Pregnancy damps thermoregulatory rhythms across timescales through neural estrogen signaling in mice

Significance

While maternal hormones have long been thought to mediate physiological changes during pregnancy, how hormonal signals coordinate changes in homeostatic rhythms remains unclear. We show that pregnancy induces an immediate and progressive attenuation of core body temperature rhythms. Using cell-specific transcriptomics, in vivo neural recordings, and targeted neuron and gene manipulations, we demonstrate that estrogen receptor alpha signaling in a thermoregulatory hub, the preoptic area of the brain, helps coordinate pregnancy-induced changes to maternal core temperature and its rhythmicity. Moreover, the silencing of estrogen-sensitive neurons and resulting elevation in maternal core temperature was associated with premature delivery.

Abstract

Pregnancy induces widespread physiological adaptations, including changes to core temperature (T_c), that may reduce fetal heat stress and enhance metabolic stability. Despite the potential importance of maternal thermoregulatory changes, the neural mechanisms organizing homeostatic rhythms during pregnancy remain poorly understood. Here, we demonstrate that pregnancy profoundly reshapes thermoregulatory dynamics across diurnal and ultradian timescales and that these changes are partially orchestrated by estrogen receptor alpha (ERα) signaling in neurons of the medial preoptic area (MPO^ERɑ). Using in vivo calcium recordings of MPO^ERɑ neurons, we show that MPO^ERɑ neural activity is synchronized with T_c oscillations in nonpregnant mice. Notably, this synchronization is lost during pregnancy and reemerges after parturition, indicating that MPO^ERɑ activity and T_c are synchronized in a manner that is dependent on reproductive state. Using chemogenetic inhibition, we show that MPO^ERɑ neurons are required for the lowering of T_c in late pregnancy. Using a genomic estrogen signaling reporter, we demonstrate that MPO^ERɑ neurons exhibit enhanced estrogen signaling in late pregnancy. This estrogen signaling helps coordinate maternal T_c changes, as selectively ablating ERα expression in the MPO increased T_c and its circadian and ultradian rhythmicity. Finally, we show that manipulating MPO^ERɑ cell activity and MPO Esr1 expression affected neonates: Silencing MPO^ERɑ neurons led to premature birth, while both silencing and gene ablation led to sex-specific changes to offspring body mass. In summary, these findings reveal a global suppression of T_c circadian and ultradian rhythmicity during pregnancy and identify ERα as a central modulator of these rhythms with consequences for fetal development.

Fetal temperature is under the control of the maternal thermoregulatory system, which must maintain a stable internal environment to support fetal development (1). Across multiple species, pregnancy is accompanied by reductions in maternal core body temperature (T_c), suggesting a conserved and potentially protective adaptation (2–7). T_c also becomes less variable during pregnancy with reduced circadian (~24 h) rhythm amplitude (5, 8). These changes are posited to protect fetuses against heat damage, support consistent fetal cell divisions, and/or conserve energy for fetal development (7, 9, 10). However, the underlying neural and hormonal mechanisms that control maternal thermoregulatory changes during pregnancy are just beginning to be identified (11, 12).

Recent work links prolactin receptor signaling in the brain to pregnancy-associated T_c reductions and fetal survival following heat exposure in mice (11), but prolactin signaling does not appear to influence T_c circadian rhythmicity (11). Therefore, other maternal hormones are likely playing a role. Estrogen receptor alpha (ERα) signaling is a strong candidate: Estrogen treatment, ERα ablation, and estrogen withdrawal in the nonpregnant state each powerfully modulate T_c, heat generation, heat dissipation, and the thermoneutral zone (TNZ)—the range of ambient temperatures in which T_c is stable without metabolic or thermoregulatory adjustments (13–18). Additionally, there is bidirectional regulation between estrogens and the circadian system that helps align physiology and behavior to reproductive needs (19). Circulating estrogen levels rise during pregnancy in both humans and mice (20–22), temporally aligning with changes in maternal T_c. Estrogens likely act within the medial preoptic area (MPO), where activation of ERα-expressing neurons (MPO^ERɑ) reduces T_c in nulliparous (i.e., never pregnant) mice (23), and where Esr1 regulates a preference for cooler ambient temperatures in postpartum mice (12).

Here, we test the hypothesis that MPO^ERɑ neurons orchestrate pregnancy-associated thermoregulatory remodeling. We report that MPO^ERɑ neurons are necessary for reductions in T_c and its circadian rhythm amplitude. In addition, we show that estrogen signaling within the POA at least partially reduces T_c and the strength of its circadian and 8-h ultradian rhythms in late pregnancy. Manipulating MPO^ERɑ neurons and estrogen signaling in these cells shortens gestational length and sex-specifically impacts pup body mass, suggesting that even subtle changes to T_c and its rhythmicity can affect pup development.

Circadian and Ultradian Core Temperature Rhythms Are Damped during Pregnancy.

