As reported previously, we mutagenized newly isogenized iso31 flies with ethyl methane sulfonate (EMS), generated independent lines, and screened F3 generation flies under 12:12 light:dark cycles for sleep phenotypes (Shi et al., 2014). One line that reproducibly showed reduced sleep was named argus (aus: after the mythological Greek giant who never slept) and subjected to further analysis. aus homozygotes showed ~600 fewer minutes of total sleep per 24 hr day compared to iso31 controls (Ryder et al., 2004) controls (p < 0.0001; Figure 1A–B). aus sleep decrease was primarily driven by inability to sustain sleep, as reflected in a significant decrease in aus homozygote sleep bout duration during both day and night (p < 0.0001; Figure 1C), while sleep bout number was unchanged during the day and increased at night (p < 0.05; Figure 1D). Sleep latency analysis showed that aus mutants took a longer time to initiate sleep after lights off at ZT12 than aus heterozygotes or controls (p < 0.01; Figure 1E). Activity index, locomotor activity per waking minute, was comparable between aus mutants and controls (p > 0.05; Figure 1F), indicating that aus is not a hyperactive mutant. aus heterozygotes showed a small decrease in sleep relative to controls (p < 0.001; Figure 1A–B), indicating that the aus mutation is slightly dominant.

As the circadian clock regulates sleep timing, and some clock mutants show changes in total sleep (Shaw et al., 2002), we tested aus behavior under constant darkness for a potential circadian phenotype. Most aus homozygotes ( > 60%) showed robust locomotor activity rhythms, indicating an intact circadian clock (Figure 1—figure supplement 1A). Similarly to other short-sleeping mutants, the overt arrhythmia in ~30% of aus homozygotes likely stems from the large reduction in sleep (Figure 1—figure supplement 1B; Shi et al., 2014). Notably, aus homozygotes displayed longer activity episodes than controls in constant darkness (Figure 1—figure supplement 1A), consistent with their short sleep under LD conditions.

As the original screen selected for recessive mutations on the third chromosome (Shi et al., 2014), mapping of the aus mutation was initiated by crossing the mutants with a line carrying multiple genetic markers on the third chromosome (ru1 h1 Diap11 st1 cu1 sr1 es ca1). Recombinant progeny were screened for sleep phenotypes and subjected to classical mapping, localizing aus distal to ebony. We then developed single nucleotide polymorphism (SNP) markers through genomic DNA sequencing of aus mutants and the genetic marker line; the aus mutation was localized between SNP markers at ~19 and 24 Mb (Figure 2A). In parallel, we subjected genomic DNAs from aus homozygotes and iso31 controls to whole-genome sequencing. DNA sequences were aligned with the Drosophila Genome Project for SNP calling. While >500,000 polymorphic sites distinguished our stocks from the reference sequence, many SNPs were common to aus and the iso31 control; these were removed from further consideration (Figure 2B). In the ~5 Mb region identified by mapping, we found 622 aus-specific SNPs, of which 10 led to amino acid changes in nine open-reading frames (ORFs) (Figure 2B).

We focused on these ORFs, knocking each down in a pan-neuronal RNAi screen using elav-GAL4 driver, and identifying two candidates that produced sleep loss. One was Neurexin 1 (nrx1, cg7050), a synapse assembly molecule that regulates fruit fly sleep (Tong et al., 2016). We ruled this candidate out, as the nrx1 knockout showed no loss of sleep amount compared to control in our hands (Figure 2—figure supplement 1A) and it complemented aus sleep loss in transheterozygotes (Figure 2—figure supplement 1B).

The other candidate gene was cg16791, in which aus mutant flies have two GC→AT transitions that are predicted to translate to A→V and R→C amino acid substitutions. Supporting its identity as the aus locus, pan-neuronal knockdown of cg16791 with elav-Gal4> Dicer and cg16791 RNAi#1 produced a severe sleep reduction comparable to aus/aus mutants (p < 0.0001; Figure 2—figure supplement 2A,B). This also suggested the aus sleep phenotype is neural in origin. To rule out RNAi off-target effects and confirm this neuron-specificity, we assessed sleep behavior in cg16791 RNAi #1 and #2 crossed to Nsyb-Gal4> Dicer2 flies. Both pan-neuronal knockdown manipulations resulted in significant decreases in total sleep compared to RNAi and Nsyb-Gal4> Dicer2 controls (p < 0.01; Figure 2C–D). These results confirm neuron-dependence of the cg16791 RNAi sleep phenotype.

