Parenting is essential for the survival and wellbeing of mammalian offspring. However, we lack a circuit-level understanding of how distinct components of this behaviour are coordinated. Here we investigate how galanin-expressing neurons in the medial preoptic area (MPOAGal) of the hypothalamus coordinate motor, motivational, hormonal and social aspects of parenting in mice. These neurons integrate inputs from a large number of brain areas and the activation of these inputs depends on the animal’s sex and reproductive state. Subsets of MPOAGal neurons form discrete pools that are defined by their projection sites. While the MPOAGal population is active during all episodes of parental behaviour, individual pools are tuned to characteristic aspects of parenting. Optogenetic manipulation of MPOAGal projections mirrors this specificity, affecting discrete parenting components. This functional organization, reminiscent of the control of motor sequences by pools of spinal cord neurons, provides a new model for how discrete elements of a social behaviour are generated at the circuit level.
Although essential for survival at a multigenerational time scale, parental care entails sacrifices without immediate benefits for the caregiver, suggesting that this behaviour is driven by evolutionarily shaped, hard-wired neural circuits1,2. Parenting, similar to other naturalistic behaviours, comprises multiple coordinated components, such as specific motor patterns, an enhanced motivation to interact with infants, distinct hormonal states and often the suppression of other social activities such as mating. We aimed to exploit the recent identification of MPOAGal neurons as a key node in the control of parenting in mice3 to uncover organizational principles of associated neural circuits. We hypothesized that the function of MPOAGal neurons in parental behaviour requires integration of external signals, such as stimuli from pups and other environmental sources, and internal hormonal and metabolic information, as well as the ability to coordinate the motor, motivational, hormonal and social components of parenting.
Identity and activity of MPOAGal inputs
To determine brain-wide inputs into MPOAGal neurons, we used rabies virus-mediated retrograde trans-synaptic tracing4 (Fig. 1a), and found that MPOAGal neurons receive direct inputs from more than 20 areas in both male and female mice (Fig. 1b, c, Extended Data Fig. 1a and Extended Data Table 1). Presynaptic neurons within the MPOA itself provided the highest fractional input (approximately 20%), and hypothalamic inputs accounted for about 60% of the presynaptic neurons, suggesting that extensive local processing occurs (Fig. 1c). MPOAGal neurons also receive inputs from monoaminergic and neuropeptidergic modulatory areas, the mesolimbic reward system, pathways associated with pheromone-processing, and hypothalamic as well as septal areas involved in emotional states (Fig. 1c and Extended Data Fig. 1a). Inputs from the paraventricular hypothalamic nucleus (PVN), a key area for homeostatic and neuroendocrine control, were particularly abundant. Notably, MPOAGal neurons did not receive direct inputs from oxytocin (OXT)-secreting PVN (PVNOXT) neurons, which are implicated in parturition, lactation and maternal behaviour1,2,5, but instead received inputs from vasopressin-expressing PVN (PVNAVP) neurons, which are associated with the modulation of many social behaviours6 and nest building7 (Fig. 1d). MPOAGal neurons also received inputs from AVP+, but not OXT+, neurons of the supraoptic nucleus (Extended Data Fig. 1d). Input fractions were similar in males and females, with a few exceptions (Fig. 1e, f and Extended Data Fig. 1a). Therefore, MPOAGal neurons appear to be anatomically well-positioned to integrate external (sensory) as well as internal (modulatory) signals that are relevant to parenting in both sexes.
Next, we investigated MPOAGal input activation during parenting according to the animal’s sex and reproductive state. In laboratory mice, virgin females and sexually experienced males and females show parental behaviours, whereas virgin males typically attack and kill pups3,8,9. We combined rabies tracing with immunostaining for the activity marker Fos after parenting in primiparous females (mothers), virgin females and fathers (Fig. 1g) and compared the Fos+ fraction of input neurons between parental animals and non-pup-exposed controls (Fig. 1h–j). Local MPOA inputs were specifically activated during parenting in all groups (Fig. 1h–j), whereas the activation of other inputs was dependent on sex and reproductive state: in parents, but not virgin females, a subset of reward-associated and modulatory inputs were activated (Fig. 1h–j). Presynaptic neurons in pheromone-processing pathways (the medial amygdala (MeA) and bed nucleus of the stria terminalis (BNST)) were selectively activated in fathers and virgin females, but not in mothers (Fig. 1h–j). Because pup-directed aggression in virgin mice is pheromone-dependent3,8, the MeA–BNST pathway might remain partially active in sexually experienced males and parental virgin females, whereas it is fully silenced only in mothers. Intriguingly, the largest number of inputs was activated in fathers (Fig. 1j), and non-overlapping subsets of inputs were activated in mothers and virgin females (Fig. 1h, i). These results suggest that MPOAGal neurons perform different computations of inputs according to the animal’s sex and reproductive state.
Input–output logic of the MPOAGal circuit
To identify MPOAGal projections and synaptic targets, we infected MPOAGal neurons with adeno-associated viruses (AAVs) encoding the fluorophore tdTomato as well as the presynaptic marker synaptophysin conjugated to GFP (Syn–GFP; Fig. 2a and Extended Data Fig. 2a). MPOAGal neurons project to approximately 20 areas in males and females (Fig. 2b, c and Extended Data Fig. 2b). Many of these regions were previously shown to be involved in maternal behaviour using pharmacological manipulations and lesions, mainly in rats10 (Extended Data Table 2). Notably, this projection map mostly overlaps with the input map defined above (Fig. 1c), revealing extensive reciprocal connectivity in parental circuits.
