Differentially Timed Extracellular Signals Synchronize Pacemaker Neuron Clocks

The four PDF-expressing LN_vs in each larval brain lobe are precursors of adult s-LN_vs. Molecular clock oscillations in larval LN_vs are normally tightly synchronized, oscillating in phase with each other so that TIM and PDP1 clock proteins are detectable in all four LN_vs at CT21 and undetectable in all four LN_vs 6 hours later at CT3 (Figure 1A) (CT, Circadian time, hours in constant darkness).

Because PER protein rhythms in adult s-LN_vs become desynchronized in Pdf¹ null mutants in DD [17], we first tested whether PDF is required to synchronize larval LN_v molecular clocks. We measured TIM protein levels instead of PER with the rationale that TIM's shorter half-life [24],[25] would allow us to detect desymchrony earlier in DD.

To visualize LN_vs in Pdf¹ mutants, we used the Gal4/UAS system [26] to express GFP in LN_vs using the Pdf-Gal4 driver. We measured TIM levels in LN_vs isolated at CT9, CT15, and CT21 on the second day in DD and at CT3 on day 3. TIM continues to oscillate in Pdf¹ mutants, indicating that the molecular clocks in their LN_vs are functional (Figure 1A–B). However, the amplitude of TIM rhythms in Pdf¹ mutants was reduced compared to controls (Figure 1B), as expected from the reduced amplitude tim RNA oscillations in Pdf¹ adult flies [27]. Closer inspection identified a mixture of TIM-positive and TIM-negative LN_vs in a single brain lobe at CT15, 21, and 3 in Pdf¹ mutants (Figures 1D and S1A; see Materials and Methods), which we term desynchronized. Elevated desynchrony likely accounts for the significantly lower average TIM levels in Pdf¹ LN_vs at CT15 and 21 than in controls (Figure 1B), in agreement with previous reports [17],[27].

We also quantified the variability within individual LN_v clusters by calculating the standard deviation in TIM levels across a single cluster. Figure 1F shows the distribution of standard deviations in TIM levels for each control or Pdf¹ LN_v cluster at CT3 and CT9. We chose CT3 because desynchronized LN_v clusters were only rarely found at this timepoint in control larvae. In contrast, TIM was detected in one, two, or three of the four LN_vs in 50% of Pdf¹ LN_v clusters at CT3 (n = 20; Figures 1D and S1A and Table S1). In subsequent experiments we used the presence of TIM in a subset of LN_vs at CT3 to indicate that an LN_v cluster had lost its normal coherent phase relationship and had become desynchronized, even if we did not observe desynchrony at other time points.

Our data show significantly more variability in TIM levels within an LN_v cluster in Pdf¹ mutants than in control larvae at CT3 (Figure 1F), reflecting desynchronized molecular clocks in Pdf¹ LN_vs. No significant increase in standard deviation was observed in Pdf¹ mutants compared to controls at CT9 (Figure 1F). Indeed, the low TIM levels at CT9 indicate that Pdf¹ LN_v molecular clocks still oscillate as shown previously [17],[27].

Because PDF signals via PdfR, we next tested whether the synchrony of larval LN_v molecular clocks is also altered in Pdfr mutants. Although overall TIM oscillations were similar between control and Pdfr^han5304 (Pdfr^han) hypomorphs, we observed higher TIM levels at CT3 in Pdfr^han than in control larvae (Figure 1A–B). As with Pdf¹ null mutants, this is because TIM was detected in 1–3 of the four LN_vs in 48% of Pdfr^han mutant LN_v clusters at CT3 (Figures 1D and S1A). We found similar results for PDP1 (Figures 1A,C,E and S1B). In contrast, TIM or PDP1 expression was detected in <5% of control LN_vs at CT3 (n = 21; Figure 1D–E and Table S1). The standard deviations in TIM and PDP1 levels are significantly elevated at CT3 in Pdfr^han mutants compared to controls (Figure 1F–G). Thus PdfR, like PDF, is required for LN_vs to stay synchronized.

In contrast to Pdf¹ LN_vs, Pdfr^han mutants did not show many desynchronized LN_v clusters at CT15 or CT21, and there was no corresponding reduction in the amplitude of TIM oscillations in Pdfr^han mutants compared to control LN_vs. This could be because Pdfr^han is a hypomorph rather than a null allele and/or because type II GPCRs tend to be promiscuous, so receptors other than PdfR may also respond to PDF [28].

We also tested whether LN_v molecular clocks required PDF to maintain synchrony under LD cycles. We measured TIM and PDP1 levels in control larvae and in Pdf¹ and Pdfr^han mutants at ZT3, but detected no TIM or PDP1 expression in LN_vs (Figure S1C). Thus, light overrides desynchrony in Pdf¹ and Pdfr^han5304 mutants, with PDF signaling required for synchronous LN_v clock oscillations only in DD.

