5-HT is synthesized in two steps: the conversion of tryptophan to 5-hydroxytryptophan (5HTP) by tryptophan hydroxylase (Trh in flies and Tph in mammals) (Kuhn et al., 1979), followed by the conversion of 5HTP to 5-HT by aromatic amino acid decarboxylase (Figure 1A). We have generated genetic tools that allow systematic studies of the serotonergic system. Four receptors were known to be 5-HT receptors in drosophila when we started this project: 5HT1a, 5HT1b, 5HT2 and 5HT7 (Colas et al., 1995; Saudou et al., 1992; Witz et al., 1990). Our bioinformatics analysis had revealed a new G-protein- coupled receptor (GPCR), annotated as CG42796 in the drosophila genome, as a new 5-HT receptor which we named as 5HT2b, with the previously known 5HT2 becoming 5HT2a (Brody and Cravchik, 2000; Gasque et al., 2013). 5HT2b was predicted to be coupled to Gq protein and shown in our collaboration to mediate 5-HT responsiveness (Gasque et al., 2013).

We constructed knockout and knockin lines for Trh and all the five receptors for 5HT in drosophila. The ends-out gene targeting strategy was used to generate null mutants (Huang et al., 2009; Rong and Golic, 2000) (Strategy I) (Figure 1B, Figure 1—figure supplement 1A and B). To visualize gene expression and to manipulate neuronal activity, we generated Gal4, LexA or Flp knock-in lines for each of these genes using the CRISPR/CAS9 method, with Gal4, LexA or Flp introduced at the starting sequence of the first exon (Strategy II) or at the end of the open reading frame (Strategy III) (Figure 1C and D, Figure 1—figure supplement 1C and D). Using Strategy II, we obtained null mutants with Gal4/Flp/LexA replacing the targeted gene, which provided strains that allowed us to manipulate neurons in mutants lacking specific genes. Strategy III enabled us to visualize the expression of specific genes faithfully and manipulated neurons without interrupting gene function and expression. The genotypes of constructed lines were confirmed by polymerase chain reaction (PCR) and sequencing (Figure 1—figure supplement 1).

We also constructed an indel mutant for Trh (Trh⁰¹) with a guide RNA targeting the catalytic center of the enzyme and deleting two basepairs (Figure 1—figure supplement 1E–F), which caused an early translational stop (Figure 1—figure supplement 1G). No 5-HT was detected in the brains of Trh⁰¹ or Trh^GKOmutants (generated by Strategy II, Gal4 insertion and replacement of gene), although it was detected in the brains of the wild type (wt) or Trh::Gal4 (generated by Strategy III) flies (Figure 1—figure supplement 2A–D).

We analyzed the sleep patterns of Trh and 5-HT receptor mutants (Figure 1E). Trh, 5HT1a or 5HT2b mutant flies were found to sleep less than the wt in both 12 hr (hr) light/12 hr dark (LD) cycles and in constant darkness (DD) (Figure 1F–H, Figure 1—figure supplement 3A–C). These three mutants exhibited reduced sleep bout duration during the day and night (Figure 1I), and also in DD phase (Figure 1—figure supplement 3D and E). Trh, 5HT1a and 5HT2b mutants also displayed prolonged latency to sleep at night (Figure 1J). Trh, 5HT1a and 5HT2b mutants showed no change in the intensity of locomotor activity when awake (Figure 1K). As reported previously, 5HT1a mutant flies had reduced sleep (Yuan et al., 2006). The sleep phenotype of 5HT2b mutants was much more severe than that of 5HT1a mutants, nearly equal to that of Trh mutants (Figure 1E and H). In DD, 5HT2b mutants showed a large decrease in sleep duration and sleep bout duration in both the subjective daytime and the subjective nighttime (Figure 1—figure supplement 3A–C). Results in males were similar to those in females except that male 5HT1b mutants exhibited decreased sleep during the daytime of LD but not during DD (Figure 1—figure supplement 4). These results indicated that 5HT2b functions to promote sleep. The difference between LD and DD was consistent with the abnormal light sensitivity of 5HT1b mutants as reported previously (Yuan et al., 2005).

