Neuropeptide‐Dependent Spike Time Precision and Plasticity in Circadian Output Neurons

The Drosophila strains utilized in this study were acquired from the Bloomington Drosophila Stock Center (Bloomington, IN, USA), except for UAS‐slob RNAi obtained from the Vienna Drosophila Research Center (VDRC: 100987). To target and visualize the PI neurons, the Dilp2‐Gal4 line (a gift from Amita Sehgal) and UAS‐CD4‐tdGFP line (BDSC: 35836) were recombined through standard genetic recombination techniques to enable use for electrophysiological recordings. We established a split‐GAL4 driver by recombining R20A02‐ad (BDSC: 70582) and R18H11‐DBD (BDSC: 69017), and UAS‐NaChBac (BDSC: 9466) and UAS‐dTRPA1 (BDSC: 26263) were used to activate Dh31⁺‐DN1p cells targeted by R20A02‐ad;R18H11‐DBD. All strains, with the exception of R20A02‐ad;R18H11‐DBD, were outcrossed into the iso31 (BDSC: 5905) genetic background at least 4 times. The flies were nourished with standard Drosophila food comprising molasses, cornmeal, and yeast. Flies were stocked in a 25°C incubator (DR‐36VL, Percival Scientific, Perry, IA, USA) following a 12 h:12 h light:dark cycle and kept at 65% humidity. All experiments involved female flies (5–8 days old) and were conducted at 25°C (except for thermogenic experiments based on 22°C and 28.5°C), adhering to all pertinent ethical regulations for animal testing and research at Case Western Reserve University.

Dh31 and Dh44 DNA were amplified from their template DNA based on gBlocks Gene Fragments (Integrated DNA Technologies) by using UltraRun LongRange polymerase (206442; QIAGEN). For PCR‐based amplification of Dh31, 5′‐AAAGCGATCGCATGACAAACCGAT‐3′ and 5′‐GTTTAAACTTAGACATCGGTCTCGG‐3′ were used as primers, and for amplification of Dh44, 5′‐AAAGCGATCGCATGATGAAAGCCACA‐3′ and 5′‐GGGCGGTT TAAACTTAATTAACGTTAT‐3′ were used as primers. These corresponded to nucleotides 1–351 or 3781–4132 in the Dh31 sequence and corresponded to nucleotides 1–1071 or 5311–6382 in the Dh44 sequence. Amplification was performed by using the following thermal program: 93°C for 3 min; 35 cycles of 95°C for 30 s, 54°C for 30 s, and 65°C for 10 min, followed by one cycle at 72°C for 10 min. The PCR products were separated by electrophoresis on a 1.5% agarose gel. The DNA was extracted from the agarose gel using the GeneJET Gel Extraction Kit (K0691; Thermo Scientific). The samples were purified to reach sufficient concentrations before proceeding to each step. Purification was performed by ethanol precipitation and the Wizard SV Gel and PCR Clean‐Up System (A9281; Promega). DNA fragments encoding Dh31 and Dh44 were subcloned into the SgfI and PmeI sites of the pF25A ICE T7 Flexi Vector (L1061, Promega) available in the kit of the TnT T7 Insect Cell Extract Protein Expression System (L1101, Promega), according to manufacturer instructions. The identities of the subcloned PCR products were verified by Sanger Sequencing analysis by using primers 5′‐CGGATGGCCTTTTTGCG TTTCTA‐3′ and 5′‐CTTCCTTTCGGGCTTTGT TAG‐3′ for Dh31, and 5′‐AAAGCGATCGCA TGATGAAAGCCACA‐3′ and 5′‐GGGCGGT TTAAACT TAATTAACGTTAT‐3′ for Dh44. For protein synthesis, the TnT T7 Insect Cell Extract Protein Expression System was used to allow both transcription and translation to occur in a single reaction. TnT T7 ICE Master Mix was mixed with 4 μg of plasmid DNA template and brought up to a volume of 50 μL with nuclease‐free water. The reactions were incubated at 30°C for 4 h to allow protein synthesis to occur. Synthesized Dh31 and Dh44 were stored at −80°C until use. These synthesized Dh31 and Dh44 were used at a final concentration of 10⁻⁶ M when they were tested alone (i.e., Dh31 or Dh44), and at a final concentration of 5 × 10⁻⁷ M when they were tested as a cocktail (i.e., Dh31 and Dh44). The selected concentration was based on a previous study (Shafer et al. 2008). As a control vehicle, TnT T7 ICE Master Mix with Luciferase ICE T7 Control DNA was used.

