Circadian rhythms in neuronal activity propagate through output circuits.

Twenty-four hour rhythms in behavior are organized by a network of circadian pacemaker neurons. Rhythmic activity in this network is generated by intrinsic rhythms in clock neuron physiology and communication between clock neurons. However, it is poorly understood how the activity of a small number of pacemaker neurons is translated into rhythmic behavior of the whole animal. To understand this, we screened for signals that could identify circadian output circuits in Drosophila melanogaster. We found that leucokinin neuropeptide (LK) and its receptor (LK-R) were required for normal behavioral rhythms. This LK/LK-R circuit connects pacemaker neurons to brain areas that regulate locomotor activity and sleep. Our experiments revealed that pacemaker neurons impose rhythmic activity and excitability on LK- and LK-R-expressing neurons. We also found pacemaker neuron-dependent activity rhythms in a second circadian output pathway controlled by DH44 neuropeptide-expressing neurons. We conclude that rhythmic clock neuron activity propagates to multiple downstream circuits to orchestrate behavioral rhythms.

INTRODUCTION

Innate behaviors such as circadian rhythms are hardwired into the nervous system, making them particularly useful to study how information flows through neuronal circuits to generate behavior. Circadian rhythms in behavior help animals anticipate predictable daily changes in the environment [1, 2] and are controlled by circadian pacemaker neurons. These neurons contain molecular clocks that drive rhythmic gene expression and set up 24hr rhythms in pacemaker neuron resting membrane potential, spontaneous firing rate and overall excitability [3]. Communication between clock neurons synchronizes their molecular clocks and adds robustness to the system [2]. Specific subgroups of clock neurons have peak neuronal activity at different times of day from each other and presumably regulate distinct output circuits to drive numerous rhythmic behaviors including locomotor activity, sleep and feeding [1]. However, how the clock neuronal network controls different output circuits remains poorly understood.

Rhythms in pacemaker neurons could propagate to downstream cells via two mechanisms. Clock neurons can act on distant cells via rhythmic hormonal signals which entrain and synchronize molecular clocks in peripheral tissues. For example, clock neurons control rhythmic glucocorticoid release from the adrenal gland into the bloodstream, which then helps reset the molecular clocks in peripheral organs such as the liver [4]. Clock neurons could also impose rhythmic activity on downstream neurons via direct neuronal communication. Although many neurons fire rhythmically in the mammalian brain, widespread clock gene expression in mammals makes it difficult to exclude a role for local clocks in these rhythms [5].

Studies of Drosophila have been instrumental in dissecting the molecular and neuronal bases of circadian rhythms [2]. However, how clock outputs are mediated in Drosophila remains poorly understood. Although peripheral clocks control rhythms in eclosion [6] and feeding [7], the clock output circuits controlling locomotor activity rhythms and sleep remain elusive. These outputs probably converge on the Central Complex (CC) [8], Pars Intercerebralis (PI) [9, 10] and the Mushroom Bodies [11-13]. One output pathway links the small LNv principal pacemaker neurons (s-LNv) to DN1p clock neurons, which then innervate a subset of PI neurons that express the DH44 neuropeptide. These DH44-expressing neurons are required for circadian rhythms [10], but how their activity is regulated by the clock network has not been addressed. A second likely clock output pathway involves CC neurons that respond to Pigment Dispersing Factor (PDF) released from LNvs, although the role of these CC neurons in circadian behavior has not yet been determined [14]. A third output pathway involves DH31 release from DN1 clock neurons to regulate sleep, but the relevant targets remain to be characterized [15].

Here, we identified an additional circadian output circuit connecting clock neurons to locomotor and sleep centers in the brain. This circuit comprises a pair of non-clock neurons expressing the neuropeptide Leucokinin (LK) and a set of downstream neurons expressing the Leucokinin Receptor (LK-R) that project to the CC. Using calcium imaging, we demonstrate that clock neurons impose 24hr rhythms on the excitability and activity of LK and LK-R neurons by neuronal communication. We also show that LK and LK-R neurons control the rhythmicity and levels of locomotor activity and sleep. In addition, we found that clock neurons also impose activity rhythms on the previously characterized DH44 circadian output neurons. Thus propagation of clock neuron electrical rhythms is a general mechanism for organizing circadian rhythms of behavior via multiple circuits.

RESULTS

Leucokinin signaling is required for circadian rhythms

We hypothesized that we could identify a novel circadian output circuit by screening for circadian behavioral defects in flies mutant for a signaling molecule and/or its relevant receptor. We chose neuropeptides since they usually have more restricted distributions than neurotransmitters and since many neuropeptides modulate neuronal activity to regulate specific behaviors, such as PDF in circadian rhythms [16, 17].

We used transgenic RNAi lines expressed via the pan-neuronal driver elav-Gal4 to knockdown Drosophila neuropeptides in the whole brain and then assayed adult locomotor rhythms in constant darkness (DD; and ). We found 4 RNAi transgenes that significantly weakened behavioral rhythms: Bursicon, SIFamide, Leucokinin (Lk) and Nplp3 (). We focused on Lk because it does not seem to be involved in development and has an intriguing expression pattern in the brain (see below).

