tailed unpaired t-tests. b-e, linear regression with 95% confidence. P values are shown in figures.

Extended Data Fig. 2 |. Lactational suppression of ERα^MBH neurons.

(a) Spontaneous firing frequency and resting membrane potential of ERα^vlVMH neurons in virgin (n = 37), pregnancy (gestation day 16–18, n = 18) and lactating mice on PPD1 (n = 37), PPD 5 (n = 31), PPD 10 (n = 27) and PPD 20 (n = 22). Neurons from 3 mice in each group. (b-e) Individual ERα^vlVMH neurons (b), ERα^ARH neurons (c), ERα^POA neurons (d), ERα^PVN neurons (e), responses to PPT as indicated by the increased firing frequency and resting membrane potential in virgin, late pregnancy (G18) and lactating mice on PPD2 and 5. Fractions indicate the responsive neuron number out of total recorded neurons. Neurons from 3 mice in each group. (f) Percentage of spontaneous firing/PPT-responsive, firing/PPT-irresponsive, silent/PPT-responsive and silent/PPT-irresponsive ERα^vlVMH subpopulations from virgin, gestation day 18 and PPD5 mice. (g) Representative traces of ERα^vlVMH neurons excited by PPT in virgin and lactating female mice. (h) The percentage of ERα, ERβ, and GPER neurons in total VMH neurons from female mice indicated by secondary analysis of scRNA-Seq data^60,67. (i) Serum levels of E2 in females during peri-parturition and lactation (peri-parturition, n = 5; PPD2, n = 4; PPD5, n = 4; PPD9, n = 4; virgin, n = 3). Data are presented as mean ± SEM and/or individual data points (a-e and i), or as part of whole bar (f) or Pie graphs (h). a, groups vs Virgin in ordinary one-way ANOVA followed by Dunnett’s multiple comparisons test. i, ordinary one-way ANOVA followed by Šídák’s multiple comparisons test. b-e, “n” is indicated in figures. P values are shown in figures.

Extended Data Fig. 3 |. Deletion of ERα from MBH neurons causes hyperprolactinemia and metabolic changes.

(a) Immunostaining of ERα in virgin female control and ERα^MBH-KO mice with deletion of ERα from the entire vlVMH but not the ARH. (b–c) Body size (b) and food intake (c) of virgin female control (n = 6) and ERα^MBH-KO mice (n = 7). (d) Food efficiency (body weight gain/accumulated food intake) of virgin female control and ERα^MBH-KO mice by 16 days after virus injection (n = 6). (e-f) Body weight (e) and food intake (f) in virgin female control and ERα^MBH-KO mice before they entered the TSE metabolic cages, 2 weeks after virus injection (control, n = 8; ERα^MBH-KO, n = 7). (g-i) Temporal levels of O₂ consumption during the dark cycle and light cycle within 3 days (g), ANCOVA analysis of O₂ consumption body weight as a covariant^66,70 (h) and predicted O₂ consumption at a presumed body weight (i) in virgin female control and ERα^MBH-KO mice measured by the TSE PhenoMaster metabolic cages (control, n = 8; ERα^MBH-KO, n = 6). (j-l) Temporal levels of CO₂ consumption during the dark cycle and light cycle within 3 days (j), ANCOVA analysis of CO₂ consumption body weight as a covariant (k) and predicted CO₂ consumption at a presumed body weight (l) in virgin female control and ERα^MBH-KO mice measured by the TSE PhenoMaster metabolic cages^66,70 (control, n = 8; ERα^MBH-KO, n = 6). (m-o) Serum PRL levels (m), body weight (n) and food intake (o) of control WT mice receiving AAV-Cre or AAV-GFP virus into the vlVMH (n = 5). Data are presented as mean ± SEM and/or individual data points. d, e, i and l, two-tailed unpaired t-tests. h and k, ANCOVA analysis by CalR. P values are shown in figures.

Extended Data Fig. 4 |. Deletion of ERα from MBH neurons causes hyperprolactinemia and neuroendocrinology changes.