We performed perforated patch‐clamp and sharp‐electrode electrophysiological recordings from a PI neuron using membrane‐coated glass electrodes (Jameson et al. 2024). We used a perforated patch clamp in order to acquire action potential firing with current‐clamp mode and KCa currents with voltage‐clamp mode. On the other hand, we used sharp‐electrode electrophysiological recordings to acquire synaptic potentials within axonal regions under tonic hyperpolarization current injection, which are hard to hold with the patch clamp technique. We used membrane‐coated glass electrodes to make recording more stable as we found that membrane‐coated glass electrodes were helpful for suppressing artifactual signal variability, which is mainly derived from access resistance fluctuations during the recording (Jameson et al. 2024). A lipid membrane was created with the use of 7.6 g/L egg yolk lecithin (440154; Sigma‐Aldrich) and 2‐mM cholesterol (C8667; Sigma‐Aldrich) by dissolving them with hexane solvent (34859; Sigma‐Aldrich) and acetone (270725; Sigma‐Aldrich) for 1 h at room temperature using an ultrasonication machine, as previously described (Jameson et al. 2024). Hexane and acetone were evaporated through pressure injection of inert nitrogen, followed by an additional incubation under the decompression chamber. The hexane/acetone solvent was entirely removed, and the egg yolk lecithin and cholesterol were transferred to a mixture of liquid paraffin/squalene (7/3, v/v), then incubated at 80°C overnight. The following day, the prepared lipid was promptly utilized for electrode preparation. Patch electrodes (12–20 MΩ) were formed using a Flaming‐Brown puller (p‐97; Sutter Instrument, CA, USA) with thoroughly cleaned borosilicate glass capillaries (OD/ID: 1.2/0.68 mm). The electrode tip was refined with a microforge (MF200; World Precision Instruments, FL, USA). Sharp electrodes (120–190 MΩ) were formed using a laser‐based micropipette puller (P‐2000, Sutter instrument) with thoroughly cleaned quartz glass capillaries (OD/ID: 1.2/0.6 mm). These electrodes were coated with the prepared phospholipids using a tip‐dip protocol (Jameson et al. 2024). Briefly, we initially dipped the electrode tip into an internal electrode solution in a custom‐made reservoir, and then the prepared lipid solution was loaded onto the surface of the internal electrode solution. Following the application of the minimum amount (< 20 μL) of prepared lipid solution onto the surface of the internal electrode solution, the electrode was vertically lifted using a micromanipulator. The electrode was prepared for electrophysiological recording through standard backfilling of the internal electrode solution by using a microfiller (MF34G MicroFil; World Precision Instruments). The internal pipette solution loaded into the tip contained 102‐mM potassium gluconate, 0.085‐mM CaCl₂, 0.94‐mM EGTA, 8.5‐mM HEPES, 4‐mM Mg‐ATP, 0.5‐mM Na‐GTP, 17‐mM NaCl, and pH 7.2 and was utilized for all patch‐clamp recording experiments. For sharp electrode intracellular recordings, a 1‐M KCl internal pipette solution was loaded into the pipette. Filtering of the internal pipette solutions was performed using a syringe filter with a pore size of 0.02 μm (Anotop 10, Whatman).

In vivo preparation for electrophysiology of PI neurons was performed as previously described (Tabuchi et al. 2018). To anesthetize the flies, they were chilled for 10 min and glued to a 0.025 mm thick metal shim using dental wax or UV light cure adhesives. The cuticle was then peeled off to expose the surface of the brain, and the tethered fly was mounted in a chamber containing Drosophila physiological saline (101‐mM NaCl, 3‐mM KCl, 1‐mM CaCl₂, 4‐mM MgCl₂, 1.25‐mM NaH₂PO₄, 20.7‐mM NaHCO₃, and 5‐mM glucose, with osmolarity adjusted to 235–245 mOsm and pH 7.2), which was pre‐bubbled with 95% O₂ and 5% CO₂. The large trachea and intracranial muscles were removed. The glial sheath surrounding the PI neurons was carefully removed using sharp forceps following the treatment of the enzymatic cocktail, collagenase (0.1 mg/mL), protease XIV (0.2 mg/mL), and dispase (0.3 mg/mL), at 22°C for 1 min, which is essentially the same condition when we conduct electrophysiological recordings from DN1 neurons. A small stream of saline was repeatedly pressure‐ejected from a large‐diameter pipette to clean the surface of the cell body under a dissection microscope. The PI neurons were visualized via tdGFP fluorescence by utilizing the PE4000 CoolLED illumination system (CoolLED Ltd., Andover, UK) on a fixed‐stage upright microscope (BX51WI; Olympus, Japan).

