Drosophila Clock Is Required in Brain Pacemaker Neurons to Prevent Premature Locomotor Aging Independently of Its Circadian Function

We first examined the effects of genetic and environmental clock disruptions on Drosophila lifespan and ARLI, assessed by monitoring the performance, at successive ages, of groups of flies in a startle-induced negative geotaxis (SING) assay. We found that the Clk^AR, cyc⁰ and tim⁰ mutations each had a similar significant but moderate impact on survival, reducing lifespan by at most 15% (S1A and S1B Fig and S1 Table). In Drosophila, ARLI was found to be faster in arrhythmic per⁰ mutants [8, 31]. Here we observed that ARLI was also moderately but significantly accelerated in the arrhythmic tim⁰ and cyc⁰ mutants, after 4 weeks of adult life (Fig 1A). Rearing Canton-S flies (controls) in LL, which stops the clock, produced a similar ARLI acceleration (Fig 1B and S1C Fig). This could be consistent with accelerated ARLI and reduced survival being general detrimental consequences of long-term circadian arrhythmicity.

ARLI in the arrhythmic Clk^AR mutant, however, was strikingly more precocious, starting as early as 10 days after adult eclosion (Fig 1C). Interestingly, locomotor decline was less rapid but still significantly faster, compared to control flies, in Clk^AR/+ heterozygous flies, which are fully rhythmic [[32] and S2 Table], while it was not affected in cyc⁰/+ heterozygotes (Fig 1D). The impact of the Clk^AR mutation on ARLI thus appears separate from its effect on rhythmicity.

In contrast, average spontaneous locomotor activity in LD during the day displayed little change between young (10- to 15-day-old) and old (31- to 36-day-old) flies in any of the four tested genotypes (S1D Fig). In young Clk^AR and cyc⁰ flies, night-time activity was higher than in controls, as previously reported for Clk^AR [32], or than in tim⁰ mutants. Although night-time activity strongly decreased with age for both Clk^AR and cyc⁰ mutants, it did not become lower than in the controls (S1D Fig).

Circadian disruption also affects sleep, and sleep disruptions by themselves negatively impact the brain [33], and may thus contribute to accelerated ARLI. Sleep in LD was similarly affected in the Clk^AR and cyc⁰ mutants, with strongly increased latency after lights-off, and reduced total and night-time sleep (S2A–S2C Fig). These results are similar to those previously reported for Clk^Jrk and cyc⁰ [34]. However, the sleep profiles of the heterozygous mutants, Clk^AR/+ and cyc⁰/+, were indistinguishable from controls (S2D Fig). In addition, knocking down wake (a gene involved in the transition from wake to sleep at the end of the day) in the LNvs, which was shown to disrupt sleep [34], did not affect the age-related impairment of the SING behavior (S2E Fig).

The Clk^AR mutation therefore produces a specific ARLI phenotype, which appears independent from its disrupting effect on circadian rhythmicity, and which does not correlate with accelerated global aging, disrupted sleep, or a major loss of motor ability.

In LL, ARLI was not significantly affected by the cyc⁰ or tim⁰ mutations (Fig 1E), suggesting that ARLI acceleration in these mutants is due to arrhythmia. In contrast, in LL ARLI was still much faster for the Clk^AR mutant than for control flies (Fig 1F). The difference between the Clk^AR mutant and the control in LL could be considered as representing the rhythm-independent effect of the Clk^AR mutation on ARLI.

Analyses of sleep showed no differences between controls and Clk^AR mutants in LL (S3A–S3C Fig). Indeed, in LL, total sleep time was similar for control, Clk^AR and cyc⁰ flies (S3B Fig). Sleep fragmentation, as assessed by average night (or presumptive night) bout duration, was stronger in LL than in LD (S3C Fig). However, in LL it was now very similar between all three strains. Thus the accelerated ARLI observed for the Clk^AR mutants in LL, relative to control or cyc⁰ flies, again does not correlate with a poorer sleep quality.