To comprehensively assess the effects of pregnancy on T_c and its rhythmicity, we continuously recorded nulliparous and pregnant C57BL/6J mice over ~3.5 wk and analyzed T_c using multiple approaches (Fig. 1 A and B). As T_c dynamically changes across pregnancy, we compared T_c on select days representing early (P1.5), mid- (P10.5), and late pregnancy (P16.5), as well as parturition (P18.5) and postpartum (PP4.5). While 24-h mean T_c did not differ between groups on these days (SI Appendix, Fig. S1A), we confirm light phase-specific effects: Mid-pregnant (P10.5) mice showed elevations in the light phase, while late-pregnant (P16.5) mice showed reductions in the dark phase (Fig. 1 C and D) (5). These changes contributed to reductions in T_c circadian rhythm amplitude in pregnant mice, as previously reported (8, 24) (SI Appendix, Fig. S1 B and C). To determine when the attenuation of T_c circadian rhythms begins and ends during pregnancy and postpartum, we compared T_c circadian amplitude between pregnant and nulliparous mice on each day and report that the decrease emerges at P0.5, progressively declines until parturition, and returns to nonpregnant levels by PP3.5 (Fig. 1E). While 100% of nulliparous mice and postpartum mice (from PP1.5) exhibited T_c circadian rhythmicity, only about 20 to 80% of pregnant mice did, with the rest classified as arrhythmic (SI Appendix, Fig. S1E). We then examined the Fourier spectra of T_c on the last 3 d of pregnancy (P15.5 to P18.5) when T_c changes are most pronounced and found reduced power in the 24-h rhythm and its higher-order harmonics, i.e., 12-h, 8-h, 6-h, and 4-h ultradian (<24-h) rhythms (Fig. 1 F and G). Wavelet analysis of representative subjects suggests that these ultradian rhythms gradually lose power across pregnancy (SI Appendix, Fig. S1D). Locomotion and temperature often covary (25), but temperature shifts in pregnancy appear to be at least partly independent from changes in locomotion (SI Appendix, Fig. S2) (26). For example, during the light phase on P10.5, T_c is elevated compared with nulliparous mice (Fig. 1C) despite similar levels of activity (SI Appendix, Fig. S2C).

Fig. 1.

Core body temperature rhythms are damped during pregnancy. (A) Experimental design used to compare time-matched T_c recordings in nulliparous and pregnant mice. (B) Curves ± CI were fit to T_c data from nulliparous (cyan; n = 7) and pregnant mice (peach; n = 7) using locally estimated scatterplot smoothing (LOESS, span = 0.025). The dashed line marks P19.5/PP0.5 (~parturition). (C) Curves ± CI were fit to 24-h T_c data from pregnant (n = 7) and nulliparous mice (n = 7) on select days of pregnancy and postpartum (LOESS, span = 0.20). AUC analysis comparing nulliparous and pregnant mice per light phase is shown. (D) Mean T_c per light-cycle phase. (E) Daily 24-h rhythm strength determined using the JTK cycle algorithm. (F and G) Mean Fourier power spectrum plots for late pregnancy show the power of peaks corresponding to circadian and ultradian rhythms. Pregnant mice show reduced power at circadian (24-h) and ultradian (12-, 8-, 6-, 4.8-h) frequencies, evident in frequency (F) and period-transformed plots (G). Statistics: Wilcoxon rank-sum tests comparing phase-specific AUC values and average T_c in (C and D). Holm-adjusted P-values are shown. Two-way ANOVA followed by Bonferroni-adjusted post hoc comparisons in (E). In (E), P < 0.05 on each day, unless otherwise shown. Wilcoxon rank-sum test in (G). At 12-h, P = 0.0033 and at all other periods, P = 0.0022 (G). Shaded rectangles indicate the dark phase. Boxplot shows the median, the interquartile range (IQR), and whiskers extending to 1.5 × IQR. Mean ± SEM and individual observations are shown in (E and G). L = light phase; D = dark phase. **P < 0.01, ns = not significant.

Enhanced Estrogen Signaling in a Thermoregulatory Node during Pregnancy.

Estrogen levels begin to rise in mid-pregnancy and peak shortly before birth (20, 21). We tested whether this increase in circulating levels corresponds to enhanced estrogen signaling in the MPO, because estrogen-sensitive neurons in this region have been tied to thermoregulation (12, 23) and maternal care (27, 28). Estrogen signaling modulates the expression of estrogen-sensitive genes, and consequently cell state and physiology, via the binding of estrogen receptors to DNA sites termed estrogen response elements (EREs) (29). We used an AAV-based reporter, “neuro-seeER,” that visualizes estrogen signaling in vivo through the binding of ERE sites that drive the expression of an enhanced green fluorescent protein (eGFP) (Fig. 2A) (30). The eGFP is destabilized with a 2-h half-life for temporal specificity. In addition, neuro-seeER contains LoxP-flanked FusionRed to mark neurons that transduced the reporter, independent of estrogen signaling levels. We delivered neuro-seeER to the MPO of Esr1^Cre females and compared the number of estrogen-responsive (i.e., eGFP+) neurons between nulliparous mice in diestrus, expected to have moderate levels of estradiol (31–33), and pregnant mice at two timepoints (P10.5 and P16.5), before and after the pregnancy-induced rise in circulating 17β-estradiol levels, respectively (20, 21). Our findings suggest that mice at P16.5, but not P10.5, have more neurons engaged in estrogen-dependent gene expression compared with nulliparous mice, based on eGFP+ counts (Fig. 2 B and E) and the proportion of FusionRed+ cells that are eGFP+ (Fig. 2D). This increase in estrogen signaling was observed despite no group differences in the number of cells expressing Esr1 (FusionRed) (Fig. 2 C and E). As estrogen signaling can be engaged by Esr1 (ERα) or Esr2 (ERβ), we used in situ hybridization to compare the number of Esr1- and Esr2-expressing neurons in the MPO in nulliparous and P16.5 mice and confirmed either no difference (Esr1) or a decrease (Esr2) in late pregnancy (SI Appendix, Fig. S3). These findings suggest that estrogen signaling is enhanced in the MPO in late pregnancy, despite no change or a decrease in the number of estrogen-sensitive neurons.

