Feedback inhibition by a descending GABAergic neuron regulates timing of escape behavior in Drosophila larvae

In a previous study, we showed that activation of all Basin neurons (Basin-1, -2, -3, and -4) induced rolling followed by fast crawling (Figure 1A–D; Ohyama et al., 2015). Here, we first examined whether optogenetic activation of all four Basins expressing the red-shifted opsin CsChrimson (using Basin-1–4 Gal4, i.e., R72F11-Gal4) could elicit the same behavior. Upon activation of all Basins, we observed rolling mostly within the first 5 s, followed by crawling (Figure 1B (top panel), Figure 1C, D). The crawling speed during activation of all Basins following rolling was ~1.5 times that of the crawling speed at baseline (Figure 1D; Ohyama et al., 2015).

Figure 1 with 2 supplements see all

Download asset Open asset

Figure 1—source data 1

Excel sheet containing statistics data from behavior tracking depicted in Figure 1E-I.

https://cdn.elifesciences.org/articles/93978/elife-93978-fig1-data1-v1.zip

Download elife-93978-fig1-data1-v1.zip

Figure 1—source data 2

Excel sheet containing raw data from behavior tracking depicted in Figure 1E-I.

https://cdn.elifesciences.org/articles/93978/elife-93978-fig1-data2-v1.zip

Download elife-93978-fig1-data2-v1.zip

To identify the neurons responsible for escape behavior (rolling and/or fast crawling), we conducted a behavioral screening of ~250 split Gal4 lines that were labeled in the central nervous system (CNS) when co-activated with all Basins. With respect to rolling, we found that activation of the split-Gal4 line, SS04185 (i.e., w¹¹¹⁸; R54B01-Gal4^AD; R46E07-Gal4^DBD), significantly reduced the probability of rolling when compared to activating only the Basins (Figure 1B, C, E, Videos 1 and 2) without affecting the crawling speed during stimulation (Figure 1D).

Video 1

Download asset

Video 2

Download asset

The likelihood of rolling upon co-activation of SS04185 neurons and Basins might decrease because activation of SS04185 neurons trigger other actions, such as crawling, head casting, hunching, or stopping, and not because they solely inhibit rolling evoked by Basins. To investigate this possibility, we examined the effect of activating only SS04185 and found that this did not induce any extra actions such as crawling, turning, hunching, or stopping (Figure 1—figure supplement 1A–D). These data suggest that co-activation of Basins and SS04185 neurons reduces rolling because SS04185 activation inhibits the Basin circuit.

Next, we explored how the quality of rolling changed during co-activation of SS04185 and Basin neurons. First, we examined the amount of time animals spent rolling during Basin activation. The average time spent rolling (percentage of the 30 s stimulation period) was 23.9% (7.2 s out of 30 s) following activation of Basins alone, whereas it was only 5.9% following co-activation of Basins and SS04185 (1.8 s out of 30 s) (Figure 1—figure supplement 1E). Additionally, the duration of each rolling bout was significantly shorter when SS04185 neurons were co-activated with Basins (Mann–Whitney U test, p < 0.001; Figure 1F).

The duration of a rolling bout could decrease because of changes in the latency to initiate rolling, latency to terminate rolling, or both. To investigate how SS04185 activation affects these temporal parameters of rolling, we analyzed the latencies for the initiation and termination of the first rolling bout. Compared to activating Basins alone, co-activating the Basins and SS04185-expressing neurons only marginally increased latency to onset of the first rolling bout (Figure 1G), whereas it markedly reduced the latency for the termination of rolling (Mann–Whitney U test, p < 0.001; Figure 1H). These data strongly suggest that SS04185-expressing neurons are involved in terminating rolling.

If the rolling module inhibits crawling, then premature termination of rolling might allow crawling to commence sooner than normal. Co-activation of SS04185 and Basins resulted in the initiation of the first crawling bout occurring earlier than when only Basins were activated (Mann–Whitney U test, p < 0.001; Figure 1I, Figure 1—figure supplement 1F). The time from the end of rolling to the start of crawling remained similar between the groups in which the Basins were activated alone and in which the Basins and SS04185 were co-activated (Figure 1—figure supplement 1G). This is consistent with the higher probability of crawling during activation of SS04185 and Basin neurons (Figure 1—figure supplement 1H). Lastly, activation of SS04185 neurons in conjunction with Basins did not change the crawling speed compared to activation of Basins alone (Figure 1—figure supplement 1I). These results collectively indicate that SS04185 activation terminates rolling and facilitates the shift to fast crawling.

To identify the neurons that express SS04185 upon CsChrimson activation, we examined the localization of SS04185-labeled neurons. We found that SS04185 split-Gal4 strongly labeled a pair of descending neurons located within the SEZ and mushroom body (MB) neurons within the brain (Figure 2A). To pinpoint which of these neurons are involved in reducing the probability of rolling (Figure 1B, C, E), we varied the level of SS04185 expression among the pair of SS04185-expressing descending neurons (SS04185-DN) and the SS04185-expressing MB (SS04185-MB) neurons (jointly with the Basins as in Figure 1). These manipulations allowed us to assess the resultant behavioral outcomes.

