Food craving-like episodes during pregnancy are mediated by accumbal dopaminergic circuits

Preparation for motherhood requires a myriad of physiological and behavioural adjustments throughout gestation to provide an adequate environment for proper embryonic development1. Cravings for highly palatable foods are highly prevalent during pregnancy2 and contribute to the maintenance and development of gestational overweight or obesity3. However, the neurobiology underlying the distinct ingestive behaviours that result from craving specific foods remain unknown. Here we show that mice, similarly to humans, experience gestational food craving-like episodes. These episodes are associated with a brain connectivity reorganization that affects key components of the dopaminergic mesolimbic circuitry, which drives motivated appetitive behaviours and facilitates the perception of rewarding stimuli. Pregnancy engages a dynamic modulation of dopaminergic signalling through neurons expressing dopamine D2 receptors in the nucleus accumbens, which directly modulate food craving-like events. Importantly, persistent maternal food craving-like behaviour has long-lasting effects on the offspring, particularly in males, leading to glucose intolerance, increased body weight and increased susceptibility to develop eating disorders and anxiety-like behaviours during adulthood. Our results reveal the cognitively motivated nature of pregnancy food cravings and advocates for moderating emotional eating during gestation to prevent deterioration of the offspring’s neuropsychological and metabolic health. Haddad-Tóvolli et al. show that food craving-like episodes in pregnant mice result from a reorganization of the dopaminergic mesolimbic circuitry, and can have long-lasting negative metabolic and neuropsychological effects on the offspring.

Preparation for motherhood requires a myriad of physio logical and behavioural adjustments throughout gestation to provide an adequate environment for proper embry onic development 1 . Cravings for highly palatable foods are highly prevalent during pregnancy 2 and contribute to the maintenance and development of gestational overweight or obesity 3 . However, the neurobiology underlying the dis tinct ingestive behaviours that result from craving specific foods remain unknown. Here we show that mice, similarly to humans, experience gestational food cravinglike episodes. These episodes are associated with a brain connectivity reorganization that affects key components of the dopami nergic mesolimbic circuitry, which drives motivated appeti tive behaviours and facilitates the perception of rewarding stimuli. Pregnancy engages a dynamic modulation of dopa minergic signalling through neurons expressing dopamine D2 receptors in the nucleus accumbens, which directly modulate food cravinglike events. Importantly, persistent maternal food cravinglike behaviour has longlasting effects on the offspring, particularly in males, leading to glucose into lerance, increased body weight and increased susceptibility to develop eating disorders and anxietylike behaviours during adulthood. Our results reveal the cognitively moti vated nature of pregnancy food cravings and advocates for moderating emotional eating during gestation to prevent deterioration of the offspring's neuropsychological and meta bolic health.
Pregnancy demands multiple physiological and behavioural adaptations to ensure a healthy embryonic development and preparation for maternal care 1 . Among them, metabolic and eating behaviour changes have been reported in diverse species [4][5][6] . A distinct example are food cravings, which are common features of human pregnancy, resulting in notable variations in maternal ingestive patterns towards consumption of highly palatable foods (HPF) 2 . Problematically, recurrent food cravings contribute to the development and maintenance of gestational overweight/obesity 3 , with potential adverse health consequences for the offspring [7][8][9] . However, despite the widespread occurrence of gestational food cravings and having been documented for centuries 10 , its underlying neurobiology remains unknown.
To verify the suitability of the mouse as a model to investigate food craving-like behaviour characteristic of pregnancy, we determined their drive to overconsume sweet-tasting items, which are predominant cravings during human gestation 2 . We exposed C57BL/6 virgin and pregnant mice to a two-bottle choice paradigm (Fig. 1a). To ensure a dynamic window of detectable preference changes, we identified the lowest sucralose and sucrose concentration that did not shift tastant selection in virgin females (Extended Data Fig. 1a,b and Fig. 1b,d). Remarkably, pregnant mice exhibited a higher inclination for both sweet compounds, suggesting increased sweetness sensitivity (Fig. 1b-e). Gestation did not disrupt the perception of the nutritional value of sugar because overall food intake was compensated (Extended Data Fig. 1c). To mimic human food cravings, we examined the motivation of pregnant mice to consume a pleasant diet in a 'limited access' paradigm 11 (Fig. 1f and Extended Data Fig. 1d). Pregnant females increased craving-like behaviour and daily food intake from the second week of pregnancy ( Fig. 1g and Extended Data Fig. 1e). These results recapitulate the accentuated ingestion of palatable food observed during human gestation, thus validating the mouse as an appropriate model.
A prevalent hypothesis posits that gestational food cravings are necessary to support embryonic growth 12 . To test this, we generated pseudopregnant females thus mimicking gestational features without actual embryonic development. Pseudopregnancy was validated by the presence of vaginal plugs and increased progesterone levels (Extended Data Fig. 1f). Pseudopregnant mice exhibited enhanced sweet sensitivity (Fig. 1b-e) and exacerbated compulsive eating similar to their genuinely pregnant counterparts (Fig. 1g), without changes in daily food intake (Extended Data Fig. 1e). These results suggest that pregnancy food craving-like behaviour does not arise to support embryonic growth.
Anxiety-like states can also promote compulsive eating 13 . Therefore, we questioned if gestation was associated with anxietylike phenotypes that could alter eating patterns. Pregnant females showed unchanged exploratory time in the anxiety-related central area and total distance in the open field test, as well as equivalent Letters NaTurE METabOlISM performance in the dark-light box test when compared to virgin females (Extended Data Fig. 1g-j). Although not exhaustive, our results suggest that the distinct ingestive habits during gestation are not caused by major anxiety-related states.
Appetite and dietary patterns are controlled by a coordinated network of multiple and distributed neuronal circuits 14,15 . Thus, we reasoned that deviations in food consumption patterns during pregnancy may have a neural basis. To assess the female mouse brain resting-state architecture and uncover temporal features of the pregnant functional connectivity, we conducted functional magnetic resonance imaging (fMRI) before, during and after pregnancy (Fig. 2a). The resting-state network was obtained by independent component analysis (ICA) set to examine 20 independent components (ICs). Among them, 13 functional networks were identified 16 (Extended Data Fig. 2). Two of these networks (that is, corticostriatal and the salience networks) showed enhanced transient activity (based on amplitude analysis) during pregnancy (Fig. 2b-e). Spatial maps identified robust functional connectivity changes in

