Abstract
Sleep is a fundamental state of behavioral quiescence and physiological restoration. Sleep is controlled by environmental conditions, indicating a complex regulation of sleep by multiple processes. Our knowledge of the genes and mechanisms that control sleep during various conditions is, however, still incomplete. In Caenorhabditis elegans, sleep is increased when development is arrested upon starvation. Here, we performed a reverse genetic sleep screen in arrested L1 larvae for genes that are associated with metabolism. We found over 100 genes that are associated with a reduced sleep phenotype. Enrichment analysis revealed sphingolipid metabolism as a key pathway that controls sleep. A strong sleep loss was caused by the loss of function of the diacylglycerol kinase 1 gene, dgk-1, a negative regulator of synaptic transmission. Rescue experiments indicated that dgk-1 is required for sleep in cholinergic and tyraminergic neurons. The Ring Interneuron S (RIS) neuron is crucial for sleep in C. elegans and activates to induce sleep. RIS activation transients were abolished in dgk-1 mutant animals. Calcium transients were partially rescued by a reduction-of-function mutation of unc-13, suggesting that dgk-1 might be required for RIS activation by limiting synaptic vesicle release. dgk-1 mutant animals had impaired L1 arrest survival and dampened expression of the protective heat shock factor gene hsp-12.6. These data suggest that dgk-1 impairment causes broad physiological deficits. Microcalorimetry and metabolomic analyses of larvae with impaired RIS showed that RIS is broadly required for energy conservation and metabolic control, including for the presence of sphingolipids. Our data support the notion that metabolism broadly influences sleep and that sleep is associated with profound metabolic changes. We thus provide novel insights into the interplay of lipids and sleep and provide a rich resource of mutants and metabolic pathways for future sleep studies.
Introduction
Sleep is conserved in all animals that have a nervous system ranging from cnidarians to mammals (Cirelli and Tononi 2008; Nath et al. 2017). Hence, sleep appears to have evolved together with the emergence of the nervous system at least 600 million years ago (Joiner 2016; Bringmann 2018). The common origin of sleep suggests that several key genes and molecular mechanisms underlying sleep are conserved. Hence, genes found to control sleep in a model animal can be studied subsequently in other species by investigating the orthologous genes.
Genetic screening is straightforward to apply in C. elegans. Thus, the molecular dissection of biological phenomena is facilitated in this system (Hodgkin 2005). Through reverse genetic screening, it was found that the AP2 transcription factor gene aptf-1 is required for sleep in C. elegans by acting in the Ring Interneuron S (RIS) neuron. RIS was identified to be crucially required for sleep during many physiological conditions and developmental stages in C. elegans, e.g. during the molt (lethargus) (Turek et al. 2013, 2016), in the adult following stress (Grubbs et al. 2020; Konietzka et al. 2020; Sinner et al. 2021) and following starvation (Skora et al. 2018; Wu et al. 2018). Impairing RIS causes a virtually complete loss of sleep during multiple conditions including during L1 arrest (Turek et al. 2013, 2016; Wu et al. 2018; Maluck et al. 2020; Sinner et al. 2021). It was shown that RIS is sleep active, i.e. it activates specifically during sleep as seen by increased intracellular calcium levels (Turek et al. 2013, 2016; Nichols et al. 2017; Wu et al. 2018; Maluck et al. 2020). The AP2 transcription factor was also found to control sleep in Drosophila (Kucherenko et al. 2016). Furthermore, there is a genetic link between Char syndrome in humans, which is caused by heterozygous loss of Tfap2b function, and a sleep disorder (Mani et al. 2005). This led us and others to study mice that are defective in various AP2 orthologs. This analysis showed that different AP2 transcription factor genes play divergent regulatory roles in mouse sleep. While the AP2 transcription factor gene Tfap2a inhibits sleep, Tfap2b is required for normal amounts and quality of sleep (Hu et al. 2020; Nakai et al. 2020; Hu and Bringmann 2023). As another example for conservation of sleep genes across C. elegans and mammals, the gain-of-function mutation of the Salt-Inducible Kinase 3-encoding gene Sik3 was found to increase sleep in mice. Loss-of-function mutation of Sik3 is lethal in mice but can be studied in C. elegans, where it reduces sleep (Funato et al. 2016).
Sleep and metabolism are thought to be tightly intertwined with metabolism controlling sleep. Sleep, in turn, controls metabolism (Anafi et al. 2019; Grubbs et al. 2020). When C. elegans hatch in the absence of food, it arrests its development at the L1 stage and switches metabolism to a mode that supports survival (Baugh and Hu 2020). This L1 arrest is associated with increased activation of the sleep-active RIS neuron, which leads to increased sleep. RIS activation in turn is required for survival of L1 arrest and the attenuation of ageing phenotypes (Wu et al. 2018). RIS supports survival by promoting the expression of protective genes such as heat shock proteins and other FOX protein transcription factor O (FOXO) targets (Koutsoumparis et al. 2022; Busack and Bringmann 2023).
Here, we performed reverse genetic screening for sleep phenotypes during L1 arrest using mutant strains of genes associated with the control of metabolism. This genetic screening uncovered a large number of mutants with altered sleep behavior that we have started to characterize and that provide a rich resource for future studies.
Methods
Caenorhabditis elegans maintenance
Caenorhabditis elegans strains were grown on Nematode Growth Medium (NGM) plates seeded with Escherichia coli OP50 and were kept at 20°C (Brenner 1974). The C. elegans strains used in this project are listed in Supplementary Table 3.
