A mechanism for circadian control of pacemaker neuron excitability

Abstract

Although the intracellular molecular clocks that regulate circadian (~24 hr) behavioral rhythms are well-understood, it remains unclear how molecular clock information is transduced into rhythmic neuronal activity that in turn drives behavioral rhythms. To identify potential clock outputs, we generated expression profiles from a homogeneous population of purified pacemaker neurons (LN_vs) from wild type and clock mutant Drosophila. We identified a group of genes with enriched expression in LN_vs and a second group of genes rhythmically expressed in LN_vs in a clock-dependent manner. Only 10 genes fell into both groups: four core clock genes including period and timeless, and six genes previously unstudied in circadian rhythms. We focused on one of these six genes, Ir, which encodes an Inward rectifier K⁺ channel likely to regulate resting membrane potential and whose expression peaks around dusk. Reducing Ir expression in LN_vs increased larval light avoidance and lengthened the period of adult locomotor rhythms, consistent with increased LN_v excitability. In contrast, increased Ir expression made adult flies largely arrhythmic and strongly dampened Period protein oscillations. We propose that rhythmic Ir expression contributes to daily rhythms in LN_v neuronal activity, which in turn feed back to regulate molecular clock oscillations.

Introduction

Forward genetic screens in mice and Drosophila have revealed a conserved mechanism for the molecular clocks that operate in key central brain pacemaker neurons to generate circadian (~24hr) rhythms in behavior and physiology. These molecular clocks consist of interlocked transcriptional-translational feedback loops that drive circadian rhythms in gene expression (reviewed by Hardin, 2011).

At the cellular level, circadian pacemaker neurons show electrical activity rhythms (Michel et al., 1993; Welsh et al., 1995; Cao and Nitabach, 2008; Sheeba et al., 2008). Electrical activity rhythms are presumably regulated by the molecular clock since the period of the firing rate of mammalian pacemaker neurons in the Suprachiasmatic nucleus (SCN) is altered in clock mutants with non-24hr rhythms (Liu et al., 1997; Herzog et al., 1998). However, direct links between the molecular clock and neuronal activity have proved elusive even though a number of channels and currents have been implicated in controlling pacemaker neuron activity,

For example, SCN firing frequency rhythms are likely regulated by BK/Kcnma1, a Ca²⁺-activated K⁺ channel, and by the Kv3.1/Kcnc1 and Kv3.2/Kcnc2 fast-delayed rectifier K⁺ channels (Itri et al., 2005; Meredith et al., 2006; Kudo et al., 2011). BK RNA levels are rhythmic in the SCN (Panda et al., 2002). Kv3.1 and 3.2 protein levels are also rhythmic (Itri et al., 2005), although the mechanism underlying this has not been established. Mice lacking either BK or both Kv3.1 and Kv3.2 display weakened behavioral rhythms although no altered periods were seen (Meredith et al., 2006; Kudo et al., 2011). However, BK rhythms appear to be widespread and uniformly phased throughout the SCN (Meredith et al., 2006), making it difficult to explain how BK could account for regional differences in firing phase (Jagota et al., 2000; Saeb-Parsy and Dyball, 2003). Recent studies in mammals indicate that the SK channel may be more important than BK in the subgroup of Per+ SCN neurons that show dramatic daily changes in their resting membrane potential (Belle et al., 2009). Indeed, diversity in clock neurons across the SCN may have obscured the identification of specific clock-regulated outputs. To avoid these issues, we decided to generate whole genome expression profiles from a single defined group of pacemaker neurons.

We chose to study the ventral Lateral Neurons (LN_Vs) from Drosophila. The adult small LN_Vs (s-LN_Vs) are the master circadian pacemaker neurons and set the pace for other fly clock neurons and for locomotor activity rhythms (Stoleru et al., 2005). However, we purified LN_vs from larval brains for three main reasons: (1) The four larval LN_vs in each brain lobe are differentiated neurons with functional molecular clocks and become the adult s-LN_vs. (2) Larval LN_vs modulate circadian rhythms in light avoidance, which peaks at dawn (Mazzoni et al., 2005; Collins et al., 2012) just as s-LN_vs drive morning activity in adult flies (Grima et al., 2004; Stoleru et al., 2004), suggesting that larval LN_vs are functionally similar to adult s-LN_vs. (3) The neuropeptide Pigment Dispersing Factor (PDF) distinguishes LN_vs from other clock neurons. However, in adult flies, PDF is also produced in the large LN_vs (l-LN_vs) which regulate sleep rather than circadian rhythms (Parisky et al., 2008; Shang et al., 2008; Sheeba et al., 2008; Chung et al., 2009). Thus using Pdf-Gal4 to mark larval LN_vs, we purified a homogeneous population of circadian pacemaker neurons for whole genome expression profiling, with the idea that genes expressed in larval LN_vs would also be expressed in the pacemaker adult s-LN_vs.