NaTurE METabOlISM
gustatory, sensorimotor and reward centres during pregnancy (Fig.  2c,e). It is noteworthy that gestation was associated with changes in key components of the dopaminergic mesolimbic circuitry (dorsal and ventral striatum, ventral tegmental area (VTA) and substantia nigra) that couple homeostatically relevant stimuli with the incentive salience of palatable foods and motivated appetitive behaviours. Based on our connectivity analysis, we explored whether gestation-associated functional mesolimbic system changes were associated with the remodelling of key molecular determinants of dopamine circuits. Dopaminergic neurons in the VTA project and release this neurotransmitter into the ventral striatum, where it binds to its receptors, dopamine D1 and D2 receptors (D1R and D2R, respectively), located mainly on medium-sized spiny neurons. Tyrosine hydroxylase (Th; rate-limiting enzyme in dopamine synthesis) and dopamine transporter (Slc6a3) gene expression, as well as the number of VTA-TH + /Fos + neurons and TH-producing neurons were unchanged in the VTA of pregnant mice (Extended Data Fig. 3a-d). Drd1 and Drd2 mRNA expression (encoding D1R and D2R, respectively), as well as Fos-dependent activity, was unchanged in the dorsal striatum (dStri) during pregnancy (Extended Data Fig.  4a-e). However, in the nucleus accumbens (NAc; part of the ventral striatum), pregnant females displayed increased Drd2 expression and an augmented proportion of Drd2 + /Fos + neurons without changes in Drd1 (Fig. 3a-d and Extended Data Fig. 4f). Enhanced accumbal Drd2 expression returned to nonpregnant status after weaning of the offspring ( Fig. 3c and Extended Data Fig. 4f). These results show that gestation specifically upregulates NAc Drd2 expression and activity of D2R neurons, suggesting that this receptor is the main dopaminergic effector involved in food craving-like behaviour.
Dopamine signalling via D2R engages two major intracellular pathways: the canonical G αi/o protein-dependent and the non-canonical β-arrestin-dependent pathways 17 . While activation of the canonical route negatively regulates cAMP production, resulting in decreased protein kinase A (PKA) activity and subsequent inactivation of dopamine and cAMP-regulated phosphoprotein 32 (DARPP-32), the alternative pathway facilitates the activation (dephosphorylation) of glycogen synthase kinase 3 beta (Gsk3β) through β-arrestin-2 and Akt (Fig. 3e) 17 . We examined whether pregnancy promoted molecular adjustments of these signalling routes. Dorsal striatal analysis revealed no changes in TH phosphorylation levels or markers for both downstream pathways (Extended Data Fig. 5a,b). In contrast, TH activity in the dopaminergic terminals that innervate the NAc was negatively modulated by gestation (decreased phosphorylation of TH-Ser40; Fig. 3f and Extended Data Fig. 5c  , pregnant mice (n = 8) and mice after pregnancy (n = 9). c, Coronal slices of whole-brain spatial mapping of the corticostriatal network, including reward and motor control areas. Axial (above, left) and sagittal (below, left) representative slices of the IC spatial map are shown. d, Shape and amplitude analysis of the salience network in virgin mice (n = 7), pregnant mice (n = 8) and mice after pregnancy (n = 9). e, Coronal slices of whole-brain spatial map of the salience network including gustatory, motor, limbic and dopaminergic control areas. Axial (above, left) and sagittal (above, right) representative slices of the IC spatial map are shown.