Because adult and larval LN_vs express Pdfr ([29],[30] and Figure S2C), the simplest model to explain how PDF promotes LN_v synchrony would be that the four larval LN_vs signal to synchronize each other via PDF and PdfR. However, Pdfr is also expressed in many non-LN_v adult clock neurons [29] and in larval DN₁s (Figure S2A,B). Thus PDF signaling to non-LN_vs could also be required for LN_v synchronization. We therefore used a Pdfr^RNAi transgene [31] to reduce Pdfr levels in subsets of clock neurons to determine where PDF signaling is required for LN_v synchronization. Expressing Pdfr^RNAi in LN_vs significantly reduced the cAMP response of LN_vs to PDF, indicating that Pdfr^RNAi likely reduces Pdfr expression (Figure S2C). UAS-Dicer-2 (UAS-Dcr-2) was co-expressed to increase RNAi efficacy in this and in all subsequent RNAi experiments unless otherwise stated, but is omitted from written genotypes in the text for simplicity.

We first targeted Pdfr^RNAi to LN_vs using Pdf-Gal4 (denoted as Pdf>). At CT3 on day 3 of DD, TIM staining revealed that 44% of Pdf>Pdfr^RNAi larvae had desynchronized LN_vs, whereas >93% of control LN_vs were synchronized (Figures 2A and S3A and Table S1). The standard deviation in TIM levels was also significantly increased in Pdf>Pdfr^RNAi larvae compared to controls at CT3 (Figure 2B). Similar results were observed for PDP1 at CT3 (Figure S3B and Table S1).

Next, Pdfr expression was reduced in all non-LN_v clock neurons using the tim-Gal4; Pdf-Gal80 driver combination (tim; Pdf-Gal80>). We found that 44% of LN_vs were desynchronized in tim; Pdf-Gal80>Pdfr^RNAi larvae (Figure S3A and Table S1). This probably underestimates the level of defective TIM oscillations, as 16% of tim; Pdf-Gal80>Pdfr^RNAi LN_v clusters showed four LN_vs expressing TIM at CT3, compared to only 6% of controls (Table S1). There is a corresponding increase in the standard deviation in TIM levels in tim; Pdf-Gal80>Pdfr^RNAi LN_v clusters compared to control LN_vs (Figure 2B). Similar results were observed for PDP1 (Figure S3A–B). These data indicate that LN_v synchrony depends on PdfR activity in both LN_v and non-LN_v clock neurons.

The non-LN_v clock neurons releasing the synchronizing signal could be the larval DN₁s, the DN₂s, or the fifth LN_v. DN₁s are the best candidates, as they project to LN_v axonal termini and modulate LN_v outputs by releasing glutamate to generate circadian rhythms in larval light avoidance [9]. Larval DN₁s also respond directly to PDF (Figure S2A).

We therefore used cry-Gal4 and Pdf-Gal80 (DN₁>) to target transgene expression exclusively to DN₁s [9]. We first tested whether DN₁ ablation affected LN_v synchrony by expressing Diptheria Toxin in DN₁s (DN₁>Dti). We found that TIM rhythms persisted in LN_vs after DN₁ ablation (Figure S3D), indicating that LN_vs do not require DN₁s for oscillations per se. However, TIM levels at CT3 on both days 2 and 3 in DD were elevated in DN₁-ablated larvae (Figure S3D). Examining TIM staining in DN₁-ablated brains in DD revealed that 50% of LN_v clusters were desynchronized at CT3 on days 2 and 3 in DD, a significant increase compared to controls (Figures 2C,D and S3A and Table S1). We observed similar increases in desynchrony of PDP1 expression when DN₁s were ablated (Figures S3A,C,E and S6C–D and Table S1) with significantly higher levels at CT3 on day 3. We did not observe desynchrony in LD cycles (Figure 2D) or at CT9, just like Pdf¹ and Pdfr^han mutants. We conclude that PDF signaling (Figure 1) and DN₁s (Figure 2) normally maintain larval LN_v molecular clock synchrony in constant darkness.

To test this model further, we sought to identify the DN₁ signal and the relevant receptor in LN_vs. Because larval DN₁s are glutamatergic [32], we tested whether reducing DN₁ glutamate levels alters LN_v molecular clock synchrony. Glutamate decarboxylase 1 (Gad1) was mis-expressed in DN₁s, to convert glutamate into GABA [9],[33], which cannot be released as DN₁s do not produce the vesicular GABA transporter. Thus misexpression of Gad1 reduces presynaptic glutamate. This manipulation does not affect DN₁ viability, and their molecular clocks still oscillate [9].

We found that overall TIM oscillations were relatively normal in DN₁>Gad1 LN_vs (Figure S4A). However, TIM levels were significantly elevated at CT3 in DN₁>Gad1 larvae (Figure S4A). This is because DN₁>Gad1 significantly increased LN_v desynchrony, determined by comparing the standard deviation in TIM and PDP1 expression with control LN_vs (Figures 3A,C,D and S4C and Table S1). Therefore, we conclude that DN₁s release glutamate to synchronize LN_v molecular clocks.

