1. Neuroscience
Download icon

Corollary discharge promotes a sustained motor state in a neural circuit for navigation

  1. Ni Ji
  2. Vivek Venkatachalam
  3. Hillary Denise Rodgers
  4. Wesley Hung
  5. Taizo Kawano
  6. Christopher M Clark
  7. Maria Lim
  8. Mark J Alkema  Is a corresponding author
  9. Mei Zhen  Is a corresponding author
  10. Aravinthan DT Samuel  Is a corresponding author
  1. Department of Physics and Center for Brain Science, Harvard University, United States
  2. Department of Neurobiology, University of Massachusetts Medical School, United States
  3. Lunenfeld-Tanenbaum Research Institute, Mount Sinai Hospital, Canada
  4. Departments of Molecular Genetics, and Physiology, University of Toronto, Canada
Research Article
  • Cited 0
  • Views 1,001
  • Annotations
Cite this article as: eLife 2021;10:e68848 doi: 10.7554/eLife.68848

Abstract

Animals exhibit behavioral and neural responses that persist on longer timescales than transient or fluctuating stimulus inputs. Here, we report that Caenorhabditis elegans uses feedback from the motor circuit to a sensory processing interneuron to sustain its motor state during thermotactic navigation. By imaging circuit activity in behaving animals, we show that a principal postsynaptic partner of the AFD thermosensory neuron, the AIY interneuron, encodes both temperature and motor state information. By optogenetic and genetic manipulation of this circuit, we demonstrate that the motor state representation in AIY is a corollary discharge signal. RIM, an interneuron that is connected with premotor interneurons, is required for this corollary discharge. Ablation of RIM eliminates the motor representation in AIY, allows thermosensory representations to reach downstream premotor interneurons, and reduces the animal’s ability to sustain forward movements during thermotaxis. We propose that feedback from the motor circuit to the sensory processing circuit underlies a positive feedback mechanism to generate persistent neural activity and sustained behavioral patterns in a sensorimotor transformation.

Introduction

Animals are able to generate behaviors that persist beyond the timescales of the inciting sensory stimuli. For example, fish continue to fixate their gaze after the onset of darkness (Seung, 1996; Aksay et al., 2007). A brief aversive stimulus can evoke prolonged escape responses in many species (Li et al., 2006; Herberholz et al., 2002). Lasting behavioral states require circuit mechanisms to turn a transient stimulus into persistent neuronal activity (Lee and Dan, 2012; Major and Tank, 2004Hoopfer et al., 2015; Inagaki et al., 2019; Kennedy et al., 2020). Theoretical studies have explored roles for recurrent circuitry, in particular positive feedback, in generating persistent neural activity (Seung, 1996). While recurrent connections are abundant in the brain, establishing causality between recurrent circuitry, persistent neural activity, and sustained behavior states has been challenging in part due to the technical difficulties in experimentally dissecting neural dynamics across entire sensorimotor pathways.

One type of recurrent connectivity that is widely observed across phyla are neuronal projections that convey motor-related signals to brain regions for sensory processing (Wurtz, 2018; Crapse and Sommer, 2008). These signals, called corollary discharge (CD) or efference copy (EC), were first proposed (Sperry, 1950; Holst and Mittelstaedt, 1950) and subsequently demonstrated (Poulet and Hedwig, 2002; Requarth and Sawtell, 2014; Schneider et al., 2014; Kim et al., 2015) as mechanisms to cancel sensory reafferents generated by self-motion. Recent studies, however, have also identified examples of motor-to-sensory feedback that are excitatory (Hendricks et al., 2012; Lee et al., 2013; Zagha et al., 2013; Fu et al., 2014). Emerging evidence from cortex- or brain-wide activity patterns has revealed widespread representation of motor states in sensory processing regions (Kato et al., 2015; Stringer et al., 2019; Aimon et al., 2019; Musall et al., 2019; Marques et al., 2020), which is likely to be caused, at least in part, by motor-to-sensory feedback. These observations suggest diverse roles for CD in sensorimotor processing.

The compact nervous system and optical accessibility of Caenorhabditis elegans make it possible to explore circuit mechanisms that underlie sustained behavioral states in intact animals (Gao et al., 2015). C. elegans requires persistent motor states to navigate variable sensory environments. During locomotion, the animal alternates between continuous forward movements (runs) and brief backward movements (reversals). When navigating through a chemical or thermal gradient, C. elegans employs a biased random walk strategy, selectively extending forward runs when moving along preferred directions and shortening runs when veering off course (Pierce-Shimomura et al., 1999; Ryu and Samuel, 2002; Iino and Yoshida, 2009; Hedgecock and Russell, 1975; Mori and Ohshima, 1995; Luo et al., 2014a). During forward runs, C. elegans also gradually steers its heading angle towards the preferred direction, a strategy called klinotaxis (Ward, 1973; Iino and Yoshida, 2009). Forward runs play a crucial role in C. elegans navigation, but the neural circuit basis for run persistence remains poorly understood (Ferrée and Lockery, 1999).

Here, we study the sensorimotor pathway that controls navigation towards warmer temperatures (positive thermotaxis). Past studies have revealed a multilayered neural circuit underlying this behavior. AFD is the thermosensory neuron that mediates both positive and negative thermotaxis (Luo et al., 2014a; Hawk et al., 2018). Its principal chemical synaptic partner, AIY, is a first-layer interneuron specifically required for positive thermotaxis. AIY responds to temperature variations due to excitatory input from AFD (Clark et al., 2006Clark et al., 2007; Narayan et al., 2011; Hawk et al., 2018). AIY sends synaptic outputs to multiple second-layer interneurons, which in turn synapse onto head motor neurons and premotor interneurons that drive forward runs or reversals. AIY has been shown to promote the speed and duration of forward locomotion (Li et al., 2014; Tsalik and Hobert, 2003). AIY is also postsynaptic to multiple other sensory neurons and is thought to play a role in navigation across different sensory modalities by controlling run duration (Gray et al., 2005; Wakabayashi et al., 2004; Tsalik and Hobert, 2003).

We probed mechanisms by which AIY biases random walks during positive thermotaxis. Imaging AIY activity in moving animals reveals that AIY encodes both temperature and motor information. Previous studies have found AIY to encode either sensory stimuli (Chalasani et al., 2007; Clark et al., 2006) or locomotory state (Li et al., 2014; Luo et al., 2014b) in different experimental paradigms. Here, we found that thermosensory response in AIY is gated by the locomotory state of the animal. In the absence of thermosensory stimuli, AIY activity reliably encodes the locomotory state. When exposed to thermal fluctuations during forward runs, AIY activity exhibits variability but tends to be excited by warming and inhibited by cooling. During reversals, AIY activity remains low and does not encode thermal stimuli.

We demonstrate that the motor state encoding in AIY represents a CD signal from premotor interneurons that drive the forward run and reversal states. This CD signal requires RIM, an interneuron that is connected with premotor interneurons. In the absence of RIM, AIY activity reliably encodes thermal stimuli regardless of the locomotory state. Loss of RIM also leads to increased thermosensory representation in premotor interneurons and motor neurons. At the behavioral level, loss of RIM led to defects in positive thermotaxis by reducing the persistence of the forward run state. Our results establish a role for CD in sustaining a motor state in variable or fluctuating sensory environments.

Results

Forward movements are sustained across thermal fluctuations during positive thermotaxis

C. elegans navigates towards temperatures that correspond to prior thermal experience. To evoke positive thermotaxis, we placed young adults cultivated at 25°C on a linear thermal gradient spanning 19–23°C (Figure 1). Consistent with earlier reports, these animals exhibited biased random walk and klinotaxis towards warmer temperatures (Figure 1B; Luo et al., 2014a; Yamaguchi et al., 2018): runs that pointed in favorable directions were lengthened (Figure 1B); forward heading angles gradually reoriented towards temperatures that correspond to prior experience (Figure 1B). Without a temperature gradient, there was no evident modulation of either run length or heading angle (Figure 1B).

Figure 1 with 1 supplement see all
Sustained forward motor state despite temperature fluctuations during positive thermotaxis.

(A) Example trajectories of wild-type C. elegans cultivated at 25°C migrating up a linear temperature gradient over 20 min. Top: schematics of the thermal gradient. Middle: trajectories of 49 animals during positive thermotaxis. The starting points of all trajectories are aligned (yellow dot) and the end points are marked by magenta dots. Bottom: a histogram of the final location of animals. (B) Left column: duration of forward runs as a function of their overall direction (vector pointing from the starting point to the end point of the run). Right column: instantaneous velocity during forward runs as a function of the instantaneous heading angle. Top row: data from animals exposed to spatial thermal gradients (top, N = 140). Bottom row: data from animals under constant temperature surfaces (bottom, N = 73). (C) Thermotaxis trajectory of a single animal during thermotaxis with alternating periods of forward movement and reversals (left), and the instantaneous heading angle over time during one extended period of forward movement within the trajectory (right). Asterisks denote periods where the heading direction pointed down the thermal gradient. (D) Histogram of temporal changes in temperature (dT/dt) experienced by animals during forward runs that ended up pointing up the temperature gradient. Data from N = 140 wild-type animals exposed to linear thermal gradient and N = 73 wild-type animals exposed to constant temperature of 21°C. Error bars are standard errors of the mean (s.e.m.).

Individual trajectories during positive thermotaxis revealed periods of forward movement that carry the animal up the temperature gradient. Although these periods of forward movement are persistent in duration, they are not always persistent in direction (Figure 1C, Figure 1—figure supplement 1). C. elegans experiences temporal changes in temperature on spatial gradients because of its own movements. Most runs – even those that orient the animal towards warmer temperatures – will involve periods of both warming and cooling stimuli because of frequent changes in movement direction (Figure 1D, Figure 1—figure supplement 1). Thus, C. elegans reveals an ability to sustain forward movement up temperature gradients despite transient cooling fluctuations.

Thermosensory representation in AIY is gated by the motor state

To uncover circuit mechanisms for sustaining forward movement against thermal fluctuations, we simultaneously imaged the calcium activity across a group of neurons involved in thermosensory processing, locomotory control, or both. These include the principal thermosensory neuron AFD, its primary postsynaptic partner AIY, the premotor interneuron AVA, the left-right pair of the head motor neurons RME, SMDV, and SMDD, and the RIM interneuron that extensively connects with premotor interneurons and motor neurons (Figure 2A, B). To measure locomotory behavior while minimizing motion artifact, we performed imaging in semi-constrained animals that exhibited sinusoidal movements akin to those observed in free-moving animals (Figure 2—figure supplement 1, also see Materials and methods). This approach allowed us to simultaneously examine the encoding of thermosensory and motor information in the same neurons.

Figure 2 with 4 supplements see all
Characterization of circuit-level neural activity in the behaving animal with fluctuating or constant temperatures.

(A) Anatomical connections among neurons implicated in positive thermotaxis, locomotory control, or both. Connectivity is inferred from both the original C. elegans connectome (White et al., 1986) and a recent reconstruction of the connectome across the developmental time course (Witvliet et al., 2020) (see http://www.nemanode.org) (B) Example maximum projection of a confocal z-stack taken from a transgenic animal expressing GCaMP6s and mCherry in neurons examined in this study. (C) Example ratiometric calcium activity trace (left panel) and histogram (right panel) of the thermosensory neuron AFD in response to oscillating temperature. (D) Average cross-correlation function between thermal stimuli and AFD activity during forward run (green) or reversal (red) states. N = 5 wild-type animals. (E) Simultaneously measured calcium activity of interneurons and motor neurons involved in thermosensory processing and/or motor control. The activity traces (left) and activity histograms (middle) are from the same sample dataset. (F) Average cross-correlation functions between the thermal stimuli and individual neurons shown in (E), conditioned on the animal in forward run (green) or reversal (red) states. N = 6 wild-type animals. Error bars are 95% CI of the mean. Wilcoxon rank-sum test was used to compare the peak mean cross-correlation values during forward runs (green) versus reversals (red). *p<0.05; **p<0.01; ***p<0.001; no asterisk p>0.05. (G) Thermal stimulus-triggered activity of the AIY interneuron during forward runs (left column) and reversals (right column) in wild-type animals (N = 5). Individual stimulus epochs from the same neuron under the given motor state were concatenated into heat maps, with the average calcium activity trace shown on top. (H) Simultaneous recording of the activity of interneurons and motor neurons under constant temperature. The histograms on the right are derived from the sample activity traces to the left. (I) Pairwise cross-correlation functions among neurons examined in (E) and (H) under oscillating (orange, N = 6) or constant temperature (blue, N = 7). Error bars are standard errors of the mean (s.e.m.). Wilcoxon rank-sum test was used to compare the peak mean cross-correlation values between data under oscillating temperature versus data under constant temperature. *p<0.05; **p<0.01; ***p<0.001; no asterisk p>0.05.

Figure 2—source data 1

wild type circuit activity under thermal stimulation.

https://cdn.elifesciences.org/articles/68848/elife-68848-fig2-data1-v3.xlsx

First, we measured the activity of the AFD thermosensory neuron and AIY, its principal postsynaptic partner, by calcium imaging in moving animals. Subjected to oscillating temperatures below the preferred temperature, C. elegans exhibits positive thermotaxis. As previously reported (Clark et al., 2006), AFD’s activity phase locks to periodic variations in temperatures, rising upon warming and falling upon cooling (Figure 2C, D). AFD activity did not covary with transitions between forward run and reversal states (Figure 2C), indicating that motor commands arise downstream of the thermosensory neuron.

Next, we simultaneously monitored AIY calcium dynamics along with components of the motor circuit known to code forward and reversal motor states (Figure 2A, B, Figure 2—videos 1 and 2). We found that AIY encodes both temperature variations in a motor state-dependent manner. During forward movements, AIY’s calcium activity on average increased upon warming and decreased upon cooling with substantial trial-to-trial variability (Figure 2E–G). During reversals, these responses were largely absent (Figure 2E).

Unlike AFD and AIY, all motor circuit neurons that we examined exhibited little to no phase-locked response to thermosensory stimulation. Instead, motor neurons reliably encoded the forward versus reversal movement states (Figure 2E, F, see the stimulus cross-correlation plots). In animals subjected to oscillating thermosensory stimulation, AVA calcium activity exhibited high and low states that correlated with backward and forward movement, respectively (Figure 2E). This observation is consistent with the known activity profile of AVA in the absence of thermal stimulation and its functional role in promoting the reversal state (Chalfie et al., 1985; Kawano et al., 2011; McCormick et al., 2011; Kato et al., 2015). Similarly, the RIM interneuron was selectively active during reversals and its activity positively correlated with AVA activity. The head motor neurons RME, SMDD, and SMDV were selectively active during forward runs. The SMD neurons exhibited alternating activity patterns, consistent with previous reports (Hendricks et al., 2012). Together, these observations indicate that, for positive thermotaxis, the sensorimotor transformation progresses through three layers of processing (Figure 2A): the thermosensory neuron encodes only thermal stimuli; the first layer interneuron encodes both thermal stimuli and motor states; the premotor and motor neurons primarily encode the motor state.

To elucidate which of these activity patterns are driven by thermosensory inputs, we next measured the activity of these neurons under constant temperature (Figure 2H). We found that the activity of all six neurons continued to be modulated by locomotory state in the same way as when exposed to oscillating temperature. Furthermore, the pairwise cross-correlation between neurons remained the same under both oscillatory and constant temperature (Figure 2I). Specifically, AIY and RME were positively correlated with one another and were anti-correlated with AVA. The head motor neurons SMDD and SMDV were strongly anti-correlated with one another, consistent with previous reports (Hendricks et al., 2012). Both SMDD and SMDV were positively correlated with RME and AIY, consistent with their role in controlling head oscillations during forward locomotion (Pirri et al., 2009). These observations indicate that the widespread encoding of motor state in neurons downstream of AFD does not require thermosensory input and may instead reflect an intrinsic circuit property.