Among the areas most intensely labelled by Syn–GFP were the PVN and anteroventral periventricular nucleus (AVPe) (Fig. 2c), which have both been implicated in the control of parenting6,11. Using rabies tracing from molecularly defined PVN cell types (Fig. 2d), we found that MPOAGal neurons project to PVNAVP, PVNOXT and corticotropin-releasing hormone (CRH)-expressing PVN neurons (PVNCRH) in both males and females (Fig. 2e–g). Furthermore, connectivity from MPOAGal neurons to PVN neurons appears sexually dimorphic, with more MPOAGal neurons projecting to PVNAVP and PVNCRH neurons in males and more MPOAGal neurons projecting to PVNOXT neurons in females (Fig. 2e–g). MPOAGal neurons might therefore exert control over parenting-promoting hormonal release in a sex-specific fashion.
Tyrosine-hydroxylase (TH)-expressing neurons in the AVPe were found to influence parenting in females via monosynaptic connections11 from AVPeTH to PVNOXT neurons. Rabies tracing from MPOAGal or AVPeTH neurons showed that whereas MPOAGal neurons do not receive monosynaptic inputs from AVPeTH neurons (Fig. 1e), AVPeTH neurons do receive direct inputs from MPOAGal neurons in both males and females (Extended Data Fig. 2e, f). Thus, MPOAGal neurons might also influence OXT secretion via a disynaptic circuit from MPOAGal→AVPeTH→PVNOXT neurons (Extended Data Fig. 2g).
We next investigated the organization of MPOAGal projections, and their activity during parenting. Injections of the retrograde tracer cholera toxin subunit B (CTB) into pairs of MPOAGal projection targets revealed few double-labelled MPOAGal neurons (Extended Data Fig. 3a–c). Moreover, retrogradely labelled cell bodies from individual projections occupied characteristic, mostly non-overlapping zones in the MPOA (Extended Data Fig. 3f, g) and conditional tracing of individual projection areas identified only minor collaterals (Extended Data Fig. 4). These results suggest that MPOAGal neurons are organized in distinct pools, each projecting to mostly non-overlapping target areas. To assess whether different MPOAGal pools, as defined by their projection sites, were equally activated during parenting, we used a Cre-dependent, retrograde canine adenovirus (CAV) to label MPOAGal subpopulations projecting to regions that have previously been implicated in parenting (12 out of 22 projections; Extended Data Table 2) and quantified their activation in parental females (Fig. 2h). Fractions of Fos+ neurons differed widely between projections, ranging from more than 50% (projections to the periaqueductal grey (PAG)) to less than 10% (projections to the ventromedial hypothalamus, Fig. 2i). A similar distribution was found in parental fathers (Extended Data Fig. 2d).
On the basis of their high projection density (Fig. 2c), high activity during parenting (Fig. 2i) and potentially diverse contributions to this behaviour (Extended Data Table 2), we selected MPOAGal subpopulations that projected to the PAG, MeA, ventral tegmental area (VTA) and PVN for further characterization. Gal+ neurons were approximately twice more likely to project to most of these candidate areas than expected from their frequency in the MPOA (Extended Data Fig. 3d, e), supporting the hypothesis that these projections have prominent roles in the control of parenting.
We next aimed to determine whether projection-defined MPOAGal subpopulations receive selected inputs from the approximately 20 identified upstream areas (Fig. 1c) or whether they uniformly integrate all inputs. We used a double-conditional approach in which rabies virus can only infect neurons that project to an area of choice12 (Fig. 2j and Extended Data Fig. 5b–d). We found that MPOAGal projections integrate broad input combinations, with characteristic sets of enriched or depleted inputs (Fig. 2k, l). This is seen for projections from the PAG, MeA, PVN and VTA, which receive similar, albeit quantitatively different, inputs (Fig. 2l). Notably, inputs from the nucleus accumbens and lateral septum, areas involved in reward and emotional responses, respectively, were specifically enriched in VTA-projecting MPOAGal neurons (Fig. 2k, l). Together, these findings suggest a circuit architecture in which broad input combinations converge onto largely non-overlapping, projection-defined MPOAGal subpopulations. These subpopulations may in turn be differentially activated during parenting by integrating across quantitatively different sets of activated inputs.
Specific activity of MPOAGal pools
We next used fibre photometry13,14 (Fig. 3a, b) to investigate whether individual MPOAGal subpopulations are active during specific parenting steps. Conditional expression of the calcium reporter GCaMP6m in MPOAGal neurons was achieved by viral injection (Extended Data Fig. 6a) and an optical fibre was implanted above the injection site (Extended Data Fig. 6b–d). The entire (pan-MPOAGal) population displayed high activity during all pup-directed parenting episodes in mothers, virgin females and fathers (Fig. 3c–g and Supplementary Video 1), but not during non-pup-directed (nest building) or passive (crouching) parenting episodes (Fig. 3h, i). MPOAGal activation was stimulus-specific: interactions with adults resulted in minimal activity (Extended Data Fig. 6k, l). Moreover, orofacial motor actions similar to pup interactions did not activate MPOAGal neurons, confirming that the observed signals were not motion-related. The tuning of MPOAGal neurons during parenting was similar in all three groups (Fig. 3q)—highlighting their common role in the control of parental interactions. Activation during pup sniffing was higher in mothers than in virgin females and fathers (Fig. 3c), possibly reflecting the very high sensitivity of postpartum females to pup stimuli15 (Extended Data Fig. 7). Furthermore, activity decreased in mothers—but not in fathers—during eating, self-grooming and sniffing of food (Fig. 3j–l). MPOAGal neurons receive their second-largest fractional input from the arcuate nucleus, a feeding control centre16 (Fig. 1c and Extended Data Fig. 1a), suggesting that inhibition from circuits controlling mutually exclusive motor patterns, such as eating and pup grooming, might cause this decrease in activity.