Larval LN_vs express two glutamate receptors: a metabotropic glutamate receptor (mGluRA, [32]) and a glutamate-gated Chloride channel (GluCl, [9]). To determine whether one of these receptors transduces the glutamate signal to synchronize LN_vs, we used RNAi transgenes previously shown to reduce expression of mGluRA or GluCl[9],[32]. We found that reducing GluCl expression in LN_vs had no effect on TIM and PDP1 oscillations or LN_v synchrony (Figures 3B–D and S4B–C). In contrast, expressing mGluRA^RNAi in LN_vs produced similar molecular phenotypes to DN₁ ablation, with elevated TIM levels at CT3 and 75% of LN_vs desynchronized (Figure 3B–D and Table S1).

As an independent way to manipulate mGluRA expression, we measured TIM levels at CT3 in LN_vs of mGluRA^112b null mutant larvae (Figures 3B–C and S4D and Table S1). We found desynchronized LN_vs in homozygous mGluRA^112b mutant larvae but not in heterozygous controls. We saw similar levels of desynchronization when measuring PDP1 levels in Pdf>mGluRA^RNAi and mGluRA^112b mutant LN_vs (Figures 3C and S4C–D). Taking these data together with our manipulations of DN₁ glutamate levels, we conclude that glutamate released by DN₁s helps synchronize LN_v oscillations via mGluRA.

We previously showed that LN_vs require GluCl rather than mGluRA for circadian rhythms in the rapid light avoidance of larvae [9]. Thus, a single neurotransmitter, glutamate, released by DN₁s has two distinct functions depending on the receptor in LN_vs that perceives the signal. Presumably the rapid action of the ionotropic receptor on LN_v excitability [9] is best suited to regulate light avoidance behavior, whereas mGluRA acts on a slower timescale to regulate clock oscillations.

LN_vs require two different signals to maintain synchrony, as reducing expression of either Pdfr or mGluRA desynchronized LN_v molecular clocks. However, we only observed an increase in desynchronized LN_v clusters at CT3 in Pdf>Pdfr^RNAi or Pdf>mGluRA^RNAi larval brains compared to controls, with most LN_v clusters remaining synchronized at CT21. This suggested that the second signal—glutamate in Pdf>Pdfr^RNAi and PDF in Pdf>mGluRA^RNAi larvae—maintains some degree of LN_v synchrony and we hypothesized that simultaneously reducing Pdfr and mGluRA expression would more strongly affect LN_v clock synchrony.

We measured TIM and PDP1 oscillations in LN_vs expressing transgenes targeting both Pdfr and mGluRA expression (Pdf>Pdfr^RNAi+mGluRA^RNAi). We found that 88% of LN_v clusters showed desynchrony in TIM protein levels at CT3 (Figures 4A–B and S5B and Table S1) and 75% for PDP1 (Figure S5A,C and Table S1). Pdf>Pdfr^RNAi+mGluRA^RNAi larvae also had significantly more desynchronized LN_v clusters at CT21 and CT3 than control larvae (Figure S5A). Thus simultaneously reducing expression of both receptors dramatically increased the percentage of desynchronized LN_vs compared to reducing Pdfr or mGluRA expression alone, indicating that PDF and glutamate signals work together to promote synchrony.

Although we observed a few individual LN_vs with high TIM levels in Pdf>Pdfr^RNAi+mGluRA^RNAi larvae, overall TIM oscillations were almost completely lost (Figure 4C). This contrasts with the robust TIM oscillations of Pdf>Pdfr^RNAi and Pdf>mGluRA^RNAi single knock-down larvae (Figure S5E). High-amplitude TIM protein oscillations in LN_vs thus depend on external signals, including PDF and glutamate, and are not fully cell-autonomous. Although PDP1 showed elevated desynchrony in Pdf>Pdfr^RNAi+mGluRA^RNAi LN_vs (Figure S5A,C), overall PDP1 oscillations were relatively unaffected (Figure S5D). Thus Pdf>Pdfr^RNAi+mGluRA^RNAi LN_vs are still partly functional. These data suggest that TIM is a more direct target than PDP1 in LN_vs for the signaling pathways that transduce glutamate and PDF signals.

Do the reduced amplitude TIM rhythms in Pdf>Pdfr^RNAi+mGluRA^RNAi double mutant larvae affect behavioral rhythms? We had previously found that light avoidance rhythms require glutamate release from DN₁s and transduction via GluCl in LN_vs [9]. Because TIM oscillations in LN_vs remained intact in Pdf>GluCl^RNAi larval brains (Figure S4B), we concluded that glutamate received by GluCl modulates LN_v outputs rather than LN_v molecular clocks [9]. Knocking down either mGluRA or Pdfr individually in LN_vs does not block TIM or PDP1 protein oscillations (Figure S5E–F) and larval light avoidance is still rhythmic, with peak levels at dawn (Figure 4D and [9]). However, we found that larvae with mGluRA and Pdfr expression simultaneously reduced in LN_vs lose light avoidance rhythms (Figure 4D). This result suggests that TIM oscillations in LN_vs are essential for light avoidance rhythms and that PDP1 rhythms alone cannot support larval rhythms. Overall, these data indicate the importance of extracellular signals for LN_vs to oscillate normally and promote rhythmic behavior.