Fig. 2.

The MPO exhibits enhanced estrogen signaling and transcriptomic remodeling in late pregnancy. (A) Neuro-seeER visualizes estrogen-dependent gene expression and was delivered to the MPO of Esr1^Cre female mice. Cre-dependent FusionRed expression confirms viral transduction, while the binding of ERE sites drives the expression of a destabilized eGFP. (B–D) eGFP+ and FusionRed+ cell counts, and their proportion, from two anatomically matched MPO sections (+0.14 and 0.02 mm from bregma). Late-pregnant females (P16.5), but not mid-pregnant (P10.5), had more eGFP+ cells (B) and proportion of eGFP to FusionRed cell counts (D) than nulliparous controls, yet no difference in Esr1-expressing cells based on FusionRed expression (C). (E) Representative images of eGFP and FusionRed fluorescence in the MPO. Gamma adjustments applied to improve signal visualization. (F) Experimental design to compare the transcriptome of Esr1-expressing cells in the MPO of nulliparous and late-pregnant (P16.5 to 18.5) mice. Top upregulated gene expression pathways are associated with cellular energy. Statistics: N per group in (B–D): n = 8 to 10 nulliparous, n = 7 to 8 P10.5, and n = 6 to 9 P16.5. One-way ANOVA in (B–D), followed by Dunnett’s multiple comparisons test if significant. Mean ± SEM shown along with individual data points. N per group in (F): n = 7 diestrous mice and n = 5 late-pregnant mice. (Scale bar, 100 μm.) MPO = medial preoptic area, 3V = third ventricle.

Because pregnancy is accompanied by enhanced estrogen-responsiveness, we transcriptionally profiled estrogen-sensitive neurons from the POA, targeting the MPO using manual dissection, in nulliparous, diestrous mice and late-pregnant mice (P16.5 to 18.5) (Fig. 2F). Esr1-expressing (FusionRed+) cells from mice that received neuro-seeER were isolated via fluorescence-activated cell sorting and sequenced (GeneWiz, South Plainfield NJ). Validating the tool, differential gene expression analysis revealed an upregulation of genes that are estrogen-responsive or encode a component of the estrogen signaling complex, such as Igf1 (34), Cited1 (35), Tgfb2 (36), and Socs2 (37) (SI Appendix, Fig. S4A). To identify broader biological pathways associated with pregnancy, we performed fast gene set enrichment analysis (FGSEA) (38) using Gene Ontology terms. A total of 533 pathways differed between nulliparous and pregnant mice with the top 20 pathways shown in SI Appendix, Fig. S5. Up-regulated pathways were associated with hormone activity (SI Appendix, Fig. S4B) and cellular energy (Fig. 2F). These findings extend our neuro-seeER results, confirming an upregulation of hormone-related genes in Esr1-expressing neurons, and suggest a change in cellular demands and/or neural activity levels during late pregnancy.

Synchronization of Core Body Temperature and MPO^ERɑ Neural Activity in Nulliparous, but Not Pregnant, Mice.

Because MPO^ERɑ neurons regulate T_c (23), and T_c is profoundly altered during pregnancy, we hypothesized that pregnancy impacts MPO neuronal activity and its temporal organization with T_c. First, we used whole-cell current-clamp electrophysiology on slices from ERɑ-ZsGreen mice to compare MPO^ERɑ neural activity between nulliparous, early-pregnant (P7.5), and late-pregnant (P16.5 to 18.5) mice. We found that neurons from late-pregnant mice show reduced firing frequency and lower membrane potential compared with nulliparous mice (SI Appendix, Fig. S6), suggesting reduced intrinsic excitability. Next, we tested whether MPO^ERɑ neural activity levels reflect T_c changes in vivo. Fiber photometry recordings of the calcium (Ca²⁺) indicator GCaMP6s were performed in nulliparous, late-pregnant (P16.5 to 18.5), and postpartum mice (Fig. 3 A–C). We recorded for up to 7.5 h in the dark phase, when T_c differences between nulliparous and late-pregnant mice were most pronounced. We found that T_c and neural activity levels were positively correlated in nulliparous mice, as apparent in group-averaged data (Fig. 3D), as well as in T_c and df/F traces (Fig. 3E) and scatterplots from representative subjects (Fig. 3I and SI Appendix, Fig. S10). Intriguingly, this synchronization was not observed in pregnant mice (Fig. 3 D, F, and J). We also recorded MPO^ERɑ neural activity on the night of parturition when T_c rises and becomes more variable. Parturient mice differed from nulliparous mice and exhibited a negative correlation between T_c and neural activity levels (Fig. 3 D, G, and K), which returned to a positive correlation in the postpartum state (Fig. 3 D, H, and L). Together, these data suggest that MPO^ERɑ neural activity is synchronized with endogenous T_c fluctuations and that this synchronization is reversible and dependent on reproductive state.

Fig. 3.

