Drosophila clock cells use multiple mechanisms to transmit time-of-day signals in the brain

Significance

Despite our growing understanding of how the fly clock network maintains free-running rhythms of behavior and physiology, little is known about how information is communicated from the clock network to the rest of the brain to regulate behavior. We identify glutamate and acetylcholine as key neurotransmitters signaling from clock neurons to the pars intercerebralis (PI), a clock output region regulating circadian rhythms of sleep and metabolism. We report a link between Drosophila evening clock neurons and the PI, and find that the effect of clock neurons on output neuron physiology varies, suggesting that the same clock cells use multiple mechanisms simultaneously to drive cycling in output neurons.

Abstract

Regulation of circadian behavior and physiology by the Drosophila brain clock requires communication from central clock neurons to downstream output regions, but the mechanism by which clock cells regulate downstream targets is not known. We show here that the pars intercerebralis (PI), previously identified as a target of the morning cells in the clock network, also receives input from evening cells. We determined that morning and evening clock neurons have time-of-day–dependent connectivity to the PI, which is regulated by specific peptides as well as by fast neurotransmitters. Interestingly, PI cells that secrete the peptide DH44, and control rest:activity rhythms, are inhibited by clock inputs while insulin-producing cells (IPCs) are activated, indicating that the same clock cells can use different mechanisms to drive cycling in output neurons. Inputs of morning cells to IPCs are relevant for the circadian rhythm of feeding, reinforcing the role of the PI as a circadian relay that controls multiple behavioral outputs. Our findings provide mechanisms by which clock neurons signal to nonclock cells to drive rhythms of behavior.

Many physiological and behavioral processes across organisms exhibit daily rhythms controlled by an internal circadian system. The temporal organization conferred by circadian clocks allows anticipation of environmental changes and the coordination of biochemical and physiological processes within and across cells and tissues. Work in Drosophila and other organisms has elucidated the molecular basis of circadian pacemaking in the brain, but we still lack a molecular understanding of how time-of-day information is relayed from brain clock circuitry to downstream output regions that regulate circadian physiology and behavior.

The Drosophila clock network consists of ∼150 neurons that express the core clock transcription-translation feedback loop genes (1). Of these, the ventrolateral neurons (LNvs) and dorsal lateral (LNd) clock neurons are important pacemakers for driving circadian rhythms in constant darkness, and coordinating the activity of other neurons in the clock network (2–5). Indeed, robust rhythms are an emergent property of the clock network as a whole (6–10). The rhythmic pattern of locomotor activity in light:dark (LD) cycles, which consists of morning and evening peaks that anticipate dawn and dusk, respectively, is also attributed to specific clock cells. The LNvs act through the neuropeptide pigment-dispersing factor (PDF) to drive the morning bout of locomotor activity while LNd clock neurons are termed evening neurons due to their role in regulating evening anticipatory activity (3, 11, 12). LNd neurons are a molecularly heterogeneous population with subpopulations expressing different signaling molecules including the classical transmitter acetylcholine and the neuropeptides ITP, NPF, and sNPF (13–16) DN1 dorsal neurons, which receive input from LNvs, also promote morning activity (17, 18) and exhibit highest activity in the predawn and morning hours (12, 19). DN1 neurons express glutamate and the neuropeptide DH31, both of which have previously described roles in regulating circadian rhythms (16, 17, 20, 21). The diverse array of signaling molecules expressed by neurons in the clock network offers many possibilities for signaling to clock output brain regions that lack their own clocks but are important for the generation of behavioral rhythm.

A major clock output region in Drosophila is the pars intercerebralis (PI), a protohypothalamic region implicated in regulating circadian locomotor rhythms and peripheral cycling (22–27). Like the hypothalamus, the PI controls aspects of sleep and feeding (28–32) and comprises a population of neurosecretory cells with heterogeneous peptide expression. PI neurons that express the neuropeptide diuretic hormone 44 (DH44), the fly ortholog of corticotropin-releasing factor, modulate rest:activity rhythms (23, 26). On the other hand, PI neurons producing Drosophila insulin-like peptides are implicated in circadian gene expression in the fat body (24, 33), but have not been linked to behavioral rhythms.