In order to determine where Clk function is required to maintain a wild-type SING behavior in aging flies, we used the Gal4-UAS system to express a Clk RNAi under the control of various Gal4 drivers. Heterozygous UAS-Clk^RNAi/+ and Gal4 driver/+ flies were used as controls. When RNAi expression was driven in all clock cells with tim-Gal4, lifespan was significantly shortened relative to both controls (S4A Fig, S3 Table). ARLI was greatly accelerated in the Clk^RNAi-expressing flies (S4B Fig). The acceleration was weaker than in the mutant, as it was not observed on days 10 and 17. This could only reflect a weaker loss of function, as Clk^RNAi also led to a weaker impact on free-running rhythms compared to Clk^AR (S2 Table). We also observed a smaller but significant effect on ARLI for the tim-Gal4/+ control(S4B Fig).

When RNAi expression was driven selectively in PDF-expressing neurons with pdf-Gal4, there was no lifespan reduction (S4C Fig, S3 Table). However, ARLI was greatly accelerated, this time with the two controls behaving similarly to wild-type flies (Fig 2A). This indicates a specific effect of Clk on ARLI, independently of any acceleration in global aging. Accelerated ARLI was also observed with pdf-Gal4 driving another independent Clk^RNAi (UAS-Clk^RNAi-R3) (S4D Fig). Its weaker effect probably reflected a weaker inhibition of Clk function, as suggested by its also weaker effect on free-running rhythms (S2 Table). cyc knock-down in the LNvs had no effect on ARLI (Fig 2B), although cyc^RNAi disrupted behavioral rhythms more than Clk^RNAi (S2 Table). This is again in line with a specific effect of Clk deficiency on ARLI, unrelated to circadian rhythmicity. The Clk^AR mutation may lead to abnormal Clk protein products [32], but the similarity between the Clk^AR ARLI phenotype and that of the two Clk RNAi indicates that accelerated ARLI is likely due to Clk deficiency rather than a gain-of-function effect.

We used additional driver lines to further pin down the neurons where Clk function is required for wild-type ARLI in aging flies. Two neuronal groups express both PDF and Clk: the small LNvs (s-LNvs) and the large LNvs (l-LNvs). The R6-Gal4 driver allowed us to knock-down Clk in the former, while the C929-Gal4 driver allowed us to do the same in the latter. Knocking down Clk in the s-LNvs did accelerate ARLI (Fig 2C), while knocking down Clk in the l-LNvs had no effect (S4E Fig). Interestingly, we observed that ablating all the PDF neurons through expression of the pro-apoptotic gene hid [see [35]] also had no effect on ARLI (Fig 2D). This indicates that the s-LNvs are not required to preserve normal locomotor aging in Drosophila and that Clk deficiency may induce an alteration in the activity pattern of these cells that would lead to accelerated ARLI.

We then asked conversely whether restoring Clk expression selectively in the PDF-expressing LNvs, using the pdf-Gal4 driver, would rescue ARLI in an otherwise Clk^AR mutant background. This was indeed the case (Fig 3A), even though these rescued flies were behaviorally arrhythmic (S2 Table), as previously reported [32]. This further demonstrates a specific effect of Clk on ARLI that is unrelated to disruption of behavioral rhythms. We wondered whether overexpressing Clk in the LNvs would modulate ARLI in a wild-type background. It did not (Fig 3B), nor did it affect behavioral rhythms (S2 Table).

One factor that has been tentatively linked to ARLI is oxidative stress [19]. We therefore estimated brain ROS levels by dye staining in control, Clk^AR and cyc⁰ flies, at various ages post-eclosion. As previously reported for oxidative damage [8], ROS levels increase with age, especially between 24 and 31 days post-eclosion, and this was true for both control and Clk^AR flies (S5A Fig). This increase may play a part in the onset of ARLI in control flies, as it is concomitant with or slightly precedes it (Fig 1A, 1C and 1D). ROS levels were higher in Clk^AR than in controls already at 10 days, and remained higher at 31 days (Fig 4A and 4B and S5B Fig). By 45 days control and Clk^AR brains displayed similarly elevated ROS levels (S5B Fig). In contrast, no increase in brain ROS levels was found for the cyc⁰ mutant relative to control (Fig 4A and 4B). Increased ROS levels are then another Clk^AR mutant phenotype which, like strongly accelerated ARLI, is not caused by clock disruption alone.