Figure 2 with 1 supplement see all

Download asset Open asset

Figure 2—source data 1

Excel sheet containing raw data from behavior tracking depicted in Figure 2D, E, G and H.

https://cdn.elifesciences.org/articles/93978/elife-93978-fig2-data1-v1.zip

Download elife-93978-fig2-data1-v1.zip

Figure 2—source data 2

Excel sheet containing statistic data from behavior tracking depicted in Figure 2C-H.

https://cdn.elifesciences.org/articles/93978/elife-93978-fig2-data2-v1.zip

Download elife-93978-fig2-data2-v1.zip

If SS04185-MB neurons are involved in the modulation of rolling, then reducing SS04185-MB expression should reduce the extent to which activation of both SS04185-DN neurons and SS04185-MB neurons decreases the probability of rolling. To test this conjecture, we expressed Killer Zipper (KZip⁺), which interferes with the binding of Gal4^AD and Gal4^DBD in SS04185-MB neurons with MB LexA line (R13F02-LexA), consequently leading to a significant reduction in CsChrimson expression in SS04185-MB neurons (Figure 2B, Figure 2—figure supplement 1A; Dolan et al., 2017; Vogt et al., 2016). When compared to KZip⁺ controls, which do not express SS04185 (Figure 2C, black bars), however, activation of SS04185 neurons with reduced SS04185-MB expression (Figure 2C, red bars on the right; Figure 2—figure supplement 1B) still reduced rolling probability (as well as the total duration of rolling [Figure 2—figure supplement 1C]) to a level no different from that of KZip⁻ controls expressing SS04185 fully in both SS04185-MB and SS04185-DN neurons (Figure 2C, dark red bars in the middle). Additionally, co-activation of MB Gal4 lines (MB247-Gal4) with Basins (without activation of SS04185-DN neurons) did not reduce the probability of rolling (Figure 2—figure supplement 1D, E). These data indicate that SS04185-DN neurons inhibit rolling.

To further test the role of SS04185-DN neurons, we investigated whether these neurons were involved in reducing the duration of each rolling bout (Figure 1A, D, F). As a result, knockdown of SS04185-MB neurons did not increase the duration of rolling bouts (Figure 2D). Furthermore, the earlier onset of crawling triggered by the activation of SS04185 neurons remained the same with knockdown of SS04185-MB neurons (Figure 2E). Collectively, these results strongly suggest that the behavioral effects on both rolling and crawling, as illustrated in Figure 1, are primarily mediated by SS04185-DN neurons.

To further ascertain the role of SS04185-DN neurons in the regulation of rolling, we employed the heat shock FlpOut mosaic expression approach. This technique allowed for controlled and sporadic expression of CsChrimson in SS04185 neurons thorough random induction of Flippase by manipulating the timing and duration of heat shock (Golic and Lindquist, 1989; Nern et al., 2015). We compared larvae subjected to activation of both SS04185-MB and SS04185-DN neurons (red, Figure 2—figure supplement 1F) with those subjected only to activation of SS04185-MB neurons (black, Figure 2—figure supplement 1G), to assess the degree to which the former showed behavioral effects. Remarkably, activation of both SS04185-MB and SS04185-DN neurons resulted in a reduction in both the probability and duration of rolling when compared to activation of SS04185-MB neurons alone (Figure 2F, G, Figure 2—figure supplement 1H, I). Furthermore, activation of both SS04185-MB and SS04185-DN neurons reduced the latency to the end of the first rolling bout and the initiation of the first crawling bout (Figure 2H, Figure 2—figure supplement 1J). These findings provide compelling evidence that SS04185-DN neurons, but not SS04185-MB neurons, play an important role in the termination of rolling. Collectively, the results suggest that a single pair of descending neurons in SS04185 is important for termination of rolling during the activation of Basins.

In a previous electron microscopy (EM) connectome study, we identified a set of neurons designated as SeIN128, whose-cell bodies in the SEZ send axonal projections throughout the thoracic and abdominal segments (Figure 3A; Ohyama et al., 2015). Our immunostaining data also showed that the cell bodies of SS04185-DN neurons are located in the SEZ, with axons bilaterally innervating the medial regions of the ventral nerve cord (VNC) from the thoracic to abdominal segments A8/9 (Figure 2A), suggesting that SS04185-DN and SeIN128 neurons are one and the same.

Figure 3

Download asset Open asset

To verify this possibility, we examined the detailed anatomy of SS04185-DN neurons by immunostaining them with several markers and compared our immunostaining images with the corresponding images obtained via EM reconstruction of the entire CNS of a first instar Drosophila larva (Ohyama et al., 2015; Winding et al., 2023). We confirmed that the projections of SeIN128 neurons are distributed within the ventromedial neural tract (one of the six major neural tracts) in Drosophila larvae (Figure 3A–C) in EM reconstruction data. We also confirmed that the cell bodies of SS04185-DN neurons were again located in the SEZ region, where the most anterior of the three neuropils in the thoracic region was marked by N-cadherin (Figure 3D). Viewed from the side (i.e., in the longitudinal or sagittal plane), both the cell bodies and axonal arbor were located ventrally (Figure 3D, far right). Immunostaining with Fasciclin2 (Fas2), which labels various neural tracts in the VNC (Grenningloh et al., 1991; Santos et al., 2007), showed colocalization of the axonal projections of SS04185-DN neurons and the Fas2-labeled ventromedial tract (Figure 3C, E). The similarity of the locations of their cell bodies and the distributions of their axonal processes suggests the identity of the SS04185-DN and SeIN128 neurons.

A previous EM study showed that SeIN128 neurons were located downstream of Basin neurons (Ohyama et al., 2015). To further confirm the identity of SS04185-DN and SeIN128 neurons, we compared the distributions of the axonal projections of SS04185-DN neurons in relation to those of several key neurons within the rolling circuit: the Basins, A00c neurons (a group of ascending neurons downstream of the Basins, and which facilitate rolling), and mdIV neurons (nociceptive sensory neurons upstream of the Basins). Immunostaining revealed that Basin projections colocalize with those of SS04185-DN neurons in both the horizontal and transverse planes (Figure 3F, top and lower panels, respectively), with the horizontal view showing that SS04185-DN projections are distributed slightly medial to those of Basins within the ventromedial tract (Figure 3F, top panels), which resembles their colocalization pattern reported in EM (Figure 3B, C, G). Similarly, we compared the distributions of SS04185-DN projections with those of A00c or mdIV projections. We found that the projections of A00c colocalize with those of SS04185-DN in a similar fashion along the rostrocaudal axis within the ventromedial tract (Figure 3H,I), with A00c projections distributed more medially than SS04185-DN projections, consistent with the distribution patterns of SeIN128 projections and A00c projections in the EM reconstruction dataset (Figure 3B, H, I). In contrast, the distributions of mdIV projections did not colocalize with those of SS04185-DN projections, as the mdIV projections were displaced more laterally relative to the SS04185-DN projections in the horizontal and transverse planes (Figure 3J, top and lower panels, respectively), consistent with the distribution patterns of SeIN128 and mdIV projections in the EM reconstruction dataset (Figure 3K). In the transverse plane, the projections of SS04185-DN neurons were also distributed dorsomedial to those of mdIV (Figure 3J, lower panel), consistent with the corresponding distribution patterns in the EM reconstruction dataset (Figure 3B, C, K).