Perforated patch‐clamp was conducted using somatic recording, following procedures described in previous studies (Nguyen et al. 2022; Jameson et al. 2024). Escin was prepared as a 50‐mM stock solution in water and was added fresh to the internal pipette solution, achieving a final concentration of 50 μM. To prevent light‐induced changes, the filling syringes were wrapped with aluminum foil due to the light sensitivity of escin. The escin pipette solutions demonstrated stability for several hours after mixing in the filling syringe, with no observable formation of precipitates. Junction potentials were nullified before the formation of a high‐resistance seal. Perforated patches were allowed to spontaneously develop over time following the high‐resistance seal formation. Once breakthrough was apparent, indicated by the gradual development of a large capacitance transient in the seal test window, monitoring of access resistance was initiated using the membrane test function. From that point onward, access resistance (Ra) was continuously monitored throughout the final stages of the perforation process until it stabilized (Ra stably < 40 MΩ). To isolate KCa currents, we perfused saline containing 10⁻⁷ M Tetrodotoxin to block voltage‐gated sodium currents and 4 × 10⁻³ M 4‐Aminopyridine to block fast‐inactivated voltage‐gated potassium currents (Tabuchi et al. 2018). The PI neurons were held at −70 mV, and two series of 200‐ms voltage pulses were delivered in 10‐mV increments between −80 and 60 mV, with a −90‐mV pre‐pulse delivered before each voltage pulse. The second series was recorded with saline containing 5 × 10⁻⁴ M CdCl₂, which abolished voltage‐activated Ca²⁺ currents. The subtraction of the current trace in the presence of 5 × 10⁻⁴ M CdCl₂ from the current trace without 5 × 10⁻⁴ M CdCl₂ was defined as KCa current. Axopatch 1D amplifier (Molecular Devices) was utilized in obtaining electrophysiological signals, which were then sampled with Digidata 1550B (Molecular Devices) under the control of pCLAMP 11 (Molecular Devices). The electrophysiological signals were sampled at 20 kHz and subjected to a low‐pass filter at 1 kHz. To characterize Dh31 and Dh44 inducible responses, we perfused apamin (10⁻⁵ M) and/or cadmium (10⁻⁴ M) together with Dh31 and Dh44 (Jedema and Grace 2004).

Intracellular recording was performed by inserting sharp microelectrodes as described in previous studies (Nguyen et al. 2022; Jameson et al. 2024). We utilized this method in order to acquire stable synaptic potentials under tonic hyperpolarization current injection. The electrode was introduced into the region characterized by dense tdGFP signals within the axonal regions of PIs in Dilp2‐Gal4 > UAS‐CD4‐tdGFP flies. During the insertion of the electrode, tdGFP signals served as the basis for initial visual inspection, while the depth of insertion was directed by alterations in sound (Model 3300 Audio Monitor, A‐M Systems) and the generation of membrane potential. To facilitate this process, “buzz” pulses were added just before the electrode was ready to cross the membrane. The duration of the pulse was determined by a technique called “advance air shooting.” If the microscope revealed that the electrode tip was physically shaking, the duration was considered excessive. We utilized the longest duration in the range where the electrodes remained stationary, typically between 2 and 5 ms. Commencement of membrane potential recordings occurred once the membrane potential had stabilized, typically requiring several minutes. Recordings were conducted with an Axoclamp 2B with HS‐2A × 1 LU headstage (Molecular Devices) and sampled with Digidata 1550B interface, controlled by pCLAMP 11 software on a computer. The signals were sampled at 10 kHz and subjected to a low‐pass filter at 1 kHz.