Motor coding in AIY is a CD signal that requires RIM

Our finding that AIY encodes thermal information in a manner that depends on motor state suggests a critical role in sensorimotor transformations during positive thermotaxis. In animals exposed to either constant or oscillating temperatures, AIY activity consistently rises at the beginning of forward runs and decays at the onset of reversals (Figure 3A, B). Moreover, AIY exhibited persistent activation, the duration of which coincided reliably with that of the forward run state (Figure 3—figure supplement 1). How does AIY, a first-order interneuron, acquire a robust motor signal?

Figure 3 with 2 supplements see all
Motor-related activity in the AIY interneuron represents a corollary discharge signal.

(A, B) Calcium activity of the AIY interneuron aligned to the onset of forward runs (left column) or reversals (right column) in animals exposed to oscillating temperature (A, N = 6) or constant temperature (B, N = 5). Each row of the heat plots represents AIY calcium activity during a single behavioral epoch. The curve on top of each panel represents activity dynamics averaged across individual epochs. Broken lines indicate the onset and offset of each behavior epochs. (C) Calcium activity of the AIY interneuron aligned to the onset of forward runs or reversals in animals expressing tetanus toxin (TeTx) specifically in AIY (N = 4). (D) Calcium activity of AIY aligned to onset or offset of AVA activation in animals immobilized by the cholinergic agonist levamisole (N = 4). The ON and OFF states of AVA activity are defined by binarizing AVA activity using the Otsu method. See Materials and methods for details. (E) Change in AIY activity before versus. after the onset of forward runs (green) or reversals (red) for datasets shown in (A) and (B). Wilcoxon Signed rank test was used to test if the change in AIY activity has a median significantly different from zero, *p<0.05; **p<0.01; ***p<0.001; n.s., non-significant.

Because AIY has an established role in promoting forward locomotion, we first tested whether motor representation in AIY arises due to feedforward output from AIY to the downstream circuit. We imaged AIY activity after blocking vesicle release from AIY through cell-specific expression of tetanus toxin (TeTx) (Figure 3C). Despite the lack of synaptic and dense-core-vesicle-dependent chemical release (Whim et al., 1997), AIY activity remained strongly coupled to the motor state, implying that AIY must receive the motor state signal.

We explored the possibility that proprioception, elicited by movement itself, underlies the calcium response in AIY. We imaged neural activity in AIY and the rest of the thermosensory circuit in immobilized animals under constant temperature. As in moving animals, AIY’s activity remained anti-correlated with neurons active during reversals (AVA) and correlated with neurons active during forward movement (RME and SMDD/V) (Figure 3—figure supplement 2). Furthermore, aligning AIY activity to the onset and offset of AVA activity revealed average activity patterns similar to that of the motor state representation in moving animals (Figure 3D, E). C. elegans movement is not required for AIY activity to reflect motor state, arguing against proprioception.

We next asked whether AIY receives CD from neurons that encode the motor command. We imaged AIY activity in moving animals upon ablation of AIY’s downstream interneurons and premotor interneurons (Figure 4A). For interneurons, we focused on AIB and RIM. AIB shares electrical synapses with RIM and the AFD thermosensory neuron (White et al., 1986). In the context of chemotaxis, AIB and RIM have been shown to regulate variability in the neuronal and behavioral response to olfactory inputs (Gordus et al., 2015). We also tested AVA and AVB, premotor interneurons that regulate reversal and forward movement, respectively. Ablations were performed by expressing flavoprotein miniSOG, which induces acute functional loss and neuronal death by photoactivation (Qi et al., 2012).

Figure 4 with 2 supplements see all
A corollary discharge (CD) pathway that requires the RIM interneuron couples AIY activity with the motor state.

(A) Change in AIY activity before versus. after the onset of forward runs (green) or reversals (red) in animals where candidate neurons for relaying the CD signal have been ablated. Data are from RIM-ablated animals (N = 9); AIB-ablated animals (N = 3); AVB-ablated animals (N = 4); AVA-ablated animals (N = 4); and AVA/AVE/RIM (nmr-1::miniSOG)-ablated animals (N = 4). Error bars are 95% CI. Wilcoxon rank-sum test was used to test if the distributions of AIY activity before and after the onset of motor states have the same median: *p<0.05; **p<0.01; ***p<0.001; no asterisk p>0.05. (B) AIY activity aligned to the onset of forward runs (left column) or reversals (right column) in the animals where neurons RIM (upper panels, N = 9) or AIB (lower panels, N = 3) have been genetically ablated. (C) AIY activity in response to optogenetic stimulation of the AVB premotor interneurons in wild-type animals grown on all-trans retinal (ATR) (top, N = 5), RIM-ablated animals grown on ATR (N = 6), and wild-type animals grown without ATR (N = 3). (D) Top: AIY activity 2.5 s before (pre-stimulus) versus 2.5 s at the end of the AVB opto-stimulation (post-stimulus) under experimental conditions shown in (D). Bottom: average change in AIY activity pre-and post-stimulation under experimental conditions shown (D). Trials are sorted into three groups based on pre-stimulation AIY activity. Same datasets as in (C). (E) AIY activity in response to optogenetic stimulation of the AVA premotor interneurons in wild-type animals grown on ATR (top, N = 4), RIM-ablated animals grown on ATR (middle, N = 5), and wild-type animals grown without ATR (bottom, N = 5). (F) Top: AIY activity 2.5 s before (pre-stimulus) versus 2.5 s at the end of the AVA opto-stimulation (post-stimulus) under experimental conditions shown in (E). Bottom: average change in AIY activity pre- and post-stimulation under experimental conditions shown in (E). Trials are sorted into three groups based on pre-stimulation AIY activity. Same datasets as in (E). For (E) and (F), error bars are 95% CI; Wilcoxon signed-rank test was used to test if the average post-stimulation change in AIY activity was significantly different from 0. *p<0.05; **p<0.01; ***p<0.001; no asterisk p>0.05.

We found that ablating AIB did not abolish the motor state representation in AIY (Figure 4A, B). Neither did the removal of the premotor interneurons AVA or AVB alone (Figure 4A, Figure 4—figure supplement 1A, B). However, AIY lost its motor state representation when we ablated RIM either by itself or in combination with other premotor interneurons (Figure 4A, B, Figure 4—figure supplement 1C).

RIM activity has been shown to be correlated with the AVA premotor interneuron that promotes reversals and anti-correlated with the AVB premotor interneuron that promotes forward runs (Kawano et al., 2011; Gordus et al., 2015; Kato et al., 2015). RIM has been shown to promote long reversals (Gray et al., 2005) and to suppress head oscillations during reversals (Alkema et al., 2005). A recent study demonstrated that RIM also promotes the stability of forward runs when hyperpolarized (Sordillo et al., 2021). To probe whether RIM is required for the motor state signal to appear in AIY, we optogenetically activated either AVA or AVB while simultaneously measuring AIY calcium activity in immobilized animals. Activation of AVB using the light-gated opsin chrimson triggered an increase in AIY calcium levels (Figure 4C, D). Activation of AVA (Klapoetke et al., 2014) triggered a decrease in AIY calcium levels (Figure 4E, F). When RIM was ablated, AIY calcium signals no longer responded to optogenetic activation of either AVA or AVB, suggesting that RIM is part of the CD pathway from the motor circuit to AIY. Without RIM, AIY activity no longer reflected or depended on the motor state, but the premotor interneurons AVA continued to encode the backward and forward movement, albeit with reduced bimodal activity (Figure 5A, C). Thus, RIM is not essential for generating motor commands, but is necessary to relay motor information to AIY, a first-layer interneuron.

Figure 5 with 2 supplements see all
Characterization of circuit-level neural activity in behaving RIM-ablated animals under fluctuating or constant temperature.

(A) Simultaneously measured activity of AIY and neurons of the motor circuit in RIM-ablated animals under oscillating temperature. Middle panels show histograms of neuron activity during forward runs (green) or reversals (red) for the dataset to the left. Right panels show average cross-correlograms between neural activity and thermal stimuli during forward runs and reversals across RIM-ablated animals (N = 3). Error bars are 95% CI of the mean. Wilcoxon rank-sum test was used to compare the peak mean cross-correlation values during forward runs (green) versus reversals (red). *p<0.05; **p<0.01; ***p<0.001; no asterisk p>0.05. (B) Thermal stimulus-triggered activity of the AIY interneuron during forward runs (left column) and reversals (right column) in RIM-ablated animals (N = 3). Individual stimulus epochs from the same neuron under the given motor state were concatenated into heat maps, with the average activity trace shown on top. (C) Simultaneously measured activity of AIY and neurons of the motor circuit in RIM-ablated animals under constant temperature (left). Right panels show histograms of neuron activity during forward runs (green) or reversals (red). (D) Violin plots showing the distribution of changes in AIY activity in response to warming (left) or cooling (right) stimuli under forward run or reversal state in wild-type and RIM-ablated animals. Dotted black lines indicate threshold values used to calculate the response probability in (E). (E) Probability that the magnitude of change in AIY activity upon warming or cooling is above defined thresholds in wild-type or RIM-ablated animals. See Materials and methods for details. For (D) and (E), N = 5 wild-type animals and N = 3 RIM-ablated animals. Error bars are 95% CI of the mean. Wilcoxon rank-sum test *p<0.05; **p<0.01; ***p<0.001; n.s., non-significant.

Figure 5—source data 1

Circuit activity under thermoal stimulation in RIM ablated animals.

https://cdn.elifesciences.org/articles/68848/elife-68848-fig5-data1-v3.xlsx

RIM-mediated CD does not depend on chemical synaptic transmission

We sought synaptic mechanisms by which RIM may contribute to the CD pathway. RIM expresses VGLUT3/EAT-4, indicating the potential involvement of glutamatergic synaptic transmission (Serrano-Saiz et al., 2013). RIM also synthesizes tyramine, a monoamine neuromodulator (Alkema et al., 2005). We thus imaged AIY activity in loss-of-function mutants for glutamatergic signaling (VGLUT3/eat-4), tyramine synthesis (TDC/tdc-1), vesicular monoamine transport (VMAT/cat-1), and peptidergic signaling (CAPS/unc-31). AIY activity co-varied with the motor state in all mutants, but the difference in AIY activity between the forward run and reversal states was less distinct in mutants defective for vesicular monoamine transport (VMAT/cat-1) or tyramine synthesis (TDC/tdc-1) (Figure 4—figure supplement 2A, B). Blocking vesicle fusion in RIM by expressing TeTx (Ptdc-1::TeTx) also attenuated the motor state representation in AIY (Figure 4—figure supplement 2A, B).

Since perturbation of chemical synaptic transmission did not fully abolish motor-related activity in AIY, neuronal communication that is independent of classic chemical synaptic transmission is likely involved in relaying CD to AIY. As previously reported (Kawano et al., 2011; Kato et al., 2015; Gordus et al., 2015), RIM activity is strongly correlated with the AVA premotor interneuron, higher during reversals and lower during forward movement. AIY, on the other hand, exhibits increased activity during forward movement (Figure 3). This sign reversal may be explained by an inhibitory input from RIM to AIY. Alternatively, RIM may play a permissive role in allowing the motor-related feedback to AIY. Taken together, our results suggest that the joint representation of sensory and motor signals in AIY arises from separate sources: feedforward input from AFD and feedback from the motor circuit that is dependent on RIM.

RIM-dependent CD promotes persistent forward states and more effective thermotaxis

RIM plays a critical role in the motor state-dependent modulation of AIY calcium activity. This prompted us to examine the effect of disrupting the CD signal on sensorimotor transformations. When RIM-ablated animals were subjected to oscillating temperatures, AIY activity was no longer coupled to the motor state, but instead reliably tracked temperature fluctuations during both forward and backward movements (Figure 5A, B, Figure 5—figure supplement 1A, B, Figure 5—video 1). Under oscillating temperature, the average duration of AIY activation was shortened compared to wild type, though no significant difference was observed under constant temperature (Figure 5D).

When RIM was ablated, we were also able to detect the representation of thermosensory oscillations in the activity pattern of the AVA premotor interneuron and the head motor neurons RME and SMDV (Figure 5A). This observation suggests that the loss of the RIM-dependent motor signal resulted in a sensorimotor circuit that becomes more susceptible to fluctuations in thermosensory input. Without RIM and the motor state encoding in AIY, fluctuations in thermosensory inputs are readily propagated to the motor circuit. Thus, the RIM-dependent CD may play an important role in sustaining neural activity states through fluctuating sensory inputs.

We tested this hypothesis by examining the effect of RIM ablation on positive thermotaxis (Figure 6A, Figure 6—figure supplement 1A). Compared to wild-type animals, RIM-ablated animals exhibited an overall reduction in thermotaxis bias (Figure 6B). These animals were specifically defective in their ability to sustain forward locomotion when moving up the thermal gradient, while their ability to gradually modify heading angle during a forward run remained intact (Figure 6C, Figure 6—figure supplement 1A). At constant temperature, run durations are similar between wild-type and RIM-ablated animals (Figure 6—figure supplement 1B). Thus, the loss of RIM specifically disrupted the animal's ability to sustain forward runs up temperature gradients.

Figure 6 with 1 supplement see all
RIM ablation disrupts positive thermotaxis and leads to increased susceptibility to sensory fluctuations.

(A) Example trajectories of RIM-ablated animals (N = 39) cultivated at 25°C and exposed to the same thermal gradient as in Figure 1A. Top: schematic of the thermal gradient. Middle: trajectories of individual animals during positive thermotaxis. The starting points of all trajectories are aligned (yellow dot) and the end points are marked by magenta dots. Bottom: a histogram of the final location of the animals at the end of the 20 min period. (B) Average thermotactic bias of wild-type (N = 140) versus RIM-ablated animals (N = 102). (C) Forward run duration as a function of forward run direction in RIM-ablated animals (blue) compared to the wild type (gray). Error bars are standard errors of the mean (s.e.m.). Wilcoxon rank-sum test was used to compare the run duration of wild-type versus RIM-ablated animals. p<*0.05, **0.01, ***0.001, p>0.05 (non-significant) for panels without asterisk. (D) Velocity profiles of wild-type (left) and RIM-ablated (right) animals aligned to the end of cooling epochs that occurred during forward runs. Heat maps are generated by concatenating velocity profiles from individual cooling epochs along the y-axis and sorting by the average velocity within the first 2 s after the offset of cooling epochs (shown as line plot to the right). Black dotted lines divide instances in which forward runs continued past the offset of cooling epochs from instances where reversals ensued within the first 2 min of cooling offset. (E) Histograms of post-cooling velocities in wild-type (top) and RIM-ablated animals (bottom). Analysis applied to same dataset as in (D). (F) Fraction of cooling epochs that were followed by transition from forward runs to reversals as a function of the duration of cooling epochs (orange: wild type, gray: RIM ablated).

If RIM-dependent motor feedback serves to filter out transient thermal fluctuations during positive thermotaxis, then loss of RIM should render the forward run state more susceptible to thermal variations. To test this prediction, we analyzed how periods of cooling affected the forward run state in wild-type and RIM-ablated animals (Figure 6D–F). Overall, cooling induced reversals at a higher probability in RIM-ablated animals compared to the wild type. This difference was particularly notable for persistent periods of cooling lasting from 7 to 30 s (Figure 6E). We also examined the probability with which forward runs end after a period of cooling (Figure 6—figure supplement 1C–E). We found that, compared to the wild type, forward runs in RIM-ablated animals are more likely to end after a period of cooling regardless of run length (Figure 6—figure supplement 1E). Together, these analyses support a role of the RIM-dependent motor feedback in promoting a persistent forward run state.