Materials and Methods

Isolation of larval neurons

For GeneChips, 3^rd instar larvae were kept in a standard LD (Light:Dark) cycle and dissections centered around ZT3 or ZT15. [ZT: Zeitgeber time in a 12:12hr LD cycle. Lights on at ZT0, off at ZT12.] ~200 brains were dissected for each biological replicate, which took ~90 minutes. Thus cells were isolated from a narrow time range, rather than precisely at ZT3 or ZT15. For qPCR, larvae were taken from constant darkness and ~50 brains per replicate dissected. Dissected brains were transferred to Schneider’s Insect Medium (Sigma) in non-stick tubes (Neptune) and kept on ice to minimize changes in gene expression. Brains were washed twice with cold PBS and dissociated as in Wegener et al. (2004) by transferring to a 50:50 mix of 1X Collagenase (Sigma): 1X Dispase II (Roche) and incubating for 2hr at 25°C. After 2hr, the dissociation solution was replaced with Schneider’s medium with 10% Fetal Bovine Serum (FBS). Brains were then triturated 100x with a pipette and strained through a 35μm nylon mesh filter. Trypan Blue exclusion indicated that ~90% of larval LN_vs were viable after dissociation. Cells in Schneider’s medium/10% FBS were kept on ice for transport to the NYU School of Medicine FACS center. To minimize contamination by cells attached to GFP+ or RFP+ cells, we used a size filter to remove cell clusters. Although Figure 1B shows a tail of cells with stronger GFP+ fluorescence than most, we only selected cells 2–3 orders of magnitude more fluorescent than most other cells. Cells were sorted directly into Arcturus PicoPure Total RNA extraction buffer. We analyzed RNA from 750 – 1,100 neurons for LN_v GeneChips, 1,000 or 10,000 GFP+ neurons for Elav GeneChips and 200–300 LN_vs for qPCR.

Figure 1. Identification of genes enriched in larval pacemaker neurons that are also time- and clock-dependent.

(A) Whole-mount of larval brain lobes with Pdf-Gal4 expressing UAS-CD8:GFP. GFP antibodies label four LN_vs in each lobe.

(B) FACS scatter plot shows the .01% most-GFP+ cells (gate R3). Collected GFP⁺ cells are roughly 2–3 orders of magnitude more fluorescent than GFP− cells of the same size (gate R9). GFP⁺ cells group together by size (data not shown).

(C) Gene expression in larval LN_vs isolated at ZT15 vs. Elav+ neurons. Each blue circle represents one of the 18,952 GeneChip probes covering the Drosophila genome. Fold change in LN_vs vs. Elav+ neurons is plotted on the x-axis and p-value (t-test) on the y-axis. Cutoff lines at 8-fold and .01 p-value show mRNAs with significantly enriched expression in LN_vs compared to Elav+ cells at ZT15. Core clock genes and Ir are highlighted with red diamonds, as is the de-enriched gene dac (dachsund).

(D) 24 mRNAs (23 unique genes) are rhythmically expressed and clock-regulated (clock-driven, p-value < 0.01, FDR < 8% and fold change > 2) when comparing LN_vs at ZT3 vs. ZT15 (195 mRNAs) and per⁰ vs. cyc⁰ (177 mRNAs). Plots below the pie charts show expression values for these 24 mRNAs (rows) across five genomic conditions with three replicates for each (columns). Values were standardized by mean centering (row mean = 0, row standard deviation =1) and assigned color-map values based on their standard deviation from the row mean. Rows were clustered using the Euclidean pairwise distance algorithm.

(E) Fold changes (FC) for clock-driven mRNAs. The charts show the clock-driven transcripts whose expression is higher at ZT15 than ZT3 (top) or higher at ZT3 than ZT15 (bottom). FC = fold change of the mean of 3 replicates each for ZT 3, ZT 15, per⁰ and cyc⁰. See Methods for statistical analyses. The “UP @” columns shows whether expression of that transcript was higher at ZT15 or ZT3 (left column) or in per⁰ or cyc⁰ LN_vs. (right). Expression of 10 of these 23 clock-driven genes was also significantly enriched in LN_vs compared to Elav+ neurons (indicated in bold). 4 of these 10 genes are core clock genes (indicated in red).

RNA amplification and analysis

For GeneChips, mRNA was amplified using the NuGen Ovation RNA Amplification System V2, and the resulting labeled single-stranded DNA hybridized to Affymetrix Drosophila 2.0 GeneChips. Hybridization, staining, and washing were as in the manufacturer’s protocol. The entire procedure was performed 3 times each for each genotype (Pdf-Gal4; UAS-CD8::GFP larvae at ZT3 and ZT15 and per⁰; PDF-RFP and Pdf-RFP; cyc⁰ larvae at ZT15).

For Quantitative Real Time PCR (qPCR), we used an amplification strategy (WT-Ovation^™ Pico System) optimized for smaller amounts of input total RNA than above to generate ~5μg of single stranded amplified unlabeled cDNA product. For each qPCR reaction, 20ng of cDNA was amplified in a Roche LightCycler. Forward and Reverse primers (F and R) and hybridization probes (P1 and P2) for quantitative PCR were designed so that one primer or probe spanned an exon/intron boundary to ensure that contaminating genomic DNA was either not amplified or measured. RNA levels were determined by comparing the time when the reaction moved into detectable exponential phase to standard curves for each primer set constructed by re-amplifying known quantities of PCR products. We normalized the absolute level of each gene in an experiment to RNA levels of Pdf (a non-cycling transcript) in that same experiment. qPCR results are an average of two or three independent experiments. For each time series plotted, the maximum value was set to 1, and other values are expressed as a fraction of the maximum. Primer sequences are in Supplementary Information.

GeneChip Data Analysis

Raw hybridization intensities from each CEL Affymetrix file (Drosophila 2.0 GeneChip) were analyzed using Matlab and accompanying bioinformatic toolbox. For the initial processing step, we used the gcrma algorithm (Irizarry et al., 2003) which incorporates a measure of nonspecific hybridization to compute an adjusted perfect match (aPM) intensity for each probe. aPMs were then quantile-normalized (Irizarry et al., 2003) across replicates and experiments to yield a single expression measure for each probeset.

