Parenting — a paradigm for investigating the neural circuit basis of behavior

Highlights • Molecularly defined nodes have been identified in parental circuits.• A functional circuit logic of parental behavior is emerging.• Parenting relies on largely similar circuitry in males and females.• Infant-directed aggression is controlled by dedicated circuits.


Introduction
Building on many decades of research in mammalian model systems, major progress has recently been made in understanding the circuit basis of parental behavior in laboratory mice (Mus musculus). Mice are ideally suited to this purpose since they exhibit robust parental care and are genetically tractable. Moreover, powerful tools for circuit mapping and interrogation are available for this species. Neuronal populations crucial for parenting have now been identified and a functional circuit diagram underlying parental behavior is taking shape. While these advances have refined previous models and revealed novel principles, they have also uncovered a considerable complexity. Key questions -such as whether parenting relies on dedicated circuits or, rather, generic circuits for social behavior -remain unaddressed. Here I review recent progress, present an emerging circuit logic of parental behavior and outline future challenges. I will first focus on neuronal populations critical for parental behavior before describing an updated functional circuit diagram for parenting. Next, I will discuss the negative regulation of parenting, with novel evidence suggesting that infant-directed aggression is an active process governed by dedicated circuits. Finally, I will outline potential avenues towards a systems-level interrogation of parental behavior.

Neuronal populations critical for parenting
Although strongly modified by experience and physiological state, parenting is an instinctive behavior that can be displayed without any prior experience [1]. For instance, a strain-dependent proportion of virgin female laboratory mice for instance will display spontaneous parental behavior upon first encountering pups, comprising essentially all components of female parental behavior (grooming, licking, crouching, nest building), with the exception of nursing [1]. Similarly, virgin males, in which vomeronasal sensing is abolished, show paternal behavior instead of pup-directed aggression [2,3]. These observations suggest that functional parental circuits are present in adults of both sexes, and that genetic programs strongly contribute to the formation of such circuits. As a consequence, nodes in these circuits are likely composed of defined neuronal populations.
The use of cell type-specific manipulations has considerably advanced our understanding of how parenting as a complex social behavior is organized at the neural level. Most investigations have focussed on brain areas previously identified as critical for parenting by classic lesion studies, such as the medial preoptic area (MPOA) or the posterodorsal medial amygdala (MeApd) [4,5]. Within these areas, neuropeptides, neurotransmitters and receptors have typically been chosen as cellular markers -especially in the hypothalamus, which is composed of a rich set of distinct neuronal cell types [6,7,8 ]. In addition, immediate early genes (IEGs, e.g. c-fos) are frequently used as indirect molecular readouts of neural activity to determine which neurons within such target areas are activated by a given behavior. These approaches have identified parentingrelevant neuronal populations and paved the way for dissecting the circuits within which these neurons function [9 ,10 ,11 ,12 ]. An initial study from Wu et al. reported that MPOA neurons expressing the neuropeptide Galanin (MPOA Gal neurons), which comprise 20% of MPOA neurons, are crucial for parental behavior in both sexes ( Figure 1) [2]. Two further studies found estrogen receptor a -expressing MPOA neurons (MPOA Esr1 ) to be critical for pup retrieval in females ( Figure 1) [10 ,12 ]. Intriguingly, MPOA Esr1 neurons also strongly affect sexual behavior in males and females [12 ].