Long-term imaging using agarose hydrogel microchambers
Long-term imaging of L1 arrested worms was performed in agarose hydrogel microchambers as described previously (Bringmann 2011; Turek et al. 2015; Koutsoumparis et al. 2022). For this, a poly-dimethyl-siloxane (PDMS) mold was activated by air plasma exposure for 30 s and was used to cast microcompartments from 1 ml of 5% high gelling temperature agarose (Sigma-Aldrich) dissolved in M9 buffer and heated to about 95°C. The chamber size was 110 × 110 × 10 μm. By using a platinum worm pick, the chambers were filled with 3-fold-stage eggs without transferring any food. The chambers were then sealed with a glass coverslip and glued with double-sided adhesive tape into a square (2 × 2 cm) opening. The opening had been milled before into a 35 × 10 mm plastic dish. The dish was then filled with 2 ml of agarose (moisture reservoir) and sealed with Parafilm M (Sigma-Aldrich) to prevent desiccation. The chambers were then incubated with the plastic dish placed upside down into an incubator at 20°C until use.
Differential interference contrast imaging screen for quiescence assessment in L1 arrest
Strains were assessed for quiescence in L1 arrest via continuous differential interference contrast (DIC) imaging. An “Eclipse Ti” microscope (Nikon) with an automated XY stage (Nikon), a digital camera (DS-Qi2, Single Lens Reflex [SLR], FX-format CMOS sensor, Nikon), and a 10×/0.45 Plan Apo λ objective lens (Nikon) were used. A red-light filter (BrightLine HC 785/62, 45 mm diameter, Semrock) was used in order to minimize behavioral changes due to illumination. A custom-made heating lid was used to keep the temperature constant at 25.5°C to avoid condensation (sample temperature was measured at 23.5°C). The software used for image acquisition and microscope control was “NIS elements” (Nikon). The length of the movies was 4 h and the frame rate was 0.2 frames per second (1 frame/5 sec). Exposure time was set to 50 ms and the resolution was 14-bit 808 × 808.
Mean pixel intensity values (that correspond to the magnitude of the displacement) were extracted for individual worms via image subtraction (Nagy, Raizen, et al. 2014), smoothened at 1% of total points (robust LOESS regression smoothing (a function in Matlab)), and divided by the time interval to obtain speed values. Quiescence bouts were detected by applying a threshold of “speed < 40% of maximum speed continuously for a duration of time >2 min”, with the custom-written MATLAB script. This threshold was selected as it most accurately describes the wild-type worm quiescence in the 110 × 110 × 10 μm chamber that was used. Then, the fraction of quiescence was calculated for each worm. Eggs that did not hatch or worms that died before or during the measurement were excluded from the analysis. Each strain was imaged in at least 2 chambers (replicates). A median number of 15 animals was screened per strain (minimum = 3). The effect size (Cohen's d), as well as 95% confidence intervals, were calculated by comparing each mutant strain to the wild-type. Precision was calculated as the reciprocal of the standard error of the mean. statistical significance was assessed by the 2-tailed Welch's t-test (assuming unequal variances) and the resulting P-values were corrected for false discovery rate by the Benjamini–Hochberg method. For a hit to be considered significant in this study, it has to satisfy the following criteria:
|Cohen's d| > 1.2 “Very large”
|95% confidence interval of mutant| > |95% confidence interval of wild-type|
The above criteria are stricter than the regular fold discovery rate (FDR) < 0.05 threshold but were chosen to account for the extent of the study and the magnitude of the biological effect observed.
Imaging of dgk-1(−) rescues and quantification of sleep
Microfluidic chambers of size 110 × 110 × 10 μm were filled with 1 egg each with the rescue and control strains as previously described (Busack and Bringmann 2023). The worms were incubated at 20°C for 48 h. Then behavior was imaged for 5 h with a framerate of 0.2 Hz. DIC imaging was conducted to assess the quiescence of the dgk-1 rescue strains using a Nikon Ti2 and NIS Element software. A 20× objective combined with a 1.5× lens was used to simultaneously image 6 × 6 adjacent microchambers containing 1 worm each. Before the start of the DIC timelapse imaging series, a single image was acquired with a 550 nm light-emitting diode (LED) (Andor Sona) and standard Texas red filters (Chroma). The mCherry signal was then utilized to distinguish between worms containing the rescue and control worms. The data was analyzed using image subtraction as described previously (Konietzka et al. 2020). To adjust for the different imaging parameters a threshold of 0.3 was set for the quiescent bout detection. Statistical analysis was conducted using Welch's t-test and the P-values were subsequently corrected for multiple comparisons using the Benjamini–Hochberg method.
Fluorescence imaging in L1 arrest
Fluorescence-tagged transgene expression was quantified by fluorescence microscopy as described (Koutsoumparis et al. 2022). For this, an “Eclipse Ti” microscope (Nikon) with an automated XY stage (Nikon), a digital camera (DS-Qi2, SLR, FX-format CMOS sensor, Nikon), and a 40×/0.75 Plan Fluor Oil objective lens (Nikon) were used combined with an additional 1.5× lens for a total magnification of 60×. A custom-made heating lid was used to keep the temperature constant at 25.5°C to avoid condensation (the sample temperature was thus around 23.5°C). NIS Elements (Nikon) software was used for image acquisition and microscope control. For GFP and mKate2 excitation, an LED system (CoolLED) was used at 490 and 565 nm, respectively. Exposure time was set to 100 ms and EM Gain to 1.0×. Light intensity was measured to be around 1.75 mW/mm2 for 490 nm and 0.517 mW/mm2 for 565 nm illumination, using an optical power meter (PM100A, Thorlabs). Standard sets of GFP and Texas Red filters were used for light filtering. The camera resolution was 14-bit 808 × 808.
Reporter transgene expression was measured every 24 h for up to 12 days. Average fluorescence intensity was quantified per worm after background subtraction, assuming that the size of the worm is 10% of the total image size. Chambers were made in the evening of the day before the start of the experiment (Day 0). That allowed for the embryo a time interval of 10–12 h to complete development and hatch but not yet starve. At least 20 animals were imaged per strain in 2–3 microchamber arrays (each forming 1 replicate). The chambers were kept in a 20°C incubator when not being imaged.
