Ultradian rhythms of CRH^PVN neuron activity, behavior, and stress hormone secretion

Significance

Ultradian rhythms are present in many biological systems; however, the underlying neural mechanisms responsible for generating these rhythms are unclear. Corticotropin-releasing hormone (CRH) neurons control corticosteroid levels in the body as well as arousal and behavior. Here, we show that CRH neurons exhibit a pronounced ultradian rhythm in neural activity over the 24-h day. CRH neural activity was highly correlated with, and predictive of, behavioral arousal. However, CRH neural activity was not as well correlated with pulses of corticosteroid secretion. Together, these data reveal the patterns of CRH neuron activity over the 24-h day and their relationship with behavior and stress hormone secretion.

Abstract

The stress axis is always active, even in the absence of any threat. This manifests as hourly pulses of corticosteroid stress hormone secretion over the day. Corticotropin-releasing hormone (CRH) neurons in the paraventricular nucleus of the hypothalamus (CRH^PVN) control both the neuroendocrine stress axis as well as stress-associated behaviors. However, it is currently unclear how the resting activity of these neurons is coordinated with both spontaneous behavior and ultradian pulses of corticosteroid secretion. To investigate this, we performed fiber photometry recordings of CRH^PVN neuron activity in Crh-Ires-Cre mice and a newly generated line of Crh-Ires-Cre rats. In both mice and rats, CRH^PVN neurons displayed an ultradian rhythm of activity with reoccurring upstates of activity approximately once per hour over the 24-h day. Upstates in activity were coordinated with increases in animal activity/arousal. Chemogenetic activation of CRH^PVN neurons was also sufficient to induce behavioral arousal. In rats, increases in CRH neural activity preceded some pulses of corticosteroid secretion but not others. Thus, while CRH^PVN neurons display an ultradian rhythm of activity over the 24-h day that is coordinated with behavioral arousal, the relationship between CRH^PVN activity and pulses of corticosteroid secretion is not one-to-one.

Ultradian rhythms are common in biological systems and are observed in unicellular organisms through to vertebrates. In mammals, ultradian rhythms exist in many systems and result in oscillating gene expression, hormone secretion, and behavior over the 24-h day (1, 2). Indeed, pronounced ultradian rhythms are observed in arousal, locomotion, body temperature, and stress hormone secretion (1–7). Current evidence suggests that ultradian oscillations in different systems are generated by distinct neural circuits which employ unique neural mechanisms to generate these rhythms.

The hypothalamic–pituitary–adrenal (HPA) axis shows a prominent ultradian rhythm of activity (8–11). This is characterized by pulses of corticosteroid secretion that occur approximately once every 60 to 90 min (7). Ultradian rhythms in the HPA axis are critical for normal physiological function in mammals and can entrain the expression of glucocorticoid- responsive genes throughout the body and brain (12). Disrupting these rhythms can disrupt numerous aspects of physiology ranging from stress axis responsiveness to cognitive function and metabolism (11, 13). Despite the importance of ultradian HPA rhythms, the mechanisms which generate them are still contentious.

Corticotropin-releasing hormone (CRH) neurons in the paraventricular nucleus of the hypothalamus (CRH^PVN) control the HPA axis. CRH^PVN neurons are activated in response to aversive stimuli (14–16) which in turn drives CRH secretion from the median eminence, activation of pituitary corticotroph cells and the subsequent release of adrenocorticotropic hormone (ACTH). ACTH travels to the adrenal gland to stimulate de novo synthesis and secretion of corticosteroids. While ultradian rhythms in CRH secretion have been observed previously (17–21), the reported period of these rhythms is variable ranging from minutes to hours. In addition, other studies show that ultradian pulses of corticosteroid secretion can be generated even if CRH secretion is clamped at a constant level (22). Therefore, questions remain as to whether ultradian patterns of CRH^PVN neuron activity exist and how they correlate with ultradian corticosteroid secretion.

In addition to projections to the median eminence to control the HPA axis, CRH^PVN neurons also have axonal projections to other brain nuclei (23–27). These central projections are important for the regulation of arousal and stress-associated behaviors (28). While there is a large body of data showing that both CRH peptide or corticosteroids regulate arousal (29), only recently has a direct role for CRH^PVN neurons been demonstrated. Specifically, optogenetic activation of CRH^PVN neurons was shown to promote wakefulness via projections to lateral hypothalamus orexin neurons (26, 27). Likewise, chemogenetic inhibition or ablation of CRH^PVN neurons decreased locomotor activity and wakefulness (26). Using fiber photometry recordings of CRH^PVN neuron activity, one study observed increases in neural activity coincident with shifts from sleep to wake (26). However, another investigation using similar methods did not observe this coincident activity (27). One limitation of these studies is that they only analyzed CRH^PVN neural activity over time frames of seconds to minutes, precluding the detection of ultradian rhythms that manifest over longer periods. What is also unclear from previous work is how spontaneous CRH^PVN neuron activity is coordinated with both behavior and ultradian pulses of corticosterone secretion.

Results

GCaMP6s Fiber Photometry Recordings Reveal Ultradian Rhythms in CRH^PVN Activity.

To record the activity of the CRH^PVN neuron population in nonstressed freely behaving mice, we used GCaMP6s fiber photometry (Fig. 1 A–C and Materials and Methods). On the day of recordings, mice were connected to the photometry system and then left undisturbed for 24 h with free access to food and water in their home cage while CRH^PVN GCaMP6s activity was recorded along with behavior with an overhead camera.

Fig. 1.

