All fly lines were maintained at 25°C under a light-dark cycle (LD12:12). Candidate genes were downregulated using dsRNAi crossed initially to the tim-gal4 driver and incorporating UAS-dicer2 into the crossing scheme to enhance the downregulation. Most of the UAS-RNAi lines were obtained from VDRC and available mutants for the genes of interest were obtained from the Bloomington stock centre.

The CPTI lines (Cambridge Protein Trap Insertions) were screened by placing 2–3 day old male flies in Trikinetics activity monitors at 25°C for 2–3 days in LD12:12 before releasing them under LL (39 μW/cm²) for a further 7–10 days. Locomotor activity was collected in 30 min time bins and rhythmicity was analyzed by autocorrelation and spectral analysis using the CLEAN algorithm. Rhythmic individuals required both analyses to be statistically significant (see Vanin et al., 2012).

Flies were maintained in Trikinetics monitors at 25°C in LD12:12 for 3 days. During the third night a 10 min light pulse (39 μW/cm²) was administered 3 h (ZT15) or 9 h (ZT21) after lights off (ZT12). The flies were kept in DD after the light pulses and locomotor behavior recorded for several days. A control group of flies of the same genotype did not receive the light pulse. Cross-correlation was used to assess the degree of phase shift by taking the locomotor data from 48 h after the light pulses i.e., the third day, and comparing each individual fly’s experimental profile against the average of the control data. This was done by shifting by one bin at a time (lag) the two sets of data against each other and calculating a correlation coefficient for each lag. The number of 30 min bins shifted that produced the maximum correlation provided the experimental phase shift, which could be either a delay or an advance. ANOVA was used to compare the phase shifts of different genotypes.

Fly heads were collected in LL at CT1, 7, 13 and 19. Western blots were performed as described previously (Sandrelli et al., 2007). α-TIM antibody (gift of Francois Rouyer, Gif, Paris) raised in rat was used at a concentration of 1:2000 with α-rat as secondary (1:10,000). α-Tubulin (1:40000) was used with α-mouse (1:6000, all Sigma-Aldrich). Image J software was used to quantify the TIM bands relative to the corresponding Tubulin.

Naturally occurring polymorphisms were studied by PCR. The ls-tim/s-tim variants were genotyped using a PCR strategy published previously (Tauber et al., 2007). We also examined polymorphisms in the jetlag (jet) gene by amplifying a 282 bp fragment of jet that may harbor two variants, jet^C and jet^R, that both cause LL rhythmicity in the ls-tim genetic background (Koh et al., 2006; Peschel et al., 2006). Both variants generate an amino acid substitution (phenylalanine to isoleucine, F209I, and serine to leucine S220L). The primers used were jet 5′-CGCGTACTCAAGCTGTCC and jet 3′-CACGCCATAGTCGGAGAT at an annealing temperature of 64°C and PCR generated fragments were directly sequenced.

This was performed for the following genotypes: Got1-YFP/qsm-gal4¹⁰⁴²⁸⁰; dsRed/+, Gs2-YFP/qsm-gal4¹⁰⁴²⁸⁰; dsRed/+, Got1-YFP/myr-RFP; repo-gal4/+, Gs2-YFP/myr-RFP; repo-gal4/+. Prior to collection at the indicated ZT22, flies from different strains were entrained for at least 2 days to LD12:12 conditions. Light intensity was ∼2500 lux. Flies with indicated genotypes were fixed in 4% paraformaldehyde/PBS (3 mM NaH2PO4, 7 mM Na2HPO4, 154 mM NaCl) for 2 h at room temperature. After fixation, the samples were washed 3× by PBS and the fly brains were dissected under a microscope. The dissected brains were collected in PBS 0.1% Triton X-100 and transferred to 4% paraformaldehyde/PBS for 30 min for post dissection fixation. Fly brains were then washed 3× in PBS 0.1% Triton X-100 and blocking with 10% goat serum in 1% PBS-T was applied for 2 h before staining with guinea pig anti-PDP-1ε (1:5000, Benito et al., 2007) in 0.3% PBS-T at 4°C for 48 hr. After washing 3× with 0.1% PBS-T, the samples were incubated at 4°C overnight with anti-guinea pig antibody conjugated with fluorophore, Alexa Fluor 647 nm (Molecular Probes) diluted 1:300 in 0.3% PBS-T. Brains were washed 4× in 0.1% PBS-T and water before being mounted in Vectashield. Samples were stored at 4°C until examination under a LSM-510 META confocal microscope (Zeiss, Germany) or a Leica SP5 confocal microscope.