Letters
NaTurE METabOlISM transduction 18 . Furthermore, the phosphorylation state of DARPP-32 and Gsk3β in the NAc of pregnant mice was attenuated, denoting engagement of both D2R-mediated dopaminergic signalling ( Fig. 3g and Extended Data Fig. 5d). Collectively, these results suggest that pregnancy is associated with a dynamic modulation of dopaminergic signalling mainly through accumbal D2R neurons at both presynaptic and postsynaptic levels.
To gain deeper insights into the role of dopamine in food craving-like behaviour during gestation, we measured this neurotransmitter and related metabolites. Dopamine content was unaltered in the dStri, but it was increased in the NAc of pregnant mice ( Fig. 3h and Extended Data Fig. 5e). While the ratio between homovanillic acid (HVA) and dopamine was unchanged, the relation between 3,4-dihydroxyphenylacetic acid (DOPAC) and dopamine (a proxy of dopamine turnover) was decreased in both brain regions in pregnancy ( Fig. 3i and Extended Data Fig. 5f). This is likely the result of the augmented dopamine concentration in the NAc. Although its biological meaning in the dStri is uncertain and presumably related with other pregnancy-related functions.
Decreased TH-Ser 19 phosphorylation in the NAc probably reflects a mechanism to counteract the increased dopamine content. This mechanism allows for faster adaptations to the oscillatory nature of gestation and is less energy demanding than transcriptomic changes. Our results have also shown an increase in Drd2 expression concomitant with enhanced D2R signalling in the NAc, suggesting a higher dopaminergic response (as a consequence of augmented dopamine in this brain region) that may drive non-homeostatic feeding. These results support the idea that the dopamine system is plastic and that particular physiological states modulate dopaminergic connectivity and function.
We next explored dopamine signalling adaptations as causal determinants of gestational food craving-like behaviour. To this end, we pharmacologically blocked dopaminergic transmission in pregnant mice (Extended Data Fig. 6a). Notably, this approach suppressed gestational food craving-like episodes, without affecting daily food intake or locomotor activity (Extended Data Fig. 6a-e). This suggests that dopaminergic receptor blockage restores the balance between D1R and D2R activity, preventing food craving episodes. Our studies showed a specific activation of D2R-expressing neurons and remodelling of D2R-mediated signalling in the NAc during gestation (Fig. 3c,d,f,g and Extended Data Figs. 4f and 5c,d). Hence, we next aimed to define the involvement of D2R-expressing neurons in this brain region following food craving-like episodes characteristic of pregnancy. To this aim, we injected inhibitory (AAV-hM4Di) Designer Receptors Exclusively Activated by Designer Drugs (DREADDs) in the NAc (and dStri as control area) of pregnant Drd2 +/+ and Drd2 Cre/+ mice (Extended Data Fig. 7a-c). This permitted Cre-mediated silencing of D2R neuron activity after clozapine-N-oxide (CNO) injection 20,21 , which in turn released their inhibitory tone on D1R medium-sized spiny neurons (Extended Data Fig. 7d-i). To exclude potential unspecific effects of CNO, both Drd2 +/+ and Drd2 Cre/+ mice were injected with this compound at a relatively low concentration that does not impact on behaviour 22 . Remarkably, this strategy normalized the exacerbated craving-like behaviour without changes in locomotor activity ( Fig. 3j and Extended Data Fig. 7j-n). This effect was region specific, as DREADD-mediated inhibition of D2R-expressing neurons in the dStri did not attenuate compulsive eating of pregnant mice ( Fig. 3k and Extended Data Fig. 7o-s). CNO administration to Drd2 +/+ mice did not modify locomotor activity or general behaviour, as assessed by visual scrutiny. Altogether, these results demonstrate that gestation-related food craving-like behaviour is mediated by accumbal D2R-dependent circuits.
To examine the potential dichotomy of D2R canonical versus non-canonical signalling underlying food craving-like behaviour, we injected the simultaneous D2R/β-arrestin partial agonist and D2R/ G αi/o antagonist UNC9994 (refs. 23,24 ) into virgin and pregnant mice. UNC9994 administration to pregnant females blocked craving-like behaviour without changes in locomotor activity (Extended Data Fig. 8a-f). Collectively, our chemogenetic and pharmacological studies suggest that gestational-specific craving-like episodes depend on the engagement of both canonical and non-canonical D2R pathways.
Maternal exposure to diverse dietary insults perturbs the development of neurocircuits implicated in psychological, appetitive and metabolic processes of the offspring 7-9 . Therefore, we questioned whether persistent food cravings during pregnancy affects the progeny's neuropsychological and metabolic status. To address this, we implemented the 'limited access' protocol throughout gestation. To characterize the metabolic consequences of this paradigm on dams, and compare it with common maternal obesity protocols, we studied pregnant females exposed to an ad libitum chow diet, ad libitum Western diet (WD) or the 'limited access' paradigm. WD-fed female mice showed a progressive body weight increase, glucose intolerance (despite unaltered fasted blood glucose and insulin levels) and higher plasma leptin (Extended Data Fig. 10a-e). These results indicate that the gestational 'limited access' and maternal obesity protocols represent entirely different paradigms from the metabolic perspective.
Male and female offspring born to dams fed either a continuous chow diet (chow-O) or the "limited access" to HPF throughout gestation (HPF-O) diet were studied. At weaning (postnatal day (P) 21), HPF-O mice of both sexes were glucose intolerant when compared to their chow-O counterparts without changes in body weight ( Fig. 4a,b). Metabolic perturbations in HPF-O mice were accentuated in adulthood, particularly in male mice, which were heavier (Fig. 4c), and exhibited increased adiposity (Fig. 4d) and persistent glucose intolerance (Fig. 4e). Next, we conducted a behavioural screening of anxiety-like phenotypes by testing the offspring to the open field, dark-light box and elevated plus maze paradigms. While individual tests showed some disparity (  Phosphorylation status is depicted by a red circle. f, NAc immunoblot assessing TH phosphorylation levels at residues Ser31 (p-TH S31 ) and Ser40 (p-TH S40 ; n = 5 mice per group). g, NAc immunoblot assessing G-protein-dependent and non-canonical β-arrestin-dependent signalling pathways showing its downstream targets, DARPP-32 phosphorylation levels (at residue pThr34; p-D-32 T34 ) and GSK3β phosphorylation levels (at residue pSer9; p-GSK3β S9 ), respectively (n = 5 mice per group). h,i, NAc neurotransmitter (DA, DOPAC and HVA) levels (h), dopamine turnover (DOPAC to DA ratio) and dopamine storage (HVA to DA ratio) (i) of virgin (n = 9) and pregnant (n = 8) females. j,k, Schematic illustration of the experimental strategy (left) and percentage of daily caloric intake (right) consumed during the 2 h of HPF access of female mice injected with hM4D(Gi) in the NAc (n = 7 Drd2 +/+ and n = 5 Drd2 Cre/+ ) and in the dStri (n = 4 Drd2 +/+ and n = 3 Drd2 Cre/+ ) before and during the second week of pregnancy. Dots represent individual sample data. Data are expressed as the mean ± s.e.m. Exact P values are shown. Statistical analysis was performed by one-way ANOVA with Tukey's multiple-comparisons test for a-d, f and g and by two-way ANOVA with Sidak's multiple-comparisons test for h-k. Drd1, dopamine receptor 1; Drd2, dopamine receptor 2; Th, tyrosine hydroxylase; DA, dopamine; V, virgin; P, pregnant. NS, not significant.