We removed low-intensity value probesets (lower 20^th percentile)prior to testing for differential expression and a second filter removed probesets showing minimal-variance across all conditions (lower 20^th percentile). Non-specific filtering when profiling homogenous tissue reduces noise introduced by the significant percentage of non-expressed and under-expressed genes, and thus improves sensitivity for detecting truly differentially-expressed genes (Blalock et al., 2003; Calza et al., 2007; Keegan et al., 2007).

Differential expression was determined using statistical analysis of variance (permuted Student’s t-test) adjusted for multiple-hypothesis testing. To estimate false discovery rate, we followed the procedure introduced by (Storey and Tibshirani, 2003) to compute q-values and positive false discovery rate FDR (pFDR). We applied cutoffs for differentially-expressed genes of p < .01 and FDR < 8%. For genes meeting these statistical criteria, we further imposed fold change cutoffs: 8-fold for enrichment and 2-fold for time-dependence or clock-regulation. Correlations between the 3 replicates for each condition were as follows with numbers indicating highest and lowest correlations: ZT3 (0.92–0.88), ZT15 (0.91–0.88), per⁰ (0.95–0.92), cyc⁰ (0.94–0.87), Elav (0.89–0.87). Raw GeneChip data has been uploaded to GEO and is accessible via: http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=nbwdnouyqyquobg&acc=GSE35752

Fly stocks and Immunocytochemistry

The Pdf-RFP transgene has 0.6kb of Pdf regulatory genomic DNA (0.5kb upstream the start site of transcription and 0.1kb downstream) fused to DNA encoding mRFP1, a monomeric soluble red fluorescent protein (Shaner et al., 2004), generously provided by Roger Tsien. DNA was injected into y w flies by the MGH CBRC Transgenic Drosophila Core. We thank Ben Collins and Dave Reeves for these flies. Other fly strains are described in Supplementary Information. The P-element in Ir^d flies was excised using standard procedures. Immunodetection of whole-mount adult brains was as previously described using a monoclonal mouse antibody to PDF (Cyran et al., 2005), guinea-pig anti-VRI (generously provided by Paul Hardin), rabbit anti-GFP (Invitrogen) to detect nYFP and rabbit anti-PER (generously provided by Jeff Hall). PER levels were measured in 5 s-LN_v clusters (each from a different brain) with typically 2–4 s-LN_vs per cluster in Figure 4.

Figure 4. Increasing Ir expression disrupts adult circadian behavioral and molecular rhythms.

(A) A P-element inserted in Ir (Ir^d) makes ~50% of flies lose behavioral rhythms and this phenotype is rescued by excision of the P element (Ir^{d rev}).

(B) Ir RNA levels in RNA isolated from whole fly heads are higher in Ir^d flies than in y w or Ir^{d rev} control flies (p< 0.05, t-test). The data are an average of 4 biological replicates isolated at ZT11. Error bars show SEM.

(C) Representative images of PER protein levels in adult s-LN_vs of Ir^{d rev} control flies (top panels) and Ir^d flies (bottom panels) at 4 time points on the 5^th day in DD. The locomotor activity of the flies stained here was not assayed and so the Ir^d flies are likely a mix of rhythmic and arrhythmic animals.

(D) Quantitation of data from C normalized to peak PER levels at CT23 in Ir^{d rev} controls. Error bars show SEM. PER levels are lower at CT23 in Ir^d (red) than in Ir^{d rev} s-LN_vs (blue, p < 0.05, t-test).

Behavioral Analysis

To assay larval light avoidance, we followed the method of Mazzoni et al. (2005) in which 15 larvae roam for 15min on a 1% agar surface in a Petri-dish. We used a light box (developed by Alex Keene) with a fluorescent tube lighting the assay plate from below that is blocked from transmitting to half the Petri dish by non-translucent dividers placed between the light source and assay plates. Light intensity was reduced to 30lux by increasing the distance of the plates from the light source using these dividers and adding neutral transmittance filters (Roscolux #397: Pale Grey). Larvae on both sides were counted after 15min. Four plates were run in parallel, with each plate representing an individual experiment. Each data point is the mean of 12 plates.

For adult circadian assays, adult flies were placed in Trikinetics locomotor assay cuvettes and entrained to light-dark (LD) cycles for at least 3 days prior to shifting to constant darkness (DD) for at least 7 days. We calculated the period of each locomotor rhythm (τ), by chi-squared periodogram analysis as previously (Price et al., 1998). The power of the rhythm is the height in arbitrary units of the periodogram peak, which quantifies the rhythm strength and is only reported for flies meeting the significance threshold (p < .01). Visual inspection of actograms was also used to determine percent rhythmic flies in Figure 4.

Results

GeneChips identify transcripts enriched in larval LN_vs

To purify master pacemaker neurons, we dissected brains at ZT3 or ZT15 from larvae with the LN_v-specific driver, Pdf-Gal4 (Park et al., 2000) expressing GFP (Figure 1A). Larval brains were dissociated into a single-cell suspension and flow cytometry used to select single GFP+ larval LN_vs. We typically obtained ~900 LN_vs from ~200 larval brains. mRNA was purified, amplified and the resulting labeled single-stranded DNA hybridized to GeneChips (Figure 1). We performed this procedure three times for each timepoint.