Adult s-LN_vs are most excitable at dawn [34],[35] and drive the morning peak of locomotor activity [6],[7]. We previously showed that the same is likely true for the larval LN_vs that control the dawn peak in light avoidance, whereas larval DN₁s most likely signal at dusk [9]. To test whether DN₁s signal at dawn or dusk to promote LN_v synchrony, we used a temperature-sensitive Shibire transgene (UAS-Shi^ts[36]) to temporally block synaptic transmission.

Shi^ts was expressed specifically in DN₁s (DN₁>Shi^ts), and larvae were maintained at the permissive temperature of 25°C for 4 days in LD and 1 day in DD. On the second day in DD, the temperature was elevated to the nonpermissive temperature of 31°C for 6 hours from either CT9 to CT15 (“CT12 shift”) or CT21 to CT3 (“CT24 shift”) to block DN₁ signaling around dusk or dawn, respectively (Figure 5A). Larval brains were dissected at CT3 on day 3 of DD (i.e., 12 hours after the end of a CT12 temperature shift or immediately after the end of a CT24 temperature shift). We found that 57% of LN_v clusters showed desynchronized TIM levels when DN₁ synaptic transmission was blocked at dusk (DN₁>Shi^ts, 31°C at CT12) compared to 7% of control LN_vs (UAS-Shi^ts/+; Figure 5A–C and Table S1). Similarly, 36% of LN_vs in DN₁>Shi^ts larvae shifted to 31°C at CT12 had desynchronized PDP1 levels compared to 0% of control LN_vs (Figure S6A–B and Table S1). In contrast, blocking synaptic transmission from DN₁s around dawn had no effect on LN_v synchrony (DN₁>Shi^ts, 0% desynchronized for TIM or PDP1 with a CT24 heat pulse; Figures 5A–C and S6A–B and Table S1). We therefore conclude that DN₁ signaling around dusk is required to synchronize LN_vs.

To further test the idea that PDF and glutamate promote synchrony at different times of day, we took advantage of the synchronizing effect of LD cycles on Pdf¹ and DN₁>Dti LN_vs (Figures 2D, S1C, and S3B and Table S1). Based on the likely timing of LN_v and DN₁ signals, wild-type LN_v clocks should have received the PDF signal at subjective dawn by CT3 on day 1 in DD, but not yet received the glutamatergic signal at subjective dusk. Thus we predicted that Pdf¹ mutants would show desynchrony at this time point, whereas DN₁>Dti LN_vs, which still receive the PDF signal, would not.

We measured TIM and PDP1 levels in LN_vs at CT3 on the first day of DD and found higher TIM and PDP1 levels and an increase in the variability of clock protein levels between LN_vs in the same cluster in Pdf¹ mutants, indicating that LN_vs are already desynchronized just 3 hours into DD (Figures 5D–E and S6C–D). In contrast, the LN_v clocks in larvae with DN₁s ablated remained synchronized at CT3 on the first day in DD, and desynchrony was first detected on day 2 (Figures 5D–E and S6C–D).

We interpret these data to mean that desynchrony in DN₁-ablated larvae requires larvae to traverse subjective dusk when the DN₁ signal is released. Because desynchrony appears on different days in Pdf¹ and DN₁-ablated larvae, this supports the model where LN_v synchrony depends on PDF received at dawn and glutamate received at dusk. This is consistent with the previously reported timing of LN_v excitability [34],[35] and of the larval LN_v and DN₁ signals that regulate light avoidance [9].

PdfR and mGluRA are both G-protein coupled receptors. PdfR signals via Gαs [18],[19],[21],[37] and mGluRA can also alter cAMP levels [38]. Because cAMP is a clock component in mammals [23] and likely also in flies [39],[40], regulation of LN_v cAMP levels by extracellular signals could maintain LN_v synchrony and promote robust TIM oscillations.

We used the FRET-based Epac1-camps sensor [19] to measure basal cAMP levels on day 2 in DD. We first assayed control LN_vs, focusing on their axonal termini near DN₁ projections [9]. We found that cAMP levels, measured by the ratio of CFP/YFP, were highest at CT24, indicating that cAMP levels normally oscillate in LN_v projections (Figure 6A). Strikingly, cAMP (CFP/YFP) oscillations were lost in the projections of both Pdf>Pdfr^RNAi and Pdf>mGluRA^RNAi larval LN_vs (Figure 6A).