NaTurE METabOlISM
showed more accentuated anxiety-like phenotypes in male HPF-O offspring (Fig. 4k). Maternal overnutrition and obesity have been linked to cognitive dysfunction of the offspring 7,9 . Male HPF-O mice exhibited a trend to display cognitive impairments in the novel object recognition test (NORT) without changes in locomotion ( Fig. 4l-n and Extended Data Fig. 10b). These results demonstrate that excessive food craving-like behaviour during gestation negatively impact offspring's neuropsychological and metabolic health.
Eating disorders often develop during adolescence 26 . To investigate the potential consequences of frequent food craving-like episodes during pregnancy following the progeny's disordered eating, we exposed adolescent (35 days old) Chow-O and HPF-O mice to a compulsive-eating paradigm (Extended Data Fig. 10c). As expected 11 , control animals did not escalate their caloric intake during the limited HPF access ( Fig. 4o) but progressively increased total daily food consumption throughout the test (Extended Data Fig. 10d). In contrast, male HPF-O mice augmented their caloric     10d). The predisposition to compulsive eating was also observed in HPF-O females ( Fig. 4o and Extended Data Fig. 10d). These findings evidenced that excessive maternal food craving-like behaviour confers offspring susceptibility to develop eating disorders during adolescence. Given that maternal high-fat diet feeding perturbs offspring dopamine signalling 27,28 , we sought to determine whether recurrent gestational food craving-like events have a similar impact. Indeed, HPF-O mice of both sexes displayed a downregulation of both Th and Slc6a3 expression in the VTA, with modest changes in the NAc and dStri (Extended Data Fig. 10e-j). Overall, the expression of dopaminergic system determinants was moderately attenuated in HPF-O, suggesting that maternal compulsivity for HPF has long-term influences in hedonic gene expression and behaviour of the offspring.
The urgent appetitive behaviours during pregnancy suggest that cravings should be conceptualized as cognitively motivated states 29 . Indeed, our results indicating that D2R-dependent circuits in the NAc mediate gestational food craving-like behaviour support this. The NAc acts as a hub integrating sensory, emotional and cognitive inputs into reward and motivated behaviours 30 . Enhanced D2R function promotes the incentive salience of food and emotional eating in mice and humans 31,32 , and has been implicated in food addiction 33 . The augmented expression and activity of accumbal D2R neurons during pregnancy might reflect particular neurobiological adjustments associated with this physiological process, in which the connectivity from cortical regions into the NAc is remodelled. This may cause a vulnerability to cope with cognitively motivated states, with subsequent intensification of food craving-like episodes.
While our findings suggest that NAc-D2R neurons are key effectors of food craving-like behaviour during pregnancy, we cannot rule out the contribution of other brain areas with afferent projections into the NAc. For example, the enhanced connectivity observed in the medial prefrontal cortex (prelimbic and infralimbic cortices) and in the insula (part of the corticostriatal and salience networks) during gestation could exert an interoceptive function, by integrating autonomic cues with emotional and motivational states 34 . Our results underline pregnancy as a physiological state able to promote plasticity of the neurocircuits connecting the prefrontal cortex and the basal ganglia. This could partially account for the conscious food cravings, as described for prevalent addictions 35 .
We show that recurrent gestational food craving-like behaviour compromises the metabolic and neuropsychological health of the progeny, enhancing the predisposition to eating and psychological disorders during adolescence and adulthood. These perturbations seem to preferentially affect male offspring. Indeed, maternal obesity is generally associated with pronounced metabolic and cognitive impairments in male than in female offspring in both mice and humans [7][8][9] . The underlying causes of such sex dimorphism demand investigation. Additionally, the majority of programming studies to date have focused on long-term maternal insults, such as diet-induced obesity, undernutrition or permanent stress 7,9 . Our results indicate that even acute HPF consumption during food craving-like episodes is sufficient to propagate detrimental health outcomes in the offspring thus advocating for controlled nutritional care during pregnancy.
Multiple theories have been proposed to explain the emergence of gestational food cravings, including fetus nourishment, nutrient replenishment after nausea, hormonal fluctuations or cultural/ psychosocial factors with inconclusive outcomes 2 . Our studies highlight biological and evolutionarily conserved attributes, which are independent of fetal growth demands, as the driving forces for food craving-like experiences. Pregnancy entails a constant fluctuation of ovarian hormone levels. A concomitant increase in progesterone and oestrogen levels positively modulate food intake, as well as reward sensitivity and dopaminergic behavioural responses 36,37 . Therefore, it is tempting to speculate that variations in ovarian hormone levels during gestation contribute to food craving behaviours.
The purpose of food-motivated behaviours during gestation remains elusive, but might have evolved to ensure the consumption of nutrient-dense and calorie-dense foods thus preventing energy deficits for periods of scarcity. In the modern lifestyle, this evolutive advantage has become detrimental and excessive food cravings may affect mother and offspring health. In summary, our findings provide the cellular and mechanistic basis of pregnancy food cravings and evidence the benefits of moderating emotional eating during gestation.

Methods
Animal care, mouse lines and diets. Mice were maintained on a temperature-controlled, 12-h light-dark cycle with free access to water and standard chow (Teklad maintenance diet 14% protein; Envigo). For the 'limited

NaTurE METabOlISM
access' paradigm, WD (40% kcal from fat and 43% kcal from carbohydrates; Research Diets) was provided for 2 h during three consecutive days a week 11 . C57BL/6 mice were bred in-house. Drd2-Cre mice have been previously described 38,39 . Cre-negative littermates were used as controls. All animal studies were performed with approval of the Universitat de Barcelona Ethics Committee, complying with local government legislation (10637).
Breeding strategy. C57BL/6, Drd2 +/+ or Drd2 Cre/+ female mice were mated at 9-12 weeks of age. Two females were housed with one male and the vaginal plug was examined daily. Presence of the plug denoted day 0.5 of gestation. Weight gain was measured weekly to confirm pregnancies. Dams were kept with their offspring until weaning. For pseudopregnant mice generation, 8-week-old C57BL/6 females were mated with vasectomized males. Presence of the vaginal plug was considered positive pseudopregnancy.

Determination of sweet compound concentration.
Eight-week-old C57BL/6 female mice were trained to drink from two water bottles until they reached a 50/50 preference. They were then presented with different concentrations of either sucralose (2.5, 1.0, 0.5, 0.1 and 0.05 mM) or sucrose (200, 100, 50 and 25 mM) and the preference of each concentration against water was tested. The concentration at which virgin females were unable to discern water from sweet compounds (50/50 intake) was determined and used in subsequent studies.