Agent-based simulations driven by a reduced model recapitulate the role of CD feedback in positive thermotaxis

To understand how CD might sustain motor states during thermotaxis, we built a minimal dynamical systems model of the thermotaxis circuit (Figure 7). In this model, temperature fluctuations encoded by a thermosensory neuron are conveyed to a downstream interneuron. The interneuron then outputs to a motor command neuron that determines the motor state. A copy of the motor command is then relayed back to the interneuron after being weighted by a feedback gain factor (g). A positive gain factor means the motor-related signal is reinforcing to the activity of the interneuron, effectively forming a positive feedback loop. A negative gain factor translates to a negative feedback loop. We allowed the gain factor to vary between 1 (positive feedback), 0 (no feedback), and −1 (negative feedback) to test the impact of the recurrent circuit motif on circuit output.

A reduced model explains the role of corollary discharge in sustaining forward locomotion during thermotaxis.

(A) Schematic of the circuit model. (B) Dynamics of the model in response to an oscillating input stimulus. Top: temporal profile of the input signal. Middle: dynamics of the model with the feedback strength set to 1 (positive feedback), 0 (no feedback), or −1 (negative feedback). (C) Simulated trajectories of navigational behavior on a 2-D arena with linear input gradient, with feedback strength set to 1 (left), 0 (middle), and −1 (right). (D) Thermotactic biases for trajectories generated by models with feedback strength equaling 1 (positive feedback), 0 (no feedback), or −1 (negative feedback). (E) Forward run duration as a function of forward run direction for the behavioral simulations in (D). (F) Fraction of cooling epochs that were followed by the termination of forward runs in simulations with the feedback gain set to 1 (blue) or 0 (black). (G) Fraction of forward runs that ended after a period of cooling in simulations with the feedback gain set to 1 (blue) or 0 (black). Error bars are 95% CI. Wilcoxon rank-sum test *p<0.05; **p<0.01; ***p<0.001.

Figure 7—source data 1

Computational model of the thermotaxis circuit.

https://cdn.elifesciences.org/articles/68848/elife-68848-fig7-data1-v3.zip

When exposed to oscillating inputs, the positive feedback model exhibited stable high and low states in both the interneuron and the motor neuron. The autocorrelative timescale of these states (a measure of persistence) outlasted the oscillatory period of the input signal (Figure 7B). This was not the case for the models with no feedback or negative feedback where the oscillatory signal remained evident in both the interneuron and the motor neuron (Figure 7B).

We then used this circuit model to simulate animal locomotion along linear thermal gradients (Figure 7C). The positive feedback model more effectively drove migration up the thermal gradient than the model with no feedback. In contrast, the negative feedback model exhibited less effective thermotaxis than the no feedback model (Figure 7C, D). The duration of forward runs was overall significantly longer and exhibited stronger dependence of run duration on run direction in the positive feedback model compared to the alternative models (Figure 7E). Lastly, forward runs were much more likely to terminate after a period of cooling in the no feedback model compared to the positive feedback model, consistent with experimental observations in RIM-ablated animals. Thus, the positive feedback model best explained the neural activity and behavioral data. These results indicate that CD in the form of positive feedback promotes persistent circuit activity states and effective navigation during time-varying thermosensory inputs.

Discussion

We have uncovered a role for CD, a feedback signal from the motor circuit, in sustaining a neural state for forward locomotion during C. elegans thermotaxis. By relaying a copy of the motor command to a sensory processing interneuron, the thermotaxis circuit encodes a recurrent loop that integrates rapidly varying thermal inputs with more slowly varying motor state signals. This integration results in stable neural activity states that sustain the behavioral state corresponding to forward locomotion. Persistent neural activity states may enable the circuit to filter out rapid fluctuations in sensory input and promote efficient navigation in a dynamic sensory environment. During positive thermotaxis up gradients, C. elegans generates sustained periods of forward locomotion that carry it up temperature gradients; CD prevents these periods of forward locomotion from being curtailed by transient negative temperature fluctuations.

In mice and flies, recurrent circuitry has a prevalent role in persistent neural activities and behavioral states. In many systems, CD has a role in suppressing neural or behavioral responses. In contrast, we show that CD is also able to reinforce the behavioral response to a sensory input. In the C. elegans circuit for positive thermotaxis, the AIY interneuron receives the motor state signal. Because AIY is postsynaptic to many sensory neurons and is required for navigation in other modalities (Wakabayashi et al., 2004; Tsalik and Hobert, 2003; Luo et al., 2014a), CD-based feedback to AIY might play a general role in many different sensorimotor pathways.

The RIM interneuron is required for the propagation of the CD signal to AIY to promote positive thermotaxis. In a recent study of olfactory responses in immobilized animals, RIM was also shown to send feedback input to AIB, another sensory processing interneuron whose activity promotes the reversal state (Gordus et al., 2015). In the olfactory pathway, RIM activity was required to correlate the activity of AIB and the AVA premotor interneuron. Silencing RIM led to more reliable odor-evoked response in AIB and odor-induced initiation of forward runs. These findings are consistent with a model where a RIM-dependent CD couples AIB activity to the motor command signal, thereby preventing AIB and its downstream circuit from passively responding to fluctuating olfactory inputs. Thus, the effect of silencing RIM in the olfactory sensorimotor pathway mirrors the impact of RIM ablation on AIY, during positive thermotaxis, shown in this study. Together, this evidence suggests a wide role for CD in promoting persistent states in sensorimotor transformation.

Our findings add to a growing body of literature from across species that motor behavior can significantly impact sensory processing (Petreanu et al., 2009; Zagha et al., 2013; Fu et al., 2014; Schneider et al., 2014; Seelig and Jayaraman, 2015; Ouellette et al., 2018; Musall et al., 2019; Stringer et al., 2019; Salkoff et al., 2020). An explicit dependence of sensory encoding on behavioral states has been shown to contribute to variability in stimulus-evoked neural and behavioral responses (Fontanini and Katz, 2008; McGinley et al., 2015). In mice performing visual and auditory tasks, cortex-wide neural activity can be dominated by movement-related signals, many of which are uninstructed (Musall et al., 2019). Importantly, these movement-related signals closely predict the inter-trial variability in neural response. In C. elegans, whole-brain imaging in both stationary and moving animals has shown brain-wide encoding of the forward run and reversal states (Venkatachalam et al., 2016; Kato et al., 2015; Nguyen et al., 2016). The roles of the AIY interneuron studied here and the AIB interneuron studied in Gordus et al., 2015 may reflect a general use of the integration of sensory and motor-related signals during C. elegans navigation. In both cases, RIM is needed to propagate the motor-related signal to sensory processing interneurons.

In addition to its postsynaptic partners predicted by the connectome, RIM releases a diverse array of neuromodulatory signals (Taylor et al., 2019). Thus, through RIM and other neurons, the motor state signal may be broadcast to many neuron types, leading to correlated activity patterns throughout the C. elegans brain (Kaplan et al., 2018). This hypothesis may be tested with whole-brain imaging after RIM inactivation. We do not fully understand the synaptic mechanism by which CD reaches AIY. We observed that loss of tyramine, a biogenic amine produced by RIM, only partially disrupted the CD signal in AIY. Broadly disrupting biogenic amine synthesis or vesicle release from RIM also yielded similarly partial defects Figure 4—figure supplement 2.

The tyramine receptor, SER-2, is expressed in many neurons including AIY. Cell-specific perturbation or rescue of SER-2 function in AIY could help test the requirement of tyramine signaling in relaying the CD signal. Other signaling molecules are likely involved as well, and more extensive molecular and cellular dissection is needed to understand how the CD signal reaches AIY. Interestingly, a recent study demonstrated that hyperpolarization of RIM extended the duration of the forward runs during spontaneous locomotion through RIM-specific function of the gap junction protein UNC-9 (Sordillo et al., 2021). Exploring the cell-specific involvement of gap junction genes in relaying CD could also elucidate the molecular basis of feedback to AIY.

Materials and methods

Molecular biology and transgenic strain construction

Promoters

Request a detailed protocol

The following promoters were used to allow neuron-specific expression of a calcium sensor, chrimson, and miniSOG. Most were generated from genomic DNA isolated from mixed stage N2 animals. Promoters include 4.8 kb (Prig-3), 0.9 kb (Pinx-1), 5.3 kb (Pglr-1), 2.9 kb (Pcex-1), 0.86 kb (Plgc-55B), and 3.1 kb (Pnmr-1) genomic sequence. All promoters except Pnmr-1 and Plgc-55B used the genomic sequence of the respective length starting immediately upstream of the predicted ATG start codon of the respective genes. For Pnmr-1, a 2 kb internal fragment that reduces the 5.1 kb nmr-1 reporter expression was removed (Kawano et al., 2011). Details on Plgc-55B can be found in Gao et al., 2015. See Appendix 1—table 1 for the full list of constructs and transgenes used in this study.

Calcium imaging

Request a detailed protocol

For AIY calcium imaging, aeaIs003 was generated by integrating olaEx1621 [Pmod-1::GCaMP6s; Pttx-3::RFP; Punc-122::mCherry]. The integrant was outcrossed against N2 for four times to generate strain ADS003 and crossed into lite-1 to generate QW1410. As in previous studies, temperature-evoked AIY activity was reliably recorded from neurites as opposed to the soma (Clark et al., 2006; Biron et al., 2006).

For AFD calcium imaging, aeaIs004 was generated by integrating an existing Ex line [Pgcy-8::GCaMP6s; Pgcy-8::RFP; Punc-122::mCherry]. The integrant was outcrossed against N2 for four times to generate strain ADS004.

For premotor interneuron and motor neuron calcium imaging, pJH3338 was constructed for calcium imaging for premotor interneurons and head motor neurons. The GCaMP6s reporter was optimized for C. elegans and contained three C. elegans introns (Lim et al., 2016; Chen et al., 2013). GCaMP6s was fused with codon-optimized mCherry (wCherry) at the C-terminus to facilitate ratiometric measurement via simultaneous imaging of GFP and RFP. The reporter expression was driven by Pglr-1 as described above. This construct was co-injected with lin-15(+) marker to lin-15(n765) animals to generate extrachromosomal transgenic array hpEx3550 and subsequently integrated to generate hpIs471. The integrated array was outcrossed against N2 wild type four times to generate ZM8558. For simultaneous AIY and premoter/interneuron imaging, hpIs471 was crossed with aeaIs003 to generate ADS027.

Neuron ablation

Request a detailed protocol

pJH2829, pJH3311, pJH2931, pJH2890, and pJH2827 were constructed for LED-based neuronal ablation for RIM, AIB, AVA (plus other neurons), AVB (plus other neurons), and AVA/AVE/AVD/RIM/PVC (plus other neurons), respectively. miniSOG fused with an outer mitochondrial membrane tag TOMM20 (tomm20-miniSOG or mito-miniSOG) (Qi et al., 2012; Shu et al., 2011). An inter-cistronic sequence splice leader (SL2) was inserted between the coding sequence of tomm20-miniSOG and codon-optimized mCherry (wCherry; a gift of A Desai, UCSD) to visualize neurons that express miniSOG and to examine the efficacy of ablation. SL2 sequence was PCR amplified off the splice leader sequence (SL2) between gpd-2 and gpd-3. These constructs were co-injected with the lin-15(+) marker in lin-15(n765) animals to generate extrachromosomal arrays hpEx2997, hpEx3464, hpEx3072, hpEx3064, and hpEx2940, respectively. With the exception of hpEx3072, other arrays were integrated to generate hpIs327, hpIs465, hpIs331, and hpIs321. All integrated transgenic arrays were outcrossed four times against N2, except hpIs327, which was outcrossed seven times against N2, before being used for behavioral analyses or to be combined with AIY calcium imaging analyses or behavioral analyses.

AIY imaging upon neuronal ablation

Request a detailed protocol

aeaIs003 was crossed with hpIs327, hpIs321, hpEx3072, hpIs331, and hpIs465, respectively, to generate ADS010, ADS014, ADS026, ADS036, and ADS046. They were used for AIY calcium imaging upon ablation of RIM, premotor interneurons (with a few other neurons), and AIB, respectively.

AIY calcium imaging upon genetic manipulation of synaptic transmission and optogenetic stimulation

Request a detailed protocol

For AIY imaging in genetic synaptic transmission mutants, QW1408, QW1409, QW1411, QW1175, and QW1415 were generated by crossing aeaIs003 into the corresponding mutant backgrounds listed in Appendix 1—table 1.

For AIY imaging upon cell-type-specific manipulation of synaptic transmission, aeaIs003 was crossed with yxIs25, xuEx1414, and kyEx4962 to generate ADS043, ADS042, and ADS013, respectively (Li et al., 2014; Zhang et al., 2005; Gordus et al., 2015).

Chrimson (Klapoetke et al., 2014) was codon-optimized and fused at C-terminus with wCherry as described (Lim et al., 2016). Chrimson expression was driven by Plgc-55B and Prig-3 to generate pHR2 and pHR6. These constructs were co-injected with Pges-1::GFP into QW1410 to generate aeaEx003 (ADS29) and aeaEx005 (ADS31) for AIY imaging upon optogenetic stimulation of AVB and AVA, respectively.

aeaEx003 and aeaEx005 were then crossed into hpIs327;aeaIs003;lite-1 to generate ADS033 and ADS035 for AIY calcium imaging in RIM-ablated animals, upon AVB and AVA stimulation, respectively.

Behavioral assays

Positive thermotaxis assay

Request a detailed protocol

L4 animals were cultivated at 25°C the night before the assay. On the day of the experiment, the behavioral arena was allowed to equilibrate until a stable linear thermal gradient spanning 19 –23°C was established. Before each assay session, a thin layer of NGM agar sized 20 cm on each side was placed on the arena and allowed to equilibrate to the temperature of the arena. Twenty young adults were collected from their cultivation plates and briefly washed in NGM buffer before they were transferred onto the thin agar. These animals were allowed to explore the assay environment for 5 min before behavioral recording starts. Afterwards, a CMOS camera positioned above the arena recorded continuously every 500 ms for 20 min. Animal trajectories were extracted from the raw behavioral recordings using custom-written LABVIEW software. Subsequent analyses were performed in MATLAB.

Spontaneous locomotion assay

Request a detailed protocol

Animals were cultivated and prepared for behavioral assay in identical manners as for the positive thermotaxis assay. The same behavioral arena, equilibrate to room temperature (22°C), was used to assay spontaneous locomotion. Behavioral recordings were conducted the same way as in the positive thermotaxis assay. Subsequent analyses were performed using the same LABVIEW software as above and subsequently in MATLAB.

Calculation of thermotactic bias

Request a detailed protocol

For each animal, the instantaneous velocity (v) and speed (|v|) were calculated from the animal's centroid positions. The velocity vector was then projected onto direction of the thermal gradient, which in this case was parallel to the negative direction of the x-axis of the behavior arena. The thermotactic bias is the ratio between the velocity projection along the thermal gradient and the instantaneous speed of the animal:

thermotactic bias =vx|v|

Calcium imaging in semi-constrained animals

Sample preparation and imaging setup

Request a detailed protocol

L4 larval animals expressing cytosolic GCaMP6s::wCherry were cultivated at 25°C the night before the imaging experiment. Immediately before the imaging session, animals were transferred to a microscope slide with a 5% agarose pad (2 mm thick). A small drop of NGM buffer was added to the agarose pad and a #1 coverslip was lowered onto the pad. This preparation allowed the animal enough mobility to execute head and (partial) body oscillations characteristic of forward runs and reversals. Under this preparation, the animal exhibits slow, local displacements, but cannot fully leave the field of view. Calcium imaging was performed on an upright spinning disc confocal microscope (Nikon Eclipse LV100 and Yokogawa CSU22) and iXon3 DU-897 EMCCD camera (Andor). High-resolution images were collected through a 40×, 0.95 NA Nikon Plan Apo lambda objective. 3D volumetric stacks were acquired in both the green (GCaMP6s) and red (wCherry) channels with an exposure of 30 ms at approximately 1.2 volumes per second.