We noticed that Pdf>mGluRA^RNAi LN_v cAMP levels were significantly higher than controls at dusk (CT12), when DN₁s signal for synchrony. This is consistent with data showing that mGluRA reduces cAMP levels by signaling via Gαi [38], thereby opposing PdfR activity [18],[37]. To test this idea, we measured the responsiveness of LN_vs to PDF with reduced mGluRA activity. We first generated a PDF response curve to determine the minimal PDF concentration that elicits an Epac1-camps response (Figure S7A–C). We then tested whether expressing mGluRA^RNAi in Pdf>Epac1-camps LN_vs altered this response (Figure 6B–C) using GluCl^RNAi as a control. We found that mGluRA^RNAi, but not GluCl^RNAi, significantly increased LN_v responsiveness to PDF (Figure 6C). Therefore, we propose that mGluRA acts in an opposite manner to PdfR and reduces intracellular cAMP.

To test if cAMP links to synchronized clock protein oscillations, we built on the recent identification of Adenylate cyclase 3 (AC3) as the specific Adenylate cyclase downstream of PdfR in LN_vs [21]. We tested whether AC3 is required for LN_v synchronization by reducing expression of AC3 using two independent RNAi lines (Pdf>AC3^{TRiP RNAi} and Pdf>AC3^{Vienna RNAi}) that reduce PDF responses in LN_vs [21]. We found that expressing each RNAi line in LN_vs desynchronized TIM expression in 35%–40% of LN_v clusters and PDP1 expression in 28%–35% of LN_v clusters (Figure S8A–B and Table S1). Reducing AC3 expression in LN_vs also significantly increased desynchrony measured by standard deviation in TIM and PDP1 expression (Figure S8C–D).

We conclude that PdfR and mGluRA regulate LN_v cAMP levels at different times of day, presumably by regulating AC3 activity. This leads to a model in which LN_v cAMP rhythms are generated by extracellular signals, with PDF/PdfR increasing cAMP via AC3 around dawn, whereas glutamate inhibits the response of LN_vs to PDF via mGluRA by inhibiting AC3 around dusk (Figure 6D). cAMP oscillations then feed into the molecular clock, affecting TIM oscillations through an unknown mechanism, which will be a topic of future research.

We next tested whether our findings from larvae held true for the more complicated adult circadian system. Because we observed the most dramatic effects on larval LN_v synchrony by simultaneously reducing Pdfr and mGluRA in LN_vs, we measured the synchrony of s-LN_v molecular clocks in Pdf>Pdfr^RNAi+mGluRA^RNAi adult flies. We found that many more Pdf>Pdfr^RNAi+mGluRA^RNAi s-LN_v clusters were desynchronized than control s-LN_vs (Figure 7A–C and Table S1), with extensive desynchrony detected at CT15 and CT21 on day 2 in DD and CT3 on day 3. TIM oscillations within Pdf>Pdfr^RNAi+mGluRA^RNAi s-LN_vs also displayed a reduced amplitude compared to control s-LN_vs, although the effect was less pronounced than in larvae (Figure 7D). We also observed significant desynchrony at CT3 when either Pdfr or mGluRA expression was reduced in LN_vs (Figure S9A). We conclude that PDF and glutamate contribute to the robustness and synchrony of LN_v oscillations in adult flies as well as in larvae.

We next tested whether Pdf>Pdfr^RNAi+mGluRA^RNAi flies displayed behavioral defects. We compared the locomotor activity of Pdf>Pdfr^RNAi+mGluRA^RNAi flies to parental flies and to Pdf>GluCl^RNAi flies to control for nonspecific effects of RNAi in LN_vs, as GluCl^RNAi does not affect larval LN_v synchrony (Figure 3). Because Pdf>Pdfr^RNAi+mGluRA^RNAi flies have ∼24 h locomotor activity rhythms in DD, we conclude that s-LN_v desynchrony does not affect period length (Figure 8A and Table S2). However, we noticed that the activity of Pdf>Pdfr^RNAi+mGluRA^RNAi flies was much less consolidated than control or Pdf>GluCl^RNAi flies, with bursts of activity visible in the subjective night when control flies are inactive (Figure 8A).

We calculated the average locomotor activity on the first 5 days in DD for each genotype. Pdf>Pdfr^RNAi, Pdf>mGluRA^RNAi, and Pdf>Pdfr^RNAi+mGluRA^RNAi flies displayed elevated levels of activity towards the end of subjective day and the beginning of subjective night (∼CT6–18) compared to control and Pdf>GluCl^RNAi flies (Figure 8B). Thus altering PDF and/or glutamate inputs to LN_vs increases nighttime activity.

To further quantify these differences in nighttime activity, we used standard measures of sleep. We found decreased overall sleep levels when mGluRA expression was reduced either alone or with Pdfr (Figure S9B). In contrast, reducing Pdfr expression alone had no significant effect on overall levels of sleep (Figure S9B). Thus, we conclude that glutamate signals to LN_vs regulate sleep levels, whereas PDF signals between LN_vs do not regulate sleep.