Brains or thoracic ganglia were fixed in 4% PFA at 4°C overnight. After several washes with phosphate‐buffered saline (137‐mM NaCl, 2.7‐mM KCl, 10‐mM Na₂HPO₄, 1.7‐mM KH₂PO₄) + 0.3% Triton X‐100 (PBST), samples were incubated with rabbit anti‐GFP (Thermo Fisher, 1:200) and mouse anti‐BRP (nc82, Developmental Studies Hybridoma Bank, 1:50) at 4°C overnight. After additional PBST washes, samples were incubated with Alexa488 anti‐rabbit (Thermo Fisher, 1:1000) for anti‐GFP staining and Alexa568 anti‐mouse (Thermo Fisher, 1:1000) for anti‐BRP staining overnight at 4°C. After another series of washes in PBST at room temperature over 1 h, samples were cleared in 70% glycerol in PBS for 5 min at room temperature and then mounted in Vectashield (Vector Labs). Confocal microscope images were taken under 10× or 63× magnification using a Zeiss LSM800.

Sleep behavior was measured with the use of consolidated locomotor inactivity, as described previously (Tabuchi et al. 2018). Female flies (5–8 days old) were loaded into glass tubes containing 5% sucrose and 2% Drosophila agar medium. Fly behavior was monitored using the Drosophila activity monitors (DAM, Trikinetics, Waltham, MA) in an incubator at 25°C with a 12 h:12 h light:dark (LD) cycle for 2 days to measure sleep. For the UAS‐dTRPA1 experiments, the baseline temperature was set at 22°C to suppress dTRPA activity. To activate dTRPA, the temperature was increased to 28.5°C during ZT12–24 on Day 1. The temperature was returned to 22°C at ZT0 on Day 2. To verify internal data consistency, this cycle was repeated on Days 3 and 4. However, the data from the second temperature increase during ZT12–24 on Day 3 was not used due to concerns about a potential history effect from the initial cycle. The first day following loading was not included in the analysis. Activity counts were collected in 1 min bins, and sleep was identified as periods of inactivity of at least 5 min. Sleeplab (Joiner et al. 2006), a MATLAB‐based (MathWorks) software, was used to quantify sleep behavior.

Statistical analyses (Table 1) were done using Prism software version 10.1.0 (GraphPad), Clampfit version 10.7 (Molecular Devices) or MATLAB 2023b (MathWorks). For comparisons of two groups of normally distributed data, unpaired t‐tests were used. Paired t‐tests were applied for comparisons of electrophysiological signals with normal distributions from the same neuron, before and after stimulation. For comparisons between multiple groups, either one‐way ANOVA with a post hoc Tukey test or a Kruskal–Wallis test with Dunn's multiple comparisons test was used for normally distributed and non‐normally distributed data, respectively. A significance threshold of p‐value < 0.05 denoted statistically significant test results, while asterisks indicated varying levels of significance (*p < 0.05, **p < 0.01, ***p < 0.001, and ****p < 0.0001). Error bars represent means ± SEM averaged across experiments.

The reliability of the spike onset was calculated using Cronbach's alpha, as previously described (Tabuchi et al. 2018). Cronbach's alpha is defined asα=kc¯v¯k−1c¯

where k represents the number of action potentials, v¯ represents the average variance of action potential onset rapidness slope, which was defined by dVm/dt measured from spike onset threshold to peak dVm/dt, and c¯ represents the average inter‐dataset covariance of action potential onset rapidness slope.

To quantify spike patterns, the coefficient of variation (CV) and CV2 of interspike intervals (ISI) were used (Holt et al. 1996). CV reflects a global measure of irregularity and is defined asCV=σ∆t∆t.

where σ is the standard deviation of ISI and ∆t. is the average time of ISI.

CV2 indicates a local measure of irregularity defined as the dispersion of the adjacent ISIs. Thus, CV2 is defined asCV2=2∆ti+1−∆ti∆ti+1+∆tiwhere Δt is an interval time of the ith ISI.