Statistical analysis was carried out as described before (Koutsoumparis et al. 2022). Worms that died before or during the measurement period were omitted from downstream analysis. Single worms with transgene intensities >Q3 + 1.5 * IQR or transgene intensities <Q1 + 1.5 * IQR for every time point of the analysis, were scored as outliers and omitted from further analysis (Q1 = 25th percentile, Q3 = 75th percentile, and IQR = interquartile range). Significance was assessed at every time point by the 2-tailed Welch's t-test (assuming unequal variances), the resulting P-values were corrected for false discovery rate by the Benjamini–Hochberg method, and FDR < 0.05 was selected as the significance threshold. In every case, error bars denote the standard error of the mean.
GCaMP imaging
GCaMP 3.35 was expressed in RIS using the flp-11 promoter and used as described before (Wu et al. 2018; Koutsoumparis et al. 2022). Three-fold-stage embryos were placed in agarose microchambers (110 × 110 × 10 μm) and imaged with a Nikon Eclipse Ti microscope, using an iXon Ultra EMCCD (1,024 × 1,024 pixels) (Andor Technology Ltd.). The camera and a 40×/0.75 Plan Fluor Oil objective lens (Nikon) were used combined with an additional 1.5× lens to achieve a total magnification of 60×. For GFP excitation, an LED system (CoolLED) provided light at 490 nm. Exposure time was set to 100 ms and EM Gain to 1.0×. Light intensity was around 1.6 mW/mm2. Standard sets of GFP and Texas Red filters were used for light filtering as described above.
Statistical analysis was carried out as described before in Koutsoumparis et al. (2022). For this, the position of the RIS neuron was detected and the intensity was quantified with a custom-made MATLAB script based on an empirically defined intensity threshold. At least 7 animals were screened per strain in 2 replicates. The area under the curve (auc) but above the baseline was calculated with Matlab. Statistical significance was assessed by the 2-tailed Welch's t-test (assuming unequal variances) and the significance threshold was set to P < 0.05.
Survival and lifespan of L1 arrested animals
We carried out survival and lifespan measurements as follows: 3-fold stage eggs were picked on 110 × 110 × 10 µm microchambers without food, made of 3% high gelling temperature agarose (Sigma-Aldrich) dissolved in M9 buffer. The chambers were stored in a 20°C incubator. Animals were scored as “dead”, if they failed to move after exposure to blue light for 2 min. Measurements took place every day. At least 8 animals were screened per strain in 3 replicates. Survival was estimated by using the Kaplan–Meier method, significance assessment was done by using the log-rank test and the significance threshold was set to P < 0.05.
Microcalorimetry
L1 arrest was induced by hypochlorite synchronization and hatching in M9. After 48 h of starvation, the worm suspension was transferred in cylindrical, sterile glass containers. Isothermal microcalorimetry was performed using two 2277 Thermal Activity Monitors and Digitam software (Thermometric AB, Jarfalla, Sweden) similar to described previously (Laranjeiro et al. 2017). The temperature was set to 24°C. A blank sample containing only M9 buffer was used to determine the background signal. Thermal equilibration was reached after about 45 min indicated by a stable signal. Measurements were taken every 1 sec for 4 h (4 replicates). Results were normalized to the total amount of worms per sample. Quantile normalization was used to remove technical variance since the measurements are close to the detection limit. Statistical significance was assessed using the 2-tailed Welch's t-test (assuming unequal variances) and the P-values were corrected for false discovery rate by the Benjamini–Hochberg method. The significance threshold was set to FDR < 0.05.
Metabolome analysis in L1 arrest
Populations of arrested L1 C. elegans larvae of strains N2 and HBR227 were grown for metabolome analysis at a final concentration of 100.000 worms/sample. In total, 3 replicates per condition were measured. Metabolome analysis by liquid chromatography–mass spectrometry was performed by the laboratory of Dr. David Meierhofer (Max Planck Institute for Molecular Genetics) as previously described (Gielisch and Meierhofer 2015). Differential abundance analysis was carried out by a custom-written MATLAB script. Briefly, principal component analysis was performed to detect any outlier samples, which were omitted from downstream analysis. Metabolites not detected (intensity = 0) in at least 1 sample were removed. Intensities were then quantile normalized and metabolites with highly variant intensities among replicates (coefficient of variation [CV] > 97.5th percentile) were removed. The remaining metabolites were assessed for statistical significance using the 2-tailed Welch's t-test (assuming unequal variances) and the P-values were corrected for false discovery rate by the Benjamini–Hochberg method. Significance thresholds were set as follows: |log2(fold change [FC])| > 0.5 and FDR < 0.1.
Results
Reverse genetic screening during L1 arrest for sleep mutants
The L1 arrest is triggered by starvation and is associated with profound changes in metabolism (Baugh 2013; Baugh and Hu 2020). As long-term starvation promotes sleep (Skora et al. 2018; Wu et al. 2018), this suggests that metabolic changes are mechanistically linked to the control of sleep. In order to find genes that control both metabolism and sleep, we performed reverse genetic screening for defects in L1 arrest sleep of mutants in genes that are known to control metabolism. We chose a total set of 503 strains. The selection encompassed genes encoding for enzymes that catalyze biochemical reactions but also genes that encode factors that control signaling in the nervous system. We chose alleles that are not lethal, thus, allowing for their behavioral analysis, and focused on alleles that are publicly available (Supplementary Table 1). aptf-1(gk794) and flp-11(tm2706) were included in the data set as positive controls for strong sleep loss-of-function mutation (Turek et al. 2013, 2016). We used agarose microchamber imaging to monitor sleep behavior in L1 larvae. We quantified behavioral quiescence by frame subtraction after 48 h of L1 arrest as a readout for sleep (Bringmann 2011; Turek et al. 2015; Wu et al. 2018; Koutsoumparis et al. 2022). 109 strains showed reduced quiescence and 8 strains showed increased quiescence (Fig. 1a Supplementary Table 1). Among the strains that had increased quiescence, the insulin receptor daf-2 loss-of-function mutant stood out in this analysis, with an increase of 59.76% in the time spent in quiescence, compared to the wild type (Fig. 1a and b, Supplementary Table 1). Hence the screen confirmed the previously described role of daf-2 in inhibiting sleep (Skora et al. 2018; Wu et al. 2018). The other strains with increased quiescence phenotypes should be considered with caution, since many of them showed visible movement defects that may have influenced the quantification of sleep.