To record the activity of projection-defined MPOAGal subpopulations, we injected MPOAGal target areas with a Cre-dependent, GCaMP6-expressing herpes simplex virus and implanted an optical fibre above the retrogradely labelled cell bodies (Fig. 3m and Extended Data Fig. 6e–h). PAG-projecting MPOAGal neurons were specifically activated during pup grooming (Fig. 3n and Extended Data Fig. 6m–q), whereas MeA-projecting MPOAGal neurons were active during most episodes of parental behaviour (Fig. 3p and Extended Data Fig. 6m–q), indicating a more general role in parenting. Consistent with their weak Fos activation after parenting (Fig. 2i), no significant activity changes were detected in VTA-projecting MPOAGal neurons (Fig. 3o and Extended Data Fig. 6m–p). Nevertheless, MPOAGal neurons signalling to VTA neurons were weakly responsive during nest entering in a subset of animals (Fig. 3o and Extended Data Fig. 6q; 4 out of 12 mice), potentially reflecting the expectation or drive to interact with pups. Taken together, these findings support the idea that MPOAGal neurons form functionally distinct modules that are tuned to specific parenting episodes.
Functionally distinct MPOAGal pools
We tested the hypothesis that MPOAGal neurons form functionally specialized pools by optogenetically activating projections to PAG, VTA and MeA during pup interactions (Fig. 4a). We virally expressed channelrhodopsin-2 (ChR2) in MPOAGal neurons (Extended Data Fig. 8a), and implanted optical fibres above MPOAGal projection targets. Optogenetic activation of MPOAGal to PAG projections at axon terminals did not affect the fraction of parental virgin females but suppressed pup attacks in infanticidal virgin males (Fig. 4b), and—consistent with MPOAGal to PAG activity during parenting (Fig. 3n)—increased pup grooming and pup-directed sniffing bouts in both males and females (Fig. 4c and Extended Data Fig. 8c). Next, we assessed the motivation to interact with pups by inserting a climbable barrier in the home cage between the test animal and pups (Fig. 4d). Activation of MPOAGal to PAG projections had no effect on the number of barrier crosses (Fig. 4d). Importantly, the effects of activation of MPOAGal to PAG projections were specific to pup interactions, and did not affect interactions with adult conspecifics (Fig. 4e, f).
By contrast, activation of MPOAGal to VTA projections did not affect pup interactions (Fig. 4g, h), but increased barrier crossing in both males and females (Fig. 4i and Supplementary Video 2), indicating an increased motivation to interact with pups. Interestingly, virgin males still exhibited pup-directed aggression after crossing the barrier, suggesting that this effect is not contingent upon the display of parenting. Nevertheless, in naturalistic situations, MPOAGal neurons and associated VTA projections are activated exclusively during parental interactions, thus specifically mediating parental drive. MPOAGal to VTA activation did not increase locomotion (Extended Data Fig. 8j, k) and did not affect interactions with intruders of either sex (Fig. 4j, k).
Finally, activation of MPOAGal to MeA projections did not affect pup-directed behaviours (Fig. 4l, m and Extended Data Fig. 7f, g)—except for a decrease in the amount of time spent in the nest in the females (Extended Data Fig. 8f)—or the motivation to interact with pups (Fig. 4n). However, this manipulation significantly inhibited male–male aggression and chemoinvestigation of a male intruder in females (Fig. 4o, p). Thus, instead of directly influencing parental behaviour, MPOAGal to MeA activation inhibits social interactions with adult conspecifics.
We tested the necessity of these subpopulations for discrete behaviours by expressing the inhibitory opsin eNpHR3.0 in MPOAGal neurons and stimulating their projections in virgin females (Fig. 4q, t, w). Consistent with ChR2 data, optogenetic inhibition of MPOAGal to PAG projections significantly reduced pup grooming and pup-directed sniffing bouts (Fig. 4s and Extended Data Fig. 8n), without affecting other behaviours (Fig. 4r and Extended Data Fig. 8n–p, u). By contrast, inhibition of MPOAGal to VTA projections specifically reduced barrier crossing frequency (Fig. 4v, u and Extended Data Fig. 8q, r, v), except for a reduction in time spent in the nest (Extended Data Fig. 8q). Finally, inhibition of MPOAGal to MeA projections did not affect interactions with an intruder (Fig. 4y) or other behaviours (Fig. 4x and Extended Data Fig. 8s, t, w). Recent findings indicate that representations of social stimuli in MeA and hypothalamic centres change significantly after sexual experience17,18. Thus, low basal activity in this circuit branch in virgin females compared to mothers may preclude further inhibition. Alternatively, or additionally, this lack of effect may result from a more complex role of the connectivity from MPOAGal neurons projecting to MeA.
Taken together, our data suggest that distinct MPOAGal pools control discrete aspects of parental behaviour in both sexes (Fig. 5). Consistent with a role of the PAG in motor aspects of maternal behaviour2, MPOAGal to PAG projections promote pup grooming. Retrograde tracing from PAG showed that MPOAGal neurons synapse with GABAergic (γ-aminobutyric-acid-releasing, inhibitory), but not glutamatergic (excitatory) PAG neurons (Extended Data Fig. 2h–j). Because the vast majority (around 90%) of MPOAGal neurons are GABAergic3, pup grooming is probably elicited by disinhibition in the PAG. Indeed, infusion of the PAG with the GABAA receptor antagonist bicuculline increases pup licking and grooming19. By contrast, MPOAGal to VTA projections specifically influence the motivation to interact with pups without affecting the quality of adult–infant interactions. This is consistent with the proposed role of the VTA in motivation20 and social reinforcement21, and complements previous findings in rats2,22. Nearby Gal+ neurons in the lateral hypothalamus promote food-seeking behaviour, despite lacking VTA projections23, further highlighting the specific role of MPOAGal neurons in parenting. Finally, we found that MPOAGal to MeA projections do not directly influence pup-directed behaviour, but instead inhibit potentially competing adult social interactions.