To further test a role for Lk signaling in circadian rhythms, we used RNAi to knockdown the Leucokinin Receptor (Lkr). This also weakened behavioral rhythms ( and ). To complement these RNAi experiments, we assayed the behavior of Lk and Lkr hypomorphic flies, which have reduced LK peptide and LK-R protein levels [18]. These mutants had weaker rhythms than heterozygous control flies. Lkr and additional Lkr alleles gave similar behavioral phenotypes as hemizygotes ( and ).

Quantifying LK peptide levels and Lkr RNA in the different mutants revealed that the strength of behavioral phenotypes correlates with the extent of knockdown: LK levels are reduced much more strongly by Lk than Lk, with only 3% of wild type LK levels in Lk flies and stronger effects on behavior than for Lk ( and ). Lkr RNA levels are reduced to similar levels in Lkr hypomorphs and Lkr knockdowns () and have similar strength behavioral phenotypes ().

Together, these RNAi and mutant analyses indicate that LK signaling is important in adult circadian rhythms. We have not been able to test a specific role for Lk in adult neurons since restricting Lk expression to adulthood did not reduce LK peptide levels (data not shown). Below we describe how manipulating the activity of LK and LK-R expressing neurons in adults alters rhythmic locomotor activity.

LK and LK-R neurons are not clock neurons

Next we tested if LK and LK-R neurons are clock neurons themselves. LK is expressed in only 4 neurons in the adult brain [19]: one pair of neurons (SELKs) in the Sub-Oesophageal Ganglion (SOG) and another pair in the Lateral Horn, the Lateral Horn LK neurons (LHLKs; ). Since the DN3 group of clock neurons is close to LHLK cell bodies, we examined LK staining with clock neuron markers. We found that LHLK neurons do not produce the essential clock protein TIMELESS (TIM, ) nor do they express tim- or per-Gal4 ( and ). We also expressed a dominant negative cycle transgene (UAS-cycΔ [20]) to block the molecular clock. UAS-cycΔ completely abolished behavioral rhythms when expressed in clock neurons via tim-Gal4, but had no effect when expressed in LK neurons ( and ).

To examine Lkr expression, we used an Lkr-Gal4 line that recapitulates endogenous LK-R expression [18]. Lkr is more widely expressed in the brain than LK and is also present in regions where clock neurons are located (). However, we could not detect TIM expression in Lkr-Gal4 expressing neurons () and locomotor rhythms were unaffected by expressing UAS-cycΔ in LK-R neurons ( and ). Thus we conclude that LK and LK-R neurons are not clock neurons.

Finally, we tested whether LK signaling affects the molecular clock in pacemaker neurons. We measured TIM and VRI oscillations in the strong Lk knockdown on the second and third days in DD and found that rhythms were very similar to control flies (). Thus LK signaling is likely downstream of the clock since Lk disrupts behavior without affecting molecular clock rhythms.

LHLK neurons project close to clock neurons

To test if LK and LK-R neurons are outputs of the clock, we wanted to determine if they communicate with clock neurons. We first analyzed their anatomy and found that LHLK projections (marked by LK staining) are very close to the dorsal projections of the s-LNv clock neurons (). To more clearly visualize LHLK projections, we used an Lk-Gal4 line that recapitulates endogenous LK expression [19] and the GFP amplification cassette FLEXAMP [21]. This revealed numerous sites of potential contact between s-LNv and LHLK projections (, inset). Several clock neuron classes converge at the s-LNv dorsal projections [22] and we also found LK staining very close to projections of DN1p and LNd clock neurons (). These data are consistent with LHLKs communicating with one or more classes of clock neurons.

Next we examined the location of LK neuron synaptic and dendritic markers. We expressed DenMark [23] and Syt::GFP [24] in LHLK neurons to simultaneously label LHLK input and output areas respectively and used PDF to label the clock network output region. We found that DenMark accumulates in LHLK neurons close to s-LNv projections with LHLK dendrites often intermingling with s-LNv projections in single confocal sections (). In contrast, we mainly found Syt::GFP in more posterior sections of LHLK neurons, which do not contain s-LNv projections (). Since s-LNvs have pre-synaptic markers all along their dorsal projections [25], our observations are consistent with the idea that LHLK neurons lie downstream of clock neurons.

Given the proximity of clock neurons and LHLKs, we wanted to test whether LK regulates circadian rhythms via LHLKs rather then the other LK-expressing neurons [19, 26]. We used apterous-Gal4 to express Lk in LHLKs but not in SELKs or Abdominal LK neurons (ABLKs) [26]. Since apterous-Gal4 > Lk flies had weaker rhythms than control flies (), we propose that LK functions as a clock output specifically in LHLK neurons. This function appears distinct from the roles of LK signaling in feeding and diuresis, which are likely mediated by the SELKs and ABLKs [18, 27].