As PI cells do not express the molecular clock machinery, time-of-day information must be relayed directly or indirectly from clock neurons to the PI. Previous work showed that the DN1 clock neurons project to two groups of PI neurons, the diuretic hormone 44 positive (DH44+) cells and the insulin-producing cells (IPCs). However, while hyperactivation of DH44+ neurons or RNAi knockdown of DH44 peptide significantly dampens rest:activity rhythms in Drosophila (23), silencing of DN1 neurons has a considerably weaker effect (34). Thus, there must be additional upstream inputs to PI neurons that maintain rest:activity rhythms in the absence of DN1 signals. We hypothesized that robust circadian control of the PI would require inputs from both morning- and evening-active clock neurons, including perhaps long-distance signals directly from LNvs. We sought to determine if evening cells project to the PI and also investigated clock-PI signaling at different times of day. We demonstrate that both DH44+ neurons and IPCs receive time-of-day–dependent inputs from both CRYPTOCHROME-negative LNds and DN1s. Surprisingly, clock neurons inhibit DH44+ neurons while activating IPCs, indicating that the same clock cells use multiple mechanisms simultaneously to drive cycling in output neurons. We show also that morning clock cells and PI-secreted insulin-like peptides are required for rhythms of feeding.

Results

LNd and DN1 Clock Neurons Have Time-of-Day–Dependent Connectivity to the PI.

Previous work demonstrated that DN1 clock neurons have physical and functional connectivity to DH44+ cells and IPCs of the PI (23, 24); however, it is unlikely that a single morning-active clock population is sufficient to confer robust circadian regulation of PI neuron activity. We used GFP reconstitution across synaptic partners (GRASP) to test the hypothesis that the LNd clock neurons are also presynaptic to PI neurons using the dvPDF-Gal4 driver with PDF-Gal80 (35–37). Although we had mixed results regarding physical connections between LNd and DH44+ neurons, both CD4-GRASP and synaptically targeted neurexin-GRASP showed contact between LNd neurons and IPCs in brains dissected in the morning (data for neurexin-GRASP are in Fig. 1). The directional information provided by neurexin-GRASP indicates that LNd neurons project to IPCs (Fig. 1 A and B).

Fig. 1.

Clock neurons have time-of-day–dependent functional connectivity to the PI. (A) nrx-GRASP between CRY-negative LNd clock neurons and IPCs. Lower is zoom view of boxed region. (B) nrx-GRASP negative control, in which neurexin-GFP^1-10 is expressed in DN1 neurons. Lower is zoom view of boxed region. (C) Left: GCaMP6m signal over time in DH44+ neurons during activation of DN1 neurons. ZT 0: n = 66 cells, 15 brains, ZT 12: n = 42 cells, 11 brains. Black bar indicates timing of ATP application. Data are represented as mean ± SEM. Right: Minimum GCaMP change (ΔF/F), points represent values in individual cells, lines represent the mean ± SD for indicated genotypes and timepoints. ZT 0 control: n = 22 cells, 7 brains, ZT 12 control: n = 35 cells, 8 brains. (D) Left: GCaMP6m signal over time in IPCs during activation of DN1 neurons. Right: Maximum GCaMP ΔF/F for each cell. Data are plotted as described in C. ZT 0: n = 60 cells, 12 brains, ZT 12: n = 45 cells, 10 brains, ZT 0 control: n = 39 cells, 7 brains, ZT 12 control: n = 23 cells, 5 brains. (E) Left: GCaMP6m signal over time in DH44+ neurons during activation of LNd neurons. Right: Minimum GCaMP ΔF/F for each cell. Data are plotted as described in C. ZT 0: n = 90 cells, 23 brains, ZT 12: n = 83 cells, 30 brains, ZT 0 control: n = 45 cells, 14 brains, ZT 12 control: n = 44 cells, 13 brains. (F) Left: Average GCaMP6m signal over time in IPCs during activation of LNd neurons. Right: Maximum GCaMP ΔF/F for each cell. Data are plotted as described in C. ZT 0: n = 73 cells, 9 brains, ZT 12: n = 92 cells, 13 brains, ZT 0 control: n = 49 cells, 8 brains, ZT 12 control: n = 44 cells, 6 brains. *P < 0.05, **P < 0.01, and ***P < 0.001 by ANOVA followed by Tukey test for all panels.

