Thirst-associated preoptic neurons encode an aversive motivational drive

To maintain homeostasis and ensure survival, physiological imbalances produce motivational drives: internal states that promote specific goal-directed behaviors and scale in duration, intensity, and valence (1–5). The classical “drive reduction” hypothesis posits that animals learn particular goal-directed behaviors to reduce the level of an aversive drive state (1, 2, 6). However, electrical stimulation experiments of putative hunger- and thirst-regulating nuclei indicate that motivational states increase the incentive value of stimuli or behaviors that led to reward (4, 7, 8). Recent experiments show that both positive and negative valence mechanisms appear to play a role in controlling feeding (9–14). By contrast, the neural mechanisms for the thirst motivational drive remain poorly understood.

The lamina terminalis of the hypothalamus has been implicated in water intake in mammals through lesion, optogenetic stimulation, and activity recording (10, 15–20). In particular, the median preoptic nucleus (MnPO) integrates blood volume, osmolarity, and hormonal inputs from the circumventricular subfornical organ (SFO) and vascular organ of the lamina terminalis (OVLT) and broadcasts this information to multiple higher brain areas (21). This architecture makes MnPO potentially well suited to play a central role in producing thirst motivational drive. However, MnPO is a highly heterogeneous nucleus that regulates body temperature, sleep, cardiovascular function, and sodium excretion, in addition to thirst (21). It remains unknown how the activity of specific populations of neurons within this area regulates thirst motivation.

Water deprivation induces robust expression of Fos in several hypothalamic nuclei, including MnPO (22, 23). To genetically access these cells, we used a new FosTRAP transgenic mouse line (24), TRAP2, in which 2A-iCreER^T2 was knocked into the Fos locus to create an in-frame fusion (25) (Fig. 1A). In TRAP2;Ai14 double transgenic mice, neuronal activation results in the expression of CreER, which enters the nucleus in response to 4-hydroxytamoxifen (4-OHT) injection and causes recombination. This results in permanent expression of tdTomato (Cre reporter from Ai14) in active neurons (24–26) (Fig. 1B). Injecting 4-OHT into TRAP2;Ai14 double transgenic mice produced many more tdTomato⁺ neurons in the MnPO of mice deprived of water for 48 hours (Thirst-TRAP) compared with water-satiated controls (Homecage-TRAP) (Fig. 1, C, D, and F). We compared Thirst-TRAP with endogenous Fos protein expression after 48 hours of water deprivation. TRAP2 labeling in MnPO had high efficiency (65% of Fos⁺ cells were tdTomato⁺) and specificity (96% of tdTomato⁺ cells were Fos⁺) (Fig. 1G). By contrast, few MnPO cells were double labeled with Fos after 48 hours of water deprivation in Homecage-TRAP mice (Fig. 1, C, E, and H). TRAP2 also efficiently and specifically labeled cells in other hypothalamic nuclei that exhibited high Fos induction upon water deprivation (fig. S1).

MnPO also contains neurons involved in thermoregulation (27), and warm ambient temperature can cause robust MnPO Fos induction (28). After Thirst-TRAP, we challenged mice with a 37°C warm environment for 4 hours before sacrifice. Warmth-induced Fos was expressed in a spatially intermingled but nonoverlapping population compared with Thirst-TRAPed cells in MnPO (Fig. 1, C, E, and H).

Because MnPO and the surrounding medial preoptic area (MPOA) are highly molecularly heterogeneous, we applied single-cell RNA sequencing (29) to determine the full spectrum of TRAPed cell types. After Thirst-TRAP, we used fluorescence-activated cell sorting (FACS) to isolate tdTomato⁺ neurons (~1% of viable cells) from microdissected preoptic hypothalamus and sequenced cDNA from each cell (Fig. 2A and fig. S2). Our microdissection included neurons in both target MnPO and the surrounding MPOA that appeared in both the Homecage- and Thirst-TRAP conditions. Dimensionality reduction and clustering of 348 single-cell transcriptomes that passed quality control revealed two clusters: a mostly inhibitory Cluster 1 (Gad1⁺ encoding a biosynthetic enzyme for γ-aminobutyric acid) (N = 172 cells) and a predominantly excitatory Cluster 2 (Slc17a6⁺ encoding a vesicular transporter for glutamate) (N = 176 cells) (Fig. 2, B to E, and table S1). Many of the top Cluster 1 genes were broadly expressed throughout the preoptic area (fig. S3A), and Cluster 1 exhibited substantial heterogeneity, including cells constitutively expressing the immediate early genes Fos, Egr1, and Junb (fig. S3B). By contrast, Cluster 2 was not clearly separable into subclusters and was characterized by high expression of specific receptors (e.g., the angiotensin II receptor Agtr1a), neuropeptides (e.g., Nxph4 and Adcyap1), and transcription factors (e.g., Etv1) (Fig. 2C). Many Cluster 2 genes were highly expressed specifically in MnPO (fig. S3A), including Nxph4, Adcyap1, and Bdnf, previously described as enriched in warm-associated MnPO neurons (27), as well as Agtr1a specific to thirst-associated neurons. Agtr1a’s endogenous agonist, angiotensin II (ANG II), has long been implicated in regulating water balance within the lamina terminalis (30). These results suggest that Cluster 2 represents a distinct cell type of thirst-associated neurons within MnPO.