The LK/LK-R circuit connects to locomotor and sleep areas

Since LHLK neurons contact LK-R neurons [18], we examined the projections of LK-R expressing neurons to determine which brain regions are the likely target of the LK/LK-R circuit. LK-R neurons form a dense and complex meshwork. However, a subset of LK-R neurons either project to or have cell bodies in brain regions implicated in controlling locomotion and/or sleep – specifically the PI and two regions of the Central Complex: the Ellipsoid Body (EB) and Fan-Shaped Body (FSB) [8, 9] (). To visualize subsets of LK-R neurons, we generated flip-out FLEXAMP clones with Lkr-Gal4. We observed FLEXAMP GFP in the FSB in all 6 clones that labeled LK-R cell bodies in the lateral horn (). These lateral horn LK-R neurons arborize in the posterior part of the brain and overlap extensively with LHLK projections (). Moreover, these LK-R arborizations are likely inputs since they are enriched for DenMark staining () and these posterior regions of the brain are where the pre-synaptic marker Syt::GFP localizes in LHLKs (). These data suggest that most LK-R neurons projecting to the FSB receive inputs from LHLKs. In contrast, LK-R outputs marked by Syt::GFP are primarily in the EB and FSB (). LK-R outputs are also present in the SOG where SELKs are found (data not shown).

We found a second Lkr-Gal4 line (Lkr) that also labels neurons in the lateral horn with pre-synaptic termini in the FSB (). Using a Lkr driver, we found that LK-RR65C07 projections in the FSB intermingle with projections from neurons labeled by 3 other FSB neuron Gal4 lines that affect locomotor activity and sleep [8, 28] (). Thus LHLKs and these lateral horn LK-R neurons have the appropriate anatomy to connect clock neurons with locomotor activity and sleep control centers.

LHLK neuron excitability is regulated by clock neurons

Next we used a functional approach to directly test connectivity and identify the direction of information transfer. We first manipulated s-LNv activity by expressing the mammalian ATP-gated cation channel P2X2. Since Drosophila neurons do not express endogenous ATP-gated channels, ATP only activates neurons expressing the P2X2 transgene [29]. We determined how this affects LHLK neuronal responses, using the genetically-encoded calcium indicator GCaMP6S [30] as a proxy for neuronal activity.

As a positive control, we first expressed P2X2 and GCaMP6S in LNvs and detected robust calcium transients in s-LNvs after perfusing ATP onto explanted brains (). We saw similar responses in the large LNvs (l-LNvs) that regulate arousal [31, 32]. To measure responses in LHLKs, we used Lk-Gal4 to express GCaMP6S and Pdf-LexA [32] to express P2X2 in LNvs. However, LNv activation did not detectably change GCaMP6S fluorescence in LHLK neurons (, red).

To test whether LNvs inhibit LHLK neurons, we first needed to identify a way to activate LHLKs. We found that the acetylcholine agonist Carbachol (CCh) induces calcium transients in LHLKs in a dose-dependent manner (). We pre-incubated brains with Tetrodotoxin (TTX) to determine if this response is direct. TTX blocks most communication via neural circuits by preventing action potentials, although graded potentials are probably unaffected. The LHLK response to CCh persisted in the presence of TTX, suggesting that CCh directly activates LHLKs ().

We then tested whether the response of LHLKs to CCh is inhibited by LNv activation, using a lower CCh concentration to be able to detect inhibition. We found that inducing LNv firing almost completely abolished the LHLK response to CCh (). This inhibition is specific since it requires P2X2 expression in LNvs (). Thus LHLK neurons are functionally post-synaptic to the clock network, consistent with their anatomy (). Our results also reveal the sign of this connection: LHLKs are inhibited by LNv activity.

CCh could generate calcium transients by activating muscarinic receptors to release internal calcium stores [33] or nicotinic receptors to depolarize the neurons and induce firing. To test whether LNvs inhibit LHLKs via intracellular signaling or via membrane excitability, we added 35mM KCl to directly generate calcium transients in LHLKs by depolarization. We found that the response of LHLKs to KCl was strongly reduced when LNvs were simultaneously activated (). Thus clock neurons inhibit LHLKs at the level of membrane excitability.

Next, we asked whether PDF neuropeptide - the main LNv output - is responsible for inhibiting LHLKs. PDF increases cAMP by activating the PDF Receptor (PDFR) in several classes of clock neurons including s-LNvs and DN1ps [16, 17]. We found that a 30 second PDF perfusion gradually increased intracellular calcium levels in s-LNvs, consistent with PDF depolarizing s-LNvs and DN1ps [16, 17]. This response is specific since it was not observed in l-LNvs (, which do not express PDFR [34].

Although LHLK neurons were not detectably activated by PDF perfusion (), pre-incubating brains with PDF dramatically inhibited their CCh response (). PDF inhibition was transient and disappeared after 15 min washout (). Thus PDF signaling can inhibit LHLKs, further evidence that LHLKs are downstream of the clock network.

However, since PDF can activate s-LNvs and other clock neurons [16, 17, 34] (), our data do not determine if PDF directly controls LHLK excitability. To test this, we used two approaches. First, we used TTX to block action potentials while applying PDF. TTX was added for 20 min prior to PDF and also throughout the experiment. We found that TTX treatment largely eliminated LHLK inhibition by PDF (), indicating that PDF acts indirectly on LHLKs. Second, we pre-incubated brains in PDF but this time with LNvs ablated via a Pdf-Dti transgene [35]. This also prevented PDF from inhibiting LHLKs (). Since PDF requires LNvs to inhibit LHLKs, we interpret this to mean that PDF activates s-LNvs () which then signal to LHLKs either via an additional s-LNv neurotransmitter [36] or indirectly via the clock network. Identifying the neurons that directly regulate LHLKs will require finding the signal that modulates LHLK excitability.

