The central molecular clock is robust in the face of behavioural arrhythmia in a Drosophila model of Alzheimer’s disease

Alzheimer’s disease (AD) is the most common cause of dementia in adults and is characterised at the microscopic level by extracellular amyloid plaques and intraneuronal tau tangles. Amyloid plaques are composed of fibrillar aggregates of a spectrum of amyloid beta (Aβ) peptides derived from the proteolytic cleavage of amyloid precursor protein (APP) (LaFerla et al., 2007). The significance of Aβ is underpinned by the numerous disease-linked mutations that dysregulate APP processing: mutations that result in a spectrum of Aβ peptides with a higher aggregation propensity have been linked to familial AD (Philipson et al., 2010), whereas sequence variation in APP that reduces Aβ production is protective (Jonsson et al., 2012). There is much evidence from cell-culture and animal-model systems (Iijima-Ando and Iijima, 2010; Philipson et al., 2010) that the conformers of Aβ that possess neurotoxic activity are likely to be soluble oligomeric species rather than the more easily detected amyloid plaques (Lesné et al., 2006; Shankar et al., 2008; Ono et al., 2009; Tomic et al., 2009; Brorsson et al., 2010; Jo et al., 2011; Tang et al., 2012; Speretta et al., 2012).

Alongside the well-recognised memory and cognitive deficits that typify AD, a substantial proportion of individuals with AD also experience circadian abnormalities, including increased daytime napping, night-time restlessness and fragmented sleep. Taken together, these clinical features constitute a dampening of the variation in day-night activity (Volicer et al., 2001; Coogan et al., 2013); furthermore, two-thirds of individuals with AD that are living at home exhibit some degree of ‘sundowning’, in which restlessness and agitation increase late in the afternoon and early evening (Prinz et al., 1982; Volicer et al., 2001). It is readily apparent that such behavioural problems are a substantial burden for both AD individuals and their caregivers.

Circadian timekeeping in animals is a cell-autonomous mechanism based on the intrinsic 24-hour-period oscillation of ‘clock gene’ products (such as PER1, PER2, CRY1, CRY2, CLOCK and BMAL1 in humans) mediated by interlocked transcriptional-translational feedback and feedforward loops (TTFLs). Such cellular circadian oscillators are present throughout the body, but those in the suprachiasmatic nucleus (SCN; ~20,000 ‘clock neurons’) of the hypothalamus are considered to be the master pacemaker in humans (Mohawk et al., 2012). The SCN neurons are divided into a dorsal shell [arginine vasopressin (AVP)-positive] and ventral core [vasoactive intestinal polypeptide (VIP)-positive areas]. Circadian oscillators in the SCN are entrained by light to keep them in synchrony with the external light-dark cycle. The SCN then converts the entrained circadian signal into coordinated physiological and behavioural outputs via multiple humoural and neuronal pathways (Mohawk et al., 2012). Importantly, circadian oscillations are self-sustaining at both molecular and behavioural levels. Therefore ‘free-running’ rhythms continue even in the absence of external cues [e.g. the constant darkness (DD)].

The behavioural abnormalities linked to AD in the clinic have been substantiated by histological changes in the SCN in postmortem brains, in particular the cell loss observed by Swaab and colleagues (Swaab et al., 1985; Swaab et al., 1988). Despite the cell loss seen in the SCN in AD brains, amyloid plaques here are sparse (Coogan et al., 2013), possibly indicating that Aβ toxicity is largely non-cell-autonomous, being derived from neighbouring cells. Concordant with this, Tate and colleagues reported reduced amplitude of behavioural rhythms in rats carrying SCN grafts of PC12 cells expressing a disease-linked variant of APP as compared with animals grafted with control PC12 cells (Tate et al., 1992). However, subsequent murine studies of AD-linked circadian locomotor abnormalities, using established model systems, has yielded a complex and sometimes contradictory picture. In particular, mice expressing mutant APP in light-dark (LD) conditions exhibit normal circadian locomotor activity (Wisor et al., 2005; Ambrée et al., 2006; Gorman and Yellon, 2010). By contrast, increased locomotor activity during resting light hours was detected in transgenic animals expressing additional mutated human γ-secretase (APP×PS1) (Duncan et al., 2012) or the combination of mutant PS1 and tau (APP×PS1×tau) (Sterniczuk et al., 2010). Furthermore, only minor deficits in free-running behaviour (DD) are detected in these AD model systems (Wisor et al., 2005; Gorman and Yellon, 2010; Sterniczuk et al., 2010). For these reasons, the role of toxic Aβ species in circadian deficits in AD remains elusive.

