Abstract
General anesthetics cause a profound loss of behavioral responsiveness in all animals. In mammals, general anesthesia is induced in part by the potentiation of endogenous sleep-promoting circuits, although “deep” anesthesia is understood to be more similar to coma (Brown et al., 2011). Surgically relevant concentrations of anesthetics, such as isoflurane and propofol, have been shown to impair neural connectivity across the mammalian brain (Mashour and Hudetz, 2017; Yang et al., 2021), which presents one explanation why animals become largely unresponsive when exposed to these drugs. It remains unclear whether general anesthetics affect brain dynamics similarly in all animal brains, or whether simpler animals, such as insects, even display levels of neural connectivity that could be disrupted by these drugs. Here, we used whole-brain calcium imaging in behaving female Drosophila flies to investigate whether isoflurane anesthesia induction activates sleep-promoting neurons, and then inquired how all other neurons across the fly brain behave under sustained anesthesia. We were able to track the activity of hundreds of neurons simultaneously during waking and anesthetized states, for spontaneous conditions as well as in response to visual and mechanical stimuli. We compared whole-brain dynamics and connectivity under isoflurane exposure to optogenetically induced sleep. Neurons in the Drosophila brain remain active during general anesthesia as well as induced sleep, although flies become behaviorally inert under both treatments. We identified surprisingly dynamic neural correlation patterns in the waking fly brain, suggesting ensemble-like behavior. These become more fragmented and less diverse under anesthesia but remain wake-like during induced sleep.
SIGNIFICANCE STATEMENT When humans are rendered immobile and unresponsive by sleep or general anesthetics, their brains do not shut off — they just change how they operate. We tracked the activity of hundreds of neurons simultaneously in the brains of fruit flies that were anesthetized by isoflurane or genetically put to sleep, to investigate whether these behaviorally inert states shared similar brain dynamics. We uncovered dynamic patterns of neural activity in the waking fly brain, with stimulus-responsive neurons constantly changing through time. Wake-like neural dynamics persisted during induced sleep but became more fragmented under isoflurane anesthesia. This suggests that, like larger brains, the fly brain might also display ensemble-like behavior, which becomes degraded rather than silenced under general anesthesia.
Introduction
There is often an assumption that, when simple animals such as insects are immobilized by sleep or anesthesia, their brains are similarly quiescent. Recent sleep studies in insects (Tainton-Heap et al., 2021), fish (Leung et al., 2019), reptiles (Shein-Idelson et al., 2016), and molluscs (Iglesias et al., 2019) present evidence to the contrary, with some stages of sleep showing levels of activity comparable to wakefulness, as has been observed during rapid eye movement (REM) sleep in mammals and birds (Miyazaki et al., 2017). Unlike sleep, deep general anesthesia is thought to produce a coma-like state in humans (Brown et al., 2011). However, certain general anesthetics, such as isoflurane and propofol, also act as sedatives on induction and have accordingly been shown to activate sleep-promoting centers in the mammalian brain, such as the ventrolateral preoptic nucleus (Moore et al., 2012). In humans, the sedative component of general anesthesia is often associated with synchronized brain activity signatures resembling slow-wave sleep (SWS) (Brown et al., 2011), consistent with the idea that these drugs are activating endogenous sleep pathways. However, a recent study using calcium imaging and intracranial recordings in mice showed that only a subset of cortical pyramidal neurons synchronize under general anesthesia (Bharioke et al., 2022), with other neurons remaining active yet uncoordinated. Similarly, another study using calcium imaging in the mouse hippocampus found that anesthetics such as isoflurane fragmented network dynamics, rather than decreasing neural activity (Yang et al., 2021). This aligns with EEG recordings showing a loss of connectivity in anesthetized humans (Lee et al., 2013; Mashour and Hudetz, 2017) and could suggest a different target mechanism than SWS induction, linked to a failure of neural coordination rather than activity.
Like other animals, flies can be anesthetized at similar concentrations that produce loss of consciousness in humans (Zalucki and van Swinderen, 2016; Karunanithi et al., 2018). Interestingly, flies also display lighter and deeper sleep stages (van Alphen et al., 2013). This raises the question of what kind of sleep general anesthetics might be promoting in flies, if they do indeed act in part as sedatives (Nelson et al., 2002; van Swinderen and Kottler, 2014). Over the past decade, there has been much work identifying sleep-promoting neurons in the Drosophila brain, potentially analogous to the ventrolateral preoptic nucleus in mammals. In particular, the dorsal fan-shaped body (dFB) has been identified as an output circuit for a “sleep homeostat” in flies (Pimentel et al., 2016); and consistent with this idea, increased dFB activity has been associated with increased sleep in flies (Donlea et al., 2011; Troup et al., 2018). However, in a recent study using whole-brain calcium imaging in Drosophila, we showed that optogenetically induced dFB sleep did not resemble spontaneous deep sleep: while sleeping flies were rendered unresponsive to external stimuli, brain activity as measured by a calcium indicator was not different from during wakefulness (Tainton-Heap et al., 2021). This suggested that dFB activation induces an “active” rather than a “quiet” sleep stage, at least under acute optogenetic activation conditions. It remains unknown whether general anesthetics such as isoflurane activate sleep-promoting neurons in the fly brain, although a previous study has shown a correlation between isoflurane sensitivity and sleep amount (Kottler et al., 2013).
Here, we used a 2-photon calcium imaging approach in tethered yet behaving flies to address how isoflurane might be achieving its anesthetic effects in this tiny model brain. We first explore whether isoflurane-induced anesthesia activates candidate sleep-promoting neurons, and we then proceed to examine whole-brain dynamics under isoflurane, during baseline activity as well as in response to visual and mechanical stimuli. We find a surprising level of minute-to-minute turnover in neural activity across the waking fly brain and observe significant differences in neural coordination between dFB-induced sleep and isoflurane anesthesia, suggesting entirely different neural mechanisms leading to loss of behavioral responsiveness in optogenetically induced sleep compared with general anesthesia.
Materials and Methods
Flies
Female Drosophila melanogaster flies were raised on yeast/sugar/agar media (1.0:1.5:0.5 g ratio) on a 12-12 light-dark cycle at 25°C. For optogenetic activation, flies were placed on food containing 0.2 mm all-trans retinal at least 24 h before experiments. Only female flies were used because they showed better survival and anesthesia recovery than males in these experiments.
A summary of the genetic reagents used is found in Table 1. For cytosolic GCaMP6s experiments (Chen et al., 2013), UAS-GCaMP6s;UAS-CsChrimson-mCherry flies were crossed to either R23E10-Gal4 (dFB expression) or R28D01-Gal4 (R3m ellipsoid body [EB] expression) flies (Jenett et al., 2012). R23E10-Gal4 experiments in Figure 1 were conducted on the same 6 flies, while R28D01 experiments were conducted on the same 9 flies.
Drosophila genetic reagents used in experiments in this study
For nuclear localized GCaMP experiments, UAS-Chrimson88;LexAop-nlsGCaMP6f;57C10-LexA flies were crossed to w1118 (anesthesia) or R23E10-Gal4 (induced sleep) flies. All nlsGCaMP anesthesia results were produced from the same set of 14 flies, and all nlsGCaMP dFB-induced sleep results were produced from the same set of 8 flies.