LK peptide does not modulate LNv excitability

We then determined how LK affects its target neurons. We first tested if LK-R neurons respond to LK peptide, focusing on the LK-R neurons with cell bodies in the lateral horn that project to the FSB. Adding LK to Lkr > GCaMP6S brains did not activate these LK-R neurons (). However, they were activated by CCh and this response was strongly reduced by pre-incubation in LK peptide (). This contrasts with non-neuronal stellate cells where LK increases intracellular calcium [37] and could be explained by differential G protein coupling in distinct cell types. We conclude that the LK-R neurons in the lateral horn are bona fide LK-responding neurons. In addition, their projection patterns strongly suggest they are downstream of LHLKs but not SELKs. Therefore we propose that the LK/LK-R network connects clock neurons to the FSB and possibly also to the EB and PI.

We also tested whether LK feeds back on clock neurons. Since we found no evidence for LK activating LNvs in Pdf > GCaMP6S brains (data not shown), we tested whether LNvs can be inhibited by LK. LNvs respond to CCh [38], but this response was unaffected by pre-incubating brains with LK ( and ). Thus LK does not affect s-LNv activity and LHLK neurons seem to act as outputs of the clock network.

Clock neurons impose rhythms on LHLK activity

LNvs and DN1ps are most depolarized and have highest spontaneous firing rates around dawn [39-41]. Since LHLK neuronal excitability is controlled by LNV firing, we speculated that LHLK neuron activity is also rhythmic (). However, the timing of peak LNv and LHLK activity should differ since LNvs inhibit LHLKs..

To test these ideas, we first measured LHLK responses to CCh at two different times in DD. We measured LHLK responses during the subjective morning (CT0-3) when LNv activity is high and the subjective evening (CT9-12) when LNv activity is low [39] [CT: Circadian Time in DD after entrainment to a 12:12 Light:Dark cycle]. We maintained individual flies in the dark until dissection to minimize exposure to light. Strikingly, we found that the LHLK response to CCh was two-fold lower in the subjective morning than subjective evening and additional time points revealed a 24hr rhythm (). Low LHLK excitability when LNv activity is high is consistent with LNvs inhibiting LHLK neurons. Oscillations in explanted brains indicate that these rhythms are not driven by locomotor activity.

LHLK excitability rhythms are likely clock-controlled since they persist in DD. To test this, we measured CCh responses in period null mutant flies (per) in which the molecular clock has stopped [1]. We found that changes in LHLK excitability were lost in per mutants, showing that these rhythms require intact molecular clocks (). This suggested that LHLK rhythms are imposed by circadian pacemaker neurons since LHLK neurons do not contain molecular clocks. We thus measured LHLK excitability rhythms in brains with LNvs ablated and found this also eliminated LHLK rhythms ().

To determine if these LHLK rhythms reflect endogenous neuronal activity, we quantified baseline GCaMP6S fluorescence in living explanted brains as a measure of spontaneous activity [42]. We observed a robust oscillation of GCaMP6S intensity with a peak around subjective dusk (CT11) and a trough at subjective dawn (CT0 and CT23, ). We did not see any changes in GFP intensity between CT0 and CT11 using a destabilized GFP transgene expressed with Lk-Gal4 ( as in ref. [43]). Thus the GCaMP6S oscillation is not due to rhythmic Lk-Gal4 expression and presumably reflects changes in spontaneous LHLK activity over 24hr.

We also measured baseline GCaMP6S levels in LHLK neurons in per mutants () and when LNvs were ablated (). Rhythms were lost in both situations, confirming that LHLK excitability is clock-controlled and driven by pacemaker neurons. We also found that artificially activating s-LNvs in the evening by applying PDF peptide decreased baseline GCaMP6S levels in LHLKs (). Thus we conclude that s-LNv firing reduces LHLK neuronal activity. PDF did not reduce LHLK GCaMP6S to the trough levels observed at dawn. This could mean that either s-LNv firing is required for >30min to fully inhibit LHLKs or that weaker s-LNv synaptic outputs at dusk [43] prevent complete LHLK inhibition. In conclusion, these results demonstrate that LHLK excitability rhythms are generated non cell-autonomously by rhythmic signaling of the clock network.

LHLK activity rhythms propagate to LK-R neurons

Next we tested if LHLK activity rhythms are transmitted to LK-R neurons. We measured LK-R responses to CCh in DD and found that LK-R neurons are more excitable at dawn than dusk (). Furthermore, LK-R excitability rhythms were abolished in per mutants (). We also measured LK-R excitability in Lkr hypomorphs to test whether LK peptide itself transmits LHLK activity rhythms to LK-R neurons. We found that LK-R excitability oscillations were dampened in Lkr mutant flies (), consistent with the hypomorphic nature of this allele and with LK modulating LK-R neurons. Thus LK-R neuron excitability is rhythmic, clock-controlled and in antiphase to LHLKs, consistent with LK peptide inhibiting LK-R neuronal activity.

We also tested if these LK-R excitability rhythms reflect endogenous rhythms in neuronal activity by measuring baseline GCaMP6S levels. We observed a robust 24hr oscillation () in antiphase to LHLKs (compare ). This oscillation was clock-dependent () and blocked in Pdf-Dti brains (), demonstrating that it originates from pacemaker neurons.