As a complement to murine models of AD, we have generated a Drosophila system to study Aβ toxicity. Instead of replicating the proteolytic processing of APP, we and others have fused the Aβ peptide with a secretion signal peptide and driven its expression in the nervous system (Finelli et al., 2004; Iijima et al., 2004; Crowther et al., 2005). Various Aβ species were expressed pan-neuronally in Drosophila using the Gal4-UAS expression system (Brand and Perrimon, 1993), and Aβ toxicity was detected using a range of biochemical, neuron-histological and behavioural assays (e.g. Jahn et al., 2011; Speretta et al., 2012; Huang et al., 2013). In this study we have combined the tools available to neurodegeneration modelling in the fly with the well-developed systems that are also available for studying circadian rhythms. The use of the fly as a model organism is justified by the many orthologies between Drosophila and human, in particular by the conserved circadian TTFLs, involving the clock genes period, timeless, clock and cycle (Allada and Chung, 2010). Circadian locomotor activity in Drosophila is controlled by ~150 clock-gene-expressing neurons (clock neurons) in the brain. As with the SCN in humans, Drosophila clock neurons can be divided into several groups (termed sLNvs, lLNvs, LNds, DN1s, DN2s, DN3s and LPNs in the fly) according to their ventral-dorsal anatomy and neuropeptide identity. Similar to the role of the neuropeptide VIP in synchronising among clock neurons in the SCN (Hastings and Herzog, 2004; Aton et al., 2005; Maywood et al., 2006), the neuropeptide PDF (pigment disperse factor), released from about 16 ventral neurons (sLNvs and lLNvs) in Drosophila, maintains robust circadian behaviour by paracrinely synchronising the molecular oscillation of clock neurons (e.g. Renn et al., 1999; Peng et al., 2003; Cusumano et al., 2009). In addition, the majority of the axons from these clock neurons project to the dorsal protocerebrum (dorsal commissure) (Helfrich-Förster et al., 2007), where they communicate with each other and to their downstream targets. Normal free-running circadian behaviour in Drosophila also requires correct signalling at these synapses (Kaneko et al., 2000; Blanchardon et al., 2001; Nitabach et al., 2002). Rezával et al. (Rezával et al., 2008) previously demonstrated that overexpression of wild-type human APP in PDF-positive ventral clock neurons (pdf >hAPP) resulted in age-dependent loss of circadian rhythm. Although Drosophila does have the γ-secretase required to process APP, it has little β-secretase-like (dBACE) activity and so the generation of Aβ peptides is inefficient (Fossgreen et al., 1998; Carmine-Simmen et al., 2009). Therefore, the circadian abnormality in pdf >hAPP flies (Rezával et al., 2008) is probably unrelated to toxic Aβ peptides. In this study, however, we have employed well-established tools for characterising the Drosophila clock system to investigate the mechanism of Aβ-mediated disruption of circadian rhythms.

To determine whether our Aβ-expressing flies (Crowther et al., 2005; Jahn et al., 2011) exhibit disturbed intrinsic circadian rhythms, we monitored their circadian locomotor activities in constant darkness (DD). By calculating autocorrelation coefficients we quantified the robustness of their circadian periodicity [arrhythmia is defined as rhythmic statistic (RS) ≤1.5] (Levine et al., 2002). The Gal4-UAS system was used to drive expression from a single transgene of each of Aβ₄₀, Aβ₄₂ and the arctic (E22G) variant of the Aβ₄₂ peptide in the Drosophila nervous system (elav>Aβ₄₀, elav>Aβ₄₂ and elav>Aβ₄₂arc; Fig. 1). While still young [2–12 days after eclosion (dae) and 12–22 dae], the Aβ₄₀ and Aβ₄₂ flies exhibited robust circadian rhythmicity in DD that was essentially identical to that observed in control flies (elav>51D). Although a subpopulation of Aβ₄₀- and Aβ₄₂-expressing flies developed arrhythmic behaviour by the age of 22–32 dae, their average RS did not differ significantly from controls (Table 1). By contrast, the overall rhythmicity of flies expressing pan-neuronal Aβ₄₂arc was significantly reduced compared with controls at all age groups (RS; Table 1). Furthermore, there was an age-related progression in dysrhythmia with a significant decline in the RS between the 2–12 dae and 22–32 dae groups (Table 1). At 22–32 dae, about 80% of the elav>Aβ₄₂arc flies were arrhythmic, whereas the majority of the age-matched controls flies remained rhythmic (Table 1). Consistent with the age-dependent decline in circadian behaviour, the arrhythmic pattern in the averaged actogram of elav>Aβ₄₂arc flies was clear by 22–32 dae (Fig. 1A). Although all flies tested exhibited age-related decline in circadian rhythmicity (Fig. 1B), those expressing Aβ₄₂arc were significantly worse than other genotypes. The appearance of discreet Aβ deposits, akin to plaques, in Aβ₄₂arc fly brains (supplementary material Fig. S1A–E) correlates with the severity of behavioural disruption. However, the amount of such large Aβ aggregates in the brain does not predict behavioural arrhythmia because, for any given Aβ genotype, there is no difference in plaque density between behaviourally rhythmic and arrhythmic flies (supplementary material Fig. S1F and see Discussion).

To exclude the possibility that the age-related decline in walking velocity in Aβ-expressing flies (Jahn et al., 2011) was confounding our observations of circadian rhythmicity, we assessed the overall daily locomotor activity. When we counted the number of beam crosses in our DAM apparatus, there were no significant differences between any genotypes at least until age 22–32 dae (Fig. 1C). This indicates that the loss of circadian activity is not a function of overall decreased locomotor activity.