Control of thermal stimulation

Request a detailed protocol

Animals were imaged on a custom-built temperature control stage where a PID controller and H-bridge amplifier (Accuthermo) drove a thermoelectric cooler (TEC) (Newark) that pumped heat into and out of a thin copper plate with a liquid-cooled water block (Swiftech) acting as a thermal reservoir. A type-T thermocouple microprobe (Physitemp) was placed on the copper plate underneath a thin steel tab. A custom written Labview program was used to specify the desired temperature waveform.

Extraction of calcium transient levels

Request a detailed protocol

To extract fluorescence intensities for individual neurons, we identified connected regions above a predefined intensity threshold and registered these regions of interest (ROI) across a movie based on spatial proximity across frames. For each ROI, we computed mean intensity of the top 30 pixels in both the green (GCaMP6s) and the red (wCherry) channel. The instantaneous activity of each neuron was computed using the following equation:

ΔR(t)/R0=(R(t)-R0)/R0

R is the ratio between the mean intensity value in the green channel and that of the red channel. R0 represents the lowest 1st percentile of R(t) values in the time series.

Event-triggered averages (ETAs) of neural activity

Request a detailed protocol

AIY activity was extracted from a defined window of time spanning the event interest (i.e., onset of thermal stimuli, transition between the forward run and reversal states, onset and offset of AVA activation). For analysis in Figure 3D, AVA ON and OFF states were defined by thresholding AVA activity using the Otsu method, which was implemented in MATLAB using the multithresh function. Data across multiple epochs of the same event were concatenated into matrices and presented as heat maps throughout the paper. Average ETA profiles were computed by averaging data from that precede or lag the event of interest by the same number of time points.

Duration of AIY activation

Request a detailed protocol

AIY activity data from each imaging session was fit using a Gaussian mixture models (see 'Statistical analysis' for implementation details). The number of Gaussian components was determined by iteratively fitting models with component number k = 1,2,3,…,6 and assessing the quality of the fit using the Bayesian information criterion (BIC). As k increases, the BIC value typically drops sharply from k = 2 to k = 3 and varies little after that. For wild-type data, models with k = 3 consistently capture the baseline AIY activities during reversals, the heightened activities during forward runs, and the large transients that occur at the onset of the forward runs or in response to warming stimuli. We thus defined the intersection point of the two Gaussian components with the lowest and the second lowest mean as the threshold above AIY is considered to be in the activated state. We then use this threshold value to binarize the AIY activity time series and compute the duration of each bout of AIY activation.

Optogenetic stimulation and simultaneous calcium imaging

Request a detailed protocol

Experimental animals expressing Chrimson were grown on NGM plates supplied with 5 µM all-trans retinal (ATR) mixed with OP50 bacteria. Control animals of the same genotypes were grown on NGM plates seeded with OP50 without ATR. The day before the experiment L4 animals were picked onto fresh plates (with ATR for the experimental groups and without ATR for the control groups). On the day of the experiment, young adult animals were prepared for imaging in the semi-constrained preparation as described above. During imaging, pulses of red light were delivered from a filtered white LED lamp. Pulse timing was controlled by MATLAB scripts. For calcium imaging, animals were illuminated with only the blue laser (488 nm) to avoid strong activation of Chrimson.

Neuron ablation

Request a detailed protocol

Transgenic animals expressing miniSOG were collected from late L1 to L2 stage onto a small NGM plate (3.5 cm diameter). The plate was placed under a blue LED spotlight (Mightex, peak wavelength 617 nm) for 40 min. Following illumination, the animals were allowed to recover for overnight at 15°C to examine the disappearance of cells. All ablation was performed using animals that carried integrated miniSOG transgens, with the exception for AVA ablation. Ablation of AVA was carried out in animals that carried an extrachromsomal array for Prig-3-miniSOG-SL2-RFP, which was subjected to random loss during somatic division. Animals used for ablation were selected for those that did not show expression (hence ablation) in a pharyngeal neuron that affects the survival of ablated animals.

Statistical analysis

Statistical tests

Request a detailed protocol

The Wilcoxon rank-sum test was used in the following comparisons: (1) comparing calcium activity upon the initiation of forward runs or reversals between wild-type animals and various neuron-ablation experiments, (2) comparing the probability of change in AIY activity upon the initiation of forward run or reversals between wild-type and AIY::TeTX animals, and (3) comparing the thermotactic bias between wild-type and RIM-ablated animals. To control for multiple comparison, p values were adjusted using the Benjamini–Hochberg correction. 95% confidence intervals were determined by bootstrapping.

Gaussian mixture model

Request a detailed protocol

Gaussian mixture models were fit to AIY activity distributions from each independent dataset using the fitgmdist function in MATLAB. The initial cluster centers were generated through the k-means++ algorithm, and the initial mixing proportions were set to uniform.

Modeling of circuit activity and behavior

Neural circuit model

Request a detailed protocol

We use a minimal model to capture the interaction between the key components of the thermotaxis circuit:

(1) Thermosensory neuron (AFD) V0:τ0dV0dt=gL0(V0VL0)+IinputR
(2) Interneuron (AIY)V1:τ1dV1dt=gL1(V1VL1)+F10V0+F12V2
(3) Motor command neurons V2:τ2dV2dt=gL2(V2VL2)+F21V1

where gL1, gL2, and gL3 are leak conductances and are non-negative. VL1, VL2, and VL3 are the resting potentials. Synaptic interactions are modeled as linear or sigmoidal functions:

(4) F10(V0)=w0V0
(5) F12(V2)=wfb(11+ek1×(V2β1))
(6) F21(V1)=w21(γ11+ek2×(V1β2))(γ21+ek3×(V1β2))

where w0, wfb, and w21 are the network weights, and gi, ki, and βi define the height, steepness, and inflection point of sigmoidal functions. The two terms in F32 represent separate groups of premotor interneurons that promote forward runs (e.g., AVB) or reversals (e.g., AVA). Equation 6 essentially performs a max operation between the two terms to determine whether the motor output favors forward runs (F32>0) or reversals (F32<0).

The above model is further simplified by setting AFD activity to its steady-state value, V0(t)V0=VL0+IinputRgL0, which reduces the model to two dimensions:

(7) τ1dV1dt=gL1V1+αIinput+F12(V2)+C1
(8) τ2dV2dt=gL2V2++F21(V1)+C2

where α=w0RgL1, C1=gL1VL1+VL0, and C2=gL2VL2.

Based on dynamical systems theory (Strogatz, 2015), the above model has two distinct stable states as long as their nullclines intersect at least three times, which can be achieved by a wide range of parameter values. Table 1 lists the parameter values that were used for network simulations in Figure 7B. Note that wfb is varied between −1, 0, and 1, corresponding to negative feedback, no feedback, and positive feedback networks.

Table 1
Parameter values used in Figure 7B.
ParameterValue
τ12/3
τ22/3
gL11
gL21
w01
wfb−1, 0, or 1
w211
k115
k25
k35
β10
β21.5
β21.5
γ11
γ21
C10.5
C20

Simulation of thermotaxis behavior

Request a detailed protocol

We used the same network model described above to drive agent behavior. The locomotory state M(t) of the animal is determined by the activity of the motor command neuron:

M(t)=1, if V2>0agent executes forward run
M(t)=0, if V2<0agent executes reversal

At the start of each forward run, the new heading direction is chosen randomly from a uniform distribution with range [-180°, 180°). During an ongoing forward run or reversal, the heading direction, θ(t), was kept constant. When a forward run ends and a reversal state starts, the heading direction changes by 180°:

(9) θ(t)={(cosθ0,sinθ0)t=0θ(tdt)M(t)=M(tdt)θ(tdt)M(t)=1 and M(tdt)=1(cosθ0,sinθ0)M(t)=1 and M(tdt)=1

These simple behavioral rules allows us to specifically model the biased random walk component of thermotaxis, while leaving the RIM-independent klinotaxis component out of the analysis.

All agents are simulated to move at constant speed (one unit length per time step) on a two-dimensional linear thermal gradient. The gradient is set to lie along the x-axis: T(x)=cTx, where x represents the x coordinate. Since AFD is known to sense temporal changes in temperature, the input current to AFD evoked by thermal stimuli is defined by Iinput(t)=cT(x(t)-x(t-Δt)), where x(t) is the instantaneous x-position of the agent.

Each simulation is initialized by setting an starting position of (0,0), an initial heading angle drawn from the uniform distribution from [−180°, 180°), an initial network state of (V1(t0)=1, V2(t0)=1), and with the animal in a forward run state. Upon numerical integration, simulated worms move autonomously in their environment for a predetermined duration (tmax).

Appendix 1

Appendix 1—table 1
Constructs and transgenic arrays.
Calcium imaging
PlasmidInjection markerTransgeneStrain
pDACR1286[Pmod-1::GCaMP6s] (25 ng/µl); pDACR63[Pttx-3::mCherry ] (25 ng/µl)pDACR218[Punc-122::dsRed] (40 ng/µl)aeaIs003(AIY) (integrated olaEx1621*)ADS003
pDACR943[Pgcy-8::GCaMP6s] (30 ng/µl);pDACR801 [Pgcy-8::mCherry] (5 ng/µl)pDACR20[Punc122::GFP] (20 ng/µl)aeaIs004 (AFD) (integrated olaEx1527)*ADS004
pJH3338[Pglr-1-GCaMP6s::wCherry]pL15EKlin-15AB genomic DNA ( 20 ng/µl)hpIs471 (premotor/motor)ZM8558
Optogenetic stimulation
pHR2[Plgc-55B-Chrimson::wCherry]pL15EK[lin-15AB genomic DNA] (80 ng/µl)aeaEx003 (AVB/others)ADS029
pHR6[Prig-3-Chrimson::wCherry]pL15EK[lin-15AB genomic DNA] (80 ng/µl)aeaEx005 (AVA/others)ADS031
Cell ablation
pJH2829[Pcex-1- MiniSOG::SL2::wCherry]pL15EK[lin-15AB genomic DNA] (20 ng/µl)hpIs327 (RIM)ZM7978
pJH3311[Pinx-1- MiniSOG::SL2::wCherry]pL15EK[lin-15AB genomic DNA] (20 ng/µl)hpIs465(AIB)ZM8484
pJH2931[Prig-3- MiniSOG::SL2::wCherry]pL15EK[lin-15AB genomic DNA] (20 ng/µl)hpEx3072 (AVA/others)ZM7198
pJH2890[Plgc-55B- MiniSOG::SL2::wCherry]pL15EK[lin-15AB genomic DNA] (20 ng/µl)hpIs331(AVB/others)ZM7297
pJH2890[Pnmr-1-MiniSOG::SL2::wCherry]pL15EK[lin-15AB genomic DNA] (20 ng/µl)hpIs321(AVA/E/D/RIM/PVC/others)ZM7054
Synaptic manipulation
Pttx-3::TeTx::mCherry Zhang et al., 2005yxIs25 (AIY)ZC1952
Ptdc-1::TeTx::mCherry Gordus et al., 2015kyEx4962 (RIM/RIC)CX14993
  1. *Gift of Daniel Colon-Ramos.

Appendix 2

Appendix 2—table 1
Strains.
For thermotaxis and locomotion assays
StrainGenotypePurposeFigure
Bristol N2Wild typeWild-type behaviorFigure 1A–D
ZM7978hpIs327Behavior upon RIM ablationFigure 6
Calcium imaging
ADS003aeasIs003AIY imagingFigures 2E–I, 3A, B
ADS004aeaIs004AFD imagingFigure 2C, D
ADS027aeaIs003; hpIs471Simultaneous imaging of AIY, AVA, RME, SMDD, SMDV, and RIMFigure 2B, E
ADS043aeaIs003; yxIs25AIY imaging, upon blockade of AIY chemical transmissionFigure 3B
ADS010aeaIs003; hpIs327AIY imaging, upon ablation of RIMFigures 4A, B, 5A, C
ADS014aeaIs003; hpIs321AIY imaging, upon ablation of RIM, AVA, AVE, AVD, and PVCFigure 4A
ADS026aeaIs003; hpEx3072AIY imaging upon ablation of AVAFigure 4A
ADS036aeaIs003; hpIs331AIY imaging, upon ablation of AVBFigure 4A
ADS046aeaIs003; hpIs465AIY imaging, upon ablation of AIBFigure 4A, B
ADS029aeaEx003; aeaIs003; lite-1(ce314)AIY imaging upon optogenetic stimulation of AVBFigure 4C
ADS031aeaEx005; aeaIs003; lite-1(ce314)AIY imaging upon optogenetic stimulation of AVAFigure 4E
ADS033aeaEx005; aeaIs003; hpIs327; lite-1(ce314)AIY imaging, upon RIM ablation and AVA stimulationFigure 4E
ADS035aeaEx003; aeaIs003; hpIs327; lite-1(ce314)AIY imaging, upon RIM ablation and AVB stimulationFigure 4C
ADS013aeaIs003; kyEx4962AIY imaging, upon disruption of RIM/RIC chemical transmissionFigure 4—figure supplement 2
ADS006aeaIs003;tdc-1(n3419)AIY imaging in tyramine/octopamine synthesis mutantFigure 4—figure supplement 2
QW1411aeaIs003; eat-4(ky5)AIY imaging in glutamate mutantFigure 4—figure supplement 2
QW1175aeaIs003; unc-31(e928)AIY imaging in dense core vesicle release mutantFigure 4—figure supplement 2
QW1408aeaIs003; cat-1(e1111)AIY imaging in biogenic amine transporter mutantFigure 4—figure supplement 2

Data availability

All data generated or analysed during this study are included in the manuscript and supporting files. Source data files have been provided for Figures 1-6. Source code has been provided for Figure 7.

References

    1. Holst E
    2. Mittelstaedt H
    (1950)
    The principle of reafference: interactions between the central nervous system and the peripheral organs
    Die Naturwissenschften 37:464–476.
  1. Book
    1. Strogatz SHSH
    (2015)
    Nonlinear dynamics and chaos : with applications to physics, biology, chemistry, and engineering
    Boulder, CO: Westview Press.
    1. White JG
    2. Southgate E
    3. Thomson JN
    4. Brenner S
    (1986) The structure of the nervous system of the nematode Caenorhabditis elegans
    Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences 314:1–340.
    https://doi.org/10.1098/rstb.1986.0056

Decision letter

  1. Manuel Zimmer
    Reviewing Editor; Research Institute of Molecular Pathology, Vienna Biocenter and University of Vienna, Austria
  2. Ronald L Calabrese
    Senior Editor; Emory University, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

In this manuscript Ji and colleagues report that information processing in a primary sensory interneuron of C. elegans is modulated by behavioral state: the AIY neuron encodes external stimuli relayed from thermo-sensory neurons exclusively during forward crawling but not during backward crawling. This modulation depends on the reversal interneuron RIM. The authors report evidence that this interaction is a corollary discharge, i.e. motor command copy, to stabilize the forward crawling state in the presence of transient aversive thermal fluctuations; thereby, corollary discharge prevents erratic forward-backward switches supporting robust thermotaxis. Both sensory and behavioral modulation of AIY was shown in previous studies, but it remained puzzling how these different inputs are utilized simultaneously in the context of a behaving animal. Moreover, the function of why a primary sensory interneuron is so strongly modulated by behavior was unclear. Similar phenomena can be observed in other model organisms ranging from worms, flies and mice but the underlying mechanisms and functions are still unclear. The current manuscript sheds more light in these directions showing that sensory processing even at its earliest stages is instantaneously modulated by behavior to support robust navigation.