Our studies to this point did not address whether cg16791 acutely regulates adult sleep, or the development of sleep regulatory mechanisms. To address this, we crossed cg16791 RNAi#1 and #2 to Actin-GeneSwitch (GS)> Dicer2, and conducted sleep experiments in the presence of food supplemented with either the gene switch activating drug mifepristone / RU486 (RU+) or ethanol vehicle control (RU-). Actin-GS> Dicer2+ RNAi#2 flies showed a robust decrease in sleep compared to RNAi and Actin-GS> Dicer2 controls on RU+; however, there was no difference between genotypes on RU- (p < 0.0001; Figure 2—figure supplement 2C). Actin-GS> Dicer2+ RNAi#2 flies also showed a within-genotype reduction of sleep on RU+ vs RU- (p < 0.001; Figure 2—figure supplement 2C), while the control genotypes did not. This shows that cg16791 regulates sleep in adulthood. RNAi#1 caused a weak trend toward RU-dependent sleep loss that did not reach significance when crossed with Actin-GS> Dicer2, likely because of weaker knockdown (data not shown); note also that RNAi#2 is predicted to have higher specificity (Figure 2—figure supplement 2D).

Having putatively identified the aus locus, we took a bioinformatic approach to hypothesize probable structure and function of the largely uncharacterized CG16791 protein isoforms. An unbiased ProDom search of the full-length CG16791 isoform-A reference sequence identified a number of possible transmembrane motifs (Supplementary file 2). A more targeted TMPred assessment of known CG16791 isoforms predicted their best-fit membrane topology with a 5-transmembrane structure, and placed the aus mutations in an internal loop region between transmembrane helices 3 and 4 (Figure 2D; Supplementary file 2). This same loop contains a variable region that distinguishes the four known isoforms of CG16791. A Deep-Loc-1.0 analysis predicted that all CG16791 isoforms are targeted to the cell membrane, and perhaps to some extent the ER/Golgi network, driven predominantly by signal sequences in the C-terminus (Supplementary file 2; Almagro Armenteros et al., 2017). Our bioinformatic analyses are experimentally supported by the isolation of CG16791 isoform-A from membrane fractions of fly head (Aradska et al., 2015). Based on these analyses, we speculated that the aus substitutions in CG16791’s internal loop cause a loss-of-function that underlies its sleep loss phenotype.

To confirm that mutated CG16791 leads to the aus sleep phenotype, we performed additional mutant analysis, as well as rescue assays. First, we obtained a P-element insertion allele of cg16791 that breaks the open reading frame of the gene (hereafter, P1). The P1 allele failed to complement the aus mutant. Thus, P1/aus trans-heterozygotes had severely reduced total sleep comparable to aus homozygotes (p < 0.01; Figure 3A), supporting the idea that the cg16791 mutations in aus are causal for the sleep phenotype.

We then used CRISPR/Cas9 to generate a cg16791 knockout, in which the first exon containing the initiating methionine was replaced with a selectable marker, Dsred (Gratz et al., 2013; Figure 3—figure supplement 1A). As homozygous knockouts were semi-lethal (0.59% survival rate; 2 / 339 flies tested), we could only reliably obtain cg16791^KO heterozygotes. Southern blot analysis confirmed a single integration of DsRed at the aus locus (Figure 3—figure supplement 1B). Behavioral analysis showed that total sleep in cg16791^KO heterozygotes is comparable to aus heterozygotes, while trans-heterozygotes of cg16791^KO and aus showed a severe reduction in total sleep, similar to aus homozygotes (p < 0.0001; Figure 3B, Figure 3—figure supplement 1C). These cg16791^KO results further support our mapping of the aus EMS allele to loss-of-function of CG16791. However, the aus EMS allele maintains function required for survival, as EMS homozygotes are viable while knockout homozygotes are semi-lethal.

The gold standard to confirm that a specific mutation drives a mutant phenotype is through a rescue experiment. We cloned two cDNA forms of cg16791 under control of a UAS (Upstream Activation Sequence): a UAS-cg16791 short form beginning with the second methionine, which lacks 51 amino acids at the N-terminus, and a full-length UAS-cg16791^FL form. Pan-neuronal (elav-Gal4) induction of either form effectively rescued sleep in aus mutants (p < 0.0001; Figure 3C, Figure 3—figure supplement 1D), proving that mutations in cg16791 indeed cause aus sleep loss. This result also suggests that the CG16791^FL N-terminus is dispensable for its sleep function. For simplicity, only UAS-cg16791 was used for later experiments.