We also calculated the local variation (LV) as alternative measures of local irregularity by computing the dispersion of the two adjacent ISIs. LV is defined asLV=1n−1∑i=1n−13ISIi−ISIi+12ISIi+ISIi+12where ISI_i is the ith ISI and n is the number of ISIs (Shinomoto et al. 2003). While both CV2 and LV compare successive ISIs, they provide different perspectives on spike train variability. CV2 is a measure of overall ISI variability, while LV focuses on the local irregularity of ISI sequences. LV is specifically designed to detect variations in local firing patterns that CV2 may miss, especially in non‐Poissonian processes (Shinomoto et al. 2009). Although the results for CV2 and LV may appear similar, the inclusion of both metrics is critical for a more complete analysis. CV2 provides insight into general ISI variability, while LV highlights local variations that are important for understanding subtle differences in firing patterns that may not be captured by CV2 alone. Thus, both metrics complement each other to provide a more robust characterization of activity patterns.

To determine the peak firing rate (Hz), the largest 1/ISI value representing the instantaneous frequency for each dataset was used. To quantify the decay time constant (т) for each condition, we used the plot of 1/ISI (Hz) versus time (s) and fitted an exponential decay model to the data.

To visualize local PSPs variability, we used a Gaussian mixture model (GMM) as described previously (Tabuchi et al. 2018). GMM assumes that data points arise from a mixture of Gaussian distributions, each defined by its mean and covariance matrix. Model parameters, including mixture weights and Gaussian component parameters, were estimated using the expectation–maximization (EM) algorithm. To preprocess the data, interval timings of PSPs were normalized by the cell‐average firing rates and aggregated across cells. Second‐order distributions were then constructed by logarithmically binning adjacent pairs of normalized intervals into a 2D histogram. Specifically, we utilized a grid of 22 × 22 bins. For different conditions, distinct logarithmic ranges were employed. We fitted 3–5 Gaussian components with full covariance to the joint second‐order distributions. For validation purposes, we generated size‐matched random samples from the model, producing joint and marginal distributions that closely resembled the training data. To create rate‐matched pairs, we first sampled an average PSP frequency from a beta distribution fitted to the PSPs of PI neurons at ZT18–20. Subsequently, we iteratively constructed novel PSP trains by selecting the next interval based on the current interval, continuing until the required number of PSPs was reached. Interval selection was performed via rejection sampling of the continuous conditional probability densities of the GMM, following the exclusion of 200 burn‐in samples. The resulting logarithmic intervals were exponentiated and normalized to yield PSP times, which were then binned into 10‐ms binary signals. To assess the processing performance of the synaptic transmission, we used the area under the curve (AUC) in receiver operating characteristic (ROC) analysis. The AUC in ROC analysis provides a single scalar value that summarizes the performance of a classification model across different thresholds, reflecting the model's ability to discriminate between different classes, making it a useful metric for assessing processing power in tasks where discrimination is important (Hajian‐Tilaki 2013). A model with an AUC close to 1 indicates strong discriminative power, while an AUC close to 0.5 indicates random guessing (Corbacioglu and Aksel 2023). Defining performance based on AUC provides a clear and interpretable measure of model effectiveness. We used the 8900 convergence ratios (from δ = 0.01 to δ = 0.9) of presynaptic to postsynaptic transmission as 8900 input arguments to the ROC analysis.

The synapses between DN1 circadian clock neurons and PI neurons play a critical role in translating circadian rhythms into physiological outputs. We used this connection as a model to study multiplexed neuropeptide mechanisms because this synapse is where Dh31 and Dh44 signaling cross. To access this synapse, we used an in vivo preparation in which the postsynaptic PI neurons are genetically labeled (Figure 1a). We focused on ZT18–20 as a specific time window for our electrophysiological recordings because this time window is associated with high‐quality sleep during the night, which is supported by a steady state of stable neural dynamics of DN1 circadian clock neurons. PI neurons have less spontaneous firing activity during the night, and ZT18–20 has a particularly low rate of spontaneous firing frequency, which is close to 0 Hz (Tabuchi et al. 2018). Despite the physical disturbances associated with live dissections of the samples, which are required for our recordings, we find the PI neuron's clock function is not disrupted. Since both Dh31 and Dh44 are associated with arousal promotion, we hypothesized that bath application of Dh31 and Dh44 would cause an increase in the firing activity of PI neurons. We performed cell‐free protein synthesis of Dh31 and Dh44 and applied the synthesized Dh31 and Dh44 to PI neurons (Figure 1b). Compared to the control (Figure 1c), the spontaneous firing of PI neurons during ZT18–20 increased with the application of Dh31 alone (Figure 1d), Dh44 alone (Figure 1e), and Dh31 and Dh44 together (Figure 1f). We found that a mixture of Dh31 and Dh44 enhanced supralinear firing of PI neurons (Figure 1f). In contrast to the spontaneous firing rate, we did not find significant changes in the resting membrane potential (Figure 1g–j).