As a positive GRASP signal is not evidence of functional contact, we used a calcium indicator-based stimulus-response assay to determine if clock cells signal to PI neurons at different times of day (Fig. 1 C–F). In addition to signaling from LNds, we asked if inputs from DN1s to PI cells are modulated by circadian time. In this assay, the adenosine triphosphate (ATP)-gated cation channel P2X2 was expressed in clock neurons to allow specific depolarization by ATP application, while the genetically encoded calcium indicator GCaMP6m was expressed in PI neurons to allow monitoring of intracellular calcium levels before and after clock neuron stimulation in acutely dissected brains (38, 39). To control for effects of possible leaky P2X2 expression, or direct effects of ATP on calcium signaling in the PI, we expressed GCaMP6m in PI neurons of flies carrying UAS-P2X2 but no GAL4 driver as a control.

Stimulation of clock neurons had divergent effects on DH44+ cells and IPCs depending on time of day. In the morning (zeitgeber time [ZT] 0 to 4), DN1 neuron activation resulted in a 21% ± 2% decrease in DH44+ cell GCaMP fluorescence, while in the evening (ZT 11 to 15) the 14% ± 1% reduction in the DH44+ GCaMP signal was indistinguishable from that in genetic controls (Fig. 1C). A time-dependent profile is also apparent for IPCs; DN1 neuron activation results in an 39% ± 10% increase in GCaMP fluorescence in the morning, and no change in the evening (Fig. 1D). It should be noted, however, that while the group response of both DH44+ cells and IPCs to DN1 activation in the morning was significant, it was also highly heterogeneous, with some cells showing strong responses while others were indistinguishable from controls (Fig. 1 C and D, Right). In contrast, the lack of response to DN1 activation in the evening in both cell groups was very consistent.

LNd neurons also show preferential morning modulation of DH44+ neurons (Fig. 1E), with 24% ± 2% inhibition of DH44+ GCaMP signaling in the morning and consistent lack of response in the evening. The IPC GCaMP response to LNd stimulation is heterogeneous in both the morning and the evening, with some neurons showing a large response to LNd activation at each time, while others show no response (Fig. 1 F, Right). Overall, in the morning there was a 58% ± 11% increase in IPC GCaMP signal; however, this was not significantly different from controls (43% ± 8%). In the evening there was a 68% ± 14% increase in IPC GCaMP signal, which was significantly different from controls (9% ± 6%). The increased GCaMP signal in morning controls may reflect the higher morning basal activity of IPCs previously described (24). The heterogeneity of the IPC response to LNd stimulation obscures some aspects of the data when averaging. Thus, we used a 10% change in GCaMP6m signal during ATP application as a cutoff to delineate responding and nonresponding neurons to allow us to compare the magnitude of response. The magnitude of the GCaMP ΔF/F increase in responding neurons was not significantly different between morning and evening time points at 87% ± 11% in the morning and 60% ± 9% in the evening. The proportion of responding neurons varied only slightly by time of day, with 57% responding at ZT 0 to 4 and 65% responding at ZT 11 to 15. We surmise that the small statistical difference between the morning and evening response does not reflect a biological change and that DH44+ neurons respond to LNds in the morning and evening.

Application of Clock Neuron-Derived Neuropeptides Alters Excitability of PI Neurons.

We next asked if neuropeptides produced by clock cells and implicated in various aspects of circadian rhythms signal to the PI to regulate behavioral rhythms. Diuretic hormone 31 (DH31), produced by dorsal clock neurons, regulates temperature preference rhythms through its canonical receptor, DH31-R, and also works in concert with PDF to regulate circadian locomotor rhythms through an unknown signaling pathway (21, 40). Bath application of 1 µM DH31 peptide onto acutely dissected GCaMP6m-expressing brains caused an acute increase in intracellular calcium in DH44+ neurons (19% ± 5%) (SI Appendix, Fig. S1 A and B). Also, 1 µM DH31 had a heterogeneous effect on IPCs, with 22/52 neurons showing an increase in intracellular calcium as measured by GCaMP6m fluorescence (22% ± 3%) and 30/52 neurons showing no change in GCaMP signal (2% ± 1%) (SI Appendix, Fig. S1 A and B). We previously reported that the IPCs are heterogeneous in their response to signals from DN1s, with some neurons showing an increase in intracellular calcium and others showing no change (24).

Neuropeptide F (NPF), produced by LNd clock neurons, regulates free-running locomotor period and the amplitude of evening activity in a light:dark cycle, and also regulates circadian gene transcription in the fat body (14, 41). As with DH31, application of 1 µM NPF increased intracellular calcium in DH44+ neurons (12% ± 6%) (SI Appendix, Fig. S1 C and D). However, NPF had little effect on IPCs (−6% ± 2%).