Before examining differences in gene expression in larval LN_vs between ZT3 and ZT15, we wanted to validate the procedure. We compared LN_v gene expression profiles with those from a heterogeneous group of larval brain neurons that produce the post-mitotic marker Elav. mRNAs whose expression was increased in LN_vs compared to Elav+ neurons were termed “enriched” (see Experimental Procedures). Pdf was the most highly enriched transcript in LN_vs: 610-fold more abundant at ZT15 in LN_vs than in Elav+ neurons (Figure 1 and Table 1a). The core clock genes period (per), timeless (tim), vrille (vri) and PAR-domain protein 1 (Pdp1) were all in the top 130 enriched mRNAs at ZT15 (close to their peak expression time) with FDR < 8%, p-values (t-test) <0.01, and at least 8-fold enrichment. The inclusion of these genes validates the purity of the sample.

Applying these cutoffs, we identified 95 mRNAs enriched in LN_vs at ZT3, and 153 mRNAs at ZT15 (Table S1), with 57 mRNAs enriched at both time points. Several of the enriched mRNAs have known roles in LN_vs. In addition to per, tim, vri and Pdp1 and the circadian photoreceptor cryptochrome (cry), the transcription factor Mef2 (53-fold enriched at ZT15) is produced in larval and adult LN_vs and regulates adult circadian rhythms (Blanchard et al., 2010). Thus the LN_v purification and RNA amplification method seems to work well.

An alternative strategy involving manual picking of Drosophila clock neurons was recently described (Nagoshi et al., 2010). We found 16 of the 63 LN_v-enriched mRNAs found by Nagoshi et al. (2010) in our ZT15-enriched datasets and these are underlined in Table S1. This overlap is considerably more than would be expected by chance and there are major differences between the two studies: Nagoshi et al. (2010) pooled data from larval and adult LN_vs (including l-LN_vs) and compared these data to a pooled reference sample of larval and adult Elav+ neurons as well as non-LN_v clock neurons. In addition, they isolated LN_vs at ZT12, 3hr earlier than our timepoint, and timing is very important in LN_vs: 61% (94/153) of the mRNAs we found enriched at ZT 15 were not enriched at ZT 3 (and see below for time-dependent LN_v expression).

GeneChips identify transcripts whose levels are time-dependent in LN_vs

CLK/CYC-regulated transcripts are low in larval LN_vs at ZT3 (early morning) and high at ZT15 (early evening). Therefore, we compared LN_v expression profiles at these time points to identify genes whose expression is time-dependent. As expected, expression of the core clock genes per, tim, vri and Pdp1 was higher at ZT15 than ZT3 (Figure 1D, E and Table S2). We identified 195 mRNAs whose expression is 2-fold different in LN_vs between ZT3 and ZT15 applying the same statistical cutoffs for p-value and FDR as for enrichment (Table S2). 87% (161/195) of these mRNAs are not enriched transcripts and are presumably expressed broadly in the larval brain as well as in a time-dependent manner in LN_vs. Only 11 of the 195 time-dependent genes we identified in larval LN_vs (including per, tim, vri, and Pdp1) are among the rhythmic genes identified by meta-analyses of the multiple whole Drosophila head microarray studies (Wijnen et al., 2006; Keegan et al., 2007), validating the purification of LN_vs prior to GeneChip analysis. A recent study came to similar conclusions about rhythmic gene expression in adult LN_vs: Kula-Eversole et al. (2010) found that most rhythmically expressed genes in l-LN_vs do not show rhythmic expression in whole fly heads.

A subset of genes whose expression is time-dependent are clock-driven in LN_vs

Rhythmically expressed genes could be driven by LD cycles rather than by the molecular clock (Wijnen et al., 2006). To distinguish light-regulated from clock-regulated genes, we isolated LN_vs from per and cyc null mutants. These mutants have opposite effects on the molecular clock since PER protein represses CLK/CYC activity (reviewed by Hardin, 2011). Therefore, if any of the 195 time-dependent mRNAs are clock-regulated, their expression should differ between per⁰ and cyc⁰ larval LN_vs. Applying the same statistical cutoffs as for the ZT3 vs. ZT15 comparison, we identified 177 “clock-regulated” mRNAs whose expression differs between per⁰ and cyc⁰ LN_vs (Table S3 and Figure 1D). 24 mRNAs have both time-dependent (ZT3 vs. ZT15) and clock-regulated expression (per⁰ vs. cyc⁰), which we term “clock-driven” (Figure 1D–E). As expected, this group includes the known direct CLK/CYC target genes: per, tim, vri and Pdp1.

Two surprising omissions from the clock-driven list are cry and Clk. cry expression was more highly enriched at ZT3 than at ZT15, but did not meet the statistical cut-offs for time-dependency. This presumably reflects its delayed time to reach peak levels in larval LN_vs since, as described below, qPCR from isolated LN_vs revealed higher cry RNA levels at CT10 than CT4 in larval LN_vs. [CT: Circadian time: time in DD after prior LD cycles]. Thus our list of clock-driven LN_v transcripts is presumably incomplete for this reason and because other transcripts (such as Clk) are lowly expressed and/or difficult to amplify.