We next sought to identify the brain region through which aus regulates sleep. We first cloned ~2 kb of the aus promoter upstream of Gal4 and used the resulting fly line, ausP2k-Gal4, to drive expression of membrane-bound GFP. GFP was expressed broadly in the fly brain, in many cell types including peptidergic neurons of the pars intercerebralis (PI), Kenyon cells of the mushroom body, optic lobe neurons and some lateral neurons (Figure 4A). Importantly, P2k-driven aus expression rescued aus/aus short sleep, indicating that P2k-Gal4 recapitulates the sleep-relevant aus expression pattern (p < 0.05; Figure 4A).

Based on the prominent representation of neuropeptidergic populations labeled by the Aus2k driver, we suspected that aus functions in peptidergic pathways to regulate sleep/arousal behavior. To test this, we assayed for rescue using the peptidergic neuron specific c929-Gal4 driver, which is inserted near the dimmed gene, a bHLH transcription factor essential for neuroendocrine cell differentiation, and which appears to express in overlapping cell populations with Aus2k-Gal4 (Figure 4A–B, white boxes) (Hewes, 2003). c929-Gal4-driven aus expression in an aus mutant homozygous background partially rescued the short sleep phenotype (p < 0.05; Figure 4B), demonstrating that aus expression in peptidergic cells contributes to sleep behavior. Furthermore, transheterozygotes for dimmed (which have impaired neuropeptidergic neuron function, including in the PI) (Hewes, 2003) and aus show a synergistic loss of sleep compared to the respective single heterozygotes, suggesting that loss of neuropeptide signaling contributes to the aus sleep phenotype (p < 0.01; Figure 4C).

In investigating the mechanism by which aus regulates sleep, we noted that CG16791 was previously identified as a protein that interacts with the cell engulfment receptor Draper (Fullard and Baker, 2015). Draper is involved in cell death associated with autophagy, the primary disposal pathway for large-scale cellular waste such as protein aggregates and damaged organelles, and an emergency nutrient source (McPhee et al., 2010). Thus, we considered the possibility that AUS plays a role in waste disposal, such as autophagy, within cells. To determine if autophagy is regulated by AUS, we conducted live-imaging experiments in aus mutants and iso31 control flies pan-neuronally by expressing a GFP-mcherry-atg8a fusion protein driven by elav-Gal4 (Figure 5A). mCherry red fluorescence, but not GFP green fluorescence, persists under low pH; thus, autophagosomes retain both GFP and mCherry fluorescence, while acidified autolysosomes (autophagosomes that have fused with lysosomes to degrade their cargoes) selectively quench GFP, leaving only mCherry fluorescence (Mauvezin et al., 2014).

First, we validated a machine learning protocol for identifying neuronal Atg8(+) puncta by comparing autophagy flux in a small cohort of elav-Gal4> UAS-GFP-mCherry-atg8a brains dissected from ZT0-2 and incubated in either 2 uM rapamycin or ethanol vehicle in AHL for 2 hr prior to imaging. As expected, given its well-characterized role as a TOR inhibitor and inducer of starvation-dependent autophagy, rapamycin pre-treatment increased total mCherry(+) puncta compared to vehicle control, with no significant difference in either the size of these puncta or the ratio of mCherry+ GFP autophagosomes to mCherry-only autolysosomes (Figure 8—figure supplement 1A–D).

We then tested aus/aus flies pan-neuronally expressing the same sensor. Neither the number nor the size of all mCherry(+) puncta was significantly different in aus mutants compared to controls (p > 0.05; Figure 5B–C), but the distribution was significantly skewed toward double-labeled puncta with a reduction in the number of mCherry-alone puncta, suggesting that inefficient lysosomal digestion drives autophagosome accumulation in aus mutants (p < 0.001; Figure 5D). It is surprising that overall mCherry(+) puncta number is not increased by aus blockade of autophagosome degradation; the most likely explanation is a compensatory reduction of autophagy initiation upstream.

Through an independent project targeted toward identifying links between sleep and neurodegeneration, we discovered a sleep phenotype in the blue cheese 58 loss-of-function allele (bchs). The bchs mutant is best known for an autophagy defect that decreases the accumulation of autophagosomes and drives neurodegeneration (Finley et al., 2003; Simonsen et al., 2007). We found that bchs increases sleep compared to iso31 control (p < 0.01), primarily by lengthening night sleep bouts (p < 0.001) (Figure 6A, and Figure 6—figure supplement 1A–C). bchs also decreases latency to sleep at nightfall (p < 0.0001), suggesting that the sleep gain reflects increased sleep need (Figure 6—figure supplement 1D). Activity index was unaffected by either dosage of bchs, ruling out defective locomotion as a confound of the sleep gain phenotype (p > 0.05) (Figure 6—figure supplement 1E). Many bchs sleep phenotypes were recessive in females and dominant in males, suggesting sexual dimorphism.