Ultradian rhythms of CRH^PVN neuron activity in freely behaving male mice. (A) GCaMP6s was targeted to CRH^PVN neurons via stereotaxic injection. (B) Fluorescence image showing the expression of GCaMP6s (green) in CRH neurons (red) within the PVN. (C) Higher magnification confocal image demonstrating colocalization of GCaMP6s (green) with CRH neurons (red) in the PVN. (D and E) Example photometry recordings of CRH^PVN activity showing spontaneous ultradian upstates over 24-h. Light off at ZT12, light on at ZT24. The dotted box in e is shown expanded in (F). (G) Interupstate interval (min) shown for night (black) versus day (orange). Individual data plotted on the Left and mean data plotted on the Right. (H) Upstate duration (min) comparison between night and day. Individual data on the Left, mean data on the Right. (I) Number of upstates observed compared between night and day.

To allow for long-duration recordings without photobleaching, we scheduled the acquisition of photometry recordings with a 3 s on/7 s off (30%) duty cycle. We averaged the data collected during the “on” period such that we effectively sampled CRH^PVN population activity once every 10 s (0.1 Hz). This approach revealed slow ultradian events which we termed “upstates” (Fig. 1 D–F). Over the 24-h day, there were on average 18.1 ± 0.9 upstate events (n = 7) with a mean duration of 27.4 ± 1.1 min each (n = 7). The frequency of these upstate events is similar to the frequency of ultradian corticosteroid pulses previously reported in rats (30, 31). The interupstate interval and upstate duration were highly variable both within a recording and between mice (Fig. 1 G and H). While interupstate interval was significantly longer during the day compared to the night when all event intervals were analyzed (P = 0.008), this was not significant if the analysis was restricted to mean data per mouse (P = 0.10). Upstate duration was not significantly different between day and night when all events were analyzed or if compared within mice (Fig. 1H, P > 0.05), neither was number of upstates (Fig. 1I, P = 0.07). In between as well as during upstates, there was ongoing neural activity, consistent with the basal activity previously reported in the CRH neuron population (16, 32, 33). However, the low sampling rate used for these experiments precluded the analysis of these fast events.

Coordination of Spontaneous CRH^PVN Activity and Behavioral Arousal.

Past work has suggested that CRH^PVN neurons are involved in regulating the sleep–wake cycle and behavioral arousal (26, 27). To determine whether there was any correlation between spontaneous CRH^PVN neuron activity and behavioral arousal in our recordings, we analyzed head movement in our 24-h video recording with DeepLabCut (34) (Fig. 2A, blue trace). Movements occurred over the entire 24-h day and ranged from small head movements while the mouse remained within their nest, to locomotion where the mouse moved around the entire cage. When we plotted head distance moved per 10 s bin, we observed periods of high levels of movement that reoccurred over the day (referred to as bouts). We analyzed the number of bouts of head movement over the day/night cycle and found that there were 9.3 ± 1.1 bouts of movement during the dark phase compared to 6.4 ± 0.6 during the light phase (n = 7, P = 0.052). We aligned the onset or offset of each bout of movement with photometry. We observed that CRH^PVN activity increased at approximately the same time as a movement bout was initiated. Likewise, CRH^PVN activity decreased at approximately the same time as a bout of movement was terminated, however, remained slightly elevated for longer (Fig. 2 B–D).

Fig. 2.

Correlation between CRH^PVN activity and behavioral arousal in male mice. (A) Example photometry recording showing spontaneous ultradian pulses across the 24-h day (orange) and corresponding 24-h head movement profile (blue). (B) Heatmap of CRH^PVN activity before and after the onset and offset of movement bouts in 7 mice. (C) Movement profiles corresponding to neuronal activity data shown in (B). (D) Time-locked analysis of neuronal activity and movement around the onset and offset of a bout of movement. Red and blue lines represent average fluorescence and movement, respectively, aligned to the onset and offset of a bout of movement. (E) State-space Granger causality (GC) analysis showing relative differences between estimated ΔF/F₀ → movement (red line) and movement → ΔF/F₀ (black line) causal influences in GCaMP6s expressing mice. Solid lines are the experimental data. Dotted lines are the shuffled and time-shifted control data. (F) Convergent cross mapping (CCM) analysis showing relative differences between estimated ΔF/F₀ → movement (red line) and movement → ΔF/F₀ (black line) causal influences in GCaMP6s expressing mice. Solid lines are the experimental data. Dotted lines are the shuffled and time-shifted control data.

To investigate the temporal relationship between GCaMP6s CRH^PVN signals and movement, we performed state-space Granger causality (GC) analysis and Convergent Cross Mapping (CCM) (Fig. 2 E and F). These test whether one time series (i.e. GCaMP) is useful in forecasting another (i.e. movement). We use these nonlinear approaches to mitigate potential issues arising from dissimilar data distributions from the two modeled time series. This analysis revealed that ΔF/F₀ was more predictive of movement (orange solid line in Fig. 2 E and F) than in the opposite direction (movement predicting ΔF/F₀, black line). This relationship was not seen for ID-shuffled and time shifted data (dotted lines in Fig. 2 E and F). This suggests the direction of coupling is driven by changes in CRH^PVN neuron activity. Both types of analysis show a predominant causal influence of ΔF/F₀ to movement, particularly at lower frequencies. Together, these data show that changes in CRH^PVN neuron GCaMP6s fluorescence are predictive of animal activity, but not vice-versa.

As a control experiment, we injected a Cre-dependent GFP-virus into the PVN of Crh-Ires-cre mice and implanted an optic fiber probe as in the GCaMP6s mice. We then recorded photometry signals and behavior using the same protocols as above. In these recordings, there was no evidence of distinct events or upstates despite bouts of animal movement (SI Appendix, Fig. S1A). We next repeated the state-space GC and CCM analysis (SI Appendix, Fig. S1 E and F). For GFP recordings, since there is no fluctuation of fluorescence coupled to neural activity, the ΔF/F₀ essentially reflects noise and background fluorescence. Consistent with this, the state-space GC estimates are low and do not indicate a preference for either direction. For the CCM analysis, correlations were low and not markedly different from the ID-shuffled and time shifted data. This suggests that the relationship observed in the GCaMP6s recordings is due to neural activity and not movement artifact.

