In vivo recording of suprachiasmatic nucleus dynamics reveals a dominant role of arginine vasopressin neurons in circadian pacesetting

We previously showed that lengthening the cellular circadian period of AVP neurons by the specific deletion of CK1δ lengthened the free-running period of circadian behavior, indicating that AVP neurons are involved in setting the ensemble period of the SCN network [23]. The remaining question was how much AVP neurons contribute to the period-setting. To elucidate this, we aimed to compare the period lengthening caused by AVP neuron-specific CK1δ deletion to that by pan-SCN CK1δ deletion. Although CaMKIIα is not abundant in the SCN [29], a particular CaMKIIα-Cre line [30] drives Cre expression in the forebrain, including the entire SCN [31]. Indeed, crossing this line to floxed Bmal1 mice resulted in >90% deletion of BMAL1 in the SCN and a complete loss of circadian behavior rhythm [31]. We also reproduced the arrhythmicity of CaMKIIα-Cre; Bmal1^flox/flox mice, confirming CaMKIIα-Cre mice as a pan-SCN Cre driver (S1A and S1B Fig).

By crossing this CaMKIIα-Cre line with CK1δ^flox/flox [8], we deleted CK1δ in the forebrain and the entire SCN neurons (CaMKIIα-CK1δ^−/−). CaMKIIα-CK1δ^−/− mice showed normal diurnal locomotor activity (home-cage activity) rhythm in the 12 h of light and 12 h of darkness (LD) condition (Fig 1A and 1B). In constant darkness (DD), CaMKIIα-CK1δ^−/− mice showed a longer free-running period (24.83 ± 0.08 h) than that of control mice (23.93 ± 0.03 h, P < 0.001, Fig 1C). By contrast, their amplitude of locomotor activity rhythm (Qp) was normal (P = 0.52, Fig 1D). The difference in the free-running period between control and CaMKIIα-CK1δ^−/− mice was almost similar to the difference between control and Avp-CK1δ^−/− mice (23.94 ± 0.03 h versus 24.72 ± 0.03 h) [23], as well as with that between control and Vgat-Cre; CK1δ^flox/flox mice (approximately 40-min elongation), another line with pan-SCN neuronal CK1δ deficiency [32]. These data suggest that AVP neurons are the principal determinant of the free-running period of the circadian behavior rhythm.

To examine the contribution of VIP neurons in the setting of the SCN ensemble period, we deleted CK1δ specifically in VIP neurons by crossing Vip-ires-Cre mice [33] with CK1δ^flox/flox (Vip-CK1δ^−/−). However, we did not observe any change in locomotor activity rhythm of Vip-CK1δ^−/− mice (Fig 2A–2D, period, control 23.93 ± 0.04 versus Vip-CK1δ^−/− 23.91 ± 0.03). This result was consistent with a previous report that the lengthening of the intrinsic TTFL period in VIP neurons by the overexpression of Clock^Δ19 did not alter the behavioral free-running period [21]. Thus, although VIP peptide act as an essential synchronizer of SCN neurons, the TTFL in VIP neurons itself may have little contribution to the ensemble period-setting of the SCN network.

We previously reported that disruption of cellular clocks in AVP neurons by deleting Bmal1 (Avp-Bmal1^−/− mice) causes severe attenuation of circadian behavior rhythm measured by home-cage activity [22]. However, most Avp-Bmal1^−/− mice still retained some circadian rhythmicity with unstable, lengthened free-running period and activity time. In contrast, Lee and colleagues reported that mice with Bmal1 deletion in NMS neurons (Nms-Bmal1^−/− mice) lost the circadian rhythm in wheel-running behavior [21]. Notably, a recent study showed that rhythms of home-cage activity and body temperature in Avp-Bmal1^−/− mice were very similar to those in Nms-Bmal1^−/− mice [34]. Thus, circadian phenotypes may be significantly different between wheel-running and home-cage activity rhythms.

Given the critical role of SCN AVP neurons in setting the free-running period, we reevaluated the circadian behavior rhythm of Avp-Bmal1^−/−and Avp-CK1δ^−/− mice by measuring wheel-running activity. The difference in wheel-running rhythm between Avp-CK1δ^−/−and control mice was consistent with that of home-cage activity rhythm [23], namely, approximately 50-min lengthening of the free-running period (control 23.73 ± 0.05 versus Avp-CK1δ^−/− 24.52 ± 0.07, P < 0.001, S2A and S2B Fig). Avp-Bmal1^−/−mice also demonstrated an impaired wheel-running rhythm similar to the previously described home-cage activity rhythm, i.e., “split” behavior with a lengthened free-running period in DD. However, in contrast to the home-cage activity, 7 out of 8 Avp-Bmal1^−/− mice gradually lost the circadian wheel-running rhythm after approximately 10 to 20 days in DD (S2C Fig). Such a gradual disappearance of the wheel-running rhythm was similar to that of Nms-Bmal1^−/− mice [21]. Therefore, the impairment of circadian behavior rhythm may be comparable between Avp-Bmal1^−/− and Nms-Bmal1^−/− mice.

To evaluate the status of cellular circadian clocks in the SCN, we next performed real-time bioluminescent cell imaging of coronal SCN slices prepared from control, Avp-CK1δ^−/−, CaMKIIα-CK1δ^−/−, and Vip-CK1δ^−/− adult mice crossed with a luciferase reporter (Per2::Luc) [35] housed in LD. In the previous study, we reported that PER2::LUC oscillations in the SCN slices of Avp-CK1δ^−/− mice did not demonstrate coherent circadian rhythm with a lengthened period, contrary to their locomotor activity rhythm [23].