Overall autophagy is impaired in both aus and bchs mutants, and yet they have opposing effects on sleep. However, we noticed that while the aus mutant increases the accumulation of autophagosomes, likely by blocking their clearance (Figure 5), the bchs mutant has been shown to decrease the accumulation of autophagosomes, by blocking the maturation of immature Atg5(+) autophagosomes and recruitment of Atg8 (Sim et al., 2019). We hypothesized that the opposite changes in the level of Atg8(+) autophagosomes in aus and bchs mutants drive their respective sleep phenotypes. If true, the bchs sleep phenotype should be epistatic to that of aus, since the bchs blockade of autophagosome maturation is expected to be upstream of aus autophagosome accumulation.

To test this, we generated transheterozygous bchs/+; aus/+ female flies and tested their sleep behavior compared to each single heterozygote. While a single allele of bchs had no effect on total sleep amount compared to control iso31 (Figure 6A), it robustly and non-additively suppressed sleep phenotypes of aus, rendering the bchs/+; aus/+ transheterozygotes statistically indistinguishable from bchs/+ for total, day, and night sleep amount, all of which were higher than aus/+ (p < 0.05; Figure 6B). Activity index was significantly higher in transheterozygotes and bchs/+ flies compared to aus/+ flies, suggesting an improvement in aus mobility with the addition of bchs that does not confound bchs suppression of aus short-sleep (Figure 6—figure supplement 1F). Our findings suggest that the bchs sleep phenotype is indeed epistatic to that of aus, consistent with their respective effects on the autophagy pathway. This epistatic relationship could not be assessed in trans-homozygous bchs/bchs; aus/aus flies because of a lethal interaction.

To confirm that aus short-sleep suppression by bchs is not due to Bchs roles in other cellular pathways including lysosome trafficking (Lim and Kraut, 2009), we assessed whether neuron-specific impairment of autophagosome maturation could rescue short-sleep in aus heterozygotes and homozygotes. Thus, we generated homozygous aus flies with elav-Gal4-driven pan-neuronal RNAi knockdown of atg5 or atg7, both of which are involved in autophagosome maturation and Atg8 recruitment/activation (Hale et al., 2013). Both pan-neuronal atg RNAis increased sleep on an aus mutant background (p < 0.05), driven selectively by increases in night sleep (Figure 6C–D). Importantly, neither RNAi increased sleep in flies lacking the aus mutation—in fact, atg7 RNAi decreased sleep in control flies—indicating that the rescue of aus did not result from an additive interaction (p < 0.001; Figure 6—figure supplement 2A,B). Impaired locomotion cannot explain either sleep rescue phenotype on the aus/aus background, as pan-neuronal atg5 RNAi flies had similar activity index to controls, while the activity index of pan-neuronal atg7 RNAi flies was intermediate between its controls (p < 0.05; Figure 6—figure supplement 2C,D). Much like the bchs mutant, atg5 and atg7 RNAi also rescue the milder sleep defect of aus/+ flies (p < 0.001; Figure 6—figure supplement 2E,F).

While the rescue of aus by neuronal knockdown of atg5 and atg7 RNAi implicated impaired autophagosome clearance as a mechanism underlying the short-sleep phenotype, we asked why these neuronal knockdowns did not produce a phenotype on their own. The bchs sleep gain could be driven by its roles in pathways aside from autophagy, so to rigorously test whether autophagy affects sleep, we conducted a targeted RNAi screen of genes with known links to various steps of autophagy for sleep behavior (Supplementary file 3, Tab 1). The use of drug-inducible geneswitch drivers allowed us to restrict manipulations to adulthood.

We first screened with pan-neuronal nsyb-geneswitch+ uas-dicer on RU+ food, and identified five RNAis for four genes that significantly increased sleep (p < 0.05; Figure 7A). These included upstream regulators that couple autophagy to starvation (Atg1, 2 RNAi’s), unfolded protein response (Bip), and ecdysone signaling (Daor) (Figure 7—figure supplement 1A,B,D,E; Hale et al., 2013). The remaining gain-of-sleep hit was Atg10, an E2 ligase-like enzyme involved in autophagosome vesicle expansion (Figure 7—figure supplement 1C; Hale et al., 2013).