Interestingly, MPOAGal to MeA projections are active during most episodes of parenting (Fig. 3p, q), suggesting that the entire behaviour, rather than specific parenting components, are broadcast by this projection to influence the vomeronasal pathway24,25,26. Specific inhibitory feedback from MPOAGal to MeA projections might impair the detection, or alter the valence, of non-pup-related social stimuli. Indeed, optogenetic stimulation of glutamatergic neurons in the posteriodorsal MeA—the MeA compartment that is most densely innervated by MPOAGal fibres (Fig. 2b)—has been shown to suppress interactions with adult conspecifics27. The projections investigated here mediate crucial, non-overlapping aspects of parental behaviour and the sum of their activity profiles matches that of the entire MPOAGal population (Fig. 3q). Thus, combined with the finding that MPOAGal neurons contact AVP-, OXT- and CRH-expressing PVN neurons (Fig. 2e–g), we have dissected circuit branches for four major—motor, motivational, social and neuromodulatory—aspects of parenting control. Other MPOAGal projections that have not been included here may have additional roles in parenting. Lastly, our tracing data suggest extensive connectivity within the MPOA (Fig. 1c), hinting at interactions between functionally specialized MPOAGal subpopulations.
Considerable progress has recently been made in identifying neuronal populations that control specific social behaviours or homeostatic functions10,16,28,29,30,31. However, little is known about how these multi-component behaviours or functions are orchestrated at the circuit level. Intriguingly, the modular architecture uncovered here for the control of parenting is reminiscent of the motor circuit motif that has been identified in the mammalian spinal cord, in which discrete phases of locomotor sequences are controlled by functionally distinct neuronal pools with highly specific connectivity patterns32. Whether other social behaviours rely on similar circuit architectures remains to be determined.
The Gal::cre BAC transgenic line (Stock: Tg(Gal-cre)KI87Gsat/Mmucd, 031060-UCD) was imported from the Mutant Mouse Regional Resource Center and has previously been described3. Cre-dependent tdTomato reporter mice (Gt(Rosa)26Sortm9(CAGtdTomato)Hze)33, C57BL/6 J, OXT-IRES-Cre, Vgat-IRES-Cre and TH-IRES-Cre mice were obtained from Jackson Laboratories. Vglut2-IRES-Cre mice were provided by B. Lowell. The AVP-IRES-Cre line has previously been described7. CRH-IRES-Cre mice were obtained from B. Lowell, J. Majzoub and Jackson Laboratories. Animals were maintained on 12 h:12 h light:dark cycle (light on: 02:00–14:00) with food and water available ad libitum. Animal care and experiments were carried out in accordance with the NIH guidelines and approved by the Harvard University Institutional Animal Care and Use Committee (IACUC).
Histology and immunostaining
Animals were perfused transcardially with phosphate-buffered saline (PBS) followed by 4% paraformaldehyde (PFA) in PBS. Brains were dissected and post-fixed in 4% PFA for 16 h, then washed in PBS for 6 h. After embedding in 4% low-melting point agarose (Thermo Fisher, 16520-050) in PBS, 60-μm coronal sections were cut on a vibratome (Leica) and mounted on Superfrost Plus slides (VWR, 48311-703) with DAPI-containing VECTASHIELD mounting medium (Vector Laboratories, H-1200). For immunostaining in 48-well culture plates, sections were permeabilized for 30 min in PBS-T (0.3% Triton X-100 in PBS), post-fixed with PFA for 10 min, and washed in PBS-T (three times, 20 min each). Blocking was carried out overnight in blocking buffer (0.3% Triton X-100, 1% BSA, 2% normal donkey serum in PBS). Incubation with primary antibodies was performed for 24–48 h on a Nutator at 4 °C. After washing in PBS-T (five times, 60 min each), secondary antibodies were added for 48 h at 4 °C. After final washes in PBS-T (five times, 60 min each), sections were mounted. Primary antibodies: goat anti-Fos (Santa Cruz, sc-52, 1:500), chicken anti-GFP (Abcam, ab13970, 1:1,000), rabbit anti-AVP (Immunostar, 20069, 1:6,000), rabbit anti-OXT (Immunostar, 20068, 1:6,000). Secondary antibodies (all from Thermo Fisher): Alexa Fluor-568 anti-goat (A-11057, 1:1,500), Alexa Fluor-555 anti-goat (A-21432, 1:1,500) and Alexa Fluor-647 anti-goat (A-21447, 1:1,500). All antibodies were incubated in PBS-T, with the exception of Fos antibody, which was incubated in PBS.
RNA in situ hybridization
Freshly dissected brains were embedded in OCT (Tissue-Tek, 4583) and frozen with dry ice. Subsequently, 16-μm cryosections were collected on Superfrost Plus slides (VWR, 48311-703) and used for mRNA in situ hybridization. Fluorescent mRNA in situ hybridization was performed mostly as described24. Complementary DNA (cDNA) of Gal or eYFP mRNA was cloned in approximately 800-base-pair segments into a pCRII-TOPO vector (Thermo Fisher, K465040). Antisense complementary RNA (cRNA) probes were synthesized with T7 (Promega, P2075) or Sp6 polymerases (Promega, P1085) and labelled with digoxigenin (DIG, Roche 11175025910) or fluorescein (FITC, Roche 11685619910). Hybridization was performed with 0.5–1.0 ng ml−1 cRNA probes at 68 °C. Probes were detected using horseradish peroxidase (POD)-conjugated antibodies (anti-FITC-POD, Roche 11426346910, 1:250; anti-DIG-POD, Roche 11207733910, 1:500). Signals were amplified using biotin-conjugated tyramide (Perkin Elmer NEL749A001KT) and subsequently visualized with Alexa Fluor-488-conjugated streptavidin (Thermo Fisher, S11223) or the TSA-plus Cy3 system (Perkin Elmer, NEL744001KT).