MPO^ERɑ neural activity and T_c are synchronized in nulliparous, but not pregnant, females. (A) Schematic of experimental design to assess the relationship between T_c and MPO^ERɑ neural activity (normalized changes in Ca²⁺-dependent fluorescence or df/F). (B) Representative image shows fiber placement with ERɑ immunohistochemistry and GCaMP6s viral expression in the MPO. (C) Overlaid viral expression (green) and the area below fiber placement (red crosses) from all mice with approximate bregma positions below. (D) Spearman’s correlation coefficients (r_s) were calculated between T_c and neural activity in nulliparous (NP), pregnant (preg; P16.5 to 18.5), parturient (part), and postpartum (post) mice. Statistical comparisons are relative to nulliparous mice. (E–H) Overlaid T_c and neural activity traces from representative subjects. (I–L) Scatterplots show T_c vs. neural activity from representative subjects. Spearman’s correlation line is shown in red and data in 1-min bins from individual mice shown as blue dots. Statistics: N per group: n = 7 nulliparous, n = 7 pregnant, n = 4 parturient, n = 6 postpartum. A linear mixed effects model (REML) was used to test for group differences in Fisher Z-transformed correlation coefficients; P-values based on sandwich-estimated SE are shown in (D). Boxplot shows the median, the IQR, and whiskers extending to 1.5 × IQR. ***P < 0.001, and ns = not significant. (Scale bar, 100 µm.) ZT = Zeitgeber time, 3V = third ventricle, MPO = medial preoptic area, mm = millimeters.

POA^ERα Neurons Are Required for Core Body Temperature Decreases in Late-Pregnant Mice.

Although MPO^ERɑ neurons show reduced firing frequency in pregnancy, they may still play a role in maintaining pregnancy-induced T_c changes. We acutely silenced MPO^ERɑ neurons and tested for their necessity, since activation of MPO^ERɑ neurons induces a torpor-like state (23), which terminates pregnancy or severely impairs fetal growth (39). To do this, we delivered control virus or Cre-dependent hM4D(Gi) to the MPO of Esr1^Cre mice. Because viral expression was observed in nearby regions, we henceforth refer to our manipulations as POA^ERɑ-specific. Deschloroclozapine (DCZ), a potent, selective, and fast-acting chemogenetic ligand (40) was administered daily to control and hM4D(Gi)-expressing mice at lights off (ZT12) starting from P14.5 until pups were observed (Fig. 4 A–C). hM4D(Gi)-expressing mice exhibited larger T_c peaks after DCZ administration, compared with control mice (Fig. 4 D–F). As a result, hM4D(Gi)-expressing mice had stronger T_c circadian rhythm amplitude (Fig. 4 G–I). Ultradian rhythms were not affected, likely due to the transient effects of daily DCZ (40). Interestingly, half of hM4D(Gi)-expressing mice delivered prematurely (Fig. 4J) and had heavier and larger male pups (Fig. 4 K and L) without changes in litter size compared with controls (Fig. 4M). Thus, our findings suggest that POA^ERα neural activity is necessary to reduce T_c in late pregnancy, and blunting their activity has consequences for offspring.

Fig. 4.

POA^ERɑ neural activity is required for the lowering of T_c in pregnant mice. (A) Experimental design for hSyn-hM4D(Gi)-mediated acute inhibition of POA^ERɑ cell activity. Chemogenetic ligand, DCZ, was orally administered daily to all mice from P14.5 until delivery at lights off (ZT12). (B) Representative image of viral expression in the POA. (C) Schematic depicts hSyn-hM4D(Gi) expression across all mice with approximate bregma positions below. (D) Curves ± CI were fit to T_c data from control (mCherry) and hM4D(Gi)-expressing mice (LOESS, span = 0.025). Dashed lines indicate DCZ administration. (E) Curves ± CI were fit to T_c in a 24-h window, averaged across treatment days (LOESS, span = 0.05). (F) Δ T_c (2-h post-DCZ−1-h pre-DCZ). (G) hM4D(Gi)-expressing mice exhibit enhanced circadian (24-h) rhythm strength. (H and I) Mean Fourier power spectrum plots show increased power at circadian (24-h), but not ultradian (12-, 8-, 6-, 4.8-h) frequencies. (J) About half of hM4D(Gi)-expressing dams delivered preterm on P18.5, while all control dams delivered by P19.5. (K–M) Male pups from hM4D(Gi)-expressing dams were heavier (K) and longer (L) than controls on postnatal day (PN) 1, despite similar litter size (M). Statistics: N per group in (K and L): n = 33 male mCherry, n = 31 female mCherry, n = 27 male Gi, n = 24 female Gi. Wilcoxon rank-sum tests were used to test group differences each day with Holm-adjusted P-values shown in (F). Two-way ANOVA in (G); pairwise comparisons were not significant on any day using Holm–Sidak adjustments. Wilcoxon rank-sum tests in (I and M), and Fisher’s exact test in (J). Aligned rank transform (ART) ANOVAs followed by Holm-adjusted ART contrasts in (K and L). Shaded rectangles indicate the dark phase. Boxplots show the median, the IQR, and whiskers extending to 1.5 × IQR. Mean ± SEM and individual observations are shown in (G and I). ***P < 0.001 and ns = not significant. (Scale bar, 100 µm.) AC = anterior commissure, MnPO = median preoptic nucleus, MPO = medial preoptic area, MPN = medial preoptic nucleus. LPO = lateral preoptic area, VMPO = ventromedial preoptic nucleus, PVN = paraventricular nucleus, SCN = suprachiasmatic nucleus, VLPO = ventrolateral preoptic nucleus, 3V = third ventricle, Och = optic chiasm, mm = millimeters.