Next, we quantified the transition between wakefulness and sleep in the evening by measuring how quickly flies fell asleep after CT12 (sleep latency). To ensure that any effects on the timing of sleep onset did not result from subtle period length differences between genotypes (Table S2), we measured sleep latency only on day 1 in DD when the phase of locomotor activity between genotypes is minimally affected by small period differences. We found that Pdf>Pdfr^RNAi+mGluRA^RNAi flies showed a significant increase in sleep latency compared to all other genotypes (Figure 8C). Their average sleep latency of 213 min compared to 113 min for UAS-Pdfr^RNAi+UAS-mGluRA^RNAi/+ control flies exceeds the 30 min period length difference between these genotypes (Figure 8C and Table S2). We observed no significant effects when either mGluRA or Pdfr expression was reduced singly (Figure 8C).

Thus, we conclude that blocking PDF and glutamate inputs to LN_vs increases evening activity and delays sleep onset timing. We did not observe a significant effect of reducing mGluRA or Pdfr expression on sleep latency under LD cycles (Figure S9C), consistent with LD cycles synchronizing larval LN_v clock oscillations (Figures 2D, 5D–E, S1C, and S3C). Although increased LN_v desynchrony may not cause the sleep latency defects observed, it is clear that normal Pdfr and mGluRA activity in LN_vs is required for normal sleep in DD. However, it is possible that desynchrony and sleep latency defects are separate phenotypes resulting from abrogated intercellular communication between clock neurons.

Feedback is an essential component in the molecular clocks that drive circadian behavior in animals [22]. Here we demonstrate the importance of feedback across the circadian neural network to synchronize individual clock neurons. We showed that larval LN_vs require two signals that cooperate to synchronize their clocks: PDF released at dawn from LN_vs themselves and glutamate released by DN₁s at dusk. The PDF signal received by PdfR in DN₁s presumably also sets the phase of the DN₁ clock (Figure S2D) [8] to correctly time glutamate release that is then perceived by mGluRA in LN_vs. Thus a feedback loop seems to exist at the circuit level, maintaining synchronized LN_v clocks in DD.

Our experiments also reveal that synchronization of larval pacemaker neurons is a very active process, as LN_v clocks were desynchronized 3 hours into the first subjective morning if they miss the dawn PDF signal. Consistent with the dual-synchronizer model, we see increased desynchrony when mGluRA and Pdfr expression is simultaneously reduced in LN_vs (Figures 4B and S5A–C and Table S1).

A DN₁ glutamate signal released around dusk is required for circadian rhythms of light avoidance when received by the ionotropic glutamate receptor GluCl in LN_vs [9]. We now show that DN₁ glutamate also promotes LN_v synchrony when received by the metabotropic glutamate receptor mGluRA in LN_vs. Thus a single neurotransmitter plays two distinct roles in the Drosophila circadian circuit depending on the receptor that receives the signal in LN_vs: a rapid behavioral response to light mediated via GluCl and longer-term regulation of the 24 hour molecular clock via mGluRA.

Although mGluRA is not required for light avoidance [9], we found that larvae with reduced expression of both Pdfr and mGluRA lose larval light avoidance rhythms. This is consistent with the loss of strong TIM protein oscillations in the LN_vs of Pdf>Pdfr^RNAi+mGluRA^RNAi larvae. These defects in the LN_v molecular clock probably alter the timing of signals from LN_vs and/or the phases of other clock neurons within the circuit. This contrasts with the role of GluCl, where glutamate received by GluCl directly regulates light avoidance by inhibiting the response of LN_vs to ACh, independent of the LN_v molecular clock [9].

Synchronization of adult s-LN_vs also depends on signaling via PdfR and mGluRA as >50% of s-LN_v clusters were desynchronized at three of the four timepoints measured when expression of both Pdfr and mGluRA was reduced in LN_vs. However, TIM oscillations in adult s-LN_vs were not as severely impaired as in larval LN_vs. The increased complexity of the adult clock neural circuit probably adds signals from neurons not present in larvae that promote synchronized and robust clock protein oscillations in adult s-LN_vs.

Because the molecular clock in adult Pdf>mGluRA^RNAi+Pdfr^RNAi flies still oscillates, it is not surprising that locomotor activity is also still largely rhythmic. However, the desynchrony and reduced amplitude of TIM oscillations in Pdf>mGluRA^RNAi+Pdfr^RNAi LN_vs correlates with increased nighttime activity and sleep latency. Desynchrony and increased activity could be independent consequences of reduced glutamate and PDF receptivity in LN_vs. An alternative possibility is that because the molecular clock regulates daily firing rhythms of clock neurons [34],, individual LN_vs remain active at the wrong time of day in a desynchronized LN_v cluster, preventing sleep. Indeed, if LN_vs are electrically coupled like SCN neurons [43], then firing of a single LN_v may cause the remaining LN_vs in that cluster to fire earlier and/or later than programmed by their molecular clock. In addition, mistimed LN_v signals in Pdf>Pdfr^RNAi+mGluRA^RNAi flies will affect the phases of other clock neurons in the circuit, which could also disrupt sleep timing.