Recombinant AAV vectors were produced by the UNC Vector Core. AAV titres ranged from 1.3 to 2.6 × 1012 viral particles ml−1, based on quantitative PCR analysis. Pseudotyped, G-deleted rabies virus4 was obtained from the Salk vector core at a titre of 4.3 × 108 viral particles ml−1. The pAAV-CAG-FLEx-Syn-GFP plasmid was provided by S. Arber and AAV1/CAG-FLEx-Syn-GFP was produced by the UNC Vector Core. The pAAV-CAG-FLEx-TCB, pAAV-CAG-FLEx-RG34, pAAV-CAG-FLExFRT-TC and pAAV-CAG-FLExFRT-RG plasmids were provided by L.L. (Stanford University), and AAV5/DJ-hSyn1-FLExFRT-mGFP35; AAV1/CAG-FLExFRT-TC and AAV1/CAG-FLExFRT-RG were packaged by the UNC Vector core. L.L. and E. Kremer provided CAV2-FLExloxP-Flp. L.S.Z. provided CAV2-FLEx-ZsGreen. AAV1/CAG-FLEx-tdTomato, AAV1/Syn-FLEx-GCaMP6m, AAV5/EF1α-DIO-hChR2(H134R)-eYFP and AAV5/EF1α-DIO-eYFP were purchased from UPenn Vector core. HSV-hEF1α-LSL1-GCaMP6m (HT) was obtained from MIT Vector Core.
Anterograde tracing experiments were performed in Gal::cre mice (or in C57BL/6 J for control experiments) at around 8–12 weeks of age. All surgeries were performed under aseptic conditions in animals anaesthetized with 100 mg kg−1 ketamine (KetaVed, Vedco) and 10 mg kg−1 xylazine (AnaSed) via intraperitoneal (i.p.) injection. Using a Nanoject II injector (Drummond Scientific), 300 nl of a 1:1 mixture of AAV1/CAG-FLEx-tdTomato:AAV1/CAG-FLEx-Syn-GFP36 (synaptophysin–GFP) was injected into the MPOA (coordinates: anteroposterior (AP): 0.0 mm from Bregma; mediolateral (ML): −0.5 mm from the midline, dorsoventral (DV): −5.05 mm) to visualize presynaptic terminals of MPOAGal neurons. Syn–GFP was chosen to distinguish presynaptic sites from fibres of passage. Analgesia (buprenorphine, 0.1 mg kg−1, i.p.) was administered for two days after each surgery. Two weeks later, mice were euthanized and dissected. In some experiments, a 1:1 mixture of AAV1/CAG-FLEx-tdTomato:AAV1/CAG-FLEx-Syn-GFP was injected to visualize presynaptic terminals of MPOAGal neurons. For quantification of synaptic density, the average pixel intensity in a target region containing presynaptic GFP+ punctae was calculated and the background was subtracted. Because injections were unilateral and no labelling was observed in most cases contralaterally, the equivalent region on the contralateral hemisphere was chosen for background subtraction; in cases where contralateral GFP+ punctae were present, an adjacent unlabelled region was chosen. Background-corrected intensities were normalized to the average pixel intensity at the MPOA injection site for each brain.
Trans-synaptic retrograde tracing
Input tracing experiments were performed in Gal::cre mice (or C57BL/6 J in control experiments) at about 8–12 weeks of age. We injected 150–200 nl of a 1:1 mixture of AAV1/CAG-FLEx-TCB:AAV1/CAG-FLEx-RG unilaterally into the MPOA. Two weeks later, 450–600 nl EnvA-pseudotyped, RG-deleted, GFP-expressing rabies virus (EnvA-ΔG-rabies) was injected into the MPOA. After recovery, mice were housed in a biosafety-level-2 (BL2) facility for four days before euthanization. Relative input strength was quantified from brain sections as follows: every second 60-μm section was imaged and cells were counted using the ImageJ CellCounter plugin. GFP+ cells on the injected hemisphere were counted and assigned to brain areas based on classifications of the Paxinos Mouse Brain Atlas37, using anatomical landmarks in the sections visualized by DAPI staining and tissue autofluorescence. In addition, all contralateral and non-assigned GFP+ cells were counted to obtain the total number of GFP+ cells. We then quantified the number of ipsilateral mCherry+ starter neurons per brain area and the total number of starter neurons. Because starter neurons are both GFP+ and mCherry+, whereas presynaptic neurons are only GFP+, the total number of starter neurons was subtracted from the total number of GFP+ neurons to obtain the total number of presynaptic neurons within the MPOA. Finally, the relative input fraction for each area was determined by dividing the number of presynaptic neurons detected in that brain area by the total number of presynaptic neurons in a given brain. Injection of starter AAVs and EnvA-ΔG-rabies into the MPOA of C57BL/6 J mice did not result in detectable background labelling (Extended Data Fig. 5a). Inputs from PAG were detected only in a subset of animals. Presynaptic AVP+ neurons in the PVN were identified as predominantly magnocellular based on cell body size38,39 and position40. Presynaptic neurons in the MPOA (Fig. 2d–g and Extended Data Fig. 2e–j) were identified as Gal+ by in situ hybridization.
Retrograde and anterograde tracing experiments were performed in the right hemisphere. However, a recent study found that the oxytocin receptor is more highly expressed in the left auditory cortex of females and that OXT binding there is crucial for pup retrieval5. We therefore investigated potential lateralization effects by tracing MPOAGal neurons in the left hemisphere. Resulting presynaptic neuron numbers and projection patterns (Extended Data Figs. 1b, 2c) were indistinguishable from those obtained after right-hemispheric tracing, suggesting that anatomical lateralization is not a dominant feature of the subcortical circuits described here.
