Dopamine and GPCR-mediated modulation of DN1 clock neurons gates the circadian timing of sleep

Recent work implicates clock neuron interactions as essential for molecular and behavioral rhythms. For example, manipulating clock neuron subpopulations can affect behavioral timing, and residual clock protein expression in only a few circadian neurons is sufficient to retain some circadian functions (11, 23, 29–31). To identify molecules within the clock network that contribute to the implied network synchrony, we isolated and sequenced fluorescence-activated cell sorting (FACS)-sorted clock neurons (clk856 > EGFP) under LD conditions at two times, namely, Zeitgeber time 2 (ZT2) and ZT14, and compared the transcriptomes to those from panneuronal adult head samples (nSyb > EGFP).

We first compared clock gene expression between the time points in the clock neurons. Consistent with expectation (32), clk mRNA levels were ∼6× higher at ZT2 than at ZT14, whereas tim mRNA levels were 22× higher at ZT14 compared to ZT2 (Fig. 1A). Housekeeping genes such as Act5C and Rpl32 were not different between time points (Fig. 1A). Not surprisingly, clock gene expression was dramatically lower in the nSyb dataset (averaged reads of all nSyb libraries show 1,428× reduced clk and a 27× reduced tim expression compared to sorted clk856-neurons; SI Appendix, Fig. S1↗), demonstrating highly efficient enrichment of clock neurons by clk856 > EGFP purification.

To address clock neuron gene enrichment more generally, we used edgeR (33) to perform differential gene expression analysis. It identified a total of 1,719 genes as differentially expressed with a false discovery rate of <0.05 between nSyb and clock neurons. Of these, 704 genes were significantly up-regulated, and 305 genes were significantly down-regulated in the clock network with at least a twofold change in amplitude (Fig. 1B). A Gene Ontology (GO)-term analysis of clock-enriched genes resulted in enrichment in expected categories like the circadian regulation of temperature homeostasis (27-fold), positive regulation of circadian sleep/wake cycle (16-fold), and circadian regulation of gene expression (11-fold). Interestingly, genes associated with signaling pathways were also found to be comparably enriched among the clock-enriched genes and include the octopamine and tyramine signaling pathways (16-fold), the serotonin receptor signaling pathway (16-fold), and the GPCR signaling pathway (12-fold). Other unrelated pathways are enriched in the nSyb neurons relative to the clock neurons (for an example, see Fig. 1C). The increased expression of genes involved in intercellular signaling pathways, including serotonin signaling and GPCRs, strongly implicate neuronal communication in clock network function.

The enrichment of GPCR signaling pathways inspired a focus on quantitative features of GPCR expression. Surprisingly, more than two-thirds of the 124 GPCR mRNAs encoded by the Drosophila genome are expressed in the clock neurons (reads per million [RPM], >3; Fig. 1D) with an almost 100× difference between the least and most expressed of these 86 mRNAs. The nSyb data had similar differences in GPCR expression levels (SI Appendix, Fig. S2↗). A direct comparison with the nSyb dataset indicates that 22 GPCRs are at least twofold up-regulated in the clock network, whereas only 2 receptors are down-regulated (Fig. 1E).

The RNA expression data suggest that many individual GPCRs are either expressed at significantly higher levels in all clock neurons or are predominantly expressed in a few neuronal subpopulations. To address these alternatives, we analyzed previous scRNAseq data for GPCR expression and focused on the 17 high-confidence clock neuron clusters (Fig. 2A); these clusters are missing most of the enigmatic DN3 clock neurons but include most if not all well-characterized clock neurons including all lateral and most dorsal clock neurons.