We also used Lkr line to drive GCaMP6S in the LK-R neuronal subset that projects to the FSB (see ). These neurons also respond to LK peptide () and their baseline GCaMP6S levels oscillate in phase with Lkr-Gal4 (), confirming that both Lkr-Gal4 lines label the same neurons. Thus rhythmic pacemaker neuron activity is propagated at least two layers deeper into the brain to generate non cell-autonomous rhythms in LK-R neurons via LK signaling.

LK and LK-R neurons control locomotor activity and sleep

Rhythms in LHLK and LK-R activity suggest that the clock network imposes rhythmic neuronal activity on locomotor and sleep control centers. To test if these neuronal rhythms are important for behavioral rhythms, we manipulated LK and LK-R neuronal activity. To manipulate the subset of LK-R neurons most likely to receive LHLK inputs, we used the more restricted Lkr. We activated LK and LK-RR65C07 neurons for 4 days using a UAS transgene expressing the heat-activated cation channel dTrpA1, which is inactive below 25°C [44]. After entraining to LD cycles at 19°C, flies were assayed in DD for 4 days at 19°C and then for 4 days at 28°C. Control flies had stronger rhythms at 28°C than 19°C, as seen previously [10, 43]. In contrast, activating LK neurons weakened behavioral rhythms at 28°C compared to 19°C, while activating LK-RR65C07 neurons blocked the increase in rhythm strength at 28°C ( and ). These data suggest that LK/LK-R neuronal activity rhythms are required for normal behavioral rhythms.

To explore the effect of LK and LK-R neurons in more detail, we performed 1-day activation experiments and also used the temperature-sensitive dominant negative Dynamin (UAS-shi) to block synaptic outputs [45]. Control flies increased their activity levels in response to heat (grey and black lines in ). This was due to increased locomotor activity while awake and decreased sleep (). In contrast, activating LK neurons dramatically reduced locomotor activity levels () by increasing the amount of sleep and reducing activity levels while awake (). Since Lk > dTrpA1 flies recovered similar activity and sleep levels to control flies on returning to 19°C, activating LK neurons does not permanently alter locomotor and sleep circuit function or render flies unhealthy ( and ).

Activating LK-RR65C07 neurons had the opposite effect to activating LK neurons, with increased locomotor activity and decreased sleep compared to controls ( and ). This effect was shorter-lived than for LK neurons and was most apparent during the first 6hr of the temperature shift (lower panel in , quantified in ). The opposite effects of activating LK and LK-R neurons are consistent with LK inhibiting LK-R excitability () and indicate that LK neurons control locomotor activity and sleep levels by inhibiting LK-R neurons.

We then inhibited synaptic transmission from LK and LK-RR65C07 neurons with shi. Surprisingly, inhibiting LK neuron synaptic transmission had almost no effect on locomotor activity or sleep ( and ). One possible explanation is that LK neurons control these behaviors via neuropeptide signaling, which may be Dynamin-independent [46]. Indeed, constitutively hyper-polarizing LK neurons with the inward rectifier potassium channel Kir2.1 [47] significantly reduced locomotor rhythm strength ().

In contrast, inhibiting synaptic transmission from LK-RR65C07 neurons reduced locomotor activity and increased sleep ( and ). This effect is the opposite of LK-RR65C07 activation and similar to activating LK neurons (). These data further support the model that LK neurons inhibit LK-R neurons which normally promote locomotor activity and inhibit sleep. These results also show that LK-R neuron signaling is required by day for normal levels of locomotor activity and sleep.

We repeated the experiments with LK-RR65C07 neurons with a heat pulse starting at CT12 and obtained very similar results to CT0-24 heat pulses: LK-RR65C07 neuron activation and inhibition mainly affected behavior during subjective day (data not shown). Thus LK-R neurons seem competent to control locomotor and sleep only at times when they are most excitable (see . The absence of phenotypes at night could be due to masking effects by heat and/or interactions with other neural pathways that override the effects of LK-R signaling during subjective night. Indeed, light dramatically delayed the effects of LK and LK-RR65C07 neuron activation and inhibition on locomotor activity during the day (). This suggests that one or more pathways downstream of light at least partially suppress the effects of interfering with LK-R neuron signaling. Thus LK-R neuron outputs are likely integrated with other pathways to shape behavioral rhythms.

LK-R expressed in Malphigian tubule stellate cells responds to circulating LK peptide released from ABLKs to regulate diuresis [37]. To test if the dTrpA1 locomotor activity phenotypes are require neuronal expression, we added elav-Gal80 to eliminate dTrpA1 expression from neurons in Lk > dTrpA1 and Lkr flies. Restricting dTrpA1 expression to non-neuronal tissues completely abolished locomotor activity and sleep phenotypes (). Thus the LK and LK-R cells controlling locomotion and sleep are neurons. Together with LK/LK-R anatomy and our functional imaging experiments, these behavioral data implicate the brain LK/LK-R circuit as a critical circadian output that regulates rhythmic locomotor activity and sleep.

Rhythms propagate in a second clock output circuit

Finally we examined a second group of clock output neurons – the DH44-expressing neurons in the PI that do not express molecular clock components but receive inputs from DN1p clock neurons [10]. We found that baseline GCaMP6S levels oscillate in DH44 neurons and this requires LNvs (). Thus rhythmic DH44 neuron activity is also imposed by pacemaker neurons. We propose that non-autonomous propagation of neuronal rhythms is a general mechanism for transmitting pacemaker neuron information.