Here, we monitored and compared the day-by-day change of individual pixels’ periods of PER2::LUC oscillations between shell and core, as well as among genotypes. To do so, we calculated the daily peak phase and period (the interval of 2 adjacent peaks) for PER2::LUC oscillations in individual pixels that covered the SCN (Figs 3 and S3). In Avp-CK1δ^−/−; Per2::Luc mice, the peak phases in the shell pixels were later than those in core pixels by approximately 5 h at the first peak (Fig 3A and 3E–3G). Also, the periods were longer in the shell (25.77 ± 0.21 h) than in the core (23.93 ± 0.13 h) by approximately 2 h for the first cycle (Fig 3A and 3B). While core pixels oscillated with relatively stable periods, those of shell pixels shortened rapidly in the subsequent cycles, as if core cells influenced the periods of shell cells (Figs 3B and S3). The results of Avp-CK1δ^−/−; Per2::Luc mice were consistent with our previous report [23].

In CaMKIIα-CK1δ^−/−; Per2::Luc mice, all SCN neurons appear to have similarly lengthened periods of PER2::LUC rhythms. Therefore, we expected that SCN explants of those mice would show coherent PER2::LUC oscillations similar to those of control mice, except for the lengthening of periods. Indeed, core pixels oscillated stably with lengthened periods (approximately 25.5 h) for 3 cycles (Fig 3B, right). However, we observed a rapid temporal change of periods in the shell pixels similar to that in Avp-CK1δ^−/−; Per2::Luc mice, except for approximately 1-h lengthening (Fig 3B, left). Namely, the peak phases in the shell pixels were later than those in core pixels by approximately 4 h at the first peak (Fig 3A and 3E–3G). In addition, the periods were longer in the shell (27.09 ± 0.32 h) than in the core (25.87 ± 0.11 h) by approximately 1 h for the first cycle. However, periods of shell pixels shortened rapidly in the subsequent cycles (Fig 3B). Thus, the coherent circadian rhythm was not maintained in PER2::LUC expression of the SCN slice cultures in CaMKIIα-CK1δ^−/− mice, which is a common phenomenon with Avp-CK1δ^−/− mice. Concordantly, the variability of the individual pixels’ periods increased in the SCN shell of Avp-CK1δ^−/− and CaMKIIα-CK1δ^−/− mice (Fig 3C). Furthermore, the peak amplitude of individual pixels’ PER2::LUC oscillations decayed more rapidly in the shell of these 2 strains of mice (Fig 3D). These observations suggested the attenuated synchronization of cellular PER2::LUC oscillations in the SCN shell of Avp-CK1δ^−/− and CaMKIIα-CK1δ^−/− mice ex vivo. In contrast, PER2::LUC oscillations in the SCN slices of Vip-CK1δ^−/−; Per2::Luc mice demonstrated no significant difference compared to control mice. This observation was consistent with their normal free-running period of circadian locomotor activity rhythm.

Lengthening the cellular circadian periods of AVP neurons could stably lengthen the free-running period of the behavioral rhythm. However, it could not lengthen the cellular PER2::LUC rhythms in the prolonged SCN cultures. Therefore, we postulated that the action of AVP neurons might be susceptible to slicing. To test this possibility, we next measured the cellular period of SCN AVP neurons in vivo by recording the intracellular Ca²⁺ ([Ca²⁺]_i) rhythm using fiber photometry [36] in Avp-CK1δ^−/− and control (Avp-Cre; CK1δ^wt/flox) mice (Fig 4). To do so, we expressed a fluorescent Ca²⁺ indicator jGCaMP7s [37] specifically in SCN AVP neurons by focally injecting a Cre-dependent AAV vector (AAV-CAG-DIO-jGCaMP7s) and then implanted an optical fiber just above the SCN (Fig 4A and 4B). The detected jGCaMP7s signal was averaged, detrended, and smoothened to extract daily [Ca²⁺]_i rhythms (see Materials and methods). We defined the time points crossing zero values as the onset and offset of the GCaMP signal and their midpoint as the peak phase. In both control and Avp-CK1δ^−/− mice, daily [Ca²⁺]_i rhythms were observed in SCN AVP neurons in both LD and DD. Their peaks were around the offset of locomotor activity (home-cage activity) (Figs 4C, 4D, 5A and S4A), as described previously for control mice [36]. The relationship between the GCaMP peak and locomotor offset remained unchanged in Avp-CK1δ^−/− mice. Thus, not only the period of locomotor activity rhythm but also the period of [Ca²⁺]_i rhythm in AVP neurons was lengthened similarly in DD (GCaMP, control 23.77 ± 0.18 versus Avp-CK1δ^−/− 24.42 ± 0.13, P < 0.05, Fig 5D; behavior, control 23.77 ± 0.10 versus Avp-CK1δ^−/− 24.43 ± 0.09, P < 0.001, Fig 5F, black). Such a lengthening of [Ca²⁺]_i rhythm period was consistently observed from the onset of DD (Fig 5H). This result indicated that the CK1δ deletion in AVP neurons caused a stable lengthening of their cellular circadian period in vivo.

We next asked whether the SCN shell and core circadian oscillations in vivo in Avp-CK1δ^−/− mice were dissociated as in the slices or maintained synchrony. Because of the importance of VIP neurons in the SCN core for maintaining the coherence of circadian oscillators in the SCN [14,15,18,38,39], we examined whether the [Ca²⁺]_i rhythm of SCN VIP neurons was altered in Avp-CK1δ^−/− mice in vivo. To specifically target the jGCaMP7s expression in VIP neurons of Avp-CK1δ^−/− mice, we first crossed Avp-CK1δ^−/− (or control) with Vip-tTA knock-in mice [40]. In Vip-tTA mice, VIP neurons specifically express tetracycline transactivator (tTA). We then injected a tTA-dependent AAV vector (AAV-TRE-jGCaMP7s) into the SCN (Fig 6A and 6B). The [Ca²⁺]_i rhythms of VIP neurons were synchronized antiphasically with locomotor activity rhythm in both control (Avp-Cre;CK1δ^wt/flox; Vip-tTA) and Avp-CK1δ^−/−; Vip-tTA mice (Figs 6C, 6D and S4B). Furthermore, the timing of locomotor onset and GCaMP offset were almost identical, while the timing of locomotor offset and VIP GCaMP onset correlated in both LD and DD (Figs 6C, 6D and S4B). In Avp-CK1δ^−/− mice, the period of VIP-neuronal [Ca²⁺]_i rhythm was lengthened in DD to the same extent as AVP-neuronal [Ca²⁺]_i and behavior rhythms (GCaMP, control 23.87 ± 0.03 versus Avp-CK1δ^−/− 24.47 ± 0.05, P < 0.001, Fig 5E). Such a lengthening of [Ca²⁺]_i rhythm period was consistently observed from the onset of DD (Fig 5I). Importantly, as a control, we verified that EGFP-expressing VIP neurons via injecting AAV-TRE-EGFP showed very little circadian oscillation of fluorescence (S4C Fig). These results suggested that the cellular circadian rhythms of AVP and VIP neurons maintained synchrony in vivo in Avp-CK1δ^−/− mice.