To test whether 5-HT functions directly in adult flies or indirectly during development, we restored 5-HT level in Trh mutant flies by feeding adults with 5HTP, which could circumvent the requirement for Trh. Two mg/ml 5HTP for 3 days restored 5HT immunofluorescence in the brains of Trh⁰¹ and Trh^GKO flies (Figure 1—figure supplement 2E–H). 5-HTP feeding rescued both daytime and nighttime sleep in 5HTP to Trh⁰¹ mutants (Figure 1—figure supplement 2I and J) and in Trh^GKO mutants (data not shown) to about wt level, indicating that 5-HT promotes sleep in adult flies.

The involvement of 5-HT in circadian rhythm has been reported previously (Nichols, 2007; Page, 1987; Rea et al., 1994; Yuan et al., 2005). We first analyzed free-running locomotor rhythms to distinguish between the roles of serotonin signaling in circadian and homeostatic sleep regulation. We have not observed circadian period changes in any of our mutants when compared to the wt in the same genetic background after entrainment to a 12 hr (hr) light/12 hr dark (LD) cycle for 3 days, followed by analysis in constant darkness (DD) for 12 days (Figure 2—figure supplement 1). Next, we investigated whether 5-HT regulated sleep homeostasis. We deprived flies of sleep for a whole night with approximately 100% efficiency (Figure 2—figure supplement 2), and measured sleep recovery over the ensuing 48 hr. Wt flies regained their lost sleep and completely recovered during the first 24 hr (Figure 2—figure supplement 3C). Trh and 5HT2b mutants exhibited a significantly lower percentage of sleep recovery rate than wt flies (Figure 2A and B). For weaker sleep deprivation, these mutants also showed impaired sleep recovery (Figure 2—figure supplement 3). However, 5HT1a mutant flies, though defective in sleep, were similar to the wt and mutants of any other serotonergic receptors in showing normal sleep recovery after sleep deprivation (Figure 2A and B), further supporting the hypothesis that involvement in regulating total sleep time is separate from involvement in regulating sleep recovery after deprivation. To test whether 5-HT functions to control sleep homeostasis in adult flies and not through indirect actions during development, we fed 5-HTP to adult flies. We found that 5HTP restored sleep homeostasis in Trh mutant flies to the normal level (Figure 2C and D).

Previous studies have shown that the activation of specific neurons could cause sleep deprivation (Dubowy et al., 2016; Seidner et al., 2015). We took advantage of this to induce sleep loss thermogenetically by expressing transgenic TrpA1 channels (UAS-TrpA1) in dopaminergic neurons (TH-Gal4). We measured the sleep loss, sleep gain and recovery rate in wt, Trh mutant and 5HT2b mutant flies after sleep loss induced by activation of dopaminergic neurons. Trh and 5HT2b mutant flies showed impaired sleep recovery when compared to wt flies (Figure 2E–J). Thus, our results indicate that 5-HT and its 2b receptor regulate sleep homeostasis in adult flies, independent of the methods of sleep deprivation.

To visualize the patterns of Trh or 5HT2b, we crossed flies carrying Trh::Gal4 and 5HT2b::Gal4 (both generated by Strategy III) with flies carrying UAS-mCD8GFP for labeling of the cytoplasmic membrane of cells expressing each gene (Figure 3A and E, Figure 3—figure supplement 1F and J), UAS-stingerGFP for nuclear labeling (Figure 3B and F, Figure 3—figure supplement 1G and K), UAS-DscamGFP for dendritic labeling (Figure 3C and G, Figure 3—figure supplement 1H and L) or UAS-sytGFP for axonal labeling (Figure 3D and H, Figure 3—figure supplement 1I and M). Trh- and 5HT2b-positive neurons were found in the brain and the ventral nerve cord (VNC). Trh::Gal4 labeled all of the previously reported 5-HT clusters in the adult brain: Anterior lateral protocerebrum (ALP), Anterior medial protocerebrum (AMP), Anterior dorsomedial protocerebrum (ADMP), Lateral protocerebrum (LP), Lateral subesophageal ganglion (SEL) and Medial subesophageal ganglion (SEM) anterior clusters; Posterior lateral protocerebrum (PLP), Posterior medial protocerebrum, dorsal (PMPD), Posterior medial protocerebrum, medial (PMPM) and Posterior medial protocerebrum, ventral (PMPV) posterior clusters (Alekseyenko et al., 2010; Pooryasin and Fiala, 2015; Sitaraman et al., 2008 , 2012; Vallés and White, 1988) and showed similar expression pattern when compared to reported transgenic strains (Alekseyenko et al., 2010). The distribution of the axons and dendrites of neurons that were positive for Trh::Gal4 could be found in the optic lobe, the olfactory lobe, the central complex and the mushroom bodies (MBs) (Figure 3C and D, Video 1).