Because of the characteristic dampening of circadian rhythm in individuals with AD, light treatment has been used in an attempt to enhance circadian rhythmicity (Coogan et al., 2013). To investigate the reversibility of the circadian arrhythmia in Aβ-expressing flies, we identified young flies that displayed arrhythmic behaviour when transferred from LD to DD. We then returned them to LD, with a 6-hour shift in the cycle, and looked to see whether rhythmic behaviour was restored. Flies with an intact circadian clock respond not only to the actual light changes in LD with a startle reflex but also anticipate dawn and dusk with 2–3 hours of increased locomotor activity (Stoleru et al., 2005; Lim et al., 2007). We used this circadian controlled ‘anticipatory ramping’ in activity as the marker of rhythmicity in these experiments. In control flies (elav>51D) exposed to LD, anticipatory ramping in behaviour was seen, as expected, before dawn (white circle, Fig. 2A) and dusk (black circle, Fig. 2A). In continuous darkness, control flies retained robust circadian behaviour, with one peak of activity during the 24-hour period (Fig. 2Aiii). In the following secondary LD cycles, control flies were able to re-synchronise with the new phase within 1 day (Fig. 2Ai,iv). Under the same conditions, elav>Aβ₄₂arc flies retained substantial rhythmic behaviour during the first LD cycles (Fig. 2Bi,ii) but became arrhythmic in the following constant darkness (cf. Fig. 1A; Fig. 2Bi,iii). Furthermore, elav>Aβ₄₂arc flies largely failed to re-synchronise and exhibited weak ramping activity in the secondary LD (Fig. 2Bi,iv). The Harrisingh anticipatory index was used to objectively quantify the degree of entrainment in elav>Aβ₄₂arc flies during LD (Harrisingh et al., 2007). We found that elav>Aβ₄₂arc flies have a much reduced anticipatory index as compared with the controls, and they were comparable to the non-anticipatory period-null mutants (per⁰¹) during the secondary LD cycles (Fig. 2Aiv,Biv,Civ,D). Taken together, we found that none of the DD-arrhythmic elav>Aβ₄₂arc flies were able to regain vigorous rhythmicity during the subsequent LD cycles, indicating that behavioural deficits are not readily remediated by re-exposure to a rhythmic 24-hour LD cycle.

Having established that DD-arrhythmic flies could not be substantially re-entrained by subsequent exposure to an LD environment, we were interested to know whether, despite this, a rhythmic environment could still provide benefits for an arrhythmic organism. To investigate this possibility, we took control and Aβ₄₂arc-expressing flies and compared their longevity in LD and continuous light (LL) conditions. Consistent with a previous study (Pittendrigh and Minis, 1972), there was a significant reduction in the lifespan for control flies in LL (48 dae elav>51D, Fig. 2E) as compared with LD (57 dae, elav>51D, P<0.001, log-rank test, n=100, as 10 sets of 10 flies, for each condition, 18% increase over LL, Fig. 2E). Remarkably, Aβ₄₂arc-expressing flies benefited identically from LD despite their arrhythmic behaviour (elav>Aβ₄₂arc, LL: 22 dae vs LD: 27 dae, P<0.001, log-rank test, n=100 for each conditions, 18% increase over LL and P<0.05; Student’s t-test, 10 sets of 10 flies, Fig. 2E). Because LL conditions disrupt both behavioural and molecular circadian oscillations in flies (Konopka et al., 1989; Marrus et al., 1996), the observed increase in lifespan on going from LL to LD might be the result of residual clock functions in elav>Aβ₄₂arc flies despite their behavioural arrhythmicity.

We used two approaches to test, at both the cellular and molecular levels, whether the central clock apparatus remains intact during the progression of Aβ toxicity even though the flies exhibit arrhythmic behaviour. The first approach involved the direct visualisation of the structural integrity of a subgroup of clock neurons (PDF-positive cells) that are essential for maintaining intrinsic rhythmicity in DD (scheme of clock neurons, Fig. 3A) (Renn et al., 1999). When we compared control flies and Aβ₄₂arc-expressing flies by counting the number of PDF-positive cell bodies, we found no differences at least until 30 dae (Fig. 3B). We also visualised two dorsal neuronal groups – LNds and DN1s – by staining for Period (Per) protein. Again, we found no evidence of clock neuron loss in elav>Aβ₄₂arc flies as compared with controls (Fig. 3C,D). Additionally, there was no correlation between the Aβ-plaque density and the number of clock neurons in elav>Aβ₄₂arc fly brains (supplementary material Fig. S1G). Despite the absence of gross structural changes in the clock neurons, we employed a second approach to look for functional deterioration. By measuring the bioluminescence derived from a Per-luciferase fusion construct, 8.0-luc:9 (Veleri et al., 2003), we were able to monitor the molecular clock as it pertains to Per protein oscillation. Previous work has indicated that the 8.0-luc:9 strain faithfully reports molecular clock oscillations in non-PDF dorsal clock neurons (DNs and LNds) in the fly brain (Hodge and Stanewsky, 2008; Sekine et al., 2008; Sehadova et al., 2009). Comparing 8.0-luc:9/elav>Aβ₄₂arc flies with equivalent control flies that do not express Aβ (Fig. 3E), we found that the rhythmic robustness of the bioluminescence was the same for both populations. To confirm that the bioluminescence from the Per-luciferase fusion construct was correctly reporting the oscillation of Per protein, we immunostained the brains of control and Aβ₄₂arc-expressing flies for Per at various times during the second day of constant darkness post-entrainment. By specifically assessing the intensity of Per staining in the clock neurons (DN1s, LNds and sLNvs) we could confirm that the behaviourally arrhythmic Aβ₄₂arc-expressing flies exhibited the same diurnal Per oscillation as we see in control flies (green staining, supplementary material Fig. S2). By contrast, there was no circadian variation in the density of the Aβ peptide deposits (magenta, supplementary material Fig. S3A).

Taken together, these data indicate that the circadian behavioural abnormalities seen in Aβ-expressing flies are not caused by loss of central molecular clock function but are more likely due to damage in the downstream pathways in the clock neurons or other distal brain region.