A reverse genetic screen on mutants associated with metabolism identifies many regulators of sleep, including dgk-1. a) Distribution of behavioral quiescence phenotypes of all mutants that where screened. Effect size (Cohen's d) is plotted vs precision. The wild-type strain is denoted in blue the 109 low-quiescence mutants in red, and the 8 high-quiescence mutants in green. Only hits with an effect size: |Cohen's d| > 1.2 and 95% confidence intervals that do not overlap with the confidence intervals of the wild type were considered significant. b) Boxplot representation of the fraction of behavioral quiescence of wild type, aptf-1(gk794), dgk-1(ok1462), and daf-2(e1370). dgk-1 loss-of-function mutation reduced quiescence by 99.93% compared to the wild type, while daf-2 loss-of-function mutation increased quiescence by 59.76%. Wild type: n = 33, aptf-1(gk794): n = 31, flp-11(tm2706): n = 26, dgk-1(ok1462): n = 25, daf-2(e1370): n = 27. Red horizontal line denotes the median value. Statistical significance was assessed with Welch's t-test, and the P-values were subsequently corrected for multiple comparisons using the Benjamini–Hochberg method ***padj < 0.001. The experiment was performed in 3 replicates. c–g) Sample mobility traces of wild-type, aptf-1(gk794), flp-11(tm2706), dgk-1(ok1462) and daf-2(e1370) larvae. Movement (shown in red) is reduced during quiescence bouts (shown in green). Loss of function of the AP2 transcription factor gene aptf-1 and DAG kinase DGKθ ortholog dgk-1 abolishes the behavioral quiescence that is associated with sleep, while loss of function of the insulin receptor daf-2 increases the time spent in behavioral quiescence.
Consistent with previous results (Wu et al. 2018; Busack and Bringmann 2023), the control mutants aptf-1(gk794) and flp-11(tm2706) showed almost no detectable sleep during L1 arrest (Fig. 1). Among the low quiescence strains, the dgk-1(ok1462) loss-of-function mutant stood out with a decrease of 99.93% in quiescence compared to the wild type (Fig. 1, Supplementary Table 1). The dgk-1(ok1462) sleep phenotype persisted after backcrossing the allele 4 times into the N2 wild-type strain. We further confirmed the decreased quiescence phenotype of dgk-1 by using an independent allele, dgk-1(sy428) (Supplementary Table 1).
Diacylglycerol (DAG) is a second messenger that is responsible for neurotransmission by regulating synaptic vesicle release. dgk-1 encodes for a homolog of the mammalian diacylglycerol kinase theta (DGKθ). dgk-1 acts by phosphorylating DAG into phosphatidic acid, an action that reduces the amount of available DAG, inhibiting neurotransmitter release. dgk-1 mutants were previously shown to have increased synaptic transmission and to be behaviorally hyperactive at the adult stage (Miller et al. 1999, 2000; Nurrish et al. 1999; Robatzek and Thomas 2000; Robatzek et al. 2001; Jose and Koelle 2005). We hence also quantified the distribution of motion activity during L1 arrest (Turek et al. 2013, 2016). Wild-type animals showed a broader activity distribution that encompassed both lower and higher levels of activity. By contrast, dgk-1 mutants were almost constantly behaviorally active at a high level, indicating hyperactivity (Supplementary Fig. 1).
Additional loss-of-function alleles resulted in low quiescence phenotypes. These mutants include the FOXO homolog daf-16, hypoxia response genes (hif-1, vhl-1, and egl-9), and the apoptosis regulator gene ced-3 (Supplementary Table 1). For 16 of the screen hits, we obtained the same direction of change in the amount of quiescence for 2 independent alleles. For increased sleep, there were 2 alleles of acl-14 and for decreased sleep, there were at least 2 alleles for ace-2, ced-3, daf-16, dgk-1, dgk-5, egl-9, elo-1, elo-3, glp-1, hyl-1, idh-2, mif-1, nnt-1, pah-1, and ser-2 (Supplementary Table 1).
We used functional enrichment analysis to identify genetic pathways among the strains that had low quiescence. For this analysis, we searched for enrichments within the list of genes that we had screened. We found the strongest enrichment (padj = 7.48 × 10−5) in sphingolipid metabolism genes (hyl-1, hyl-2, F33D4.4, sms-3, hpo-13, sptl-3, gba-1, and plpp-1.2) (Supplementary Table 1) (Deng and Kolesnick 2015; Watts and Ristow 2017). For hyl-1, a clear low quiescence phenotype was obtained for 2 independent alleles, and for gba-1 a trend for decreased quiescence was observed in a second allele that bordered statistical significance (Supplementary Table 1). The presence of a low quiescence phenotype in multiple mutants of the sphingolipid metabolism pathways strongly supports the involvement of this pathway in promoting sleep.
Thus, our screen revealed a substantial number of strains with mutations in metabolic genes that result in low quiescence. This supports the previous notion that genes that control metabolism also control sleep. dgk-1 impairment produced the strongest phenotype, but also several other mutants including those affecting sphingolipid metabolism and genes encoding enzymes displayed consistently reduced sleep phenotypes. The involvement of dgk-1 and the sphingolipid metabolism pathway in promoting behavioral quiescence is supported by multiple mutants in our dataset. For dgk-1, a sleep phenotype is observed in 2 loss-of-function alleles, for the sphingolipid metabolism pathway 8 strains carrying 8 different alleles result in significant sleep loss phenotypes, thus, strongly supporting the idea that sphingolipid metabolism promotes sleep. For the additional hits from the genetic screen, it will be necessary to test whether the suspected mutations are causal to the sleep phenotype, for example by backcrossing the allele, by testing additional alleles, or by rescue experiments.