The lack of overlap between the genes showing time-dependent and clock-dependent expression is also surprising and remains to be fully explained. Interestingly, per, tim, vri, and Pdp1 are the only overlap between the 24 clock-driven genes in larval LN_vs and the ~100 genes with circadian expression identified by a meta-analysis of multiple whole head microarray studies (Wijnen et al., 2006). While this supports the idea that novel information comes from using a single cell type for expression profiling, added power will come from additional time points and replicates. Nevertheless, the spatial-resolution in the LN_v dataset and resulting small list of clock-driven LN_v genes will permit detailed studies of their contribution to LN_v function.

Circadian gene expression profiles

To validate the GeneChip data, we focused on genes whose expression is clock-driven and enriched. These are the four core clock genes (per, tim, vri and Pdp1) and six genes previously unstudied in LN_vs: Ir, which encodes an Inwardly rectifying K⁺ channel; CG33275, predicted to encode a Rho-GEF; Pka-C3 and three uncharacterized genes (CG14521, CG42238 and CG14853).

Figures 2A and 2B show the expression profiles for 9 of the genes that are both clock-driven and enriched as well as for Pdf, a non-cycling enriched gene. Rhythmic expression in LD and clock-regulation made it likely that all of these genes would also be rhythmically expressed in constant darkness (DD), the hallmark of circadian gene regulation. This has already been demonstrated in larval and/or adult LN_vs for tim, vri, Pdp1 and cry RNA (Price et al., 1998; Yang and Sehgal, 2001; Peng et al., 2003; JB, unpublished data). To validate rhythmic LN_v expression for the previously unstudied genes and to test for rhythms in DD, LN_vs were isolated from wild-type larvae at four timepoints on the second day in DD after prior entrainment to LD cycles. RNA was isolated from FACS-sorted LN_vs as for the GeneChip experiments. mRNA was then amplified using a method that includes random priming and expression levels were measured by qPCR. Thus the amplification and quantification strategies differ from the GeneChip studies, which used a poly-T driven amplification.

Figure 2. Clock-driven transcript profiles across time, genotype, and cell type.

(A) Hybridization signal (y-axis) for the CLK/CYC direct target genes per, tim, vri and Pdp1. The signal represents the mean normalized and log-transformed hybridization intensity for the 3 replicate experiments at each timepoint. Error bars indicate standard error of the mean (SEM). The hybridization signal to Pdf is also shown.

(B) Signals for five additional enriched and clock-driven genes in LN_vs.

(C) Real-time qPCR on amplified LN_v RNA for Ir (blue diamonds), Pdp1 (purple triangles), cry (grey squares) and per (green squares) at four time points on day 2 of constant darkness (circadian time, CT). For all qPCR experiments, RNA levels were normalized to non-cycling levels of Pdf, with the maximum value set to 1.0 in each time series. Each data point is the average of two or three independent biological replicates with error bars representing SEM. For each mRNA shown here and in panel D, we tested for a significant difference in expression (one-way ANOVA) across the time series. Significant differences in expression was detected for all transcripts (p < .05) except for per and Pka-C3, which showed differences only between peak and trough by t-test only (p < .005 for per and p < .01 for Pka-C3).

(D) Real-time qPCR curves as in C for five additional enriched clock-driven genes: Pka-C3 (blue diamonds), CG14251 (orange squares), CG14853 (green triangles), CG33275 (red squares) and CG44238 (black squares).

Figure 2C shows the results of qPCR using primers to amplify cry, Ir, per and Pdp1 from RNA isolated at four different time points in DD. The results show circadian rhythms in gene expression for all four genes. We detected peak levels of cry at CT10 and trough levels at CT16, whereas Ir, per and Pdp1 RNAs have a different phase with peak levels at CT10/CT16 and trough levels at CT4 and CT22. The results in Figure 2D also show robust oscillations in DD for all 5 genes: Pka-C3, CG14853 and CG14521 are all at much higher levels at CT10 than CT22 (as seen for cry); CG33275 seems to oscillate in phase with per, Ir and Pdp1, with high levels at CT16 and trough levels at CT4. CG42238 RNA levels are highest at CT22. These qPCR data extend the conclusions from the GeneChip experiments since strong circadian rhythms in gene expression were detected for all 6 novel LN_v genes.

Altered expression of Ir in LN_vs affects circadian behavior

Clock-driven oscillations in Ir expression could be one mechanism by which LN_vs generate circadian rhythms in excitability since Inward rectifier K⁺ channels regulate resting membrane potential in mammals (Hille, 2001; Kuhlman and McMahon, 2006). More Inward rectifiers at the cell membrane should increase K⁺ efflux, hyperpolarizing the cell, while decreased Ir levels should decrease K⁺ efflux and depolarize the cell. The temporal profile of Ir expression (Figure 2C) is consistent with electrophysiological measurements of adult s-LN_vs, which become progressively depolarized towards the end of the night when Ir expression is low, and then start to hyperpolarize again after dawn when Ir RNA levels begins to rise (Cao and Nitabach, 2008).

We first tested a role for Ir in LN_v function using larval light avoidance. In this assay, larvae are placed in the middle of a Petri dish half of which is exposed to light and the number of larvae on the dark side is counted 15 minutes later. Wild type larvae show a robust preference for the dark side, with ~70% in the dark after 15min at bright light. LN_vs are downstream of the larval visual system and their neuronal activity rapidly increases with light exposure. This response is elevated when LN_vs are made hyper-excitable by expressing NaChBac, a low threshold voltage-gated sodium channel (Yuan et al., 2011). This likely explains how larvae with Pdf-Gal4 expressing NaChBac in LN_vs avoid low light levels much better than control larvae (Collins et al., 2012; Figure 3A).