These observations illustrate several important considerations when using genetic markers for circuit-level studies of behavior: (1) Genetic markers are necessarily imperfect, that is, not all neurons activated by, or involved in controlling, a given behavior, express a single marker. Conversely, not all marker-expressing neurons are involved in a given behavior. Neuropeptide expression can be associated with functional specialization (e.g. somatostatin-positive or parvalbumin-positive interneurons, oxytocinergic and vasopressinergic secretory neurons), but such populations are typically involved in narrowly described physiological functions. In contrast, circuits for complex behaviors are unlikely to be defined by single markers. Pragmatic considerations, for example, the availability of Cre mouse lines with restricted expression patterns, seem to underlie marker choice in some cases.
(2) In cases where a marker is expressed by the majority of neurons within a brain area (e.g. >50% of MPOA neurons are Esr1-positive (Figure 1) [11 ] and 70% of MeApd neurons are GABAergic [13]), the fact that the neurons in question express a marker might be largely irrelevant. Since individual brain areas participate in many behaviors and physiological functions, manipulation of a large fraction of neurons in an area would be expected to result in context-specific effects. This might explain why optogenetic activation of MPOA Esr1 neurons elicits context-dependent sexualbehavior or parental-behavior ( Figure 1) [12 ]. Another prediction is that manipulating variable fractions of a broad population (e.g. by tuning illumination levels in optogenetic experiments) would result in different phenotypes. In cases where the large majority of neurons within an area is manipulated, the conceptual advance over classic, non-cell type specific approaches is questionable. Screening for markers with high enrichment ratios, that is, controlling for relative frequency of markerpositive neurons within an area can address this limitation (see [2]). (3) Immediate early genes such as c-fos are slow (minutes-hours) and only provide an indirect readout of neural activity. Also, it remains incompletely understood which neuronal activity patterns result in their activation in vivo [14]. IEG-positive and marker-positive neurons thus only partially reflect parenting-relevant neural populations. These limitations also apply to other systems, such as Esr1-expressing neurons in the ventrolateral ventromedial nucleus of the hypothalamus (VMHvl Esr1 ), which have prominent roles in aggression [15] but also food intake, physical activity and thermogenesis [16,17].
Single-cell and spatial transcriptomics approaches now offer the opportunity to further define neuronal populations based on location, anatomical connectivity and gene expression profile [8 , [18][19][20][21][22]. Several recent studies have used such approaches on hypothalamic populations [6,8 ]. For instance, Moffitt et al. recently assembled a spatially resolved molecular atlas of the MPOA, identifying distinct MPOA Gal subpopulations [8 ]. In order to functionally exploit such refined molecular identities, better genetic access to such neuronal populations is required. At present, neurons characterized by expressing single marker genes are typically targeted using recombinaseexpressing mouse lines. Only a handful of orthogonal recombinases (Cre, Flp, Dre, FC31, Vika) are currently available [23][24][25][26]. Of those, Cre accounts for the vast majority and the generation of new lines is slow and expensive. Genetic intersections therefore remain challenging and impractical. Alternatively, conditional Neural circuits underlying parental behavior Kohl 85   viral tools, especially adeno-associated viruses (AAVs), can be used. While their limited packaging capacity (4.7 kb) often precludes the incorporation of promoter fragments large enough to drive cell-type specific transgene expression (but see e.g. [27,28]), enhancer sequences have been shown to be suitable for this purpose [27,29]. Such approaches have the potential to give access to more specific, behaviorally relevant neuronal populations in the future.