How about the phase relationships between AVP-neuronal, VIP-neuronal, and behavioral circadian rhythms? Overall, the peak phases of GCaMP signals in LD were similar between control and Avp-CK1δ^−/− mice for both AVP and VIP neurons (AVP: control, ZT 2.74 ± 0.57; Avp-CK1δ^−/−, ZT 3.82 ± 0.74, Fig 5A, top; VIP: control, ZT 4.98 ± 0.54; Avp-CK1δ^−/−, ZT 5.99 ± 0.17, Fig 5A, bottom). AVP neurons peaked earlier than VIP neurons (control, P < 0.05; Avp-CK1δ^−/−, P = 0.07) in LD. These peak phase relationships were essentially maintained even in DD. Due to the lengthening of the free-running period, the peak phase of AVP-neuronal [Ca²⁺]_i in projected ZT was significantly later in Avp-CK1δ^−/− during days 8 to 10 in DD (control, projected ZT 0.72 ± 0.69; Avp-CK1δ^−/−, projected ZT 6.15 ± 0.65, P < 0.001, Fig 5B, top) but not different in CT (circadian time) defined by the onset of locomotor activity as CT12 (control, CT 3.16 ± 0.66; Avp-CK1δ^−/−, CT 4.43 ± 0.76, Fig 5C, top). The situation of VIP-neuronal [Ca²⁺]_i in DD was similar (control, projected ZT 4.48 ± 0.22; Avp-CK1δ^−/−, projected ZT 9.47 ± 0.36, P < 0.001, Fig 5B, bottom; control, CT 5.64 ± 0.17; Avp-CK1δ^−/−, CT 6.23 ± 0.44, Fig 5C, bottom). Thus, the phase relationship between [Ca²⁺]_i rhythms of AVP and VIP neurons remained unaltered in Avp-CK1δ^−/− mice. Note that the TTFL period was artificially lengthened only in AVP but not in VIP neurons. These results suggested that AVP neurons can regulate the period of cellular circadian rhythm of VIP neurons in vivo.

To test whether the AVP peptide is essential for SCN AVP neurons to convey their cellular period to other SCN neurons, we selectively disrupted the Avp gene in these neurons using in vivo genome editing. First, we further crossed Avp-CK1δ^−/− mice with Rosa26-LSL-Cas9-2A-EGFP mice, which express SpCas9 Cre-dependently [41]. We then injected an AAV vector expressing gRNA targeting the Avp (AAV-U6-gAvp-EF1α-DIO-mCherry) or a control AAV (AAV-U6-gControl-EF1α-DIO-mCherry) into the SCN of Avp-CK1δ^−/−; Rosa26-LSL-Cas9 mice (S5A Fig) [41,42]. Injecting AAV-U6-gAvp-EF1α-DIO-mCherry efficiently reduced AVP immunoreactivity in the SCN (S5B–S5F Fig). However, the Avp knockdown did not significantly reduce the lengthening of the free-running period of these mice (S5G–S5I Fig). Thus, AVP peptide may be dispensable for transmitting the period length of the cellular clocks of AVP neurons to the entire SCN network.

To regulate the calcium rhythm of VIP neurons, SCN AVP neurons should have some functional connectivity with VIP neurons. Therefore, we tested the projection of AVP neurons to VIP neurons by anterograde tracing. To do so, we injected an AAV vector expressing the Synaptophysin-fused-GFP reporter in a Cre-dependent manner (AAV-EF1α-DIO-Synaptophysin::GFP) [43] into the SCN of Avp-Cre mice. Synaptophysin::GFP-positive fibers of SCN AVP neurons were scattered throughout the SCN (Fig 7A). We confirmed at least some, but not many, appositions of these fibers onto VIP neurons (Fig 7B). This result was consistent with a previous study reporting that AVP fibers make sparse contacts onto VIP neurons, whereas they make numerous contacts onto GRP and calretinin neurons; the latter two, in turn, make dense contacts onto VIP neurons [44].

Lastly, we investigated the functional connectivity between AVP and VIP neurons by recording [Ca²⁺]_i response of VIP neurons to optogenetic activation of SCN AVP neurons in vivo. To this end, we expressed ChrimsonR-mCherry, a red light-gated cation channel [45,46], and jGCaMP7s in AVP and VIP neurons, respectively, by injecting AAV-CAG-FLEX-ChrimsonR-mCherry and AAV-TRE-jGCaMP7s into the SCN of Avp-Cre; Vip-tTA mice (Fig 7C and 7D). Strikingly, optogenetic activation of SCN AVP neurons around ZT22, when [Ca²⁺]_i in AVP and VIP neurons is relatively low, acutely increased jGCaMP7s signal in VIP neurons as measured by fiber photometry (Fig 7E and 7F). These data suggest that SCN AVP neurons can activate VIP neurons in vivo.

Previous studies utilizing neuron type–specific genetic manipulations of cellular periods have suggested the importance of NMS neurons, AVP neurons, Drd1a neurons, and VPAC2 neurons in setting the ensemble period of the SCN [21,23,24,25,47]. In contrast, although VIP is a critical synchronizer for SCN neurons, the TTFL in VIP neurons itself may contribute little to the pacesetting of the entire SCN network.