The dopamine receptor DopEcR transcript is highly expressed in the nSyb as well as the clk856 dataset with high transcript levels in all clock neuron clusters (Fig. 2B). The differential expression patterns of more poorly expressed GPCR transcripts within the clock neuron population were similarly impressive. For example, CNMaR appears to be almost exclusively expressed in one DN1p and the DN2 cluster, whereas Gaba-B-R3 is highly enriched in the sLNv cluster. sNPFR is expressed more broadly but still mostly in the clusters defining the DN1ps, the lLNvs, and the DN3s (Fig. 2B). A differential GPCR expression pattern is also apparent when comparing all GPCR expression using scRNAseq despite dramatic variation in expression levels (SI Appendix, Fig. S3↗).

To confirm this cell specificity in an independent way, we used GAL4 lines in which the GAL4 sequence was integrated into endogenous GPCR-expressing loci and used to express nuclear GFP (UAS-stinger) (34). The overall expression levels of individual GPCRs correlated nicely with the nSyb sequencing data (SI Appendix, Fig. S2↗). For example, some GPCRs such as the aforementioned DopEcR appear to be expressed almost panneuronally within the brain, whereas others are not expressed in the brain or only in one to two cells per hemisphere (SI Appendix, Fig. S2↗).

To detect the expression of specific GPCRs within the clock neuron network, we focused on GPCRs with striking expression patterns from our single-cell data and determined the overlap of PER and GFP by costaining knock-in lines with anti-PER. The single-cell data indicate that DH31-R is highly enriched in the lLNvs and slightly enriched in the sLNvs and the DN1ps. The GAL4-knock-in line is strikingly consistent with this pattern; three out of four lLNvs were GFP-positive, whereas only one out of four sLNv neurons was labeled, explaining the differences in expression levels. As predicted, no LNds were labeled, but we could also not detect GFP expression in the DN1 neurons in most brains. Similar cell type specificity was also observed with sNPFR, which mainly labeled DN1 neurons and one out of four lLNvs (Fig. 2C). The data taken together suggest that GPCRs are strongly differentially expressed within the clock network. Notably, differences appear even between individual cells within an anatomically mostly uniform cluster like the lLNvs (SI Appendix, Fig. S4↗).

Might these GPCR expression differences be sufficient to define clock cell identity within the clock network? We originally identified 17 high-confidence clock neuron clusters based on common highly variable genes across different time points and conditions (27). Here, we first used Seurat, an unsupervised clustering method, and reproduced the previously published 17 high-confidence clock neuron clusters (Fig. 2D). We then used GPCR expression alone for clustering, omitting the contribution of all other genes expressed in the individual cells. This unsupervised GPCR-based clustering also identified 17 distinct clusters (Fig. 2D). To our surprise, many of these newly generated clusters could be mapped onto the previously published dataset (27), indicating that the GPCR-generated clusters are of biological and anatomical significance. For example, the newly identified cluster 1 includes ∼90% of this previously identified sLNv cluster. This is also the case for other clock clusters like the DN1s (Fig. 2D). The data therefore indicate that GPCR expression alone is sufficient to define the identify of most clock neurons and also suggest that each cluster may exhibit a unique neuropeptide response pattern.

The likely contribution of GPCRs to clock neuron identity and the neuropeptide requirement for clock neuron synchrony inspired the development of a general strategy to eliminate any GPCR in a neuron-specific manner. Relevant to this goal, we and others recently showed that CRISPR-Cas9-based mutagenesis is superior to RNA interference (RNAi)-mediated gene expression knockdown in the fly brain (29, 30). As the former also does not require strongly expressing driver lines, it is much more amenable to highly neuron-specific split-GAL4 lines.

The Drosophila genome encodes 124 GPCRs. They are like mammalian GPCRs and react to a variety of stimuli, including biogenic amines, neurotransmitters, neuropeptides, and even light (Fig. 3A) (4). To mutate these receptors, we generated UAS-guide lines, each of which expresses three guides targeting the coding sequence of an individual GPCR. Three guides have previously been shown to efficiently mutate eye tissue and provide high mutagenesis efficiency by compensating for potential noncutting guides (28); this strategy also worked well to remove PER and TIM from the clock system (29, 30). We used clk856-GAL4 to drive the expression of GFP, Cas9, and the guides of interest to mutate individual GPCRs in most of the clock network. The goal was to generate double-strand breaks within the coding sequences of GPCR genes. Given that repair is error prone, this strategy should generate small deletions of variable sizes and frame shifts, resulting in nonfunctional GPCRs.