The precociously elevated ROS levels in Clk^AR brains could be hypothesized to cause their precocious ARLI phenotype. However, we found no increase in ROS levels in the brains of pdf>Clk^RNAi flies relative to their two controls (Fig 4C), even at 31 days of age, although the former flies displayed accelerated ARLI well before that age (Fig 2A). Conversely, whereas restoring Clk expression selectively in the LNvs in a Clk^AR background rescued ARLI (Fig 3A), it did not reduce brain ROS levels relative to the two controls (Fig 4D). There is thus no correlation between elevated brain ROS levels and accelerated ARLI in Clk-deficient flies (Clk^AR and pdf>Clk^RNAi), indicating that a global increase in brain oxidative stress is neither necessary nor sufficient to cause their accelerated ARLI phenotypes.

Because dopaminergic circuits are involved in locomotor control in flies (see above in the Introduction), we wondered whether the Clk^AR mutation could affect dopaminergic neurons. The adult Drosophila brain contains eight clusters of dopaminergic neurons [36]. We counted the number of these neurons in most classes (S6 Fig) except the protocerebral anterior median (PAM) cluster, which was difficult to quantify precisely because of its large size. The number of neurons immunoreactive for tyrosine hydroxylase (TH-IR neurons), the rate-limiting enzyme for dopamine synthesis, was not significantly different in the brains of 10-day-old Clk^AR, cyc⁰ and control flies (Fig 5A), at an age when the effect of the Clk^AR mutation on SING is still very small (see Fig 1C).

In contrast, in 31-day-old flies, we observed a significant and selective loss of TH-IR neurons in the PPL1 cluster of Clk^AR mutants relative to control, while cyc⁰ brains were unaffected (Fig 5B and S7 and S9A Figs). A similar loss was also observed in 31-day-old Clk^AR/+ heterozygotes (Fig 5C and S7 Fig), as well as in pdf>Clk^RNAI flies where Clk expression was knocked down in the LNvs, relative to both controls (Fig 5D and S7 Fig). That loss, like the accelerated ARLI phenotype (Fig 3A), was rescued by Clk expression restricted to the PDF neurons (S10A Fig). In contrast, the size of several serotonergic clusters in a broad dorso-lateral protocerebral region [see [37]], where the PPL1 is located, did not differ between aging Clk^AR and control flies (S11 Fig). Accelerated ARLI due to impaired Clk function appears thus selectively associated with a loss of TH-IR neurons in the PPL1 cluster, although we cannot rule out that a few PAM neurons are also affected (see also below).

We observed that the PPL1 neuronal cell bodies are located close to the s-LNv dorsal projections, as they turn medially in the protocerebrum (S9A Fig), suggesting that these dopaminergic neurons could be influenced by the s-LNvs, through a direct or paracrine connection. Note that projections from the s-LNv appeared wild-type in the Clk^AR, Clk^RNAi- and cyc^RNAi-expressing flies, while they were disrupted in cyc⁰ flies (S7–S9 Figs), as previously reported for both cyc⁰ and Clk^Jrk [38, 39]. Indeed one of the reasons that prompted us to analyze the Clk^AR rather than the Clk^Jrk mutant was to avoid such a developmental effect, another one being the known dominant-negative character of Clk^Jrk [40].

To test the possibility that PPL1 neuronal loss may involve the activation of apoptotic pathways, we used the H99 deficiency, which removes a pro-apoptotic gene cluster including hid, rpr and grim. H99/+ heterozygous flies are viable, but lack almost all embryonic programmed cell death [41]. No decrease the number of TH-IR PPL1 neurons was observed in H99/+, Clk^AR flies relative to H99/+ controls, which contrasted with Clk^AR alone (S10B Fig). Although this does not demonstrate that PPL1 neurons actually undergo apoptosis in aging Clk^AR brains, it indicates that the activation of apoptotic pathways is indeed involved in the mutant phenotype.

The main signaling molecule secreted by the s-LNv neurons is the neuropeptide PDF. It was recently reported that expression of a dominant-negative form of the circadian kinase doubletime (dbt^K/R) in the LNvs led to transient daily caspase activation, prominently in the optic lobes, by a circadian-independent mechanism that requires PDF receptor signaling [42]. Transient caspase activation was also observed in head extracts of the Clk^Jrk mutant [42]. Here we found that the Clk^AR mutation was unable to accelerate ARLI in the absence of the PDF receptor (Fig 6A). Remarkably, in a Clk^AR mutant background, the lack of PDF receptor also fully rescued the loss of dopaminergic neurons in the PPL1 cluster (Fig 6B).