We conclude that the morphological findings for SS04185-DN neurons, together with data on the distribution of their axonal projections in relation to that of Basin, A00c, and mdIV neurons, strongly suggest the identity of SS04185-DN and SeIN128 neurons.

A previous study that reconstructed larval neurons involved in the rolling circuit showed that Basin-2 and A00c neurons (in the VNC) make excitatory synaptic contacts onto SeIN128 neurons (in the CNS), which in turn make reciprocal inhibitory synaptic contacts onto Basin-2 and A00c neurons (Figure 4A, Figure 4—figure supplement 1A; Ohyama et al., 2015). These data suggest that SeIN128 neurons are directly activated by Basin-2 and A00c (which also receives inputs from Basin-1, Basin-2, and Basin-4).

Figure 4 with 2 supplements see all

Download asset Open asset

Figure 4—source data 1

Excel sheet containing raw data depicted in Figure 4C.

https://cdn.elifesciences.org/articles/93978/elife-93978-fig4-data1-v1.zip

Download elife-93978-fig4-data1-v1.zip

To assess the functional significance of these synaptic connections between SeIN128 neurons and Basins or A00c, we activated either Basins or A00c neurons and examined the resultant green fluorescent protein (GFP) calmodulin (CaM) protein (GCaMP) signaling in SeIN128 neurons. Specifically, after expressing CsChrimson in Basins and A00c neurons and GCaMP in SeIN128 neurons, we used a two-photon microscope (920 nm laser) and monitored GCaMP signaling in SeIN128 neurons during illumination of a specimen with a 620-nm light-emitting diode (LED) for 1 s (0.04–1.4 µW/mm²), which activated either Basins or A00c neurons. GCaMP signals in SeIN128 neurons increased in an intensity-dependent manner when either Basins and A00c were activated (Figure 4B, C) but not when larvae were not fed retinal (Figure 4—figure supplement 1C, D). Peak activity occurred at around 3 s after the onset of LED stimulation, which was similar to the results when Basins or A00c neurons were stimulated (Figure 4B, C). Finally, both Basin and A00c stimulation resulted in linear dose-dependent increases in SeIN128 firing (Figure 4—figure supplement 1B). These results are consistent with the notion that SeIN128 neurons are downstream of Basins and A00c neurons.

To compare the neural responses between Basins and SeIN128 or A00c neurons, we recorded neural activity in A00c neurons with GCaMP while stimulating Basin neurons in the same experimental setting. Although A00c neurons displayed a similar dose-dependent increase in peak axonal firing as the intensity of optogenetic stimulation of Basin neurons increased, unlike SeIN128 neurons they showed no delay in peak firing activity (Figure 4D, E, Figure 4—figure supplement 1B), suggesting that A00c and SeIN128 neurons function differently in the rolling circuit.

We then investigated the anatomical locations of the synaptic outputs and inputs of SeIN128 neurons, and found that, whereas their outgoing projections primarily make synaptic contacts along the anterior–posterior nerve axis, the inputs coming from other neurons are mainly located in the SEZ (Figure 3A). On the other hand, SeIN128 neurons make axo-axonal contacts onto Basin-2 neurons (Figure 4—figure supplement 2A–G) as well as A00c neurons: that is, their axons make synaptic contacts with the dorsal and medial processes of Basin-2, which correspond to their axonal compartments (Figure 4—figure supplement 2E–G). These data suggest that the delay of SeIN128 activity may be caused by multi-synaptic connections involving the SEZ or a feedback loop involving axo-axonal connections between SeIN128 and Basin-2 or A00c.

The results thus far indicate that, activation of SeIN128 neurons inhibits rolling (Figure 1A–C); SeIN128 neurons receive functional inputs from Basin-2 and A00c (Figure 4A–C); and SeIN128 neurons make anatomical connections onto Basin-2 and A00c (Figure 4A). These findings suggest that SeIN128 neurons might be inhibitory. To test this possibility, we performed immunostaining experiments and found that SeIN128 neurons colocalized with glutamic acid decarboxylase (Gad)-positive neurons but not with acetylcholine- or glutamate-positive neurons, suggesting that SeIN128 neurons are GABAergic inhibitory neurons (Figure 5A, Figure 5—figure supplement 1A,B).