Recently, Hamnett and colleagues demonstrated that lengthening cellular periods in VPAC2 neurons, which were marked by a specific Cre driver line, also lengthens the free-running period of the circadian wheel-running rhythm [25]. However, PER2::LUC rhythms of SCN slices did not reflect the behavioral period of these mice. Such inconsistency between in vivo and ex vivo periods is similar to our current and previous observations for AVP neuron–specific lengthening of cellular periods. Furthermore, these manipulated VPAC2 neurons were mostly confined in the SCN shell and include approximately 85% of AVP neurons [28]. Therefore, their and our findings implicate that the ability of VPAC2/AVP neurons to set the SCN ensemble period in vivo is lost in slices. Interestingly, they reported that VPAC2 neurons additionally require the contribution of VIP neurons to dictate the SCN slice period [28]. This view also explains that the cellular period lengthening in NMS neurons, including both AVP and VIP neurons, successfully lengthened the periods of both SCN slices and behavior [21]. On the other hand, an essential significance of our current study is the demonstration that in vivo AVP neurons alone can control the circadian rhythms of other SCN neurons, such as VIP neurons, and determine the SCN ensemble period to control behavior rhythm.

Indeed, it was recently reported that the optogenetic stimulation of SCN AVP neurons advances the circadian behavior rhythm [48]. In this study, we also demonstrated that the optogenetic stimulation of these neurons immediately increased VIP neuronal [Ca²⁺]_i in vivo. However, whether this effect is mediated by a direct or indirect connection of AVP neurons to VIP neurons remains elusive. An SCN connectome analysis reported only sparse projections of AVP neurons directly onto VIP neurons [44]. Instead, most GRP and calretinin neurons receive AVP fibers and send projections to VIP neurons in the SCN. Taken together, AVP neurons may functionally connect to VIP neurons indirectly via GRP or calretinin neurons.

So, what made the difference in periods between in vivo and ex vivo? AVP neurons may require direct or indirect connections with extra-SCN regions, lost in slices, to exert influence on the entire SCN. Another possibility, which seems more appealing to us, is that slicing damaged the usual connections and humoral environment inside the SCN so that AVP neurons are unable to transmit period information to other SCN neurons. It is also formally possible that the difference is due to more general factors, such as PER2::LUC rhythm resetting by culture and lower survival of CK1δ-deficient cells in slices. Another technical limitation of this study may be the methodological difference in how the period was calculated between ex vivo (PER2::LUC) and in vivo ([Ca²⁺]_i) measurements. Although Ca²⁺ imaging has been successfully performed in neonatal organotypic SCN cultures [49,50], we are unaware of any Ca²⁺ imaging studies in adult SCN cultures. Similarly, techniques to measure the PER2::LUC rhythm of VIP neurons in Avp-CK1δ^−/− mice in vivo are currently unavailable. Nevertheless, [Ca²⁺]_i and PER2::LUC are likely to oscillate with the same period but different phases most of the time [49,50].

We previously showed that GABA from AVP neurons is unnecessary for transmitting their cellular period to the behavioral period [36]. Here, we show that AVP is also dispensable for the circadian pacesetting in Avp-CK1δ^−/− mice. This finding is consistent with previous reports that genetic and pharmacological blockade of AVP signaling may attenuate interneuronal communication in the SCN but causes little change in the period of behavior or SCN slice PER2::LUC rhythms [51–55], with the exception of one study reporting that high doses of V1a or V1b receptor antagonists lengthened the PER2::LUC periods in SCN slices [56]. However, it is still possible that AVP and GABA act as redundant transmitters to mediate the AVP cellular period. Alternatively, other peptides expressed in SCN AVP neurons, such as NMS, Cholecystokinin, and Prokineticin 2 [21,57–59], may be responsible for the ensemble period-setting. In general, peptides are often engaged in volume transmission rather than synaptic transmission [60]. Therefore, the peptidergic transmission by AVP neurons might be more diffusible and attenuated in slice cultures. Also, because AVP neurons are distributed around the core, the ratio of AVP neurons to core cells was smaller in coronal sections containing both shell and core than in vivo. Such a condition may be another reason why the cellular period of AVP neurons was not reflected throughout the SCN in slices.

In this context, it was surprising that PER2::LUC rhythms were dissociated between shell and core in the SCN slices of CaMKIIα-CK1δ^−/− mice. Because, in contrast to Avp-CK1δ^−/− mice, all SCN neurons were supposed to lack CK1δ. A simple explanation for this observation is that CK1δ contributes to the cellular period-setting differently between shell and core cells. Its deficiency might lengthen the cellular period more in the shell than in the core, resulting in chimerism of the cellular periods within the SCN, as in Avp-CK1δ^−/− mice. It is technically challenging to precisely measure the intrinsic cellular period of each type of CK1δ^−/− SCN neurons (e.g., AVP neurons and VIP neurons) under conditions in which extracellular inputs are completely excluded [10,11,61]. The initial periods of PER2::LUC rhythms in the culture might reflect but were not the same as the intrinsic cellular periods. This is because there should be some intercellular communications that could affect the cellular periods.

The lengthening of the behavioral periods by approximately 0.8 to 0.9 h in Avp-CK1δ^−/− and CaMKIIα-CK1δ^−/− mice seemed to be less than the approximately 1.5- to 2-h lengthening observed in CK1δ^−/−peripheral cells, such as primary mouse embryonic fibroblasts (MEFs) and liver explants [8]. Nevertheless, such an apparent reduction of period lengthening in behavior rhythm may be attributed to the nature of in vivo SCN network. Indeed, the circadian period alteration caused by a series of CRY1 mutations was much smaller in behavior rhythms than in cellular oscillations of MEFs, converging close to 24 h in behavior [62].

In any case, this study demonstrated the importance of measuring the dynamics of the SCN network in vivo. Although the slice preparations are beneficial for examining the spatiotemporal organization of cellular clocks in the SCN, there is a large gap between the SCN slice and behavioral rhythms. Indeed, inconsistencies between the behavioral free-running period and the period of slice Per1-Luc, Per1-Gfp, or PER2::LUC oscillations have also been reported [22,63–67]. In vivo recording of the SCN dynamics will play a pivotal role in filling this gap. By combining genetically modified mice and AAV vectors utilizing Cre/loxP and Tet systems [40], we successfully measured the in vivo cellular oscillations of AVP and VIP neurons separately in mice with AVP neuron–specific gene knockout. Such a dual-targeting strategy would be potent for studying the interaction between 2 different types of cells within the SCN, a small but complicated network consisting of multiple types of neurons and glial cells.