To test the possibility that loss of TH-IR neurons in the PPL1 cluster may impair SING in older flies, we first eliminated at least part of these neurons by expressing the pro-apoptotic gene hid under control of the TH-D1-Gal4 driver [43]. This accelerated SING decline (Fig 7A), comparably to the effect of knocking down Clk in the s-LNvs (Fig 2A). Similar results were obtained with the TH-D'-Gal4 driver (Fig 7B), which is more specific for dopaminergic neurons [43]. However, TH-D'-Gal4 targets other dopaminergic neurons, outside the PPL1, in the PPL2 and PPM3 clusters, which may contribute to the observed effect on ARLI. Therefore we used an alternative strategy [27]: inactivating Drosophila TH by targeted RNAi. By driving TH RNAi expression with MZ840-Gal4, which is expressed in only one PPL1 neuron (MB-V1) and in other non-dopaminergic brain cells [44], the two MB-V1 neurons should be the only affected ones. We observed that TH knock-down with MZ840-Gal4 also accelerated ARLI (Fig 7C), again comparably to the effect of Clk knock-down in the s-LNvs (Fig 2A). The impact of reducing Clk function in the s-LNvs is thus phenocopied by reducing the number or function of PPL1 neurons, and particularly of the MB-V1 pair, consistent with other neurons (including dopaminergic ones within the PAM) not playing a major effect in the accelerated ARLI of Clk^AR flies.

We report here that both the Clk^AR mutation and reducing Clk function in the PDF-expressing LNvs led to faster decline of a startle-induced locomotor response in aging flies, and to dopaminergic neuron disruption selectively in the PPL1 cluster. A significant acceleration of ARLI was also observed in heterozygous Clk^AR/+ flies, which are fully rhythmic. Such Clk gene-specific phenotypes were indeed completely clock-independent, as they were not observed when the circadian clock was stopped or perturbed in other ways, such as in LL conditions, or by cyc⁰ and tim⁰ mutations. In contrast, we found that spontaneous locomotor activity, sleep, and lifespan were similarly affected in the Clk^AR and cyc⁰ mutants.

Other clock-independent functions of clock genes or the LNvs have been reported in Drosophila, e.g. in reproduction [45, 46], responses to cocaine [47, 48] and sensitivity to sleep deprivation [49]. In the latter case, Cyc seems to play a specific role, as Clk, per, and tim mutants do not display hypersensitivity to sleep deprivation like the cyc⁰ mutant. In what cells Cyc is required to protect the flies from the effects of sleep deprivation is not known. However, in most studies, Clk and cyc mutants had the same or similar phenotypes [34, 40, 50–53]. Clk and Cyc are co-expressed in all brain clock neurons, and neither one appears to be expressed in any other brain neurons [54]. Ectopic expression of Clk but not Cyc is sufficient to induce a functional clock in other brain neurons; it does require Cyc function, however [55, 56]. It remains to be investigated whether Clk requires another partner in circadian neurons to prevent accelerated ARLI and dopaminergic neuron loss.

We observed that removing the PDF-expressing neurons had no effect on ARLI, in contrast to the striking phenotype induced by Clk inactivation in the same neurons. This suggests that Clk inactivation does not inhibit but rather overactivates PDF-neuron signaling. Both the locomotor and neuronal loss phenotypes of the Clk^AR mutant were indeed rescued in the absence of PDF receptor. Recently, the Clk^Jrk mutation, but not per⁰, was shown to transiently activate the Dronc caspase in fly heads, in response to light during the day. Furthermore, the expression of a dominant negative form of the circadian kinase doubletime (dbt^K/R) in the LNvs was shown to activate the Dronc caspase in large parts of the brain, independently of the clock itself but dependent on PDF receptor signaling [42]. It is therefore quite possible that Clk inactivation in the s-LNvs can induce PDF receptor-dependent caspase activation in target or neighbouring cells, that will eventually lead, directly or indirectly, to dopaminergic neuron loss (see below).