DISCUSSION

How does the clock network regulate downstream circuits? We show that the LHLK, LK-R and DH44 neurons downstream of the Drosophila clock network display clock-dependent activity rhythms in explanted brains although these neurons have no molecular clocks themselves. The loss of LHLK, LK-R and DH44 neuronal activity rhythms after LNv ablation demonstrates that these rhythms originate from pacemaker neurons (see ). In addition, PDFR-expressing neurons in the EB display a circadian rhythm in their cAMP response to acetylcholine which partly depends on PDF [14]. Thus clock output pathways relay rhythmic information to several different brain regions using diverse signals.

Function of LK/LK-R signaling

Behavioral analyses of Lk and Lkr mutants and neuronal manipulations implicate the LK/LK-R circuit in organizing locomotor activity and sleep over time. Specifically, we found that LK-R neurons promote locomotor activity and reduce sleep (). This function seems distinct from the diuretic function of LK/LK-R signaling which is likely controlled by LK release from ABLKs [19]. Indeed, we found that locomotor behavior was disrupted with RNAi targeting LK only in LHLKs but not in SELKs and ABLKs and also when manipulating LK-R specifically in neurons. Other functions of LK/LK-R signaling such as regulating feeding [18] are unlikely to affect locomotor rhythms since Lk and Lkr mutants ingest normal amounts of food [18] and since blocking feeding rhythms does not alter locomotor activity rhythms [7]. However, we cannot completely rule out that LK/LK-R regulation of feeding affects locomotor and sleep behaviors given the precedent of orexin/hypocretins regulating both energy intake and arousal in vertebrates [48].

LK-R neurons intermingle with neurons that promote locomotor activity in the FSB [8, 28]. However, LK-R neurons projecting to other locomotor centers such as the PI and EB might also contribute to circadian behavior. The locomotor activity-promoting role of LK-R neurons is consistent with their neuronal activity profile determined by GCaMP: they are more excitable and active around dawn, when flies have high locomotor activity. Together with our analysis of LHLK and pacemaker neuron connections, these observations suggest a model in which signaling from the clock network inhibits LHLK neurons at dawn to allow LK-R neurons to signal and promote locomotor activity (). Supporting this model, downregulating Lk and Lkr by RNAi interferes with morning anticipatory behavior but has no effect in the evening (data not shown).

In addition to being inhibited by LHLKs at dusk/night, we found evidence suggesting that LK-R neuron outputs are blocked by additional unidentified signaling pathways since activating them at night did not affect locomotor behavior. Some of these pathways may be downstream of light, which partially suppresses LK-R-driven locomotor activity during the day. The exact timing of LK and LK-R firing likely also depends on additional non-circadian inputs and probably differs from the windows of excitability imposed by clock neurons. Additional work will be required to determine how the different circuits downstream of the clock interact to organize circadian behaviors.

In conclusion, our experiments reveal a mechanism to temporally control behavior: Pacemaker neuron electrical rhythms are propagated through downstream neuronal circuits that control specific components of circadian behavior.

ONLINE METHODS

Fly strains

The following Drosophila melanogaster fly strains have been described previously: y (ref. [18]), y (ref. [18]), y w; ; Lkr (MI06336), y w; ; Lkr (MI08640), w (ref. [49]), w (ref. [49]), w (ref. [49]), y w; Lk-Gal4 (ref. [19]), y w; ; Lkr-Gal4 (ref. [18]), Pdf (ref. [50]), tim(UAS)-Gal4 (referred to here as tim-Gal4 , ref. [51]), per-Gal4 (ref. [52]), Ap-Gal4 (ref. [53]), UAS-nlsGFP (from C. Desplan), elav-Gal4; UAS-Dcr-2 (ref. [54, 55], the UAS-Dcr-2 transgene was included to enhance RNAi effectiveness), y (TRiP JF01816), y (TRiP JF01956), 10×UAS-mCD8::GFP (ref. [56]), FLEXAMP cassette: y w, UAS-Flp; tub-Gal80ts / CyO ; act-FRT-stop, y+-FRT-LexA, 13×lexAop-myr-GFP (ref. [21]), Clk4.1-Gal4 (ref. [57]), Mai179-Gal4 ; PdfGal80 (ref. [58, 59]), w; L / CyO; UAS-Syt::GFP, UAS-DenMark (ref. [23]), w; UAS-cycΔ (ref. [20]), w; 20×UAS-GCaMP6S (ref. [30]), w; ; Pdf-LexA, lexAop-P2X (ref. [60]), y per; (ref. [61]), Pdf-Dti (ref. [35], we verified that Pdf-Dti ablates all adult LNvs by immunostaining, n=8 brains, data not shown), UAS-dsGFP (ref. [62]), DH44-VT-Gal4 VDRC (ref. [10]), w (ref. [47]), w; UAS-dTrpA1 (ref. [44]), +;UAS-shi (ref. [45]), w and w (ref. [63], Janelia R65C07), 121Y-Gal4, C5-Gal4, c584-Gal4, described in ref. [8, 28], elav-Gal80 (from S. Sweeney, ref. [64]).