To examine roles for clock neuron-derived neuropeptides in modulating circadian locomotor output through DH44+ neurons, we examined locomotor rhythms in constant darkness (DD) after DH44-Gal4–driven RNAi knockdown of DH31 and NPF receptors. We saw no consistent change in total activity, period, or rhythm strength relative to parental controls in constant conditions; this is consistent with previous studies of DH31, which did not find effects on freerunning rhythms, and suggest that effects of NPF on behavior do not depend upon DH44 neurons (SI Appendix, Fig. S2 A–C). Note that a behavioral difference would be considered significant only if the experimental line is significantly different from both parental controls. We also considered the possibility that PDF acts directly on the PI; although LNvs are not known to directly contact the PI, they project close to it in the protocerebrum, and so limited diffusion of PDF is possible. As noted above, PDF signaling is implicated in freerunning rhythms and also in the maintenance of a characteristic morning-evening pattern in light:dark cycles; pdfr mutants have reduced morning anticipation and phase shifted evening anticipation in LD conditions (42). While knockdown of PDFR in DH44 cells did not affect rhythms in constant conditions, it reduced the evening peak of activity in a 12:12 light:dark cycle (SI Appendix, Fig. S3).

Acetylcholine and Glutamate Are Required for Clock-to-PI Signaling.

Two major fast neurotransmitters are produced by the clock neurons upstream of the PI: glutamate from DN1 neurons, and acetylcholine from a subset of LNd neurons (13, 16, 20, 34). To examine the role of clock-secreted fast transmitters in the clock-to-PI circuit, we conducted the stimulus-response paradigm described above in the presence of specific neurotransmitter receptor blockers. Specifically, we asked if pretreatment with a blocker attenuated the effect of P2X2 stimulation of upstream clock neurons on downstream PI intracellular calcium levels. All experiments were conducted in the morning, as this was the time of maximal PI neuron response for three out of four connections. We hypothesized that NMDAR or mGluR blockers would be most effective in attenuating the response of PI neurons to DN1 stimulation, while mAChR or nAChR blockers would attenuate the response to LNd stimulation. To our surprise, the most effective blocker was unique to each of the four connections tested and did not necessarily correlate with the neurotransmitter believed to be released by the upstream clock population (Fig. 2).

Fig. 2.

Acetylcholine and glutamate receptor blockers attenuate clock-to-PI signaling. (A) Left: GCaMP6m signal over time in DH44+ neurons during activation of DN1 neurons in the presence of the mGluR blocker LY341495. Control data for ATP activation only are replotted from Fig. 1C (orange). Black bar indicates timing of ATP application; gray bar indicates timing of blocker application. Data are represented as mean ± SEM. Right: Minimum GCaMP change (ΔF/F), points represent values in individual cells, lines represent the mean ± SD for indicated receptor blocker. Control data for ATP inhibition only are again replotted from Fig. 1C (orange). N_atropine = 28 cells from 8 brains. N_LY341495 = 38 cells from 9 brains. N_MK-801 = 21 cells from 6 brains. N_tubocurarine = 31 cells from 8 brains. (B) Left: Data are plotted as in A for IPCs during activation of DN1 neurons in the presence of the mGluR and AChR blockers. Control data for ATP activation only are replotted from Fig. 1D (orange). Right: Maximum GCaMP ΔF/F for individual cells. N_atropine = 40 cells from 8 brains. N_LY341495 = 50 cells from 8 brains. N_MK-801 = 51 cells from 9 brains. N_tubocurarine = 22 cells from 4 brains. (C) Left: Data are plotted as in A for DH44+ neurons during activation of LNd neurons in the presence of the mGluR and AChR blockers. Control data for ATP activation only are replotted from Fig. 1E (orange). Right: Minimum GCaMP ΔF/F for individual cells. N_atropine = 19 cells from 10 brains. N_LY341495 = 48 cells from 8 brains. N_MK-801 = 48 cells from 16 brains. N_tubocurarine = 46 cells from 13 brains. (D) Left: Data are plotted as in A for IPCs during activation of LNd neurons in the presence of the mGluR and AChR blockers. Control data for ATP activation only are replotted from Fig. 1F (orange). Right: Maximum GCaMP change (ΔF/F) for individual cells. N_atropine = 52 cells from 8 brains. N_LY341495 = 62 cells from 9 brains. N_MK-801 = 72 cells from 10 brains. N_tubocurarine = 62 cells from 10 brains. All data were collected at ZT 0 to 4. Asterisks indicate comparisons between each drug treatment and no-drug control (far Left); **P < 0.01 and ***P < 0.001 by ANOVA followed by Tukey test for all panels.