Of 147 YFP lines screened, 7 showed rhythmicity under LL at levels between 50 and 94%. Figure 1 shows the average locomotor histograms for the 7 lines plus a control line and examples of individual spectral analyses of locomotor records. Table 1 provides the statistical analyses of these data. Each CPT line was also tested in constant darkness (DD) and all were rhythmic although the line CPTI00051 in which the insertion lies in the Rab11 locus also showed a significantly longer free-running period in DD (26.0 ± 0.3 h). Table 1 shows the identity of the genes trapped in each of these lines and the free-running period in LL. We then obtained the available UAS-RNAi lines or any mutants that existed for these seven loci. Table 2 shows that RNAi driven by timgal4 for Got1 (Glutamate oxaloacetate transaminase 1) and Gs2 (Glutamine synthetase 2) and the available mutant males w/Y; P(GT1)Got1/P(GT1)Got1 and wP(GT1)Gs2/Y, confirmed the LL rhythmic phenotype observed in Got1^YFP and Gs2^YFP (Supplementary Figure S1). In the other lines either the mutants could not confirm the phenotype, or the RNAi revealed high levels of LL rhythmicity, but so did their corresponding UAS-RNAi parental controls, suggesting leakage and/or a background effect (Supplementary Table S1). One interesting exception was Glycogenin (CPTI-000902) which gave extremely high levels of LL rhythmicity for both the gene trap and the RNAi (95%) although the UAS-RNAi control line also generated 56% LL rhythmicity. Several of the VDRC UAS-RNAi lines also gave high levels of rhythmicity in LL suggesting that they carry additional genetic variant(s) within the circadian light input pathway (although not those for Gs2 nor Got1, Tables 1, 2).

As both Got1 and Gs2 are involved in glutamate metabolism, we focused on these genes for the rest of the study. Mutations in both genes involved the w;P(GT1) element. To test whether the w;P(GT1) background was sensitized for LL rhythms, we examined another three w;P(GT1) insertions in Tom7, AdhAdhr, and prtp. All three w;P(GT1) lines gave low levels of rhythmicity in LL, 22, 20, and 16% respectively (Table 2). Consequently the genetic background that provided the w;P(GT1) screen was not of itself responsible for the high levels of LL rhythmicity in Got1 and Gs2.

We also crossed the Got1 w;P(GT1) mutant females to w¹¹¹⁸ (white-eyed) mutant males and then assessed the LL rhythmicity in the F2 generation where we could distinguish by eye color the mutant homozygotes, heterozygotes and the white-eyed wild-type. We observed 61% LL rhythmicity in the mutant homozygotes, 44% in heterozygotes (mean periods were 28–29 h) and considerably less LL rhythmicity in the wild-type (Table 2). When we performed a similar cross to the sex-linked Gs2 (wPGT1) mutant we observed 63% LL rhythmicity in hemizygous mutant males and 31% rhythmicity in the wild-type white-eyed F2 males (Table 2). The LL rhythmicity of both Gs2 and Got1 mutants after randomizing the genetic background in this way was ∼20% less than the ∼80% we observed with the original YFP gene traps (Table 1), suggesting further polymorphisms that were enhancing or reducing the LL effects.

The best known polymorphism in the light input pathway involves the naturally occurring s-tim/ls-tim variant, in which the ls-tim allele reduces circadian photosensitivity (Sandrelli et al., 2007; Tauber et al., 2007). We therefore addressed the tim status of the w;P(GT1) lines and w¹¹¹⁸ using PCR. The original Gs2^YFP and Got1^YFP alleles whose results are shown in Table 1 are in the ls-tim background, whereas w¹¹¹⁸ carries s-tim. Consequently, the decrease in levels of LL rhythmicity in the F2 crosses with w¹¹¹⁸ could be due, in part, to the segregating ls-tim/s-tim variant (Table 2). We repeated the crosses of each mutant to w¹¹¹⁸ and inspected the ls-tim/s-tim status of each fly after it had also been investigated for LL locomotor rhythmicity. Table 3 reveals that for both genes, LL rhythmicity is associated with either ls-tim homo- or heterozygosity. Almost all the ls-tim homozygotes and half the heterozygotes are LL rhythmic. For Gs2, 5 homozygous s-tim individuals were arrhythmic in LL. Unfortunately we did not obtain any s-tim homozygotes for Got1. Based on these results it was necessary to verify the tim status of the three “control” w;P(GT1) induced mutants Tom7, AdhAdhr, and prtp which did not show LL rhythmicity. Their tim genotyping by PCR revealed that they were all ls-tim homozygous (Table 2). Consequently the ls-tim polymorphism does not by itself cause high levels of LL rhythmicity, supporting previous studies (Sandrelli et al., 2007), yet it does enhance the LL rhythmicity of the Gs2 and Got1 w;P(GT1) variants.