Projection-specific trans-synaptic retrograde tracing
For projection-specific trans-synaptic retrograde tracing (cTRIO (cell-type-specifically tracing the relationship between input and output))12, 300–500 nl of CAV2-FLExloxP-Flp was injected into identified target areas of MPOAGal neurons (for coordinates, see Extended Data Table 1) in 8–12-week-old Gal::cre mice. During the same surgery, 300–600 nl of a 1:1 mixture of AAV1/CAG-FLExFRT-TC:AAV1/CAG-FLExFRT-RG12 (starter AAVs) was injected into the MPOA. This combination of Cre-dependent, Flp-expressing CAV and Flp-dependent starter AAVs renders MPOAGal neurons projecting to a specific target area susceptible to subsequent infection with G-deleted, EnvA-pseudotyped rabies virus. Two weeks later, 450–500 nl of EnvA-ΔG-rabies was injected into the same MPOA coordinate. After recovery, mice were housed in a biosafety-level-2 (BL2) facility for four days before euthanization. Injection of starter AAVs without CAV did not result in expression (Extended Data Fig. 5b, c). However, because the injection of all cTRIO tracing viruses into C57BL/6 J mice resulted in background expression near the injection site (Extended Data Fig. 5d), the following areas were excluded from analysis: MPOA, BNST, AH, PVN and supraoptic nucleus (SON). This background labelling is probably due to low levels of Cre- or Flp-independent expression of TVA-mCherry and RG12.
We quantified the connectivity of each MPOAGal projection to its inputs using a multinomial regression model (response: neuron counts in each input area, factors: MPOAGal projections). The baseline category in the model was represented by the mean input fraction across all experiments. Reported effects are therefore relative to a randomly chosen projection and the P values reported in Fig. 2k, l are obtained from a normal distribution in which the z scores are the effects of the multinomial regression divided by their corresponding standard errors. To test for differences in the multinomial distribution of input to target region projections, the least-square means from the multinomial regression model was computed using the lsmeans package in R and used to run all pairwise comparisons.
MPOAGal input activity screen
To determine which fraction of MPOAGal inputs is activated during parental behaviour, viral injections were performed as described in ‘Trans-synaptic retrograde tracing’. Animals were single-housed until behavioural testing four days later with two pups (see ‘Parental behaviour assay’). For the equivalent experiments in mothers and fathers, 8–12-week-old Gal::cre males and females were paired up 10 days before injection of starter viruses and returned to their home cage where they remained until three days after injection of EnvA-ΔG-rabies when either the father and litter (for testing of mothers) or the mother and litter (for testing of fathers) were removed from the home cage. Parents underwent behavioural testing on the following day, that is, four days after injection of EnvA-ΔG-rabies. Typically around 80% of virgin females and more than 90% of mothers and fathers were parental. Ninety minutes after onset of retrieval, mice were deeply anaesthetized with isoflurane and rapidly perfused transcardially with 30 ml of ice-cold PBS, followed by 30 ml of ice-cold PFA (4% in PBS). Brains were dissected and post-fixed in PFA (4% in PBS) at 4 °C for 16 h. On the next day, brains were rinsed with cold PBS and 60-μm coronal sections were prepared with a vibratome (Leica VT1000 S). Sections were further post-fixed in PFA (4% in PBS) at room temperature for 10 min and immunostaining against Fos was performed (see ‘Histology and immunostaining’). Only brains from mice that performed all steps of pup-directed parental behaviour (sniffing, retrieval, grooming, licking, crouching) were processed. Animals that were habituated in the test arena but not exposed to pups served as negative controls. Unpaired t-tests were used to assess activation of input areas between parental and control animals and P values were adjusted for multiple comparisons using the Benjamini–Hochberg method (false-discovery rate (FDR) < 0.05).
Previous studies have reported that the basic properties of ΔG-rabies-infected neurons are not altered until seven days after infection41,42 and likewise, effects of rabies on (transgene) expression levels have only been reported seven days after infection43. Because animals were tested and perfused four days after rabies infection in our study, neuronal physiology and Fos activation should be mostly unaffected. Because we reliably observed Fos immunostaining in rabies+ neurons (Fig. 1g–j), rabies infection per se does not preclude activity-dependent Fos expression after four days. However, rabies infection could theoretically upregulate Fos expression in infected neurons, resulting in an overestimation of activated input neurons in our dataset. To address this possibility, we compared Fos+ cell numbers in the MPOA of unilaterally rabies-injected mothers between the injected (ipsilateral) and the non-injected (contralateral) hemisphere (Extended Data Fig. 1c, top). We found that numbers of Fos+ neurons were not significantly different between hemispheres (Extended Data Fig. 1c, bottom; P = 0.43; paired t-test; n = 6). Therefore, rabies infection is unlikely to strongly affect Fos+ expression in our experimental paradigm.
MPOAGal projection activity screen
To determine the activation of individual MPOAGal projections during parental behaviour, 300–500 nl of CAV2-FLEx-ZsGreen was injected into identified MPOAGal target areas in 8–12-week-old Gal::cre females. Animals were single-housed one week after injection. Behavioural testing with two pups (see ‘Parental behaviour assay’) was performed three weeks after injection to allow for efficient retrograde transport of the virus. For the equivalent experiments in fathers, 8–10-week-old Gal::cre virgin males were individually paired up with females for four days, injected and subsequently returned to the female. Two to three days after pups were born (around three weeks after injection), and one day before testing, the female and pups were removed from the cage. Testing, brain collection and immunostaining were performed as described in ‘MPOAGal input activity screen’. Because MPOAGal neurons are not activated in non-pup-exposed mice3, negative controls were not performed in these experiments.
Axon collateralization experiments
In order to assess axon collateralization of MPOAGal neurons (Extended Data Fig. 4), Gal::cre mice received injections of 300–500 nl of CAV-FLExloxP-Flp into an MPOAGal target site (for coordinates, see Extended Data Table 1) and of 600 nl of AAV5/hSyn1-FLExFRT-mGFP into the MPOA. Mice were euthanized eight weeks later and the signal was amplified by anti-GFP immunostaining.