Figure 3. Reduced Ir levels in LN_vs affect larval and adult behavior.

(A) Foraging 3rd instar larvae were tested for avoidance of the light half (30 lux) of a Petri dish. The percentage of larvae on the dark side of the dish after 15 min is shown as a measure of light avoidance. Control y w, UAS-NaChBac/+ and UAS-Ir:RNAi/+ larvae do not avoid light at this intensity and are not significantly different from each other (p>0.1 for all comparisons, t-test). Red dotted line indicates 50% light avoidance i.e. when larvae cannot differentiate between 30lux light and darkness. Larvae with the Pdf-Gal4 driver and either UAS-NaChBac2 (Pdf > NaChBac) or UAS-Ir:RNAi (Pdf > Ir:RNAi) avoid light significantly better than the three control strains (p<.01). These experiments were performed between ZT 14 and ZT 16. Bars show averages of 12 plates each with ~15 larvae in each plate. Error bars show SEM.

(B) Real-time qPCR on amplified LN_v RNA from either control larvae (UAS-Ir:RNAi/+) or with Pdf-Gal4 expressing UAS-Ir:RNAi (Pdf > Ir:RNAi) at ZT15. RNA levels were normalized to noncycling levels of Pdf, with the maximum value set to 1.0. Ir RNA levels are lower in Pdf > Ir:RNAi than in control LN_vs (p < 0.0005, t-test). Data are the average of two independent experiments. Error bars show SEM.

(C) Actograms showing adult locomotor activity in DD. Top panels: Pdf-Gal4/+ controls have a shorter period than flies with Pdf-Gal4 expressing UAS-Ir:RNAi (Pdf > Ir:RNAi, p < .01). Pdf > Ir:RNAi flies also have longer periods than flies with Pdf-Gal4 expressing a control RNAi (p < .01, Table S4). Bottom panels: Flies with tim-UAS-Gal4 expressing UAS-Ir:RNAi in all clock neurons (tim > Ir:RNAi) have longer periods than the tim-UAS-Gal4, Pdf-Gal80 combination expressing UAS-Ir:RNAi in all clock neurons except LN_vs (tim, Pdf-Gal80 > Ir:RNAi, p < .01) or control tim-UAS-Gal4/+ or UAS-Ir:RNAi/+ flies (p < 0.01, Table S4). Period (chi-squared analysis) and SEM are reported for flies with significant activity rhythms.

(D) Immuno-stainings with UAS-nuclear-YFP (Kimura, 2005) reporting expression from the Ir-Gal4 enhancer trap. Adult brains were dissected at ZT15 and stained using antibodies to GFP (to detect YFP, green), PDF (to mark LN_vs, red) and VRI (to mark all clock neurons, blue). Top panels show LN_vs. Nuclear YFP was detected in all four PDF+ s-LN_vs in all 10 brains examined, in 2–3 of the 5 l-LN_vs but never in the 5^th PDF- s-LN_v. YFP was detected in 1 of the 6 dorsal Lateral Neurons (LN_ds, lower left panel), but not in any dorsal neuron groups (DN₁, DN₂ or DN₃, lower right panel).

We tested the effect of knocking-down Ir expression specifically in LN_vs using Pdf-Gal4 to express a UAS transgene that expresses RNAi directed to Ir (UAS-Ir:RNAi) at ZT15, when Ir expression is high. The results in Figure 3A show that knocking-down Ir in LN_vs made larvae super-sensitive to low light (30lux), compared to control UAS-Ir:RNAi or Pdf-Gal4 larvae. The similar light avoidance phenotypes of larvae in which LN_vs either express NaChBac or Ir:RNAi are consistent with the idea that Ir levels in LN_vs are inversely related to LN_v excitability.

To quantify the efficacy of Ir:RNAi, LN_vs were isolated by FACS from either control larvae (UAS-Ir:RNAi alone) or with Pdf-Gal4 expressing UAS-Ir:RNAi at ZT15. mRNA was amplified and gene expression measured by qPCR. The results in Figure 3B show a 3.8-fold decrease in Ir transcript abundance in larval LN_vs expressing Pdf-Gal4 and UAS-Ir:RNAi, verifying that the UAS-Ir:RNAi transgene reduces Ir RNA levels in LN_vs.

Altered Ir expression in adult LN_vs changes circadian locomotor activity

One reason for profiling larval LN_vs was to identify key genes expressed in adult s-LN_vs without the confounding issue of co-purifying l-LN_vs. To test a role for Ir in adult LN_vs, we measured the locomotor activity rhythms of flies with Pdf-Gal4 expressing UAS-Ir:RNAi. These flies had strong 25.0hr long period locomotor rhythms in DD compared to control flies with periods between 23.6 and 24.5hr (Figure 3C, Table S4). Long-period behavioral rhythms are also seen when LN_vs are hyper-excited via UAS-NaChBac, which leads to a dominant 25.5hr behavioral period as well as complex behavioral rhythms (Nitabach et al., 2006), or via UAS-dnATPase [a dominant-negative Na⁺/K⁺-ATPase α subunit (Sun et al., 2001), Table S4]. The similar period phenotypes of flies with Pdf-Gal4 expressing UAS-Ir:RNAi, UAS-NaChBac or UAS-dnATPase are consistent with decreased Ir expression increasing LN_v excitability and further support the idea that Ir expression levels regulate LN_v excitability.