To accomplish broad and strong knockdown of autophagy genes, we repeated the same screen with actin-geneswitch. As expected, this approach yielded more gain-of-sleep hits (P < 0.05; Figure 7B), including several autophagosome maturation proteins: two distinct RNAis encoding Atg8b (one of two ubiquitin-like homologs that label mature autophagosomes), and single RNAis for Atg12 and Atg7 (Figure 7—figure supplement 1H,I,J,K; Hale et al., 2013). Notably, the atg7 hit was the same allele used to rescue aus, suggesting that sleep loss from its knockdown with elav-Gal4 in the absence of aus reflects dosage and/or developmental compensation effects (Figure 6 and Figure 6—figure supplement 1B). Additional RNAi hits encoded proteins involved in autophagy initiation by multiple pathways (Aduk, Atf6, Daor, Dram, Wacky); autophagosome nucleation (Atg14, Dor); and facilitating autophagosome-autolysosome fusion (also Dor) (Figure 7—figure supplement 1F,G,L,M,N,O,P; Hale et al., 2013; Lindmo et al., 2006; Montagne, 2016). All of these hits consistently increased sleep, in the case of Dor likely because of an epistatic effect on autophagosome nucleation (Lindmo et al., 2006).

To validate these results, we back-crossed our highest confidence hits (atg1 RNAi’s #1,4 and atg8b RNAi’s #1,2) to iso31 and closely assessed sleep with crosses to nsybGS and actinGS (respectively). Both atg1 RNAi crosses increased total sleep (p < 0.0001; Figure 7C–D). The nsybGS> dcr,atg1 RNAi#1 increased sleep largely RU-dependently, while nsybGS> dcr,atg1 RNAi#4 increased sleep largely RU-independently, suggesting a leaky GS/RNAi combination (Figure 7C–D). Neither total mean bout length nor total bout number was significantly increased in either cross (Figure 7—figure supplement 2A,B,G,H). But both atg1 RNAi crosses had many flies with massive single night-time sleep bouts, and on RU+ food we found consistently longer night (p < 0.01) and longest (p < 0.0001) sleep bout lengths, with significantly lower night bout number (p < 0.05), suggesting that consolidation of night sleep drives overall sleep gain in these flies (Figure 7—figure supplement 2A,B,C,G,H,I). Sleep latency at nightfall was consistently decreased in both atg1 RNAi crosses on both foods (p < 0.01), with an even stronger decrease on RU+ vs RU- (p < 0.001; Figure 7—figure supplement 2D,J). Food-independent increased waking activity in atg1 knockdowns excludes the possibility that sleep increases are derived from sickness (p < 0.05; Figure 7—figure supplement 2E,K). Finally, qPCR quantification confirmed knockdown of atg1 by our RNAi alleles in actinGS> dcr,atg1 RNAi flies (p < 0.05; Figure 7—figure supplement 2F,L).

Both actinGS> dcr,atg8b RNAis robustly and RU-dependently increased night sleep, but only RNAi#2 increased total sleep and day sleep after back-crossing (Figure 7E–F). Both atg8b RNAis RU-dependently increased mean sleep bout length (p < 0.05), driven disproportionately by longest bout (p < 0.05; Figure 7—figure supplement 2M-O,S-U). Sleep latency at nightfall was marginally decreased on RU+ food compared to RU- in actinGS> dcr,atg8b RNAi flies, but not control genotypes (p < 0.05; Figure 7—figure supplement 2P,V). Neither atg8b RNAi cross had significantly different waking activity compared to both controls on either RU+ or RU- food (Figure 7—figure supplement 2Q,W). As qPCR did not consistently detect atg8b even in control fly extracts, suggesting very low expression, atg8b RNAis may effect their sleep gain by knockdown of atg8a through conserved sequences. This was supported by qPCR quantification of atg8a cDNA in actinGS> dcr,atg8b RNAi flies (p < 0.05; Figure 7—figure supplement 2R,X).

In sum, pan-neuronal atg1 and whole-fly atg8 knockdown phenotypes largely recapitulate the key hallmarks of bchs phenotypes: (1) sleep gain disproportionately driven by night sleep, (2) sleep consolidation, and (3) decreased sleep latency at nightfall (Figure 6 and Figure 6—figure supplement 1). This supports our attribution of the bchs sleep phenotype to its autophagy effects and, more generally, the sleep promoting effects of blocking autophagosome formation. RU-dependence of many phenotypes demonstrates that perturbing autophagosome formation in adulthood is sufficient to drive changes in sleep.