Chemogenetic Activation of CRH^PVN Neurons Leads to Elevations in Home Cage Activity.

We showed that CRH^PVN neuron activity precedes and predicts increases in behavioral arousal. We next tested whether artificial activation of CRH^PVN neurons could change home cage behavior. Crh-Ires-Cre mice were bilaterally injected with AAV to drive the expression of either hM3Dq to excite CRH^PVN neurons, hM4Di to inhibit CRH^PVN neurons or tdTomato as a control virus. After a minimum of 3 wk recovery, mice were habituated to having their home cage placed in a white testing box with an overhead camera. On the day of testing, mice were given a single injection of deschloroclozapine (DCZ) and home cage activity was recorded. Prior to DCZ injection, mice in all groups displayed episodic bouts of activity. Following DCZ injection, hM4Di and control mice continued to show episodic bouts of activity (Fig. 3 A and C). However, hM3Dq mice showed a pronounced increase in home cage activity that remained elevated for approximately 2 h (Fig. 3B). We compared the mean level of home cage activity in 30 min bins (Fig. 3D). A two-way ANOVA revealed no significant effect of group [F_(2,17) = 2.67, P = 0.09], but a significant effect of time [F_(2.9,46.8) = 11.53, P < 0.0001] and a significant interaction [F_(22,177) = 5.14, P < 0.0001]. Home cage activity was significantly different between the hM3Dq and other groups at 30 min and 1 h postinjection (P < 0.01). On a subsequent day, the effect of DCZ on corticosterone levels was tested. hM3Dq mice showed a significantly higher level of corticosterone (126.4 ± 12.0 ng/mL) compared to both hM3Di mice (55.0 ± 9.0 ng/mL, P < 0.0001) and control mice (40.8 ± 2.25 ng/mL, P < 0.0001). Control and hM4Di corticosterone levels were not significantly different (P = 0.498). Together, these data suggest that activation of CRH^PVN neurons in the home cage environment can by itself drive increases in behavioral arousal in mice that disrupts ongoing ultradian patterns of movement. However, inhibition of CRH^PVN neurons does not prevent on-going bouts of spontaneous home cage activity. One caveat of this experiment was that the hM4Di virus expression spread to the area surrounding the PVN, whereas the hM3Dq expression was restricted to the PVN.

Fig. 3.

Chemogenetic activation of CRH^PVN neurons increases home cage activity. (A) Activity profiles for three different hM4Di mice injected with DCZ (dotted line) at ZT17. Spontaneous bouts of movement are seen both before and after DCZ injection. (B) Activity profiles for three different hM3Dq mice injected with DCZ (dotted line) at ZT17. Movement becomes markedly elevated following DCZ injection. (C) Activity profiles for three different control mice injected with DCZ at ZT17. Spontaneous bouts of movement are observed both before and after DCZ injection. (D) Animal movement (plotted in 30 min bins) is significantly elevated following DCZ injection in hM3Dq mice (red) compared to hM4Di (blue) and control mice (black). All experiments were in male mice.

Coordinated CRH^PVN Activity and Behavior in Crh-IRES-Cre Rats.

The above data show a clear relationship between spontaneous CRH^PVN activity and spontaneous behavior in mice over the 24-h day. The period of CRH^PVN activity over the 24-h day appeared similar to the period of ultradian corticosterone pulses previously reported in rats (22, 31). To investigate this further, we generated a Crh-Ires-Cre rat so that we could perform CRH^PVN fiber photometry, behavioral analysis, and repetitive blood sampling.

Crh-Ires-Cre rats were generated using CRISPR-Cas9 technology. Cre recombinase was expressed under the control of the endogenous Crh gene by inserting an internal ribosomal entry site (IRES)-Cre cassette immediately after the translation stop site of the Crh gene. We observed Cre mRNA labeling in brain regions where Crh is known to be expressed, including in the PVN (SI Appendix, Fig. S2 and Table S1). Immunohistochemistry for cre-recombinase protein showed labeling in the PVN of Crh-Ires-Cre rats (SI Appendix, Fig. S3). To determine the exact colocalization of Crh and Cre mRNA, we performed dual-label RNAscope in the PVN. This showed that on average 81.9 ± 3.9% of Crh-labeled neurons coexpressed Cre and that 97.8 ± 0.5% of Cre labeled neurons coexpressed Crh (Fig. 4 A and B). To determine whether knock-in of Ires-Cre affected basal corticosterone levels, we measured morning and evening corticosterone with tail vein blood sampling. We observed a clear elevation of corticosterone levels in the PM samples compared to AM samples of both wildtype and Crh-Ires-Cre rats. There was a significant effect of time of day [F_(1,3) = 23.55, P = 0.0003], but no effect of genotype [F_(1,14) = 3.329, P = 0.0895] and no significant interaction [F_(1,13) = 0.7100, P = 0.4147; SI Appendix, Fig. S3]. Likewise, adrenal weight was not significantly different between wildtype and Crh-Ires-Cre rats (P = 0.4701). We next injected Cre-dependent GCaMP6s AAV into the PVN of Crh-Ires-Cre rats. To determine whether GCaMP6s were targeted to CRH neurons, we performed RNAscope for Crh with immunohistochemistry for GCaMP. 58.34 ± 0.075% of Crh positive neurons expressed GCaMP6s while 87.71 ± 0.036% of GCaMP6s expressing neurons coexpressed Crh mRNA (SI Appendix, Fig. S3). All experiments with fiber photometry and blood sampling were performed in female rats. Females were chosen due to the fact that they have larger amplitude corticosterone pulses that are easier to detect with repetitive blood sampling (35).

Fig. 4.