Two-bottle taste preference test.
Eight-week-old C57BL/6 female mice fed with standard diet were individualized and trained to drink from two water bottles until 50/50 preference. They were then subjected to a two-bottle paradigm 40,19 to test their preference for either sweet tastant (0.5 mM sucralose or 25 mM sucrose) or water during three experimental periods: (1) before pregnancy, (2) throughout pregnancy with measurements every week, and (3) 1 week after the weaning of the offspring (after pregnancy). Mice were water deprived overnight. At 09:00, two bottles containing sweet compound or water were provided. Mice were allowed to drink for 3 h, without access to food. Total volume intake was measured and water was provided ad libitum until 19:00. Taste preference was calculated as the ratio between the sweet liquid consumption and the total volume ingested. Sweet preference and total food intake were measured daily. The position of the bottles was changed every day to exclude the effects of position. Control groups with virgin and pseudopregnant females were run in parallel.
'Limited access' paradigm. The 'limited access' paradigm was based on previous studies 11 . Briefly, 8-week-old C57BL/6, Drd2 +/+ or Drd2 Cre/+ female mice were individualized and habituated to a WD during a 5-d period. After that, females had ad libitum access to chow and compulsive eating was induced by limited access to a WD for 2 h a day (at the end of the light cycle) for three consecutive days per week during two experimental periods: (1) before pregnancy and (2) throughout pregnancy with measurements every week. Maternal food intake was measured daily. The percentage of daily caloric intake consumed during the 2 h of intermittent WD access was considered a proxy for food craving-like behaviour.

Maternal physiology measurements.
For the GTT, embryonic day (E)14.5-16.5 pregnant females fed with a chow diet (n = 8) or a WD (n = 8) or exposed to the 'limited access' paradigm (n = 8) were injected intraperitoneally (i.p) with d-glucose (2 g per kilogram body weight) after 6 h of fasting. Blood glucose levels were measured using a glucometer (Nova Pro Biomedical). Plasma insulin and leptin levels were measured after fasting (6 h) with commercial ELISA kits (Crystal Chem). Body weight was measured weekly.
Progesterone serum levels. Plasma samples were collected from C57BL/6 female mice at three different time points: (1) before pregnancy, (2) throughout pregnancy with measurements every week and (3) 1 week after the weaning of the offspring (after pregnancy). Control groups with virgin or pseudopregnant females were run in parallel. Levels of progesterone were determined using commercial kits (Crystal Chem).
Magnetic resonance imaging acquisition, processing and analysis. Each mouse was scanned at three time points: (1) before pregnancy (n = 7), (2) during the second week of pregnancy (n = 8) and (3) 1 week after the weaning of the offspring (after pregnancy; n = 6). Experiments were conducted on a 7.0T BioSpec 70/30 horizontal animal scanner (Bruker BioSpin), equipped with an actively shielded gradient system (400 millitesla per metre), 12-cm inner diameter). The receiver coil is a surface coil for the mouse brain. Animals were sedated (4% isoflurane in 30% oxygen and 70% nitrogen), placed in a supine position in a Plexiglas holder with a nose cone for administering anaesthetic gases and fixed using tooth and ear bars. Eyes were protected from dryness with Siccafluid ophthalmologic fluid. Once placed in the holder with constant isoflurane (1.5%), a subcutaneous bolus of medetomidine (Domtor, Orion Pharma; 0.3 mg per kg body weight) was injected. For the next 15 min, the percentage of isoflurane was progressively decreased to 0.5%. Then, a continuous perfusion of 0.6 mg per kg body weight per hour of medetomidine was started and maintained until the end of the acquisition session. After completion of the imaging session, 1 µl g −1 of atipamezole (Antisedan, Orion Pharma) and saline were injected to reverse the sedative effect and compensate fluid loss. Localizer scans were used to ensure the accurate position of the head at the isocentre of the magnet. Anatomical T2 RARE images were acquired in coronal orientation with effective time of echo (TE) = 33 ms, time of repetition (TR) = 2.3 s, RARE factor = 8, voxel size = 0.08 × 0.08 mm 2 and slice thickness = 0.7 mm. Resting-state fMRI was acquired with an EPI sequence with TR = 2 s, TE = 19.4, voxel size 0.21 × 0.21 mm 2 and slice thickness = 0.5 mm. In total, 420 volumes were acquired resulting in an acquisition time of 14 min.
The resting-state fMRI acquisition was processed to extract functional brain networks by ICA and evaluate differences in connectivity between pregnant and nonpregnant periods. Each fMRI acquisition was pre-processed including slice timing, motion correction, pass-band frequency filtering, spatial normalization to a mice brain atlas template and smoothing (ANTs and Python 3). All the pre-processed images were considered to extract the group ICA using FSL MELODIC. The resulting ICs were compared with the functional networks described 16 to identify the components corresponding to functional networks of interest. Afterwards, dual regression was performed to identify the network components and their spatial distribution in each individual brain and estimate the network shape and amplitude differences 41 . Network amplitude was defined as the standard variation of the time-series corresponding to the IC, and shape differences were analysed based on the mean value of the z-score of the subject IC within the spatial map of the component (threshold at z = 2.3). The Kruskal-Wallis test was applied to identify differences between shape and amplitude values in the pregnant period versus nonpregnant period in the selected ICs.
Fluorescence in situ hybridization. C57BL/6 female mice (virgin, pregnant (during the second week of pregnancy) and after pregnancy) were transcardially perfused with 4% paraformaldehyde (PFA). Dissected brains were post-fixed in 4% PFA at 4 °C for 24 h and cryoprotected. Brains were cut at 20 µm on a cryostat and collected into four series (one in every four sections) in SuperFrost Plus Gold slides (Thermo Fisher) and subsequently stored at −80 °C.
Fluorescence in situ hybridization for the simultaneous detection of Drd1, Drd2 and Fos mRNA was performed using RNAscope. All reagents were purchased from Advanced Cell Diagnostics. All incubation steps were performed using the Advanced Cell Diagnostics HybEz hybridization system.
On assay day, one section series cut throughout the dStri and NAc was selected. From each animal, one section was incubated with the negative control probe to enable calculation of background. Slides were washed in PBS, baked at 60 °C for 30 min and post-fixed with 4% PFA for 15 min. Sections were then dehydrated and baked for an additional 30 min at 60 °C and submerged into boiling Target Retrieval reagent for 5 min. The slides were dehydrated in 100% ethanol, allowed to air dry for 5 min and placed into an RNAscope holder. Sections were treated with Protease III for 30 min at 40 °C. Probe hybridization, amplification and detection were performed according to the manufacturer's protocol. The colour module chosen labelled the Fos probe with Atto 550, the Drd1 probe with Atto 647 and the Drd2 probe with Atto 488. Sections were counterstained with DAPI and coverslipped with ProLong Gold Antifade Mountant (Thermo Fisher) and stored in the dark at 4 °C.
Images were taken using a confocal Leica DM 2500 microscope, equipped with a ×40/1.15 oil objective and using a zoom of 2×. Z-stacks of 1 µm of either the dStri or the NAc were captured bilaterally from four rostral-to-caudal sections for each animal (n = 3 animals per group). Laser intensities were kept constant throughout the entire image acquisition process. Images were imported into Fiji (National Institutes of Health) where maximum intensity projections were made. To acquire the minimum intensity value for analysing the expression of Drd1 and Drd2, the threshold for probe recognition was calculated as the mean cell intensity present in the negative control sections + 3 × s.d. All labelling above this value was considered to be true signal 42 . Brightness and threshold were adjusted in all images. For quantification, 8-10 Drd1+ or Drd2+ neurons per section were manually selected and Drd1 and Drd2 particles were counted. After quantification, the presence of Fos expression was determined.