The precocious increase in ROS levels in Clk^AR mutant brains could also contribute to caspase activation. Again, the clock and Cyc do not seem to be involved as ROS levels were not affected in the cyc⁰ mutant. However, our targeted Clk rescue and knock-down experiments suggest that a global increase in brain ROS levels accelerates ARLI only marginally, if at all, in this context. The global ROS increase may contribute to some differential effects of the Clk^AR versus cyc⁰ mutations (e.g. on sensitivity to sleep deprivation, since elevated ROS levels may induce protective mechanisms, see e.g. [57]), or of the Clk^AR mutation versus LNv-restricted expression of Clk^RNAi (such as the one we found on longevity).

Our results suggest the existence of a link between the s-LNvs and the PPL1 cluster of dopaminergic neurons that involves PDF-receptor signaling (Fig 8). Inverse links between clock and DA neurons have already been demonstrated: dopamine strongly modulates cAMP levels in l-LNvs, via different receptor subtypes, while s-LNvs are much less responsive [58]. The identity of the DA neurons afferent to the l-LNvs is still unknown.

We observed that the PPL1 cluster lays close to the s-LNv projections to the dorsal protocerebrum. s-LNv projections were impaired in the cyc⁰ mutant, as previously shown [38, 39], but not in the Clk^AR mutant. This difference does not seem to play a part in the differential impact of the two mutants on ARLI, since s-LNv projections appeared completely wild-type when expressing RNAi for either cyc or Clk in the PDF neurons. Although cyc^RNAi was more effective than Clk^RNAi in disrupting circadian activity rhythms, it did not accelerate ARLI at all. As no PDF processes were observed in apposition to TH-IR neurons in the posterior dorsal brain [59], signaling to the PPL1 may well be paracrine rather than synaptic, consistent with both anatomical [60] and functional [61] evidence for such signaling mechanism by PDF. On the other hand, PDF receptor in that protocerebral region appears exclusively expressed in 3 LNd clock neurons, as judged from anti-MYC labeling of a PDF receptor-MYC fusion [62]. The LNds might thus be intermediary neurons mediating the influence of the s-LNvs on the PPL1.

What happens to the PPL1 neurons when Clk function is compromised in the LNvs? Similar questions were raised in fly models of PD, where the size of various TH-IR neuronal clusters often appeared reduced (see e.g. [63–68]), but actual cell loss was sometimes contested [69, 70]. Here we report that reducing TH levels in specific neurons was sufficient to accelerate ARLI, as was also shown in our previous study [27]. In the latter, accelerated ARLI was attributed not to cell loss, but to age-related alterations in contacts onto the mushroom bodies from a subset of ~15 PAM neurons, when they express α-synuclein (a protein involved in PD). In agreement with that, we found that locomotion was already affected in 10-day-old Clk^AR mutants, before any visible reduction in dopaminergic cell numbers. The MB-V1 neuron, which is currently our best candidate among PPL1 neurons as mediator of Clk deficiency-induced ARLI acceleration, also projects to the mushroom bodies [71]. It remains to be seen whether this specific neuron is among those that disappear or lose TH-IR in aging flies when Clk is down-regulated in the LNvs, and whether the loss of other PPL1 neurons would be sufficient to accelerate ARLI.

The PPL1 cluster was reported to be selectively reduced in flies expressing α-synuclein under control of TH-Gal4 [66], a driver that does not express strongly in the PAM [23]. That reduction was rescued in an apoptosis-deficient genetic background [66], similarly to what we observed for the effect of the Clk^AR mutation, or by overexpression in dopaminergic neurons of either glutathione S-transferase Gst1 [66], the Nrf2 transcription factor [67] or the endosomal recycling factor Rab11 [72]. Whether these latter effectors can also protect the PPL1 neurons in Clk-deficient flies, as well as how apoptotic pathways may affect PPL1 neurons in these contexts, remain open questions.