Behavioral analyses

Flies were raised on regular cornmeal medium and entrained to 12:12hr LD cycles for at least 3 days before transfer to DD. Flies were raised and assayed at 25°C except for experiments involving UAS-dTrpA1 or UAS-shi in which flies were raised at 19°C and assayed at 19°C and 28°C. Male flies assayed for behavior were ~5-10 days old. No randomization or blinding was used when preparing and analyzing behavioral experiments, but controls were performed in parallel. Locomotor activity was recorded using the DAM system (TriKinetics, Waltham, MA). Manual inspection of actograms was performed for each fly to exclude flies that died during an experiment. Rhythm strength (power) and period were analyzed by ClockLab in Matlab using chi-squared analysis. All other analyses (actograms, activity and sleep profiles, total activity and waking activity) and statistical tests were performed using custom-written scripts in IgorPro (Wavemetrics) as in ref. [65]. Sleep was defined as at least 5 min of inactivity: 0 beam crossings in a 5 min data window.

Minimum required sample sizes were determined empirically, no statistical methods were used. Behavioral experiments were repeated at least twice. The RNAi screen was initially performed with 8 flies but genotypes with potentially interesting phenotypes repeated to obtain >16 flies. For dTrpA1 and shi experiments, larger sample sizes were required since most experimental treatments lasted only one day. Thus these experiments were performed with 32 flies at least twice. To determine which statistical test to compare experimental flies to parental controls, we first ran Levene's test to determine if variances were equal and found that they were often unequal. Thus we used the non-parametric Kolmogorov-Smirnov test (KS test) to determine if experimental flies differed from controls. dTrpA1 or shi manipulations were considered to have a significant effect when experimental flies were statistically different from both controls (p<0.05).

Immunocytochemistry

Adult brains were dissected in PBS, fixed for 45 min in 4% formaldehyde in PBS, rinsed 3x in PBS + 1% Triton and washed for ~2hr in PBS + 1% Triton. Primary antibodies were incubated in PBS + 0.5% Triton + 4% horse serum overnight at 4°C. Secondary antibodies were incubated for 2hr at room temperature and rinsed overnight at 4°C. Brains were mounted in SlowFade (Invitrogen). Primary antibodies used were: rabbit anti-LK at either 1:1000 or 1:10000 for quantification (ref. [18]), mouse anti-PDF 1:50 (ref. [66]), rat anti-TIM 1:250 (from A. Sehgal), Guinea Pig anti-VRI 1:1500 (from P. Hardin), sheep anti-GFP 1:500 (Novus Biologicals NB100-62622), rabbit anti-RFP 1:500 (Invitrogen R10367) and mouse anti-RFP 1:100 (MBL 8D6). Alexa Fluor (Invitrogen) secondary antibodies were all used at 1:200. Confocal stacks were acquired with a Leica SP5 confocal microscope with a 20x water immersion objective and processed in ImageJ. Anatomical analyses were performed on male and female flies (~2-10 days old), with a minimum of 8 brains imaged on both left and right sides for each experiment. Anatomical observations were highly reproducible from brain to brain.

TIM staining in were performed in brains dissected at ZT22 (TIM levels high in clock neurons) except for DN2s where brains were dissected at ZT10 (. Flip-out FLEXAMP clones () were generated with tub-Gal80; Lkr-Gal4 by transferring developing larvae at 29°C for ~3hr to allow transient UAS-Flp expression from Lkr-Gal4.

Protein level quantification was performed on ~3-5 day old male flies entrained to LD cycles. Quantification of LK, VRI and TIM levels in wild type and Lk mutants and of dsGFP in wild type flies were performed using IgorPro (Wavemetrics) on 8-bit images (i.e. pixel intensity ranging from 0-255). Average pixel intensity (integrated intensity / area) was measured for individual cell bodies using manually-defined ROIs on z projections. Background intensity was measured for each image and subtracted from the corresponding cell bodies. Sample sizes were determined empirically based on the relatively low variability observed between brains as previously [65]. We imaged 8 brains (16 LHLK cell bodies) to quantify LK levels in and 9-10 brains (>60 cell bodies) for TIM and VRI in for each data point. To quantify LK levels in Lk brains, LHLK cell bodies were first identified using higher laser power since they were almost undetectable and then imaged using regular acquisition settings. s-LNvs were identified using PDF staining and distinguished from l-LNvs by size. No randomization or blinding was used when preparing and analyzing immunostaining.

Quantitative Real Time PCR (qPCR)

Lkr mRNA levels were measured by qPCR using a standard curve constructed as in ref. [67]. RNA was extracted from male whole heads (40 heads/extraction, flies 3-5 days old) using PureLink RNA Mini Kit (Ambion). Sample sizes were based on ref. [67]. Each data point consisted of 2 biological replicates with 3 technical replicates (6 samples total). No randomization or blinding was used when preparing and analyzing qPCR. Reactions were performed using the LightCycler RNA Master HybProbe kit (Roche) with Calmodulin as a loading control [43]. Primers and probes were synthesized by TIB Molbiol (Adelphia, NJ).