In vivo calcium activity of AVP neurons was close to the sine curve (Fig 4C). In contrast, the calcium signal of VIP neurons was close to on and off square pulses (Fig 6C), which nearly delineated the (subjective) day. Such daily patterns of [Ca²⁺]_i were consistent with the previous reports [18,36,68]. In single SCN neurons in culture, clock gene expression rhythms are quasi-sinusoidal, whereas the neuronal activity rhythms are often quasi-rectangular in shape. [Ca²⁺]_i rhythms may be the intermediate and variable, which seems to be regulated by both the intercellular neuronal network and the intracellular modulators linking the TTFL to [Ca²⁺]_i [11,12,61,69–71].

By analogy, AVP-neuronal [Ca²⁺]_i rhythm may be regulated primarily by the intracellular TTFL mechanism, while neuronal firing may contribute more significantly to VIP-neuronal [Ca²⁺]_i. As discussed previously [28], some mechanisms in addition to the neuronal activity and Ca²⁺ influx, such as cAMP responsive element (CRE)-dependent transcription [72], may be essential for modulating TTFL in VPAC2/AVP neurons. The more rectangular [Ca²⁺]_i rhythm in VIP neurons is like an on-and-off switch. This shape may be suitable for setting time frames according to the LD cycle for locomotor activity, sleep-wakefulness, and other bodily functions.

As discussed above, AVP neurons are likely the principal pacesetter of the SCN network clock. In addition, reassessment of the behavioral rhythm in mice with AVP neuron–specific Bmal1 deletion (Avp-Bmal1^−/− mice) revealed that their wheel-running rhythm was disrupted similarly to mice with NMS or VPAC2 neuron–specific Bmal1 deletion [21,25].

Altogether, we propose the following model to integrate findings in the previous reports and this study. AVP neurons function as the primary oscillatory part of the SCN network, determining the period and taking charge of the oscillation itself. However, the cellular clocks of AVP neurons may be sloppy and not necessarily self-sustaining. Therefore, the sustained oscillation of the cellular clocks of AVP neurons may need to be driven by other SCN neurons. VIP neurons likely play this role, couple AVP neuronal population, and transmit information about external light to regulate phases of AVP-cellular rhythms. In this way, VIP is critical for circadian pacemaking. On the other hand, the TTFL of AVP neurons regulates the period of rhythmic VIP release. Thus, the functional differentiation and reciprocal interaction between cell types in the SCN network exert the central clock function.

All experimental procedures were approved by the Kanazawa University Animal Experiment Committee (approval numbers: AP-173836, AP-224337), the Kanazawa University Safety Committee for genetic recombinant experiments (approval numbers: Kindai 6–2124, 6–2598), the Institutional Animal Care and Use Committee at the Meiji University (approval number: IACUC20-126), and Meiji University Recombinant DNA Experiment Safety Committee (approval number: D-19-4). The study was carried out in compliance with Fundamental Guidelines for Proper Conduct of Animal Experiment and Related Activities in Academic Research Institutions under the jurisdiction of the Ministry of Education, Culture, Sports, Science and Technology (Notice No. 71 of 2006), and Standards relating to the Care and Keeping and Reducing Pain of Laboratory Animals (Notice of the Ministry of the Environment No. 88 of 2006).

Avp-Cre and Vip-tTA mice were reported previously [22,40]. This Avp-Cre line is a transgenic mouse harboring a modified BAC transgene, which has an insertion of codon-improved Cre recombinase gene immediately 5′ to the translation initiation codon of exogenous Avp gene in the BAC but without manipulation of the endogenous Avp loci in the mouse. Thus, the strain is different from the widely used Avp-ires2-Cre (JAX #023530) that shows hypomorphic expression of AVP [73], as well as from another Avp-ires-Cre line [74]. CK1δ^flox (JAX #010487) [8], Bmal1^flox mice (JAX #007668) [75], Vip-ires-Cre (Vip^tm1(cre)Zjh/J, JAX #010908) [33], and Rosa26-LSL-Cas9-2A-EGFP mice (JAX #026175) [41] were obtained from Jackson Laboratory. CaMKIIα-Cre (Camk2a::iCreBAC) [30] was obtained from the European Mouse Mutant Archive (EMMA #01153). The Per2::Luc reporter mice were provided by Dr. Joseph Takahashi [35]. All lines were congenic on C57BL/6: Avp-Cre and Vip-tTA mice were generated on C57BL/6J; CK1δ^flox (N>12), Bmal1^flox (N>9), Vip-ires-Cre (N>7), Per2::Luc (N>10), and Rosa26-LSL-Cas9-2A-EGFP (N>7) were backcrossed to C57BL/6J before use; CaMKIIα-Cre was backcrossed to C57BL/6N (N>10), then crossed to CK1δ^flox and Per2::Luc. We compared the conditional knockouts with controls whose genetic backgrounds were comparable. Avp-Cre, CaMKIIα-Cre, Vip-ires-Cre, Per2::Luc, Vip-tTA, and Rosa26-LSL-Cas9-2A-EGFP mice were used in hemizygous or heterozygous condition. We used both male and female mice in our experiments (S1 Data). Whether we pooled data from both sexes or analyzed them separately, the conclusions we reached remained the same. Mice were maintained under a strict 12-h light/12-h dark cycle in a temperature- and humidity-controlled room and fed ad libitum.

