PER is the major repressor of the core TTFL of the molecular clock. Most clock neurons in culture show cytoplasmic enrichment of PER and TIM (Figures 2C–E), resembling the state of clock neurons around ZT14, a time when CLK/CYC-target gene transcription is at its maximum. Together with the finding that no periodic PER-TDT nuclear localization is observed in cultured clock neurons (Figure 2F), these results suggest that the feedback inhibition of CLK/CYC transcriptional activity by nuclear PER is impaired in isolated clock neurons, resulting in the constant upregulation of CLK/CYC-mediated transcription. Constant rise of 3 × 69-VNP levels in wt dissociated neurons (Figure 1B) fits this model. Importantly, 3 × 69-VNP levels exhibit fluctuations but not a steady increase in pdfr mutant cells in dispersed culture (Figure 4A). This observation rules out that the steady increase of 3 × 69-VNP levels in wt dissociated neurons is caused by a general impairment of protein degradation systems. Furthermore, it indicates that CLK/CYC-mediated transcription sustains in a stochastic manner in dissociated neurons even in the absence of PDF signaling.

Cytoplasmic enrichment and the absence of rhythmic nuclear accumulation of PER suggest an impairment in nuclear translocation or the stability of nuclear PER. PER nuclear entry is controlled by multiple kinases, including CK2 (Helfrich-Forster, 2005; Chiu et al., 2008), SGG (Martinek et al., 2001), and probably a still unknown proline-directed kinase (Ko et al., 2010). PER phosphorylation by DBT promotes PER degradation, which is prevented by binding to TIM (Kloss et al., 1998; Ko et al., 2002; Kim et al., 2007). Phosphorylated PER enters the nucleus as PER/TIM dimer, where dephosphorylation by PP1 (Fang et al., 2007) and PP2A (Sathyanarayanan et al., 2004) prevents the phosphorylation at the sites promoting the binding of the E3 ubiquitin ligase SLIMB and eventual degradation (Chiu et al., 2008). Therefore, some of these post-translational mechanisms are probably dysregulated in the dispersed clock neurons.

Our results also suggest that signaling downstream of PDF/PDFR may work differently in isolated clock neurons, which exacerbates the dampening of the molecular rhythms. In brain explants during the night, PDF enhances CLK/CYC-mediated transcription via neuronal activity-dependent but c-AMP independent mechanism. At the same time, PDF signaling stabilizes PER via cAMP-PKA dependent mechanism (Li et al., 2014; Seluzicki et al., 2014; Sabado et al., 2017). The latter mechanism seems to be inhibited in dispersed clock neurons, as the addition of PDF to the culture medium or pdfr mutation did not affect PER-TDT expression profile (Figures 3, 4B). Thus, PER is probably less stable in dissociated neurons even in wt culture than in vivo, which also impairs negative feedback on CLK/CYC-mediated transcription.

We previously showed that the addition of PDF in the medium increases the 3 × 69-VNP levels in dissociated neurons, whereas PDF upregulates 3 × 69-VNP levels only during the night in brain explants (Sabado et al., 2017). Collectively, these results postulate that transcriptional response to PDF is negatively regulated during the day in the brain. In contrast, our results here indicate that PER levels are insensitive to PDF signaling in dissociated culture, whereas PDF stabilizes PER during the night in brain explants. Therefore, the nighttime-specific effect of PDF on PER stabilization probably occurs via a positive regulatory mechanism that happens only within the intact circadian circuit. These negative and positive gating of molecular clock response to PDF signaling requires cycling clocks, which appear to be the emergent property of the intact circadian circuitry (Supplementary Figure 3). Intriguingly, recent studies using in vivo Ca²⁺ imaging demonstrated that PDF controls the “morning” and “evening” phases of the Ca²⁺ rhythms in different subpopulation of pacemaker neurons, independent of their molecular phases (Liang et al., 2016, 2017). Thus, our results add to a growing body of evidence that PDF differentially regulates molecular and neuronal rhythms through complex and network-dependent mechanisms.