As there are no reliable antibodies for many of the receptors, we verified the strategy with a targeted genomic sequencing approach. As the clock neurons were simultaneously labeled with GFP as well as mutated by the CRISPR-Cas9 system, we FACS sorted and analyzed 2,000 GFP-positive cells, which should have been mutated, and 2,000 GFP-negative cells, which should remain wild type (Fig. 3B). We designed three sets of primers flanking each of the guide binding sites to allow for gene-specific amplification. As three guides are being used at the same time, there can either be small deletions in the area of guide binding (Fig. 3C, PCR1 to 3) or larger deletions if multiple guides cut at the same time, resulting in different DNA fragment combinations (Fig. 3C, PCR4 to 6). We analyzed five randomly chosen target genes on three different chromosomes (PDFR and Tre1 on the X-chromosome, mAchR-A and CG15614 on the second chromosome, and CrzR on the third chromosome) to avoid possible biases from chromosome location. PDFR served as a positive control; guide-mediated mutagenesis of the clock network with its guides completely reproduced PDFR full body mutant (han⁵³⁰⁴) phenotypes (35, 36).

All three PDFR guides generated deletions of variable sizes at the predicted cut sites in GFP-positive cell DNA (Fig. 3D). Moreover, there were large deletions as indicated by a genomic fragment representing a deletion of several thousand base pairs (Fig. 3C, PCR5). GFP-negative cells in contrast showed no deletions in the investigated area, suggesting that there is no background mutagenesis due to leaky expression in nontarget cells. We then assayed CG15614. Like for PDFR, all three guides generated deletions, whereas GFP-negative cells were unaffected (Fig. 3C). Tre1 and CrzR had similar results, whereas only two out of the three guides for mAchR-A created deletions (SI Appendix, Fig. S5↗).

The efficacy of the three individual guide sequences to generate small deletions varied substantially, from between 0.7% (as mentioned above for one guide of mAchR-A) to more than 50%, with no evident chromosome or location bias. When an individual guide failed to generate mutations, the other two guides efficiently mutated the gene of interest; this indicates the importance of using several guides to compensate for potential noncutters. There was also a reduced frequency of bigger deletions of coding sequences. For example, two genes (CrzR and mAchR-A) did not show big deletions, whereas 2% to 6% of the reads from other genes reflected big deletions. It is important to note that these percentage are based on the sequencing of pooled neurons from several animals and therefore do not reflect a single, mutated neuron. Nonetheless, the collective data support the original assertion based on eye color essays (28), namely, that combining three guides per gene is an effective strategy to manipulate genes of interest.

To exploit this functional library, we focused on the DN1ps. This specific group of dorsal neurons influences several aspects of Drosophila behavior. This is because manipulating these cells affects activity in the morning and during the siesta as well as in the evening. In addition, DN1ps affect sleep and connect to sleep centers within the central complex (24–26). Recent work also showed that this group of neurons can be subdivided into six independent clusters with functions that are not yet clearly understood. The molecules that affect sleep within these neurons are also mostly unknown. To address this question, we used clk4.1M-GAL4, which expresses in 8 to 12 of the 15 DN1ps per hemisphere. We first reproduced previous experiments; activating these neurons significantly altered sleep in the middle of the day, the prominent siesta of male flies (SI Appendix, Fig. S6↗).

To identify candidate DN1p GPCRs, we turned to our single-cell data and identified 21 GPCRs enriched in DN1ps compared to the other clock cells. We then performed a behavioral screen in which we compared the behavior of control flies with flies harboring mutated GPCRs in their DN1ps. The control flies expressed Cas9 but no guides in these neurons (clk4.1M > Cas9).