Figure 5 with 1 supplement see all

Download asset Open asset

Figure 5—source data 1

Excel sheet containing raw data depicted in Figure 5D.

https://cdn.elifesciences.org/articles/93978/elife-93978-fig5-data1-v1.zip

Download elife-93978-fig5-data1-v1.zip

Figure 5—source data 2

Excel sheet containing statistics data depicted in Figure 5C and D.

https://cdn.elifesciences.org/articles/93978/elife-93978-fig5-data2-v1.zip

Download elife-93978-fig5-data2-v1.zip

We reasoned that if Gamma-aminobutyric acid (GABA) in SeIN128 neurons is necessary for inhibiting rolling, then selectively knocking down GABA secretion in SeIN128 neurons should enhance rolling. When we expressed RNA interference (RNAi) HMS02355 in SeIN128 neurons to knock down vesicular GABA transporter (VGAT) expression and suppress the release of GABA, the population-level rolling probability increased from 23.6% to 45.2% (Figure 5B, C; Kallman et al., 2015; Zhao et al., 2019). We confirmed the effect of HMS02355 by immunostaining: pan-neural HMS02355 expression decreased GABA and VGAT expression in the neuropil (Figure 5—figure supplement 1C, D). The control group (only Basins expressing CsChrimson with VGAT RNAi HMS02355 but without SS04185) showed a lower probability of rolling (23.6%) compared to similar genotypes without VGAT RNAi HMS02355 (Figure 2C, F). This indicates that VGAT RNAi HMS02355 background reduces the probability of rolling. Furthermore, the duration of each bout of rolling increased from 0.8 to 1.4 s (Figure 5D). These data support the idea that SeIN128 neurons inhibit rolling via GABAergic transmission.

To further test whether the release of GABA upon activating SeIN128 neurons is necessary for inhibiting rolling, we expressed tetanus toxin (TNT) in SeIN128 neurons to block synaptic transmission. Silencing SeIN128 neurons via TNT while triggering rolling by optogenetically activating Basin neurons via R72F11-LexA>LexAop-CsChrimson significantly increased the probability of rolling compared to controls (Figure 6A, B). Silencing SeIN128 neurons via TNT extended the duration of each rolling bout, as well as the total rolling duration, in each larva (Figure 6C, D). We also examined the rolling-escape crawling sequence upon silencing SeIN128 neurons, and found that the time to offset of rolling and the time onset of crawling were both delayed relative to controls (Figure 6E, F).

Figure 6 with 1 supplement see all

Download asset Open asset

Figure 6—source data 1

Excel sheet containing raw data from behavior tracking depicted in Figure 6C-F.

https://cdn.elifesciences.org/articles/93978/elife-93978-fig6-data1-v1.zip

Download elife-93978-fig6-data1-v1.zip

Figure 6—source data 2

Excel sheet containing statistics data from behavior tracking depicted in Figure 6C-F.

https://cdn.elifesciences.org/articles/93978/elife-93978-fig6-data2-v1.zip

Download elife-93978-fig6-data2-v1.zip

Given that TNT is expressed constitutively during development, long-term compensatory changes in the nervous system could have contributed to alterations in the parameters of rolling and crawling. To test whether similar results could be replicated with the use of a temporally specific intervention, we expressed shibire^ts1 (shi^ts1) in SeIN128 neurons to block synaptic transmission at temperatures above 30°C (van de Goor et al., 1995; Kitamoto, 2001). Silencing SeIN128 neurons via shibire^ts1 increased the probability of rolling from 60.4% to 79.7% (Figure 6—figure supplement 1A, B). The total duration of rolling per animal during stimulation increased from 10 to 12 s (Figure 6—figure supplement 1C). Although the duration of each rolling bout, the time to onset of the first rolling bout, and time to onset of the first crawling bout did not differ from those of controls (Figure 6—figure supplement 1D, E, G), the time to offset of the first rolling bout was delayed relative to controls (p = 0.013 for Figure 6—figure supplement 1F). Together with the results showing that activation of SeIN128 neurons inhibits rolling, these findings suggest that the activity of SeIN128 neurons is important in controlling the duration of rolling and the shift to crawling.

Given that Basins receive axo-axonal inputs from SeIN128 neurons and GABA signaling in SeIN128 neurons inhibits rolling, we next used RNAi to test whether Basins receive GABAergic signals from SeIN128. We hypothesized that knockdown of GABA receptors in Basin neurons would increase the probability and duration of rolling at the population level. To knock down ionotropic GABA-A receptors (GABA-A-R) and G-protein-coupled GABA-B receptors (GABA-B-R1 and GABA-B-R2), we tested Basin neurons with GABA-A-R, GABA-B-R1, and GABA-B-R2 RNAi lines (i.e., HMC03643 for GABA-A-R, HMC03388 for GABA-B-R1¹, JF02989 for GABA-B-R1², and HMC02975 for GABA-B-R2, respectively). For all RNAi lines, the rolling probability at the population level increased from 80% to 90% or even higher (Figure 7A), while the total rolling duration at the individual level increased for each larva throughout the stimulation window (Figure 7—figure supplement 1A). All GABA receptor knockdown groups showed significant increases in rolling duration across multiple bouts (Figure 7B); all groups except for GABA-B-R1¹ showed a reduced time to onset of the first rolling bout (Figure 7—figure supplement 1B); and only the GABA-B-R2 and GABA-A-R groups showed a delayed offset of the first rolling bout (Figure 7—figure supplement 1C). None of the groups differed from controls in the time to onset of the first crawling bout (Figure 7—figure supplement 1D). The greatest increase in the probability and duration of rolling was seen during knockdown of ionotropic GABA-A-R (Rdl), suggesting that Rdl contributes most to the inhibition of Basin neurons (Figure 7A, B).

Figure 7 with 1 supplement see all

Download asset Open asset

Figure 7—source data 1

Excel sheet containing raw data from behavior tracking depicted in Figure 7B and C.

https://cdn.elifesciences.org/articles/93978/elife-93978-fig7-data1-v1.zip

Download elife-93978-fig7-data1-v1.zip

Figure 7—source data 2

Excel sheet containing statistics data from behavior tracking depicted in Figure 7B and C.

https://cdn.elifesciences.org/articles/93978/elife-93978-fig7-data2-v1.zip

Download elife-93978-fig7-data2-v1.zip

To investigate whether SeIN128 neurons actually inhibit Basins, we recorded the activity of all Basins during activation of SeIN128 neurons. We compared GCaMP signaling in the Basins when they were co-activated with SeIN128 neurons (experimental treatment) or when they were activated alone (control treatment), with the intensity of optogenetic stimulation varied from 0.04 to 1.4 µW/mm². We found that Basins in the experimental group showed reductions in GCaMP signaling by 11–36% compared to those in the control group (Figure 7C, Figure 7—figure supplement 1E, F). The reductions were observed at all stimulation intensities when contrasting peak GCaMP responses, and statistically significant at intensities of 0.3 and 0.5 µW/mm² (Figure 7C, Figure 7—figure supplement 1E, F). Collectively, these data support the idea that SeIN128 neurons directly inhibit the activity of Basins via GABA.