Of note, unlike the l-LNvs, the s-LNvs do not express the neuropeptide amidation enzyme PHM; hence, the PDF released from the s-LNvs might be either non-amidated or amidated by another amidation machinery (Helfrich-Forster, 2009, Taghert et al., 2001). Since larval LNvs are the precursors of adult s-LNvs, we used non-amidated PDF for all the pharmacological experiments in this and our previous study (Sabado et al., 2017). Non-amidated PDF can trigger PDFR-dependent, time-of-day- and activity-dependent response of molecular clockwork in cultured larval brains as well as TTX-sensitive CLK/CYC-mediated transcription in cultured clock neurons (Sabado et al., 2017). Thus, although non-amidated neuropeptides have shorter half-lives and are less bioactive than the amidated ones, non-amidated PDF is bioactive at least in these assays. Nevertheless, since PDF secreted from larval LNvs and adult s-LNvs may as well be amidated (Taghert et al., 2001; Park et al., 2008), the results of these assays might not fully represent the native activity of endogenous PDF. It will be important in future studies to validate whether non-amidated PDF can activate PDFR and also test the effect of amidated PDF in the same assays.

While no cells in dissociated culture exhibit circadian rhythms in 3 × 69-VNP expression, approximately one-tenth of clock neurons display PER-TDT fluctuations with periods in the circadian range (Figures 1, 4). These results suggest that, certain post-transcriptional mechanisms involved in PER protein stability or transcriptional regulation via per regulatory sequences other than the CRS is still functioning in dispersed clock neurons. These regulatory mechanisms prevent over-accumulation of PER and even produce PER protein oscillations in the circadian range in some cells. Our results thus showcase the emerging notion that post-transcriptional controls reinforce robustness of circadian oscillation (Lim and Allada, 2013).

A previous work in mammals has shown that, in low-density dispersed culture, only ~18% of SCN neurons display circadian rhythms with periods ranging from 15 to 35 h (Webb et al., 2009). This is reminiscent of our observation that ~13% of isolated fly clock neurons exhibited PER protein rhythms. Moreover, it is noteworthy that we did not detect obvious cell-type specificity as to which neurons display circadian rhythms in dissociated culture. This implies that it is probabilistic rather than deterministic whether or not a neuron displays PER rhythms in isolation. Similarly, it was reported that isolated SCN neurons sustain or lose rhythmicity in random fashion (Webb et al., 2009). Thus, both Drosophila and SCN neurons seem to be poor oscillators that require network interactions to generate robust circadian rhythms. Whether or not a given pacemaker neuron sustain circadian molecular rhythms isolated from the intact circuit probably depends on the cell's physiological state, such as ion conductance and excitability. It has been proposed that the sensitivity of SCN neurons to the intercellular communication provides the circadian network the plasticity to promptly respond to changes in environmental conditions and produce a coherent output (Ko et al., 2010; Webb and Oates, 2016). It may also be a common principle of neuronal clocks in Drosophila and mammals.

Drosophila were reared at 25°C on a corn-meal medium under 12:12 h light-dark cycles (LD). The 1982clk-gal4 line (Gummadova et al., 2009), The GAL4 enhancer trap line gal1118 (Blanchardon et al., 2001), cry⁰ (Dolezelova et al., 2007), and pdfr⁵³⁰⁴ (Hyun et al., 2005), UAS-mCD8::VENUS and UAS-mCD8::RFP, 3 × 69-Venus and per-TdT lines (Sabado et al., 2017) have been previously characterized.

Brains of LD-entrained non-wandering L3 larvae were dissected in ice-cold saline solution (Jiang et al., 2005) at ZT11. Dissociated neuron culture was performed as previously described (Kuppers-Munther et al., 2004; Saad et al., 2012) with the following modifications. In brief, the dissected brains were enzymatically treated with 50 unit/mL papain (Worthington) and mechanically dissociated. The cell suspension was then plated on glass-bottom dishes (35 mm MatTek petri dish, 10 mm microwell with 0.16/0.19 mm coverglass) coated with Concanavalin A (Sigma). Approximately 3.1 ~ 4 × 10⁵ cells, which corresponds to the suspension derived from ca. Ten larval brains, were plated in each dish (4 ~ 5 × 10⁵ cells/cm²). Once the cells were attached to the glass bottom, the dish was flooded with SM^active (Supplementary Table 1). The dissociated neuron culture was incubated at 25°C in 80% relative humidity and in constant darkness for 2 days prior to time-lapse imaging. Time-lapse imaging was carried out in the same culture conditions.

For pharmacological experiments, the neuropeptide PDF (custom-made by Chi Scientific, H-NSELINSLLSLPKNMNDA-OH; 2 μM) or the vehicle (DMSO) was added to the cell culture medium just before the start of the time-lapse imaging.