Imaging
All imaging was performed with a Bergamo series 2 multiphoton microscope (Thorlabs) with a Ti:Sapphire Mai Tai DeepSee laser (Spectral Physics) tuned to 920 nm. Laser power was controlled by a pockels cell and electro-optic modulator (Conoptics). A 40× water immersion objective (Nikon CFI APO 40XW NIR) was attached to a Piezo controlled motor to move through the z plane. Laser power at the sample ranged from 10 to 18 mW, ramped linearly as depth increased. GCaMP fluorescence was detected with a gallium arsenide phosphite photomultiplier tube (Thorlabs PMT2000) after filtering through a 525/25 nm bandpass filter (Thorlabs MF525-39).
For cytosolic dFB GCaMP anesthesia experiments, recordings were captured at a resolution of 512 × 256 pixels across 18 z slices (6 µm each) at 2.79 Hz. For cytosolic EB GCaMP anesthesia experiments, recordings were captured at a resolution of 512 × 256 pixels in 11 5 µm z slices at 3.66 Hz. For optogenetic activation of CsChrimson (Klapoetke et al., 2014), a 617 nm LED was delivered via the objective at 10 mW (∼0.3 mA/mm2) in 30 Hz 5 ms pulses.
For nuclear-targeted GCaMP anesthesia experiments, recordings were captured at a resolution of 512 × 256 pixels in 13 6 µm z slices at 3.66 Hz. For nuclear-targeted GCaMP sleep experiments, recordings were captured at a resolution of 256 × 256 in 18 6 µm z slices at 2.77 Hz. For optogenetic activation of Chrimson88 (Strother et al., 2017), a 617 nm LED was delivered via the objective at 10 mW (∼0.3 mA/mm2) in 40 Hz 5 ms pulses.
Experimental setup and stimuli
Flies were tethered to a custom holder (Weir et al., 2016) using UV-cured dental cement and dissected as described by Tainton-Heap et al. (2021). Briefly, a window in the dorsal cuticle of the head was removed with forceps and a syringe tip to expose the brain. The imaging chamber above the brain was filled with room temperature extracellular fluid containing 103 mm NaCl, 10.5 mm trehalose, 10 mm glucose, 26 mm NaHCO3, 5 mm C6H15NO6S, 5 mm MgCl2 (hexa-hydrate), 2 mm sucrose, 3 mm KCl, 1.5 mm CaCl (dihydrate), and 1 mm NaH2PO4.
Tethered flies were placed on a 5-mm-diameter air-supported ball where they could freely walk. Fly behavior during experiments was recorded with an infrared camera (Firefly MV 0.3MP, Point Gray) at 30 fps under infrared LED illumination. Movement was calculated with a pixel subtraction method (Yap et al., 2017) using custom MATLAB code. Flies were continuously exposed to a 2 L/min air flow via a silicon tube placed next to the fly, which was also used for isoflurane delivery (Cohen et al., 2016).
A 50 ms 10 psi air puff was delivered to the fly via a 3 mm tube to determine binary responsiveness during wakefulness and anesthesia (Cohen et al., 2016). Responses were manually scored as responding or not responding depending on fly movement following the stimulus. The UV stimulus was delivered by a 405 nm LED (Thorlabs M405L4) filtered through a 405/2 nm bandpass filter (Thorlabs FB405-10), and terminated with a fiber optic cable placed 1.5 cm from the front of the fly. The LED was driven with custom hardware to deliver a 7 Hz sine wave for 2 s at 1 mW.
Experimental protocol
nls-GCaMP activity under anesthesia
Imaging occurred in three sequential recordings: spontaneous (10 min/2200 frames), UV stimulus response, and mechanical stimulus response (100 s/370 frames each) as shown in Figure 2A. During the stimulus periods, four stimuli were presented 20 s apart. Flies were then exposed to 2% isoflurane for 5 min (quantified as in Zalucki et al., 2015), and then tested for response to the air puff stimulus to confirm loss of responsiveness. The three recordings were then repeated while the fly remained exposed to isoflurane.
nls-GCaMP sleep
Spontaneous activity was recorded for 15 min (2500 frames) during a baseline period followed by dFB optogenetic activation for 15 min (2500 frames). A single UV stimulus was delivered halfway into these recording periods, and for this reason we chose to analyze the 7 min before this stimulus to avoid confounding factors. The raw data from these dFB experiments has been previously used in (Tainton-Heap et al., 2021), with different processing and analysis methods.
Cytosolic-GCaMP activity under anesthesia
Three 1 min recordings with a 15 s baseline, 30 s optogenetic stimulation, and 15 s recovery were delivered before anesthesia induction. Flies were then exposed to 2% isoflurane to render them unconscious (Zalucki et al., 2015), while behavior was recorded for loss of movement. Loss of responsiveness was confirmed following at least 5 min of isoflurane delivery by testing the grabbing reflex of the fly following retraction of the air-suspended ball. Flies were confirmed unresponsive by lack of a grabbing reflex.
Imaging analysis (nuclear GCaMP)
Images were motion-corrected in the x-y axis using the FIJI plugin align_slices_in_stack with subpixel registration using a SD projection as a reference slice (Tseng et al., 2011). Quality checks for movement artifacts were performed by taking SD projections across time. Recordings with distortions in cell shape in these images were rejected from further analysis. Recordings with z axis movement could not be corrected and were discarded. Cell segmentation was performed using watershed segmentation of a SD projection across the entire time series (Tainton-Heap et al., 2021). ROIs >20 pixels were rejected.
Neuron time-series data were extracted using the mean grayscale fluorescence of all pixels within each ROI. Data were smoothed across three frames with the Pandas function rolling. The time series for each neuron was z-scored, and periods of activity were defined in spontaneous recordings as any time points where fluorescence exceeded 3 SDs for at least three rolling frames.
Active neurons in spontaneous recordings were defined as those that had at least three time points of activity above the threshold (defined above) during the 10 min recording, or 2 in the 7 min dFB recording. The percent active metric was defined as the number of active neurons divided by the total number of ROIs per fly. The activity level metric was calculated from the mean number of frames each active neuron spent above the activity threshold. Overlap metrics for shared cell identity between conditions were calculated by dividing the number of common active neurons by the total number of active neurons in the first condition. For the self-similarity measure in Figure 4E, H, spontaneous recordings were divided into 10 1-min bins, and the overlap was measured for each pairwise combination of bins. Overlap by time (see Fig. 4D,G) was found by averaging bins together that were the same time distance apart. Total fly body movements during baseline were tallied for each fly using cumulative pixel change (Yap et al., 2017), and this total was correlated across flies with their average neural activity, using the Pearson correlation coefficient.
To determine neuron responses to the air puff and UV stimuli, we modified our detection criteria as responses could be time-locked to stimulus onset. Neurons were considered responsive if they had a z score fluorescence change >2 for at least 3 frames within 12 frames (2 s UV stimulus), or 6 frames (0.5 s air puff stimulus) of stimulus onset, and they responded to at least 2 of the 4 stimuli. The number of responses during baseline were compared with the number during isoflurane, and also to unstimulated control time points in between stimulus delivery with a two-way ANOVA and Sidak's multiple comparisons test. One fly was excluded from UV analysis because of high levels of background fluorescence artifact during stimulus onset, resulting in large numbers of false positive responses. Some flies had small regions of optic lobe neurons in the periphery of the recording, and these areas were masked before image segmentation.