The behavior of w;P(GT1) mutants for Gs2 and Got1 resembles that of Veela flies which carry the jetlag variant jet^c, and are rhythmic in LL only in the presence of the less light-sensitive LS-TIM isoform (Peschel et al., 2006). We therefore investigated the Gs2, Got1 and w¹¹¹⁸ mutants for the presence of jet^c or the rare allele, jet^r. PCR of a 282 jet bp fragment that includes both variant sites followed by sequencing did not reveal any polymorphism (data not shown).

We then genotyped the other lines for Got1 and Gs2 as well as the parental RNAi lines for the tim and jet variants. Table 2 shows that the flies that were downregulated for Gs2 with timGal4 (s-tim) would segregate both tim alleles as UAS 32929 was polymorphic, yet they gave >80% LL rhythmicity. The downregulated lines for Got1 were heterozygous s-tim/ls-tim or s-tim homozygous (Table 2). Clearly having at least one copy of ls-tim enhances LL rhythmicity, but UAS control flies that are ls-tim homozygotes do not give high levels of LL rhythmicity (see line 108247) so the very high levels we observe with Gs2 and Got1 variants represent these alleles interacting with tim. None of these lines were polymorphic for the jet alleles.

We also placed the Gs2 and Got1 alleles on the s-tim background by crossing to w¹¹¹⁸ and generating s-tim homozygous lines for Gs2 and Got1 from the F2 generation using PCR. We observed that on this genetic background, only 20 and 21% respectively of the males were rhythmic in LL, further underscoring the interaction of these two genes with the ls-tim polymorphism (Table 2).

Western blots were performed from fly heads for the w;P(GT1)Got1 and wP(GT1)Gs2 mutants and compared to Canton-S. Levels of TIM were compared with tubulin under free running LL conditions for four time points, CT1, 7, 13, and 19. Two replicate blots were performed and both gave almost identical results (Figure 2). Levels of TIM compared to tubulin cycled with a 3–5 fold peak-to-trough amplitude for both Got1 and Gs2 mutants, with a clear peak at CT19 whereas in wild-type the relative amplitude was blunted to 1.5–2 fold with a peak at either CT13 or CT19. Furthermore TIM was approximately 2–4 times as abundant in the mutants compared to wild-type under LL and very similar to wild-type flies maintained in LD cycles (Figure 2). These results reinforce the view that under LL the two mutants block the normal light-induced degradation of TIM (Myers et al., 1996) but nevertheless maintain high amplitude TIM cycling.

We also examined the effects of 10 min light pulses on the phase of the locomotor cycle in the two w;PGT1 Got1 and Gs2 mutants and compared them to the w;P(GT1)AdhAdhr mutant as a control. We observed that light pulses at ZT15 (3 h after lights off in a LD12:12 cycle) delayed the clock by 3 to 3.5 h for Gs2 and AdhAdhr, whereas the phase delay for Got1 was significantly reduced to 2 h (Figure 3). Similarly, the light pulse at ZT21 late at night caused an advance of 1.2–1.4 h for Gs2 and AdhAdhr, whereas this was again significantly reduced to 0.6 h for Got1. All these lines are in the less light-sensitive ls-tim background (Sandrelli et al., 2007); nevertheless, the smaller phase shifts under these conditions appear to be specific to Got1.

We attempted to define the relevant clock neurons that were mediating the LL rhythmicity of Got1 and Gs2 mutants. We therefore used the UAS-RNAi constructs to downregulate each gene’s expression in different clock neuronal clusters. Knockdown in the peptidergic neurons including the l-LNvs but excluding other clock neurons using the c929-Gal4 driver resulted in arrhythmicity in LL, thus excluding glutamate from within the l-LNvs from maintaining rhythms in LL (Table 4). Similarly the great majority of flies were arrhythmic in LL when the mai79-gal4 driver was used, which drives expression in the s-LNvs and the three CRY-positive LNds and possibly one of the two DN1a neurons (Grima et al., 2004). When we combined tim-gal4 with cry-gal80 and targeted expression to the DN1p, DN2, and DN3, and three LNd CRY-negative cells in the dorsal region, Got1 downregulation generated 62.5% rhythmicity in LL and 53% for Gs2. We also combined tim-gal4 with Pdf-gal80 and restricted downregulation to all the LNds and DNs. In these cases a much higher level of 73% of flies were rhythmic in LL for Got1 and a remarkable 90% were rhythmic for Gs2. This was in spite of the flies segregating the s-tim variant. The parental controls for these crosses were considerably less rhythmic in LL than in the experimental flies (Table 4). Figure 4 illustrates the correlation between LL rhythmicity and anatomical expression for the two downregulated genes. It is clear that the DNs play a major role, particularly those that are CRY-negative but with another significant component arising from the CRY-negative and possibly CRY-positive LNds.