Decision letter after peer review:

[Editors’ note: the authors submitted for reconsideration following the decision after peer review. What follows is the decision letter after the first round of review.]

Thank you for submitting your work entitled "Corollary Discharge Promotes a Sustained Motor State in a Neural Circuit for Navigation" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by a Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: M Eugenia Chiappe (Reviewer #3).

Our decision has been reached after a detailed follow-up consultation between the reviewers. Based on these discussions and the individual reviews below, we are convinced that your manuscript is potentially very important and a strong candidate for publication in eLife. Therefore, we reject your manuscript now but encourage you to address the concerns and resubmit at a later stage. If you decide to re-submit the paper, please provide a point-by-point response to this letter and the reviewer's comments.

Please pay particular attention to some of the major concerns that came-up also in the consultation session, which we summarize here:

1. The reviewers are concerned about how unconstrained the animals are in your assays. Based on the data you provide, they conclude that the animals must have been recorded under almost immobile conditions. Since your major conclusions are contingent on activity recordings made in freely behaving animals this caveat needs to be addressed by (a) showing data demonstrating unconstrained movement, i.e. locomotion speed and /or posture kymographs and (b), eventually provide additional activity data obtained from sufficiently unconstrained animals. It might be sufficient to record from individual neurons, like AIY, as long as this permits more natural movement and confirms your major results.

2. The ethological model lacks sufficient support by your data and analyses. The alternative model of an efference copy from RIM simply inhibiting AIY during reversals seems very likely. However, you propose an unexplained sign-inversion in the interactions between RIM and AIY, and it is unclear to us how such an interaction can lead to a persistent feedback gain from the motor circuit to AIY, subsequently maintaining forward motor state. To address this, you could show more directly that in RIM(-) animals reversal initiations are tightly locked to sensory input. These data should be retrievable from your thermotaxis assays and AFD/AIY imaging data, including the suggested new experiments above.

3. The reviewers raise concerns that some experiments were not performed with sufficient repetitions and lack statistics, therefore require more validation.

4. In the discussion, we agreed that if the ethological model finds better support in a new manuscript, deciphering the molecular pathway in more detail suggested by reviewer #2, might be of less priority. However, alternatively you could tone down the major statements of corollary discharge function and focus your studies on a more complete molecular characterization of RIM-AIY interactions.

Reviewer #1:

In this manuscript Ji and colleagues report that a primary sensory interneuron, AIY, is modulated by behavioral state (forward versus reverse crawling), which depends on the reversal interneuron RIM. The authors suggest that this interaction is a corollary discharge, i.e. motor command copy, to stabilize the forward crawling state in the presence of transient aversive sensory fluctuations during thermotaxis. Both sensory and behavioral modulation of AIY was shown in previous studies, but it remained puzzling how these inputs converge onto AIY and which circuits are involved. Moreover, the function of why a primary sensory interneuron might be so strongly modulated by behavior was unclear. Similar phenomena were observed in other model organisms ranging from worms, flies and mice and hence currently draw a lot of attention in the neuroscience community since mechanisms and functions are still pretty unclear; therefore, this work is potentially very important, and I am very enthusiastic about the main findings. However, at the current stage the study is very preliminary and suffering in various experiments from insufficient repetitions, lack of proper analyses and statistics. While my comments below are extensive, they require merely more experimental repetitions and additional analyses. Addressing them is essential for publication, but I believe this can be done within reasonable time.

1. Figure 1: during positive thermotaxis it is not surprising that animals experience short negative dT/dt fluctuations while performing forward runs. This is indeed expected, and animals can easily deal with this, both considering the classical random biased walk strategy (since switch to reversal in this model is probabilistic, a fraction of episodes of negative dT/dt will not elicit terminating the forward state). To convince readers of this manuscript that there is indeed an interesting and surprising observation in this data, the authors should perform some additional analyses showing how episodes of continuous negative and positive dT/dt are distributed in their duration, e.g. in a 2D histogram. The authors then should help the reader not familiar with the literature and refer to the reports showing that thermosensory neurons are indeed sensitive in such an operating range.

Moreover, an analysis should be performed showing that episodes of negative dT/dt occur indeed throughout forward runs, not just at the beginning or at the end.

2. Inspecting the Ca++-imaging data in moving animals (e.g. Figure 2), I conclude that these animals must be crawling at extremely slow rates. Recent work showed that SMD neuronal activity is tightly locked to head bending phase (e.g. (Hendricks et al., 2012; Kaplan et al., 2019; Yeon et al., 2018)), which in this paper appears to be as slow as 2min (see Figure 2D, 5A). This is 1-2 orders of magnitude slower than freely crawling worms under standard laboratory conditions (e.g. compare to SMD imaging data in ref (Kaplan et al., 2019) or in semi-restrained animals (Hendricks et al., 2012)). Moreover, AVA and RIM Ca++ traces in this study exhibit minutes-lasting plateau states, which is characteristic for immobilized animals and typically not seen in freely crawling animals (see (Kato et al., 2015) for a comparison of these neurons between immobilized and freely crawling conditions). In conclusion, animals in this study must be nearly immobilized. I assume this is because animals have been placed between a coverglass and a 5% agarose pad, the latter having an unusual high concentration. To a non-expert eLife reader these caveats are quickly overseen.

2a. The authors should provide data about crawling speed in these experiments.

2b. The authors should provide an in-depth explanation why they chose such extreme conditions; perhaps to avoid movement artefacts given the slow 1Hz acquisition rate? Or is AIY modulation only detectable in these extreme conditions with strong and prolonged RIM activity phases?

2c. Were thermotaxis assays in Figure 1 performed under these conditions (coverglass, 5% agarose)? Are animals able to perform thermotaxis under these conditions?

3. While AFD activity traces in Figure 2A look convincing, additional analysis is required for AIY and other inter- motor-neurons to demonstrate sensory responses and gating by behavioral state. The peak in power spectrum alone does not suffice to conclude that AIY etc. activity is stimulus locked. The AIY activity trace in Figure 2—figure supplement 1 from immobilized unstimulated worms indicates interesting dynamics in AIY, that certainly would appear as a peak in a power spectrum.

3a. I suggest showing phase aligned averages of the neurons and a statistical test showing that their activity gets significantly entrained.

3b. How do power spectra and phase-averages differ in forward versus reverse?

3c. Why does AIY show a peak at 0.067 and not 0.033Hz, like AFD?

3d. An equivalent control dataset from non-stimulated animals is essential.

4. The finding that AIY is modulated by reversal neuron RIM in immobilized worms is crucial for the main conclusion of this paper, since it distinguished corollary discharge / efference copy from behavioral feedback. However, this statement is based only on a single observation shown in Figure 2—figure supplement 1 and lacks any quantification. Is AIY activity different in RIM high versus low states? This should be scrutinized via an analysis like it was done for reversal states in Figure 3; with multiple animal repetitions and appropriate statistics. In direct comparison to freely moving worms, this would be an important result deserving a spot in a main Figure. Without this extra data and analysis, the authors' hypothesis that AIY modulation is corollary discharge / efference copy versus behavioral feedback is not sufficiently supported.

5. I am wondering how reversal and forward states were annotated in AVA or AVB ablated animals respectively, which should lack these behaviors according to a large body of literature. If AVA and AVB were successfully ablated, the analysis in Figure 4B should not be possible. Also, ablation of RIM and AIB can affect behavioral state durations and switching frequencies. The authors should provide these data along with evidence that the cells were indeed specifically ablated, thus they must validate and characterize all of their ablation strains. Data like in Figure 4C,D should be shown as well form the other ablations.

6a. Data shown in Figure 4E come from partially very low numbers of independent experiments and no statistics are provided. These shortcomings must be addressed.

6b. Moreover, stimuli were applied at very different pre-stimulus AIY activity baselines. More data would definitely strengthen this result substantially.

7. I find the result that RIM and SMD neurons show stimulus evoked activity fluctuations in RIM(-) animals very interesting, but this result needs more substantial quantifications (see comment (3)).

8. Figure 6: the effect of RIM ablation on thermotaxis is very interesting but there are many ways how this manipulation could affect the behavior. Additional analysis of the data could provide more compelling evidence supporting their model: consistent with their model run durations are decreased (Figure 6E). If the function of RIM corollary discharge is indeed to maintain prolonged forward states in spite of negative dT/dt epochs, run terminations should be coinciding with these events, but not in control animals. See also comment (1) for analyses suggestions.

9. Modulation of primary sensory circuits by behavioral state seems to extend to many other neurons in the worm brain (Kato et al., 2015), and this seems to be a more general phenomenon seen in other organisms (e.g. Aimon et al., 2019; Musall et al., 2019; Salkoff et al., 2019; Stringer et al., 2019). See also for a recent discussion (Kaplan et al., 2018). The current manuscript, in my opinion, is pretty important providing data hinting at a function of these phenomena. This is just my recommendation, but I think the authors could make more impact with their story by discussing their findings in the context of these recent discoveries.

Reviewer #2:

Ni Ji and colleagues apply quantitative analysis of behavior and mutli-neuronal calcium imaging to identify and examine a corollary discharge circuit that filters sensory input based on motor state during thermotaxis. The main claims of the paper are that RIM reports this motor state (the corollary discharge signal) to the first layer interneuron AIY, which acts as a sensory filter to ignore temperature fluctuations encoded by AFD during reversals:

1. AIY encodes temperature and motor state information

2. Motor state representations in AIY arise from a corollary discharge signal originating from RIM.

3. CD allows positive feedback to sustain forward locomotion

The first and second claims are well-supported, while the ethological model is less well-developed and was not rigorously tested. Issues with the data, particularly what seem like unrealistically long reversal states make some results difficult to interpret.

Overall, I am convinced by the data showing that RIM broadcasts a motor state signal (the CD) that filters or modulates upstream sensory pathways. This is a nice result but given the systemic effects of RIM/tyramine on sensory neurons and interneurons and the prior results showing that RIM ablation affects reversal rates even under isotropic conditions, I don't think the argument that it is supporting efficient thermotaxis is very well supported. Identifying the relevant receptor on AIY would bring better resolution to the circuit analysis.

1. The behavior and activity patterns look unusual for locomoting animals. 2A shows reversals that last 30-60 seconds, typical reversals during navigation are <5s. I looked at the dataset and indeed the F/R ethogram in 2A is incorrect-there are many more state changes than represented in the figure. However, the state durations for reversals are still very long (median >30s for reversals, just under 30s for forward). It does not seem like this can be correct for a freely moving animal and is inconsistent with the tracks shown in 1A.

2. As the authors note, SMDD/V activity has been reported to match head movements (Hendricks et al., 2012; Yeon, Kim et al., 2018; Kaplan, Thula et al., 2019), and so in panel 2D I would expect oscillatory activity that matches the forward gait in the traces and appears at ~1Hz in the power spectrum. As in 2A (and throughout), the extremely long "reversal" state durations look more typical of restrained animals.

3. Figure 2A and 2D. Do the colors in the activity histogram correspond to forward and reverse states? If so this should be indicated and perhaps match the color used for the ethogram. I am not sure having the Ca++ traces superimposed on the stimulus indicator and ethogram improves clarity, but it's fine.

4. The power spectra are too cramped and low-res in Figure 2D to be legible. It would be useful to compare spectra (or perform temp-AIY cross correlations) from stimulus cycles that occur during reversals vs forward epochs, particularly to support the claim about AIY.

5. How many animals are represented in the summary data (histograms/spectra) in Figure 2? Statistical tests on summary data should be reported.

6. Because AIY shows discrete Ca++ peaks, a quantification of warming cycles that do and do not produce an AIY response (as a binary output) in forward and reverse states might be more clear or intuitive than the (also small and cramped) histogram.

7. There is a formatting error in the legend of 2A (fstim, stim should be all subscript)

8. Label forward and reverse states in panel 2D as in 2A.

9. What do the colors mean in 2B?

10. I think the standard wiring diagram does not have an EJ between AFD and AIZ, as shown in 2B and 4A (though these authors may know better)?

11. Figure 2—figure supplement 1. Same questions as above regarding the very long reversals and lack of SMDD activity matching forward gait. The "right" side of panel A is labeled panel B and panel B as described in the legend and line 141 (neuronal cross correlations with RIM) is missing.

12. Figure 3. Consider using a perceptually uniform color map for heat maps (here and elsewhere).

13. Spell out corollary discharge in the Figure 3 legend.

14. While AVA correlates perfectly with reversals, it is less clear how well RIM does under different conditions (Gray et al., 2005; Guo et al., 2009; Piggott et al., 2011, Gordus et al., 2015)..… AVA activation/reversals can happen while RIM is inactive, and RIM activation can occur in the absence of a reversal, and in fact can lower the probability of a stimulus-evoked reversal (Gordus), while ablating RIM can increase reversal frequency (Gray, Piggott). While RIM Ca++ does correlate pretty well with reversal, and it's clear here that it is inducing sensory filtering at AIY, it's possible that it is not exclusively or necessarily reversals. This should be considered.

15. LGC-55 and TYRA-2 are expressed in AIB and AIZ; SER-2 and TYRA-2 are expressed in AIY (https://cengen.shinyapps.io/SCeNGEA/). Given that SER-2 mediates inhibition of synaptic transmission in GABAergic motor neurons (as some of these authors showed), it seems a promising receptor to mediate RIM's inhibitory effect. Was this checked? Identifying the AIY receptor would resolve some points below.

16. The principle claim of the paper is that the CD signal from RIM communicates a motor state to AIY to filter AFD input during reversals. If the effect is in large part tyramine-mediated, there are receptors for this modulator on many or most of the sensory neurons upstream of AIY, as well as on its interneuron synaptic partners. Are AFD responses normal during manipulation of RIM? Could other sensory neurons have altered responses to temperature that impact AIY? It would be show that a tyramine receptor acts in AIY to directly respond to RIM signaling (probably SER-2) and show that expression in AIY is necessary/sufficient to filter AFD signaling during reversals.

17. Should indicate color coding for WT and RIM- in 6E/F in the panel.

18. Previous results showing that ablating RIM can increase reversal frequency under some conditions, including an isotropic environment (Gray, Piggott), complicate the interpretation here that shorter run lengths in RIM-ablated animals under thermal fluctuation are due to the sensory filtering observed in AIY. Do RIM- animals reverse more under these conditions in the absence of thermal stimulation? While the need to maintain forward movement during fluctuations is clear, it is less clear why an animal (or AIY at least) would need to ignore AFD input during reversals. I'm particularly concerned about this aspect given the unusually long reversal states observed here.

Reviewer #3:

Here, Ji and colleagues study a sensory-driven navigational circuit and ask how sustained locomotor-related signals arise from self-generated, rapid changes in sensory stimuli driven by the navigational strategy of the worm to successfully reach its goal. The question on the relation among sensory cues, internally generated signals, and motor strategies, is poorly understood for any moving animal. Therefore, I consider the work very timely and important for a wide neuroscience audience.

I think this work is important and heroic: examining the presence and role of internal signals is no easy duty, and I think the authors are taking advantage of the model system beautifully. The propose role for a CD signal presented here is a very interesting and exciting idea, and consistent with the effect on behavior on the ablation of RIM (shown to be the source of motor-related signals of the AIY interneuron). However, the authors present some statements that I am not sure are shown directly from their experimental data. The concern is really about the gap between the description of the activity in AIY, a convincing metric of the presence of a persistent motor activity, and the interpretations of the data by the authors. While RIM ablation induces bleed through of sensory signals in motor cells, this effect may be due to the connections between RIM and motor cells per se rather than a consequence of the presence of a positive feedback CD signal at AIY level.