dgk-1 acts mostly in cholinergic and also in tyraminergic neurons
dgk-1 is broadly expressed in most if not all neuron types (Miller et al. 1999; Nurrish et al. 1999; Taylor et al. 2021). To test in which neuron types dgk-1 is required to allow for sleep, we performed rescue experiments by transgenically expressing dgk-1 in specific subsets of neurons in a dgk-1(ok1462) background and quantified total sleep time as well as sleep bout frequency and sleep bout length during L1 arrest. For this experiment, we used promoters that express in a large fraction of neurons (the rgef promoter expresses in about 100 neuronal cell types), in cholinergic neurons (the unc-17 promoter expresses in about 60 cholinergic neuronal cell types (Pereira et al. 2015; Taylor et al. 2021)), in GABAergic neurons (the unc-47 promoter expresses in about 17 different neuronal cell types including in the sleep neuron RIS (Gendrel et al. 2016; Taylor et al. 2021)), in glutamatergic neurons (the eat-4 promoter expresses in about 48 different neuronal cell types (Serrano-Saiz et al. 2013; Taylor et al. 2021)), in serotonergic neurons (the tph-1 promoter expresses in about 7 neurons (Taylor et al. 2021; Barrett et al. 2022)), in dopaminergic neurons (dat-1 promoter (Gaglia and Kenyon 2009; Taylor et al. 2021)), in tyraminergic neurons (the tdc-1 promoter expresses in Ring Interneuron M (RIM) and Ring Interneuron C (RIC) (Alkema et al. 2005; Taylor et al. 2021)), in the sleep neuron ALA (the flp-24 promoter expresses in 24 different cell types and most strongly in ALA (Nath et al. 2016; Taylor et al. 2021)), and in nmr-1-expressing neurons (nmr-1 encodes for an N-methyl-D-aspartate receptor that expresses in interneurons involved in locomotion (Brockie et al. 2001; Taylor et al. 2021)). The rescue was observed most strongly when dgk-1 was driven in almost all neurons (rgef promoter) and cholinergic neurons (Fig. 2). While expression of dgk-1 in cholinergic neurons rescues the total amount of sleep, the architecture of sleep was different from wild type with shorter and more frequent bouts in the rescue strain. In addition to the rescue in cholinergic neurons, we observed a small partial rescue by expressing dgk-1 in tyraminergic neurons (Fig. 2). These results suggest that the effects of dgk-1 on sleep are distributed across multiple neuron types. As there was no rescue in either GABAergic and flp-24-expressing neurons, dgk-1 appears to not be sufficient for sleep if expressed in the sleep neurons RIS and ALA (Van Buskirk and Sternberg 2007; Turek et al. 2013). Instead, dgk-1 is sufficient for sleep if expressed in cholinergic and tyraminergic neurons that are known to be active during wakefulness behavior (Nichols et al. 2017; Skora et al. 2018; Maluck et al. 2020; Busack and Bringmann 2023). In light of the known role of dgk-1 in limiting synaptic transmission, these results suggest that dgk-1 might act to limit excitability in these wakefulness neurons, which allows for sleep.
dgk-1 is required for RIS calcium activation transients during L1 arrest, likely through the inhibition of synaptic transmission
dgk-1 is a negative regulator of synaptic transmission and is expressed broadly in many neurons (Taylor et al. 2021). Thus, dgk-1 might also impair the functionality of sleep-controlling circuits that include the functioning of the sleep-active RIS neuron. Different hypotheses are conceivable for how dgk-1(−) might cause sleep loss. It might impair the activity of circuits that act upstream of and activate RIS, e.g. command interneurons (Maluck et al. 2020). In such a scenario the activation transients of RIS should be reduced. Alternatively, dgk-1(−) might cause a sleep deficit while RIS activates normally but would be unable to inhibit downstream wakefulness circuits. Such an effect has been observed, for example, in mutations that impair the sleep-promoting neurotransmitters that are released by RIS (Turek et al. 2013, 2016). In this second scenario, RIS would activate normally but the activation transients would not be associated with sleep bouts any longer. To test whether dgk-1 causes aberrations in RIS activation, we performed calcium imaging in RIS with GCaMP inside microchambers (Fig. 3) (Wu et al. 2018; Koutsoumparis et al. 2022). The average GCaMP activity of RIS showed no differences in the dgk-1 loss-of-function backgrounds (Fig. 3a, b, e). Strikingly, however, the characteristic RIS activation transients that are visible in the wild type were almost completely absent in dgk-1 (Fig. 3a, b, f, g). Consistent with this result, the reduction of speed that was observed by RIS activation in the wild type was abolished in dgk-1. Thus, the inverse correlation of RIS activity and speed was lost in dgk-1 (Fig. 3h). Hence, RIS activation is impaired in dgk-1, potentially explaining the lack of quiescence bouts. It is known that arousal levels control sleep and can inhibit the activation of RIS (Raizen et al. 2008; Choi et al. 2013; Iwanir et al. 2013; Schwarz and Bringmann 2013; Turek et al. 2013; Nagy, Tramm, et al. 2014; Singh et al. 2014; Nichols et al. 2017; Wu et al. 2018; Maluck et al. 2020). Thus, a hypothetical explanation for the dgk-1 phenotype would be that increased global synaptic transmission in dgk-1 mutant animals prevents the activation of RIS, and might in addition also impair the shutdown of wakefulness circuits.
dgk-1(−) rescue in cholinergic and tyraminergic neurons restores sleep. a) Expression in cholinergic neurons is sufficient to retrieve the wild-type fraction of quiescence. Expression of dgk-1 in tyraminergic neurons results in a partial yet significant rescue of dgk-1(−). Statistical significance was assessed with Welch's t-test, and the P-values were subsequently corrected for multiple comparisons using the Benjamini–Hochberg method n.s.: not significant, *padj < 0.05, ***padj < 0.001. b) The average quiescent bout length is increased when dgk-1(−) is rescued in cholinergic or tyraminergic neurons. Statistical significance was assessed with Welch's t-test, and the P-values were subsequently corrected for multiple comparisons using the Benjamini–Hochberg method n.s.: not significant, *padj < 0.05, **padj < 0.01. c) The average quiescent bout frequency is increased when dgk-1(−) is rescued in cholinergic or tyraminergic neurons. Statistical significance was assessed with Welch's t-test, and the P-values were subsequently corrected for multiple comparisons using the Benjamini–Hochberg method n.s.: not significant, *padj < 0.05, ***padj < 0.001.