Our results show that Clk inactivation in the main Drosophila pacemaker neurons, the s-LNvs, has dramatic consequences for a locomotor response and select dopaminergic neurons maintenance in aging flies, independently of the less prominent but significant defects induced by circadian rhythm disruptions. Such progressive, circadian clock-independent effects both add to, and contrast with, previously described links between Clk and dopaminergic signaling. The acute nocturnal hyperactivity and reduced sleep of Clk^Jrk mutants, for instance, were attributed to increased dopaminergic transmission [73]. However, they involved Cyc as well, and the l-LNvs rather than the s-LNvs. In mice, a Clk mutant also displays decreased sleep and hyperactivity, as well as a mania-like behavior, presumably owing to increased activity of dopaminergic neurons in the ventral tegmental area [74, 75]. However, contrary to flies, Clk is expressed in such neurons, where its loss increases expression of genes involved in dopaminergic signaling, including TH [74, 75], consistent with the observed phenotypes.

Interestingly, the combined deletion of Clk and its paralog Npas2 induced severe age-dependent astrogliosis in the mouse brain, leading to degeneration of synaptic terminals, neuronal oxidative damage and impaired expression of several redox defense genes [76]. Neuron and glia-specific inactivation of BMAL1, the mouse ortholog of Cyc, produced similar phenotypes. In Drosophila, our results now indicate that Clk regulates s-LNv activity, independently of its role in the circadian machinery, and that the loss of this regulation leads to progressive dysfunction of specific brain dopaminergic neurons, by mechanisms that involve PDF receptor signaling. Further deciphering the wide-ranging effects of Drosophila Clk dysfunction, whether circadian clock-dependent or not, or cell-autonomous or not, could shed new light on the regulation of dopaminergic neuron survival in both flies and mice.

Flies were maintained on standard medium at 25°C, under a 12:12 Light/Dark cycle (LD 12:12), or under constant light. Light intensity at the level of the flies was in the range 500–3000 lux. The following clock mutants were backcrossed for five generations to a wild-type (Canton-S) genetic background before use: Clk^AR [32], cyc⁰ [77], pdf⁰ [35], tim⁰ [78]. cyc⁰, pdf⁰ and tim⁰ are null mutations, whereas Clk^AR is a strong hypomorph that is behaviorally arrhythmic but with detectable Per oscillations in peripheral tissues [32]. Other strains included the PDF receptor mutant han⁵³⁰⁴ [79], the H99 deficiency (Bloomington Drosophila Stock Center (BDSC) strain #1576) and the following Gal4 drivers: C929-Gal4 [80], MZ840-Gal4 [81], pdf-Gal4 [35], R6-Gal4 [82], TH-D’-Gal4 and TH-D1-Gal4 [43], tim-Gal4 [83] and UAS strains: UAS-Clk-B19 (C. Michard-Vanhée, B. Richier and F. Rouyer, personal communication), UAS-Clk^RNAi (TriP HMJ02224, BDSC #42566) [84], UAS-Clk^RNAi-R3 (7391R-3 strain, National Institute of Genetics, Japan), UAS-cyc^RNAi (TriP HMJ02219, BDSC #42563), UAS-hid [85], UAS-TH^RNAi (TriP JF01813, BDSC #25796), UAS-Wake^miR1 [34]. H99, Clk^AR recombinant chromosomes were obtained via standard Drosophila genetics. As they are homozygous lethal, like the H99 parental chromosome, they were identified by assaying the rhythmicity of progeny from crosses between H99 Clk^AR / TM3(Sb) and Clk^AR flies.

Locomotor activity experiments were performed using commercial activity monitors (TriKinetics) placed in incubators equipped with standard white, fluorescent low-energy tubes. Young (10 day-old male flies) or older (30 day-old males) males were maintained 5–6 days under LD 12:12 (Light-Dark 12h:12h), and then switched to at least 5–6 days of constant darkness, all on 5% sucrose-agar medium at 25°C. Data analysis was performed with the FaasX software, as described previously [86]. For constant darkness experiments, analysis started the second day of constant darkness. Histograms represent the distribution of the activity through 24 h in 30 min bins, averaged for n flies over 4–5 cycles. All behavioral experiments were repeated 2–3 times to verify reproductibility.

3–4 days old male flies were placed individually in Trikinetics glass tubes, video monitored under infra-red illumination during 3 days in LD, then placed for at least 5 days in LL. The images were processed using pySolo video software [87] to determine the distance travelled (in pixels) for each minute of the day. Sleep was defined as 5 min of more of immobility [88]. At least 40 flies in 2–3 replicates were analyzed for each genotype.