Anti-PER and Anti-TIM immunostaining of the cultured neurons was performed as previously described with some modifications (Nawathean et al., 2005). After 2 or 4 days in culture, the medium was discarded, and the culture was washed with phosphate-buffered saline (PBS); then, the cells were fixed with 1% paraformaldehyde in PBS for 20 min at room temperature. The cells were then washed with 0.1% Triton in PBS, and 1:1,000 rabbit anti-PER and 1:2,500 rat anti-TIM (gift from F. Rouyer) in blocking buffer were used as the primary antibodies. The secondary antibodies, anti-rabbit-Alexa-633 and anti-rat-Alexa-555 (Invitrogen), were diluted 1:500 in the blocking buffer.

A Leica TCS SP5 tandem scanner confocal microscope was used for fluorescence imaging. The same parameter settings were applied to image all the samples of the same type. Immunostained cultured neurons were scanned using a × 63 water-immersion objective with the galvo scanner at 400 Hz. For time-lapse imaging, × 63 objective was used and the images were acquired every 3 h for 48 h using the resonant scanner at 8,000 Hz with high-sensitivity HyD detectors. The bi-directional scan was used together with 8 × line average. A Z-section step of 1.7 μm × 8 was used for imaging dissociated cultured neurons. A 514 nm laser was used to excite VENUS fluorophore (0.68 μW/cm² to image VNP, 0.40 μW/cm² for mCD8::VENUS), and a 561 nm laser was used for the TdTOMATO and mRFP fluorophores (14.8 μW/cm² for PER-TDT, 5.71 μW/cm² for mCD8::RFP). Laser intensity was measured at the level of the sample with a microscope slide power meter (Thorlabs, S170C).

The fluorescence intensities in immunostained cultured neurons and the time-lapse movies of the cultured neurons were measured with the Imaris software (Bitplane). A 3D mask built from the fluorescent signal emitted by the clock neurons over the time course was created using the surface tool of Imaris. The same threshold settings, including background subtraction, were applied to all the samples of the same type. The intensity SUM was then extracted from the statistical data that were automatically generated by the program.

Fluorescence intensity time series data were normalized to the value at t = 0 except for the raster plots for intensity fold-change. Heatmaps representing the fluorescence intensity time course were generated with an in-house R script. Each row represents the data of a single cell. Rows were ordered by the highest intensity over 24 h. To generate raster plots for intensity fold-change, the value at t = n + 1 was divided by the value of t = n, and the log2 of the fold-change value was plotted over time using an in-house R script. The log2 fold-change values were color-coded, in which positive fold-changes are shown with a magenta gradient, no change is shown in black, and negative changes are shown with a green gradient. The hierarchical clustering analysis (complete linkage method) was performed to find similar clusters and visualize the plots accordingly.

For rhythm analysis, a combination of manual inspection and Maximum Entropy Spectral Analysis (MESA) was used to detect rhythmicity and estimate the period. The data of the entire recording period (48 h) were subjected to the analysis. MESA analysis was performed using an in-house R script containing the spec.ar function. MESA was chosen because, unlike other methods, it is adapted for the detection of the rhythms in short or noisy time series (Dowse et al., 1989; Levine et al., 2002; Klarsfeld et al., 2003). First, the intensity time course plots were generated using Excel or Prism (GraphPad Prism version 6.0c for Mac, GraphPad Software, San Diego California USA, www.graphpad.com). Because we found that 6th-order polynomial regression models best fit the data, we superimposed the 6th-order polynomial trend lines to the graphs to facilitate the detection of rhythmicity. Then, the time series data that have circadian rhythms, defined by the peak-to-peak interval from 18 to 36 h, were identified by manual inspection. However, the polynomial trend lines were not used for determining the period. In parallel, the intensity time series data without normalization were analyzed with MESA without filtering. The time series data were scored as circadian only when both the manual and MESA analyses detected the rhythms with a period between 18 and 36 h.

Statistical analyses were performed with GraphPadPrism software to compare the expression of antibodies or transgenes in different genetic backgrounds or pharmacological conditions. Non-parametric tests were used when the data did not follow a normal distribution. To compare the effect of one condition (genetic backgrounds or pharmacological conditions) parametric or non-parametric T-Tests or one-way ANOVA with a correction for multiple comparisons (depending on the number of data sets to compare) were used. For live imaging experiments, time was an added condition, hence, two-way ANOVA with a correction for multiple comparisons were utilized. Details of the used statistical tests are given in the figure legends.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.