Male and female CaMKIIα-CK1δ^−/− (CaMKIIα-Cre; CK1δ^flox/flox) (Fig 1), Vip-CK1δ^−/− (Vip-ires-Cre; CK1δ^flox/flox) (Fig 2), and Avp-CK1δ^−/−; Rosa26-LSL-Cas9-2A-EGFP (Avp-Cre; CK1δ^flox/flox; Rosa26-LSL-Cas9-2A-EGFP) (S5 Fig) mice, aged 8 to 20 weeks, were housed individually in a cage placed in a light-tight chamber (light intensity was approximately 100 lux). Spontaneous locomotor activity (home-cage activity) was monitored by infrared motion sensors (O’Hara) in 1-min bins as described previously [22]. CK1δ^flox/flox and CaMKIIα-Cre; CK1δ^wt/flox littermates were used together as the control for CaMKIIα-CK1δ^−/− mice. CK1δ^flox/flox and Vip-ires-Cre; CK1δ^wt/flox littermates were used as the control for Vip-CK1δ^−/− mice. Because 2 control lines (CK1δ^flox/flox and Cre; CK1δ^wt/flox) for each conditional knockout behaved similarly, we regarded them together as control. Actogram, activity profile, and χ² periodogram analyses were performed via ClockLab (Actimetrics). The free-running period was measured by periodogram for days 10 to 24 (Figs 1 and 2) or the last 10 days in DD (S5 Fig). The activity time was calculated from the daily activity profile (average pattern of activity) of the same 15 days using the mean activity level as a threshold for detecting the onset and the offset of activity time [22]. Wheel-running activity was measured for Avp-CK1δ^−/− (Avp-Cre; CK1δ^flox/flox), Avp-Bmal1^−/− (Avp-Cre; Bmal1^flox/flox), and control mice as described previously [76]. Briefly, each mouse was housed in a separate cage equipped with a running wheel (12-cm diameter; SANKO, Osaka, Japan). The cages were placed in ventilated boxes; the light intensity at the bottom of the cage was 200 to 300 lux. The number of wheel revolutions was counted in 1-min bins. A chronobiology kit (Stanford Software Systems) and ClockLab software (Actimetrics) were used for data collection and analyses. The free-running period was measured by periodogram for the last 7 days in DD (S2 Fig).

The Avp-CK1δ^−/−, CaMKIIα-CK1δ^−/−, and Vip-CK1δ^−/− mice were further mated with Per2::Luc reporter mice [35] (Avp-Cre; CK1δ^flox/flox; Per2::Luc, CaMKIIα-Cre; CK1δ^flox/flox; Per2::Luc, Vip-Cre; CK1δ^flox/flox; Per2::Luc) and compared to control mice (CK1δ^flox/flox; Per2::Luc). Male and female mice aged 8 to 17 weeks were housed in LD before sampling. Coronal SCN slices of 150 μm were made at ZT8–9 with a linear slicer (NLS-MT; Dosaka-EM). The SCN tissue at the mid-rostrocaudal region was cultured, and its PER2::LUC bioluminescence was imaged every 30 min with an exposure time of 25 min with an EMCCD camera (Andor, iXon3), as described previously [23].

Images were analyzed using ImageJ and MATLAB (MathWorks). The resolution of images was adjusted to 4.6 μm/pixel for further analyses. Bioluminescence values in individual pixels were detrended by subtracting 24-h moving average values and then were smoothened with a 15-point moving average method. The middle of the time points crossing the value 0 upward and downward was defined as peak phases, whereas the maximum value between these 2 time points was defined as amplitude. The intervals between 2 adjacent peak phases were calculated as the periods [22]. The values of the regions lateral to the SCN were regarded as background. Therefore, pixels with values lower than the background were eliminated for further analyses. Periods of individual pixels’ oscillation were calculated for every cycle. Then, square regions of interest (ROIs, 15 × 15 pixels) were defined in the shell and core of the left and right SCN. Pixels with a period shorter than 4 h or longer than 40 h were eliminated to calculate the mean period, amplitude, and peak phase. Pixels in the left and right ROIs were first processed separately. Then, the averaged values of these two were considered representatives of individual mice.

The AAV-2 ITR containing plasmids pGP-AAV-CAG-FLEX-jGCaMP7s-WPRE (Addgene plasmid #104495, a gift from Dr. Douglas Kim and GENIE Project) [37] and pAAV-TRE-EGFP (Addgene plasmid #89875, a gift from Dr. Hyungbae Kwon) [77] were obtained from Addgene. pAAV-TRE-jGCaMP7s and pAAV-U6-gAvp-EF1α-DIO-mCherry were described previously [40,42]. pAAV-U6-gControl-EF1α-DIO-mCherry contains spacer sequences from pX333 (Addgene plasmid #64073, a gift from Dr. Andrea Ventura) instead of gRNA sequences for Avp. pAAV-EF1α-DIO-synaptophysin::GFP [43] was kindly provided by Dr. Richard D. Palmiter, University of Washington. pAAV-CAG-FLEX-ChrimsonR-mCherry was generated by replacing a NheI-AscI fragment containing EGFP of pAAV-CAG-FLEX-EGFP [36] with a NheI-AscI fragment containing ChrimsonR-mCherry PCR-amplified from pAAV-TRE-ChrimsonR-mCherry, which was acquired from Addgene (#92207, a gift from Dr. Alice Ting) [46]. As a negative control for the optogenetic study, we injected AAV-EF1α-DIO-hM3Dq-mCherry, which was generated with a plasmid pAAV-EF1α-DIO-hM3Dq-mCherry provided by Dr. Bryan Roth, University of North Carolina [78]. Recombinant AAV vectors (AAV2-rh10) were produced using a triple-transfection, helper-free method and purified as described previously [22]. The titers of recombinant AAV vectors were determined by quantitative PCR: AAV-CAG-DIO-jGCaMP7s, 3.4 × 10¹³; AAV-TRE-jGCaMP7s, 6.3 × 10¹¹; AAV-TRE-EGFP, 5.7 × 10¹¹; AAV-U6-gAvp-EF1α-DIO-mCherry, 6.4 × 10¹²; AAV-U6-gControl-EF1α-DIO-mCherry, 2.6 × 10¹²; AAV-EF1α-DIO-synaptophysin::GFP, 1.2 × 10¹³; AAV-CAG-FLEX-ChrimsonR-mCherry, 1.5 × 10¹³; and AAV-EF1α-DIO-hM3Dq-mCherry, 4.5 × 10¹² genome copies/ml. Stereotaxic injection of AAV vectors was performed as described previously [22]. Two weeks after surgery, we began monitoring the mice for their locomotor activity.