For the DN1-to-DH44+ neuron connection, all blockers tended to attenuate the reduction in intracellular calcium caused by DN1 neuron stimulation. However, only the mGluR blocker LY341495 caused a statistically significant change in the GCaMP response (Fig. 2A). In some instances, LY341495 uniquely reversed the effect of DN1 neuron stimulation, such that we observed an average increase in DH44+ neuron intracellular calcium in the presence of the blocker (Fig. 2A).

For the DN1-to-IPC connection, all blockers tested significantly attenuated the increase in intracellular calcium caused by DN1 neuron stimulation. The nAChR blocker (tubocurarine), mAChR blocker (atropine), and NMDAR blocker (dizocilpine) all completely attenuated the IPC GCaMP response to DN1 neuron stimulation (Fig. 2B). Administration of the mGluR blocker LY341495 substantially reduced the IPC GCaMP response, but was slightly less effective than the other blockers (Fig. 2B).

For the LNd-to-DH44+ neuron connection, all blockers tested trended toward an attenuation in the reduction in intracellular calcium caused by LNd neuron stimulation. However, only the mAChR blocker (atropine) and the mGluR blocker (LY341495) resulted in a statistically significant attenuation of the DH44+ neuron GCaMP response to LNd simulation (Fig. 2C).

For the LNd-to-IPC connection, all blockers tested significantly attenuated the increase in intracellular calcium caused by LNd neuron stimulation. Without any blocker, the IPC response to DN1 stimulation by ATP was heterogeneous, but resulted in an average increase in GCaMP fluorescence (Fig. 1D) In the presence of mAChR and NMDAR blockers (atropine and MK-801) the ATP-induced increase in intracellular calcium was completely blocked (Fig. 2D). The mGluR and nAChR blockers (LY341495 and tubocurarine) significantly reduced the ATP-induced increase in intracellular calcium compared to controls, but IPCs still responded with an increase in calcium above baseline (Fig. 2D).

To examine roles for clock neuron-derived glutamate or acetylcholine modulation of circadian output through DH44+ neurons, we examined locomotor rhythms in DD after DH44-Gal4–driven RNAi knockdown of acetylcholine and glutamate receptors (SI Appendix, Fig. S4). mAChR-A knockdown with two RNAi lines resulted in a significant increase in period, primarily due to outlier effects (SI Appendix, Fig. S4B). One mGluR RNAi knockdown line lengthened period and reduced rhythmicity, but this was not recapitulated by other RNAi lines against the same target (SI Appendix, Fig. S4C). In general, we conclude that knockdown of any single glutamate or acetylcholine receptor in DH44+ neurons is not sufficient to induce alteration in behavioral rhythms, though that does not rule out participation of these receptors in concert with other redundant signaling mechanisms.

Clock Signaling through IPCs Alters Feeding Rhythms and Food Intake.

We next asked if the timed clock inputs into the PI could account for the phase of behavioral rhythms. Importantly, morning activation of the IPC population by DN1s corresponds to the morning peak of firing we previously reported for these cells (24). Given that feeding is also maximal in the morning and electrical activity in IPCs is also sensitive to feeding (23), we asked if insulin-like peptides themselves are circadian output molecules regulating rhythmic feeding. We used the Activity Recording CAFE (ARC) assay (43) to quantify single-fly feeding for 72 h, with the first day in a 12:12 LD cycle and the following 2 d in DD. We first validated that our wild-type (WT) iso31 flies show rhythmic feeding and that this rhythm is lost in the absence of a functional clock, e.g., in flies lacking the period gene (per⁰¹) (Fig. 3 A and B). As expected, WT flies exhibited feeding rhythms in LD that were maintained in DD. per⁰¹ flies lack feeding rhythms in both LD and DD, despite the fact that per⁰¹ flies exhibit rhythmic locomotor activity in LD (44). Additionally, per⁰¹ flies ate significantly more food per day than WT flies throughout the assay (Fig. 3B).

Fig. 3.