In the studies above, we measured the activity of all Basins while manipulating the activity of SeIN128 neurons. Connectome and behavioral analyses indicate, however, that of the four types of Basins, only Basin-2 and Basin-4 receive nociceptive input from mdIV and trigger rolling (Ohyama et al., 2015). Moreover, as noted above, an examination of the larval connectome (Ohyama et al., 2015; Winding et al., 2023) revealed that Basin-2 both receives axo-axonal inputs from SeIN128 neurons and sends excitatory projections to the same SeIN128 neurons, whereas a similar examination revealed that Basin-4 neither receives inputs from, nor sends any outputs to, SeiN128 neurons. Therefore, we hypothesized that activation of SeIN128 neurons would inhibit rolling elicited by Basin-2 activation and modify the temporal parameters of rolling, but not affect rolling elicited by Basin-4 activation.

We first examined the pattern of rolling evoked by optogenetically activating Basin-2. Basin-2 activation induced multiple bouts of rolling throughout the stimulation window (Figure 8—figure supplement 1A). Furthermore, the rolling elicited by Basin-2 activation tended to be sustained (Figure 8—figure supplement 1A). Next, to determine how SeIN128 activation affects the pattern of rolling elicited by Basin-2 activation, we optogenetically activated SeIN128 neurons and Basin-2 simultaneously. As expected, compared to the probability of rolling in control animals in which only Basin-2 was activated, the probability of rolling in experimental animals in which Basin-2 and SeIN128 neurons were simultaneously activated was significantly lower (66.7% vs 24.4%; Figure 8A, Figure 8—figure supplement 1D). We also examined other parameters of rolling, including the time from the start (onset) of stimulation to the onset of the first rolling bout, termination (offset) of the first rolling bout, and onset of the first crawling bout, as well as the duration of the rolling bout (i.e., the time from its onset to its offset). Consistent with the hypothesis that SeIN128 activation inhibits Basin-2 activity, the duration of the rolling bout significantly decreased (Figure 8B, Mann–Whitney U test, p = 0.0034, Cohen’s d = 0.351) and the time to onset of the first rolling bout significantly increased in experimental animals compared to controls (Figure 8—figure supplement 1E; Mann–Whitney U test, p < 0.001). In addition, as expected, the time to offset of the first rolling bout (Figure 8C; Mann–Whitney U test, p = 0.0047, Cohen’s d = 0.607) and time to onset of the first crawling bout (Figure 8D; Mann–Whitney U test, p = 0.0074, Cohen’s d = 0.548) both significantly decreased in experimental animals compared to controls. Collectively, these findings suggest that Basin-2 neurons play a major role in mediating the effects of SeIN128 activation on rolling induced by optogenetic activation of all Basin neurons.

Figure 8 with 1 supplement see all

Download asset Open asset

Figure 8—source data 1

Excel sheet containing raw data from behavior tracking depicted in Figure 8B-D and F-H.

https://cdn.elifesciences.org/articles/93978/elife-93978-fig8-data1-v1.zip

Download elife-93978-fig8-data1-v1.zip

Figure 8—source data 2

Excel sheet containing statistics data from behavior tracking depicted in Figure 8B-D and F-H.

https://cdn.elifesciences.org/articles/93978/elife-93978-fig8-data2-v1.zip

Download elife-93978-fig8-data2-v1.zip

To ascertain our expectation that SeIN128 activation would have little if any effect on the pattern of rolling elicited by Basin-4 activation, given the absence of any identifiable synaptic contacts between Basin-4 neurons and SeIN128 neurons based on available information on the larval connectome, we also carried out the same analyses as those described above for rolling elicited by Basin-2 activation. We examined the pattern of rolling evoked by optogenetically activating Basin-4, and found that this manipulation induced rolling mostly within the first 5 s of stimulation (Figure 8—figure supplement 1B, F). Consequently, at the population level, rolling elicited by Basin-4 activation was transient compared to the rolling elicited by Basin-2 activation (compare Figure 8—figure supplement 1A, B; Figure 8—figure supplement 1C).

We then assessed whether SeIN128 activation would affect rolling elicited by Basin-4 activation. Surprisingly, compared to control animals, the probability of rolling in experimental animals was significantly lower (66.7% vs 26.8%; Figure 8E), much as was the case for rolling elicited by Basin-2 activation. We also examined the other rolling parameters, and found that the duration of the rolling bouts (Figure 8F; Mann–Whitney U test, p = 0.032, Cohen’s d = 0.248), time to offset of the first rolling bout (Figure 8G; Mann–Whitney U test, p < 0.0047, Cohen’s d = 0.427), and time to onset of the first crawling bout (Figure 8H; Mann–Whitney U test, p < 0.001, Cohen’s d = 1.039) all significantly decreased in experimental animals compared to controls, although the effect sizes were smaller compared to those observed for rolling elicited by Basin-2 activation. The time to onset of the first rolling bout, however, did not significantly differ between experimental animals and controls (Figure 8—figure supplement 1G). These findings suggest the possibility that sites further downstream of Basin-4 neurons may be involved in inhibitory processes that affect rolling elicited by Basin-4 activation.