Although these results underline the importance of Aβ-mediated degradation of the clock output pathways, the studies showing cell loss in the SCN in AD postmortem samples indicate that damage to the central clock neurons is a possible cause for circadian rhythm deficits (Wu and Swaab, 2007). For this reason we were interested to know how the flies would respond to a range of Aβ insults that were restricted to the clock system. To this end we used the driver lines timeless-gal4, which drives expression in all clock cells including neurons and glia (Fig. 4A), and pdf-gal4, which drives expression in PDF-positive clock neurons (Kaneko and Hall, 2000). As with elav-gal4, the clock-neuron-specific expression of the less-aggregation-prone Aβ isoforms Aβ₄₀ and Aβ₄₂ had few or no behavioural consequences. Surprisingly, both pdf>Aβ₄₂arc and tim>Aβ₄₂arc expression exhibited robust behaviour; this is in contrast to the arrhythmicity induced when Aβ₄₂arc was expressed ubiquitously (cf. Fig. 4Bi,Ci and Fig. 1A). To test the possibility that the absence of any phenotype is due to lower expression levels for pdf and tim lines, as compared to elav, we measured gal4-dependent expression of GFP as a control. When we quantified GFP fluorescence specifically in PDF-positive neurons we found that the expression levels were essentially identical for all three (elav, pdf and tim) drivers (arrows in supplementary material Fig. S4A,B and quantification in S4C). The only remarkable difference was that, for elav-driven expression, the levels of GFP in the bulk of the brain (that is the non-clock neurons) were higher, as expected (supplementary material Fig. S4D). Taken together, these data indicate that Aβ expression that is restricted to clock cells is insufficient to trigger circadian abnormalities.

Although Aβ₄₂arc expression in clock cells was compatible with normal circadian behaviour, we tested whether the highly toxic tandem Aβ₄₂ construct (TAβ₄₂) could induce arrhythmia under similar conditions. TAβ₄₂ consists of two repeats of the Aβ₄₂ sequence linked by a glycine-rich 12mer linker peptide. We have previously shown that TAβ₄₂ has an oligomer-rich aggregation mechanism (Speretta et al., 2012) and indeed elav-driven expression at 25°C results in developmental lethality (data not shown). By contrast, the TAβ₄₀ variant, although equally aggregation prone, does not form oligomers and is essentially non-toxic (Speretta et al., 2012). When TAβ₄₂ was expressed in all clock cells using the tim driver, we observed significant arrhythmic behaviour as compared with tim>Aβ₄₂arc and non-Aβ controls (ages between 10 dae and 32 dae; Table 1 and Fig. 4Bii). Further restriction in the scope of TAβ₄₂ expression was achieved by using repo-gal80 to suppress expression in the glial subset of tim-positive cells (tim,repo-gal80, supplementary material Fig. S5; Fig. 4A). Although more rhythmic than the tim>TAβ₄₂ flies at the same age, tim,repo-gal80>TAβ₄₂ flies showed a reduced rhythmic percentage and reduced rhythmicity as compared to controls (supplementary material Fig. S5B) without any potentially confounding decrease in average locomotor activity (supplementary material Fig. S5C). Remarkably, even TAβ₄₂, when driven by pdf-gal4, was insufficient to induce arrhythmic behaviour (Fig. 4Cii; Table 1), indicating that the roles of Aβ expression outside PDF neurons are likely to be of primary importance. Moreover, cell-autonomous effects of Aβ are likely to be a relatively minor contributor to circadian abnormalities (Fig. 4D).

The preceding data indicate that pan-clock expression, driven by tim-gal4, is the most clock-restricted domain that generates robust TAβ₄₂-mediated circadian arrhythmia. To assess the functional and structural integrity of the clock system under these conditions we again employed a number of molecular and cellular techniques. In tim>8.0-luc:9/TAβ₄₂ flies, aged between 19 dae and 29 dae, we found that the central molecular clock retained its rhythmicity (Fig. 5A) despite the behavioural arrhythmia and significant clock-cell loss induced by expression of TAβ₄₂ (Fig. 5B; supplementary material Fig. S6). A similar pattern of behavioural arrhythmia and cell loss was observed when TAβ₄₂ was restricted to clock neurons (tim,repo-gal80>TAβ₄₂, Fig. 5C; supplementary material Fig. S5).

We also studied Aβ₄₂-linked synaptic dysfunction in clock neurons in two ways using confocal microscopy; firstly, we immunostained the brains for the PDF peptide, as a marker of PDF-positive clock neurons, and, secondly, we assessed the intensity of a chimaeric GFP-synaptotagmin construct that accumulates presynaptically in all clock neurons. In both cases we quantified the signal intensity at the dorsal termini, a major site of clock neuron axonal projection. We found that the presence of TAβ₄₂ markedly reduced PDF peptide and GFP signals in these areas, indicating a paracrine abnormality in PDF neurons and presynaptic dysfunction in clock neurons (Fig. 5D). Given the relatively robust character of the molecular clock signal in the face of extensive neuronal dysfunction, and appreciable neuronal death, it seems likely that the central molecular clock continues to ‘tick’ until the cells die and are physically lost. In other words, the molecular clock seems to be the most robust feature of the clock system during Aβ pathology, being interrupted only by neuronal dysfunction and death.

So far we have demonstrated that even TAβ₄₂ does not induce arrhythmia when expressed exclusively in PDF neurons (Fig. 6Ai); by contrast, use of a tim-gal4 driver that includes glia, pdf neurons and other nearby clock neurons does cause arrhythmia (Fig. 6Aii). Therefore, we were interested to determine whether the behavioural disturbance consequent on the expression of Aβ is mediated by cell-autonomous mechanisms, or not. To achieve the expression of TAβ₄₂ throughout the clock system but specifically not in the PDF-positive neurons that control the DD rhythm (Fig. 6Aiii), we used the tim,pdf-gal80 driver. In this experiment we find that, although the number of PDF neurons is not affected by the expression of TAβ₄₂, we do see a reduction in PDF peptide staining in the dorsal terminus of PDF-positive neurons (Fig. 6B). The hypothesis that PDF neurons are dysfunctional, despite not expressing Aβ themselves, was justified by the finding that the flies also exhibited behavioural arrhythmia in DD (Table 1).