dgk-1 is required for the activation of RIS. a) RIS calcium activity sample trace of an animal that expresses GCaMP3 in RIS in a wild-type genetic background. b) RIS calcium activity sample trace of an animal that expresses GCaMP3 in RIS in a dgk-1(ok1462) genetic background. c) RIS calcium activity sample trace of an animal that expresses GCaMP3 in RIS in an unc-13(s69) mutant genetic background. d) RIS calcium activity sample trace of an animal that expresses GCaMP3 in RIS in a dgk-1(ok1462); unc-13(s69) double mutant genetic background. e) The reduced quiescence of dgk-1(ok1462) is suppressed by unc-13(s69). Statistical significance was assessed with Welch's t-test, and the P-values were subsequently corrected for multiple comparisons using the Benjamini–Hochberg method ***padj < 0.001, **padj < 0.01. f) dgk-1(ok1462) results in reduced integrated RIS activity levels and this effect is suppressed by unc-13(s69). The area under the curve was calculated. Statistical significance was assessed with Welch's t-test, and the P-values were subsequently corrected for multiple comparisons using the Benjamini–Hochberg method ***padj < 0.001. g) dgk-1(ok1462) results in a reduced number of RIS activity transients and this effect is suppressed by unc-13(s69). Statistical significance was assessed with Welch's t-test, and the P-values were subsequently corrected for multiple comparisons using the Benjamini–Hochberg method ***padj < 0.001. h) dgk-1(ok1462) results in loss of anticorrelation of RIS activity levels and speed. Statistical significance was assessed with Welch's t-test, and the P-values were subsequently corrected for multiple comparisons using the Benjamini–Hochberg method ***padj < 0.001. Wild type: n = 15, dgk-1(ok1462): n = 16, unc-13(s69): n = 14, dgk-1(ok1462); unc-13(s69): n = 12. The experiment was performed in 2 replicates.
dgk-1 limits synaptic transmission that depends on unc-13 (Nurrish et al. 1999; McMullan et al. 2006). We hence tested whether the reduction-of-function mutation unc-13(s69), which strongly reduces synaptic transmission and behavioral activity (Rose and Baillie 1980; Richmond et al. 1999), restored RIS calcium transients. unc-13(s69) arrested L1 larvae moved only a little and were almost always in a state of mobility quiescence, which rather seems to present paralysis and not a sleep state. Also, RIS calcium transients were reduced (Fig. 3c, e, f). dgk-1; unc-13 double mutant larvae were also quiescent during most of the time and had reinstated RIS calcium transients that were lower in amplitude but occurred more frequently than in the wild type, resulting in an overall similar integrated RIS activity (Fig. 3d, e, f). As unc-13 mutant animals were mostly immobile due to the broad synaptic transmission defect, no conclusion can be drawn regarding the rescue of sleep behavior. As the RIS calcium transients were partially rescued, our data suggest that a major function of dgk-1 is to limit synaptic transmission to allow for RIS activation.
dgk-1 loss-of-function mutation results in broad aberrations of physiology during L1 arrest
We have previously shown that impairment of RIS reduces starvation resistance during L1 arrest and impairs the expression of protective genes such as hsp-12.6, which encodes a small heat shock protein that is part of a set of genes that are thought to support L1 arrest survival (Hsu et al. 2003; Wu et al. 2018; Koutsoumparis et al. 2022). We hence tested whether dgk-1 mutant animals are also impaired in starvation survival as well as in hsp-12.6 expression during L1 arrest. For these experiments, we measured survival during the arrest until all animals had died and also measured the expression of hsp-12.6 with a fluorescent reporter transgene (Baugh and Sternberg 2006; Wu et al. 2018; Koutsoumparis et al. 2022). We found that dgk-1 loss-of-function mutation was associated with reduced survival and reduced hsp-12.6 expression during L1 arrest (Fig. 4). This result is consistent with the idea that loss of RIS function results in reduced L1 arrest survival and protective gene expression. As the effect of dgk-1 mutation is larger than the effects of aptf-1 deletion, the effects of dgk-1 likely stem not only from the impairment of sleep but also from additional functions of dgk-1.
dgk-1 is required for survival and protective HSP-12.6 expression during L1 arrest. a) Loss of function of dgk-1 reduces the lifespan in L1 arrest. Statistical significance was assessed with the log-rank test, **P < 0.01 and n.s.: not significant. Wild-type: n = 25, aptf-1(gk794): n = 29, dgk-1(ok1462): n = 24. The experiment was performed in 3 replicates. b) dgk-1(ok1462) strongly reduced HSP-12.6 expression based on a fluorescent reporter transgene. Statistical significance was assessed by Welch's t-test, and the P-values were subsequently corrected for multiple comparisons using the Benjamini–Hochberg method, ***padj < 0.001 for time >2 days; Wild type: n = 95 (3 replicates), aptf-1(gk794): n = 95 (3 replicates), dgk-1(ok1462): n = 23 (2 replicates). Error bars denote the standard error of the mean (in aptf-1(gk794) and dgk-1(ok1462) the error bars are so small that they are not visible).
RIS impairment is associated with a metabolic shift and a depletion of sphingolipids
Sleep loss typically is associated with increased activity of excitable cells, hence causing increased energy expenditure, for example through neuronal and muscle activity (Schmidt 2014; Schmidt et al. 2017). Also, sleep loss can cause qualitative metabolic changes and it has been suggested that a core function of sleep is the control of metabolism (McEwen 2006; Tu and McKnight 2006; Scharf et al. 2008; Vetrivelan et al. 2012; Anafi et al. 2019; Grubbs et al. 2020). Microcalorimetry analysis in C. elegans showed that more than half of the energy cost is spent on locomotion, with muscle activity being the largest consumer of energy (Laranjeiro et al. 2017). It is well known that C. elegans shuts down locomotion activity and muscle activity during different types of sleep (Raizen et al. 2008; Schwarz et al. 2012; Hill et al. 2014; Wu et al. 2018; Busack and Bringmann 2023), but the energy cost that is associated with sleep loss has not yet been determined in C. elegans.