5HT2b::Gal4 labeled more than 500 neurons in the brain (Figure 3E and F). The axons and dendrites were found in the central complex, the olfactory lobe, the optic lobe, the subesophageal ganglion and the ventrolateral protocerebrum, but not the MBs (Figure 3G and H, Video 2). A recent study of 5HT2b expression using the MiMIC system labeled the same brain regions and showed a similar expression pattern (Gnerer et al., 2015). We also examined the expression patterns of Trh^GKO, 5HT2b^GKO (generated by Strategy II, Gal4 insertion and replacement) and 5HT2b-LexA (generated by Strategy II, LexA insertion and replacement), and found patterns of expression similar to that of the Gal4 line generated by Strategy III (Figure 3—figure supplement 1A–E).

Previous studies have implicated different brain regions, such as the MB (Joiner et al., 2006; Pitman et al., 2006; Sitaraman et al., 2015), the dFB (Donlea et al., 2011, 2014; Liu et al., 2012; Pimentel et al., 2016; Ueno et al., 2012), large ventral lateral clock neurons (lLNv) (Chung et al., 2009), dorsal neurons (DN1) (Guo et al., 2016; Kunst et al., 2014), the ellipsoid body (EB) (Liu et al., 2016), and the pars intercerebralis (PI) (Crocker et al., 2010) in controlling sleep duration and homeostasis (Figure 4A). We used the intersectional strategy to examine whether 5HT2b was expressed in any of these regions. Sparse expression was found in the MB and extensive expression in the dFB when 5HT2b-LexA strains were combined with LexAop-Flp, UAS-FRT-stop-FRT-mCD8GFP and different region-specific Gal4 lines (Figure 4—figure supplement 1B). To further confirm the functional roles of the 5HT2b gene, we expressed 5HT2b RNA interference (RNAi) in 5HT2b neurons and different brain regions. RNAi efficiency was validated by qRT-PCR (Figure 4—figure supplement 1A). RNAi-mediated knockdown of either 5HT2b or dFB neurons reduced sleep relative to controls, whereas knockdown of 5HT2b in other regions had no significant effect (Figure 4A). 23E10-GAL4 and 23E10-LexA lines labeled approximately 40 neurons projecting their axons to the dFB (Figure 4B and C) (Donlea et al., 2014; Manning et al., 2012). Two to four dFB neurons were labeled on each side of the adult fly with intersection of 5HT2b and 23E10 (Figure 4D and E, Video 3). We also used 23E10-Gal4 and 5HT2b-LexA to drive the expression of UAS-StingerGFP and LexAop-tdTomato separately (Figure 4F). Only a single dFB neuron on each side of the adult fly was positive for both green fluorescent protein (GFP) and red fluorescent protein (RFP) (Figure 4G and H). They constitute a single pair of 5HT2b-positive neurons in the dFB.

To visualize 5HT2b protein expression, we constructed a 5HT2b protein-trap knockin strain by fusing a superfolder GFP to the carboxyl terminus of 5HT2b protein (Pédelacq et al., 2006). We found the pattern of protein expression to be similar to that of 5HT2b gene expression (Figure 3E and Figure 4J). We also found both strong anti-serotonin immunofluorescence and 5HT2b^sfGFP staining signal in the dorsal FB region (Figure 4I–K). To determine whether the ExFl2 neurons contribute to the 5HT2b^sfGFP staining signal in the dorsal FB region, we crossed UAS-5HT2b^sfGFP to 23E10-Gal4, and found strong GFP in the presynaptic region of ExFl2 neurons, but not in the postsynaptic region nor in the cell body (Figure 4L–N). These results suggest that the 5HT2b receptor is located presynaptically in ExFl2 neurons.