The clock controls daily feeding rhythms via PDF+ morning cells. (A) Average consumption in microliters per fly per hour for wild type (black, n = 23) and per⁰¹ clock mutant (red, n = 20) over 3 d. Error bars indicate SEM. Greyscale bars at the Top indicate light conditions. White: lights on; light gray: lights off (subjective day), dark gray: lights off (night). JTK_cycle P values were calculated by 24-h day for each genotype; significance is indicated below lighting bars. P_JTK < 0.001 = ***; P_JTK > 0.05 = not significant (n.s.). (B) 12-h food intake in microliters during the day (ZT/circadian time [CT] 0 to 12) vs. night (ZT/CT 12 to 23) for flies shown in A. Day vs. night feeding was compared within each genotype, and 24-h food intake was compared between genotypes by Student’s t test. *P < 0.05, **P < 0.01, and ***P < 0.001. (C) Average consumption per fly per hour as in A for PDF-Gal4 > UAS-Kir2.1 (orange, n = 24), Gal4 control (black, n = 25) and UAS control (gray, n = 25). (D) 12-h food intake for flies shown in C, plotted as in B. There were no significant differences in 24-h food intake between genotypes. (E) Average consumption per fly per hour for as in A for dvPDF-Gal4; PDF-Gal80 > Kir2.1 (blue, n = 31), Gal4 control (black, n = 31) and UAS control (gray, n = 31). (F) 12-h food intake for flies shown in E, plotted as in B. There were no significant differences in 24-h food intake between genotypes.

As the clock populations relevant for feeding rhythms have not been identified, we examined the contribution of pdf+ “morning” cells and LNd “evening” cells to rhythmic food intake (Fig. 3 C–F). Suppression of morning cell activity by expression of the hyperpolarizing inwardly rectifying potassium channel Kir2.1 resulted in a loss of feeding rhythms in both LD and DD conditions, while controls remained rhythmic in both conditions (Fig. 3C). Total daily food intake was not altered by pdf-Gal4–driven expression of Kir2.1 (Fig. 3D). Suppression of evening cell activity, however, did not result in a loss of feeding rhythm in LD or DD and did not change daily food intake compared to controls (Fig. 3 E and F).

To assess the role of insulin-like peptides in regulating circadian feeding, we examined DILP2 and DILP2,3 insertion mutants (ΔDILP2 and ΔDILP2,3) in the ARC assay (Fig. 4). In the DILP2 mutant (Fig. 4 A and B), rhythmic feeding was maintained in constant darkness, though the timing of the daily feeding peak was delayed relative to controls, from CT3 to CT6. There was no significant difference in 24-h food intake between ΔDILP2 mutant and controls. Because loss of one or more ILPs can result in compensatory up-regulation of the remaining ILPs (45), the contribution of PI-derived ILPs is best studied using a triple knockout of ILPs 2, 3, and 5. However, the ΔDILP2,3,5 flies have multiple metabolic and reproductive defects that make them unsuitable for use in the feeding assay; hence we used ΔDILP2,3 flies. Loss of both DILPs 2 and 3 resulted in not only a loss of rhythm, but also a substantial reduction in total food intake (Fig. 4 C and D). Control flies ate 0.82 ± 0.06 μL/d, while ΔDILP2,3 flies ate 0.51 ± 0.03 μL/d. While ΔDILP2,3 flies maintain a low amplitude rhythm in LD, this rhythm in immediately dampened in DD and is not detectable by JTK_cycle. The role of morning clock cells and DILP2 in the feeding rhythm is consistent with the timing of morning cell input to IPCs and with the phase of the feeding rhythm.

Fig. 4.

Brain-derived insulin regulates timing and amount of food intake. (A) Average consumption in microliters per fly per hour for DILP2^−/+ control (black, n = 54) and DILP2^−/− insertion mutant (magenta, n = 69) over 3 d. Error bars indicate SEM. Greyscale bars at the Top indicate light conditions. White: lights on; light gray: lights off (subjective day); dark gray: lights off (night). JTK_cycle P values were calculated by 24-h day for each genotype; significance is indicated below lighting bars. P_JTK < 0.001 = ***; P_JTK > 0.05 = not significant (n.s.). (B) 12-h food intake in microliters during the day (ZT/CT 0 to 11) vs. night (ZT/CT 12 to 23) for flies shown in A. Day vs. night feeding was compared within each genotype, and 24-h food intake was compared between genotypes by Student’s t test. *P < 0.05, **P < 0.01, and ***P < 0.001. There were no significant differences in 24-h food intake between genotypes. (C) Average consumption per fly per hour for as in A for DILP2,3^−/+ control (black, n = 67) and DILP2,3^−/− insertion mutant (purple, n = 66). (D) 12-h food intake for flies shown in C, plotted as in B, with statistical comparisons for 24-h consumption above.