Feedback inhibition occurs when an excitatory neuron sends projections to an inhibitory neuron, which in turn sends projections back onto the same excitatory neuron, often at its presynaptic terminals (Isaacson and Scanziani, 2011; Kapfer et al., 2007; Ray et al., 2020; Stokes and Isaacson, 2010; Yoshimura and Callaway, 2005). The hallmark of feedback inhibition lies in its ability to modulate the duration and magnitude of incoming excitatory signals, thereby fine-tuning neural responses and maintaining homeostasis (Kapfer et al., 2007; Papadopoulou et al., 2011; Stokes and Isaacson, 2010; Yoshimura and Callaway, 2005). Compared to the fast temporal dynamics of feedforward inhibition, in which an inhibitory neuron directly inhibits an excitatory neuron downstream of it, the temporal dynamics of feedback inhibition are slower, primarily due to the added synaptic delays (two or more) following activation of an excitatory neuron (Papadopoulou et al., 2011; Ray et al., 2020; Stokes and Isaacson, 2010). The slow temporal dynamics serve to inhibit the sustained neural activity and magnitude of incoming excitatory signals (Papadopoulou et al., 2011; Ray et al., 2020; Stokes and Isaacson, 2010).

In this study, we showed that SeIN128 neurons are descending neurons whose main inputs arrive in the brain and SEZ regions, and whose outputs target the VNC. We also found that SeIN128 neurons receive excitatory inputs from Basin-2 as well as its downstream neuron A00c, and in turn send inhibitory projections back to these neurons in the VNC, potentially establishing a feedback inhibition motif that modulates the nociceptive rolling circuit. The interplay we observed among SeIN128 neurons, Basin-2, and A00c is consistent with this view. Our findings revealed that activation of SeIN128 neurons has a suppressive effect on Basin-2 activity and, notably, on the duration of rolling. These observations support the idea that feedback inhibition is critical in regulating the temporal aspects of nociceptive responses.

Neurons form a wide variety of neural networks that perform various computations in the brain. Typically, a neuron receives inputs via axo-dendritic synapses (i.e., contacts made by the axon terminals of an upstream neuron with its dendrites), which play a role in the spatial and temporal computations that lead to the firing of action potentials. Less commonly, the axon terminals of an upstream neuron may contact the soma (i.e., via axo-somatic synapses) or axon (i.e., via axo-axonal synapses) of a downstream neuron (Palay, 1956; Pinault et al., 1997; Zheng et al., 2018). Axo-axonal synapses have a subtle effect on neurotransmission at the network level because the activity in presynaptic neurons does not alter the membrane potential (Cattaert and El Manira, 1999; Guo and Hu, 2014; McGann, 2013). Axo-axonal synapses mainly affect the release probability of neurotransmitter vesicles in response to an action potential triggered in the postsynaptic neuron (McGann, 2013; Oleson et al., 2012).

Recent studies suggest that the activity of axo-axonal synapses can prevent the transmission of action potentials. For example, it has been reported that, neurotransmission mediated by type-B muscarinic receptors at lateral axo-axonal connections between Drosophila Kenyon cells is critical for stimulus specificity learning in Drosophila Manoim et al., 2022; inhibitory axo-axonal connections between Chandelier cells and CA1 pyramidal cells are important for activity-dependent plasticity (Pan-Vazquez et al., 2020; Schneider-Mizell et al., 2021); and GABAergic axo-axonal interneurons in the amygdala are crucial for generating action potentials in the principal output cells (Veres et al., 2023). Furthermore, EM connectome analyses of the entire larval brain reveal that ~70% of all synapses in Drosophila larvae are axo-dendritic whereas ~30% are axo-axonal, suggesting that the latter may have considerable influence over network function (Winding et al., 2023).

In this study, we found a feedback connection between SeIN128 and Basin-2 mediated by axon-axonal synapses (Figure 4—figure supplement 2E–G). The slow increase of SeIN128 activity in response to Basin-2 or A00c activation could potentially occur because of these axo-axonal connections. Alternatively, the slow response in SeIN128 may involve as yet unidentified indirect connections from Basin-2 or A00c to the main inputs in the SEZ region. This delayed activity may play an important role in the feedback inhibition of Basin-2 activity and consequently in the termination of rolling.

Although the current study focused on rolling, activation of SS04185 neurons appeared to influence other escape behaviors. First, during co-activation of SS04185 and Basins (Figure 1B, lower panel), the frequency of crawling following the initial rolling bout increased, whereas when only the Basins were activated, animals typically showed multiple rolling bouts during the 30-s stimulation period (Figure 1B, upper panel). This observation might be attributed to the strong stimulation induced by Chrimson activation of Basins alone that interrupts crawling via the intrusion of repeated rolling bouts, compared with the co-activation of SS04185 neurons and Basins that reduces the intrusion of rolling. More notably, after either co-activation of SS04185 neurons and Basins (Figure 1B, lower panel) or activation of SS04185 neurons alone (data not shown), the frequency of turning increased upon the cessation (offset) of stimulation, but not following the activation of Basins alone (Figure 1B, upper panel). This second observation suggests the possibility that activation of SS04185 neurons leads to sustained inhibition of turning throughout the stimulation period, which when released upon the offset of stimulation, results in a rebound in the frequency of turning beyond baseline levels. Alternatively, activation of SS04185-MB neurons alone may independently trigger the increase in turning frequency following the offset of stimulation (Figure 1B, Figure 2—figure supplement 1K). A comprehensive examination of this question, however, is beyond the scope of the present study.

Previous studies have shown that, Basin-2 and Basin-4 receive both chordotonal sensory and nociceptive sensory inputs, and in addition, play a critical role in escape behavior (Ohyama et al., 2015). Here, we investigated the differences between rolling induced by activation of Basin-2 or Basin-4. We found that activation of Basin-2 induced rolling that was sustained. Furthermore, activation of SeIN128 neurons reduced the duration of rolling induced by co-activation of Basin-2, which resulted in a delay in the onset of rolling and an earlier termination of rolling. These data indicate that activation of Basin-2 serves to maintain rolling. Connectome data indicate that SeIN128 neurons provide inhibitory input onto Basin-2, which is consistent with the finding that SeIN128 activation reduces the duration of rolling.