A similar long period rhythm was observed when Ir-RNAi expression was widened to include all adult clock neuron groups using the tim(UAS)-Gal4 driver (Figure 3C, Table S4). This was surprising because tim(UAS)-Gal4 usually gives stronger phenotypes than Pdf-Gal4 and because hyper-exciting all clock neurons with tim(UAS)-Gal4 and UAS-NaChBac makes flies arrhythmic (Collins et al., 2012). This suggested that Ir might not play a role in other clock neurons. To test this idea, UAS-Ir:RNAi was expressed in all clock neurons except LN_vs, using the tim(UAS)-Gal4 driver in combination with a Pdf-Gal80 transgene that inhibits Gal4 activity specifically in LN_vs (Stoleru et al., 2004). The results in Figure 3C and Table S4 show that flies in which Ir:RNAi was restricted to non-LN_v clock neurons have normal circadian rhythms. This contrasts with expressing NaChBac only in non-LN_v clock neurons, which dramatically constricts the time when flies are active (Collins et al., 2012). Thus Ir does not seem to regulate non-LN_v clock neurons and is perhaps not even expressed in non-LN_v clock neurons.

To test where Ir is expressed, we used a Gal4 enhancer trap inserted in Ir (NP2554, hereafter referred to as Ir^Gal4). Ir-Gal4 flies were crossed to a UAS-nuclear-YFP transgene (Kimura, 2005). Brains were dissected at ZT15 and stained for YFP, PDF (to mark LN_vs) and VRI (to mark all clock neurons). Nuclear YFP was detected in all four adult PDF+ s-LN_vs (Figure 3D) in all 10 brains examined. We found YFP expressed in 2–3 of the 5 l-LN_vs but not in the 5^th PDF- s-LN_v. We found YFP expressed in one of the six LN_ds (Dorsal lateral neurons), but in none of the three dorsal neuron (DN) groups, although it is difficult to completely exclude expression in DNs since they are numerous and reside in different focal planes. The relatively broad expression of Ir-Gal4 in the brain is consistent with Ir being one of only 3 Inward Rectifier K⁺ channels in the Drosophila genome and underlines the importance of purifying LN_vs prior to expression profiling to detect rhythmic Ir expression. Although this Ir^Gal4 enhancer trap line might not precisely reflect Ir expression, the pattern seen is consistent with the genetic data above indicating that s-LN_vs are the main locus where Ir acts in circadian rhythms. This interpretation is consistent with Nagoshi et al., (2010) who detected high-level Ir expression in larval and adult LN_vs but not in other clock neurons. In addition, Kula-Eversole et al., (2010) found that Ir is rhythmically expressed in adult LN_vs.

Increased Ir expression disrupts behavioral rhythms and the LN_v molecular clock

As an additional test of the importance of Ir in circadian rhythms, we assayed a novel Ir P-element insertion line (Ir^d08240, referred to hereafter as Ir^d). We found that 47% of Ir^d flies were arrhythmic. This phenotype was reverted by excising the P-element (Ir^{d rev}, Fig. 4A). To understand how Ir^d affected Ir expression, we measured Ir levels in RNA extracted from whole fly heads. The results in Fig. 4B show a modest but significant increase in Ir RNA in Ir^d flies compared to control (1.6–1.8-fold). Ir^{d rev} flies had the same Ir expression levels as control flies, indicating that Ir over-expression likely underlies the behavioral defects. However, we have not determined how strongly Ir is over-expressed in Ir^d LN_vs. Heterozygous Ir^d flies also have a higher incidence of arrhythmicity than control flies, consistent with increased Ir expression via this gain-of-function mutation (DM & JB, data not shown).

Since the Ir^d mutation does not target LN_vs specifically, we measured molecular clock oscillations in LN_vs to test if these are affected by increased Ir expression. Over-expressing Ir in LN_vs should hyper-polarize these cells. Interestingly, the behavioral phenotype of Ir^d mutants is similar to LN_vs expressing the dORKΔ K⁺ channel (Nitabach et al., 2002), which causes LN_v molecular clocks to run-down with the constitutively-expressed Pdf-Gal4 driver (Nitabach et al., 2002; Nitabach et al., 2005). A mammalian inward rectifier K+ channel (mKir2.1) expressed in LN_vs also hyperpolarized LN_vs and caused their molecular clocks to run down (Nitabach et al., 2002; Nitabach et al., 2005; Wu et al., 2008), leading to the idea that the LN_v molecular clock requires LN_v membrane activity for robust rhythms in DD. However, this idea was challenged when Depetris-Chauvin et al., (2011) found that expressing mKir2.1 only in adulthood via an inducible Pdf-Gal4 made flies arrhythmic but did not dramatically affect their molecular clocks. However, it has not yet been tested how strongly the constitutive and inducible mKir2.1 expression affect LN_v resting membrane potential. The long periods seen with Ir:RNAi (Fig. 3C) suggest that Ir, as an endogenous ion channel, likely affects molecular clock oscillations and contributes to period determination. To test how Ir over-expression affects the LN_v molecular clock, we measured PER protein rhythms in the s-LN_vs of Ir^d and Ir^{d rev} adult flies on day 5 in DD. The results in Figure 4C–D show that peak PER protein levels at CT23 are reduced in Ir^d flies (t-test, p<0.05). Taking these data together with the modest period-lengthening in Ir:RNAi flies leads us to propose that Ir is not just a clock output, but likely affects LN_v excitability which, in turn, feeds back to regulate the molecular clock by, as yet, unidentified mechanisms.