Our findings above indicated that autophagy, in particular autophagosome levels, regulate sleep amount. To determine whether sleep, in turn, regulates autophagosome accumulation, we first live-imaged neuronal autophagy flux in brains of flies carrying elav-Gal4> UAS-GFP-mCherry-atg8a at ZT0-2 and ZT12-14 (Figure 8A). In the early night, there were more total mCherry(+) puncta than in the early day, with no significant difference in the size of mCherry(+) puncta or the ratio of mCherry+ GFP autophagosomes to mCherry-only autolysosomes, suggesting a potential role for sleep:wake state in regulating the production of autophagosomes (p < 0.05; Figure 8B–D). However, this experiment left ambiguous whether sleep contributed to the observed day/night effect. To address this, we mechanically sleep-deprived (SD) flies of the same genotype overnight for at least 12 hr, and compared autophagy flux in SD vs control flies at ZT0-2 (Figure 8E). SD flies had significantly more total mCherry(+) puncta compared to controls, with no significant difference in the size of mCherry(+) puncta or the ratio of mCherry+ GFP autophagosomes to mCherry-only autolysosomes (Figure 8F–H).

To complement our SD data and mitigate possible confounding effects from the stress of mechanical perturbation, we also assayed effects of increased sleep on autophagy. We flipped flies onto food laced with either gaboxadol or water vehicle at ZT0-1, and compared ZT1-12 sleep and ZT12-14 autophagy flux in the same flies (Figure 8I). As previously reported, gaboxadol treatment markedly increased sleep (Figure 8—figure supplement 1E; Berry et al., 2015). Gaboxadol flies had significantly fewer total mCherry(+) puncta compared to controls, with no significant difference in either the size of mCherry(+) puncta or the ratio of mCherry+ GFP autophagosomes to mCherry-only autolysosomes (Figure 8J–L).

These data showing that wake increases and sleep decreases autophagosome number in wild-type fly neurons (Figure 8) were unexpected because they could be interpreted as sleep-promotion by autophagosomes, while our mutant and RNAi data indicate that high neuronal autophagosome number decreases sleep and low neuronal autophagosome number promotes sleep (Figures 1 and 5—7). As discussed below, we believe that the phenotypes of the mutants/RNAis reflect sustained high or low levels of autophagosomes not seen during a normal daily cycle (Figure 9).

The argus mutant line was obtained in a chemical mutagenesis screen as described previously (Shi et al., 2014). Several Drosophila lines used to interrogate the argus allele, including aus2k-Gal4, both cg16791 over-expression lines, and the cg16791 Crispr mutant, were developed by our laboratory (see below). Mapping stocks, insertion mutants, some RNAi, and Gal4 lines were acquired from the Bloomington Drosophila Stock Center at Indiana. Other RNAi lines were acquired from the Vienna Drosophila Resource Center in Austria or the Kyoto Stock Center in Japan. See Supplementary file 1, Tab1 and Supplementary file 3, Tab 1 for details including stock center ID, genetic background, and figure-by-figure breakdown of all Drosophila lines used in this manuscript.

Flies were housed individually in glass tubes in Percival incubators. Beam-break activity was recorded with the Trikinetics DAM system (http://www.trikinetics.com/). Pysolo (http://www.pysolo.net) and custom Matlab software were used to analyze and plot sleep patterns (Gilestro and Cirelli, 2009; Hsu et al., 2020). All flies were entrained prior to and maintained on a 12 hr:12 hr light:dark cycle for all behavior experiments, except where otherwise noted. Most behavior experiments examined behavior in flies that were ~3–5 days old at the start of recording for durations of up to 6 days, except where otherwise stated. Behavior experiments including homozygous aus groups examined sleep in flies of all groups that were ~3–7 days old at the start of recording for durations of up to 6 days. We expanded the acceptable age range for experiments including aus to allow us to maximize collections from a number of crosses with the aus allele that yielded few progeny.

Classical genetic mapping with phenotypic markers was conducted for the argus allele very similarly to how we previously isolated redeye (Shi et al., 2014). The minimal overlap narrowed down the location of argus to the region distal of ebony.

SNP mapping: Genomic DNA of homozygous recombinants was subject to SNP analysis. SNP19M and SNP24M primers (Supplementary file 1, Tab2) were used for PCR amplification, and identified nucleotide polymorphism between wild type and the marker line. Scoring of recombinant progeny for aus further narrowed the locus to a ~ 5 M bases region between SNPs.