The dampening, and eventual disintegration, of circadian behaviour in individuals with AD constitutes one of the most distressing clinical features of the disorder. Although there is evidence that cell loss in the SCN is correlated with such symptoms of AD, it is not known whether damage to the molecular clock, the clock neurons, or rather to the output pathway, underpins the behavioural deficits. To address these questions we have expressed Aβ peptides pan-neuronally in the fly brain and assessed the consequences for circadian behaviour. During the lifespan of our flies, we found that pan-neuronal Aβ₄₀ and Aβ₄₂ expression had no effect on circadian locomotor behaviour. By contrast, expressing the Arctic variant of Aβ₄₂ (Aβ₄₂arc) resulted in profound age-dependent behavioural arrhythmia as evidenced by a progressively increasing arrhythmic sub-population, reduced overall rhythm robustness in DD (Fig. 1) and loss of anticipatory activity in LD (Fig. 2), recapitulating the dampening of behavioural rhythms in individuals with AD (Volicer et al., 2001) and demonstrating a loss of circadian regulation that could be considered as the fly equivalent of ‘sundowning’ (Fig. 2D). Flies were not aged past 35 days because thereafter control flies exhibit circadian abnormalities (Rezával et al., 2008; Luo et al., 2012; Rakshit et al., 2012; Umezaki et al., 2012) and the differences between Aβ-expressing and control flies become less clear.

The more marked consequences of Aβ₄₂arc expression are thought to be due to the increased aggregation propensity of this peptide and to its oligomer-rich aggregation mechanism (Nilsberth et al., 2001; Whalen et al., 2005; Luheshi et al., 2007). Our previous studies have shown that the fly reports the presence of such oligomeric aggregates and these species correlate more closely with climbing and longevity deficits than do large insoluble aggregates (Luheshi et al., 2007; Speretta et al., 2012). Likewise, the circadian behavioural abnormalities demonstrated here also correlate more closely with oligomer formation (Aβ₄₀≤TAβ₄₀≤Aβ₄₂<Aβ₄₂arc<TAβ₄₂, see Figs 1 and 4) than they do with total Aβ plaque density in the fly brains (supplementary material Fig. S1). Therefore, we reason that oligomeric species instead of large aggregates are the likely cause of the circadian deficits. These findings are concordant with similar experiments in the mouse, where animals expressing a disease-linked APP mutation, with or without mutant PS1 and tau, were found to retain essentially normal DD circadian behaviour throughout life (Wisor et al., 2005; Sterniczuk et al., 2010). Because these mice represent a model of wild-type Aβ₄₂ overproduction, this might explain why, like in the elav>Aβ₄₂ flies, little or no circadian abnormality was observed (Table 1). It is possible that increasing the levels of oligomeric aggregates, by introducing additional copies of the wild-type Aβ₄₂ transgene, or by ageing the flies for longer, might also elicit circadian abnormalities. However, as it stands, our fly model of Aβ₄₂arc toxicity is the first experimental animal that robustly recapitulates the progressive circadian deficits found in AD.

One of the remarkable conclusions of our study is that central clock neurons survive and continue to exhibit circadian oscillation of at least one clock-related protein in the face of Aβ-induced behavioural arrhythmia. This was most apparent in flies expressing pan-neuronal arctic Aβ₄₂ (Aβ₄₂arc). Despite exhibiting locomotor arrhythmia from ~12 dae, these flies had morphologically normal central clock neurons until at least 30 dae. Specifically, the PDF-positive sLNv neurons, which are essential for DD rhythmicity, remained intact. Furthermore, the non-PDF clock neurons continued to express the Per-luciferase reporter construct (8.0-luc:9) in a circadian pattern that was essentially identical to rhythmic control flies (Fig. 3E). Concordant with this finding, immunostaining of Per protein in Aβ₄₂arc-expressing flies also confirmed its normal circadian oscillation in clock neurons during constant darkness and despite Aβ-induced arrhythmia (in DN1s, LNds and sLNvs; supplementary material Fig. S2). When we subsequently restricted the neuronal expression of Aβ to the clock system using tim-gal4 we observed equivalent behavioural arrhythmia; however, in this context, Aβ₄₂arc was insufficiently potent. Instead, the highly toxic tandem Aβ₄₂ construct was required to induce arrhythmia and was accompanied by some loss of both sLNv and ILNv neurons. Despite the expression of this highly toxic Aβ construct, and the consequent loss of PDF neurons, the molecular oscillations in dorsal clock neurons continued unabated (Fig. 5). Taken together, our data indicate that cell-autonomous Aβ toxicity is insufficient to disrupt the oscillation of the central molecular clock.

Notable was the lack of a circadian phenotype even when we expressed the highly toxic TAβ₄₂ in PDF neurons alone. Indeed, the resistance of the PDF neurons to Aβ₄₂ expression has been remarked upon in passing by others (DiAngelo et al., 2011). When we expanded the expression of TAβ₄₂ to include all clock neurons, by using tim,repo-gal80, the flies became behaviourally arrhythmic. Indeed, further expansion to also include tim-positive glial cells (using tim-gal4), resulted in a progressive circadian degradation in TAβ₄₂ flies and so supports a role for glia in modulating clock neuronal activity (Ng et al., 2011). Such evidence highlights the contribution of circuits peripheral to the central clock neurons in mediating Aβ-linked circadian locomotor deficits, particularly in the ageing brain (Nakamura et al., 2011; Luo et al., 2012).