Determining energy consumption in the face of sleep loss requires a manipulation that impairs sleep with a high level of specificity. dgk-1 is broadly required for nervous system function, including neuronal transmission (Nurrish et al. 1999; McMullan et al. 2006). Consequently, dgk-1 loss-of-function mutants display broad phenotypes related to nervous system function that include hyperactivity and suppression of sleep. These broad phenotypes limit the specificity of dgk-1 mutants and thus their suitability for the study of heat production upon sleep loss, as phenotypes observed in dgk-1 cannot be causally attributed to sleep function, but would likely relate to other nervous system function or hyperactivity. Hence, we used aptf-1(−) for this experiment, as aptf-1(−) has a highly specific and well-characterized sleep loss phenotype that is caused by impaired RIS function (Turek et al. 2013). Like dgk-1(−), aptf-1(−) causes a virtually complete lack of sleep during L1 arrest (Wu et al. 2018). It does not, however, cause any hyperactivity in the waking state (Turek et al. 2013; Wu et al. 2018).
We performed isothermal microcalorimetry analysis using large populations of arrested L1 larvae to measure the heat output of wild-type larvae and of larvae with sleep loss caused by aptf-1(−) (Turek et al. 2013; Wu et al. 2018). Sleep impairment roughly doubled heat production (Fig. 5a), which is consistent with these worms surviving for about only half of the time. A part of the increased heat production of sleep-impaired worms can likely be attributed to the increased time spent in locomotion. Swimming adult worms consume about 3 times as much energy as paralyzed worms (Laranjeiro et al. 2017). As 48 h-arrested L1 larvae sleep for about 25% of the time, one might thus expect a 20% increase in metabolic output in nonsleeping mutant worms measured over a representative time slot. Hypothetically assuming that no energy is spent during sleep at all one might expect a maximal increase of energy output by 33% in nonsleeping worms. As impairing RIS does not cause any hyperactivity (Turek et al. 2013; Wu et al. 2018), however, the doubling of heat production might hypothetically be explained by altered metabolism.
Sleep loss during L1 arrest is associated with altered metabolism and decreased sphingolipid levels. a) Heat production is increased in aptf-1(−) in 48 h-arrested L1 larvae. The experiment was performed in 4 replicates. Statistical significance was assessed with Welch's t-test, ***P < 0.001. b) Scatterplot overview of metabolomics data of wild-type vs aptf-1(−) animals. Statistically significantly upregulated metabolites are labeled in green, and statistically significant downregulated metabolites are shown in red. Multiple species of sphingomyelins (shown in blue) are downregulated in aptf-1(−) animals. Statistically not significant entries according to the selected thresholds are marked in gray. c) Simplified overview of the sphingolipid metabolism pathway. Indicated on the respective reaction are the genes associated with low quiescence phenotypes.
Only a few individual metabolites such as lipids and ATP have been measured directly in the context of sleep in C. elegans (Grubbs et al. 2020). To our knowledge, no systematic metabolomic analysis of sleep has yet been conducted in C. elegans. We hence determined the metabolomes of 2 day-arrested L1 wild-type and aptf-1(−) larvae using mass spectrometry. One hundred twenty metabolites could be identified in both aptf-1(−) and wild type after 48 h starvation, of which 33 metabolites were present at statistically significant different concentrations (Fig. 5b and c and Supplementary Table 2).
Among the 15 metabolites that were upregulated in aptf-1(−) are multiple molecules that are associated with catabolic processes and increased oxidative stress. These include adenosine monophosphate, which is a breakdown product of ATP, uric acid, and allantoin, which are component of the nucleic acid degradation pathway, succinic acid, which is an intermediate of the tricarboxylic acid cycle cycle and substrate of complex II of the electron transport chain. Among the upregulated amino acids were L -arginine, L -phenylalanine, and L -tryptophan. Several carnitines were upregulated (butyrylcarnitine, propionylcarnitine, and methylmalonylcarnitine) as well as 5′-methylthioadenosine, a component of the methionine salvage pathway. Choline is present in multiple metabolic pathways and hence the metabolic pathway from which the upregulation derives is not clear. It could stem from the degradation or reduced synthesis of, e.g. phospholipids like phosphatidylcholines and sphingomyelins, acetylcholine that might be caused by increased cholinergic neurotransmission during sleep loss, or S-adenosylmethionine. Also, 2 species of sphingomyelins, the longest and the shortest form, were present at an increased level (sphingomyelin (d18:1/26:0) and sphingomyelin (d18:1/12:0)). Glutathione was increased while oxidized glutathione was unaltered, indicating an increase in the reductive capacity, which might indicate a shift toward the production of more antioxidants (Ferguson and Bridge 2019).
The largest group among the 18 downregulated metabolites were sphingomyelins, with 14 identifiable species that were depleted in aptf-1(−) (sphingomyelin (d18:0/16:1), sphingomyelin (d18:1/26:1)/sphingomyelin (d18:0/26:1), sphingomyelin (d18:0/18:0), sphingomyelin (d18:0/12:0), sphingomyelin (d18:0/22:3), Sphingomyelin (d18:0/14:0), sphingomyelin (d18:0/22:1(OH)), sphingomyelin (d17:1/24:0), sphingomyelin (d19:1/24:1), sphingomyelin (d18:1/18:1)/sphingomyelin (d18:1/18:1), sphingomyelin (d18:1/22:1), sphingomyelin (d18:1/22:0)/sphingomyelin (d18:0/22:1), sphingomyelin (d18:0/26:0), sphingomyelin (d18:1/24:1)). Sphingomyelins are lipids that take part in the composition of the cell membrane and can act as signaling molecules (Liu et al. 2014; Deng and Kolesnick 2015). Other species of sphingolipids and ceramide could not be detected in the metabolomes. Also, uracil, 1 of the 4 nucleobases in the nucleic acid RNA, glycerol 3-phosphate, a building block for glycerolipids, and 2 amino acids, L -methionine, and L -proline were detected as less abundant. The reduction of methionine together with the increase in glutathione might speculatively indicate a shift in the transsulfuration pathway toward the production of more glutathione. Overall, the metabolomic analysis of aptf-1(−) indicates a shift in metabolism that includes depletion of sphingolipids.