Circuit logic of parenting
Behaviors are encoded by dynamic activity patterns in brain-wide circuits. Although specific neuronal populations can neither be necessary nor sufficient for any given behavior [30], the identification of parenting-relevant neuronal populations has recently precipitated rapid advances in our understanding of how parenting is orchestrated at the circuit level [9 , 12 ,31,32 ]. Lesion studies and pharmacological manipulations, primarily in female rats, have found many brain areas to be involved in parenting [1,9 ,33,34]. Importantly, each of these areas is also critical for other social and non-social behaviors.
Based on these seminal studies, a circuit model for parenting was proposed in which two opposing pathways mediate the activation and inhibition of parenting, respectively . These neurons project to, and receive inputs from, more than 20 brain areas in a circuit exhibiting extensive reciprocity [9 ]. Importantly, MPOA Gal neurons form projection-defined subpopulations, each receiving inputs from essentially all input areas (Figure 2) [9 ]. The parallel organization of MPOA Gal projections is similar to what has been described for agouti-related peptide-expressing neurons in the arcuate nucleus (Arc Agrp neurons) [35], but contrasts with, for example, VMH Esr1 or PeFA Ucn3 neurons (see 'Negative regulation of parenting'), which predominantly send out branched projections [36,37]. Corresponding with this segregated organization, different MPOA Gal pools are active during different episodes of parenting, and control distinct motor, motivational and hormonal aspects of parenting ( Figure 2) 12 ]. VTA-mediated pup retrieval might be a consequence of acutely increased parental motivation (stimulation of MPOA Esr1 neurons also elicits retrieval of rubber pups [12 ]), but further experimental evidence is needed to address the role of this projection. While it remains to be shown whether these projectiondefined MPOA subpopulations have separable genetic identities (see e.g. [8 ]), these results indicate that discrete components of a complex behavior can be isolated at the circuit level.
In addition to such efforts to trace parenting-relevant circuits in an inside-out manner, i.e. starting from neuronal populations deep in the brain, another possibility is to define parental circuits in an outside-in manner, starting from the sensory periphery. Such efforts have encountered both methodological and conceptual hurdles. One technical challenge is the absence of suitable reagents for anterograde trans-synaptic circuit tracing, although progress has recently been made in this regard [38]. Other limitations are of a conceptual nature: Because of their presumed ability to 'trigger' instinctive behaviors, pheromonal cues have long been proposed to be processed along dedicated, stimulus-specific neural circuits from the sensory periphery into the brain (labeled lines) [39]. Pup-emitted pheromones are thought to promote pup-directed aggression, since ablating vomeronasal organ (VNO) function elicits paternal behavior in otherwise infanticidal virgin males [2,3]. The identification of pup-specific vomeronasal receptors (VRs) might therefore constitute entry points into labeled line circuits into the brain. However, a recent study found that neither pup-sensitive vomeronasal receptors nor associated cues are pup-specific [40 ]. Instead, such receptors are also tuned to adult chemosensory signals, and pup recognition relies on a combination of physical and chemical traits (see 'Negative regulation of parenting') [40 ]. These findings thus call into question the existence of labeled lines for pheromone-triggered behavior [39,41], and therefore the possibility of an outside-in identification of parental circuits.
In summary, considerable progress has been made in uncovering the functional circuit architecture underlying parental behavior. Key emerging principles are that these circuits are enormously complex, overall remarkably similar between the sexes (but see [32 ,42]), and that specific aspects of parenting can indeed be assigned to discrete circuit elements [9 ,43]. It will be interesting to investigate how this circuitry interacts with neural systems controlling other instinctive behaviors (or whether they largely overlap), how information is processed between successive circuit nodes and how experience and physiological states affect their function.  [40 ]. Together, these results suggest that several VRs (and, correspondingly, VNO neuron types) contribute to the detection of infant cues. Surprisingly, however, these VRs are also activated by adult cues, and pup recognition requires a combination of chemical and tactile cues [40 ]. Furthermore, the chemical stimuli detected by Vmn2r65 and Vmn2r88 are rather unexpected: submandibular gland protein C, expressed in salivary glands of pups and adult females, and hemoglobins, which are ubiquitously found in social environments, especially after parturition [40 ]. These results indicate that VNO cues emitted by infants are ambiguous, and that adults use multisensory information for pup recognition.
How are pro-infanticidal stimuli processed deeper in the brain? Vomeronasal information is relayed to the MeA via the accessory olfactory bulb (AOB) before reaching hypothalamic areas, such as the BNST or MPOA (Figure 2) [51]. Chemosensory signals from both VNO and the main olfactory system are presumably integrated by MeA neurons [52], but it remains unclear where and how these signals interact with haptic and other types of sensory information to form pup representations ( Figure 2 [37]. While silencing of PeFA Ucn3 neuronal activity in virgin males blocks infanticide, activation of these neurons elicits infant-directed neglect in virgin females [37]. Intriguingly, PeFA Ucn3 neurons receive direct inputs from (almost exclusively inhibitory) MPOA Gal neurons [2], suggesting that infanticide-promoting circuits might be actively suppressed in parental animals.
Altogether, these observations indicate that (1) infantdirected aggression relies on dedicated circuits which are likely distinct from those mediating male-male aggression, (2) these circuits directly interact with parental circuits -potentially in a mutually inhibitory fashion, and (3) similar neural mechanisms control infant-directed aggression in males and females. It will be exciting to further dissect the circuit mechanisms underlying infantdirected aggression, to investigate how stress promotes this behavior in females, and to address which plasticity mechanisms govern the switch from infanticide to parenting in males.