Spontaneous elevations in CRH^PVN activity correlate with bouts of behavioral arousal in female Crh-Ires-Cre rats. (A and B) Example images of RNAscope dual in situ hybridization in the PVN region for Cre (blue) and Crh (red) mRNA. (C) Example photometry recording of CRH^PVN activity showing spontaneous ultradian activity across the 24-h day in a freely behaving rat (Top) along with simultaneous animal movement tracking (Bottom). (D) Heatmap representation of CRH^PVN activity before and after the onset and offset of movement bouts from 4 GCaMP-expressing rats. (E) Heat map movement profiles corresponding to neuronal activity data shown in (D). (F) Average CRH^PVN activity (red) and animal movement (blue) 10 mins before and after the onset and offset of each movement bout. (G) Cross-correlation analysis demonstrating a peak correlation coefficient of 0.51 (red line) at 10 secs (CRH^PVN activity preceding animal movement). No clear peak was observed in the time-shifted data (black line). (H) State-space GC analysis showing relative differences between estimated ΔF/F₀ → movement (red line) and movement → ΔF/F₀ (black line) causal influences. Solid lines are the experimental data. Dotted lines are the shuffled and time-shifted control data. (I) CCM analysis showing relative differences between estimated ΔF/F₀ -> movement (red line) and movement → ΔF/F₀ (black line) causal influences. Solid lines are the experimental data. Dotted lines are the shuffled and time-shifted control data.

We next performed PVN fiber photometry on GCaMP6s expressing Crh-Ires-Cre rats over 28 h along with video recording of behavior. As with recordings in mice, we ignored the first 4 h after connecting the optic fiber. Similar to recordings from mice, we observed recurrent upstates of CRH^PVN neuron activity in rats that had a frequency of 1.0 ± 0.9 events per hour (n = 4, Fig. 4C). However, we observed that the magnitude of signals was smaller in rats compared to mice and upstates were less obvious. We next analyzed head movement in our 24-h video recording with DeepLabCut. Consistent with our observations in mice, bouts of head movements showed an ultradian rhythm that appeared coordinated with upstates of CRH^PVN activity. We aligned the onset or offset of a bout of movement with photometry and observed a similar relationship to that seen in mice (Fig. 4 D–F). Cross-correlation between ΔF/F₀ and head movement revealed a peak correlation coefficient of 0.51 (red line) at 10 secs (CRH^PVN activity preceding animal movement). No clear peak was observed in control time shifted data.

To further investigate the temporal relationship between GCaMP6s CRH^PVN signals and movement in rats, we performed state-space GC analysis and CCM. Similar to what we observed in mice, the orange line (ΔF/F₀ predicting movement) is consistently higher than the other direction (movement predicting ΔF/F₀; black), confirming the direction of coupling is driven by changes in CRH^PVN neuron activity (Fig. 4 H and I).

To measure pulsatile corticosterone secretion, we performed jugular vein cannulation in Crh-Ires-Cre rats. After recovery from surgery, animals were connected to a custom-built automated blood sampling system. This system could automatically take 25 to 40 μL blood samples once every 7.5 or 15 min. Following collection of each blood sample, the same volume of saline was automatically infused back into the rat. Using this system, we were able to collect blood samples and detect spontaneous ultradian corticosterone pulses with temporal kinetics similar to that previously reported (30, 31) (SI Appendix, Fig. S4). In a separate cohort, we performed jugular vein cannulation in animals that had previously undergone surgery for GCaMP6s fiber photometry. We then measured CRH^PVN activity, behavior, and blood corticosterone over a 6 to 8 h time window in freely behaving Crh-Ires-Cre rats. As can be seen from the representative recordings, bursts of CRH^PVN activity (Fig. 5 A and D) preceded a pulse of corticosterone (Fig. 5 B and E) in many instances. However, in other instances, bursts of CRH^PVN activity occurred without any subsequent corticosterone pulse. There were also cases where corticosterone pulses occurred without clear bursts of CRH^PVN activity. We next performed cross-correlation to assess the relationship between CRH^PVN activity and corticosterone. On average, the cross-correlation coefficient between CRH neuronal activity and corticosterone secretion peaked at −15 min time lag with a coefficient of 0.73 ± 0.05 (n = 4). This suggests that CRH neuronal activity and corticosterone secretion are correlated on average and that CRH^PVN activity precedes corticosterone by 15 min (Fig. 5G). CRH^PVN activity and animal movement were also highly correlated, with a peak at 0 min time lag (Fig. 5H).

Fig. 5.

Ultradian CRH^PVN activity correlates with pulsatile corticosterone secretion under unstressed conditions. (A) Example photometry recording of CRH^PVN activity with simultaneous blood sampling for corticosterone (CORT) (B) and recording of rat movement (C) in female Crh-Ires-Cre rats. Ultradian rhythms can be seen in all three measures. The black line shows the slow component of the photometry recording extracted using decomposition. The dashed line in b shows missing samples (sampling rate of 1 sample/15 min). Red stars indicate ultradian pulses identified using MATLAB. (D) Example photometry recording of CRH^PVN activity from a different rat along with simultaneous blood sampling for corticosterone (E) and recording of rat movement (F). Blood sampling in (E) was performed at a rate of 1 sample every 7.5 mins. (G) Cross-correlation analysis of CRH^PVN neuron activity and CORT secretion show a peak correlation coefficient of 0.73 at −15 min (CRH^PVN activity precedes corticosterone secretion). (H) Cross-correlation analysis of simultaneous CRH^PVN activity and animal movement show a peak correlation coefficient of 0.43 at 0 mins.

Discussion

Here, we show that CRH^PVN neurons exhibit an ultradian rhythm in excitability over the 24-h day. Increases in CRH^PVN activity were predictive of increases in animal movement and chemogenetic activation of these neurons increased locomotion/movement. However, chemogenetic inhibition of CRH^PVN neurons did not change movement patterns. In a new line of Crh-Ires-Cre rats, we found that pulses of corticosterone secretion were sometimes, but not always, preceded by elevations of CRH^PVN neural activity. However, correlation analysis revealed that on average, increases in CRH^PVN activity preceded corticosterone rises by approximately 15 min. Together these data reveal that CRH^PVN neurons display an ultradian rhythm of activity over the 24-h day that is coordinated with behavioral arousal, however, the relationship between CRH^PVN activity and pulses of corticosteroid secretion is not one-to-one.