To calculate mean degree, the z-scored times series fluorescence trace of every active neuron was correlated pairwise using the Pearson correlation coefficient. A temporally shuffled negative control was used to determine statistically significant correlations. Each activity trace was split into 10 equal segments and randomly shuffled to create a temporally disordered time series. Each disordered time series was then pairwise correlated and the correlation values recorded, for 1000 repetitions. The 99th percentile correlation value from these repetitions was used as the threshold value for significant correlations in the original correlation data, where correlated neurons were defined as connected. Mean degree was calculated by summing the number of connected partners each neuron had (degree), divided by the total number of active neurons. To determine mean degree following temporal shifts in the time-series data, each activity trace was shifted forward in time for 1-10 frames and then the above correlation procedure was repeated. This was performed for every neuron sequentially, such that the degree of each neuron now reflected its correlated partners after the time shift. Mean degree could then be calculated for all time-shifted degrees.
Mean distance of each neuron to its correlated partners was found by calculating the Euclidean distance to each partner and dividing by the total number of partners. For each fly, mean distance of all neurons was normalized to the maximum distance and compared with the normalized mean degree, so that variation between flies in their distributions of distances and degrees could be compared. Neurons with low degree were defined as those within 0.1 of the normalized degree. Neuron distances were defined as low <0.3 normalized distance, medium between 0.3 and 0.6, and far as >0.6.
Imaging analysis (cytosolic GCaMP)
ROIs around cell axons, dendrites, and 2 cell bodies were manually segmented from z projections of the recordings. The mean grayscale values for the ROIs were measured, and the 2 cell bodies were averaged together. dF/F for the optogenetic activation experiments was calculated using the mean of the first 20 frames (before red light delivery) as F. The three optogenetic activation recordings were averaged together. For anesthesia induction, fluorescence was normalized to the mean peak grayscale values for 5 s during the middle of optogenetic activation. Fluorescence time-series data for pre and post anesthesia were aligned for each fly using the time point from the behavioral data where the fly stopped actively moving. Mean fluorescence pre and post anesthesia was calculated using 10 s either side of this time point. In some flies, movement could be seen after losing control of the air-suspended ball and occasional small twitches, and these movements were manually zeroed.
Experimental design and statistical analyses
All statistical comparisons were conducted in GraphPad Prism. Comparisons using ANOVA were corrected for multiple comparisons using Sidak. Statistical tests used for each comparison and number of flies are described in the figure legends and in Results. Error bars and shading represent SEM. Statistical significance was set at p < 0.05.
Data and code availability
The data underpinning the publication are stored in UQRDM via UQ eSpace, the institutional data storage repository of the University of Queensland. The metadata in UQ eSpace is indexed by common search engines (e.g., Google) as well as by the national data discovery platform Research Data Australia and also by Data Citation Index.
Results
Isoflurane anesthesia does not activate sleep-promoting dFB neurons
Tethered, female flies were exposed to isoflurane gas while walking on an air-supported ball (Fig. 1A). Anesthesia induction was determined behaviorally when flies stopped moving (Fig. 1B) and stopped responding to a mechanical stimulus (see Materials and Methods), as determined by a pixel subtraction method from the filmed behavior (Yap et al., 2017). Only flies that became immobile and unresponsive were used for subsequent brain imaging experiments, and all flies recovered behaviorally from isoflurane anesthesia after ∼30 min, which aligns with our previous work (Troup et al., 2019). To examine the sleep-promoting R23E10-Gal4 circuit (Donlea et al., 2014), we imaged the fly's central brain through a window cut into the back of the head (Tainton-Heap et al., 2021). Since distinct compartments of the R23E10 neurons might display different levels of GCaMP activity, we imaged dendrites and cell bodies in addition the dFB structure itself (Fig. 1C). Before the anesthesia experiment, we activated R23E10 neurons with red light, to establish a baseline level of activity that anesthesia induction could be compared and normalized to. We confirmed that red light directed to the brain indeed activated these sleep-promoting neurons throughout, in their cell bodies and dendrites as well as the dFB (Fig. 1D, top). As expected, optogenetic activation of these neurons was associated with a transient loss of behavioral activity (Fig. 1D, bottom). We then investigated these neurons under isoflurane anesthesia, to see if they displayed any level of activity comparable with optogenetic sleep induction (Fig. 1E). We saw no change in GCaMP activity in any neural compartment, either immediately before or immediately after flies stopped moving as a result of anesthesia (Fig. 1E, top). When normalized to the peak activity that these neurons displayed with optogenetic activation, it is evident that anesthesia induction is not associated with any significant increase in activity in these purported sleep-promoting neurons (Fig. 1E, bottom). These results suggest that the sedative component of isoflurane's mechanism of action does not involve activation of these specific sleep-promoting neurons in Drosophila.
Isoflurane anesthesia does not activate sleep-promoting dFB neurons. A, Tethered flies walking on an air-supported ball were exposed to 2% isoflurane. B, Example movement trace from a single fly following exposure to isoflurane. Loss of movement (orange line) was used as the behavioral endpoint of anesthesia. C, Flies expressing GCaMP6s and CsChrimson in dFB neurons (R23E10-Gal4) were imaged using a 2-photon microscope, with ROIs selected for cell bodies, dendrites, and axons (dFB). Scale bar, 50 µm. D, Mean fluorescence change (dF/F) during optogenetic activation (red bar, top) and corresponding fly movement (bottom). n = 6 flies, 3 stimulations each. E, Mean fluorescence activity before (black) and following loss of movement (orange), normalized to peak optogenetic activation (D). n = 6 flies, two-way ANOVA with Sidak's multiple comparisons. F, Flies expressing GCaMP6s and CsChrimson in EB R3m neurons (R28D01-Gal4), with ROIs selected for cell dendrites, axons (EB), and cell bodies. Scale bar, 50 µm. G, Mean fluorescence change (dF/F) during optogenetic activation (red bar, top) and corresponding fly movement (bottom). n = 9 flies, 3 stimulations each. H, Mean fluorescence activity before (black) and following loss of movement (orange), normalized to peak optogenetic activation (G). n = 9 flies, two-way repeated-measures ANOVA with Sidak's multiple comparisons. Fly image in A from SciDraw. Fly head in C and F created with BioRender.com.
A confounding feature of optogenetically induced dFB sleep is that it shows little sleep inertia: flies awaken immediately after the red light is turned off (Fig. 1D), which we have suggested is indicative of a different kind of sleep quite unlike spontaneous “quiet” sleep (Tainton-Heap et al., 2021; Van De Poll and van Swinderen, 2021). Since isoflurane anesthesia, like sleep, is also characterized by inertia (Fig. 1B, right), we were interested to test other quiescence-inducing circuits that might more closely reproduce this aspect of spontaneous sleep. A number of other sleep-promoting circuits have been uncovered in Drosophila in addition to the dFB, for example neurons in the EB (Liu et al., 2016), including R3m ring neurons (Aleman et al., 2021). In a behavioral screen for sleep-promoting neurons (Kirszenblat et al., 2019), we uncovered an EB R3m ring neuron driver, R28D01-Gal4 (Fig. 1F), that when activated rendered flies completely immobile during as well as after red light exposure (Fig. 1G), thereby providing a comparison for dFB-induced sleep. As for the dFB neurons, activity in these EB neurons remained completely unchanged by isoflurane-induced anesthesia (Fig. 1H). This suggests that neither dFB nor R3m-EB neurons are activated by isoflurane anesthesia, although it remains possible that other sleep-promoting neurons in these central brain circuits are responsive to the sedative effects of isoflurane. However, the failure of finding any effect whatsoever of isoflurane on the activity levels of identified sleep-promoting neurons or associated quiescence-inducing neurons prompted us to use a broader imaging strategy to investigate the anesthetic's effect on neural activity and dynamics across the fly brain.