Towards a systems-level investigation of parental behavior
A key insight from recent studies is that parenting, as well as other instinctive behaviors, rely on highly complex, unexpectedly malleable, and potentially overlapping circuits [9 , 35,59,60]. It remains unclear whether parental behavior is controlled by parenting-specific circuits or rather by general-purpose social behavior circuits that are state-specifically and/or context-specifically engaged.
Distinguishing between these scenarios will require the use of systems neuroscience approaches and the integration of anatomical, functional and behavioral data.
First, single cell and spatial transcriptomics approaches have the potential to identify novel genetic entry points into parenting-relevant neuronal populations, and to uncover plasticity mechanisms within these populations. For instance, Moffitt et al. recently used a massively multiplexed in situ hybridization pipeline (MERFISH) to create a cell atlas of the preoptic area, defining novel cell types and subdividing MPOA Gal neurons into ten transcriptionally and spatially distinct clusters [8 ]. Second, refined anatomical approaches will help uncover further motifs in parental circuits, thereby guiding future functional investigations. Improved viral vectors now enable more specific, efficient and permanent access to defined neurons and circuits [61][62][63][64][65]. However, viral tracing approaches typically visualize connectivity between hundreds to thousands of neurons, thereby obscuring cellular-level anatomical diversity. Individual neurons can be reconstructed by serial two-photon tomography after sparse neuronal labeling, which revealed strikingly complex morphologies and brain-wide projection patterns [66,67]. However, this approach is is highly time-consuming, resource-intensive and laborious. High-throughput, sequencing-based strategies, such as MapSeq [68] are expected to give complementary insights into the organizational principles of parentingrelevant circuits. Third, rather than investigating these circuits one node at a time, addressing dynamic information processing at brain-wide scales will be necessary to understand the neural computations underlying parenting and other instinctive behaviors. High-density recordings from thousands of individually resolved neurons across the brain, will be instrumental for tracking information flow within circuits [69][70][71][72]. Lastly, deep learning approaches now allow for automated, markerless tracking of animals under varying experimental conditions, thereby greatly reducing the time required to analyze behavioral video recordings [73][74][75]. These methods have facilitated behavioral tracking, but behavioral classification remains challenging (e.g. pup grooming versus chemoinvestigation), especially for social interactions involving several subjects. Further improvements to these algorithms, assay-specific behavioral classifiers, and optimization of experimental conditions will without doubt result in increasingly automated behavioral quantification.
Fully leveraging these methodologies will put us in a position to address key questions in neuroscience, such as the degree of plasticity within neural circuits thought to be hardwired, how robustness and plasticity are balanced in such systems, and whether circuits for different behaviors are separate or highly overlapping. Thus, insights into the neural mechanisms underlying parental behavior have the potential to broadly contribute to our general understanding of how evolutionarily sculpted circuits control instinctive behaviors.

Conflict of interest statement
Nothing declared.

53.
Trouillet A-C, Keller M, Weiss J, Leinders-Zufall T, Birnbaumer L, Zufall F, Chamero P: Central role of G protein Gai2 and Gai2+ vomeronasal neurons in balancing territorial and infantdirected aggression of male mice. Proc Natl Acad Sci U S A 2019, 116:5135-5143 Trouillet et al. observe that conditional ablation of Gai2 (which abolishes pheromonal responses in a subset of VNO neurons) enhances male-male aggression, but surprisingly decreases infant-directed aggression, with distinct neuronal activity profiles in downstream brain areas. Together with Ref. [37], this study provides evidence that these two types of aggression are controlled by different circuit mechanisms. In this follow-up study to Ref.
[56] (which reports that BNSTrh lesions suppress infanticidal behavior), the authors hypothesize that the transition from infant-directed affiliative behavior to aggression in males results from plastic changes in the BNSTrh. Using whole-cell patch clamp recordings, they find differences in evoked excitatory synaptic currents between adults and juveniles. This further highlights the BNSTrh as an important locus mediating the adaptive change from parental to infanticidal behavior in male mice.