Discussion
In summary, our reverse genetic screen provides a valuable resource of sleep-controlling genes including dgk-1, components of the sphingolipid pathway, as well as many other genes associated with metabolism. These genes will be investigated in detail in future studies. It will be interesting to test which of these genes regulate sleep specifically during L1 arrest and which of these genes regulate sleep also during other conditions. Also, the underlying mechanisms will need to be solved for sleep during different conditions.
The requirement of dgk-1 for the generation of sleep and physiological benefits associated with sleep might indicate that a negative regulation of synaptic transmission is required for the activation of the RIS neuron and thus for sleep induction. Activation of EGFR signaling produces the second messenger DAG through phospholipase (PLC). EGFR signaling has been shown to promote sleep by activating the ALA and RIS neurons upon cellular stress (Van Buskirk and Sternberg 2007; Hill et al. 2014; Konietzka et al. 2020). The neuronal knockdown of EGFR during L1 arrest does not potently suppress sleep (Konietzka et al. 2020). Hence it seems unlikely that the phenotype of dgk-1(−) is caused by a suppression of EGFR signaling in ALA or RIS and it seems likely that dgk-1(−) acts through a different pathway.
Insomnia in human patients and sleep loss in genetic animal models often is associated with difficulty to shut down arousal circuits. Therapies of insomnia hence often aim at lowering activity of the arousal system (Halasz 1998; Kume et al. 2005; Meerlo et al. 2008; Schwartz and Roth 2008; Bereshpolova et al. 2011; Liu et al. 2012; Cho and Sternberg 2014; Chrousos et al. 2016; Pedersen et al. 2017). Arousal can inhibit RIS (Nichols et al. 2017; Wu et al. 2018; Maluck et al. 2020). Galpha q signaling inhibits DGK-1 in order to promote synaptic vesicle exocytosis. Hence, manipulations that increase Galpha q signaling promote hyperactivity and also inhibit sleep (Schwarz and Bringmann 2013; Trojanowski et al. 2015). DGK-1 might thus hypothetically act by reducing inhibitory inputs onto RIS by inhibiting UNC-13. DGK-1 might hypothetically also be required for shutting down arousal circuits during sleep. The rescue experiments suggest that DGK-1 is important for sleep by (probably negatively) controlling the activity of cholinergic and tyraminergic neurons. It appears that the control of the activity of these neurons is important for the generation of a sleep state. Further experimentation will be required to solve the detailed circuit mechanisms of this function of DGK-1.
Sleep controls energy metabolism, and energy metabolism in turn is a core regulator of physiological activity and arousal in C. elegans as well as in other species (Kilduff et al. 1993; Siegel 2009; Skora et al. 2018; Wu et al. 2018; Anafi et al. 2019). Our microcalorimetric and metabolomic analysis suggests that sleep might preserve energy in part by altering metabolism. Multiple metabolites were altered in their abundance in association with sleep loss, inviting future studies to mechanistically dissect these metabolic pathways during sleep. Together with the results of the genetic screen, these metabolomic data indicate that RIS is required for the presence of sphingomyelins, and sphingolipid genes in turn are required for sleep behavior. This could hypothetically be explained by aptf-1(−) causing a reduction of sphingomyelins as part of global depletion of energy stores and lipids, including sphingomyelins. Glycerol 3-phosphate, a starting material for de novo synthesis of glycerolipids, is also reduced in aptf-1(−), which is consistent with a reduction in lipid species. A depletion of sphingolipid species could in turn interfere with lipid-based signaling that might hypothetically underlie the control of sleep. Further experimentation will be required to dissect the details of these regulatory circuits. With this study, we are providing a plethora of mutants and metabolites that will in the future form a basis for additional dissection of the interplay of metabolism and sleep.
Data availability
Key C. elegans lines that were used in this project are available at the Caenorhabditis elegans Genetics Center. Other C. elegans strains or additional reagents generated for this study are available upon reasonable request. Data used for this manuscript is present in the manuscript and its supplemental Supplementary Tables 1–4. MATLAB scripts that were generated for this study are available at https://github.com/a-koutsoumparis/paper2023.
Supplemental material available at GENETICS online
Acknowledgments
We thank the Caenorhabditis Genetics Center supported by the National Institutes of Health Office of Research Infrastructure Programs (P40 OD010440) for strains. We thank Eva Naumann, Gamze Naz Öztan, and Lina Paola Baena Lozada for assisting in the creation and maintenance of some of the strains used in this study.
Funding
his work was supported by the European Research Council (Starting Grant ID: 637860, SLEEPCONTROL), by the Max Planck Society, and by the Deutsche Forschungsgemeinschaft (BR 4710/5-1, BR 4710/7-1).
Author contributions
A.K. executed and analyzed experiments related to Figs. 1, 3, 4, 5, Supplementary Fig. 1 and Supplementary Tables 1–2, prepared the corresponding figures, and edited the paper. I.B. executed and analyzed the measurement for Fig. 2, prepared Fig. 2, and edited the paper. C.-K.C. and Y.H. provided the dgk-1 rescue strains and edited the paper. B.B. carried out the microcalorimetry measurements and analysis together with A.K. and edited the paper. D.M. carried out the metabolome analysis and edited the paper. H.B. conceived the study and experiments, acquired funding, supervised the project, and wrote the paper. All authors agreed to the final version of the manuscript.
Literature cited
Author notes
Conflicts of interest The author(s) declare no conflict of interest.