Our conclusions are concordant with Pallier and colleagues, who studied the R6/2 Huntingdon’s disease (HD) mouse (Pallier et al., 2007) and found that a Per-luciferase reporter continued to oscillate in ex vivo SCN preparations despite the behavioural arrhythmia of the donor animals. These investigators, like us, concluded that the primary target for circadian disruption in the R6/2 mouse is external to the central pacemaker. Such parallels between AD and HD models indicate that damage to the communication between central pacemaker neurons and other neuronal circuits might be a pathological feature that is common to neurodegenerative disorders. In the human context, the central clock neurons in the SCN are thought to use the pineal gland as one of their downstream targets; in this regard it is interesting to note that one consequence of AD is that the secretion of melatonin from the pineal becomes arrhythmic (Wu et al., 2006).

The manifestation of circadian locomotor rhythmicity in the fly requires the synchronisation of the molecular clocks in central clock neurons, downstream neurons and in the peripheral tissues. All this is thought to be dependent on both peptidergic paracrine and synaptic communication. Our work has particularly implicated dysfunction in the dorsal protocerebrum as an important pathological mechanism, on the basis of reduced PDF staining and synaptotagmin-GFP (sytGFP) intensities in this area (Fig. 5D). When determining the mechanisms underlying Aβ toxicity on pacemaker PDF neurons, we have demonstrated that PDF peptide signal is reduced by expressing Aβ not in PDF neurons themselves but in neighbouring cells that are in communication with PDF neurons (i.e. tim,pdf-gal80>TAβ₄₂, Fig. 6Aiii). Such non-cell-autonomous toxicity of Aβ has long been suspected (Busche et al., 2008); however, to our knowledge, this is the first in vivo demonstration of toxicity in one neuron as a consequence of Aβ being expressed explicitly by its neighbours. Considering the importance of PDF peptides in synchronising clock neurons under DD conditions, such reduced PDF signals likely contribute to the behavioural arrhythmia in tim,pdf-gal80>TAβ₄₂ flies. Furthermore, the cell-autonomous toxicity of Aβ on PDF neurons likely had no role in generating these circadian phenotypes (Fig. 6A).

Nevertheless, Aβ-mediated arrhythmia cannot be equated to a pure loss of the PDF signal, because killing PDF neurons by expressing hid (pdf>hid, Table 1) results in characteristic short period rhythms in the remaining rhythmic subpopulation, something that we do not see in any of our Aβ-expressing flies. Notably, the behavioural abnormalities in tim>TAβ₄₂ and tim,pdf-gal80>TAβ₄₂ flies resembles the phenotype of tim>tetanus-toxin, in which synaptic blockade results in arrhythmic flies despite an intact molecular clock (Kaneko et al., 2000). In summary, our findings signify Aβ-mediated damage to both axonal outputs and paracrine signalling in clock neurons.

It has previously been documented that animals in a rhythmic LD environment live longer than those exposed to rhythm-disrupting light cycles (Pittendrigh and Minis, 1972; Davidson et al., 2006; Park et al., 2012). Similarly, loss of normal circadian behaviour in humans is associated with increased morbidity and mortality (Paudel et al., 2010; Tranah et al., 2011). Of particular interest recently has been the finding that behavioural and molecular arrhythmia in per⁰¹ flies is accompanied by increased oxidative stress (Krishnan et al., 2012) and this could increase neurotoxicity in AD (Rival et al., 2009). However, it is unclear whether it is the circadian behaviour pattern per se, or alternatively an entrained molecular clock, that prolongs life. In this regard we have made the interesting observation that profoundly arrhythmic Aβ₄₂arc flies live longer when exposed to LD as compared to those in the clock-disrupting LL environment. Indeed, the proportional increase in median survival is identical for both control and arrhythmic Aβ flies on going from LL to LD. In the absence of a visible behavioural correlate, it seems likely that an entrained molecular clock is beneficial and it is not the behavioural rhythms that prolong life.

Nevertheless, our data agree with the findings of Park and colleagues who concluded that the harmonious interaction of endogenous and environmental rhythms is optimal for longevity, something that was lost in their per1per2 double-null mice (Park et al., 2012). Much is still unclear though; for example, we do not know whether entrainment needs to be central, or whether entrainment of one or more peripheral tissue clocks is sufficient. We can also speculate that the harmonious interplay of endogenous rhythms and behavioural activity might have an important role in protecting the organism from the oxidative stress that is a key feature of both arrhythmic organisms and individuals with AD. In particular, the circadian variation in antioxidant proteins such as peroxiredoxins (O’Neill et al., 2011; Edgar et al., 2012) might be timed to best protect the organism from the stress of oxidative cellular metabolism. Indeed it is notable that our arrhythmic Aβ-expressing flies show a general dampening in circadian behaviour, being moderately active throughout a 24-hour cycle. By contrast, per⁰¹ flies, which have a genetically impaired molecular clock, seem to respond to the light phase with activity and to dark with relative inactivity (Fig. 2Civ), being entirely arrhythmic only in continuous dark. Comparing the oxidative consequences of behavioural arrhythmia in these two contexts could provide interesting insights into pathogenesis of AD.

Nevertheless, these findings have implications for the environment that we provide for individuals with AD; indeed, it is already established that good circadian light hygiene can result in circadian behavioural improvements (Coogan et al., 2013). By contrast, our work has shown that, once Aβ-expressing flies became arrhythmic in a dark environment, light-dark cycling can no longer significantly restore circadian behaviour patterns. Despite this, our work has indicated that the benefits of a clean light-dark environment might not be expressed as improved behavioural endpoints. Our research points to potentially disease-modifying benefits of an entrained molecular clock, possibly as a consequence of reduced oxidative damage, something that is known to characterise AD from its earliest stages (Nunomura et al., 2001; Markesbery et al., 2005).