Double Th and Fos immunofluorescence.
Selected 20-μm-thick sections (one of four sections) throughout the VTA (bregma between −3.10 mm and −3.70 mm) of virgin, pregnant (during the second week of pregnancy) or post-pregnancy C57BL/6 female mice (n = 3 per group) were blocked with 2% chicken serum in PBS + 0.1% Triton X-100 + 3% BSA and incubated with rabbit anti-Fos antibody (1:500 dilution; Synaptic Systems) in blocking solution overnight at 4 °C. As secondary antibody, a chicken anti-rabbit Alexa Fluor 488 (1:300 dilution; Life Technologies) in PBS + 0.1% Triton X-100 + 3% BSA was used. After washing, slices were blocked with 2% donkey serum in PBS + 0.1% Triton X-100 and incubated with sheep anti-Th antibody (1:600 dilution; Sigma) overnight at 4 °C. As secondary antibody, a donkey anti-sheep Alexa Fluor 594 (1:300 dilution; Life Technologies) in PBS + 0.1% Triton X-100 was used. For quantification, representative images throughout the VTA of each animal were acquired using a Leica DMI 6000B confocal microscope equipped with a ×20 objective. The total number TH + neurons and the proportion of TH + /Fos + was counted manually using Fiji (ImageJ) software.

Pharmacological blockage of dopamine receptors.
Eight-week-old C57BL/6 female mice were i.p. injected with the D1R/D2R dopamine receptor antagonist cis-(Z)-flupentixol dihydrochloride (flupentixol, 0.4 mg per kg body weight; Sigma) or saline (vehicle) and exposed to the 'limited access' paradigm or open field test 30 min later. The drug was prepared 5 min before the tests.
Pharmacological assessment of D2R signalling dichotomy. Eight-week-old C57BL/6 female mice were i.p. injected with the D2R dopamine receptor β-arrestin agonist/Gi antagonist UNC9994 (2 mg per kg body weight; Axon MedChem) or vehicle (10% DMSO and 20% cyclodextrin in saline) and exposed to the 'limited access' paradigm or open field test 10 min later.
Offspring studies: general considerations. Pregnant and lactating dams (n = 9 chow fed and n = 11 exposed to the 'limited access' paradigm) underwent weekly follow-ups of body weight and food intake. Litter size was adjusted (between P1 and P4) to six to eight pups to ensure standardized nutrition until weaning. At weaning, one to three male and female offspring from each litter were randomly subdivided for subsequent studies. For physiological and behavioural tests, the offspring from five chow-fed females and five females exposed to the 'limited access' paradigm were used. For compulsive-eating predisposition studies, the offspring from four chow-fed females and six females exposed to the 'limited access' paradigm were used.
Offspring physiological measurements. For the GTT, mice were i.p. injected with a single bolus of d-glucose (2 g per kg body weight) after 6 h (P21 animals; n = 8-10 mice per group for each sex derived from n = 5 dams per group) or overnight (10-week-old; n = 7-8 mice per group for each sex derived from n = 5 dams per group) fasting. Blood glucose levels were measured using a glucometer (Nova Pro Biomedical).
Compulsive-eating predisposition in the offspring. Offspring of mothers continuously subjected to a chow diet (n = 4) or the 'limited access' paradigm (n = 6) were weaned at 21 d of age. At this time point, male (n = 5 per group) and female (n = 5 per group) pups from both experimental groups were separated and switched to standard chow diet until P30. Males and females of similar body weights were subjected to a limited access WD feeding to assess compulsive-eating-like behaviour. Briefly, chow and HPF offspring were habituated to WD during a 5-d period. P35 offspring were exposed to a 'limited access' paradigm (2 h a day, three times a week) during 4 weeks without food restriction 11 . Weekly body weight and daily food intake (chow and WD) were measured. The percentage of daily caloric intake consumed during the 2 h of intermittent WD access was considered a proxy for compulsive-eating behaviour.
Behavioural procedures. Before each test, mice were acclimatized to the behavioural room and arena for 1 h. The arena was cleaned with 70% ethanol before and after every trial. Light intensity was adapted to each task. A video camera was used to track the movement of each animal. Videos were recorded and analysed with video-tracking software (SMART v3.0, Panlab). For gestational studies, the experiments were conducted in virgin (n = 10), pregnant (during the second week of pregnancy; n = 12) and post-pregnancy (n = 10) females. For offspring studies, the experiments were conducted in 12-to 14-week-old male (n = 8) and female (n = 7-8) offspring from chow (n = 7-8 offspring for each sex derived from five dams) or 'limited access' HPF (n = 8 offspring for each sex derived from five dams) mothers.
Open field test. Mice were placed in the centre of a dark methacrylate arena (35 × 35 × 35 cm) and allowed to freely explore it for 15 min. Animals were tested under a low-intensity light (<30 lux) to avoid stress. Total distance and the time spent in the corner and centre of the arena was scored.