Artificial activation of dFB neurons induces sleep (Donlea et al., 2011; Ueno et al., 2012), whereas silencing of them reduces sleep (Kottler et al., 2013; Liu et al., 2012). We confirmed these results by activating 23E10-labeled dFB neurons with UAS-Nachbach (Luan et al., 2006), which used a bacterial sodium channel to increase neuronal excitability (Figure 5—figure supplement 1A and B). We crossed 23E10-Gal4 with UAS-head involution defective (UAS-Hid), expressing a protein causing cell death (Zhou et al., 1997), to ablate the 23E10 neurons (Figure 5—figure supplement 1C and D). Sleep duration was reduced in flies in which 23E10 neurons were ablated (Figure 5—figure supplement 1E and F). Flies without 23E10 neurons lost sleep homeostasis: showing no sleep rebound after 12-hr overnight sleep deprivation (Figure 5—figure supplement 1G and H).

We next tested the roles of 5HT2b- and dFB- intersectional neurons. We used Gal80 driven by the Tublin promoter to suppress 23E10-Gal4-driven expression of UAS-mCD8GFP, or of effectors such as Nachbach and Hid, with the Gal80 gene flanked by FRT recombination sites (FLP-out Gal80; Tub-FRT-Gal80-FRT). Gal80 could efficiently suppress Gal4 expression (Figure 6—figure supplement 1C). When combined 5HT2b-LexA was used to drive the expression of LexAop-Flp, FLP-mediated recombination caused GAL80 to be flipped out, resulting in Gal4 expression in Flp-containing 5HT2b neurons. Consequently, GAL4 was effective exclusively in 5HT2b and 23E10 intersectional neurons (Figure 6—figure supplement 1A and B) (Gordon and Scott, 2009). Sleep was greatly increased after 23E10 and 5HT2b intersectional neurons were activated (Figure 5A–C). Sleep was reduced in flies in which 23E10 and 5HT2b intersectional neurons were ablated (Figure 5A–C). Sleep homeostasis was impaired after ablation of 23E10 and 5HT2b intersectional neurons (Figure 5D–F).

To examine whether the 5HT2b receptor functions in dFB neurons, we restored 5HT2b cDNA to 5HT2b knockout mutant flies by UAS-5HT2b (Figure 6A). 5HT2b^GKO is a Gal4 KI line generated by Strategy II which was used to drive the expression of 5HT2b cDNA in all 5HT2b neurons, whereas 23E10 Gal4 was used to drive 5HT2b expression in dFB neurons (Figure 6A). Both significantly increased sleep duration at nighttime but neither affected daytime sleep (Figure 6B and C), indicating that 5HT2b in dFB neurons is sufficient to promote nighttime sleep. Sleep recovery after 12-hr deprivation in 5HT2b mutant flies was rescued by either 5HT2b expression in all 5HT2b positive neurons or 5HT2b expression in 23E10 neurons (Figure 6D and E). We also used RNAi knockdown of the 5HT2b gene in dFB neurons to test the necessity of 5HT2b. A dFB gene knockdown strain showed shortened sleep duration and impaired sleep homeostasis (Figure 6K–O). These results indicate that 5HT2b in dFB neurons is sufficient to regulate sleep homeostasis.

To determine whether a single pair of dFB neurons that were positive for the 5HT2b receptor were necessary for sleep homeostasis, we used 5HT2b^GKO to drive the expression of 5HT2b cDNA in all 5HT2b neurons, but with 23E10-LexA to drive the expression of LexAop-Gal80, thus suppressing 5HT2b expression in only one pair of dFB neurons (Figure 6F). Dual reporters for 23E10 and 5HT2b were used to test the efficiency of Gal80 suppression, which revealed a single pair of neurons that were positive for both 23E10 and 5HT2b (Figure 6—figure supplement 1D and E), but no double-positive neurons in the presence of LexAop-Gal80 (Figure 6—figure supplement 1F and G), indicating that the Gal4 was efficiently suppressed in the dFB by LexA-Gal80. This allowed us to express 5HT2b in all 5HT2b-positive neurons except this single pair of dFB neurons. The sleep duration of flies in which 5HT2b^GKO drove the expression of 5HT2b cDNA in all 5HT2b neurons. except one pair of dFB neurons, in the 5HT2b knockout background was significantly decreased, to a level similar to that of the 5HT2b knockout mutants (Figure 6G and H). Furthermore, flies lacking 5HT2b receptor in dFB neurons had no sleep recovery after 12-hr deprivation (Figure 6I and J). In addition, we rescued the expression of 5HT2b with Gal80 in dFB neurons in the background of 5HT2b knockdown to test the sufficiency of 5HT2b. Both sleep duration and homeostasis were restored in this strain (Figure 6K–O). These results indicate that 5HT2b in a small subset, probably a single pair, of dFB neurons is necessary for the regulation of sleep homeostasis.