Estrogen Signaling in the POA Coordinates Decreases in Core Body Temperature and Its Rhythmicity during Pregnancy.

Subsets of POA^ERɑ neurons coexpress genes shown to influence T_c, such as Bdnf, Adcyap1, Brs3, and Prlr (11, 41–43). To dissect the specific role of estrogen signaling in pregnancy-induced thermoregulatory changes, we knocked out Esr1 in the POA (Fig. 5 A–C). The effects of POA^Esr1 KO on T_c were dynamic (Fig. 5D). POA^Esr1 KO mice had subtle elevations in T_c during the dark phase on P16.5 compared with controls, but showed no difference on P10.5 (Fig. 5 E and H). The timing is consistent with our neuro-seeER findings indicating enhanced estrogen signaling at P16.5, but not P10.5 (Fig. 2). Moreover, while 24-h mean T_c did not differ (SI Appendix, Fig. S7A and Table S1), POA^Esr1 KO mice exhibited increased T_c circadian rhythm strength and amplitude than controls across pregnancy (SI Appendix, Fig. S7 B and C). Focusing on late pregnancy, we detected higher power at frequencies associated with both circadian (~24-h) and 8-h ultradian rhythms in POA^Esr1 KO mice compared with controls (Fig. 5 J and K). The effects of POA^Esr1 KO on T_c do not appear to be driven by gross motor activity as activity counts did not differ between groups in mid- or late pregnancy (SI Appendix, Fig. S8). Thermogenesis may be altered as POA^Esr1 KO mice had increased BAT mass after birth, despite no change in body mass, although further exploration is warranted (SI Appendix, Fig. S9). Female pups birthed from POA^Esr1 KO dams were smaller than those from controls (Fig. 5L), without a difference in gestation length or litter size (SI Appendix, Table S1). In summary, our data demonstrate that Esr1 expression in the POA is required to fully exhibit the pregnancy-induced damping of T_c and its rhythmicity.

Fig. 5.

Estrogen signaling ablation in the POA increases T_c and its circadian and ultradian rhythmicity, blunting pregnancy-induced changes. (A) Esr1^f/f mice were stereotaxically delivered control or Cre virus in the POA to ablate Esr1 expression. Representative images of ERɑ labeling and viral expression in control and Esr1 KO mice. (B) Viral targeting of Cre recombinase with approximate bregma positions indicated below. (C) ERɑ cell count density (per 1,000 pixels²) in the MPO in anatomically matched sections (~bregma 0.02). (D) Curves ± CI were fit to T_c data across pregnancy (LOESS, span = 0.025). (E and F) Curves ± CI were fit to T_c data within a 24-h window on P10.5 (E) and P16.5 (F; LOESS, span = 0.20). (G and H) Mean T_c per light-cycle phase on P10.5 (G) and P16.5 (H). (I) Daily rhythm strength determined using the JTK cycle algorithm. (J and K) Mean Fourier power spectra revealed increased power at circadian (24-h) and ultradian (8-h) frequencies in Esr1 KO mice compared with controls. (L) Offspring body weight on postnatal day (PN) 1. Statistics: N per group: n = 8 controls and n = 8 Esr1 KO in (C); n = 11 controls and n = 8 to 9 Esr1 KO in (D–H, J, and K); n = 7 controls and n = 6 Esr1 KO in (I); n = 28 control males, n = 23 control females, n = 32 Esr1 KO males, n = 18 Esr1 KO females in (L). t tests were used in (C). Wilcoxon rank-sum tests comparing phase-specific AUC values in (E and F). Wilcoxon rank-sum tests in (G, H, and K). Two-way ANOVA, followed by Bonferroni-adjusted pairwise comparisons in (I). ART ANOVA followed by Wilcoxon rank-sum tests for preplanned pairwise comparisons in (L). Shaded rectangles indicate the dark phase (D). Boxplots show the median, the IQR, and whiskers extending to 1.5 × IQR. Mean ± SEM are shown in (C, I, and K). *P < 0.05, ***P < 0.001, ns = not significant. (Scale bar, 100 µm.) AC = anterior commissure, MnPO = median preoptic nucleus, MPO = medial preoptic area, MPN = medial preoptic nucleus. LPO = lateral preoptic area, VMPO = ventromedial preoptic nucleus, PVN = paraventricular nucleus, SCN = suprachiasmatic nucleus, VLPO = ventrolateral preoptic nucleus, 3V = third ventricle, Och = optic chiasm, mm = millimeters.

Discussion

Here, we show that T_c rhythms are flattened across multiple timescales during pregnancy, indicating broad remodeling of thermoregulatory control. Although the weakening of harmonic, ultradian rhythms (12-h, 8-h, 6-h, and 4.8-h) possibly reflect the damping of the 24-h fundamental, there are indications that at least some ultradian rhythms are independently regulated (44, 45). Irrespective of their interdependence, damping of circadian and ultradian T_c rhythms during pregnancy may represent an adaptive strategy to stabilize the fetal thermal environment during critical developmental windows. As maternal and fetal temperatures are tightly coupled (10) and fluctuations in maternal T_c are transmitted to the fetus (46, 47), reducing rhythmic amplitude and variability could minimize thermal stress, particularly during periods of rapid cell proliferation, differentiation, and organogenesis. A more thermally stable internal environment may also buffer the fetus from oscillations in maternal metabolism, limiting fluctuations in oxygen consumption, mitochondrial activity, and reactive oxygen species production that could otherwise compromise developmental fidelity (10, 48, 49).