The loss of strong TIM protein oscillations in Pdf>Pdfr^RNAi+mGluRA^RNAi larval LN_vs is surprising, as molecular clock oscillations in pacemaker neurons are often regarded as cell-autonomous. Our data extend conclusions from the SCN showing that the VIP receptor, VPAC2R, is required for synchronized molecular clocks [14]. By removing a second receptor simultaneously and by restricting our analyses to a defined subset of pacemaker neurons, we demonstrate that specifically blocking input signals to master pacemaker neurons has a dramatic effect on core clock protein oscillations.

We observed a much stronger effect on TIM than on PDP1 oscillations in Pdf>mGluRA^RNAi+Pdfr^RNAi larval LN_vs. It may be that TIM oscillations are relatively easily modified, allowing information from outside the cell to be integrated into the molecular clock, whereas a more robust PDP1 oscillation prevents LN_vs overreacting to external stimuli. It is well-documented that TIM can be regulated at the posttranslational level in addition to transcriptional control [22], whereas PDP1 protein levels closely follow Pdp1 RNA levels [44]. We propose that external signals mediated via PdfR and mGluRA mainly regulate the clock posttranslationally, and this is supported by recent findings [45]–[47]. Testing this idea will require developing a combination of transcriptional and translational reporter genes.

VIP and VPAC2R synchronize the mammalian SCN in a cAMP/Ca²⁺-dependent manner [14],[16],[23]. VPAC2R is expressed more broadly than VIP, and some SCN neurons express both VIP and VPAC2R [48]. This is highly reminiscent of Drosophila, where Pdfr is found in both PDF+ and PDF– clock neurons [29]. Because VIP/VPAC2R and PDF/PdfR are functionally similar and because both mediate synchronization of pacemaker neurons, discoveries about the roles of PDF/PdfR in Drosophila should be relevant to understand how synchrony is maintained across circadian neural circuits in general. In both flies and mammals, a reciprocal relationship between synchrony and clock protein amplitude seems to allow pacemaker neurons to be more precise and robust timekeepers than individual neurons.

Our data reveal a remarkable degree of conservation of clock circuit properties between mammals and Drosophila, echoing the conserved molecular basis of the circadian clock. The mechanisms promoting LN_v synchrony in flies mirror the signaling pathways that make the SCN a more robust oscillator than other mammalian clock cells (reviewed in [4]). VIP and PDF are both required to synchronize the molecular clocks in different neurons, both promote robust oscillations of clock proteins within clock neurons, and they both likely signal through Gαs and Adenylate cyclase [4]. We have not yet determined the signaling pathways downstream of cAMP that link to clock protein oscillations, but they likely include PKA and/or Epac, which affect circadian rhythms in flies and mammals [23],[49],[50]. Recent data show that PKA lies downstream of PDF and cAMP in Drosophila clock neurons (see Figure 9). In addition, Epac can regulate MAP kinase signaling, which is interesting because MAP kinase has also been proposed to lie downstream of PDF [51].

Our data provide evidence that the external signals that drive cAMP oscillations are received at different times of day. In the SCN, the amplitude of the cAMP rhythm is amplified by increased VIP signaling at dawn. cAMP levels decrease at dusk via falling VIP release and a release of the inhibition of Gαi/o by RGS16 [5]. However, the behavioral phenotypes of Rgs16^−/− mice are modest, suggesting that additional signaling mechanisms operate. Based on the similarity of the mammalian and Drosophila systems, we predict that a second signal released around dusk is also required for normal SCN function. Two possible signals are GABA [52] and glutamate perceived via its metabotropic receptor [53].

In Drosophila, different clock neuron groups respond to specific environmental inputs such as light or temperature [10],[11]. Thus the true function of the cAMP oscillator in flies and mammals may be to integrate information from diverse clock neurons into the molecular clocks of all clock neurons, generating a single time of day for an animal.

The following stocks used in this article have been described previously: Pdf¹[15], Pdfr^han5304[37], cry13-Gal4[54], cry39-Gal4[55], Pdf-Gal80[6], tim(UAS)-Gal4 (referred to in the text as tim-Gal4 for simplicity [56]), Pdfr-Gal4^GMR18F07[57], UAS-Dti[58], Pdf_0.5-Gal4[59], UAS-CD8::GFP[60], UAS-Epac1-camps[19], UAS-mGluRA^RNAi[32], UAS-Dicer-2[61], UAS-Gad1[33], UAS-GluCl^RNAi (v105724) [9],[62], UAS-Pdfr^RNAi (v42724) [61], UAS-Shi^ts[36], mGluRA^112b[63], UAS-AC3^{Vienna RNAi} (v33217), and UAS-AC3^{TRiP RNAi} (JF03041) [21]. UAS-mGluRA^RNAi, UAS-Pdfr^RNAi, Pdfr^han, mGluRA^112b, Pdf¹, [Pdf-Gal4; Dcr-2], [Pdf-Gal80; cry-Gal4], UAS-Dti, and UAS-Shi^ts stocks all carry the ls-tim allele [64]; thus, differences in TIM expression observed are not due to an inability of flies to express specific tim isoforms.