Male flies were maintained on standard medium at 25°C and under either LD 12:12, or constant light from adult day 1 until death. 50 animals/bottle in triplicate were tested for each genotype in a given experiment, and each experiment was performed at least twice.

Male flies were aged under the same conditions as for the lifespan assays. SING assays were performed as described previously [27, 64]. Groups of 10 flies were placed in a vertical column (25 cm long, 1.5 cm diameter) with a conic bottom end and left for about 20–25 min for habituation. Then, for each genotype, 5 columns were tested individually by gently tapping down the flies (startle), which normally respond by climbing up. Each fly group was assayed three times at 15 min intervals.

Results are the mean and SEM of the scores obtained with the 5 independent groups of flies per genotype. The performance index (PI) for each column was calculated as follows: ½[1 + (n_top-n_bot)/n_tot], where n_tot is the total number of flies in the column, n_top is the number of flies that have reached at least once the top of the column (above 22 cm) during a 1 min interval, and n_bot is the number of flies that never left the bottom (below 4 cm). ARLI was monitored as described previously by testing SING performance weekly over 6 weeks, starting on day 10 after eclosion [27]. Dead flies were replaced by substitutes of the same age. Experiments were repeated 2 to 3 times at different periods of the year.

Immunostaining was performed essentially as previously described [26]. Adult flies of the desired ages were briefly washed in 70% ethanol before brain dissection in ice-cold Drosophila Ringer Ca²⁺ free solution. Brains were then fixed for 1h with shaking at room temperature in fixative containing 4% paraformaldehyde in PBS. Brains were then washed 3 x 20 min with PBS and incubated overnight in PBS/0.5% Triton X-100 + 2% BSA, at 4°C. Staining with primary antibody was carried out overnight at 4°C in PBS/0.5% Triton X-100 + 2% BSA. Brain were then washed 3 x 20 min with PBS/0.5% Triton X-100, and incubated for 2h at room temperature with secondary antibody diluted in PBS/0.5% Triton X-100 + 2% BSA. Finally, 3 x 20-min washes were performed using PBS.

Primary antibodies used included: mouse monoclonal anti-TH (Immunostar, 1:1000), rabbit polyclonal anti-PDF (gift from F. Rouyer's lab, 1:1000), rabbit polyclonal anti-5-HT (Sigma, 1:1000). Secondary antibodies included: anti-mouse or anti-rabbit conjugated to Alexa Fluor 488 or 555 (Invitrogen Molecular Probes, 1:1000).

In situ ROS detection was performed using a dihydroethidium (DHE) dye (Life technologies) following a previously described protocol [89] we adapted to whole-mount Drosophila brains. Briefly, male flies were aged under the same conditions as for lifespan assays and dissected in Schneider’s insect medium. The brains were then incubated with DHE for 5 minutes in the dark, fixed for 5 minutes in 7% formaldehyde in 1 X PBS, and immediately imaged on a confocal microscope, as indicated below. Relative ROS levels were measured by quantification of the dye fluorescence using the Fiji software [90].

Fly brains were mounted on slides using as antifade reagent either Prolong Gold (Life Technologies), for brain immunostaining, or Vectashield (Vector Laboratories), for ROS measurements. Brains were visualized and images acquired with a Nikon A1R confocal microscope. A minimum of 10–15 brains was scored over at least 2 trials. Laser, filter and gain settings remained constant within each experiment, and channels were scanned sequentially. Confocal Z-stacks were analyzed using ImageJ software: dopaminergic cells were counted for each cluster, and whole brain average intensity levels were measured for ROS detection.

For lifespan assays, survival curves were generated and compared using the log-rank (Mantel-Cox) test, and 150 animals were tested per genotype, with each experiment performed 2–3 times. For SING assays and fluorescence quantification, the mean and SEM were calculated for each trial and two-way Anova with post-hoc Tukey comparisons was used. Staining data were analyzed by one-way Anova and Tukey’s pairwise comparisons. Statistical analysis of sleep was performed using the Kruskal-Wallis test. GraphPad Prism 6 was used for all statistical analyses. Significant values in all figures: *: p<0.05, **: p<0.01, ***: p<0.001.