Using ex and in vivo neural activity recordings and a loss-of-function manipulation at the cellular level, we identified estrogen sensitive neurons in the POA as mediating the suppression of T_c in late pregnancy. Our fiber photometry data suggest that neural activity levels in MPO^ERɑ neurons track nocturnal fluctuations in T_c in nulliparous mice, with higher T_c values associated with higher neural activity. This pattern is consistent with warm-responsive neurons in the POA, which promote cooling and show higher activity at higher temperatures (50). Since activation of MPO^ERɑ neurons lowers T_c (23), we initially predicted that MPO^ERɑ neurons would show enhanced activity during pregnancy. Instead, we found reduced neuronal excitability in late-pregnant mice relative to nulliparous mice. Our findings align with recent work showing that galanin-expressing MPO neurons, coexpressing Esr1 (27), also exhibit decreased baseline firing rates in pregnant mice (51). Notably, Ammari et al. report that while most galanin-expressing MPO neurons become silenced during pregnancy, a subset becomes markedly more excitable in an Esr1-dependent manner (51). Together, these findings suggest that pregnancy may induce a redistribution of excitability within Esr1+ populations rather than a uniform suppression. Given the heterogeneity of Esr1+ neurons (27) and the observation that silencing increased maternal T_c, it is possible that a specialized subset of more excitable Esr1+ neurons may actively suppress T_c during pregnancy.

Next, we used a novel estrogen signaling reporter and viral-mediated ERɑ knock-out to show that estrogen signaling in the POA at least partially contributes to maternal thermoregulatory changes. POA Esr1 KO mice had similar 24-h mean T_c as controls throughout pregnancy, however, they showed stronger circadian and 8-h ultradian T_c rhythms. At P16.5, but not P10.5, we also observed mild elevations in T_c specifically in the dark phase, aligning with enhanced neuro-seeER expression. Because our manipulations reduced, but did not fully prevent, maternal T_c changes, ERɑ signaling likely shapes T_c and its rhythms in coordination with other mechanisms. Consistent with this idea, recent work shows that prolactin receptor (Prlr) signaling in the POA lowers T_c in pregnant mice, but does not alter circadian T_c amplitude (11). We propose that ERɑ signaling orchestrates circadian and ultradian thermoregulatory rhythms through Prlr-independent mechanisms while partially contributing to reduced T_c in pregnancy through Prlr, in line with evidence that ERɑ regulates Prlr expression (52).

We also show that manipulations that partially elevate T_c and restore thermoregulatory rhythmicity during pregnancy have consequences for offspring. POA^ERɑ neuronal silencing increased rates of premature birth and body size in males, whereas Esr1 ablation decreased body size in females. These sex-specific growth responses may reflect differences in the timing of the manipulations and/or the rhythmic components affected: Silencing was restricted to late gestation and primarily strengthened circadian rhythmicity, whereas KO effects spanned pregnancy and enhanced both circadian and ultradian rhythms. These findings align with human studies linking prenatal heat exposure to altered fetal growth (53) in sex-specific ways (54), and with the hypothesis that female offspring reduce growth in response to mild gestational stressors, whereas males increase growth to meet these challenges (54, 55). We cannot exclude contributions from altered early maternal–pup interactions, although POA Esr1 KO dams were observed nursing pups.

In conclusion, our findings position thermoregulatory rhythmicity as a tunable and state-dependent feature of female physiology and suggest that rhythmic attenuation represents an underappreciated adaptation of pregnancy shaped by estrogen signaling. The simultaneous damping of circadian and ultradian T_c rhythms during pregnancy suggests a coordinated reorganization of temporal control, rather than a circadian clock failure. Indeed, circadian clock genes appear to retain rhythmic expression in the anterior hypothalamus during pregnancy (56). Instead, pregnancy may shift the output of T_c-regulating neurons into a low-amplitude mode that minimizes physiological variability. Thus, pregnancy may preserve central circadian clock function while modulating the amplitude of its outputs, perhaps by weakening coupling between the SCN, thermoregulatory circuits in the POA, and peripheral oscillators in metabolic and thermogenic tissues (8, 56, 57). Estrogen signaling in the POA may serve as the cue to flatten rhythms across timescales. Viewed in this light, pregnancy supports a model of state-dependent circadian flexibility in which mammals may dynamically adjust their rhythmicity to meet physiological demands [i.e., growing fetus(es)]. Together, these results suggest that temperature rhythms, in addition to temperature alone, may be a critical dimension of maternal–fetal health.

Materials and Methods

Ethics and Husbandry.