Next, we quantified intrinsic membrane excitability by evoking activity through current injection (Figure 2a). While the evoked mean firing rate did not increase with either Dh31 or Dh44 alone (Figure 2c,d), the mixture of Dh31 and Dh44 showed a supralinear increase, demonstrating a strong increase in elicited firing frequency (Figure 2e). To further characterize the effects of Dh31 and/or Dh44 applications on PI neuronal intrinsic firing properties, we also analyze peak firing rate and decay time constant. The maximum 1/ISI value, representing instantaneous frequency, was taken as the peak firing rate. The decay time constant (τ) was determined by an exponential decay model fitting to the 1/ISI vs. time plot. Peak instantaneous frequency did not show a significant difference in the control vehicle (Figure 2f) and Dh31 application (Figure 2g), but the Dh44 application showed a significant increase in the peak instantaneous frequency (Figure 2h). Similar to the analysis of spontaneous firing activity (Figure 1) and the mean frequency of the evoked firing (Figure 2e), the mixture of Dh31 and Dh44 showed a more dramatic increase in the peak instantaneous frequency (Figure 2i). The decay time constant, on the other hand, showed a similar level of reduction compared to the control (Figure 2j) in the applications of Dh31 alone (Figure 2k), Dh31 alone (Figure 2l), and the mixture of Dh31 and Dh44 (Figure 2m). Together, based on results from Figures 1 and 2, these data suggest that Dh31 and Dh44 synergistically interact to enhance the action potential firing of PI neurons during ZT18–20.

To investigate the biophysical origins underlying the Dh31/Dh44‐induced change in action potential firing, we analyzed the kinetics of the action potential waveform (Figure 3a). We focused on analyzing action potential waveforms from spontaneous firing because the kinetics were artifactually distorted in evoked firing (Figure 2a), which is commonly observed in invertebrate unipolar neurons due to the distance between the current injection site and the action potential generation site (Gouwens and Wilson 2009). The action potential waveforms exhibit variation induced by interactions with Dh31, Dh44, Dh31/Dh44, Dh31/Dh44 with apamin (10⁻⁵ M), and Dh31/Dh44 with cadmium (10⁻⁴ M). Cadmium and apamin were used to selectively inhibit calcium and SK channels, respectively (Jedema and Grace 2004). The amplitude and threshold of action potentials were not significantly different in the presence of Dh31/Dh44 (Figure 3b,c), but the reduction of amplitude was observed only in the presence of cadmium (Figure 3b), and the increase of threshold was observed only in the presence of cadmium (Figure 3c). A non‐significant trend in greater afterhyperpolarization (AHP) amplitude was induced by a mixture of Dh31 and Dh44 (Figure 3d). Interestingly, the change in AHP amplitude was only sensitive to cadmium but not apamin. Based on the potential relationship between action potential waveform kinetics and firing precision (Axmacher and Miles 2004; Naundorf et al. 2006; Niday and Bean 2021), we quantified the reliability of action potential timing by using internal consistency of onset rapidness (Tabuchi et al. 2018), which we computed using Cronbach's alpha. We found that the reliability of spike timing was significantly enhanced in the mixture of Dh31 and Dh44, which was reduced by cadmium (Figure 3e). Taken together, these data suggest multifactorial aspects of the biophysical origins underlying the Dh31/Dh44‐induced change in action potential firing to make spike timing of PI neurons more reliable in the presence of the Dh31 and Dh44 mixture.