We validated this conclusion by examining the colocalization between marker genes and TRAPed preoptic neurons in situ. Using three-color amplified single-molecule fluorescence in situ hybridization (smFISH) (31), we simultaneously mapped the locations of tdTomato⁺, Gad1⁺ (inhibitory), and Slc17a6⁺ (excitatory) neurons. Whereas 90% of tdTomato⁺Gad1⁺ neurons were outside MnPO, 91% of tdTomato⁺Slc17a6⁺ neurons were within MnPO (N = 2 mice, Fig. 2, F and G). Double smFISH between tdTomato and top Cluster 2 genes Adcyap1, Agtr1a, or Nxph4 revealed that for each marker gene, tdTomato⁺ cells represented less than a third of the total population of marker⁺ cells within MnPO (double⁺/marker⁺ = 28, 31, and 28%, respectively). However, the vast majority of tdTomato⁺ cells within MnPO coexpressed each marker gene (double⁺/tdTomato⁺ = 91, 89, and 98%, respectively, N = 3 mice; pie chart in Fig. 2, H to J). Thus, Thirst-TRAPed MnPO neurons appeared to constitute a single, molecularly defined cell type.

If these Thirst-TRAPed MnPO neurons encode thirst motivational drive, then their suppression in water-deprived animals should inhibit, while their reactivation in water-satiated animals should elicit, water consumption. We delivered virally encoded Cre-dependent iC++-2A-eYFP (enhanced yellow fluorescent protein) (32) (inhibitory opsin) or ChR2-eYFP (33) (excitatory opsin) to the MnPO of TRAP2 mice and induced recombination after 48 hours of water deprivation, then delivered light to MnPO via a fiber implant (Fig. 3, A and B). Photoinhibition reduced water consumption in water-deprived Thirst-TRAPed mice but not water-deprived Homecage-TRAPed mice (Fig. 3C). By contrast, photoactivation of these neurons in water-satiated animals rapidly evoked water consumption, which terminated upon cessation of stimulation (Fig. 3D). The rate of water consumption scaled with the frequency of photoactivation [one-way analysis of variance (ANOVA) test for linear trend; P < 0.001] (Fig. 3E), whereas Homecage-TRAPed animals never drank water when photostimulated (Fig. 3F). Patch-clamp recordings in acute slices showed that ChR2-expressing Thirst-TRAPed MnPO neurons responded with high fidelity to photoactivation at 20 Hz (fig. S4, A and B), and optogenetic stimulation 90 min before sacrifice induced a drastic increase in Fos expression, specifically in Thirst-TRAPed MnPO (fig. S4, C and D). Stimulation of Thirst-TRAPed neurons in water-satiated mice did not cause undirected gnawing or licking behaviors (movie S1).

To determine whether activation of MnPO neurons simply causes water consumption or can motivate mice to perform an operant task for water, we water-restricted mice and trained them in a fixed ratio 1 (FR1) operant task, where each lever press leads to a unit of water reward. After training, mice were allowed free access to water and tested on the task while water-satiated. Photoactivation in water-satiated mice rapidly induced lever-pressing for water (Fig. 3G), the rate of which scaled with the frequency of stimulation (one-way ANOVA test for linear trend; P < 0.001) (Fig. 3H), whereas stimulation of Homecage-TRAPed neurons had no effect on behavior (Fig. 3I).

Classical theories of learning and motivation suggest that deviations from homeostatic set points are aversive and that animals perform motivated behaviors to reduce such aversive states (1, 2, 4). Consistent with this theory, real-time place preference assay revealed that mice vigorously avoided the photoactivation of Thirst-TRAPed but not Homecage-TRAPed neurons (Fig. 3, J and K). These data suggested that a reduction in the activity of Thirst-TRAPed MnPO neurons alone might be reinforcing. Indeed, mice learned to vigorously lever-press to turn off photoactivation of Thirst-TRAPed (but not Homecage-TRAPed) neurons over several 30-min sessions (Fig. 3L). Thirst-TRAPed mice would lever-press consistently for the whole duration of the session without any signs of satiation (Fig. 3M). The number of rewarded lever presses scaled with the frequency of stimulation (one-way ANOVA test for linear trend; P < 0.05) (Fig. 3N), suggesting that higher levels of Thirst-TRAPed MnPO neuron activity are more aversive and that the reduction in activity is more reinforcing.