We used 6 Avp-CK1δ^−/− × Vip-tTA (Avp-Cre; CK1δ^flox/flox; Vip-tTA) mice, 4 Avp-CK1δ^−/− (Avp-Cre; CK1δ^flox/flox) mice, and 11 control mice (8 Avp-Cre; CK1δ^wt/flox; Vip-tTA and 3 Avp-Cre; CK1δ^wt/flox). We combined the data of Avp-Cre; CK1δ^flox/flox; Vip-tTA and Avp-Cre; CK1δ^flox/flox for AVP neurons recording considering the lack of intergroup differences. Additional 4 mice (Vip-tTA) were used for control experiments to measure EGFP signal in VIP neurons. The mice were anesthetized by administering a cocktail of medetomidine (0.3 mg/kg), midazolam (4 mg/kg), and butorphanol (5 mg/kg) and were secured at the stereotaxic apparatus (Muromachi Kikai). Lidocaine (8%) was applied for local anesthesia before making the surgical incision. We drilled small hole in the exposed region of the skull using a dental drill. We injected 0.5 to 1.0 μL of the virus (AAV-CAG-DIO-jGCaMP7s or AAV-TRE-jGCaMP7s) (flow rate = 0.1 μL/min) at the right SCN (posterior: 0.5 mm, lateral: 0.25 mm, depth: 5.7 mm from the bregma) with a 33 G Hamilton Syringe (1701RN Neuros Syringe, Hamilton) to label AVP neurons. Subsequently, we placed an implantable optical fiber (400 μm core, N.A. 0.39, 6 mm, ferrule 2.5 mm, FT400EMT-CANNULA, Thorlabs) above the SCN (posterior: 0.2 mm, lateral: 0.2 mm, depth: 5.2 to 5.4 mm from the bregma) with dental cement (Super-bond C&B, Sun Medical). The dental cement was painted black. Atipamezole (0.3 mg/kg) was administered postoperatively to reduce the anesthetized period. The mice were used for experiments 2 to 7 weeks after the virus injection and optical fiber implantation. Their ages were 3 to 10 months old, including both male and female.

A fiber photometry system (COME2-FTR, Lucir) was used to record the calcium signal of SCN neurons in freely moving mice [36,79]. Fiber-Coupled LED (M470F3, Thorlabs) with LED Driver (LEDD1B, Thorlabs) was used as an excitation blue light source. The light was reflected by a dichroic mirror (495 nm), went through an excitation bandpass filter (472/30 nm), then to the animal via a custom-made patch cord (400 um core, N.A. 0.39, ferrule 2.5 mm, length 50 cm, COME2-FTR/MF-F400, Lucir) and the implanted optical fiber. We detected the jGCaMP7s fluorescence signal by a photomultiplier through the same optical fibers and an emission bandpass filter (520/36 nm); furthermore, we recorded the signal using Power Lab (AD Instruments) with Lab Chart 8 software (AD Instruments). The excitation blue light intensity was 2.5 to 100 μW at the tip of the patch cord of the animal side. We recorded the same for 30 s every 10 min in 2 weeks to reduce photobleaching. During the recording, the mouse was housed in a 12-h light–dark cycle for more than 5 days (LD condition) and then moved to continuous darkness for approximately 10 days (DD condition) in a custom-made acrylic cage surrounded by a sound-attenuating chamber. A rotary joint for the patch cord was stopped during the recording to prevent artificial baseline fluctuation. The animal’s locomotor activity was monitored using an infrared sensor (Supermex PAT.P and CompACT AMS Ver. 3, Muromachi Kikai).

The detected jGCaMP7s signal was averaged within a 30-s session [36]. To detrend the gradual decrease of the signal during recording days, ±12 h average from the time (145 points) was calculated as baseline (F). The data were subsequently detrended by the subtraction of F (ΔF). Then, the ΔF/F value was calculated. To determine the peak phase of jGCaMP7s calcium signal, ΔF/F were smoothened with a 21-point moving average, then the middle of the time points crossing value 0 upward and downward were defined as peak phases. Additionally, the intervals between peak phases were defined as the periods [22]. A double-plotted actogram of jGCaMP7s or EGFP signal was designed by converting all ΔF to positive values by subtracting the minimum value of ΔF. Subsequently, these values were multiplied with 100 or 1,000 and rounded off. The plots were made via ClockLab (Actimetrics) with normalization in each row. A double-plotted actogram of locomotor activity was also prepared and overlaid on that of jGCaMP7s signal.

The onset and offset of locomotor activity were determined using the actogram of locomotor activity. Initially, we attempted to automatically detect the onset and offset; however, it was followed by a manual visual inspection, along with modifications by the experimenter. To calculate CT of the peak phases of GCaMP signal, we have defined the regression line of locomotor activity onsets as CT12.

We confirmed the jGCaMP7s expression and the position of the optical fiber by slicing the brains in 100 μm coronal sections using a cryostat (Leica). The sections were mounted on glass slides with a mounting medium (VECTASHIELD HardSet with DAPI, H-1500, Vector Laboratories) and observed via epifluorescence microscope (KEYENCE, BZ-9000E).

We used 8 Avp-Cre; Vip-tTA mice (n = 4 for ChrimsonR, n = 4 for control, both male and female). We injected 1.0 μL of the mixture of viruses (AAV-TRE-jGCaMP7s with AAV-CAG-Flex-ChrimsonR-mCherry or AAV-EF1α-DIO-hM3Dq-mCherry and AAV-TRE-jGCaMP7s) at the right SCN (posterior: 0.5 mm, lateral: 0.25 mm, depth: 5.7 mm from the bregma) and then implanted an optical fiber (400 μm core, N.A. 0.39, 6 mm, Thorlabs) above the SCN (posterior: 0.2 mm, lateral: 0.2 mm, depth: 5.3 mm from the bregma) with dental cement. The mice were used for experiments more than 2 weeks after the surgery.