Through a phenomenological model, which was inspired by their experiments (combining analysis of behavior, physiology, and activity manipulations), they describe how the presence of a positive feedback CD signal can account for a persistent motor-related signal at an interneuron of the circuit, and the biased behavior of the worm. While this is a very exciting idea, I am not convinced that the experimental evidence supports this view clearly.

1. The authors show the presence of an internally generated signal at AIY reporting motor state, and convincingly show that RIM activity is required for this signal.

2. However, the increase of AIY activity at onset of Forward runs, and it decrease at onset of Reversals can be equally explained by either the a conventional efference copy (EC)-like signal that would be associated with reversals (since AIY is part of the forward pathway) or the proposed positive feedback CD signal (PF-CD) that would be associated with forward runs, and induce persistent activity in AIY. I see no effort in the manuscript to distinguish between these two possibilities. An EC like signal can also explained the increase in sensory cues in the response profile of AIY under oscillating temperature cues in the absence of activity in RIM (Figure 6A).

3. Optogenetic experiments are more consistent with an inhibition from reversal pathways (EC-like phenomena) than an excitation signal from forward runs (PF-CD like phenomena) (Figure 4). figure 4E, I really cannot see the effect the authors describe for the manipulation of AVB (either in the presence or absence of a functional RIM), whereas the description of the effect of the manipulation of AVA is closer to what we see in the data.

4. There is a missing metric for persistent activity in AIY, or a clear explanation of whether bimodality would be such metric. If this is the case, the authors make a very weak argument on how such a metric would measure RIM-dependent persistent motor-sate activity in AIY. If I understood correctly, bimodality should represent a more persistent activity at a particular motor state. But it is not very clear how bimodal the AIY activity is under the presence vs the absence of RIM activity, for example, if one compares figure 5A with Figure 2 Suppl Figure 1 (left column, mobile animals at constant T). Therefore, perhaps it is important to show the corresponding distribution for the supplementary figure to be able to compare the critical metric that is then quantified in Figure 5B.

5. Perhaps they could use the model to predict the effect on behavior of an EF vs PF_CD function. While the latter was examined, it is unclear how the worm behavior departs from a model under an EC signal at AIY

6. I suggest the authors make less of strong statements when the experimental evidence is not shown directly (for example, in line 235, they describe that "motor-related signals and Temperature-related signals reinforce each other": this is only shown through the model).

7. An exciting observation is the bleed through of sensory signals to motor cells within the circuit, and what potential impact these may convey at the level of the behavior of the worm (Figure 6). AIY has increased sensory components, but the lack of an EC could also induce such an effect, for example, by extending the time the neuron responds to T fluctuations.

8. What is the impact of making AIY insensitive to the motor state on other postsynaptic elements of the circuit? The authors show that motor cells themselves respond to sensory fluctuations likely affecting the persistent behavioral state. But is this a consequence of the lack of motor state sensitivity of AIY or a consequence of the lack of activity of RIM, a neuron that is presynaptic to motor cells? In conveying the idea of a PF-CD signal, it is important to distinguish these scenarios. A neuron that integrates information from AIY and AFD, and that is presynaptic to motor cells may be an ideal candidate. For example, AIB is very well suited for this distinction. If AIY activity contained both a persistent signature of motor state and thermal signals, then AIB sensory information should be modulated by the AIY's indirect input. On the other hand, if the motor state modulation in AIY is related to an EC signal, AIB cell sensory activity should not be strongly modulated by the lack of such EC signal (given electrical coupling of AIB and AFD). If this is not a possible experiment, then the strength of the statements should be adjusted accordingly, and the discussion on the presence of a conventional EC signal should be present.

https://doi.org/10.7554/eLife.68848.sa1

Author response

[Editors’ note: the authors resubmitted a revised version of the paper for consideration. What follows is the authors’ response to the first round of review.]

Reviewer #1:

In this manuscript Ji and colleagues report that a primary sensory interneuron, AIY, is modulated by behavioral state (forward versus reverse crawling), which depends on the reversal interneuron RIM. The authors suggest that this interaction is a corollary discharge, i.e. motor command copy, to stabilize the forward crawling state in the presence of transient aversive sensory fluctuations during thermotaxis. Both sensory and behavioral modulation of AIY was shown in previous studies, but it remained puzzling how these inputs converge onto AIY and which circuits are involved. Moreover, the function of why a primary sensory interneuron might be so strongly modulated by behavior was unclear. Similar phenomena were observed in other model organisms ranging from worms, flies and mice and hence currently draw a lot of attention in the neuroscience community since mechanisms and functions are still pretty unclear; therefore, this work is potentially very important, and I am very enthusiastic about the main findings. However, at the current stage the study is very preliminary and suffering in various experiments from insufficient repetitions, lack of proper analyses and statistics. While my comments below are extensive, they require merely more experimental repetitions and additional analyses. Addressing them is essential for publication, but I believe this can be done within reasonable time.

We thank the reviewer for recognizing the significance of our study. As suggested, we have substantially augmented our datasets and added analyses that strengthen our conclusions.

1. Figure 1: during positive thermotaxis it is not surprising that animals experience short negative dT/dt fluctuations while performing forward runs. This is indeed expected and animals can easily deal with this, both considering the classical random biased walk strategy (since switch to reversal in this model is probabilistic, a fraction of episodes of negative dT/dt will not elicit terminating the forward state). To convince readers of this manuscript that there is indeed an interesting and surprising observation in this data, the authors should perform some additional analyses showing how episodes of continuous negative and positive dT/dt are distributed in their duration, e.g. in a 2D histogram. The authors then should help the reader not familiar with the literature and refer to the reports showing that thermosensory neurons are indeed sensitive in such an operating range.

Moreover, an analysis should be performed showing that episodes of negative dT/dt occur indeed throughout forward runs, not just at the beginning or at the end.

We agree. In any biased random walk, an animal will experience thermal fluctuations. With the minimal rules of a biased random walk, an animal would be able to migrate in the correct direction. We assert that a biased random walk can work better with an additional feature. If runs that happen to be pointed in the right direction are insulated from thermal fluctuations, they would not end prematurely and navigation would improve. This is borne out by numerical simulation and phenomenological modeling.

As requested, we have more extensively analyzed the trajectories of animals performing thermotaxis. These data indicate that the worm does suppress its sensitivity to thermal fluctuations, enabling the animal to sustain runs that carry it in the correct direction.

As now shown in Figure 1-Supplement 1, over half of forward runs exhibited by wild-type animals involved at least one cooling episode with a minimum drop in temperature exceeding 0.001 C that did not occur at the beginning or the end of the run (Figure 1-Supplement 1A). Most of these episodes were shorter than 2-3 s (Figure 1-Supplement 1B) and involved temperature drops greater than 0.01 C (Figure 1-Supplement 1C). Previous work has shown that C. elegans reliably responds to temperature change as small as 0.005 C/s1. Negative thermal fluctuations in the middle of forward runs up temperature gradients are common and within the detectable range for C. elegans. This raises the question: how is the response to these fluctuations suppressed so that the run continues?

In our revision, we clarify that the behavior we investigate is a biased random walk with an augmentation. We found that animals are more likely to have persistent runs during forward movements in favorable directions. Because the principal thermosensory neuron in C. elegans is exquisitively sensitive to rapid temperature changes, there must be a neural mechanism that filters rapidly fluctuating thermal input during the forward motor state.

We found that the mechanism for suppressing thermal fluctuations involves feedback from the motor circuit to a first-layer interneuron. This feedback, in effect, turns persistent forward movements during positive thermotaxis into an attractor-like behavioral state that is less sensitive to transient negative thermal fluctuations.

The revised manuscript clarifies our contribution and conclusion. Thank you for the advice that led to these improvements.

2. Inspecting the Ca++-imaging data in moving animals (e.g. Figure 2), I conclude that these animals must be crawling at extremely slow rates. Recent work showed that SMD neuronal activity is tightly locked to head bending phase (e.g. (Hendricks et al., 2012; Kaplan et al., 2019; Yeon et al., 2018)), which in this paper appears to be as slow as 2min (see Figure 2D, 5A). This is 1-2 orders of magnitude slower than freely crawling worms under standard laboratory conditions (e.g. compare to SMD imaging data in ref (Kaplan et al., 2019) or in semi-restrained animals (Hendricks et al., 2012)). Moreover, AVA and RIM Ca++ traces in this study exhibit minutes-lasting plateau states, which is characteristic for immobilized animals and typically not seen in freely crawling animals (see (Kato et al., 2015) for a comparison of these neurons between immobilized and freely crawling conditions). In conclusion, animals in this study must be nearly immobilized. I assume this is because animals have been placed between a coverglass and a 5% agarose pad, the latter having an unusual high concentration. To a non-expert eLife reader these caveats are quickly overseen.

2a. The authors should provide data about crawling speed in these experiments.

2b. The authors should provide an in-depth explanation why they chose such extreme conditions; perhaps to avoid movement artefacts given the slow 1Hz acquisition rate? Or is AIY modulation only detectable in these extreme conditions with strong and prolonged RIM activity phases?

Reviewer 1 is correct. In experiments where the animal was moving very slowly, they were semi-constrained. The head region moved forward or backward without leaving the field of view at high magnification. Under high mechanical load, animals exhibited much slower head oscillations.

We used slowly moving animals during calcium imaging to optimize signal-to-noise and reduce movement artifact. This is particularly crucial for the AIY interneuron where sensory encoding is only detectable in its neurite and not its soma. This requires imaging at considerably higher resolution than other methods we have used2.

We more explicitly describe the semi-constrained preparation. We have also added new analyses that demonstrate how we infer motor states in semi-constrained animals during calcium imaging (Figure 2-Supplement 1). With our high-resolution of a portion of the head, we cannot use the axial speed of the animal that we and others use at lower magnification. Here, we segment motor states using the phase difference in the movement of two body landmarks (the AIY soma versus neurite). We confirm that this metric works when segmenting motor states in freely moving animals (Figure 2-Supplement 1A,B), providing us with a means of assessing forward vs backward movement in our calcium imaging experiments with semi-constrained animals (Figure 2-Supplement 1D,E). In both cases, the calcium dynamics in AIY neurite is positively correlated with axial (forward) velocity (Figure 2-Supplement 1C,F).

In the revision, we more clearly describe how we infer the forward or backward state based on the relative movement of internal markers, not by directly tracking crawling speed. Given that the undulation is roughly 0.1Hz, as the Reviewer rightly points out, we can infer that the crawling speed in calcium imaging experiments is roughly ten-fold lower than that of typical behavioral assays.

2c. Were thermotaxis assays in Figure 1 performed under these conditions (coverglass, 5% agarose)? Are animals able to perform thermotaxis under these conditions?

All behavioral assays were performed on large (20 cm X 20 cm), unseeded, 2% agarose plates with no coverslips placed on top. In the absence of bacterial food and the larger plate size, conditions used for assaying thermotaxis behavior are identical to those used for cultivating the animals.

3. While AFD activity traces in Figure 2A look convincing, additional analysis is required for AIY and other inter- motor-neurons to demonstrate sensory responses and gating by behavioral state. The peak in power spectrum alone does not suffice to conclude that AIY etc. activity is stimulus locked. The AIY activity trace in Figure 2—figure supplement 1 from immobilized unstimulated worms indicates interesting dynamics in AIY, that certainly would appear as a peak in a power spectrum.

3a. I suggest showing phase aligned averages of the neurons and a statistical test showing that their activity gets significantly entrained.

3b. How do power spectra and phase-averages differ in forward versus reverse?

3c. Why does AIY show a peak at 0.067 and not 0.033Hz, like AFD?

3d. An equivalent control dataset from non-stimulated animals is essential.

We agree. The power spectrum is not the best measure of stimulus-locked activity in AIY in response to thermal oscillations. The power spectrum may be affected by motor-related oscillations from locomotion and is less reliable for analyzing signals in short time windows, e.g. a short bout of forward or backward movement.

Throughout the revision, we now use the more direct measure of cross-correlation between neural activity and the temperature waveform. We compare this cross-correlation between the forward and reverse motor state. This improved analysis is easier to understand and confirms our previous conclusion, as shown in new Figures 2, 5. We conclude that AIY does not exhibit stimulus-locked response to temperature up-sweeps or down-sweeps in wild type animals, while it does exhibit phase-locked response to the thermal inputs in RIM-ablated animals.

4. The finding that AIY is modulated by reversal neuron RIM in immobilized worms is crucial for the main conclusion of this paper, since it distinguished corollary discharge / efference copy from behavioral feedback. However, this statement is based only on a single observation shown in Figure 2—figure supplement 1 and lacks any quantification. Is AIY activity different in RIM high versus low states? This should be scrutinized via an analysis like it was done for reversal states in Figure 3; with multiple animal repetitions and appropriate statistics. In direct comparison to freely moving worms, this would be an important result deserving a spot in a main Figure. Without this extra data and analysis, the authors' hypothesis that AIY modulation is corollary discharge / efference copy versus behavioral feedback is not sufficiently supported.

We have now analyzed AIY activity profile across multiple immobilized animals. To test whether AIY still carry motor-related signals, we aligned AIY activity to the onset and offset of AVA activation. In Figure 3C-D, we show that AIY activity rises around the activation and dips around the inactivation of the premotor interneuron AVA (note that the activation of AVA, defined by binarizing AVA activity using the Otsu method, may not be a perfect predictor of the onset of fictive reversal states). In addition, we present the average cross-correlations of AIY activity with other neurons in immobilized animals in Figure 3-Supplement 2B,C, which shows that the anti-correlation between the activity of AIY and that of AVA remains unchanged in immobilized animals.

5. I am wondering how reversal and forward states were annotated in AVA or AVB ablated animals respectively, which should lack these behaviors according to a large body of literature. If AVA and AVB were successfully ablated, the analysis in Figure 4B should not be possible. Also, ablation of RIM and AIB can affect behavioral state durations and switching frequencies. The authors should provide these data along with evidence that the cells were indeed specifically ablated, thus they must validate and characterize all of their ablation strains. Data like in Figure 4C,D should be shown as well form the other ablations.

It is not true that ablation of single premotor interneurons such as AVA or AVB completely abolishes reversals or forward movement. These premotor interneurons are active during and contribute to these motor states, but animals lacking AVA can still crawl backward and animals lacking AVB can still crawl forward as noted in the first study to ablate these neurons3 and subsequent studies of the contribution of these premotor interneurons to locomotion4. Animals without premotor interneurons can still move, but less well.

In Figure 4B and Figure 4 Supplement 1A-C, we show the detailed behavioral-state-triggered AIY activity profiles for every ablation strain used in this study.

In our ablation experiments, we used the flavoprotein miniSOG to induce neuronal death by photoactivation5. To verify death, we expressed miniSOG in neurons along with cytoplasmic RFP. To ablate neurons, the transgenic strain was exposed to LED light as early larva and animals were examined in adults. We have verified the process of miniSOG-induced neuronal ablation using RFP markers6. Upon LED exposure, neurons die and disintegrate. Red debris is gradually cleared by other tissues, most prominently by muscles. Complete clearance takes approximately two days, by which time the larva have become adults. The clearance of neuronal RFP signals in adults is the criterion for successful ablation of the neuron, and is not attributable to photobleaching.

6a. Data shown in Figure 4E come from partially very low numbers of independent experiments and no statistics are provided. These shortcomings must be addressed. (6b) Moreover, stimuli were applied at very different pre-stimulus AIY activity baselines. More data would definitely strengthen this result substantially.

We have substantially augmented our optogenetic and control datasets and analyses.

In the new Figure 4C-F, we present both the raw heat maps and analyses quantifying the red-light-evoked response in AIY. The new data allow us to differentiate the effect of optogenetic stimulation at different pre-stimulus activity levels of AIY. We show that optogenetic activation of AVB induced a strong increase in AIY activity when AIY exhibited low pre-stimulus activity levels. Optogenetic activation of AVB evoked weak to no increase when AIY exhibited medium to high pre-stimulus activity levels. These data suggest that AIY cannot be activated further if its activity is already near its maximum.