Deep Sequencing: Illumina paired-end DNA library kit was used to make genomic DNA libraries of iso31 and aus homozygotes. The libraries were amplified ten times through PCR prior to Illumina Hi-Seq analyses. SNP calling algorithm identified polymorphisms.

Aus promoter Gal4 constructs: Aus2kGal4 primers were used to amplify the aus 2 kb promoter region from genomic DNA derived from iso31, and cloned into pBPGw (Addgene #17574). aus cDNA clones: UAS-aus primers were used to amplify a truncated aus CDS from an iso31 cDNA library and cloned into a pUAST-attB vector. UAS-aus^FL primers were used to amplify the full-length aus cDNA from an iso31 cDNA library and cloned into a pUAST-attB construct.

The PhiC31 integration system was adapted to target ausP-Gal4 constructs or UAS-aus(cDNA) constructs onto attP40 site on the 2^nd chromosome or attP2 site on the 3^rd chromosome.

Two gRNAs designed to generate a CG16791/aus knockout allele using the CRISPR/Cas9 system were cloned into pCFD4 (Addgene#49411) (Port et al., 2014). Separate primer sets were used to amplify and verify the target sequence. gRNA and primer sequences in Supplementary file 1, Tab 2. pHD-DsRed-attp-CG16791 vector: Approximately 1 kb upstream and downstream of the argus gene (CG16791) were PCR amplified using iso31 genomic DNA as a template. The 5’ CG16791 arm was PCR amplified. The PAM sequence CCG inside the 5’ arm was changed to GCG in the reverse primer (see underline) to prevent potential cutting by Cas9. The 3’ CG16791 arm was amplified by cloning primers, and the PAM sequence inside 3’ arm was changed from CCT to GCT using PAM elimination primers to prevent potential cutting by Cas9. PCR products of the 5’ and 3’ CG16791 arms were cloned into a SmaI site in the pBS-KS vector. After the construct was confirmed by sequencing with T7 and T3 primers, 5’ and 3’ arms were processed with AarI and SapI restriction enzymes, respectively, and inserted into AarI and SapI sites in pDsRed-attP (Addgene#51019). See Supplementary file 1, Tab2 for primer sequences.

The pCFD4-CG16791 gRNAx2 vector and pHD-DsRed-attp-CG16791 vector were mixed to final concentrations of 0.1 μg/μl and 0.5 μg/μl, respectively and injected into vas-Cas9 embryos by the Rainbow transgenic service. A single G0 male was crossed with Chr3 balancer virgin females to establish the line. Only G1 flies expressing DsRed in the eye were tested by extraction of gDNA followed by PCR. Further confirmation was done by southern blotting. The correct gene targeting lines were saved for testing in behavior assays. Knock-out flies (CG16791^KO) were back-crossed with the iso31 strain for several times and tested for behavior.

DNA Isolation: Flies (~15) were homogenized in DNA extraction buffer (100mMTris pH7.5; 100 mM EDTA; 100 mM NaCl; 0.5 % SDS). gDNAs were then isolated by sequential LiCl/KAc and isopropanol precipitations, and resuspended in TE for subsequent analysis.

Southern blot analysis: Roche Digoxin kit (Cat# 11093657910) was used to label DsRed DNA probes generated by PCR, using primers recorded in Supplementary file 1. Genomic DNA was digested with restriction enzymes and separated on 1% agarose gel before transfer to a nylon membrane. Digoxin labeled probe was hybridized with the membrane at 42°C overnight. After washing, the membrane was exposed with a chemi-luminescence reaction through anti-Digoxin conjugated alkaline phosphatase (Cat# 11093274910).

RNA: Adult fly heads (~15) were subject to Trizol extraction (Ambion). High-capacity cDNA reverse transcription kits (Applied Biosystems) were used to make cDNA libraries.

Autophagy RNAi Screen for Sleep Behavior actinGS+ dicer and nsybGS+ dicer were separately crossed to RNAi’s for genes with known roles in autophagy (Supplementary file 3). We initially measured total sleep in up to 16 female flies on 5% sucrose-agar food laced with 500 uM Sigma-Aldrich mifepristone / RU486 (Cat#: M8046) in ethanol vehicle (RU+ food), averaging sleep across days 4–5 of exposure to drug. Crosses with a median total sleep two hours or more higher or lower than both GS+ dicer and RNAi controls were considered possible hits (primary criterion; Supplementary file 3). In cases where different RNAi’s for the same gene gave initial hits of opposite direction, we excluded both to avoid probable RNAi off-target effects (secondary criterion; Supplementary file 3). Finally, remaining possible hits were re-run a second time, measuring sleep under the same conditions as the initial screen. Crosses statistically different in the same direction from both controls in the combined runs (tertiary criterion) were considered RNAi hits (Figure 7—figure supplement 1).