(a) Immunohistochemistry staining of pSTAT5 in the hypothalamus of virgin female control and ERα^MBH-KO mice. 3 V, third ventricle. (b) H&E staining of the pituitary and immunofluorescence of PRL in the pituitary of virgin female control and ERα^MBH-KO mice. The scale bar is 200 μm. (c) Boxed areas of (b) are magnified as indicated. The scale bar is 50 μm. (d) Immunohistochemistry staining of ERα in the hypothalamus of virgin female control and ERα^MBH-KO mice 4 months (2 months chow diet followed by 2 months HFD) after surgery. The scale bar is 200 μm. (e-h) Serum levels of GH (e), IGF1 (f), ACTH (g) and TSH (h) in virgin female control (n = 6) and ERα^MBH-KO (e, n = 8; f, g, h, n = 7) mice 4 months (2 months chow diet followed by 2 months HFD) after surgery. (i) Serum levels of TSH in virgin female control and ERα^MBH-KO mice on chow diet 6 weeks after surgery (n = 8). (j) H&E staining of iWAT in virgin female control and ERα^MBH-KO mice by 4 months after surgery. The scale bar is 500 or 100 μm. (k-m) Serum levels of P4 (k), LH (l) and FSH (m) in virgin female control (k, n = 4; l, n = 6; m, n = 6) and ERα^MBH-KO (k, n = 6; l, n = 7; m, n = 7) mice 4 months (2 months chow diet followed by 2 months HFD) after surgery. Data are presented as mean ± SEM and/or individual data points. h, two-tailed unpaired t-tests. P value is shown in the figure.

Extended Data Fig. 5 |. Deletion of ERα from vlVMH neurons increases PRL levels.

(a) Immunofluorescence staining of ERα in the hypothalamus of control side and ¼ ERα^vlVMH-KO side from one virgin female mouse. The scale bar is 200 μm. (b) Quantification of ERα deletion in the half vlVMH KO side compared to the control side (n = 5). (c) The appearance of mammary glands in female virgin control and virgin ¼ ERα^vlVMH-KO mice. (d) Body weight at 12 weeks after virus injection and (e) average of 4 days-food intake of virgin control and ¼ ERα^vlVMH-KO mice by 12 weeks after virus injection (n = 5). (f) Serum PRL level of virgin control and ¼ ERα^vlVMH-KO mice at 6 weeks and 12 weeks after virus injection (n = 5). Data are presented as mean ± SEM and individual data points. f, ordinary one-way ANOVA followed by Dunnett’s multiple comparisons test. P values are shown in the figure.

Extended Data Fig. 6 |. Activation of ERα^vlVMH neurons inhibits lactational hyperprolactinemia and hyperphagia.

(a) Bright-field illumination and fluorescence for mCherry of a recorded ERα^vlVMH neuron in a brain slice of ERα^vlVMH-hM3Dq mouse (up), and a representative trace of ERα^vlVMH neuron in ERα^vlVMH-hM3Dq mouse treated with CNO (down). The scale bar is 20 μm. (b) The spontaneous firing frequency (n = 6) and resting membrane potential (n = 8) of ERα^vlVMH neurons before and after treatment of CNO, and after the wash out of CNO. (c) c-fos expression in the MBH of a ERα-Cre mouse that received the injection AAV-DIO-hM3Dq-mCherry virus into one side of the vlVMH (hM3Dq-mCherry) and the injection of control AAV-DIO-mCherry virus into the other side of the vlVMH. Increased c-fos expression in the hM3Dq-mCherry side indicates activation of ERα^vlVMH neurons, compared to the control mCherry side 2 hours after injection of CNO. Black arrows indicate magnified images. White arrows indicate neurons that co-express hM3Dq-mCherry and c-fos in the magnified picture of the white boxed area. Scale bar is 200 or 100 μm. (d) The c-fos quantification results of (c) (n = 4). (e) Serum PRL levels of control (n = 12) and ERα^vlVMH-hM3Dq (n = 14) lactating dams on PPD2. (f) Serum PRL levels of control (n = 12) and ERα^vlVMH-hM3Dq (n = 10) lactating dams on PPD9. (g-i) Body weight (g), daily food intake (h) and BAT temperature (i) of control (g, n = 15; h, n = 12; i, n = 5) and ERα^vlVMH-hM3Dq (g and h, n = 11; i, n = 5) mice during pregnancy before CNO injection. (j-m) Body weight (j), daily food intake (k), BAT temperature (l) and averaged body weight of pups (m) in control (j, n = 15; k, n = 14; l, n = 10; m, n = 8) and ERα^vlVMH-hM3Dq (j, n = 21; k, n = 15; l, n = 7; m, n = 9) mice following the injection of CNO during lactation (excluding those with loss of all pups). (n-p) Maternal care time spent on nursing (n), grooming (o) and nest building (p) of control (n = 5) and ERα^vlVMH-hM3Dq lactating dams (n = 4) on PPD5. Data are presented as mean ± SEM and/or individual data points. b, two-tailed paired t-tests. d, two-tailed unpaired t-tests. e and f, two-way RM ANOVA by multiple comparisons test with uncorrected Fisher’s LSD. k, two-way ANOVA followed by Šídák’s multiple comparisons test with Geisser-Greenhouse correction. P values are shown in figures.