Discussion

The Drosophila PI was previously described as a protohypothalamic region that serves as a circadian output hub (23, 24, 26, 32); while inputs from DN1 clock neurons were shown, relevance of other clock neurons or the signaling molecules involved in transducing time-of-day cues from the clock to the PI were not described. We demonstrate here that time-of-day information arrives at the PI from multiple clock neuron populations through cholinergic and glutamatergic signaling. Further, we identify an output role for the IPCs of the PI in regulating circadian feeding, which depends on inputs from morning-active, but not evening-active, clock neurons.

The response of PI neurons to clock neuron stimulation is time-of-day dependent. DH44+ neurons are inhibited—i.e., exhibit a reduction in intracellular calcium—by both DN1 and LNd neuron stimulation in the morning, with little to no effect of stimulation of clock neurons in the evening (Fig. 5). Given that LNds drive the evening peak of locomotor activity (8), it is surprising that their stimulation only inhibits DH44+ neurons in the morning; however, inhibition in the morning is consistent with the normal pattern of activity of DH44+ neurons. DH44+ neurons are evening active in both LD and DD conditions (25, 27), and while they regulate both the morning and evening peaks of locomotor activity in LD, their contribution to the evening peak is larger (23, 26). It is likely that DH44+ neurons, which are implicated in several processes (23, 26, 28, 30), receive nonclock inputs that drive locomotor activity, and they require silencing by the clock to inhibit locomotion outside the morning and evening activity peaks. Thus, silencing by DN1s may serve to delineate the morning peak of activity; on the other hand, early day silencing of DH44+ neurons by LNds may be one mechanism by which LNds drive evening activity.

Fig. 5.

Model for time-of-day–dependent modulation of PI neurons by clock populations. (A) In the morning, LNd and DN1 clock provide inhibitory signals to DH44+ neurons and excitatory signals to some IPCs. This may contribute to the increased activity of IPCs and decreased activity of DH44+ neurons at this time. (B) At night, LNd neurons continue to provide excitatory signals to IPCs, but no longer inhibit DH44+ neurons. The reduction in LNd-mediated inhibition may contribute to the night time activity of DH44+ neurons. DN1 neurons, which are active in the morning, provide no modulatory signal to either PI population at night. Increased activity of a neuronal population is indicated by the surrounding glow.

The response of IPCs to DN1 stimulation is time-of-day dependent (Fig. 5). The response to LNd stimulation also shows an apparent difference between morning and evening, but, in fact, some cells respond at both times. The IPC population is larger than the DH44+ population and is known to be heterogeneous in terms of basal activity and signaling molecule expression (24, 46, 47). The functional relevance of LNd signaling to IPCs is unclear, but could be linked to the role of these cells in promoting arousal (48). On the other hand, the DN1 inputs likely contribute to the circadian rhythm of feeding. As noted above, this is consistent with the higher firing of IPCs in the morning and the morning peak of the feeding rhythm in w¹¹¹⁸ flies (24, 49), as well as with our finding here that DILPs are required for rhythmic feeding. In further support of this idea, we report that feeding rhythms depend upon the activity of morning-active LNvs, which signal through DN1s, but not evening-active LNds. Our finding of loss of feeding rhythms on yeast/sucrose diet with genetic ablation of two insulin-like peptides differs from the findings of Dreyer et al. (32) who observed maintenance of rhythmic feeding on sucrose-only food upon adult silencing of IPCs. We speculate that developmental changes caused by genetic ablation of insulin(s) result in disorganized feeding patterns and/or that rhythmic protein, but not sugar, feeding is regulated by brain insulins.