Spike timing reliability determined by AHP and onset rapidness is closely related to spike patterns in neuronal cells (Scuri et al. 2005; Stiefel et al. 2013; Dumenieu et al. 2015; Trinh et al. 2019; Cui et al. 2022). To investigate whether Dh31/Dh44 applications also influence spike patterns, we analyzed spike frequency adaptation, which is one of the mechanisms shaping spike patterns. The spike frequency adaptation was quantified from the ratio of the 1st ISI to the nth ISI, and an index value closer to 1 would indicate that no spike adaptation occurred, whereas an index value closer to 0 would indicate significant spike adaptation with prolongation of the ISIs (Figure 4a). The Dh31/Dh44 interact and convey significant spike irregularity compared to Dh31 alone and Dh44 alone (Figure 4b–d). CV (Figure 4b), LV (Figure 4c), and CV2 (Figure 4d) each provide complementary information about the temporal structure of neuronal spike trains: CV is sensitive to changes in the overall firing rate, but LV is a measure of variability in the timing of spikes within local time windows, and CV2 is a measure of spike irregularity that considers pairs of consecutive ISIs. Nevertheless, we found that the application of cadmium and apamin decreased Dh31/Dh44‐enhanced spiking irregularity in all metrics (Figure 4b–d). One of the contributing factors in shaping AHP and spike patterns can be KCa channel conductance (Engbers et al. 2012; Stiefel et al. 2013; Sahu and Turner 2021), and KCa channels are known to play a critical role in shaping the activity rhythms of PI neurons (Ruiz et al. 2021). Therefore, we directly quantified KCa channel conductance in PI neurons by voltage‐clamp recordings (Figure 4e). Similar to our previous AHP analysis (Figure 3d), we found that a non‐significant trend in greater K_Ca conductance amplitude was induced by a mixture of Dh31 and Dh44 (Figure 4f,g). SLOB (slowpoke‐binding protein) is known as a key mediator in modifying K_Ca conductance (Shahidullah et al. 2009; Jepson et al. 2013). Thus, we tested if genetic manipulation of SLOB in PI neurons changes Dh31/Dh44‐induced changes in K_Ca conductance. Strikingly, we found that Dh31/Dh44‐induced changes in K_Ca conductance were completely eliminated by introducing SLOB RNAi in PI neurons (Figure 4f,g). These results suggest that Dh31/Dh44‐induced changes in K_Ca conductance were mediated by SLOB in PI neurons.

To reveal if Dh31/Dh44 impacted synaptic drive, we examined postsynaptic potential (PSP) parameters such as amplitude, inter‐event interval, cumulative probability, and probability density. The voltage traces of PSPs are altered when Dh31 and Dh44 are introduced, with high increased activity when both Dh31/Dh44 are present (Figure 5a). The cumulative probability varied for each condition when recording across PSP amplitudes (Figure 5b). The amplitude of PSP for the condition with Dh31 alone and Dh44 alone was significantly greater compared to the control (Figure 5c). The amplitude of PSP for the condition with both Dh31 and Dh44 was significantly greater than not only the control but also Dh31 alone and Dh44 alone (Figure 5c). The addition of apamin in the mixture of Dh31 and Dh44 brought the effects back to the level of Dh31 alone and Dh44 alone, but not to control (Figure 5c).

Amplitude changes in PSP generally reflect postsynaptic changes in response to neurotransmitters (Turrigiano et al. 1998), and we next asked whether the frequency of PSP, which is more related to presynaptic events, was changed. To characterize potential presynaptic changes, the frequency of PSP was analyzed. The probability density during PSP inter‐event intervals for the Dh31/Dh44 and Dh31/Dh44/apamin groups was decreased (Figure 6a). By quantifying the PSP inter‐event interval for each condition, we found that Dh31/Dh44/apamin was significantly different from all other conditions in that its inter‐event interval was longer, indicating a reduction in synaptic activity (Figure 6b). Taken together, these data suggest that Dh31 and Dh44 interact synergistically to enhance the PSPs of PI neurons, which may be mediated by both presynaptic and postsynaptic mechanisms. To identify the computational basis for how Dh31/Dh44 alters the temporal dynamics of synaptic potentials, we implemented two types of computational models. We first used a Gaussian Mixture Model (GMM). Interval timings of two adjacent PSPs were projected into two‐dimensional space, and GMM was used to estimate the geometry of their probability distribution (Figure 6c). We found the emergence of an additional cluster in the presence of Dh31 or Dh44 compared to the singular cluster in the control. Furthermore, in the presence of Dh31 and Dh44, we found the emergence of multiple clusters which was partially diminished by the administration of apamin. We next conducted receiver operating characteristic (ROC) to estimate the processing performance of interval timing of postsynaptic activity of PI neurons. The higher the area under the curve (AUC) in the ROC analysis, the higher the discriminative power, while an AUC close to 0.5 indicates a random chance level (Corbacioglu and Aksel 2023). We found that the interval timing of PSPs in the presence of Dh31 or Dh44 alone showed a higher AUC compared with the control, but their AUC was lower than the interval timing of PSPs in the presence of both Dh31 and Dh44 with the effect partially reduced by apamin (Figure 6d).