On the other hand, activation of Basin-4 induced rolling that was transient, which was then followed by rapid crawling. Furthermore, activation of SeIN128 neurons reduced the probability of rolling but did not affect the duration of rolling (Figure 8F). This suggests that activation of Basin-4 is important for the induction of rolling, but not its maintenance. The behavioral effects of co-activating SeIN128 and Basin-4, together with connectome data indicating the lack of connections between SeIN128 neurons and Basin-4, suggest that these descending neurons target neurons downstream of Basin-4 neurons.

The dendritic regions of SeIN128 neurons are located in the SEZ and brain, suggesting that SeIN128 neurons receive other inputs from the SEZ and brain neurons. In this study, we did not examine these inputs. Connectome data indicate that MB output neurons project onto SeIN128 neurons (Ohyama et al., 2015). Given the well-established role of MB neurons in learning, this finding suggests that SeIN128 neurons could play a role in experience-dependent modulation of rolling. Two recent studies have shown that descending neurons inhibit nociceptive neurons (Nakamizo-Dojo et al., 2023; Oikawa et al., 2023). Specifically, one study showed that insulin signaling modulates escape behavior by activating GABAergic descending neurons that inhibit nociceptive sensory neurons (Nakamizo-Dojo et al., 2023), whereas the other demonstrated an inhibitory mechanism mediated by the neuropeptide Drosulfakinin, a homolog of cholecystokinin in mammals (Oikawa et al., 2023). Whether SeIN128 neurons are also influenced by insulin signaling or Drosulfakinin, however, remains to be seen.

In this study, we did not investigate how crawling is initiated after rolling. Recent studies of the motor circuits underlying rolling and crawling suggest that different premotor neurons are involved in driving each action (Cooney et al., 2023; Kohsaka, 2023; Zarin et al., 2019). When we co-activated SS04185 and Basins, the duration of rolling decreased and the latency to onset of crawling decreased. These data are consistent with the notion that the rolling circuit inhibits the crawling circuit. It would be of interest to examine if premotor crawling neurons are inhibited during rolling, and if so, how they are activated following Basin inhibition to trigger crawling.

In summary, our study delineates a neuronal ensemble consisting of a set of descending inhibitory neurons, a first-order interneuron (Basin-2), and an ascending neuron (A00c) in fruit fly larvae, which functions as an inhibitory feedback circuit that regulates the probability and duration of rolling, and thereby facilitates the transition from rolling to crawling. This work represents another example of how detailed analyses of connectomes and functional analyses of neural and behavioral activity can identify mechanistic explanations of behavioral phenomena at the level of neural circuits—in this case, how neuronal ensembles generate behavioral sequences.

All D. melanogaster stock lines used in this study were raised on Bloomington Drosophila Stock Center cornmeal food. Flies were maintained in a humidity- and temperature-controlled chamber kept at 18 or 25°C, 40% humidity, and set to a 12-hr light/dark cycle. All crosses for experiments were reared at 25°C and 40% humidity.

First instar Drosophila larvae were heat shocked in water bath at 37°C for 12 min as previously reported (Nern et al., 2015). With the precise temporal and temperature control of heat shock, larvae with the genotype of w⁺, hs(KDRT.stop)FLP/13xLexAop2-IVS-CsChrimson::tdTomato; R54B01-Gal4.AD/72F11-LexA;20xUAS-(FRT.stop)-CsChrimson::mVenus/R46E07-Gal4.DBD had sporadic CsChrimson::mVenus expression driven by SS04185 split GAL4. As a result, the ratio of the larvae with SS04185-DN and SS04185-MB expression to those with only SS04185-MB expression was 1:1. Each individual larva was individually examined with optogenetic stimulation and behavior analysis. After behavioral experiments, mVenus expression in CNS was confirmed under the fluorescence microscope.

To optogenetically stimulate neurons, embryos were collected for 24 hr and larvae were raised on fly food plates with 0.2 mM trans-retinal (Toronto Research Chemicals, R240000). The larvae were kept in the dark at 25°C for 4 days to grow to the third instar stage. Before the experiment, food plates with larvae were rinsed with a 15% sucrose solution to separate the larvae from the food. Larvae were then moved to a sieve, washed with water, dried, and placed evenly on 2% agar plates. The agar plate with animals were placed under a camera in the arena of the behavior rig.

Larvae were tracked in real-time using Multi Worm Tracker (MWT) software (Swierczek et al., 2011, http://sourceforge.net/projects/mwt). The contour, spine, and center of mass for each larva were generated and recorded by MWT as a function of time. From these tracking data, the key parameters of larval motion were computed using Choreography software (a component of the MWT software package which measured the behavioral parameters offline) as described previously (Ohyama et al., 2013; Ohyama et al., 2015). The behavioral parameters generated by the Choreography algorithm include speed, crabspeed (i.e., speed perpendicular to the body axis), body curvature, angle of head bending, body length, body width, area (dorsal view), and bias (i.e., fractional excess of time spent moving in one direction). In this offline process, objects that were tracked for less than 5 seconds or moved less than one body length of a larva were rejected. We refer the reader to the open-source package for further details of the software implementations for the above calculations.