All gal4 and gal80 lines expressing in the Drosophila clock system are gifts from Prof. Ralf Stanewsky (Queen Mary, London, UK), including tim-gal4 (27), tim-gal4 (62), tim-gal4 (67), tim-gal4 (86), tim-gal4 (27),pdf-gal80 and pdf-gal4 (Kaneko and Hall, 2000; Chen et al., 2011). Various UAS-Aβ lines, including UAS-Aβ₄₀, UAS-Aβ₄₂, UAS-Aβ₄₂arc, UAS-TAβ₄₀ and UAS-TAβ₄₂, were previously generated by the site-specific Phi31C system using acceptor line 51D (therefore as the background control) and backcrossed to w¹¹¹⁸ (Jahn et al., 2011; Speretta et al., 2012). The acceptor site is marked by RFP; however, we recently noticed that 20% of our w¹¹¹⁸;51D line lost this signal during backcrossing. The pan-neuronal expression of Aβ is driven by elav-gal4^c155 (Crowther et al., 2005). The fly strain, tim,repo-gal80, expressing Gal4 exclusively in timeless clock neurons, were generated by combining two transgenes: tim-gal4 (67) and repo-gal80(N18) (Awasaki et al., 2011). The UAS-synaptotagmin-GFP fly strain is a gift from Dr Cahir O’Kane (Department of Genetics, University of Cambridge). All the flies are reared in cornmeal food vials at 25°C and 70% humidity with continuous light-dark cycles (14 hour:10 hour).

The Drosophila circadian locomotor assay is adapted from that described previously (Chen et al., 2011). No more than 20 adult male flies of each genotype are aged in a cornmeal food, replaced every other day, before being transferred individually to a glass tube containing 2% w/v agar and 5% w/v sucrose. The age of the flies is expressed as the day after eclosion (dae) in this study. One-dimensional locomotor activity of each individual fly is then detected continuously by summing the beam crosses every 30 minutes in an automated infrared beam monitoring system (DAM system, Trikinetics, Waltham, USA). The DAM apparatus is placed in a temperature-regulated incubator (Model 200, LMS Ltd, UK), in which the light condition is regulated by a compact fluorescent lamp (660 lumen, Eveready, UK) controlled by an external 24-hour timer.

For detecting intrinsic circadian locomotor rhythm, flies at a given age are first entrained by 3 days of 12 hour:12 hour light-dark cycles (LD cycles) followed by 7 days of constant darkness (DD) at 25°C. The overall level of locomotor activity for individual flies was calculated by averaging the beam crosses/30 minutes over the 10-day duration of the experiment. The time-series of daily activity (actogram) for each fly is plotted and analysed using the Flytoolbox in MATLab software (Levine et al., 2002). The free-running period of the individual time-series under constant darkness is determined by an autocorrelation base method (Levine et al., 2002), in which the rhythmicity statistic (RS) value is derived as a measure of rhythmic robustness. The RS value is the ratio of the autocorrelation coefficient value of an activity time-series to its 95% confidence interval of sampling error (Levine et al., 2002). The mean and median of the RS values are calculated for all tested genotypes (Table 1). All flies with an RS value ≤1.5 are classed as arrhythmic (Levine et al., 2002; Chen et al., 2011). The rhythmic percentage is the fraction of flies that achieve an RS >1.5. The mean and median of the period (hour) are calculated for each genotype from all rhythmic individuals. D’Agostino and Pearson omnibus normality test (GraphPad) was used to verify the normality of each dataset in this study before using parametric statistics; otherwise, non-parametric statistics have been applied. Non-parametric one-way ANOVA (Kruskal-Wallis with Dunn’s comparison post-test) or Student’s t-test are used to analyse differences in rhythmicity (RS) and the period between various genotypes and age groups (GraphPad software, Prism).

To assess whether LD rhythm can be re-entrained following DD, 14-dae elav>Aβ₄₂arc and elav>51D flies are first entrained by 4-day LD cycles, followed by 6 days of DD and a secondary series of four LD cycles with a 6-hour phase delay as compared with the primary LD. Flies with an RS value ≤1.5 are defined as arrhythmic during DD. The average actogram are plotted by Flytoolbox for all LD and DD sessions. In LD, the anticipatory activity ramps at dark-light (morning) and light-dark (evening) transitions are visualised using histograms plots. Quantification of the evening anticipation is chosen because of a clear difference in evening ramping activity between elav>51D flies and the arrhythmic period-null mutants (per⁰¹) (see Stoleru et al., 2004) and also it is the fly equivalent of the ‘sundowning’ behaviour that is particularly significant for individuals with AD (Volicer et al., 2001). The anticipation quantification is based on Harrisingh/Individual Index (Harrisingh et al., 2007) by calculating the ratio of the total activity during the 3 hours before light-dark (evening) transitions to those in the 6 hours before the transitions for the first (2–4 day) and second (12–14 day) LD cycles.

Flies containing UAS-Aβ variants are crossed with elav-gal4^c155 driver lines. Female progeny are collected on the day of eclosion and mated for 24 hours before rearing in either 12-hour LD cycles or constant light (LL) at 29°C. The longevity was analysed as described previously (Crowther et al., 2005) with each assay being comprised of ten tubes of ten flies each (total 100 flies for each genotype). The statistical significance in median survival between LL and LD conditions was determined in two ways: (1) by using the estimates from the 100 individuals with the log-rank test (n=100) and, (2) more conservatively, by using the non-paired Student’s t-test for ten population median survival derived from the ten tubes of ten flies for each conditions. Statistical significance was set at P<0.05.