Discussion

Expression profiling of Drosophila pacemaker neurons

Circadian rhythms offer one of the best examples of how genes regulate animal behavior. However, there have been relatively few insights into how molecular clocks control pacemaker neuronal activity rhythms. Since circadian rhythms in mRNA levels are widespread across organisms (e.g. Harmer et al., 2000; Storch et al., 2002; Wijnen et al., 2006), rhythmic expression of key output genes that underlie rhythmic pacemaker neuron activity is an attractive idea. Our approach to identifying clock-regulated output genes differed from most previous circadian expression profiling studies by starting with a homogeneous population of behaviorally-relevant pacemaker neurons. Single-cell type expression profiling is clearly a powerful approach because the vast majority of rhythmically expressed transcripts identified in larval LN_vs (this study) or adult l-LN_vs (Kula-Eversole et al., 2010) differ from those showing rhythmic expression in whole fly heads.

Ir contributes to the regulation of LN_v excitability

We focused on Ir because of the well-described roles of Inward Rectifier K⁺ channels in resting membrane potential and neuronal activity (Hille, 2001). Since Ir transcript levels are high at CT10 and CT16 (around dusk) and low at CT22 and CT4 (around dawn), LN_vs should be less excitable at dusk than at dawn. This inference correlates well with electrophysiological measurements of adult s-LN_v resting membrane potential, which becomes progressively depolarized towards the end of the night (Cao and Nitabach, 2008). Low Ir expression in cyc⁰ mutants and high expression in per⁰ mutants is consistent with data indicating that larval LN_v excitability is high when CLK/CYC activity is low (Collins et al., 2012). Ultimately these ideas will require measuring when Ir protein is present and functional. We note that the ~365aa cytoplasmic C-terminus of Ir has 63 Serines and Threonines, suggesting that phosphorylation may also regulate Ir activity. Low Ir RNA levels in LN_vs at dawn, and thus high LN_v excitability, are consistent with LN_vs promoting: (i) larval light avoidance, which peaks around dawn (Mazzoni et al., 2005), and (ii) the morning peak of adult locomotor activity in LD cycles (Grima et al., 2004; Stoleru et al., 2004). The presence of multiple E boxes (potential CLK/CYC binding sites) in the Ir regulatory region (MR & JB, data not shown), its phase of expression and the effect of the per⁰ and cyc⁰ mutations on Ir RNA levels support the idea that Ir is a direct CLK/CYC target, although this remains to be formally tested.

Although Ir is rhythmically expressed in LN_vs, we propose that other clock neurons have different mechanism(s) to link their molecular clocks to neuronal activity. Larval LN_vs and DN1s have opposite relationships between CLK/CYC activity and excitability: LN_vs are most excitable when CLK/CYC activity is low and DN1s are most excitable when CLK/CYC activity is high (Collins et al., 2012). Cell-type specific clock outputs have already been observed in mammals, where similarly phased molecular clocks lead to rhythmic expression of outputs that differ extensively between tissues (e.g. Storch et al., 2002). Similarly, we propose that CLK/CYC regulate distinct sets of output genes in different clock neurons.

A holistic model for regulating LN_v output

While Ir affects circadian behavior, the rhythmic physiology of LN_vs probably derives from multiple genes and/or redundant mechanisms, perhaps explaining why forward genetics has revealed little about clock outputs. Our LN_v GeneChips identify additional clock-driven genes with potential roles in clock output, including amon, an enzyme that cleaves inactive neuropeptide precursors, Snap (Soluble NSF attachment protein), which helps recycle synaptic vesicles from the plasma membrane and Rdl, a GABA-activated chloride channel.

It will be interesting to determine how similar the mechanisms for regulating circadian rhythms in excitability are between Drosophila and mammals. For the SCN, rhythms in excitability can be sub-divided into rhythms in resting membrane potential and action potential frequency (Kuhlman and McMahon, 2006). No channels have yet been identified that regulate SCN membrane potential, but firing frequency is likely regulated by BK, Kv3.1 and Kv3.2 K⁺ channels (Itri et al., 2005; Meredith et al., 2006; Kudo et al., 2011). Although mice lacking either BK or both Kv3.1 and Kv3.2 display weakened behavioral rhythms, no altered periods were seen (Meredith et al., 2006; Kudo et al., 2011). However, manipulating the channel(s) that determine resting membrane potential in SCN neurons may change period length because artificially hyper-polarizing SCN slices in vitro with low K⁺ altered period and/or lead to loss of mPer1-luciferase rhythms (Lundkvist et al., 2005), suggesting a link between membrane potential and the molecular clock. At least for Drosophila, the period-altering phenotypes with Ir knockdown and LN_v hyper-excitation (e.g. via NaChBac) blur the conventional distinction between clock outputs and inputs.

In conclusion, whole-genome profiling of purified LN_vs presents an exceptional opportunity to link the transcriptional profile of a single neuronal type to animal behavior. Our results also suggest multiple levels of regulating LN_v activity and give us a short list of genes for targeted manipulation in LN_vs. Generating transcriptional profiles of individual neuronal types should be broadly applicable in neurobiology to analyze how defined neuronal populations use regulated transcription to shape, for example, sex-specific behavioral traits (Ryner et al., 1996) and other complex cognitive processes (Klausberger and Somogyi, 2008).