Standard immunocytochemical procedures were followed (Patel, 1994). Briefly the CNSs of Drosophila larvae were dissected in phosphate-buffered saline (PBS). After dissection, tissues were fixed with 4% paraformaldehyde for 20 min, washed with PBS three times and then washed with 0.4% Triton X-100 in PBS (PBST) twice. Samples were incubated at room temperature with a blocking solution (5% normal goat serum [NGS]) for 1 hr. Next, the samples were incubated with the primary antibody solutions at 4°C overnight and washed for 15 min six times. Specially, anti-VGAT was incubated for 48 hr to compensate for the permeability. The primary antibodies were diluted at concentrations of 1:3000 for chicken anti-GFP; 1:1000 for rabbit anti-GFP, rabbit anti-GABA, and rabbit anti-dsRed; 1:200 for rabbit anti-VGAT; 1:50 for rat anti-Elav, 1:50 for mouse nc82; and 1:20 for rat anti-DN-Cadherin, mouse anti-Fas2, mouse anti-choline acetyltransferase (ChAT), and rabbit anti-VGLUT in 5% NGS. CNS samples were then incubated with a secondary antibody solution at 4°C overnight and washed 15 min for six times. The secondary antibodies, including anti-chicken Alexa488, anti-rabbit Alexa488, anti-mouse Alexa568, anti-rabbit Alexa568, and anti-rat Alexa568, were all diluted at the concentration of 1:500. These samples were mounted in VECTASHIELD antifade mounting medium and imaged by a Zeiss LSM 710 confocal microscope with a 20×/NA0.8 objective lens (Zeiss) and Zen digital imaging software (Zeiss). For quantifying the expression of VGAT or GABA expression, laser power and gains are consistent between samples. All images were processed using Fiji software (https://imagej.new/Fiji, ImageJ, NIH Bethesda).

Larval CNS image stacks were processed with FIJI. For Figure 5—figure supplement 1C,D, four to six slices along the z dimension were averaged. The neuropil for VGAT or GABA channels at segments A4 to A6 and cell body regions for elav staining were manually selected as regions of interest (ROIs). The mean intensity of VGAT or GABA was measured and normalized by the mean value of elav staining.

The CNSs of third instar larvae were dissected out in cold Baines external physiological saline (135 mM NaCl, 5 mM KCl, 5 mM TES, 36 mM sucrose, 2 mM CaCl₂·2H₂O, 4 mM MgCl₂·6H₂O, pH 7.15), and secured on a poly-L-lysine coated cover glass placed in a small Sylgard plate.

Functional calcium imaging experiments were performed on a customized two-photon microscope equipped with a Galvo-Resonant Scanner (Cambridge) controlled by Scanimage software (mbf BIOSCIENCE) using a 40×/0.80NA water immersion objective (LUMPlanFL, Olympus). A Mai Tai, Ti:Sapphire Ultrafast Laser (Spectra Physics) tuned to 925 nm was used for excitation of GCaMP protein. Fluorescence signals were collected with photomultiplier tubes (Hamamatsu) after bandpass filtering. Images were acquired by the Galvo-Resonant Scanner for a single plane of the CNS.

Each larva was stimulated by a 620-nm LED (Thorlabs) through the objective three times with a 30-s interval between periods of stimulation. Every stimulus consisted of a 30-ms pulse given every 100ms for a total of 1 s. Light intensity was measured to be 0.8–1.4 μW/mm². Images were acquired at a resolution of 512 × 512 pixels with a frame rate of 30 fps. Fluorescence intensities were averaged to 6 fps and processed in FIJI, and analyzed in MATLABwith customized scripts (available at https://github.com/Jiayi2019/2-photon-analysis; copy archieved at Zhu, 2024b). ROIs were determined by the standard deviation of the full recording. ΔF = (F − F₀)/F₀. F₀ is the average of images taken 10 frames (i.e., 1.7 s) before stimulation. F is the mean value of the fluorescence in the ROI averaged every five frames from the start of the 5 s period before stimulation to end of the 15 s period after the onset of each stimulation. For each larva, ΔF is obtained through averaging the ΔF during the three stimulation periods. The peak ΔFs were the maximal values selected from the onset of stimulation to 15 s after stimulus onset.

The probabilities for each response were analyzed by Chi-square tests. For the other parameters, when multiple groups were tested, their normality was examined first. If the normality assumption was rejected, Kruskal–Wallis tests were performed for multiple group variance comparisons, followed by multiple-comparison-corrected Wilcoxon–Mann–Whitney U tests as post hoc pairwise comparisons. If normality was met, analysis of variance was performed for variance comparisons and multiple-comparison-corrected Student’s t-tests were utilized for pairwise comparisons. For two group comparisons, the Wilcoxon–Mann–Whitney U test was conducted if the normality assumption was offended, and the Student’s t-test was applied if normality was met. All analyses were conducted with MATLAB.

One or more of authors of this paper self-identifies as a member of the LGBTYQ+ community.

The funders had no role in study design, data collection, and interpretation, or the decision to submit the work for publication.

We thank JW Truman, A Cardona, and M Zlatic for generating and sharing the split Gal4 lines. J Truman and M Zlatic for constructive inputs on the project. Confocal images were collected at the McGill University Advanced Bio Imaging Facility (ABIF), RRID:SCR_017697. We also thank the Bloomington stock center for providing fly stocks. We thank members of Ohyama lab for critical comments on the manuscript. This work was funded by McGill University, the Natural Sciences and Engineering Research Council (NSERC, RGPIN/04781-2017), the Canadian Institute of Health Research (CIHR, PTJ-376836), the Fonds de recherche du Quebéc – Nature et technologies (FRQNT, 2019-N-25523), and the Canada Foundation for Innovation (CFI, CFI365333). JCB was supported by Fonds de recherche du Quebéc-Santé (FRQS) graduate training award.

1. Sent for peer review: November 15, 2023
2. Preprint posted: November 21, 2023
3. Reviewed Preprint version 1: January 26, 2024
4. Reviewed Preprint version 2: July 8, 2024
5. Version of Record published: August 28, 2024

You can cite all versions using the DOI https://doi.org/10.7554/eLife.93978. This DOI represents all versions, and will always resolve to the latest one.

© 2024, Zhu et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.