The fly brain remains active and responsive during isoflurane anesthesia
An alternative approach to investigating anesthetic effects within an anatomically defined circuit is to record anesthetic effects across large populations of distinct neurons within a defined brain volume. To image neural activity across the central fly brain, we expressed a nuclear-localized GCaMP in all neurons, as done previously for assessing spontaneous sleep stages in this same strain (Tainton-Heap et al., 2021). We followed a similar protocol as in Figure 1 for inducing isoflurane anesthesia in these flies, except that we also tested neural (and behavioral) responsiveness to visual and mechanical stimuli during baseline wakefulness and during subsequent isoflurane anesthesia (Fig. 2A). For each fly, we were able to detect hundreds of neurons that displayed a range of activity levels during baseline wakefulness (Fig. 2B). Active neurons (or ROIs) were identified if they satisfied key criteria relating to their signal-to-noise ratio, the duration of the calcium transients, and recurrence within a trace (Fig. 2C; see Materials and Methods). While some neurons correlated with locomotion, overall neural activity was not correlated with total fly movements across flies (r(12) = −0.35, p = 0.22; see Materials and Methods), and we observed ongoing patterns of neural activity in the waking fly brain even when animals were briefly quiescent (Fig. 2D). We found that isoflurane anesthesia had surprisingly little effect on this large-scale readout of neural activity (see Fig. 2E for traces of the exact same ROIs as for baseline in Fig. 2B), with most active neurons still active when flies were completely immobilized under isoflurane (see same single neuron example in Fig. 2F compared with Fig. 2C) and their average activity levels unchanged (Fig. 2G, left, t(13) = 0.1964, p = 0.847, paired t test). However, we did note that overall, the percentage of active neurons decreased moderately yet significantly, from ∼18% active during wakefulness to 15% active under isoflurane anesthesia (Fig. 2G, middle, t(13) = 3.256 p = 0.0063, paired t test).
The effect of isoflurane on whole-brain neural activity and responsiveness. A, Broad-scale volumetric imaging using a nuclear localized GCaMP was performed during 10 min of spontaneous activity, followed by four 2 s UV visual stimuli and four 0.5 s air puff stimuli. B, Example activity heatmap from 200 randomly selected neurons in a single fly. C, Example neuron from B, showing activity over the recording. Green bars represent periods of detected activity (see Materials and Methods). D, Corresponding behavioral activity for the same fly during the spontaneous period. E, Activity heatmap of the same neurons as in B, under isoflurane anesthesia. F, Activity of the same neuron shown in C, under anesthesia. G, The effect of isoflurane (orange) on activity level (number of frames spent active, left), activity number (% cells active, middle), and behavioral responsiveness as measure by responses to an air puff stimulus (right). n = 14 flies. **p < 0.01 (paired t tests). H, Activity heatmap of responses to the UV stimuli for the same fly shown in B-F. I, Average response profile (purple) for all 345 responding neurons across all 4 stimuli for the fly shown in H. Gray represents individual responses. J, Corresponding behavioral activity for the same fly during the stimulation period. In this example, the fly has not moved. K, Activity heatmap of responses to the UV for the same neurons in H, under isoflurane anesthesia. L, Average response profile for the same neurons, under isoflurane. M, Comparison of mean activity levels (left) and the number of UV-responsive neurons (right) during baseline (black) and isoflurane (orange). n = 13 flies. **p < 0.01; paired t test (activity levels) and two-way repeated-measures ANOVA with Sidak's multiple comparisons (neurons responding) (see Materials and Methods). N, Activity heatmap of responses to the air puff stimuli for the same fly shown in spontaneous and UV panels. O, Example average response profile (blue) for all 34 responding neurons across all 4 stimuli. Gray represents individual responses. P, Corresponding behavior during the 4 air puffs. Q, Activity heatmap of responses to the air puff for the same neurons in N, under isoflurane. R, Average response profile (blue) for the same neurons, under isoflurane. Gray represents individual responses. S, Comparison of mean activity levels (left) and the number of air puff-responsive neurons (right) during baseline (black) and isoflurane (orange). n = 14 flies. *p < 0.05, paired t test (activity levels) and two-way repeated-measures ANOVA with Sidak's multiple comparisons (neurons responding) (see Materials and Methods). Color intensity on heatmaps represents z score fluorescence values, where the darkest are values ≥3, and negative values have been zeroed (white) for display.
There is an expectation that the sleeping or anesthetized insect brain should display diminished responsiveness to external stimuli (Kaiser and Steiner-Kaiser, 1983; van Swinderen, 2006; Cohen et al., 2016; Tainton-Heap et al., 2021). We therefore next investigated whether the central brain neurons in our imaging preparation also become unresponsive to visual and mechanical stimuli, since flies have clearly been rendered behaviorally unresponsive by the anesthetic (Fig. 2G, right). To evoke visual responses in central fly brain neurons, we used a brief ultraviolet (UV) stimulus (Fig. 2A, middle) as we have shown previously that this stimulus evokes a robust response in the fly brain (Tainton-Heap et al., 2021). Accordingly, we found numerous UV-responsive cells in all tested flies (Fig. 2H; data are from the same fly as in Fig. 2B), that fit our criteria for identification (see Materials and Methods). We detected a significantly higher number of active neurons during stimulus onset compared with control time points in between stimuli, indicating the responses were specific to the UV stimulus (214.15 ± 25.78 vs 73.61 ± 4.41, t(12) = 16.43, p < 0.0001, Sidak multiple comparisons with two-way ANOVA). Averaged together, UV-responsive cells displayed a peak of calcium activity during stimulus onset (Fig. 2I). The UV stimulus did not typically evoke a behavioral response (Fig. 2J). When we anesthetized flies with isoflurane and examined the exact same UV-responsive neurons identified during wake, we observed a loss of responsiveness to UV: fewer of the neurons that were UV-responsive during wakefulness also responded to the UV stimulus under isoflurane anesthesia (Fig. 2K,L). This at first suggested that neural responsiveness to the UV stimulus was diminished by isoflurane, when we only looked at the same set of neurons. However, when we compared all of the neurons responsive to the UV stimulus in all flies, in isoflurane versus baseline wakefulness, we were surprised to find that the average number of UV-responsive neurons actually increased significantly under isoflurane anesthesia (Fig. 2M, right, t(12) = 4.128, p = 0.0042, Sidak multiple comparisons with two-way ANOVA), from an average of 214 ± 26 during wakefulness to 240 ± 33 during anesthesia. This suggests that, on average, a different set of neurons are responding to the UV stimulus under isoflurane anesthesia compared with wakefulness. Additionally, neurons responded as robustly to the UV stimulus under anesthesia as during wakefulness (Fig. 2M, left), when all UV-responsive neurons were considered rather than just the ones identified during wakefulness.