A fiber photometry system (FP3002, Neurophotometrics) was used to record the calcium signal of SCN neurons with optogenetic stimulation in freely moving mice [80,81]. Excitation light sources were a 470-nm LED for detecting calcium-dependent jGCaMP7s fluorescence signal (F470) and a 415-nm LED for calcium-independent isosbestic fluorescence signal (F415). The duration of excitation lights is 50 ms, and the onsets of the excitation timing of LEDs were interleaved. The lights passed through excitation bandpass filters, dichroic mirrors, and then to the animal via fiber-optic patch cords (BBP(4)_400/440/900-0.37_1m_FCM-4xFCM_LAF, MFP_400/440/LWMJ-0.37_1m_FCM-ZF2.5_LAF, Doric Lenses) and the implanted optical fiber. Subsequently, both signals were detected using a CMOS camera through the optical fibers, dichroic mirrors, and emission bandpass filters. The recorded signals were acquired using Bonsai software, with a sampling rate of 10 Hz for each color. The excitation intensities of the 470-nm and 415-nm LED at the animal side’s patch cord tip were 130 μW and 90μW, respectively. Additionally, a 635-nm red laser (inside the fiber photometry system FP3002) was transmitted through the same optical fibers with an intensity of 2 mW. During the 720-s recording at ZT22, optical stimulation (635 nm, 50 ms pulse, 5 Hz, 120 s, 600 pulse count) was applied in the middle of the recording. Throughout the experiment, the mice were housed in a custom-made acrylic cage surrounded by a sound-attenuating chamber and maintained in a 12-h light–dark (LD) cycle.

The recorded data were interleaved to eliminate artifacts caused by red laser stimulation, and half of it was discarded. Consequently, the final sampling rate for the jGCaMP7s fluorescence signals at the 470-nm light excitation (F470) and the 415-nm excitation (F415) was 5 Hz. Ratio (R) was defined as the ratio between F470 and F415 (F470/F415) for calibration and reducing motion artifacts. The baseline ratio (R₀) was calculated as the mean R-value during the pre-laser stimulation period (−30 s to 0 s from the start of the laser stimulation, 150 points). R₁ was calculated as the mean R-value during the late period of laser stimulation (90 s to 120 s from the start of the laser stimulation, 150 points). The response ΔR/R₀% was computed as the difference between R₁ and R₀ (ΔR) divided by R₀ (ΔR / R₀). All data analysis was done using MATLAB. After the recordings were completed, we histologically confirmed the jGCaMP7s and ChrimsonR-mCherry expressions and the position of the optical fiber.

We used 15 Avp-CK1δ^−/−; Rosa26-LSL-Cas9-2A-EGFP mice for the Avp knockdown study (S5 Fig, 9 or 6 for gAvp or gControl) and 2 Avp-Cre mice for the Synaptophysin::GFP tracing study (Fig 7A and 7B). Immunostaining was performed as described previously [22,40,42]. Mice were killed approximately at ZT8 by transcardial perfusion of PBS followed by 4% paraformaldehyde (PFA) in PBS. Then, the mice brains were postfixed in the 4% PFA at 4°C overnight, followed by immersion in 30% sucrose solution at 4°C for 2 days. Serial coronal brain sections (30 mm thickness) were made with a cryostat (CM1860, Leica) and collected in 4 series—one of which was further immunostained. For immunofluorescence staining, sections were washed with PBS containing 0.3% Triton X-100 (PBST) and blocked with PBST plus 3% BSA (blocking solution). Then, slices were incubated overnight with the designated primary antibodies in the blocking solution at 4°C. Antibodies used were rabbit anti-AVP antibody (1:4,000; AB1565, Merck Millipore) and rabbit anti-VIP antibody (1:1,000; #20077, Immunostar). Then, slices were washed with PBST, followed by incubation with the designated secondary antibodies in blocking solution for 4 h. Secondary antibodies used in this study were Alexa Fluor 488–conjugated donkey anti-rabbit IgG antibody (1:2,000, A-21206; Thermo Fisher Scientific) and Alexa Fluor 594–conjugated donkey anti-rabbit IgG antibody (1:2,000, A-21207; Thermo Fisher Scientific). After incubation with secondary antibody, slices were washed with PBS, mounted on slide glasses, air dried, and coverslipped using Mounting Medium (H-1500, Vector Laboratories; Dako Fluorescence Mounting Medium, Agilent Technologies). Images were taken using an epifluorescence microscope (BZ-9000, Keyence) or a confocal microscope (Fluoview Fv10i, Olympus). The AVP expression levels were quantified by Photoshop (Adobe) as follows [40]. First, the images were transformed to grayscale. Then, the mean intensities of pixels within the SCN were calculated. Finally, the values of the region lateral to the SCN were regarded as background and were subtracted from those of the SCN. When slices from Avp-CK1δ^−/−; Rosa26-LSL-Cas9-2A-EGFP mice were immunostained for AVP in green, green fluorescence was emitted from both AVP immunostaining and EGFP. Therefore, to evaluate the AVP expression level, EGFP fluorescence was quantified in negative control slices, i.e., mock-treated slices without incubation with anti-AVP antibody, and compared to the fluorescent intensity of slices incubated with anti-AVP antibody.

All results are expressed as mean ± SEM. For comparisons of 2 groups, two-tailed Student or Welch t tests or Mann–Whitney U tests were performed. For comparisons of multiple groups with no difference of variance by Bartlett test, three-way or two-way repeated measures ANOVA or one-way ANOVA followed by post hoc Ryan test were performed. For comparisons of multiple groups with difference of variance by Bartlett test, nonparametric tests, Kruskal–Wallis test with post hoc Mann–Whitney U test with Bonferroni correction, Friedman rank sum test followed by Wilcoxon signed rank test with Bonferroni correction, and Wilcoxon signed rank test were performed with EZR, a graphical user interface for R [82] or ANOVA4. For circular data, Rayleigh test, Watson–Williams test, and Harrison–Kanji test were performed with Circstat MATLAB Toolbox for Circular Statistics [83]. All P values less than 0.05 were considered as statistically significant. Only relevant information from the statistical analysis was indicated in the text and figures.