Mice expressing tdTomato in Gal+ neurons (Gal::cre+/−loxP-Stop-loxP-tdTomato+/−) received pairwise injections of 50–100 nl of 0.5% (wt/vol) fluorescently labelled cholera toxin B subunit (CTB-488, Thermo Fisher C22841, CTB-647, Thermo Fisher C34778). After seven days, brains were collected, fixed and 60-μm sections prepared. Individual sections were fixed again in 4% PFA for 10 min. The fraction of double-labelled, tdTomato+, Gal+ neurons in the MPOA was quantified. In control experiments, a 1:1 mixture of CTB-488 and CTB-647 was injected into MeA or PAG.
Imaging and image analysis
Samples were imaged using an Axio Scan.Z1 slide scanner (Zeiss), and confocal stacks were acquired on an LSM 880 confocal microscope (Zeiss). Image processing was performed using custom routines for the Fiji distribution of ImageJ. For most tracing experiments, every second section was imaged, with the exception of MPOAGal projection activity and CTB-tracing experiments, where every MPOA-containing section was imaged and analysed.
Parental behaviour assay
Before behavioural testing animals were housed individually for 5–7 days unless otherwise specified. Experiments started at the beginning of the dark phase and were performed under dim red light. Testing was performed in the home cage (with the exception of locomotion assays, see below) and preceded by a 30-min habituation period. Two 1–4-day-old C57BL/6 J pups were placed in different corners opposite the nest. Once retrieval occurred, a timer was started. Each test was recorded using a multi-camera surveillance system (GeoVision GV-1480) and behaviours were scored by an individual blind to the genotype using the Observer 5.0 or XT 8 software (Noldus Information Technology).
Fibre photometry (fluorometry) was performed as previously described44. For photometry recordings, 8–12-week-old Gal::cre+/−loxP-Stop-loxP-tdTomato+/− mice were used. For pan-MPOAGal recordings, 400–500 nl of AAV1/Syn-FLEx-GCaMP6m (Upenn Vector Core) was injected into the MPOA; for projection-specific recordings, 600–700 nl of hEF1α-LS1L-GCaMP6m, a Cre-dependent, retrograde, long-term herpes simplex virus (LT-HSV) was bilaterally injected into MPOAGal target areas. During the same surgery, a custom 400-μm fibre-optic cannula (Doric Lenses) was implanted into the MPOA (for coordinates, see Extended Data Table 1). For recordings in mothers and fathers, animals were paired up five days before surgery, to ensure that pups were born around three weeks after virus injection. One day after surgery, animals were returned to their mating partner. The implanted animal’s mating partner and offspring were removed 3–5 h before recordings. Virgin female mice were single-housed seven days before the first recording session and thereafter between experiments. Recordings were made 2–4 weeks after the surgery under IR illumination in the home cage of the mouse. Mice were briefly (around 10 min) habituated in the recording setup before 8–10 pups (1–4 days old) were introduced into the cage. Recording sessions typically lasted 10–20 min, with at least two days between sequential recordings. The implant was coupled to a custom patch cord (Doric Lenses) to simultaneously deliver 473-nm excitation light from a DPSS laser (Opto Engine LLC), passed through a neutral density filter (4.0 optical density, Thorlabs) and to collect fluorescence emission. Activity-dependent fluorescence emitted by cells in the vicinity of the implanted fibre tip was collected by a 0.65 NA microscope objective (Olympus), spectrally separated from the excitation light using a dichroic mirror (Chroma), passed through a band pass filter (ET500/50, Chroma) and focused onto a photodetector (FDS10X10, Thorlabs) connected to a current preamplifier (SR570, Stanford Research Systems). Another band pass filter (ET600/20) in front of a second photodetector/preamplifier was used to collect tdTomato fluorescence. Owing to considerable bleed-through of the GCaMP signal into the tdTomato channel, we chose not to use the tdTomato recording trace to normalize our data, instead opting for a set of behavioural controls for motion artefacts (see below). The preamplifier output voltage signal was collected by a NIDAQ board (PCI-e6321, National Instruments) connected to a computer running LabVIEW (National Instruments) for signal acquisition. Video recordings were acquired at 15 frames per second and the signal from the optical fibre was sampled at 1 kHz. A TTL-triggered photodiode next to the cage was used to align videos and voltage recording traces.
Analysis was performed using custom MATLAB (MathWorks) routines. Only recordings with a stable baseline were included in our analysis. The raw signal over each entire recording session was divided by the mean of a Gaussian fit to the distribution of GCaMP to normalize the baseline over the recording session. Since the increase in GCaMP signal preceded event detection in some cases (for example, see Fig. 3c), z scores were calculated using the period from −5 to −2 s before event detections as baseline and from 0 to 3 s from event detection as signal. For statistical analyses (that is, t-tests, ANOVA), we considered a value of P ≤ 0.05 significant. Behaviours were scored manually off-line by an experimenter blind to the photometry recording data. The responses to a stimulus type within a session (typically 5–10 trials per behaviour type) were averaged, and these session averages across mice were used as data displayed in Fig. 3 and Extended Data Fig. 6.
We performed a set of behavioural controls to address the possible contribution of motion artefacts to the recorded signal. In all of the following cases, (orofacial) motor actions highly identical to pup interactions did not result in detectable increases in GCaMP fluorescence intensity. No increase in signal was observed when animals retrieved or sniffed a pup-sized cracker (Fig. 3j), during eating (Fig. 3k) or during self-grooming (Fig. 3l). In addition, no increase in signal was detectable when animals retrieved bedding material to the nest (Fig. 3h). Finally, chemoinvestigation of accessible versus inaccessible pups resulted in different GCaMP responses (from −5 to 0 s period before sniffing, Extended Data Fig. 6i, j). Therefore, the increases in signal intensity observed during pup interactions very probably represent actual activity changes rather than motion artefacts.