As expected, Cas9 expression in the DN1ps did not affect fly behavior; they showed the canonical sleep pattern with consolidated sleep at night and during the siesta with almost no sleep in the morning and the evening; this reflects the standard bimodal activity pattern (Fig. 3E). We then focused on the siesta and quantified sleep levels between ZT3 and ZT9. Male flies sleep extensively during this time, leading to a median of ∼5.5 h of siesta sleep in control flies. Of the 21 mutated strains, 4 significantly reduced their siesta sleep compared to the control group by one-way ANOVA followed by a post hoc Tukey test (Fig. 3F). One GPCR is rh7, which reproduces known whole-body mutant phenotypes (37). Another is TrissinR, which reduced sleep by half an hour. Its relevant ligand, Trissin, is expressed in two LNds, suggesting that intraclock neuron communication, LNd to DN1p, is relevant to sleep regulation (27). Eliminating two different dopamine receptors, namely, Dop1R1 and Dop1R2, also reduced siesta sleep, each by approximately half an hour.

The sleep reduction by mutating Dop1R1 and Dop1R2 in the DN1ps was surprising, as dopamine is traditionally activity promoting (38–40). We therefore addressed this sleep effect in more detail by dividing the 24-h day into four equal sections of 6 h each, as follows: ZT21 to 3 for morning sleep, ZT3 to 9 for siesta sleep, ZT9 to 15 for evening sleep, and ZT15 to 21 for nighttime sleep (Fig. 4A). As expected, flies slept more during the siesta and at night compared to morning and evening (Fig. 4B). Quantifying sleep with Dop1R2 mutated in DN1ps showed that sleep was only significantly reduced during the siesta in the experimental flies (P < 0.01) compared to the Cas9 (clk4.1M > Cas9) and guide (clk4.1M > Dop1R2-g) control flies. Mutating Dop1R1 in the DN1ps had very similar effects (SI Appendix, Fig. S7↗).

Will mutating both Dop1R1 and Dop1R2 at the same time have an even stronger effect on sleep? Although siesta sleep is still significantly reduced compared to both control strains with no effect at any other time of day, we did not observe any additive effect; the double-mutated strain had a similar effect compared to the single mutant strains. (Fig. 4C and D).

To provide further support for this approach, we expressed Cas9 and guides against Dop1R2 in the dorsal fan-shaped body and reproduced the increased siesta sleep phenotype observed by Pimentel et al. (41) using the same dFSB driver 230000000000 and RNAi knockdown of Dop1R2 (SI Appendix, Fig. S8↗). The results taken together indicate that the target receptors and cells dictate the effect of dopamine, which can promote sleep as well as wake.

The sleep patterns indicate that the siesta is terminated earlier in flies with mutated dopamine receptors compared to controls, suggesting that dopaminergic input contributes to timing the end of the siesta. Notable in this context is the traditional per^S mutant strain; it has a very short free-running period with a similar LD phenotype, e.g., the evening peak occurs during the daytime (41, 42). Yet there was no effect of removing both dopamine receptors from the DN1ps on period length or rhythmicity; experimental groups as well as controls were identical to wild-type flies (Fig. 4E and F), suggesting that changes in clock speed are not responsible for the sleep phenotypes. The data indicate that dopaminergic input onto DN1p neurons enhances sleep during the siesta, likely caused at least in part by delaying the onset of evening activity.

How might dopaminergic modulation of DN1ps impact the siesta? To address this question, we first used the trans-Tango technique to label target neurons through its anterograde transsynaptic circuit tracing (43). We applied this technique to the dopaminergic system (TH-GAL4) and labeled downstream neurons with GFP. Because there were many GFP-positive neurons, we costained with anti-PER and investigated colocalization of GFP and PER immunoreactivity. The dopaminergic system appears to contact a variety of clock cells including PDF cells, 3 out of 6 LNds, 2 of 2 DN2s and on average 4 of 15 DN1ps, confirming that the DN1s are indeed among the direct downstream partners of TH cells (SI Appendix, Fig. S9↗).