For individual genes with multiple consistent RNAi hits (atg1 and atg8b), we backcrossed five generations to iso31, then ran a more detailed analysis of sleep in females on both RU+ and ethanol vehicle laced (RU-) food, using the same crosses that gave hits in our screen. To confirm knockdown of target transcripts, we also crossed these alleles to actinGS+ dicer and harvested RNA from pools of 5 RU+ fed whole female flies with a Qiagen RNeasy Miniprep Plus kit (Cat# 74134). gDNA was removed by both included eliminator columns, and on-column Qiagen RNase-free DNAse treatment (Cat# 79254). RNA was reverse transcribed with Lifetech Superscript II Reverse Transcriptase (Cat# 18064071). cDNAs for putative RNAi target genes and alpha-tubulin were amplified using Lifetech SYBR Green PCR mix (Cat# 4364346) and primers in Supplementary file 1 on an Applied Biosystems ViiA7 qPCR machine. We calculated relative transcript levels by ddCT.

Brains from approximately 1 week old adult female flies singly housed on our lab’s standard yeast-molasses food were dissected and mounted in chilled artificial hemolymph (108 mM NaCl; 5 mM KCl; 2 mM CaCl₂; 8.2 mM MgCl₂-6H₂O; 4 mM NaHCO₃; 1 mM NaH₂PO4-H₂O; 5 mM trehalose; 10 mM sucrose; 5 mM HEPES; 265mOsm and pH7.5) (Cohn et al., 2015). They were live imaged embedded in vacuum grease with a 40 X water immersion objective at 1.3 X digital zoom under a Leica confocal microscope at Alexa488 (green) and Alexa594 (red) wavelengths. Z-stacks containing ~60 μm of the central brain starting from the tips of the antennal lobes were captured.

Ilastik machine learning software was trained to isolate all mCherry(+) puncta from our Z-stacks (Berg et al., 2019). Briefly, for each experiment an equal number of representative brains from each group were marked for signal and noise in the red channel by a human scorer to train the Ilastik algorithm. Slices from the front, middle, and back of each stack were used, taking care to mark a range of diverse examples of signal and background. A similar number of markings were made between groups, to avoid biasing the algorithm. Ilastik’s prediction of signal and background was then reversibly overlaid on unmarked sample sections and visually inspected for accuracy by the human scorer. Once the algorithm passed inspection, simple segmentations of all brains were generated by Ilastik to define puncta and background for input into ImageJ. ImageJ was then used to measure mCherry(+) puncta count and size, and each mCherrry puncta’s green channel intensity from GFP. We then thresholded to background GFP intensity within each brain, and counted mCherry(+) puncta with green fluorescence intensity exceeding background to determine autophagosome and autolysosome percentages. The Ilastik algorithm was validated by quantifying autophagy following treatment with the autophagy-inducer rapamycin (see below).

Brains for Ilastik validation were incubated in artificial hemolymph supplemented with either 2 μM LC Laboratories rapamycin (Cat#: R5000) in ethanol vehicle, or ethanol vehicle alone, for ~2 hours prior to imaging. The drug condition was maintained for each group throughout imaging.

For sleep deprivation, flies were placed in DAM monitors in locomotor tubes filled with fresh yeast-molasses food on top of a mechanical deprivator. During the night preceding imaging, flies were shaken for a period of 2 s every 20 s to disrupt sleep, as previously described (Toda et al., 2019).

For sleep induction, flies were flipped from regular yeast-molasses food onto yeast-molasses food supplemented with either Sigma-Aldrich gaboxadol hydrochloride (Cat#: T101) in water vehicle diluted to 0.1 mg/mL final concentration, or water vehicle alone, during ZT0-1. Flies were maintained on the supplemented food for ~12 hr before imaging from ZT12-14. Sleep was recorded for at ~11 hr after flip onto drugged food, to verify that we observed gaboxadol-induced sleep gain as previously described in the same flies whose brains were imaged (Berry et al., 2015; Dissel et al., 2015).

Statistics were run in GraphPad Prism or JMP software. Shapiro-Wilkes tests were used to assess normality of each group for each individual experiment. Multiple-comparison correction was appropriately applied where multiple comparisons tested multiple hypotheses, but not where multiple comparisons were made to test a single hypothesis, as in non-geneswitch Gal4 driven RNAi and rescue experiments conducted in the manuscript (Shaffer, 1995).