The finding that nAChR and mGluR signaling contribute to inhibition in one case (input from DN1 and LNds on to DH44 neurons) and activation in another (input from DN1 and LNds on to IPCs) raises the question of what determines the nature of the effect. The most straightforward explanation would be a difference in the receptor subtypes expressed in the two cell groups. Alternatively, other signaling components specific to each cell type could account for the differential response. Regardless, it is clear that the actions of known neurotransmitters can be regulated by the clock in different ways to effect rhythmic output. As noted above, acetylcholine is secreted by the LNds and likely also DN1s, but to date, the only known source of glutamate in the clock network are the DN1s (16, 20). This suggests that glutamatergic input from the DN1s can modulate the effect of LNds on PI neurons.

Despite the effect of clock cell-secreted neuropeptides on PI cells and the role of fast neurotransmitters in signaling from clock cells to the PI, knockdown of neuropeptide/neurotransmitter receptors in the PI has little effect on behavioral rhythms (SI Appendix, Fig. S2). There are many possible explanations for lack of a phenotype—low mRNA levels remaining after knockdown (we verified knockdown) are sufficient for function; receptor subtypes act redundantly in the regulation of rhythms; knockdown in the PI alone is insufficient to yield a phenotype; or these molecules affect physiological rhythms other than those of rest:activity. While DH31 and NPF have been implicated in the control of rest:activity rhythms, loss of acetylcholine and glutamate have not been linked to behavioral rhythms (14, 17, 21, 50). Although knockdown of receptors for these molecules in the PI did not yield a phenotype, we saw reduced evening activity with DH44-driven knockdown of PDFR. This is in contrast to the effect of loss-of-function pdf mutants in which evening activity remains high, but is shifted ∼1 h early (42). It is possible that DH44+ cell-mediated reduced evening activity is compensated in pdfr mutants. PDF is secreted by LNvs, which are not known to synapse onto the PI, but do project to the dorsal brain, suggesting diffusion of PDF across a small region. Limited diffusion of PDF is supported by a previous study showing that overexpression of PDF in cells that project to the dorsal brain produces behavioral phenotypes (51).

These findings elucidate some of the complexity underlying circadian control in neural circuits. Through the use of neuropeptides and fast neurotransmitters coupled with time-of-day–specific actions on downstream neurons, clock cells are able to drive multiple outputs at varying timescales. Behavioral outputs require that rhythmic patterns of activity be propagated through many parts of the brain, and this is likely the case. We previously showed that hugin+ neurons, which are downstream of the PI, show rhythmic peptide release, indicating that cycling is propagated to second order output neurons (26). In addition, clock neurons are linked to rhythmic neural activity in locomotor centers of the fly brain (52). While a rhythm in spiking activity may be sufficient to transmit some outputs, we know that other variations in electrical activity (e.g., temporal coding from DN1 neurons) may also contribute to circadian behaviors (53). As in mammals, different output rhythms also map to different output circuits in flies (54). Thus, Drosophila take advantage not only of diverse neural output mechanisms to regulate aspects of circadian behavior, but also use distinct neuropeptidergic output cells to regulate locomotor and feeding rhythms. DH44+ neurons are important for locomotor, but not feeding, rhythms, while the opposite is true for IPCs. At the same time, proximity of output neurons controlling locomotor activity with those that control feeding likely allows for integration of multiple signals to coordinate behavior, thus contributing to organismal fitness.

Materials and Methods

Detailed materials and methods are provided in SI Appendix, SI Materials and Methods.

Fly Stocks.

UAS/Gal4 lines and mutants used for behavior and immunohistochemistry are described in SI Appendix. See SI Appendix, Table S1 for a list of the complete genotype for the animals used in each experiment.

P2X2 Activation and GCaMP Imaging.

GCaMP6m imaging in response to P2X2 activation in acutely dissected brains was performed as previously described (24, 26). Fly entrainment procedures and detailed methods are described in SI Appendix.

Drosophila Activity Monitoring Assay.

Rest:activity rhythm assays were performed with the Drosophila Activity Monitoring System (Trikinetics) as described previously (55). Fly entrainment procedures and detailed methods are described in SI Appendix.

Activity Recording CAFE Assay.

Single-fly circadian feeding was assessed using the automated recording CAFE assay (43). Detailed methods are described in SI Appendix.

Immunohistochemistry, GRASP, and Confocal Microscopy.

Fly lines, entrainment conditions, antibodies, and detailed methods are described in SI Appendix.

Statistical Analysis.

Sample sizes and statistical details of each experiment can be found in the figure legends. All statistical tests were performed in OriginPro 2020. Additional detailed statistical methods are provided in SI Appendix.

Data Availability

All relevant data are within the manuscript and its SI Appendix.