Examining the 6 DN1p RNA expression clusters in more detail indicated that Dop1R1 and Dop1R2 are primarily expressed in 4 clusters, numbers 6, 7, 15, and 18 (Fig. 5A). Intriguingly, four is identical to the number of tyrosine hydroxylase neurons targeting DN1ps. Importantly, the size of these four clusters is rather small, indicating that several of them may only contain a single cell (27). We also considered the two remaining DN1p clusters, namely, clusters 1 and 4. The biggest cluster (#1) is the only one that expresses AstC. We showed previously that AstC is expressed in four DN1 neurons, which fits well with the size of this cluster (10, 27). Knockdown of AstC affected the timing of the evening peak in summer or winter days, suggesting that these neurons also contribute to evening activity, at least as a function of different seasons (10). In addition, this is the only cluster expressing TrissinR, which also produced a siesta phenotype in our behavioral screen (SI Appendix, Fig. S7↗). The data taken together indicate that several and perhaps most DN1 clusters contribute to evening activity.

We exploited the striking difference in Vglut expression between morning and evening DN1p cells to further address how the DN1ps regulate evening activity (Fig. 5A). Combining a per-AD with Vglut-DBD labeled on average seven to eight DN1s per hemisphere, consistent with the expected number of neurons in these clusters (SI Appendix, Fig. S10↗). Trans-Tango experiments with this split-GAL4 revealed that these seven to eight DN1p neurons primarily target neurons in the dorsal part of the brain (Fig. 5B). They include DN1s, DN2s, and DN3s, and this split-GAL4 also targets all LNds in the lateral part of the brain; they are a major controller of evening activity. These data reinforce previous results (25) indicating that the glutamatergic subset of DN1ps likely controls evening activity and sleep at least in part through their interactions with other clock neurons. Importantly, the upstream influence of dopamine on these cells and their functions widens the influence of the environment and brain state over the siesta and evening cell timing (see Discussion).

Flies used in this study are listed in SI Appendix, Table S1↗. All flies were raised at 25 °C in a temperature-controlled incubator in LD 12:12 h.

To generate UAS-guideRNA flies, we used the pCFD6 vector (addgene #73915, described in ref. 28). In short, we generated three guides targeting the coding sequence of each GPCR. To identify possible target sites and avoid off-target effects, we used the optimal target finder developed by C. Dustin Rubinstein, Ed O'Connor-Giles, and Kate M. O'Connor-Giles (53). Gene-specific guide sequences were then incorporated into the primers (SI Appendix, Table S2↗), and the protocol described in ref. 28 was followed. Correct clones were identified by colony PCR followed by Sanger sequencing. Plasmids were injected into the attP1 site on the second chromosome (Bloomington Drosophila Stock Center [BDSC]: 8621) by Rainbow Transgenic (Rainbow transgenic flies Inc.). Individual flies were crossed to w¹¹¹⁸ (BDSC: 3605) and screened for red eye color. Red eyed flies were balanced using w;CyO/Sco;MKRS/TM6B (BDSC: 3703).

Neurons of interest were genetically labeled with eGFP using nSyb-GAL4 (for all neurons) or clk856-GAL4 (for clock neurons). About 2-wk-old male flies were entrained to the appropriate LD cycles. Brains of flies of the appropriate genotypes were dissected at ZT02 and ZT14 in ice-cold Schneider’s medium (SM) and stored on ice until all dissections were completed (about 10 for nSyb and 30 for clk856). Brains were incubated with an enzyme solution (SM with 0.75ug/ul collagenase and 0.04 μg/ul dispase) at room temperature (RT) for 30 mins and then washed with SM, and triturated in SM. The volume was then brought up to ∼1 mL and filtered through a 40-μm filter fitted on a FACS tube. GFP+ neurons of interest were isolated using the FACS Melody instrument (BD) with sorting gates set by comparing to a GFP-negative neuron sample. A total of 500 clock neurons or 1,000 nSyb neurons were collected per sample in 100 µL lysis buffer (Dynabeads mRNA direct kit) and frozen on dry ice immediately after collection. Two sets of neurons were collected from each dissociated sample.