For AVA optogenetic activation, we observed a rapid drop in AIY activity when AIY had medium to high pre-stimulus activity levels.

In contrast, optogenetic activation of AVA or AVB in RIM-ablated animals yielded no significant response in AIY. These new data substantiate the crucial requirement for RIM for AVBand AVA-activation evoked responses in AIY. These results point to a critical role for RIM in relaying signals from pre-motor interneurons to AIY.

7. I find the result that RIM and SMD neurons show stimulus evoked activity fluctuations in RIM(-) animals very interesting, but this result needs more substantial quantifications (see comment (3)).

We have added these quantifications in the state-specific correlation between the stimulus and activity for all neurons that we studied in RIM-ablated animals Figure 5A. These analyses reveal significant correlations between the thermal stimuli and the activity of AVA, SMDD, and SMDV.

We agree with the reviewer that these correlations that emerge specifically in RIM- animals is interesting. This observation is consistent with our model. RIM-mediated motor feedback, by driving persistent neural activity states, helps to insulate motor neurons from the effects of transient or fluctuating sensory inputs.

8. Figure 6: the effect of RIM ablation on thermotaxis is very interesting but there are many ways how this manipulation could affect the behavior. Additional analysis of the data could provide more compelling evidence supporting their model: consistent with their model run durations are decreased (Figure 6E). If the function of RIM corollary discharge is indeed to maintain prolonged forward states in spite of negative dT/dt epochs, run terminations should be coinciding with these events, but not in control animals. See also comment (1) for analyses suggestions.

We agree. In the new Figure 6D-F we have more carefully and thoroughly quantified the effect of cooling (dT/dt < 0) on run termination. In wild type animals, cooling epochs that occurred during a forward run led to run termination (i.e. switching from positive to negative velocity) about 25% of the time. In RIM-ablated animals, this likelihood increases to 37%.

We found that the probability of negative dT/dt leading to run termination depends on the duration of the cooling epoch. In both wild type and RIM-ablated animals, cooling periods longer than 7 seconds were more likely to be followed by the termination of the run. For longer cooling epochs, the increase in RIM-ablated animals from the wild-type is more striking: nearly 2-fold for epochs between 7-15 seconds (Figure 6F). We observed no change in the frequency of run termination for sustained cooling epochs (> 30s) after RIM ablation. RIM plays a role in sustaining forward movements for brief but not sustained cooling epochs.

In a complementary analysis, we examined the probability of forward runs being preceded by a period of cooling (Figure 6-Supplement 1C-E). We found that this probability increased from about 29% in wild type animals to 47% in RIM-ablated animals. Consistent with our model, cooling epochs are more likely to truncate forward runs in RIM-ablated animals.

Taken together, our new analyses confirm that RIM ablation increases the susceptibility of forward runs to be terminated by negative thermal fluctuations. Our interpretation is that the forward run is an attractor-like state that is mediated by RIM-dependent motor feedback. This feedback reduces the sensitivity of forward movements to transient thermal fluctuations or noise. This is described in the revision.

9. Modulation of primary sensory circuits by behavioral state seems to extend to many other neurons in the worm brain (Kato et al., 2015), and this seems to be a more general phenomenon seen in other organisms (e.g. Aimon et al., 2019; Musall et al., 2019; Salkoff et al., 2019; Stringer et al., 2019). See also for a recent discussion (Kaplan et al., 2018). The current manuscript, in my opinion, is pretty important providing data hinting at a function of these phenomena. This is just my recommendation, but I think the authors could make more impact with their story by discussing their findings in the context of these recent discoveries.

We agree. A growing body of literature across species report the encoding of behavioral states in sensory areas. Studies in flies and the electric fish have revealed a role for motor feedback in canceling sensory inputs caused by self-motion. Our study has uncovered another functional role for motor feedback, generating attractor-like persistent behavioral states that filter transient sensory fluctuations. We now reference these pertinent studies that provide context for our work.

Reviewer #2:

Ni Ji and colleagues apply quantitative analysis of behavior and mutli-neuronal calcium imaging to identify and examine a corollary discharge circuit that filters sensory input based on motor state during thermotaxis. The main claims of the paper are that RIM reports this motor state (the corollary discharge signal) to the first layer interneuron AIY, which acts as a sensory filter to ignore temperature fluctuations encoded by AFD during reversals:

1. AIY encodes temperature and motor state information

2. Motor state representations in AIY arise from a corollary discharge signal originating from RIM.

3. CD allows positive feedback to sustain forward locomotion

The first and second claims are well-supported, while the ethological model is less well-developed and was not rigorously tested. Issues with the data, particularly what seem like unrealistically long reversal states make some results difficult to interpret.

Overall, I am convinced by the data showing that RIM broadcasts a motor state signal (the CD) that filters or modulates upstream sensory pathways. This is a nice result but given the systemic effects of RIM/tyramine on sensory neurons and interneurons and the prior results showing that RIM ablation affects reversal rates even under isotropic conditions, I don't think the argument that it is supporting efficient thermotaxis is very well supported. Identifying the relevant receptor on AIY would bring better resolution to the circuit analysis.

We appreciate the positive assessment and suggestions for improvement.

1. The behavior and activity patterns look unusual for locomoting animals. 2A shows reversals that last 30-60 seconds, typical reversals during navigation are <5s. I looked at the dataset and indeed the F/R ethogram in 2A is incorrect-there are many more state changes than represented in the figure. However, the state durations for reversals are still very long (median >30s for reversals, just under 30s for forward). It does not seem like this can be correct for a freely moving animal and is inconsistent with the tracks shown in 1A.

2. As the authors note, SMDD/V activity has been reported to match head movements (Hendricks et al., 2012; Yeon, Kim et al., 2018; Kaplan, Thula et al., 2019), and so in panel 2D I would expect oscillatory activity that matches the forward gait in the traces and appears at ~1Hz in the power spectrum. As in 2A (and throughout), the extremely long "reversal" state durations look more typical of restrained animals.

Reviewer One raised the same comments, which we have addressed with new experiments and analyses. See above.

3. Figure 2A and 2D. Do the colors in the activity histogram correspond to forward and reverse states? If so this should be indicated and perhaps match the color used for the ethogram. I am not sure having the Ca++ traces superimposed on the stimulus indicator and ethogram improves clarity, but it's fine.

We agree. We have now modified the colors of the histograms to match those used in the ethogram.

4. The power spectra are too cramped and low-res in Figure 2D to be legible. It would be useful to compare spectra (or perform temp-AIY cross correlations) from stimulus cycles that occur during reversals vs forward epochs, particularly to support the claim about AIY.

We agree. As described in our response to Reviewer One, we have replaced the power spectra analyses with the better cross-correlation correlation analyses between stimulus-triggered calcium activity for AIY and other neurons with the thermal stimuli during forward movement and reversals (new Figure 2D and Figure 5A). Thank you for the suggestion. These analyses more clearly represent the phase-locked response of AIY to thermal inputs that occurs in RIM-ablated animals.

5. How many animals are represented in the summary data (histograms/spectra) in Figure 2? Statistical tests on summary data should be reported.

The activity histogram in Figure 2 and Figure 5 are shown only for the sample dataset to the left. The cross-correlograms and the newly added stimulus-triggered averages, however, are generated from data across multiple animals. We have now included the number of animals corresponding to these analyses in the figure captions.

6. Because AIY shows discrete Ca++ peaks, a quantification of warming cycles that do and do not produce an AIY response (as a binary output) in forward and reverse states might be more clear or intuitive than the (also small and cramped) histogram.

We agree. We now provide much more detailed information about the AIY response to warming and cooling in Figure 2D.

7. There is a formatting error in the legend of 2A (fstim, stim should be all subscript)

This power spectrum has been supplanted by cross-correlation analyses.

8. Label forward and reverse states in panel 2D as in 2A.

We have incorporated these state labels.

9. What do the colors mean in 2B?

The colors only highlight the neurons of key interest.

10. I think the standard wiring diagram does not have an EJ between AFD and AIZ, as shown in 2B and 4A (though these authors may know better)?

This is correct, the original connectome does not report gap junctions between AFD and AIZ, but we have observed gap junctions in all of connectomes that we recently reconstructed across the developmental time course (see http://www.nemanode.org)7.

11. Figure 2—figure supplement 1. Same questions as above regarding the very long reversals and lack of SMDD activity matching forward gait. The "right" side of panel A is labeled panel B and panel B as described in the legend and line 141 (neuronal cross correlations with RIM) is missing.

The moving worms in this analysis are semi-constrained, giving rise to the very long reversals. The figure legends in the original submission had an error. We now provide neuronal cross-correlations for all multineuron imaging datasets in the resubmission (see Figure 2 and Figure 3-Supplement 2).

12. Figure 3. Consider using a perceptually uniform color map for heat maps (here and elsewhere).

In the revised figures, the heat maps for temperature change, locomotion state and calcium activity is used consistently throughout. We wish to point out that, in Figure 2E and 5A, the warming periods might appear reddish as result of a visual illusion. The actual colors corresponding to warming are shades of yellow as indicated by the blue-yellow heat map.

13. Spell out corollary discharge in the Figure 3 legend.

Done.

14. While AVA correlates perfectly with reversals, it is less clear how well RIM does under different conditions (Gray et al., 2005; Guo et al., 2009; Piggott et al., 2011, Gordus et al. 2015)..… AVA activation/reversals can happen while RIM is inactive, and RIM activation can occur in the absence of a reversal, and in fact can lower the probability of a stimulus-evoked reversal (Gordus), while ablating RIM can increase reversal frequency (Gray, Piggott). While RIM Ca++ does correlate pretty well with reversal, and it's clear here that it is inducing sensory filtering at AIY, it's possible that it is not exclusively or necessarily reversals. This should be considered.

RIM has been studied in many different contexts, and we cite a number of these previous studies. In our hands, we have never observed an increase of RIM activity that was not correlated with reversals during spontaneous movement. RIM calcium dynamics may differ in other experimental paradigms. This is now discussed in the revised Discussion.

15. LGC-55 and TYRA-2 are expressed in AIB and AIZ; SER-2 and TYRA-2 are expressed in AIY (https://cengen.shinyapps.io/SCeNGEA/). Given that SER-2 mediates inhibition of synaptic transmission in GABAergic motor neurons (as some of these authors showed), it seems a promising receptor to mediate RIM's inhibitory effect. Was this checked? Identifying the AIY receptor would resolve some points below.

16. The principle claim of the paper is that the CD signal from RIM communicates a motor state to AIY to filter AFD input during reversals. If the effect is in large part tyramine-mediated, there are receptors for this modulator on many or most of the sensory neurons upstream of AIY, as well as on its interneuron synaptic partners. Are AFD responses normal during manipulation of RIM? Could other sensory neurons have altered responses to temperature that impact AIY? It would be show that a tyramine receptor acts in AIY to directly respond to RIM signaling (probably SER-2) and show that expression in AIY is necessary/sufficient to filter AFD signaling during reversals.

We thank the reviewer these suggestions. We think that they offer excellent starting point to reveal the molecular and signaling underlying of this feedback mechanism. However we don’t expect these pathways are linear enough to be solved by one or two single mutant analyses. Many neurons express the same receptors (e.g. SER-2), and one single neuron expresses many receptors (e.g. SER-2 and TYRA-3). Previously studies have revealed extensive redundancy in these pathways. We have incorporated these comments in the discussion for our future studies in the revised manuscript.

17. Should indicate color coding for WT and RIM- in 6E/F in the panel.

18. Previous results showing that ablating RIM can increase reversal frequency under some conditions, including an isotropic environment (Gray, Piggott), complicate the interpretation here that shorter run lengths in RIM-ablated animals under thermal fluctuation are due to the sensory filtering observed in AIY. Do RIM- animals reverse more under these conditions in the absence of thermal stimulation? While the need to maintain forward movement during fluctuations is clear, it is less clear why an animal (or AIY at least) would need to ignore AFD input during reversals. I'm particularly concerned about this aspect given the unusually long reversal states observed here.

There have been conflicting reports on the effect on manipulating RIM. In their independent studies of C. elegans locomotion, Kawano and Zhen (unpublished results) reached these conclusions: during spontaneous movements, activation of RIM is always associated with reversals, anatomic ablation of RIM does not alter spontaneous motor states; constitutive silencing of RIM with Kir2.1 or TWK-18 does not alter spontaneous motor states. The results reported in this study are consistent with their independent earlier assessments. In our discussion, we now point out that there have been conflicting reports, and some of this conflict may be due to differences in the experimental paradigm.

Reviewer #3:

Here, Ji and colleagues study a sensory-driven navigational circuit and ask how sustained locomotor-related signals arise from self-generated, rapid changes in sensory stimuli driven by the navigational strategy of the worm to successfully reach its goal. The question on the relation among sensory cues, internally generated signals, and motor strategies, is poorly understood for any moving animal. Therefore, I consider the work very timely and important for a wide neuroscience audience.

I think this work is important and heroic: examining the presence and role of internal signals is no easy duty, and I think the authors are taking advantage of the model system beautifully. The propose role for a CD signal presented here is a very interesting and exciting idea, and consistent with the effect on behavior on the ablation of RIM (shown to be the source of motor-related signals of the AIY interneuron). However, the authors present some statements that I am not sure are shown directly from their experimental data. The concern is really about the gap between the description of the activity in AIY, a convincing metric of the presence of a persistent motor activity, and the interpretations of the data by the authors. While RIM ablation induces bleed through of sensory signals in motor cells, this effect may be due to the connections between RIM and motor cells per se rather than a consequence of the presence of a positive feedback CD signal at AIY level.

Through a phenomenological model, which was inspired by their experiments (combining analysis of behavior, physiology, and activity manipulations), they describe how the presence of a positive feedback CD signal can account for a persistent motor-related signal at an interneuron of the circuit, and the biased behavior of the worm. While this is a very exciting idea, I am not convinced that the experimental evidence supports this view clearly.

1. The authors show the presence of an internally generated signal at AIY reporting motor state, and convincingly show that RIM activity is required for this signal.

We thank Reviewer 3 for the positive comments.

2. However, the increase of AIY activity at onset of Forward runs, and it decrease at onset of Reversals can be equally explained by either the a conventional efference copy (EC)-like signal that would be associated with reversals (since AIY is part of the forward pathway) or the proposed positive feedback CD signal (PF-CD) that would be associated with forward runs, and induce persistent activity in AIY. I see no effort in the manuscript to distinguish between these two possibilities. An EC like signal can also explained the increase in sensory cues in the response profile of AIY under oscillating temperature cues in the absence of activity in RIM (Figure 6A).

We agree with the reviewer that both phenomena can in principle arise from either negative feedback (NFB) or positive feedback (PFB) from the motor circuit. This ambiguity arises in part from the inherent limit of calcium imaging, which reports neural activity as a change in fluorescence relative to a baseline.

We thus re-examined our results for evidence that could differentiate between the NFB and the PFB model. We found two sets of evidence supporting the latter. First, reversal-associated inhibition to AIY, as part of the NFB-EC model, should not affect how AIY responds to thermal stimuli during the forward run. Instead, we found that the thermosensory response in AIY became more reliable during both forward runs and reversals upon RIM ablation (Figure 5D). This observation argues against a pure NFB model where AIY is simply inhibited during reversals. Through a dynamical systems model (Figure 7), we show that our results are better explained by a positive feedback network, which exhibits bi-stable dynamics that are resilient to rapid input fluctuations under both the ON (i.e. forward run) and OFF (i.e. reversal) states.