A key finding from this study is the high correlation between CRH neuronal activity and movement under resting, unstressed states. Causality analyses indicated that CRH^PVN neuron activity predicted changes in physical activity. This suggests that CRH^PVN neurons might be an important node in the complex network that governs arousal state. This is in line with previous work showing that CRH^PVN neurons are involved in the regulation of arousal/wakefulness (26, 27). CRH^PVN neurons also synthesize glutamate and other studies that have manipulated PVN vglut neurons have observed similar results (36, 37). In addition, past work has also shown that CRH^PVN neurons regulate stress-related behaviors and affective states (14, 15, 23). CRH^PVN neuron projections to the lateral hypothalamus appear to be particularly important for many of these actions (23, 26, 27). However, CRH^PVN neurons have projections to other brain regions including the amygdala and periaqueductal gray (25) and these projections are also likely to be important.

The most extensive CRH^PVN neuron projection is to the external zone of the median eminence where these neurons release CRH peptide to control the HPA axis (38). Despite these neurons being widely accepted as master controllers of the HPA axis, there is still debate regarding their role in controlling the pulsatile, ultradian pattern of corticosteroid secretion. This debate has stemmed from past experiments which have shown that ultradian corticosteroid pulses can persist following disconnection of the hypothalamus from the pituitary in sheep (39) or that corticosterone pulses can be generated with constant CRH infusions in rats (22). This has led to the proposal of a subhypothalamic pulse generator mechanism for pulse generation (22, 40). Prior to this study, it was unknown whether the CRH^PVN neuron population exhibited ultradian rhythms of activity in vivo that are coordinated with corticosteroid release. Here, we demonstrate that CRH^PVN neurons do exhibit ultradian rhythms in activity over the 24-h day in both rats and mice. In rats, we demonstrate that in many cases, these bursts of activity precede pulses of corticosterone release. Cross-correlation analysis also demonstrated a correlation between CRH^PVN activity and corticosterone with the peak in CRH^PVN neuron activity preceding corticosterone by 15 min. These data strongly suggest that bursts of CRH^PVN neuron activity contribute to ultradian pulses of corticosterone release. However, as noted above, the relationship is not one-to-one. This could be due to several reasons. First, we recorded CRH^PVN neuron activity from only one side of the PVN. It is possible that some pulses of corticosterone release were driven by bursts of activity from the unmonitored contralateral PVN. Second, it is likely that ultradian patterns of CRH^PVN neuron activity cooperate with a subhypothalamic pulse generator mechanism which leads to the final pattern of corticosteroid secretion. Future work using both mathematical models (40) and artificial manipulations of CRH^PVN neuron activity will be needed to address this.

Unlike circadian rhythms, ultradian rhythms do not appear to be controlled by a central master clock. Instead, the circuit mechanisms appear to be unique to each system (41, 42). Past work has shown that ultradian pulses of corticosterone persist when animals are exposed to constant light as well as following ablation of the suprachiasmatic nucleus (43). While the circuit mechanisms controlling ultradian activity of CRH^PVN neurons and corticosterone are not clear (8), the subparaventricular zone (SPZ) has been implicated as a possible region important for generating these rhythms (44). Specifically, Wu et al. observed ultradian rhythms of activity in PVN and SPZ neurons using Ca²⁺ imaging in cultured brain slices (44). These rhythms were blocked with tetrodotoxin and glutamate receptor antagonists suggesting that they require local network activity. However, the detailed circuit mechanisms driving these rhythms remain unresolved.

In summary, we have demonstrated that the CRH^PVN neuron population displays an ultradian rhythm of activity over the day-night cycle. Spontaneous elevations of CRH^PVN neuron activity are coordinated with increases in behavioral arousal. We also show a temporal correlation between elevations of CRH^PVN activity and pulses of corticosterone secretion, with CRH^PVN activity preceding corticosterone. However, CRH^PVN neuron activity and pulses of corticosterone are not one-to-one. Together, these data reveal the patterns of CRH^PVN neuron activity over the 24-h day and their complex relationship with behavior and stress hormone secretion.

Materials and Methods

Animals.

All animals were housed under a 12 h light/dark cycle in individually ventilated cages with ad libitum access to food and water. All Experiments were conducted in accordance with the New Zealand Animal Welfare Act and approved by the University of Otago Animal Welfare and Ethics Committee.

Stereotaxic Surgery for Mice.

Adult (10 to 12 wk old) male Crh-Ires-Cre (45) or Crh-Ires-Cre;Ai14 (tdTomato reporter) mice (46) were anesthetized with 2% isofluorane and placed in a stereotaxic frame. Adeno-associated virus (AAV) encoding GCaMP6s (AAV1.CAG.Flex.GCaMP6s.WPRE.SV40) or GFP (AAV9.Syn.DIO.EGFP.WPRE.hGH) was stereotaxically injected unilaterally into the PVN via a Hamilton syringe (−0.8 mm AP, −0.25 mm ML, −4.5 mm DV) at a volume of 1μL over 10 min (Fig. 1 A–C). A fiberoptic cannula (400 μm core, 0.48 N.A.; Doric Lenses) was then implanted at the same coordinates and secured using adhesive dental cement. For DREADD experiments, mice received bilateral injections into the PVN and did not receive fiber optic implants. For these experiments, mice were injected with either AAV9-hSyn-DIO-hM3Dq-mCherry (hM3Dq), AAV9-hSyn-hM4Di-mCherry (hM4Di) or AAV9-FLEX-tdTomato (control). All mice were given carprofen (5 mg/kg) and lidocaine (2%) during surgery and allowed to recover for at least 3 wk before experimental recordings.