PolyA mRNA was isolated from the frozen cell samples using the Dynabeads mRNA direct kit (Thermo fisher 61011). Subsequently, complementary DNA (cDNA) was prepared using the method described in Picelli et al. (54). cDNA integrity and concentration were assessed using a High Sensitivity D5000 ScreenTape (Agilent 5067-5592). A total of ∼500 pg of cDNA was used as the input to make sequencing libraries with the Illumina Nextera XT DNA Library Preparation Kit (FC-131-1096) with nine PCR cycles. Final libraries were quantified on a High Sensitivity D1000 ScreenTape on the TapeStation (5067-5584).

Libraries were run on the Illumina NextSeq 550 sequencing system. Reads were aligned to the dm6 version of the Drosophila genome using STAR (55). PCR duplicates were removed using Picard Tools (Picard Toolkit 2019. Broad Institute, GitHub Repository, https://broadinstitute.github.io/picard/↗; Broad Institute). Differential expression analysis between nSyb neurons and clock neurons was performed using the Bioconductor package edgeR (33). Raw data were submitted to the Gene Expression Omnibus (GEO) repository and are available using the accession number GSE202407↗.

For details on scRNAseq procedures, refer to Ma et al. (27). In order to identify differentially expressed GPCRs in clock neurons, we first computed all marker genes in each cluster using the FindAllMarkers function of the Seurat package. Using a negative binomial generalized linear model, the batch effect from sequencing depth and conditions was regressed out. We next used an adjusted P value significance of 0.05 and fold change cutoff of 1.25. GPCRs matching these criteria are regarded as differentially expressed in clock neurons. Their expression was plotted by the ComplexHeatmap package.

Annotated single-cell clustering data were used as the basis for GPCR-based reclustering using Seurat V4 in R (56). For the downstream analysis, only cells from the 17 high-confidence annotated clusters were used. GPCR-based reclustering was done by restricting the variable features to GPCRs by setting the features argument of the ScaleData Seurat function to all genes identified as GPCRs from Hanlon and Andrew (4) that were detectable in all timepoints. For the standard clustering, the FindVariableFeatures function with default settings was used to determine variable genes.

For both approaches, principal components (PCs) were determined using the RunPCA function, and the first 20 PCs were selected for clustering based on visual inspection of the ElbowPlots. Communities were generated using the standard workflow functions FindNeighbors, FindClusters, and RunUMAP with default settings. The resolution argument of the FindClusters argument was set at the default of 0.8 after experimentation with resolutions as low as 0.5. Higher values for resolution and increased numbers of PCs did not improve clustering results separately or in combination for the GPCR clustering.

The cell cluster identities published in Ma et al. (27) were stored in the Seurat metadata and used to assess the correspondence between the annotated clusters and the GPCR-based clustering. In general, the GPCR clusters correspond to a single annotated cluster and vice versa. There are two exceptions, as follows: the fusion of the two LNd clusters 9 and 12 into a single cluster and the fission of the DN1p cluster 4 into two similarly sized clusters. The split of the DN1p cluster is characterized by a nearly twofold difference in average levels of FMRFaR, which is expressed in subsets of DN1p neurons.