Dark-light box test.
Our protocol was based on previous studies 44 . The test apparatus consisted of a methacrylate arena (35 × 35 × 35 cm) divided into a small dark (safe) compartment and a large strongly illuminated (200 lux; aversive) compartment. A door connected both compartments. Mice were placed in the dark compartment and allowed to freely move between the two chambers for 5 min. Video tracking data were analysed to calculate the time spent in each compartment and the latency to enter the illuminated compartment.
Elevated plus maze test. Our protocol was based on previous studies 44 . The apparatus used was a cross-shaped four-arm maze, with two open arms and two closed arms (25 × 5 cm). This structure was elevated 60 cm above the floor. Mice were placed in the centre of the apparatus, facing towards a closed arm, and their behaviour was video recorded for 5 min. Time spent in the closed arms and open arms was analysed.

NaTurE METabOlISM
Novel object recognition test. The NORT was based on Leger et al. 45 . The test was conducted in a methacrylate (35 × 35 × 35 cm) arena under a 20-lux environment. NORT consisted of three phases: habituation, training and test. During habituation (day 1), mice were allowed to explore the arena for 10 min. In the training phase (day 2), mice were allowed to explore two identical objects (equidistantly spaced) for 10 min. The test phase was conducted 24 h later to measure long-term memory. In this phase, one of the objects was replaced by a new one and the mouse was allowed to explore them for 10 min. The position of the two objects was constant across sessions. Discrimination indices were calculated as: (time novel − time familiar )/ (time novel + time familiar ). We also scored total exploratory time (time novel + time familiar ) and total distance travelled. Animals that showed freezing behaviour or exhibiting <5 s of exploratory behaviour were excluded. The trial was video recorded and analysed off-line.
Statistical analysis. Data are expressed as the mean ± s.e.m. Two-group one-factor comparisons were performed using a two-tailed unpaired Student's t-test. Three-group one-factor comparisons were performed using a one-way ANOVA, and two-factor or three-factor comparisons were performed using two-way or three-way ANOVA followed by Tukey's multiple-comparison test when computing confidence intervals for every comparison or the Holm-Šídák test when not. Factor results (in relation to the variables group, time and/or the interaction group:time in the female studies; Fig. 1 and Extended Data Figs. 1 and 9) are shown next to the graph (lack of this information means that no significant differences were found). For offspring studies ( Fig. 4 and Extended Data Fig. 10), we implemented a linear mixed-effects model analysis to account for statistical dependence among individuals originating from the same dam. In the case of non-repeated measures, comparisons were conducted between experimental 'diet' and offspring 'sex' (fixed effects) while setting the variable 'dam' as a random term. For physiological and compulsive-eating predisposition studies, when repeated measurements were recorded over time, comparisons were made between diet, offspring sex and time of record (fixed effects). The variables 'dam' and 'pup ID' were considered as random factors. The statistical significance of experimental treatments was assessed by ANODE. Given the substantial number of statistical comparisons involved in the compulsive-eating study, all P values were adjusted using the Benjamini-Yekutieli test thus minimizing the false discovery rate. Results from all post hoc analyses are depicted on top of the graph bars using asterisks. Z-scores were calculated by dividing the difference between the individual data determined in each experiment (x) and the mean of the control group (m), to the standard deviation of the control group (s) (z = (x − m)/s). We defined chow diet as the control group. Complementary measures (open field, dark-light box and elevated plus maze) were integrated by implementing a z-normalization across diverse behavioural tests as previously described 23 . Analysis was performed with GraphPad v8 and R v4.0.2 (lme4, car and multcomp packages). P < 0.05 was considered statistically significant.
Reporting Summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support the findings of the study are available from the corresponding authors upon reasonable request. Source data are provided with this paper. All other data are available in the main text or the Supplementary Information.

NaTurE METabOlISM
Letters NaTurE METabOlISM Extended Data Fig. 1 | Sweet taste perception, food craving-like behaviour and anxiety-like states during pregnancy. a, Preference for diverse sucralose concentrations (0.05-2.5 mM). Black arrow indicates the selected concentration in which virgin female mice were unable to distinguish sucralose from water (n = 5 mice per group/ average of 5 measurements per week). b, Preference for diverse sucrose concentrations (25-200 mM). Black arrow indicates the selected concentration in which virgin female mice were unable to distinguish sucrose from water (n = 5 mice per group/ average of 5 measurements per week). c, Daily food intake of chow when pregnant females (during the second and third week of pregnancy) were exposed to the two-bottle paradigm. Water (n = 32), Sucrose (n = 29), Sucralose (n = 31). d, Schematic illustration of HPF "limited access" paradigm. e, Total daily caloric intake of virgin (n = 8), pseudopregnant (n = 22) and pregnant (n = 13) mice. f, Serum progesterone levels of virgin (n = 5), pseudopregnant (n = 15) and pregnant (n = 8) mice throughout the study. g, h, Recorded parameters to assess open field performance in virgin (n = 10), pregnant (n = 12) and after pregnancy (n = 10) mice, including percentage of time spent per zone (g), and total distance travelled (h). i, j, Recorded parameters to assess dark-light box performance in virgin (n = 8), pregnant (n = 12) and after pregnancy (n = 9) mice, including the latency to cross to the light compartment (i) and the time spent in the light compartment (j). Dots in all panels represent individual sample data. Data are expressed as mean ± SEM. Exact P values are shown. Statistical analysis was performed by one-way ANOVA with Tukey's multiple comparisons test for c, g, h, i, j and by two-way ANOVA with Tukey's multiple comparisons test for e, f. When the factors Group, Time and/or the interaction Group:Time were considered significant, results are shown with significant factor or the interaction effect next to it (f).  signalling is not altered in the dStri of pregnant mice. a, Immunoblot assessing TH phosphorylation levels at residues Ser31 (p-TH S31 ) and Ser40 (p-TH S40 ) in dStri extract from virgin (n = 5), pregnant (n = 5) and after pregnancy (n = 5) mice. b, Immunoblot assessing G protein-dependent and non-canonical β−arrestin-dependent signalling pathways showing its downstream targets, DARPP-32 phosphorylation levels (at residue pThr34; p-D-32 T34 ) and GSK3β phosphorylation levels (at residue pSer9; p-GSK3β S9 ), respectively, in dStri extract from virgin (n = 5), pregnant (n = 5) and after pregnancy (n = 5) mice. c, Complete immunoblot assessing TH phosphorylation levels at residues Ser31 (p-TH S31 ) and Ser40 (p-TH S40 ) in NAc extract from virgin (n = 5), pregnant (n = 5) and after pregnancy (n = 5) mice. d, Complete immunoblot assessing G protein-dependent and non-canonical β−arrestin-dependent signalling pathways showing its downstream targets, DARPP-32 phosphorylation levels (at residue pThr34; p-D-32 T34 ) and GSK3β phosphorylation levels (at residue pSer9; p-GSK3β S9 ), respectively, in NAc extract from virgin (n = 5), pregnant (n = 5) and after pregnancy (n = 5) mice. e-f, Neurotransmitter (DA, DOPAC and HVA) levels (e), dopamine turnover (DOPAC to DA ratio) and dopamine storage (HVA to DA ratio) (f) in the dorsal striatum (dStri) of virgins (n = 9) and pregnant (n = 8) females. Dots in all panels represent individual sample data. Data are expressed as mean ± SEM. Exact P values are shown. Statistical analysis was performed by one-way ANOVA with Tukey's multiple comparisons test for a, b and by two-way ANOVA with Sidak's multiple comparisons test for e, f. dStri: dorsal striatum; Th: tyrosine hydroxylase; DA: dopamine; PKA: protein kinase A; DARPP-32: dopamine-and cAMP-regulated phosphoprotein, Mr 32 kDa; Gsk3β: glycogen synthase kinase 3 beta; NAc: nucleus accumbens; DOPAC: 3,4-Dihydroxyphenylacetic acid; HVA: homovanillic acid.