Fiber Photometry for Mice.

Photometry recordings of GCaMP6s fluorescence were acquired using Synapse software and a RZ5P processor from Tucker-Davis Technologies (TDT, Alachua Florida) and optical components purchased from Doric Lenses (16, 42). Excitation LEDs (465 nm blue and 405 nm violet) were sinusoidally modulated at 211 and 531 Hz, respectively. Excitation wavelengths were relayed through a filtered fluorescence minicube (spectral bandwidth: 460 to 490 nm and 405 nm) to a 400 μm 0.48 NA fiberoptic cable connected to the mouse. Light power for the 465 nm wavelength at the fiber tip was 35 to 45 μW. A single emission (filtered at 500 to 550 nm) was detected using a femtowatt photoreceiver (2151, Newport) with a lensed fiber cable adapter.

On the day of recordings, mice were connected to the photometry system and then left undisturbed for 28 h with free access to food and water in their home cage while CRH^PVN GCaMP6s activity was recorded along with behavior with an overhead camera. Connecting mice to the photometry system involved handling, which is a stressor. For this reason, we excluded the first 4 h of recordings following connection and only analyzed the last 24 h of photometry data. For these recordings, photometry data were acquired in a scheduled recording mode, where both LEDs (405 nm and 465 nm) were switched on for 3 s and off for 7 s. This scheduled recording mode was chosen to reduce photobleaching GCaMP6s over extended recording periods.

All experiments were conducted in the animal’s home cage, which was placed in a custom-made apparatus (40 cm length, 40 cm width, 40 cm height) with white walls and transparent lid. Mice were habituated to the testing room and apparatus for 7 consecutive days prior to experimental manipulations. On the day of recording, mice had their optic fiber implant attached to an optic cable attached to a swivel. They were then left undisturbed for 28 h with free access to food and water in their home cage.

Manipulation of CRH^PVN Neuron Activity with DREADDs.

At least 3 wk after surgery, mice underwent habituation followed by behavioral recording. All experiments were conducted in the animal’s home cage, which was placed in a custom-made apparatus (40 cm length, 40 cm width, 40 cm height) with white walls and transparent lid. Mice were habituated to the box for 4 d before experiment. On the day of the experiment, animal behavior was recorded using a Logi digital camera suspended above the testing box. Animals were left undisturbed until 5 h into the dark phase of the light cycle when they received a single IP injection of DCZ (0.3 mg/ml in DMSO, 1 mg/kg) and placed back into their home cage.

At least one week following behavioral testing, mice received an IP injection of DCZ (0.3 mg/ml) and were returned to their home cage for one hour. They were then removed from their home cage and placed in a novel environment (40 cm × 40 cm white box as previously used but without their home cage, bedding, food, etc.) for 5 min. They were then returned to their home cage for 15 min before a tail tip blood sample was taken. Corticosterone levels were subsequently determined by ELISA (see below).

Generation of Crh-Ires-Cre Rats.

A Sprague-Dawley Crh-Ires-Cre knock-in rat line (SD-Crh^tm1(Cre)Kji)(RRID:RGD_401976372) was generated using CRISPR technology by Cyagen Biosciences Inc. Cre recombinase was expressed under the control of the endogenous Crh gene by inserting an internal ribosomal entry site (IRES)-Cre cassette immediately after the translational stop site of the Crh open reading frame.

Immunohistochemistry and RNAscope.

Rats were terminally anesthetized with sodium pentobarbital. Upon loss of the pedal withdrawal reflex, rats were transcardially perfused with 0.9% heparinized saline solution followed by 4% paraformaldehyde (PFA). Brains were postfixed in 4% PFA for 24 h before being saturated sequentially with 10% sucrose, 20% sucrose, and 30% sucrose. After which, the brains were sectioned coronally on a cryostat at 30 μm thickness for Cre recombinase immunohistochemistry and 12 μm thickness for Cre and Crh mRNA RNAscope. Slices for immunohistochemistry were stored in tris-buffered saline (TBS) at 4 °C before being transferred to cryoprotectant whereas slices for RNAscope were collected on Superfrost plus slides and were stored at −80 °C.

Immunohistochemistry was used to label Cre recombinase in the PVN region in Crh-IRES-Cre rats. Brain slices were incubated in ethylenediaminetetraacetic acid (EDTA) at 90 ºC for 15 mins and then thoroughly rinsed in TBS before being quenched using hydrogen peroxide (3% hydrogen peroxide, 40% methanol, 57% TBS). Next, slides were placed in blocking buffer (0.4% Triton-X + 0.25% bovine serum albumin (BSA) + 2% normal goat serum (NGS) in TBS) for 90 min and proceeded with primary antibody incubation in rabbit anti-Cre recombinase [G. Schütz, Heidelberg (47)] diluted 1:5,000 in blocking buffer for 1 h at room temperature and then 48 h at 4 °C. Slices were then washed extensively in TBS before being incubated with biotinylated goat anti-rabbit IgG (H + L) secondary antibody (ThermoFisher Cat# B-2770, RRID: AB_2536431,1:500 in TBS + 0.4% Triton-X + 0.25% BSA) followed by incubation in Vectastain Elite A/B solution (Vector laboratories, PK6100). Cre labeling was visualized by a chromogenic reaction using nickel sulfate- 3,3'-diaminobenzidine chromogen solution. Slices were dried and mounted in dibutylphthalate polystyrene xylene (DPX) mounting medium and imaged with a bright field microscope.