To analyze the potency of our guide library, we established a targeted genomic sequencing approach similar to the 16S metagenomic sequencing library preparation protocol (Illumina 15044223). In short, we generated three pairs of primers for each gene (SI Appendix, Table S3↗). Each pair of primers was designed to amplify 230- to 270-bp-long genomic regions centered around the predicted guide cut sites. To analyze the potency of our guide library to mutate genes in a cell-specific manner, we labeled the clock network with GFP (clk856 > EGFP) and expressed Cas9 along with the guide of interest. We FACS sorted (see above) 2,000 GFP+ and 2,000 GFP− cells and extracted DNA using a DNA extraction buffer. We then used this extract and performed a PCR with all six primers for a gene of interest using an ExTaq polymerase. After Dynabeads mediated cleanup, adapters were ligated followed by another round of cleanup. Final libraries were quantified on a High Sensitivity D1000 ScreenTape on the TapeStation (5067-5584). Libraries were then pooled and sequenced using the miseq platform (Genewiz Inc.). Libraries were analyzed using Crispresso2 (57).

Two- to 7-d-old male flies were individually placed into glass tubes with food (2% agar, 4% sucrose) on one end and a plug to close the tube on the other end. These tubes were subsequently placed into Drosophila Activity Monitors (DAMs; Trikinetics Inc.), and a computer recorded the number of infrared light beam interruptions in 1-min intervals. Flies were recorded for 1 wk in LD 12:12 followed by DD for another week. Each experiment was performed at least twice.

To allow for proper entrainment to the LD cycle, we used the last 4 d of LD to analyze sleep, defined as 5 min of inactivity (58, 59). We then created average sleep profiles by averaging the amount of sleep across days and flies for each genotype in half-hour bins. To analyze sleep in more detail, we split the sleep amount into four sections of 6 h each, as follows: morning (ZT21 to ZT3), siesta (ZT3 and ZT9), evening (ZT9 and ZT15), and night (ZT15 to ZT21). Individual values were plotted as scatter plots, and statistical analysis was performed using a one-way ANOVA followed by a post hoc Tukey test (http://www.astatsa.com↗). P < 0.05 compared to all control groups was considered significant.

To analyze changes in rhythmicity, we performed a χ² analysis for rhythmicity and analyzed the speed of the clock. Individual values were compared using a one-way ANOVA followed by a post hoc Tukey test (astatsa.com); P < 0.05 compared to all control groups was considered statistically significant.

Two- to 7-d-old male flies were fixed for 2 h 45min in 4% PFA in PBST (phosphate buffered saline including 0.5% TritonX). After rinsing 5× for 10 min each in PBST, brains were dissected and blocked for 2 h in 5% normal goat serum in PBST. Primary antibodies were applied overnight at RT. Primary antibodies were the following: chicken anti-GFP (1:1,500, abcam), rabbit anti-PER (1:1,000; 60), mouse anti-PDF (1:1,000, DSHB), and rat anti-RFP (1:500, chromotec). After rinsing 5× with PBST, secondary antibodies (1:200, Alexa Fluor, Fisher Scientific) were applied for 2 h at RT followed by rinsing 5× for 10 min each with PBST. Brains were subsequently mounted on glass slides using Vectashield mounting medium (Vector laboratories) and imaged using the Leica SP5 confocal microscope. For PER quantification we used a 3×3-pixel area to determine the staining intensity of neurons and corrected by measuring three different background intensities as previously described (61).

The trans-Tango technique allows for anterograde transsynaptic tracing (43). UAS-RFP, transTango; QUAS-GCaMP flies were crossed to either TH-GAL4 (SI Appendix, Fig. S9↗) or per-AD Vglut-DBD split GAL4 (Fig. 5B) flies to identify the respective downstream targets. The upstream neurons (TH-GAL4 or per-AD Vglut-DBD) were labeled with RFP, whereas their synaptic downstream partners were labeled with GCaMP. Two- to 7-d-old male flies were stained using anti-GFP (to visualize GCaMP) and anti-period (to identify clock neurons) using the immunostaining protocol described above.

RNAseq data have been deposited in GEO (GSE202407↗) (62).