We extended our analysis to include downregulation of the metabotropic glutamate receptor DmGluRA. We used three different VDRC RNAi lines and crossed them to tim-gal4. All three lines gave >50% rhythmicity in LL with ∼24 h periods (Table 4). We did not, however, observe any lengthening of period in these knockdowns in DD (Supplementary Table S2) as reported by Hamasaka et al. (2007), even though we used three independent UAS-RNAi lines. Glutamate serves as the precursor for the synthesis of the inhibitory GABA neurotransmitter and this reaction is catalyzed by glutamate decarboxylase (GAD). We downregulated Gad using tim-gal4 and observed a high level of LL rhythmicity. The parental controls for all these manipulations were considerably less rhythmic under LL (Table 4). We genotyped the UAS-RNAi lines, DmGluRA and Gad for tim and jet polymorphisms. All UAS-DmGluRA lines were homozygous ls-tim, whereas UAS-Gad RNAi lines were polymorphic and showed all three tim genotypes. All lines were monomorphic for wild-type jet. The w:tim-Gal-4 line they were crossed to was homozygous s-tim. Therefore the experimental flies with UAS-DmGluRA were heterozygous (ls/s-tim) whereas those with UAS-Gad were either heterozygous or homozygous for both tim alleles (Table 4).

To investigate the spatial distribution of Got1 and Gs2 in the fly brain, we examined the endogenous YFP signal in the gene-trap Got1^YFP/ + and Gs2 ^YFP/ + flies. Overall the Got1 and Gs2 derived YFP expression patterns are ubiquitous in various areas between neuropils and in the brain cortex (e.g., around medulla, lobula, and mushroom body, Figure 5). This result is consistent with the earlier high throughput study including both YFP trap lines (Knowles-Barley et al., 2010). We did not detected any gross spatial differences of expression between Got1-YFP and Gs2-YFP. Notably the expression patterns for Got1-YFP and Gs2-YFP appeared to include glial cells. To confirm this observation, we examined overlaps between YFP and the glial reporter repo-gal4 (Awasaki et al., 2008) by driving myr-RFP (membrane tethered RFP by myristoylation signal fusion, from Henry Chang). Consistently, we found Got1 and Gs2 positive cells overlapping with glial cells in brain cortex and between neuropils (Figures 6A–C).

To explore the relationship among YFP positive cells and clock neurons, we applied an antibody against PDP1ε as a nuclear marker for clock neurons (Benito et al., 2007). Both YFP signals were detected peripherally to the cell bodies of the PDP1ε positive neuron, as well as in cells near clock neurons (Figures 7, 8). Within clock neurons, we detected clear cytoplasmic YFP signals in LNvs, LNds, and DN1s (Figures 7, 8). Although the behavioral data suggested that CRY-negative dorsal clock neurons including half of DN1s, DN2s and most DN3s may be responsible for the LL phenotype, we did not detect clear cytoplasmic YFP signals in DN2s and DN3s under higher magnification (Figures 7, 8). We did observe YFP signal surrounding quasimodo (qsm +) expressing DN2s and DN3s (“×” in Figure 9 and (Chen et al., 2011). Pericellular YFP signals regularly overlapped with repo>myrRFP (asterisks in Figures 7, 8), suggesting these signals could be derived from glial cell processes which were previously identified to surround neurons (Awasaki et al., 2008; Ng et al., 2011). Taken together with the behavioral results, these data imply that the LL rhythms observed in Got1^YFP and Gs2^YFP flies may be derived from abnormal glutamate metabolism in CRY-negative DN1s, the LNds and/or glial cells.

Finally, in order to investigate glial contributions to the LL phenotype we downregulated Got1 and Gs2 in glia using repo-Gal4 with timgal4 driven RNAi as controls. We observed that entrainment in LD cycles was normal, but most of the repo-gal4 flies for both Gs2 and Got1 knockdown became arrhythmic almost immediately in LL, whereas flies carrying tim-gal4 crossed to the same UAS RNAi constructs were considerably more rhythmic in LL (Table 4). We conclude that the LL rhythmic phenotype is predominantly caused by neuronal rather than glial glutamate signaling.