DN1 clock neurons are known to be one of the major presynaptic inputs to PI neurons (Barber et al. 2016; Tabuchi et al. 2018), and the subset of DN1 clock neurons expressing Dh31 is known to promote wakefulness (Kunst et al. 2014). DN1 clock neurons are known to be heterogeneous in their role of promoting sleep and/or wakefulness (Guo et al. 2018; Lamaze et al. 2018), so we decided to establish a split‐GAL4 driver to define Dh31‐DN1ps neurons. R20A02‐ad expresses the p65 activation domain associated with Dh31, and R18H11‐DBD expresses the DNA binding domain of GAL4 associated with PDF receptors (Dionne et al. 2018). Thus, we used R20A02‐ad;R18H11‐DBD as an intersectional approach to genetically extract a putative subset of Dh31⁺‐DN1p clock neurons. We found that R20A02‐ad;R18H11‐DBD>UAS‐thGFP flies label DN1p cells (5.1 cells ± 1.5, N = 90 brains) and a few uncharacterized cells in VNC (Figure 7a). To investigate the potential involvement of Dh31‐positive DN1 clock neurons, we first assessed whether genetic neural activation in R20A02‐ad;R18H11‐DBD flies would lead to changes in sleep and activity patterns using UAS‐NaChBac (Figure 7b). No significant difference in daytime (ZT0–12) sleep was found, but a significant decrease in nighttime (ZT12–24) sleep was found in flies expressing UAS‐NaChBac in putative Dh31‐DN1ps (Figure 7c). We also found an increase in sleep bout number during nighttime in flies expressing UAS‐NaChBac in putative Dh31‐DN1ps (Figure 7d). We also analyzed locomotor profiles (Figure 7e) and found increases in both nighttime activity time (Figure 7f) and activity level (Figure 7g) in flies expressing UAS‐NaChBac in putative Dh31‐DN1ps; it is possible that the expression of NaChBac drove hyperactivity. We also quantified activity and sleep profiles of flies having NaChBac alone, without R20A02‐ad;R18H11‐DBD, but we did not find any significant difference compared to control (Figure S1).

Constitutive expression of UAS‐NaChBac can cause artifactual effects not only during experiments but also during development. To exclude this possibility, we also performed acute activation of Dh31‐DN1ps based on thermogenetics. As a thermogenic tool, we used dTRPA1, a temperature‐sensitive ion channel, which can be activated by raising the temperature (Hamada et al. 2008). We found that activating the putative Dh31‐DN1p neurons through temperature elevation significantly reduced sleep in R20A02‐ad;R18H11‐DBD>UAS‐dTRPA1 flies when compared to R20A02‐ad;R18H11‐DBD>iso31 controls during the nighttime (Figure 8a,b), although we found that these flies exhibited reduced sleep during the daytime even without temperature elevation (Figure 8b,c). However, we found no significant difference in sleep during the nighttime without temperature elevation (Figure 8c), suggesting that the reduction in sleep during the nighttime was certainly due to thermogenetic activation. In addition to sleep amount, we found facilitated nighttime sleep fragmentation by thermogenetic activation, as evidenced by an increase in the number of sleep bouts (Figure 8d). We also analyzed activity profiles (Figure 8e) and found an increase in active time during nighttime (Figure 8f), but we found the daily activity level during wakefulness was reduced (Figure 8g), which was opposed to chronic activation of putative Dh31‐DN1ps by expressing NaChBac. We also performed another thermogenetics experiment to compare R20A02‐ad;R18H11‐DBD>UAS‐dTRPA1 flies with UAS‐dTRPA1 alone control flies (Figure S2), and found essentially similar results when comparing R20A02‐ad;R18H11‐DBD>UAS‐dTRPA1 flies to R20A02‐ad;R18H11‐DBD>iso 31 control flies (Figure 8).

The peer review history for this article is available at https://www.webofscience.com/api/gateway/wos/peer‐review/10.1111/ejn.70037↗.