Extended Data Fig. 7 |. Metabolic changes induced by central or systemic supplement of E2 during lactation.

(a-b) The number of pups survived with systemic (a) or central supplement (b) of E2. (c-d) Body weight of lactating dams with systemic (c) or central (d) supplement of E2. (e-f) Daily food intake of lactating dams with systemic (e) or central (f) supplement of E2. (g-h) The average body weight of pups raised by lactating dams with systemic (g) or central (h) supplement of E2. (systemic: vehicle, n = 15; E2 0.5 ug/day, n = 7; E2 2.4 ug/day, n = 10. central: n = 5). (i-j) H&E staining of mammary glands enmeshed in the iWAT of lactating dams with systemic supplement of vehicle (i) or E2 (j). Data are presented as mean ± SEM. a, e and g, two-way ANOVA followed by Šídák’s multiple comparisons test with Geisser-Greenhouse correction. P values are shown in figures.

Extended Data Fig. 8 |. Lactational remodulation of gene profiles of ERα^vlVMH neurons.

(a) Electrophysiology recordings of the three ERα^vlVMH subpopulations (virgin firing vs PPD5 firing vs PPD5 silent) subjected to Patch-Seq. Resting membrane potential (left) and firing frequency (right) of these three ERα^vlVMH subpopulations and their responsiveness to PPT (n = 10, neurons from 3 mice per subpopulation). (b-c) Volcano plot and (b) GO analysis (c) of the differentially expressed genes between virgin firing ERα^vlVMH neurons and PPD5 firing ERα^vlVMH subpopulation. (d-e) Volcano plot and (d) GO analysis (e) of the differentially expressed genes between PPD5 firing subpopulation and PPD5 silent ERα^vlVMH subpopulation. (f) Heatmap of Patch-Seq-identified genes that potentially mediate ERα downstream signal pathways. (g) Secondary analysis of published scRNA-Seq data of VMH neurons⁶⁰ identified 4 potassium channels that displayed the same changes as Patch-Seq data. These 4 genes were expressed at higher levels in lactating ERα^VMH neurons compared to virgin ERα^VMH neurons or lactating non-ERα^VMH neurons (virgin ERα⁺, n = 3932; lactating ERα⁺, n = 5462; virgin ERα⁻, n = 9373; lactating ERα⁻, n = 9776; all information in the Supplementary Table 1). (h) Percentage of ERα^VMH neurons out of the whole VMH neurons in virgin and lactating female mice based on the secondary analysis of published scRNA-Seq data⁶⁰(n = 7). (i) Heatmap of Patch-Seq-identified inhibitory channels between OVX-V and OVX-E mice by reanalyzing published RNA-Seq data from the VMH of OVX-V and OVX-E mice¹⁸. Data are presented as mean ± SEM and/or individual data points. a-bar graphs, one-way ANOVA followed by Dunnett’s multiple comparisons test or two-tailed unpaired t-tests. g, Wald test followed by Bonferroni correction. P values are shown in figures.

Extended Data Fig. 9 |. Lactational remodulation of PRL signaling in ERα^vlVMH neurons.

(a-b) Secondary analysis of the population (a) and expression (b) of Prlr in the VMH⁶⁰(n = 7). CPM, counts per million. (c) Secondary analysis of the expression of Prlr in OVX-V (n = 3) vs OVX-E2 (n = 4) female mice¹⁸. Data are presented as mean ± SEM and individual data points. (d) The expression of Prlr, Stat5a, Stat5b and receptors of other sex hormones in ERα^VMH neurons from our Patch-Seq data. (e) Graphic summary. Lactational suppression of ERα^vlVMH neurons suppresses BAT thermogenesis, promotes feeding and maintains lactational hyperprolactinemia. Created in BioRender. Wang, C. (2025) https://BioRender.com/i72c521.

Supplementary Material

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s42255-025-01268-z.

Data availability

All data supporting this publication are available in the Source Data published alongside the paper. A reporting summary for this article is available as a Supplementary Information file. Patch–seq data are accessible from the NCBI GEO database (GSE284575). Source data are provided with this paper.