Flies were reared at 25°C and 60% humidity and were kept in a 12-hr light-12-hr dark cycle (or in constant darkness when specified). For thermogenetic activation of neurons, flies were raised and loaded into the visual-recording system at 22°C. Temperature was changed to 28°C when neural activation to place. Flies were backcrossed into a Canton S for at least five generations. UAS-Hid, Dilp2-Gal4, Pdf-Gal4 and MB247-Gal4 were generously provided by Aike Guo (Institute of Biophysics, CAS). Tub > Gal80 > was provided by Jing Wang (UCSD). UAS-StingerGFP and LexAop-tdTomato were gifts from Barry Dickson (Janelia Research Campus, HHMI). UAS-TrpA1 was provided by Paul Garrity (Brandeis University). UAS-mCD8GFP, UAS-StingerGFP, UAS-DscamGFP, UAS-sytGFP, UAS-DenMark, UAS-NachBach, UAS-FRT-stop-FRT-mCD8GFP, 23E10-Gal4, 23E10-LexA, 69F08-Gal4, 18H11-Gal4, LexAop-Gal80, TH-Gal4 and LexAop-Flp were obtained from the Bloomington Stock Center. UAS-5HT2b-RNAi (KK102356) was obtained from the Vienna Drosophila Resource Center (VDRC). Table 1 contains the complete genotypes used in the figures.

5-HTP was obtained from Sigma. Wt or Trh mutant flies were raised with normal food during the larval stage and immediately transfered to 5HTP food (5% sucrose, 2% agar and 2 mg/ml 5HTP) after eclosion. For immunostaining analysis, after three days of feeding with 5HTP-containing food, flies were dissected for immunohistochemistry. For behavioral analysis, flies were fed with 5HTP food after eclosion (usually 3–5 days), then transferred to glass tubes with 5HTP food for sleep analysis.

The UAS-5HT2b DNA construct was generated by subcloning 5HT2b cDNA into the pACU2 vector, with the 5HT2b cDNA from the Vosshall lab at Rockefeller University and the pACU2 vector from the Jan Lab at UCSF (Gasque et al., 2013; Han et al., 2011). Superfoder GFP cDNA was subcloned from the Addgene plasmid (No.60904).

The end-out method was used to generate mutant lines (Huang et al., 2009). Gal4 was introduced to replace the gene by homologous recombination. In each of the targeting constructs, the entire first exon with the first ATG codon of the gene (or the entire gene) was deleted. At least the first transmembrane domains of the seven transmembrane domains in a receptor were also deleted to ensure null mutants. The 5’ homologous arms (~5 KB) and the 3’ homologous arm (~3 KB) were inserted into pGX2 vectors by homologous recombination repair. We named the receptor mutants using Strategy I as 5HT1a^Gal4, 5HT1b^Gal4, 5HT2a^Gal4, 5HT2b^Gal4 and 5HT7^Gal4. We also introduced loxp and attp sites along with Gal4, which enabled us to flip out the introduced sequence when crossing with a cre strain, leaving a functional attp site in the gene targeted region (Figure 1B). We named the 5HT2b mutant with introduced sequence flipped out 5HT2b^attp (Figure 2G–H and Figure 6A–J).

The Crispr/Cas9 system was used to generate Trh indel mutant flies. The gRNA and Cas9 mRNA were designed and generated as described previously (Yu et al., 2013). The gRNA targeted the catalytic center of the Trh enzyme. We named the Trh indel mutant Trh⁰¹. PCR analysis and DNA sequencing were used for further confirmation.

Two different gRNAs were used to generate deletion in genome regions of interest. The sgRNA sequence was subcloned into a U6b-sgRNA vector to generate gRNA in fly embryos (Ren et al., 2013). The nos-Cas9 flies were used for injection. The 5’ homologous arms (~2.5 KB) and the 3’ homologous arm (~3 KB) were inserted into a pBSK vector for homologous recombination repair. Arm fragments were amplified from the wt fly genome by Phanta Max DNA polymerase (Vazyme Biotech Co. Ltd.). The Gal4, Flp or LexA effector sequence was introduced between the two homologous arms and 3p3-RFP was used as the fluorescent marker for screening (Xue et al., 2014a, 2014b). We named the Trh with Gal4 knockin using Strategy II strain Trh^GKO, and the Trh Gal4 knockin using Strategy III strain Trh::Gal4. We named the 5HT2b with Gal4 knockin using Strategy II strain 5HT2b^GKO, the 5HT2b LexA knockin using Strategy II strain 5HT2b-LexA, and the Gal4 knockin using Strategy III strain 5HT2b::Gal4. PCR analysis and DNA sequencing were used for further confirmation. Table 2 contains the primers and sgRNAs used.