Lkr-F: AAATGCGGACCGTGACA

Lkr-R: GGACGTGCCCTAAGTGGAT

Lkr-FL: GGTATTCACGCTGACCGCCA--FL

Lkr-LC: LC640-TGCAATCGATCGGCATAGGGCC--PH

Calcium imaging

We chose the GCaMP6S variant for two reasons. First, GCaMP6S has slower kinetics than GCaMP6F and GCaMP6M [30], which makes it easier to detect calcium transients in many neurons in a large field of view at slow scanning speed – even though the slow time course of GCaMP6S responses (>5min) is not physiological. Second, GCaMP6S is more sensitive than GCaMP6F and GCaMP6M [30], making it easier to detect subtle changes in neuronal activity, especially for baseline GCaMP6S fluorescence.

3-5 day old adult male flies entrained to LD cycles were anesthetized on ice and dissected in hemolymph-like saline (HL3, ref. [34]). For measuring GCaMP6S responses to drugs, brains were gently pressed against a glass slide coated with Poly-L Lysine (Sigma) and mounted in a Bioptechs FCS3 perfusion chamber. HL3 flow across the brain was established and maintained at ~1ml/min by gravity. Brains were allowed to recover for ~5min in the chamber before an experiment. Test compounds (0.5ml) were injected into the tubing system using a syringe and 3-way stopcock. Compounds were perfused for ~30sec and started at slightly different times depending on experiments because of small changes in flow rate and tubing length. The timing of drug perfusion is indicated by the position of the grey bar on GCaMP6S line graphs. To measure baseline GCaMP6S intensity, live brains were mounted in HL3 medium on glass slides coated with Poly-L Lysine and imaged immediately in one z-stack. No randomization or blinding was used when preparing and analyzing calcium imaging experiments.

ATP (Sigma) and Carbachol (Sigma) were dissolved directly in HL3. KCl was used at 35mM after ref. [68]. PDF and LK peptides were synthesized by PolyPeptide Group (San Diego, CA), dissolved in DMSO, diluted to final concentration in HL3 and used within 1 day. Three different batches of PDF were used during this study and had different efficacies, as previously noted [14]. The batch used in was more potent and thus used at lower concentration (20 μM) than the batches used in (100 μM).

TTX (Tetrodotoxin citrate, Abcam) was dissolved in HL3 and included in the main HL3 flow throughout the relevant experiments and while test compounds were injected. Electrophysiological recordings of l-LNv neurons show that TTX completely eliminates action potentials within 1 min of application on brain explants [40]. We used a high TTX concentration to completely inhibit action potentials as in ref. [38, 69]. TTX works in our preparation since it eliminates the inhibitory effect of PDF on LHLKs (see ).

Pre-incubation with PDF, LK or TTX was in a 1ml drop/well of HL3 prior to mounting in the chamber. For the PDF+TTX experiment (), brains were first incubated for 20 min in TTX and then for another 20 min in PDF+TTX (or vehicle+TTX). PDF washout () was performed in the perfusion chamber after the first CCh stimulation.

GCaMP6S imaging was performed with an Olympus two-photon system with a Mai-Tai laser (Spectra Physics) at 920 nm and a 10x water immersion objective. z stacks (~20 slices at 5 μm intervals) were acquired every 30 seconds for 10 minutes. Maximal z projections were used to quantify fluorescence in individual neuronal cell bodies over time. 12-bit images were used (i.e. pixel intensity ranging from 0-4095). Subsequent data processing was performed using custom-written scripts in IgorPro (Wavemetrics). Individual traces were normalized to initial fluorescence (F/F) and averaged across samples. The line graphs show the average GCaMP6S fluorescence (thick line) +/− standard error of the mean (thin vertical lines) plotted versus time. The mean maximum change in GCaMP6S fluorescence (Max. ΔF/F) was calculated by averaging the peak F/F determined for each trace of a given sample. For each experiment, the sample sizes are indicated as n=x;y where x and y are the total number of cell bodies and brains quantified per sample, respectively. No statistical method was used to determine minimum required sample sizes, they were based on ref. [38] and also determined empirically. Generally, 8 brains were imaged for each data point over at least two experiments performed on different days. Additional brains were imaged when the results were variable from one day to another and all results pooled for analysis. We verified that the difference in Max. ΔF/F between samples was due to differences in absolute peak GCaMP6S intensity, not to differences in initial GCaMP6S intensity for each experiment.

Statistics were performed in IgorPro (Wavemetrics). Since samples often had unequal variances, we used the KS test to compare responses to drugs to avoid the problem of requiring normal distributions and equal variances. For baseline GCaMP6S experiments, we used Kruskal-Wallis ANOVA to determine if there was a significant rhythm in the data.

Low expression levels from Lk-Gal4 meant that we used flies homozygous for both Lk-Gal4 and UAS-GCaMP6S to image GCaMP6S in LHLKs, except for Pdf-Dti experiments. For the latter, we decreased the scan speed (1 z-stack per minute) to compensate for decreased signal intensity. To measure LHLK, LK-R and DH44 neuron excitability rhythms and baseline GCaMP6S levels in DD, flies were entrained to LD for at least 3 days and assayed on the first day in DD. Individual flies were maintained in DD and anesthetized on ice immediately before dissection under visible light. In total, brains were exposed to visible light for <5 min before measuring excitability and for between 5-10min before measuring baseline GCaMP6S.