The definitive test for an anesthetized state is a complete loss of behavioral responsiveness. We therefore next sought to find some correspondence between loss of behavioral responsiveness to a mechanical stimulus (air puffs, Fig. 2A, right; Fig. 2G, right) and loss of responsiveness in neural activity. Our brain imaging method revealed a number of neurons that were recurrently responsive to air puff stimuli (see Materials and Methods), although these were notably fewer than the number that were responsive to the UV stimulus (Fig. 2N,O; data are from the same fly as in Fig. 2B,H). As expected, most awake flies responded behaviorally to the air puff, as evidenced by increased movement in the seconds following most of the stimuli (Fig. 2P; Fig. 2G, right, baseline). As for the UV stimulus, few of the neurons that responded to air puffs during wakefulness also responded under isoflurane anesthesia (Fig. 2Q,R), although these neurons remained active in general. However, unlike the UV stimulus, we observed a significant decrease in the average number of air puff-responsive neurons under isoflurane anesthesia (Fig. 2S, right, t(26) = 2.691, p = 0.0244, Sidak multiple comparisons with two-way ANOVA). As for UV, air puff-responsive neurons were as robust under anesthesia as during wakefulness (Fig. 2S, left). Together, these results indicate that, when flies are behaviorally anesthetized by isoflurane, neural activity in the central brain can remain more responsive to some stimuli (e.g., visual) than others (e.g., mechanical). We were nevertheless surprised that both sets of stimuli still evoked robust responses in the anesthetized fly brain (249 ± 33 neurons on average for UV and 25 ± 8 neurons for air puff), albeit probably in a different set of neurons through the time course of the experiment. We therefore next examined more closely how neural activity patterns might be changing, by tracking neural identities over time.
Responses to stimuli under isoflurane involve different neurons than wake
We were able to track the activity of the same ROIs throughout each experiment (see Materials and Methods), which allowed us to assess the overlap in neural identities between responsive neurons during waking (Fig. 3A, baseline) versus anesthetized conditions (Fig. 3B, isoflurane). As predicted from our preceding observations, UV-responsive ROIs during isoflurane anesthesia are mostly comprised of a different group of neurons compared with waking, with only a small subset (9.17 ± 1.99%) responding to the UV stimulus under both the waking and isoflurane conditions (Fig. 3C,D). The same analysis performed for the air puff stimulus revealed a similar trend (Fig. 3E–G), although the overlap between conditions was closer to zero (Fig. 3H). Not surprisingly, considering these are different sensory modalities, there was also very little overlap between UV and air puff-responsive neurons during waking (1.02 ± 0.27%).
Neural identity changes under isoflurane anesthesia. A, All UV-responsive cell locations for an example fly during baseline condition. Each dot represents the centroid of the responsive neuron ROI, with the z dimension collapsed. Grayscale background image is a time and z projection of the recording, with minimum and maximum gray values adjusted for clarity. Scale bar, 50 µm. B, All UV-responsive cell locations for the same fly under isoflurane. C, Baseline and isoflurane UV cell locations merged. Green dots represent cells responsive under both conditions. D, Average overlap between baseline and isoflurane conditions for UV-responsive cells for all flies, n = 13. E, All air puff-responsive cell locations for the same fly as A, during baseline condition. F, All air puff-responsive cell locations for the same fly under isoflurane. G, Baseline and isoflurane air puff cell locations merged. In this example, there are no cells in common. H, Average overlap between baseline and isoflurane conditions for air puff-responsive cells for all flies, n = 14. I, Active cell locations for baseline spontaneous condition for the same fly, in a single z slice. J, Active cell locations for isoflurane spontaneous condition, in the same slice. K, Baseline and isoflurane active cells merged. Green dots represent cells active during both conditions. L, Average overlap between baseline and isoflurane conditions for active cells (all slices) for all flies, n = 14.
The low overlap between stimulus-responsive neurons in our isoflurane data made us question the level of overlap for neurons that were spontaneously active (i.e., active, but not responding to either stimulus) under isoflurane versus baseline, since one explanation for the instability of neural responses to stimuli might be that isoflurane anesthesia engages a different group of active neurons than wakefulness. Indeed, we found that many different neurons were spontaneously active under isoflurane compared with wake (Fig. 3I–K); however, the overlap was higher (38.48 ± 2.06%) compared with the stimulation conditions (Fig. 3L). Since the level of overlap probably scales with the number of neurons involved (e.g., fewer neurons responded to the air puff, so the likelihood for overlap is smaller), it is possible that the instability of neural responses to stimuli reflects a more general aspect of neural dynamics across the fly brain. We therefore next examined neural turnover rates.
Neural turnover is less diverse under isoflurane anesthesia
Less than half of the neurons that were spontaneously active under isoflurane were also active during baseline wakefulness (Fig. 3K), indicating that mostly different neurons were active under both conditions. This raised the question of whether neural identities were constantly changing or whether an overlapping subset persisted through time. To address these questions, we first tracked changes in ROI identities from minute to minute for spontaneously active neurons during 10 min of wakefulness. The percent overlap in ROI identities could be compared with every preceding minute, as well as with every other minute bin within a baseline or isoflurane experiment (Fig. 4A,B). We discovered a surprisingly high level of instability or turnover from one min to the next for awake flies (∼35% overlap, or ∼65% new neurons every minute), and this was also the case for anesthetized flies (Fig. 4C). However, differences between the two conditions emerged when we compared more distal time bins. In awake flies, neural identities were more likely to keep changing through time, such that active ROIs were even less likely to be overlapping for time bins that were further apart (Fig. 4C, left; Fig. 4D, black line). In contrast, under isoflurane anesthesia, the level of neural identity overlap remained higher (∼30%) across time (Fig. 4C, right; Fig. 4D, orange line). This suggests that, although the fly brain remains mostly active under isoflurane anesthesia (Fig. 2E,G), and neural turnover is similar from minute to minute (Fig. 4D, temporal distance 1), a less diverse set of neurons is involved. Indeed, when we compared the overall percent overlap between active neurons in either condition, we found that significantly more neurons were the same identity under isoflurane anesthesia than during wakefulness, which we quantified as “self-similarity” (Fig. 4E, t(13) = 4.141, p = 0.0012, paired t test). To conclude, isoflurane anesthesia did not decrease minute-to-minute neural turnover rates, but rather decreased the diversity of the pool contributing to successive rounds of neural activity.
Neural identity changes through time. A, Active cell identity was compared between all pairwise 1 min periods within the spontaneous recording. B, Example active neurons (black dots) for one fly in a single z slice in 1 min bins. Green represents cells active between two bins. Scale bar, 50 µm. C, Pairwise comparison of cell identity overlap between minute bins during baseline and isoflurane anesthesia. n = 14 flies. D, Mean overlap of bins which share the same distance in time apart (1-9 min) for baseline (black) and isoflurane (orange). n = 14 flies. Shaded region represents SEM. E, Self-similarity as an average measure of pairwise overlap for baseline (black) and isoflurane (orange) spontaneous recording. n = 14 flies. **p < 0.01 (paired t test). F, Pairwise comparison of cell identity overlap between minute bins during baseline and induced sleep. n = 8 flies. G, Mean overlap of bins which share the same distance in time (1-6 min) apart for baseline (black) and sleep (red). n = 8 flies. Shaded region represents SEM. H, Self-similarity as an average measure of pairwise overlap for baseline (black) and sleep (red) spontaneous recording. n = 8 flies, paired t test. ns, not significant.