Mouse studies were carried out in approved facilities at the University of California Los Angeles (UCLA) or Pennington Biomedical Research Center at Louisiana State University (LSU). All procedures were approved by the UCLA or LSU Institutional Animal Care and Use Committee (IACUC) and followed the NIH guide for the care and use of laboratory animals. Mice were housed under 12:12 light–dark cycle [zeitgeber time (ZT) 0 = 06:00 or 07:00 AM] at an ambient temperature of 22 to 23 °C, and ~30% relative humidity. Mice had ad libitum access to water and breeder chow (Mouse Diet 9F 5020, 20% protein, 9% fat, LabDiet; St. Louis, MO). Experimental mice were typically between 7 and 12 wk of age at the start of the experiments.

Mice.

C57BL/6J mice (JAX Stock #000664) were obtained from the Jackson Laboratory. Esr1^Cre mice (JAX stock #017913) were maintained heterozygous on a C57BL/6J background (58). Esr1^f/f mice were maintained homozygously on a mixed B6;129P background (59). Female mice underwent stereotaxic viral microinjection and probe implantation surgeries approximately a week apart. After recovery, singly housed nulliparous female mice were either paired with males (pregnancy groups) or left unpaired (nulliparous groups). Paired females were checked daily for the presence of a copulatory plug, which denoted pregnancy day 0.5. Pregnancy onset was further verified by characteristic reductions in the circadian amplitude of T_c (24) and weight gain. Parturition typically occurred during the night around P19.0, with most litters present by the morning of P19.5, which also corresponds to postpartum day (PP) 0.5.

Stereotaxic surgery.

Viral constructs and infusion volumes are listed in SI Appendix, Table S3. AAVs were targeted to the MPO in adult mice at the following coordinates: x = ±0.35 to 0.40 mm; y = +0.30 to 0.40 mm, and z = 5.20 mm from the brain. Viruses were infused at a rate of 5 nL/s using pulled glass needles attached to Hamilton syringes. Although Esr1 expression is abundant in the MPO, viral expression extended to adjacent nuclei, such as the median preoptic nucleus and lateral preoptic area. Therefore, we refer to our chemogenetic and KO manipulations as POA-specific and include targeting illustrations for all experiments. Importantly, no viral expression was observed in the SCN, consistent with the reported low expression of Esr1 in this region in rats and mice (60–62).

Neuro-seeER.

Estrogen responsiveness was visualized using a recently developed dual-function AAV (30). Neuro-seeER contains estrogen response elements (EREs), which upon binding by ligand-bound estrogen receptors, drive the expression of a destabilized eGFP. The eGFP has a ~2-h half-life, allowing for temporally specific readouts of estrogen-dependent gene expression. Neuro-seeER also contains a Cre-dependent FusionRed under the control of the human synapsin (hSyn) promoter which labels transduced neurons. Neuro-seeER was targeted to the MPO of Esr1^Cre mice. After 10 to 14 d, a subset was paired with studs and the rest were used as nulliparous controls. Estrous staging was performed in control mice by assessing cell composition after vaginal lavage, which was conducted daily for at least two consecutive, estrous cycles between ZT 2 and 4 using 0.9% saline (63). Nulliparous mice were killed humanely in diestrus with moderate (31, 33) to high (32) levels of circulating estradiol expected, relative to other estrous stages, while pregnant mice were euthanized on either P10.5 or P16.5 (most at ZT 8 to 9.5). Brains were imaged on a Nikon Eclipse Ti-2E microscope. The Image J Stitching plug on was used to stitch together images (64). eGFP+ and FusionRed+ cells were manually counted in the MPO in two anatomically matched sections using the Cell Counter Plug-in. Two experimenters, blind to group identity, analyzed each section. Animals or hemispheres with poor stereotaxic targeting of the MPO were removed from analysis, as was one outlier detected using the GraphPad Grubb’s test. To account for differences in viral transduction, the proportion of eGFP+ cells to FusionRed+ cells are also reported per group.

Flow-Sequencing.

Adult Esr1^Cre female mice were stereotaxically delivered neuro-seeER to the MPO via stereotaxic surgery and were euthanized in diestrus (control; between ZT3 and 6.5) or late pregnancy (P16.5 to P18.5; between ZT3 and 9) after 3.5+ wk of viral expression. Mice were deeply anesthetized with isoflurane and rapidly decapitated. Brains were immediately placed in a chilled brain matrix, and a 2-mm-thick coronal section, containing the POA, was sliced using RNAse-free razor blades. The tissue slab was immediately placed in chilled calcium-free Hibernate A medium, and the MPO was microdissected and enzymatically digested. The tissue underwent fluorescence activated cell sorting (FACS) to isolate live, nucleated FusionRed+ (i.e., Esr1+) neurons (see SI Appendix, Supplementary Methods for details). Sequencing of these neurons was performed at GeneWiz.

Transcriptomic Analysis.

Downstream analyses of RNA-seq data were performed using https://usegalaxy.org (v25.0.2) and RStudio (v2023.09.1+494). Raw reads were checked for quality using FastQC (v0.74), and adapter sequences and low quality bases were removed using Trimmomatic (v0.39) (65). Cleaned reads were quasi-mapped to the Mus musculus reference transcriptome (GRCm39) using Salmon (v1.10.1) (66, 67) with default parameters set to estimate transcript-level abundance. Differential gene expression analysis was performed using DESeq2 (v1.40.2) (68) to compare pregnant mice to nulliparous controls. Volcano plots and pathway enrichment analyses (38) were performed in RStudio. The “fgsea” package was run with a fixed random seed of 42, and genes were compared against Gene Ontology gene sets (69, 70) with set size parameters between 10 and 500. Genes were ranked based on signed P values, based on log₂ fold-change (FC) [sign(log₂FC) × – log₁₀(P-value)]. Volcano plots were generated using log₂FC and −log₁₀ of the nonadjusted P-value.