Gal::cre mice 8–12 weeks of age were used in these experiments. Because potential increases in parental behaviour would be difficult to detect in already highly parental mothers and fathers, we performed these experiments in virgin animals, in which a higher dynamic range of parental interactions can be assessed. Animals were exposed to two pups in their home cage (see ‘Parental behaviour assay’) and those that attacked (virgin males) or initiated parental behaviour (virgin females) within 15 min were selected for surgery. We injected 700 nl of AAV5/EF1α-DIO-hChR2(H134R)-eYFP (activation) or AAV5/EF1α-DIO-eNpHR3.0-eYFP (inhibition) bilaterally into the MPOA and in the same surgery a dual fibre-optic cannula (300 μm, 0.22 NA, Doric Lenses) was implanted 0.4–0.5 mm above the respective MPOAGal projection target (Extended Data Table 1) and affixed to the skull with dental cement. Mice were tested 3–5 weeks after injection to allow for efficient expression of ChR2 or eNpHR3.0 into axon terminals. On testing day, the implant was connected to an optical fibre attached to either a 473-nm laser (150 mW, Laserglow Technologies) or a 460-nm LED (50 W, Prizmatix) for optogenetic activation, or a 589-nm laser (300 mW, Opto Engine LLC) for inhibition, via a commutator. Animals were tested in either stimulation or non-stimulation trials in randomized order, with two days between trials. In addition, the order in which animals were tested during each experimental session was randomized. In pup exposure experiments, two C57BL/6 J pups, 1–3 days of age, were introduced to the test animal’s home cage in each corner furthest from the nest after 10 min of habituation. For activation experiments, blue light (473 nm) was delivered in 20-ms pulses at 20 Hz for 1–4 s whenever the animal contacted a pup with its snout. The light power exiting the fibre tip was 5 mW, which we calculated as providing an irradiance of 5–10 mW mm−2 at the target region (using the brain tissue light transmission calculator provided by the Deisseroth laboratory, http://www.stanford.edu/group/dlab/cgi-bin/graph/chart.php). For loss-of-function experiments, constant yellow light (589 nm) was delivered at 8–10 mW at the fibre tip, amounting to an estimated irradiance of 15–20 mW mm−2 at the target. Each trial lasted up to 10 min but when virgin males attacked and wounded a pup, the trial was ended and the pup was euthanized.
The following behaviours were scored and quantified: pup sniffing, grooming and licking, pup retrieval to the nest, aggression (animal grabs the pup violently and attempts to bite), crouching (animal hovers above the pup in the nest), nest building and time spent in the nest. For the motivation assay, following a 10-min habituation period a transparent barrier was inserted into the home cage, dividing the cage into a nest and a pup compartment. Next, 4–5 pups were introduced into the pup compartment and 473-nm light was delivered in 20-ms pulses at 20 Hz for 4 s every 10 s for a total of 6 min. Locomotion was assessed in a 36 × 25-cm arena over a period of 5 min. In stimulation trials, 473 nm light (20 ms, 20 Hz) was delivered to the implant for 4 s every 20 s, equivalent to the stimulation administered during a typical pup interaction trial. The position of the animal was tracked and analysed by Ethovision XT 8 software (Noldus) to calculate the average velocity and moved distance. For intruder assays, an 8–12-week-old C57BL/6 J intruder of the opposite sex (receptive virgin female, as determined by vaginal smear, or sexually experienced male) was introduced into the resident mouse cage and 473-nm light was delivered in 20-ms pulses at 20 Hz for 1–4 s whenever the animal contacted the intruder with its snout. Sniffing and grooming durations were scored over a period of 5 min, aggression was scored during a 10-min period. After behavioural testing, animals were transfused transcardially and fibre placement as well as efficient light transmission were verified.
Statistics and reproducibility
Data were analysed by two-tailed, unpaired or paired Student’s t-test, by two-tailed Fisher’s exact test or by χ2 test if not indicated otherwise, using Graph Pad Prism 7 for Mac OS, MATLAB or R. Statistical details are given in the respective figure legends. Experiments were independently performed twice (Figs. 1b–f, 2e–g, k, l, 3c–l, 4 and Extended Data Figs. 1, 2a–d, i, j, 3d, e, 4b–f, 7, 8), three (Figs. 1g–j, F2b, c, h, i, 3n–p and Extended Data Fig. 6b–d) or four times (Extended Data Fig. 6f–h).
Further information on experimental design is available in the Nature Research Reporting Summary linked to this paper.
The code that supports the findings of this study is available from the corresponding author upon request.
The data that support the findings of this study are available from the corresponding author upon request.
Publisher’s note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
We thank S. Sullivan for help with behaviour and mouse husbandry. E. Kremer (Montpellier) and R. Neve (MIT) provided viral vectors. E. Soucy and J. Greenwood helped design motivation assay. R. Hellmiss and K. Wilbur helped with illustrations. H. S. Knobloch-Bollmann provided advice on PVN cell types. We thank the members of the Dulac laboratory for comments on the manuscript. This work was supported by a Human Frontier Long-Term Fellowship, an EMBO Long-Term Fellowship and a Sir Henry Wellcome Fellowship to J.K., Fondation pour la Recherche Médicale grant SPE20150331860 to B.M.B., a NIH K99 Award and a NARSAD Young Investigator Award to A.E.A., a Howard Hughes Gilliam Fellowship to B.M.-R., a Harvard Mind Brain and Behavior faculty grant to N.U. and NIH grant 1R01HD082131-01A1 to C.D. C.D. and L.L. are investigators of the Howard Hughes Medical Institute.
Nature thanks V. Grinevich and the other anonymous reviewer(s) for their contribution to the peer review of this work.
Extended data figures and tables
About this article
Anatomical Science International (2019)