RNAscope 2.5 HD Duplex assay (Advanced cell diagnostics Inc., USA) was performed according to the manufacturer’s protocol to colabel Crh and Cre mRNA in the brain (Crh probe Cat# 318931; Cre probe Cat# 423321). Brain slices were baked at 60 °C for 30 mins and then incubated in prechilled 4% PFA for 15 mins before being dehydrated sequentially in 50, 70, and 100% ethanol. Slides were incubated in RNAscope hydrogen peroxide for 10 mins at room temperature and then submerged in target retrieval solution at 98 to 102 °C for 5 mins. Slides were then dehydrated in 100% ethanol and air-dried. Immedge™ hydrophobic barrier pen was used to draw a hydrophobic barrier around each slice. One drop of Protease Plus was added to each slice and then incubated for 30 mins at 40 °C. Slides were washed in distilled water before being incubated with the appropriate probe for 2 h at 40 °C. After incubation, slides were kept in 5X saline-sodium citrate buffer overnight at room temperature.

On the following day, slides went through 6 hybridization steps using hybridize Amp 1 to 6 solutions. After the sixth hybridization step, Red solution mix containing 1 part of Red-B in 60 parts of Red-A was applied to each slice to detect the red signal (Cre mRNA). Four more hybridization steps were performed before adding Green solution mix containing 1 part of Green-B in 60 parts of Green-A to each slice to detect green signal (Crh mRNA). Slices were thoroughly washed in wash buffer and then washed in distilled water before being submerged in a 10% hematoxylin staining solution for 2 s. Slides then went through multiple wash steps in tap water, 0.02% ammonia water, and then tap water again. Slides were dried at 60 ºC and then mounted using VectaMount mounting medium. The mounting medium was allowed to dry before slides were examined under a standard bright field microscope at 20 to 40× magnification. In the PVN, cells with only Crh mRNA expression, cells with only Cre mRNA expression and cells with both Crh and Cre mRNA expression were counted to calculate the coexpression levels. Cells were classed as expressing mRNA if they had three or more stained puncta.

To label for Crh mRNA and GCaMP6s protein, Crh-Ires-Cre rats that were previously injected with GCaMP6s-AAV were anesthetized and perfusion fixed. Brain sections of the PVN were first processed with RNAscope Multiplex Fluorescent Detection kit v2 (Cat #323110) according to the manufacturer’s protocol. Day-1 slide processing was same as described above for RNAscope 2.5 HD Duplex assay. On day-2, three hybridization steps were performed at 40 °C in the hybridization oven using FL v2 A 1 to 3 solutions followed by three HRP steps including the addition of FL v2 HRP-C1 (Crh probe in C1 channel), fluorophore Cyanine 3 (Akoya BioSciences, # SKU NEL7444001KT), and FL v2 HRP blocker. Immunolabeling with GCaMP6s was continued immediately by pretreating the sections with 4% cold PFA for 5 min and placing the slides in blocking buffer for 60 min followed by primary antibody incubation (chicken anti-GFP (1:2,000) RRID: AB_10000240, diluted in blocking buffer) for 24 h at 4 °C. The brain slices were thoroughly washed in TBS and then transferred into secondary antibody which contained goat-anti-chicken conjugated with Alexa Fluro-488 (RRID: AB_2543096) diluted 1:500 in TBS+ 0.4% Triton-X + 0.25% BSA. The brain slices were then thoroughly washed in TBS before being cover slipped with Prolong Antifade Gold (ThermoFisher, P36930).

Stereotaxic Surgery for Rats.

Female heterozygous Crh-Ires-Cre rats (2 to 3 mo old) were anesthetized with isoflurane (5% in an induction chamber, and then 1 to 3% during the surgery) and placed in a stereotaxic frame. A 1 μl Hamilton syringe at −1.5 mm AP, +1.5 mm ML, −7.5 mm DV with a 10° was inserted into the brain to inject 0.7 μl of AAV1.CAG.Flex.GCaMP6s.WPRE.SV40 (PennVector Core) at a 100 nl/min rate into the PVN. A fiber implant (400/430-0.48_8.5mm_MF1.25_FLT, Doric Lenses, Quebec, Canada) was lowered into the brain at a 10-degree angle (−1.6 mm AP, +1.7 mm ML, −6.8 mm DV) and secured using dental cement.

Fiber Photometry in Rats.

Optical recordings of GCaMP6s fluorescence were acquired using a custom software acquisition system with optical components purchased from Doric Lenses (16, 42). Excitation light-emitting diodes (LED) (465 nm and 405 nm) were sinusoidally modulated at 211 Hz and 531 Hz, respectively. Excitation wavelengths passed through a filtered fluorescence mini cube (spectral bandwidth: 460 to 490 nm and 405 nm) to the rat via a 400 μm, 0.48 N.A. optic patch cord. The light power of the 465 nm LED at the fiber tip was 35 μW, and the 405 nm LED was set to 15 μW light output. Emitted fluorescence was filtered (500 to 550 nm) and was collected onto a photoreceiver (2151, Newport).

Fiber photometry recordings were conducted in a custom round home cage (35 cm in diameter, 40 cm in height). All recordings started 4 h before ZT12, and the rats were then left undisturbed for 28 h. During the 28-h recording, both 405 nm and 465 nm LED were set to turn on for 3 s and then to turn off for 7 s to minimize bleaching.

Photometry Analysis.

The isobestic 405 nm signal was scaled and fitted to the 465 nm signal using a second-order polynomial fit. The ΔF/F₀ was calculated by subtracting the fitted 405 nm signal from the smoothed 465 nm signal and then dividing it the line of polynomial fit. For determining the onset, offset, and hence the interupstate interval, upstate duration, and upstate counts, we used a frequency-decomposed signal to automate the detection of upstates. Specifically, a multiresolution wavelet (sym4) decomposition was carried out at 6 levels, where the sum of the two “fastest” components (i.e. higher frequency) was used to detect sharp changes in the data – the putative onset/offsets of the upstates. As this sum of faster oscillations is insensitive to the sustained change in activity, we used the slowest component of the decomposition to provide the threshold needed to define a sustained change in activity, so a peak-finding algorithm can identify the first peak of the fast summed activity preceding the low-to-high threshold to be the “onset” of the upstate, and the first fast peak after the high-to-low (following a low-to-high threshold crossing of the slow component) to be the “offset” of the upstate. With these definitions of upstate onset and offset, we then derive the measures of interupstate interval, upstate duration, and upstate counts.