We have previously proposed that flies engage in different kinds of sleep, and that a form of active sleep is engaged when dFB neurons are optogenetically activated (Tainton-Heap et al., 2021). In the current study, we have shown that the dFB is not activated during isoflurane anesthesia (Fig. 1C), lending support to the view that dFB-induced sleep involves different processes than those subserved by general anesthesia or “quiet sleep” (Tainton-Heap et al., 2021). To further investigate this question, we therefore next examined neural turnover during optogenetically induced dFB sleep, to see if it behaved more like wakefulness or anesthesia. We activated the dFB using the same UAS/Gal4 drivers as before (Fig. 1C), in addition to pan-neuronal nuclear GCaMP expression to visualize neural turnover across the brain when flies were induced to sleep (see Materials and Methods) compared with baseline wakefulness (Fig. 4F). As before, we found that awake flies displayed a steady rate (>50%) of neural turnover from one min to the next, and this was also the case during dFB-induced sleep (Fig. 4F). However, unlike isoflurane anesthesia, there was no difference in the overlap through time, with new sets of neurons continuously at play during sleep as well as wake (Fig. 4G,H, t(7) = 0.3567, p = 0.7318, paired t test). This indicates that, for this specific neural readout during dFB-induced sleep, the brain is more similar to wakefulness than during isoflurane anesthesia, although under both conditions flies are inert and unresponsive to mechanical stimuli. To gain a better understanding of how neural dynamics might differ under isoflurane anesthesia, we next examined their correlation structure.
Decreased neural correlations across time under isoflurane anesthesia
Cortical neurons in mammals are often coactive as transient ensembles that fire together or in sequence (Carrillo-Reid and Yuste, 2020). One way to reveal ensemble dynamics in calcium imaging datasets is by investigating the correlation structure of neurons (Sporns, 2018; Yang et al., 2021) (Fig. 5A). Often, different calcium activity transients might seem correlated through time (Fig. 5B), and whether this is a significant correlation can be determined by permutation statistics (see Materials and Methods). The average number of other active ROIs that a neuron is correlated with determines its “degree,” and the average degree for all neurons in a population (or in our case, a fly brain) defines its “mean degree” (Fig. 5A, right). However, neurons can also be correlated through time, meaning that another neuron is often active immediately before or after. Such delayed correlation, which probably represents a functional property of any waking brain (Carrillo-Reid and Yuste, 2020), would be completely missed by simple mean degree calculations, as different pairs of neurons might be correlated at a different time (Fig. 5C). To address this temporal aspect of correlation structure, we performed mean degree calculations across 10 frame shifts (∼3 s) of our calcium imaging data across the fly brain (across each optical plane as well as among optical planes and corrected for imaging delays, see Materials and Methods). We examined shifts across 3 s because this encompasses the peak time of our GCaMP signal (Tainton-Heap et al., 2021). For our isoflurane anesthesia experiments, we found that simple mean degree (no frame shift) decreased significantly compared with baseline conditions (Fig. 5D, t(13) = 3.082, p = 0.0087, paired t test), meaning that calcium transients became less synchronized, on average. Repeating these analyses with temporally shifted baseline data significantly decreased this measure of neural connectivity, confirming the relevance of temporal structure to our data in waking flies (Fig. 5E, black line and corresponding dots, t(52) = 9.602, p ≤ 0.0001, Sidak multiple comparisons with two-way ANOVA). In contrast, temporally shifted isoflurane data did not decrease the mean degree any more than was already evident in nonshifted data (Fig. 5E, orange lines and dots, t(52) = 2.414, p = 0.0751, Sidak multiple comparisons with two-way ANOVA). This suggests that isoflurane exposure has a similar effect to shifting the temporal organization of waking data. Observing this significant difference in correlation structure between anesthesia and waking, we next questioned how dFB-induced sleep affected these whole-brain connectivity metrics compared with baseline wakefulness (Fig. 5F). Consistent with our previous study (Tainton-Heap et al., 2021), we detected no significant change in connectivity when flies were induced to sleep by optogenetic activation of the R23E10 neurons (Fig. 5F, t(7) = 0.01557, p = 0.9880, paired t test), and we observed a similarly significant decrease in connectivity compared with wakefulness when data were shifted in time (Fig. 5G, t(28) = 5.881, p < 0.0001, Sidak multiple comparisons with two-way ANOVA). This shows that dFB sleep preserves temporal structures that are similar to wakefulness.
Decreased neural connectivity under isoflurane anesthesia. A, Neurons were pairwise correlated to find connected partners. The average number of partners per neuron is the mean degree. B, Example neuron (black) with its two connected partners (top) and their corresponding activity traces (middle). Bottom, Two minutes of correlated activity between two of the neurons. Scale bar, 50 µm. C, Pairwise correlations were performed following the shifting of each neuron trace in time by 1-10 frames, which resulted in changes in connectivity. D, Mean degree changes from baseline (black) to isoflurane (orange). n = 14 flies. **p < 0.01 (paired t test). E, Effect of time shifts on mean degree during baseline (black) and isoflurane (orange). Left, The effect of each shift (frames). Right, Comparison of mean degree without shifts to a 10 frame shift. n = 14 flies. ****p < 0.0001 (two-way repeated-measures ANOVA with Sidak's multiple comparisons between each frameshift to 0). Shaded region represents SEM. F, Mean degree during baseline (black) and induced sleep (red). n = 8 flies, paired t test. G, Effect of time shifts on mean degree during baseline (black) and sleep (red). Left, The effect of each shift (frames). Right, Comparison of mean degree without shifts to a 10 frame shift. n = 8 flies. ****p < 0.0001 (two-way repeated-measures ANOVA with Sidak's multiple comparisons between each frameshift to 0).
Decreased neural correlations across space under isoflurane anesthesia
We next asked whether physical distance between cell soma determined which correlated pairs of neurons were most affected under isoflurane anesthesia, knowing the average distance for every neuron with respect to all of its significantly correlated partners (Fig. 6A; see Materials and Methods). Thus, some neurons are mostly correlated with nearby neurons (e.g., in the same brain hemisphere) (Fig. 6A, left), while other neurons reveal long distance correlations across the brain (Fig. 6A, right). With each active neuron thus providing two distinct metrics, correlation degree, and average distance to partners, we hypothesized three possible effects of isoflurane on the correlation structure: more distant neurons could become less correlated, closer neurons could become less correlated, or all neurons could become less correlated regardless of distance (Fig. 6B–D, left). Interestingly, we found individual examples of all three possibilities (Fig. 6B–D, right). In Fly 1, for example (Fig. 6B, right), most correlations with middle-distance neurons were abolished under isoflurane, with closer neurons dominating the correlation structure under anesthesia. However, the opposite was evident in another fly (Fig. 6C, right), while an overall decrease in mean degree across most distances was observed in a third fly (Fig. 6D, right). To simplify our analyses, we combined all distances into three groups (close: <0.2 of normalized distance; medium; and far: >0.6 normalized distance) and asked which correlation groups were most affected by isoflurane anesthesia. We uncovered a significant decrease in mean degree only in the middle-distance group (Fig. 6E, t(39) = 3.484, p = 0.0037, Sidak multiple comparisons with two-way ANOVA). Interestingly, ROIs that were close together were never well correlated under either condition (Fig. 6E, close; Fig. 6B–D, right). Indeed, proximity of cell soma in the insect brain has little relation with neuronal projections into the neuropil (Strausfeld, 1976). However, this also confirms the validity of our image segmentation protocol, as illegitimately split neurons would have revealed an erroneously high mean degree for adjoining ROIs.