Luciferase assays are modified from Chen et al. (Chen et al., 2011). Briefly, male flies expressing Per-luciferase protein fusion, 8.0-luc:9 (Veleri et al., 2003), and Aβ variants were generated from crossing elav-gal4;8.0-luc:9 (this study) or 8.0-luc:9;tim-gal4 [obtained from Stanewsky (Hodge and Stanewsky, 2008)] to UAS-Aβ strains. Flies of each genotype are loaded in a white 96-well microtiter plate containing 5% w/v sucrose, 1% w/v agar and 15 mM luciferin (L-8220, Biosynth AG, Switzerland). Bioluminescence emitted from the flies was measured in a Packard Topcount Multiplate Scintillation Counter at 25°C for 2–3 days of LD before entering DD. Data were plotted and analysed using BRASS Version 2.1.3 (Locke et al., 2005). Fast Fourier transform-non-linear least squares (FFT-NLLS) was also performed by BRASS to estimate the period and the relative amplitude of each luciferase time-series. Relative amplitude error (Rel-amp error) values are used as a measure for rhythm robustness: if >0.7 then the individual luciferase activity would be assigned as arrhythmic (Stanewsky et al., 1997). Non-parametric one-way ANOVA are performed to analyse the significant difference (P<0.05) in the estimated period, amplitude and Rel-amp error among genotypes.

The immunohistochemistry protocol is modified from that previously described (Hermann et al., 2012). After 3 days entrainment in LD conditions, male flies of each genotype at the given age were fixed at the indicated ZT and CT (defined below) in 4% w/v paraformaldehyde/PB0.1%T (0.1 M phosphate buffer, pH 7.4 with 0.1% v/v Triton X-100) at room temperature for 2.5 hours. ZT denotes zeitgeber time with ZT0 indicating dawn and ZT12 dusk during LD cycles. CT denotes constant time, with CT00 indicating subjective dawn and CT12 subjective dusk. After fixation, the samples are washed three times with PB at room temperature (RT). The whole brain was dissected out and blocked with 10% v/v goat serum in PB with 0.5% v/v Triton X-100 (PB0.5%T) for 2 hours at RT and stained with monoclonal mouse anti-PDF (1:1000, PDFC7, Developmental Studies Hybridoma Bank, USA) and polyclonal rabbit anti-Aβ_1–16 (1:500, SIG-39322, Covance) in PB0.5%T at 4°C for 48 hours. After washing six times in PB0.1%T, the samples are incubated at 4°C overnight with Alexa-Fluor-647-conjugated anti-rabbit and Alexa-Fluor-488-conjugated anti-mouse antibodies (Molecular Probes) diluted 1:300 in PB0.5%T. For Per and Aβ double staining (supplementary material Fig. S2), rabbit anti-Per (1:1000, gift from Ralf Stanewsky, QMUL) and monoclonal mouse anti-Aβ_1–16 (1:500, 6E10, Covance) are used and the secondary antibodies are Alexa-Fluor-647-conjugated anti-mouse and Alexa-Fluor-488-conjugated anti-rabbit (Molecular Probes). Brains were washed six times in PB0.1%T before being mounted in Vectashield. Samples were stored at 4°C until examination under a Nikon Eclipse C1si confocal microscope.

Image stacks are acquired along the anterior-posterior axis (z-axis) of the fly brain for each genotype from the confocal microscope with 40× magnification. ImageJ software was used to process and analyse all images. Large (lLNvs) and small (sLNvs) PDF neurons are identified by their anatomical location, size and PDF-peptide-positive staining. The number of the two neuron groups was counted separately per brain hemisphere (normally four per hemisphere of each neuronal group) (Helfrich-Förster et al., 2007) in the indicated genotype. The ratio of brain hemispheres with four PDF-positive neurons versus less than four was calculated. The significance of the differences was calculated using the χ²-test in GraphPad Prism.

The brightest PDF signal in the dorsal terminus of sLNvs in individual brains was identified across the z-axis by tuning signal gain around saturation. A single confocal image was then taken below signal saturation with the same laser intensity and signal gain across all samples of the indicated genotypes. The PDF signal (S_pdf) was quantified by a ROI (region of interest) mask in ImageJ software. The same ROI mask was then moved to brain areas with no PDF signal to quantify the background signal (S_b), which was then subtracted from S_pdf. The average of S_pdf values, corrected in this way, was calculated for all individuals of the indicated genotypes and the difference among genotypes are determined by non-parametric one-way ANOVA.

Image stacks containing PDF neurons are captured along the z-axis. The mean grey scale pixel intensities of GFP within PDF neurons were calculated by individual ROI circular masks outlined by the PDF-positive cell body (GFP_pdf). The average GFP signal in the observed brain area (GFP_b) for the indicated genotype was calculated from all the image stacks in a fixed field of view (318 μm²). Both GFP_pdf and GFP_b signals were documented and compared among genotypes by non-parametric one-way ANOVA.

We thank Drs Ralf Stanewsky and Elena Carbognin for their helpful discussions and Drs Werner Wolfgang and Kalina Davies for their assistance with luciferase assays in SBCS, Queen Mary University of London. We thank Drs Ralf Stanewsky, Takeshi Awasaki, Sara Imarisio, Cahir O’Kane and Yuu Kimata for providing fly strains and equipment.