For each sample, 20 flies were ground on ice. RNA and cDNA were generated with the PrimerScript RT Master Mix (Takara). qRT–PCR was performed using TransStart Green qPCR SuperMix (Transgen Biotech) and the 96-well Applied Biosystems 7900HT Real-Time PCR System.

Female or male flies aged 4–6 days were transferred into monitor tubes (5 mm × 65 mm) containing normal food. 48 flies were usually used in each group in our experiments. Female flies were used for sleep analysis if not specifically mentioned. Light was on from 9:00am to 9:00pm. Locomotor activity was monitored in the LD cycle using a video system (Gilestro and Cirelli, 2009; Yi et al., 2013; Zimmerman et al., 2008). Infrared LEDs were used for constant illumination for video recording. Images were acquired every second (sec) and processed by Visual C++ to determine the location of the flies. Sleep was defined as no detectable movement for 5 min or longer. Sleep parameters were calculated using Matlab (MathWorks, Natick, MA). Sleep duration, sleep bout duration, sleep latency and wake activity were analyzed for each 12-hr period of LD and DD for each condition. Female flies were sleep deprived by the mechanical method using a steering engine, which randomly shook for 20 s in 3 min (Shaw et al., 2002). Sleep deprivation efficiency was recorded by a DAM system. Sleep loss was defined as total sleep duration last night. Individual sleep rebound was quantified hourly for 24 hr by dividing the cumulative amount of sleep regained by the sleep duration last night. Each experiment was repeated twice or three times.

Male flies aged 3–6 days were housed individually in 65 mm glass tubes containing normal food. Locomotor activity was measured in TriKinetics DAM systems for 12 days in constant darkness. Periodogram analysis was completed with the Actogram J plugin (Schmid et al., 2011). Period length was determined by periodogram analysis.

Adult female flies of 5–10 days of age were collected and their CNS dissected in phosphate buffered saline (PBS). The brains were subject to 4% paraformaldehyde (wt/vol) fixation in the PBS for 2–4 hr and washed three times with the wash buffer (3% NaCl in 0.1% PBT) for 10 min at room temperature. Samples were transferred to 10% normal goat serum (vol/vol, diluted in 1% PBT) for overnight blocking at room temperature and incubated with antibodies (diluted in 10% normal goat serum and 0.1% PBT) at 4°C overnight. After washing samples three times for 15 min with wash buffer at room temperature and incubating samples with secondary antibody (1:500) at 4°C overnight, we mounted the samples with FocusClear (CelExplorer Labs CO) and imaged them on a Zeiss LSM710 confocal microscope. Images were processed using Imaris (Bitplane AG, Zurich, Switzerland, 640 RRID: SCR_007370). The following antibodies were used, chicken anti-GFP (1:1000; Abcam Cat# 13970, RRID:AB_300798), rabbit anti-RFP (1:2000; Clontech Cat# 632496, RRID: AB_10013483), rabbit anti-5-HT (1:2000; Immunostar Cat# 20080, RRID:AB_572263), mouse anti-nc82 (1:50; DSHB Cat# 2314866, RRID: AB_2314866), rat-anti Myc (1:500, Transgen). Secondary antibodies were AlexaFluor488 anti-rabbit (Life technologies Cat# A11008, RRID:AB_10563748), AlexaFluor488 anti-chicken (Life technologies Cat# A11039, RRID:AB_2534096), AlexaFluor546 anti-rabbit (Life technologies Cat# A11035, RRID:AB_2534093), and AlexaFluor633 anti-mouse (Life technologies Cat# A21052, RRID: AB_141459).

Pairwise comparisons were evaluated by Student’s t test. ANOVA with Holm-Sidak corrections for multiple comparisons was used to test hypotheses involving multiple groups. Statistical analyses were performed with Prism 5 (GraphPad).