Isoflurane anesthesia targets transient neural ensembles across the fly brain. A, Example neurons with their significantly correlated partners which can be close (left) or distant (right). Mean distance is calculated from the distances of all partners. Scale bar, 50 µm. B, Normalized degree and average corresponding distance to correlated partners for each significant neuron in an example fly (right) during baseline (black) and isoflurane (orange). This example is a case where neurons with higher degree tend to be closer together (left) under isoflurane. C, Normalized degree and distance of each neuron in a different fly (right), where the example shows neurons with higher degrees tend to be further apart under isoflurane (left). D, Normalized degree and distance of each neuron in an example fly where distance between connected neurons does not generally change under isoflurane; however, degree decreases. E, The effect of neuron distance (close, medium, and far; see B, right) on connectivity (degree) during baseline wakefulness (black) and during anesthesia (orange). n = 14 flies. **p < 0.01 (two-way repeated-measures ANOVA with Sidak's multiple comparisons). F, The effect of degree (low and high; see B, right) on neuron distance. n = 14 flies, two-way repeated-measures ANOVA with Sidak's multiple comparisons. G, Connectivity during baseline (black) and anesthesia (orange) in neurons active across both conditions (overlap), and neurons active only within a condition (nonoverlap). n = 14 flies. **p < 0.01 (two-way repeated-measures ANOVA with Sidak's multiple comparisons). H, The effect of neuron distance (close, medium, and far) on connectivity (degree) during baseline wakefulness (black) and during sleep (red). n = 8 flies, mixed-effects model with Sidak's multiple comparisons. I, The effect of degree (low and high) on neuron distance. n = 8 flies. *p < 0.05 (two-way repeated-measures ANOVA with Sidak's multiple comparisons). J, Connectivity during baseline (black) and sleep (red) in neurons active across both conditions (overlap), and neurons active only within a condition (nonoverlap). n = 8 flies, two-way repeated-measures ANOVA with Sidak's multiple comparisons.
To confirm our finding that physical distance was not an important factor in our analyses, we binned all neurons into two correlation groups, defined by low mean degree and high mean degree (all were significant correlations for both groups), and examined if isoflurane changed average distance among neurons within each ensemble. It did not (Fig. 6F, t(52) = 0.6, p = 0.7985 (low), t(52) = 0.1663, p = 0.9827 (high), Sidak multiple comparisons with two-way ANOVA). Finally, we examined whether isoflurane affected more stable versus more transient ensembles. Earlier, we showed that neuronal identities changed through time, although with greater overlap under isoflurane (Fig. 4D). We therefore questioned whether the loss of connectivity observed under isoflurane was a property of the more rapidly changing (nonoverlapping) groups of neurons, or the unchanging (overlapping) groups of neurons. We found that connectivity was significantly decreased in the nonoverlapping group and remained the same in the neurons that persisted in being active through time (Fig. 6G, t(26) = 0.7014, p = 0.7392 (overlap), t(26) = 3.782, p = 0.0016 (nonoverlap), two-way ANOVA, n = 14). This suggests that the more dynamic ensembles in the fly brain (meaning, the ones with regularly changing neural identities) are the ones losing connectivity under general anesthesia. These are also more connected ensembles, on average (Fig. 6G, nonoverlap). In stark contrast, induced dFB sleep had no effect on connectivity metrics partitioned by distance (Fig. 6H, t(20) = 0.4552, p = 0.9585 (close), t(20) = 1.966, p = 0.1783 (medium), t(20) = 1.360, p = 0.4664 (far), Sidak multiple comparisons with mixed-effects model), a small effect when partitioned by degree (Fig. 6I, t(14) = 3.283, p = 0.0109 (low), t(14) = 1.487, p = 0.2929 (high), Sidak multiple comparisons with two-way ANOVA) and no effect when partitioned by neural identity (Fig. 6J, t(14) = 0.3208, p = 0.939 (overlap), t(14) = 0.0091, p = 0.999 (nonoverlap), Sidak multiple comparisons with two-way ANOVA, n = 8).
Discussion
Loss of behavioral responsiveness is a defining feature of both sleep and general anesthesia. In humans and other mammals, it is known that these comparably inert states can nevertheless look quite different at the level of brain activity (Hennevin et al., 2007). It is now understood that many general anesthetics act as sedatives by potentiating inhibitory channels in the brain (Nelson et al., 2002; Franks, 2008), and with some drugs this can be transiently associated with oscillatory brain activity resembling SWS (Brown et al., 2011). This has led to the hypothesis that general anesthetics act on endogenous arousal pathways, thus exploiting a feature of the brain, namely, SWS. This view has been supported by studies showing that general anesthetics activate sleep-promoting circuits in the mammalian brain (Zecharia et al., 2009; Moore et al., 2012), raising the question whether these drugs act similarly in all animals that sleep, from flies to humans.
The discovery of sleep-promoting neurons in Drosophila (Donlea et al., 2011) provided an opportunity to examine how fly brain activity might change when flies are induced to sleep on command, much like acutely applying a powerful sedative. At the level of local field potential (LFP) recordings, studies show that GABA-acting sedatives such as Gaboxadol or anesthetics such as isoflurane decrease LFP activity in the fly brain (van Swinderen, 2006; Dissel et al., 2015; Cohen et al., 2016; Yap et al., 2017), and in this way resemble spontaneous sleep in this insect (Nitz et al., 2002). This is consistent with LFP recordings from other animals, which also show decreased LFP amplitude under anesthetic exposure (Zalucki and van Swinderen, 2016). Surprisingly, acute activation of the well-studied sleep-promoting dFB neurons in Drosophila does not decrease LFP activity; instead, both thermogenetic and optogenetic approaches reveal similar LFP activity as during wake (Yap et al., 2017; Troup et al., 2018). If isoflurane works like a sedative in the fly brain by acting on endogenous arousal circuits, as it appears to do in mammalian brains (Franks, 2008), then it might be expected that sleep-promoting neurons in Drosophila should be activated at least on general anesthesia induction. We did not see any evidence of increased R23E10 activity on isoflurane exposure. Further, unlike isoflurane anesthesia, dFB-induced sleep was never different from wakefulness when we tracked neural activity across the fly brain (although we did not examine stimulus-responsive cells in our current dFB analyses). Unchanged baseline dynamics supports our view that acute activation of the dFB promotes an “active” sleep stage (Tainton-Heap et al., 2021), distinct from the “quiet” electrophysiological brain states evident when flies are exposed to GABAergic sedatives or general anesthetics (van Swinderen, 2006; Dissel et al., 2015; Yap et al., 2017).