Jugular Surgery and Automated Blood Sampling in Rats.

Female heterozygous Crh-Ires-Cre rats aged 3 to 4 mo were used. Rats were anesthetized with isoflurane (5% in an induction chamber, and then 1 to 3% during the surgery) and positioned in the dorsal position. A thermoplastic polyurethane elastomer catheter was inserted into the jugular vein through a skin incision. The jugular vein catheter was then tunneled subcutaneously and exteriorized through an incision between the scapulae. The jugular vein catheter was filled with sterile heparinized saline (50 IU/mL) which was flushed twice a day.

On the day of the blood sampling, the jugular catheter tubing was connected to a one-way swivel (Instech) mounted on a stand via a spring. This was connected to a custom-built automated blood sampling system. This consisted of a peristaltic pump (Minipuls 3 peristaltic pump, Gilson), three separately controlled one-way solenoid valves, a custom-built XY stage (with 96-well plate) and a raspberry pi that controlled the sample collection. A new well on the 96-well plate was used for each blood sample collected. Once set up, the system collected blood samples automatically without the need of experimenter intervention.

Two different blood sampling protocols were used. For the first protocol, 1 sample was collected every 15 mins for 6 h (24 samples). Each sample contained approximately 20 μL of saline and 40 μL of blood. The second protocol collected 64 samples across 8 h (1 sample every 7.5 mins). Each sample contained approximately 20 μL of saline and 25 μL of blood. As the blood samples collected showed slight variation in dilution, different amounts of 0.9% saline (50 IU heparin per mL) were added to each sample to reach a common 1:10 dilution (1 part blood, 9 parts saline). The amount of saline required was determined by measuring the hematocrit of each sample.

ELISA.

Corticosterone concentrations were measured using the DetectX Corticosterone Enzyme Immunoassay kit (Arbor Assays, Ann Arbor, MA, USA Cat# K014, RRID AB_2877626) according to the manufacturer’s instructions. The assay sensitivity was 26.54 ng/mL and the intra-assay coefficient of variation for this set of experiments was 4.81%.

Animal Movement Tracking.

DeepLabCut was used to track animal movement (34, 48). The right and left ears of the mouse, the optic fiber cannula and the base of the tail were manually labeled in 25 frames chosen through k-means clustering. A ResNet-101 based neural network with default parameters was created and then trained until the loss plateaus for each different video resolution. The network was then used to analyze videos with specific resolutions, and the XY position of each body part was exported as a CSV file. In Matlab, we first normalized all coordinate data to the dimension of the highest input video resolution (1,024 × 576 pixels) and took the median value between the ears and the cannula as a single tracking point. Large tracking jitters (>20 px displacement between time points) were removed and interpolated XY coordinates were submitted to an 11 point median filter for smoothing.

Rat movements were also tracked using DeepLabCut. A ResNet-50 based neural network with default parameters was trained for 250,000 iterations on 19 manually labeled frames (left ear, right ear, and fiber optic cannula were labeled) for each rat individually. The network was then used to analyze videos from that specific rat. The centroid of the labeled body parts was calculated. The displacement of the centroid was reported as animal movement in millimeters.

Movement-Triggered Analysis.

Since the tracking and photometry data are not sampled at the same rate, we downsampled tracking data by taking the median distance and speed from 10 s windows. Using this downsampled tracking, multiresolution decomposition was done with a symlets 4 wavelet (sym4) at 6 levels. We summed the two fastest components to extract the sharp changes in the tracking data. To map these sharp transient points to bouts of activity, we also used the residual (slowest component) of the decomposition and used these values as a threshold to detect coincidental sharp changes in movement (fast components) that is sustained (slow components). The timestamps of these transitions were collected and used for movement-triggered analyses and the determination of behavioral upstates.

Causality Analyses.

State-space causality: Downsampling by taking the median of windowed data results in the loss of information and likely exacerbates the moving average feature of our tracking and photometry data. Although no clear indications of nonstationarity can be observed from either the photometry and tracking data, sharp, bistable state transitions render tests for heteroscedasticity inconclusive. To account for these potential biases, we performed GC analysis using a state-space model. The method for state-space GC was adopted from Barnett and Seth, 2015 (49). Consistent with various generalizations of GC, linear state-space model uses optimal model order from the Bayesian information criterion to estimate Kalman states from a singular decomposition weighted by future/past canonical correlation weights of the inputs (i.e., movement and photometry data). With defined Kalman states, the observation matrix, state transition matrix, innovations covariance matrix, and Kalman gain matrix are derived to generate a state-space model in the innovations form. These parameters are used to generate error covariance matrices to parse out causal influences between frequency spectra calculated from autoregressive modeling of the state-space model for tracking and photometry data. Causal influences from movement to photometry data, and vice versa, are determined by the relative mutual information after accounting for covariances from the other.

Convergent cross-mapping (CCM): CCM is a time-embedding technique to uncover linear and nonlinear correlations between two inputs. Similar to a state-space approach, the lagged version of the input is used to construct a manifold, of which should show high correspondence if both inputs belong to the same system. Correlation between tracking data mapped onto a nearest-neighbor weighted estimate of it from the shadow manifold of photometry data is taken as a measure of the relative causal influence tracking data has on photometry data. With our data, we used a lag term derived from the optimal model order used in the state-space GC and cumulative block length of 60 min to test the convergence of correlations between across time.

Statistics.

Statistical analyses were performed with Prism and MATLAB. Paired t-tests were used to analyze circadian difference in ultradian CRH neuronal activity. Two-way ANOVA was used to analyze the effect of DCZ treatment. All data are presented as mean ± SEM. *P < 0.05, **P < 0.01, and ***P < 0.001.

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

All study data are included in the article and/or SI Appendix.