All immunocytochemistry was carried out as in [44]. We used the following antibodies: rat αTIM (from Amita Sehgal), rabbit αPDP1 [44], mouse αPDF [65], and rabbit αGFP (Sigma, St. Louis, MO). Images were scanned on a Leica SP2, SP6, or SP8 confocal microscope, with the same microscope used for a single experiment. The beginning and end of TIM staining was used to establish the limits of confocal stacks. The mean staining intensity for each channel for each neuron in every confocal stack was quantified using FIJI (http://pacific.mpi-cbg.de/wiki/index.php/Main_Page↗), with background levels of staining for each channel subtracted to control for variation in staining between brains. For each time course, the mean staining intensities for all LN_vs in each brain lobe were averaged to give a single value for an LN_v cluster. The average staining intensities per brain were then averaged to generate the time courses shown.

We used two methods to measure LN_v synchrony. In a simple binary method, we used a cutoff of 20 arbitrary units (au) above background to determine if a cell produced TIM or PDP1 or not. We chose 20 au as it is the lowest number where protein levels are convincingly visible above background. An LN_v cluster was then scored as “desynchronized” if they contained a mixture of LN_vs with and without detectable TIM or PDP1, or “synchronized” if all four LN_vs were the same. To more precisely quantify desynchrony, we also calculated the standard deviation in TIM or PDP1 mean staining intensities between individual LN_vs within a single LN_v cluster, producing a standard deviation in TIM or PDP1 staining intensity to use as a proxy for the level of desynchrony, allowing statistical comparisons of the data.

Larval light avoidance assays were carried out as in [9]. For adult locomotor activity experiments, adults were entrained to 12∶12 LD cycles at 25°C for at least 3 days before transfer to DD. Locomotor activity was recorded using the DAM system (TriKinetics, Waltham, MA).

Basal levels of Epac1-camps FRET were used to measure cAMP levels. Larval brains were dissected and mounted in hemolymph-like saline. To minimize the time from dissection to imaging (∼1 hour), different genotypes were removed from DD, dissected, and scanned in the same order. LN_v projections were imaged for CFP (460–490 nm) and YFP (528–603 nm) on an SP5 Leica confocal using a TD 458/514/594 dichroic 63× lens and 3× digital zoom at 100 Hz and 1024×1024 resolution, after excitation with a 458 nm laser. CFP and YFP levels were quantified using the Leica software. Background measurements of CFP and YFP were subtracted from raw CFP and YFP measurements and an average CFP to YFP ratio calculated for each image. For LN_v projections, five boutons in each image were quantified for CFP and YFP as above and averaged to give a value per projection.

Live cAMP imaging was performed on larval LN_vs as described in [66]. Briefly, larval brains were dissected in hemolymph-like saline and mounted to the bottom of a 35-mm Falcon culture dish lid (Becton Dickenson Labware, Franklin Lakes, NJ), fitted with a Petri Dish Insert (PDI, Bioscience Tools, San Diego, CA). Brains were allowed to settle for 5–10 min to reduce movement during imaging. Images were acquired on an Olympus FV1000 laser-scanning microscope (Olympus, Center Valley, PA) through a 60× (1.1N/A W, FUMFL N) Objective (Olympus, Center Valley, PA) using Fluoview software (Olympus). The Epac1-camps FRET sensor was imaged by scanning frames at 1 Hz with a 440-nm laser. An SDM510 dichroic mirror was used to separate CFP and YFP emission. Regions of interest were drawn around single LN_v cell bodies in Fluoview. Peptides were bath applied using a micropipette after 30 s of baseline imaging. PDF was dissolved in 0.01% DMSO and vehicle controls consisted of 0.01% DMSO delivered at the same volume as peptide applications (45 µL bath application into 405 µL hemolymph-like saline). The lowest PDF dose that evoked a consistent response (100 nM) was used to assay differential responses of LN_vs in which PDF or glutamate receptors had been knocked down in Figure 6. PDF was used at 100 µM in Figure S2. For each assay, no less than five larvae were imaged. Only one hemisphere was imaged per brain, and 1–4 LN_v were imaged per brain. Processing and analysis of Epac1-camps data was as described [67].

Fly locomotor activity was recorded in 5 min bins, using the DAM system (TriKinetics). Data analysis was performed using custom-written scripts in IgorPro (Wavemetrix). Sleep was defined as periods of immobility >5 min. Sleep latency was calculated for each fly on each day as the time from CT12 until the first sleep bout. Locomotor activity was calculated as the average number of beam crossings per 30 min bins and averaged for each genotype over the first 5 days in DD.