Although LFP recordings might suggest that the fly brain is less active under isoflurane anesthesia (van Swinderen, 2006; Cohen et al., 2016), our results here suggest that this view is not entirely accurate and probably reflects a limitation of electrophysiology. Measuring extracellular activity by electrophysiology requires groups of neurons to fire synchronously, to create electrical fields that can be detected with an electrode. Diminished LFP activity could therefore also indicate uncorrelated neural activity which does not summate sufficiently to create a detectable LFP. Using calcium imaging, we found that neural soma innervating the fly central brain remain mostly active under isoflurane anesthesia, although flies become inert and behaviorally unresponsive. A loss of correlated calcium activity in the central brain is nevertheless consistent with the decreased LFP activity observed in previous studies (van Swinderen, 2006; Cohen et al., 2016), and provides an explanation for why the fly brain appears to go “quiet” under general anesthesia. Similarly, spontaneous “deep” sleep in flies is also associated with less correlated calcium activity across the fly brain (Tainton-Heap et al., 2021) and decreased LFP activity (van Alphen et al., 2013; Yap et al., 2017), suggesting that both “quiet” sleep and general anesthesia in flies are defined in part by decreased neural coordination. Although the fly brain is smaller and organized differently than the mammalian brain, our observation is consistent with studies showing a loss of neural coordination (rather than activity) in anesthetized or sleeping rodents (Hennevin et al., 2007; Yang et al., 2021).
One of the more surprising results from our study is that even evoked neural activity in response to some visual stimuli does not decrease under isoflurane anesthesia. Rather, a largely nonoverlapping population of neurons responded to the UV stimulus. In contrast, we have previously found that dFB-induced sleep decreased responsiveness to a UV stimulus in the fly central brain (Tainton-Heap et al., 2021), although only one UV stimulus was used in that previous study whereas four in succession were probed in this study. For mechanical stimuli, although fewer neurons responded, there remained again a largely nonoverlapping population of evoked neural activity. Although it is possible that a different set of neurons responds to external stimuli in anesthetized flies, a more plausible explanation for these findings is that neural activity in the fly brain is in constant flux, with sparsely firing (or responding) neurons being recruited into dynamic ensembles characterized by relatively short temporal durations (Turner et al., 2008; Lin et al., 2014). Indeed, when we examined only spontaneous neural activity, it was clear that over half of the active neurons are changing from minute to minute. A similar finding also using whole-brain nlsGCaMP imaging was reported in zebrafish, where turnover of spontaneous activity was found to represent preferred network states tuned to behaviorally relevant features (Romano et al., 2015). In the mammalian brain, visual stimuli also appear to recruit shifting neural ensembles (Miller et al., 2014), with less than half remaining active across two successive imaging sessions (Perez-Ortega et al., 2021). This is probably a feature of any neural network: even the relatively simple pyloric system of the crustacean stomatogastric ganglion shows evidence of neurons being continuously recruited into different ensembles to achieve continuity of function (Prinz et al., 2004). An alternative explanation for the dynamics we uncovered in the fly brain might be that ensemble-like behavior simply reflects ongoing undetected microbehaviors, which might also explain why we observed no correlation between neural activity and total movement during waking, across flies. While we did find neurons that were correlated to gross movement (e.g., locomotion) during baseline recordings, these were typically a minority (<10%) of the total active neurons per fly. Other recent pan-neuronal imaging studies in Drosophila have found significant correlation between brain activity and locomotion (Aimon et al., 2022; Brezovec et al., 2022), although those studies examined neuron/movement correlations on a finer scale, and used cytosolic GCaMP while ours was restricted to cell bodies. Since fly behaviors are unlikely to persist under isoflurane anesthesia (van Swinderen, 2006) while neural activity clearly does, our findings highlight the importance of tracking networks rather than single neurons when assessing changes in brain activity associated with altered behavioral states. Loss of responsiveness at the level of single neurons may be misleading and indicate instead that another network has been engaged.
While similar minute-to-minute changes in neural activity were evident during both waking and anesthesia, the overall contributing population was significantly less diverse under isoflurane. Similarly, there is evidence that isoflurane decreases the repertoire of microstates in the mammalian brain (Wenzel et al., 2019). Interestingly, the loss of connectivity (or correlation) that best characterizes the neural fragmentation we observed under isoflurane anesthesia was only evident in the more transient (nonoverlapping) and more highly connected neurons, rather than in the neurons that were persistently active throughout an experiment. This suggests that isoflurane specifically fragments the larger, more transient neural ensembles, rather than the smaller, longer-lasting ensembles. In mice, only a minority of cortical neurons form stable ensembles across time, with most belonging to more transiently correlated groups (Perez-Ortega et al., 2021). Our results in Drosophila hint at fundamentally similar brain dynamics. Whether these isoflurane-sensitive transient ensembles represent potential correlates of perception in flies could be tested with more sophisticated visual experiments probing for attention-like responses in the brain (Grabowska et al., 2020), in and out of anesthesia or sleep.
The effect of isoflurane on Drosophila neurons that we have uncovered is surprisingly subtle: neurons are not silenced, rather a transient subpopulation become less correlated and less diverse. This conclusion could, of course, reflect our superficial coverage: since we used nuclear-localized (nls) GCaMP, only the cell bodies were tracked in our experiments, and it is thought that in insects these may not always accurately reflect activity in other cellular compartments (Tuthill, 2009), such as the dendrites (but for a comparison between cytoplasmic and nuclear localized GCaMP, see Jung et al., 2020). While nlsGCaMP has been informative for studying ensemble-like behavior in response to visual stimuli in other animals such as zebrafish (Thompson and Scott, 2016), it remains possible that calcium imaging in the nuclei of could also reflect other biological processes (Shemesh et al., 2020). Nevertheless, our calcium imaging findings are consistent with measures of LFP activity that uncovered altered neural dynamics under isoflurane anesthesia in flies, such as impaired feedback mechanisms (Cohen et al., 2018) or the collapse of integrated information structures (Leung et al., 2021). Together, these studies suggest that general anesthetics cause a loss of responsiveness by impairing a network-level property of brain function, rather than neural activity per se. This agrees with findings in other animal models as well as humans (Hudetz, 2012; Awal et al., 2018; Yang et al., 2021), suggesting a common mechanism.
How might general anesthetics produce a loss of correlation among neurons, whether these are in nematode worms (Awal et al., 2018), flies (this study), mice (Yang et al., 2021), or humans (Hudetz, 2012)? While the activation of endogenous sleep pathways in the brain has been proposed as a likely first stage in general anesthesia induction (Franks, 2008), this does not seem a sufficient explanation for anesthesia in animals that lack sleep-promoting circuits (Kirszenblat and van Swinderen, 2015; Zalucki and van Swinderen, 2016), nor does it explain why loss of behavioral responsiveness is more complete under surgical doses of general anesthetics. While sleep induction is certainly a component of general anesthesia, the loss of integrated information is increasingly understood to better define the anesthetized state in humans (Mashour and Hudetz, 2017), and we propose that the same is true for any brain. While anesthetic effects on sleep pathways largely involve postsynaptic mechanisms centered on GABAA receptors (Franks, 2008), it is possible that the altered neural dynamics that characterize the anesthetized state also have a presynaptic explanation (van Swinderen and Kottler, 2014; Hemmings et al., 2019). General anesthetics, such as isoflurane and propofol, impair synaptic release in all animals where this has been investigated (Herring et al., 2009, 2011; Baumgart et al., 2015; Bademosi et al., 2018; Karunanithi et al., 2020; Hines and van Swinderen, 2021). It therefore seems plausible that impaired neurotransmitter release might alter coordination across the brain, leading to a fragmentation of the transient neural ensembles required for normal waking behaviors.
Footnotes
This work was supported by National Health and Medical Research Council GNT1164879 and National Institutes of Health R01 Grant NS076980.
The authors declare no competing financial interests.
- Correspondence should be addressed to Bruno van Swinderen at b.vanswinderen{at}uq.edu.au