NaTurE METabOlISM
Letters NaTurE METabOlISM Extended Data Fig. 6 | Pharmacological blockage of pan-DR receptors reverts pregnancy-specific increase in HPF consumption. a, Schematic illustration of experimental strategy (left) and percentage of daily caloric intake (right) consumed, during the 2-hours of HPF access, by female mice treated with saline or pan-DR antagonist flupenthixol before and during the second week of pregnancy (n = 11 mice/group). b, Total daily caloric intake (chow + HPF) of female mice treated with saline or pan-DR antagonist flupenthixol before and during the second week of pregnancy (n = 11 mice/group). c-e, Recorded parameters to assess open field performance in female mice treated with saline and pan-DR antagonist flupenthixol, including global activity (c), time spent immobile (d), and total distance travelled (e) (n = 3 mice/group). Schematic of the open field test highlights center (purple) and corner (cream) zones is shown. Dots in all panels represent individual sample data. Data are expressed as mean ± SEM. Exact P values are shown. Statistical analysis was performed by two-way ANOVA with Sidak's multiple comparisons test for a, b, and by unpaired t-test for c, d, e. BP: before pregnancy; W2: second week of pregnancy.

NaTurE METabOlISM
Extended Data Fig. 8 | Pharmacological treatment with D2R biased ligand uNC9994 reverts pregnancy-specific increase in HPF consumption. a, Schematic illustration of experimental strategy (left) and percentage of daily caloric intake consumed (right), during the 2-hours of HPF access, by female mice treated with either vehicle or D2R biased ligand UNC9994 before pregnancy (BP) and during the second week of pregnancy (W2) (n = 6 mice/group). b, Total daily caloric intake (chow + HPF) of female mice treated with either vehicle or D2R biased ligand UNC9994 before pregnancy and during the second week of pregnancy (n = 6 mice/group). c-f, Recorded parameters to assess open field performance in female mice treated with either vehicle or D2R biased ligand UNC9994, including global activity (c) (n = 4 for vehicle and 5 for UNC9994) time spent immobile (d) (n = 7 mice/group) total distance travelled (e) (n = 7 mice/group) and percentage of time spent in each compartment (f) (n = 7 mice/group). Schematic of the open field test highlights center (purple) and corner (cream) zones. Dots in all panels represent individual sample data. Data are expressed as mean ± SEM. Exact P values are shown. Statistical analysis was performed by two-way ANOVA with Sidak's multiple comparisons test for a, b, and by unpaired t-test for c, d, e and by two-way ANOVA with Sidak's multiple comparisons test for f. BP: before pregnancy; W2: second week of pregnancy; VEH: vehicle.

NaTurE METabOlISM
Extended Data Fig. 9 | Characterization of maternal physiology under diverse nutritional paradigms. a, Body weight of females exposed to either ad libitum chow (n = 4), "limited access" paradigm (n = 5) and ad libitum Western diet (n = 6). Females were crossed with Chow-fed males at 12 weeks of age. b, Glucose tolerance test at E14.5-E16.5 pregnant females exposed to either ad libitum chow (n = 4), "limited access" paradigm (n = 5) and ad libitum Western diet (n = 6). c-e, Blood glucose levels (c), plasma insulin (d), and plasma leptin (e) of E14.5-E16.5 females exposed to either ad libitum Chow (n = 4), "limited access" paradigm (n = 5) and ad libitum Western diet (n = 6) after 6 hours of fasting. Dots in all panels represent individual sample data. Data are expressed as mean ± SEM. Exact P values are shown. Statistical analysis was performed by two-way ANOVA with Tukey's multiple comparisons test for a and b. Green P values refer to the comparison between limited access group and Western diet-fed group. Yellow P values refer to the comparison between chow-fed group and Western diet-fed group. Statistical analysis was performed by one-way ANOVA with Tukey's multiple comparisons test for c, d, e. When the factors Diet, Age/Time and/or the interaction Age/Time:Diet were considered significant, results are shown with significant factor or the interaction effect next to it.