The activity representing this aversive drive state must interface with other brain systems to produce coordinated goal-directed behavior. We found prominent axonal projections to several other subcortical areas, including lateral hypothalamus (LH), paraventricular hypothalamus (PVH), paraventricular thalamus (PVT), and supra-optic nucleus (SON) (fig. S5A). Optogenetic stimulation of MnPO increased the number of Fos⁺ cells in PVT, PVH, and SON of Thirst-TRAPed mice relative to Homecage-TRAPed mice (fig. S5, B to D). Stimulation of ChR2-expressing Thirst-TRAPed MnPO neuron axon terminals in PVH, PVT, and LH all produced both water-drinking and FR1 lever-pressing in water-satiated mice (fig. S5, E to G). Even though individual MnPO neurons displayed limited collateralization (fig. S6), we cannot rule out the possibility that some of these effects were caused by back-propagating action potentials spreading from the stimulated area to other brain areas via collaterals. Finally, Cre-dependent transsynaptic rabies tracing from Thirst-TRAPed MnPO neurons revealed inputs from PVH, PVT, and thirst-related hypothalamic nuclei such as SFO and SON, as well as from the central amygdala (CeA) and parabrachial nucleus (PBN), which could relay other forms of sensory, valence, or satiety information (34, 35) (fig. S7).

Finally, we investigated the in vivo dynamics of Thirst-TRAPed MnPO neurons during thirst-motivated behaviors. MnPO neurons respond to changes in blood osmolarity and circulating ANG II concentration, presumably via inputs from OVLT and SFO (36, 37). MnPO neuron activity is presumed to decrease upon water intake but could potentially exhibit one of the following patterns (Fig. 4A). First, the activity decrease could be purely anticipatory: When animals sense the access to water, their MnPO activity immediately decreases. Second, these neurons could integrate water consumption over time. Third, these neurons could remain active and continue to drive behavior until a satiety signal is engaged that shuts off MnPO activity.

We expressed Cre-dependent GCaMP6 (38) in Thirst-TRAPed MnPO neurons and used fiber photometry to record changes in their activity during behavior (39) (Fig. 4, B and C). Free water consumption by thirsty mice rapidly decreased MnPO neuronal activity within ~1 min, similar to that of SFO^Nos1 (20) neurons, with the decrease coinciding with the period spent licking for water (Fig. 4, D and E). Mice licking an empty bottle had no decrease in activity (Fig. 4, E and I). MnPO neurons also increased activity in response to 3 M saline injection and had no response to isotonic phosphate-buffered saline injection (fig. S8). However, because thirsty mice rapidly consume freely available water, it was difficult to discriminate between the various models relating MnPO activity to motivated behavior.

To reduce the rate of water consumption, we trained these mice on an FR1 operant task and recorded activity during task performance while the mice were water restricted. Lever-pressing for water resulted in a gradual decrease in MnPO activity over many minutes, much more slowly than during free drinking (Fig. 4G). The time when activity reached a minimum coincided with a decrease in the rate of licking and lever-pressing (Fig. 4G). When mice were put on an extinction schedule, with no water delivered for lever-pressing, there was no decrease in activity (Fig. 4, G and I). Control mice expressing YFP had no change in fluorescence during any behavior (fig. S9). We measured how quickly Thirst-TRAPed MnPO neuron activity decreased in an even slower FR6 task, where six lever presses yielded a unit of water. In this condition, the rate of reinforcements was lower than for FR1, as was the rate of decrease of MnPO activity over time (Fig. 4, H and J). Directly comparing the rates of licking over time between free access to water, FR1, and FR6, we found that mice tended to lick for progressively longer across these conditions (Fig. 4J), and activity correspondingly decreased more slowly as mice received water at a slower rate (Fig. 4, K and L). This slow time course of activity decrease, which correlated with progressive water intake, is inconsistent with the anticipation and satiation models; instead, MnPO neuron activity appears to integrate water intake over time.

Genetic access of activated neurons has emerged as an important tool to dissect neural circuit function (40). Using TRAP2 (25), we show that the physiological imbalance following water deprivation produces a motivational drive encoded in the activity of a molecularly defined neuronal type within hypothalamic MnPO. This drive is aversive, scales with the activation frequency of thirst-associated MnPO neurons, and persists in driving behavior over long periods of time. The real-time reduction in activity of these neurons is reinforcing and, as such, is capable of conditioning instrumental behavior. The activity of these neurons integrates the recent history of water intake to adaptively regulate goal-directed behavior. When thirsty, animals perform actions that lead to water consumption, which reduces the aversive MnPO activity by some amount. These actions are repeated until enough water is consumed that the level of aversion ceases to evoke behavior. In doing so, an association is formed between particular actions and reduction of an aversive state, which would contribute to making those actions more likely to be repeated when the animal is thirsty again in the future. Together, our results suggest a mechanism for the implementation of thirst motivational drive in MnPO neurons that resembles the classical drive-reduction hypothesis (1, 2, 4). MnPO thirst neurons are connected to a variety of other brain regions (figs. S5 to S7) that could translate thirst drive into specific goal-directed actions. The extent to which thirst-motivated behaviors arise from negative-valence drive reduction mechanisms working in conjunction with separate positive-valence incentive mechanisms in these downstream circuits remains to be explored.