Fiber Photometry.

Esr1^Cre mice received DIO-GCaMP6s and a fiber implant in the left hemisphere during the same surgery. GCaMP6s was targeted to the MPO where ERɑ+ cells are concentrated. The fiber (400 μm) was implanted 5.1 mm deep from bregma and affixed using dental cement anchored with a screw. After recovery, mice were implanted with a probe recording T_c in 1-min bins. After a 4-wk+ delay, fiber photometry recordings were performed in the home cage from ZT 12:15 until ~20:00 (about ~7.5-h of recording), except for the postpartum group (recorded until ~18:30). Most nulliparous mice were in metestrus the morning of the recording session, while a separate group of late-pregnant mice were between P16.5 and P18.5. Pregnant mice were also recorded on P18.5 and included in the parturition group if they delivered pups. Postpartum recordings were performed in dams with culled pups, 5 to 24 d after birth (median = 13.5 d postbirth). Recordings were performed as described previously (23, 71, 72). For the calcium (Ca²⁺)-dependent signal, a 470 nm LED was used, while Ca²⁺-independent isosbestic control signal used a 405 nm LED. LED light was delivered at low power and resulted in approximately 2 μW of 470 nm light and 2 μW of total light at the tip of the fiber. The frame rate was set to 4 Hz with alternating LEDs, as previously described (42). Neural data were reported as normalized changes in fluorescence over average fluorescence (df/F; see SI Appendix, Supplementary Methods for details). Signals were then down-sampled by averaging over consecutive 1-min bins and aligned with T_c data. Spearman’s correlation coefficients (r_s) were then computed between the time-aligned neural and temperature vectors. These correlations were done during the entire recording sessions (including for the parturition group). For statistical analyses, correlation coefficients were Fisher-Z transformed [z = arctanh(r_s)] using RStudio. Using a linear mixed model allowed us to use all recordings and account for nonindependence that resulted from one nulliparous subject that was recorded twice. The linear mixed model was fit using restricted maximum likelihood (REML) with mouse ID as a random factor. In addition, we implemented a sandwich estimator (CR2) to provide SE robust to small n and unequal variance between groups.

Chemogenetic Silencing.

DCZ (powdered form) was mixed in peanut butter for a stock concentration of 0.4 mg/g. Mice were administered 0.7 mg/kg of DCZ. Mice were weighed 2- to 3-h before the experiment began to determine the dose. The drug was orally administered in the home cage. Most mice readily ate the peanut butter within 5-min after 1-2 habituation runs using just peanut butter. Pregnant mice expressing either control virus (mCherry) or hM4D(Gi) in the POA were all administered DCZ at ZT12 from P14.5 to 18.5. The day following birth, mice were euthanized. As a high number of DCZ-treated pregnant mice delivered prior to the last dose, temperature analysis for that experiment was restricted to P14.5 to 17.5.

Telemetry Analysis.

Temperature and activity data were processed and analyzed in RStudio (2025.09.0+387; R v4.5.1). “Dplyr,” “lubridate,” “tidyverse,” “ggplot2,” and “patchwork” packages were used for processing and visualization. Individual mouse T_c traces are shown in SI Appendix, Figs. S11–S13. “Lme4,” “emmeans,” “car,” “afex,” “rstatix,” “pracma,” and “clubSandwich” packages were used for statistical analyses. For visualization, locally estimated scatterplot smoothing (LOESS) was applied to T_c data. The LOESS span determines the proportion of the observations (recorded every 5-min) used in the local regression sliding window, i.e., the degree of smoothing. T_c data were analyzed using area under the curve (AUC) and mean T_c analyses. AUC and mean T_c were computed per light phase for each mouse. AUC was computed by numerically integrating values across time using the “trapz” function from “pracma.” Group differences in AUC and mean T_c values were tested using Wilcoxon rank-sum tests per light phase. Group differences in changes in T_c (ΔT) were also analyzed using Wilcoxon rank-sum tests. Experiment-specific details provided in SI Appendix, Supplementary Methods.

JTK and Fourier analysis of T_c circadian and ultradian rhythms.

Nitecap software (73) was used to implement JTK_CYCLE (74) to test the strength of rhythmicity per day using the Jonckheere–Terpstra–Kendall (JTK) algorithm with period constrained to approximately 24-h. Rhythms with Benjamini–Hochberg adjusted JTK q < 0.05 were considered significantly rhythmic; traces with q ≥ 0.05 were classified as not significantly rhythmic, i.e., arrhythmic. JTK amplitude was computed by fitting a moving quadratic regression and taking the difference between the peak and trough height of the smooth curve in 24-h intervals.

Raw and unbinned T_c data were also analyzed using Fourier-based spectral methods to identify oscillatory patterns as specified in ref. 45. Analyses were performed in MATLAB (vR2023b) using the Fast Fourier Transform function and the Continuous Wavelet Transform function. Power distributions for specific frequency components were compared between groups using the Wilcoxon rank-sum test.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

RNA sequencing data have been deposited in GEO—Gene Expression Omnibus (accession number GSE346233) (75) and dryad (76).