The second piece of evidence comes from examining the amplitude of AIY response to thermal stimuli. We focus specifically on warming-evoked response in AIY during forward runs, which should not be impacted by any reversal-associated inhibitory EC signal. As shown in the new Figure 5D, we observe in the wild type a long-tailed distribution, reflecting the occasional large increases in AIY activity upon warming. These large activity transients were abolished in RIM-ablated animals. This observation cannot be explained by a reversal-associated NFB model, as these responses occur during forward runs when AIY is already in its high state. A plausible explanation, from a dynamical systems perspective, is that there is an imaginary eigencomponent associated with the high state of AIY activity8, which could allow AIY activity to transiently leave the stable attractor state and reach even higher levels.

Together, this evidence supports the requirement of positive feedback in sustaining AIY activity during forward runs. Our results, however, do not exclude the existence of an NFB signal. As mentioned in the new Discussion, a recent study from Cori Bargmanns group has shown that RIM drives reversals through its chemical output, while stabilizing forward runs through its gap junction outputs 9. Thus, it is possible that the RIM-dependent motor feedback found in our study also has more than one function. We have incorporated the above analysis and discussion points in the revised manuscript, and we thank Reviewer 3 again for prompting us to think more deeply about our circuit model.

On a related note, we have observed that both classical and recent literature use the terms efference copy and corollary discharge as largely synonymous concepts. As we describe in the new Introduction, von Holst and Mittelstaedt proposed the term efference copy and R.W. Sperry coined corollary discharge in 1950 to describe similar phenomena in flies and fish. Both terms referred to an internal copy of the motor signal that functions to cancel sensory inputs evoked by self-motion. While subsequent studies in fish and flies tend towards the original usages of EC or CD in their respective fields, studies in other species use EC and CD interchangeably when referring to inhibitory motor feedback signals. As we clearly describe in the introduction, we follow the general usage of the term corollary discharge and note that we consider it a synonym for ”efference copy”.

3. Optogenetic experiments are more consistent with an inhibition from reversal pathways (EC-like phenomena) than an excitation signal from forward runs (PF-CD like phenomena) (Figure 4). figure 4E, I really cannot see the effect the authors describe for the manipulation of AVB (either in the presence or absence of a functional RIM), whereas the description of the effect of the manipulation of AVA is closer to what we see in the data.

We apologize for this mistake in our original submission. The heat map showing the AIY response to AVB stimulation was accidentally replaced with the heat map for the “no ATR control for AVA stimulation”. We have corrected this mistake in the new Figure 4C-F.

In addition, we have included more trials for each experimental condition. In Figure 4D,F, we quantify the stimulus-evoked response in AIY as a function of its pre-stimulus activity levels. We show that optogenetic activation of AVB induced a strong increase in AIY activity at low pre-stimulus activity levels, moderate increase for medium pre-stimulus activity, and no response at high pre-stimulus activity. This dependence likely reflects a ceiling effect, where AIY activity is not driven above a maximal value. We observe the opposite effect for the optogenetic stimulation of AVA. AIY exhibited a drop in activity upon AVA stimulation when AIY had medium to high pre-stimulus activity levels. RIM ablation shows that RIM is required for both AVA- and AVB-evoked changes in AIY activity.

These new data are consistent with our model that motor-state activity in AIY represents a feedback signal from premotor interneurons, and that this feedback is RIM-dependent.

4. There is a missing metric for persistent activity in AIY, or a clear explanation of whether bimodality would be such metric. If this is the case, the authors make a very weak argument on how such a metric would measure RIM-dependent persistent motor-sate activity in AIY. If I understood correctly, bimodality should represent a more persistent activity at a particular motor state. But it is not very clear how bimodal the AIY activity is under the presence vs the absence of RIM activity, for example, if one compares figure 5A with Figure 2 Suppl Figure 1 (left column, mobile animals at constant T). Therefore, perhaps it is important to show the corresponding distribution for the supplementary figure to be able to compare the critical metric that is then quantified in Figure 5B.

We thank the reviewer for this reasonable suggestion. In the new manuscript, we examined in closer detail the distribution of AIY in WT and RIM-ablated animals. As shown in Figure 3 Supplemental 1 and Figure 5 Supplemental 1, we found that in both WT and RIM-ablated animals, the distributions of AIY activity can be well approximated by a Gaussian mixture model consisting of three Gaussian distributions. The three Gaussian distributions can be viewed as representing low, intermediate and high activity states. In WT animals, the low state is most stable (represented by a high, narrow peak close to zero) while the high state (arising from large transients often observed at run onset) is broadly distributed. If one takes the low state as the OFF state and takes the sum of the intermediate and high states as the ON state, then the AIY ON state coincides closely with the forward run state in wild type animals Figure 3 Supplemental 1C.

In RIM ablated animals, the distribution of AIY activity is also best described by a mixture of three Gaussians (Figure 5 Supplemental 1C). Compared to the WT, the Gaussian associated with the low activity state is broader and the Gaussian corresponding to high activity represents a much smaller fraction of the full distribution. These changes reflect a less stable low state and the disappearance of large activity transients. If we again take the low state as the OFF state and takes the sum of the intermediate and high states as the ON state, we find that the duration of the AIY ON states were significantly shortened in RIM ablated animals under oscillating temperature (Figure 5D). This pattern is consistent with our earlier finding that forward run durations were shortened in RIM-ablated animals when exposed to varying temperature (Figure 6C and Figure 6 Supplement 1B). Interestingly, we did not find the duration of AIY activity to change significantly in RIM ablated animals under constant temperature, nor did the duration of forward runs. It is possible that other mechanisms are involved in regulating forward run durations in the absence of food and strong sensory inputs. In fact, C. elegans are known to gradually transition from frequent short forward runs (i.e. the local search state) to persistent forward runs (i.e. the global search state) within the first 20 minutes after transfer to an off-food environment without sensory gradients10.

Together, these results suggest RIM sustains AIY activity as well as prolongs the forward state in the presence of fluctuating thermal inputs. These new analyses on AIY distribution and duration of activation replace our previous analysis using an index of bimodality.

5. Perhaps they could use the model to predict the effect on behavior of an EF vs PF_CD function. While the latter was examined, it is unclear how the worm behavior departs from a model under an EC signal at AIY

We thank the reviewer for this suggestion. The goal of our model was indeed to explore the functional properties of different circuit architectures. As shown in the new Figure 7, we drew comparisons between networks models with the positive feedback (PFB), no feedback, or negative feedback (NFB) from the motor command neurons. We show that: (1) under oscillating sensory input, only the PFB model exhibited sustained ON and OFF states, as observed for AIY in WT animals (Figure 7B) in an agent-based simulation of thermotaxis, the PFB model exhibited stronger dependence of forward run duration on heading direction (i.e. biased random walk) and more efficient thermotaxis overall (Figure 7C-E). Furthermore, comparing the simulated trajectories generated through the PFB model and the model without motor feedback, we observe a stronger association between cooling stimuli and termination of forward runs. This observation echoes the difference in thermotaxis behavior between WT and RIM ablated animal shown in Figure 6D-F and Figure 6-S1C-E.

6. I suggest the authors make less of strong statements when the experimental evidence is not shown directly (for example, in line 235, they describe that "motor-related signals and Temperature-related signals reinforce each other": this is only shown through the model).

We have modified our statements to stay consistent with the data.

7. An exciting observation is the bleed through of sensory signals to motor cells within the circuit, and what potential impact these may convey at the level of the behavior of the worm (Figure 6). AIY has increased sensory components, but the lack of an EC could also induce such an effect, for example, by extending the time the neuron responds to T fluctuations.

As shown in the new Figure 5D, we found that the thermal response in AIY become more reliable during both forward runs and reversals. While a negative feedback EC signal could explain the increase in thermal response during reversals in RIM ablated animals, it could not explain the more reliable response during forward runs. As we demonstrate in the computational model (Figure 7B), a positive feedback network can generate bi-stable states, thereby preventing individual neurons from responding to rapid input fluctuations during both states.

8. What is the impact of making AIY insensitive to the motor state on other postsynaptic elements of the circuit? The authors show that motor cells themselves respond to sensory fluctuations likely affecting the persistent behavioral state. But is this a consequence of the lack of motor state sensitivity of AIY or a consequence of the lack of activity of RIM, a neuron that is presynaptic to motor cells? In conveying the idea of a PF-CD signal, it is important to distinguish these scenarios. A neuron that integrates information from AIY and AFD, and that is presynaptic to motor cells may be an ideal candidate. For example, AIB is very well suited for this distinction. If AIY activity contained both a persistent signature of motor state and thermal signals, then AIB sensory information should be modulated by the AIY's indirect input. On the other hand, if the motor state modulation in AIY is related to an EC signal, AIB cell sensory activity should not be strongly modulated by the lack of such EC signal (given electrical coupling of AIB and AFD). If this is not a possible experiment, then the strength of the statements should be adjusted accordingly, and the discussion on the presence of a conventional EC signal should be present.

Indeed, RIM is densely connected with many inter- and motor neurons and could affect behavior in different ways. Since AFD is the only thermosensory neuron known to contribute to positive thermotaxis11 and the only sensory neuron shown to phase-lock with thermal stimuli12, the stimulus-locked activity that emerges in the motor circuit after RIM ablation is likely downstream of AFD.

AIY and AIB are the two primary post-synaptic partners of AFD. A recent study showed that AIB receives excitatory feedback input from RIM and exhibits positively correlated activity with both RIM and the premotor neuron AVA13. Silencing RIM restored reliable response in AIB to olfactory stimuli. Since AIB is known to promote reversals, the motor-related feedback from RIM to AIB serves as a positive feedback signal, similar to the RIM-dependent input to AIY reported in our study. Since AIB receives input directly from RIM, examining its activity will not directly inform the role of motor representation in AIY. An ideal experiment in this case would be to directly measure voltage dynamics in AIY in WT and RIM-ablated animals. This may be possible when more reliable voltage indicators or electrophysiological techniques become accessible but is outside the scope of this study.

References

1. L. Luo, D. A. Clark, D. Biron, L. Mahadevan, and A. D. T. Samuel. Sensorimotor control during isothermal tracking in Caenorhabditis elegans. The Journal of Experimental Biology, 209(3):4652–4662, 2006

2. V. Venkatachalam, N. Ji, X.Wang, C. Clark, J. K. Mitchell, M. Klein, C. J. Tabone, J. Florman, H. Ji, J. Greenwood, A. D. Chisholm, J. Srinivasan, M. Alkema, M. Zhen, and A. D. T. Samuel. Pan-neuronal imaging in roaming Caenorhabditis elegans. Proceedings of the National Academy of Sciences of the United States of America, 113(8):E1082-8, 2016

3. M. Chalfie, J. E. Sulston, J. G. White, E. Southgate, J. N. Thomson, and S. Brenner. The neural circuit for touch sensitivity in Caenorhabditis elegans. Journal of Neuroscience, 5(4):956–964, 1985. ISSN 02706474. doi: 10.1523/jneurosci.05-04-00956.1985

4. T. Kawano, M. D. Po, S. Gao, G. Leung,W. S. Ryu, and M. Zhen. An imbalancing act: Gap junctions reduce the backward motor circuit activity to bias C. elegans for forward locomotion. Neuron, 72(4):572–586, 2011

5. Y. B. Qi, E. J. Garren, X. Shu, R. Y. Tsien, and Y. Jin. Photo-inducible cell ablation in Caenorhabditis elegans using the genetically encoded singlet oxygen generating protein miniSOG. Proceedings of the National Academy of Sciences of the United States of America, 109(19):7499–7504, 2012

6. S. Gao, S. A. Guan, A. D. Fouad, J. Meng, T. Kawano, Y. C. Huang, Y. Li, S. Alcaire, W. Hung, Y. Lu, Y. B. Qi, Y. Jin, M. Alkema, C. Fang-Yen, and M. Zhen. Excitatory motor neurons are local oscillators for backward locomotion. eLife, 7:1–32, 2018

7. D. Witvliet, B. Mulcahy, J. K. Mitchell, Y. Meirovitch, D. R. Berger, Y. Wu, Y. Liu, W. X. Koh, R. Parvathala, D. Holmyard, R. L. Schalek, N. Shavit, A. D. Chisholm, J.W. Lichtman, A. D. T. Samuel, and M. Zhen. Connectomes across development reveal principles of brain maturation in C. elegans. bioRxiv, 2020

8. S. H. S. H. Strogatz. Nonlinear dynamics and chaos : with applications to physics, biology, chemistry, and engineering. Westview Press, Boulder, CO, 2015. ISBN 0813349109

9. A. Sordillo, C. I. Bargmann, T. Rockefeller, C. Z. Initiative, and R. City. Behavioral control by depolarized and hyperpolarized states of an integrating neuron by. bioRxiv, pages 1–54, 2021. doi: 10.1101/2021.02.19.431690

10. A. L´opez-Cruz, A. Sordillo, N. Pokala, Q. Liu, P. T. McGrath, and C. I. Bargmann. Parallel Multimodal Circuits Control an Innate Foraging Behavior. Neuron, 102(2):407–419.e8, 2019. ISSN 10974199. doi: 10.1016/j.neuron.2019.01.053

11. M. Ikeda, S. Nakano, A. C. Giles, L. Xu, W. S. Costa, A. Gottschalk, and I. Mori. Context-dependent operation of neural circuits underlies a navigation behavior in Caenorhabditis elegans. Proceedings of the National Academy of Sciences of the United States of America, 117(11):6178–6188, 2020

12. D. A. Clark, D. Biron, P. Sengupta, A. D. T. Samuel, Clark, D. Biron, P. Sengupta, and A. D. T. Samuel. The AFD sensory neurons encode multiple functions underlying thermotactic behavior in Caenorhabditis elegans. Journal of Neuroscience, 26(28):7444–7451, 2006. ISSN 1529-2401. doi: 10.1523/jneurosci.1137-06.2006

13. A. Gordus, N. Pokala, S. Levy, S. W. Flavell, and C. I. Bargmann. Feedback from network states generates variability in a probabilistic olfactory circuit. Cell, 161(2):215–227, 2015

https://doi.org/10.7554/eLife.68848.sa2

Article and author information

Author details

  1. Ni Ji

    Department of Physics and Center for Brain Science, Harvard University, Cambridge, United States
    Present address
    Department of Brain and Cognitive Sciences, MIT, Cambridge, United States
    Contribution
    Conceptualization, Resources, Methodology, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Writing - original and revised manuscript
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-7870-0678
  2. Vivek Venkatachalam

    Department of Physics and Center for Brain Science, Harvard University, Cambridge, United States
    Present address
    Department of Physics, Northeastern University, Boston, United States
    Contribution
    Resources, Software, Methodology, Writing - original draft, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-2414-7416
  3. Hillary Denise Rodgers

    1. Department of Physics and Center for Brain Science, Harvard University, Cambridge, United States
    2. Department of Neurobiology, University of Massachusetts Medical School, Worcester, United States
    Contribution
    Resources
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0565-1940
  4. Wesley Hung

    1. Lunenfeld-Tanenbaum Research Institute, Mount Sinai Hospital, Toronto, Canada
    2. Departments of Molecular Genetics, and Physiology, University of Toronto, Toronto, Canada
    Contribution
    Resources
    Competing interests
    No competing interests declared
  5. Taizo Kawano

    1. Lunenfeld-Tanenbaum Research Institute, Mount Sinai Hospital, Toronto, Canada
    2. Departments of Molecular Genetics, and Physiology, University of Toronto, Toronto, Canada
    Contribution
    Resources
    Competing interests
    No competing interests declared
  6. Christopher M Clark

    Department of Neurobiology, University of Massachusetts Medical School, Worcester, United States
    Contribution
    Resources
    Competing interests
    No competing interests declared
  7. Maria Lim

    1. Lunenfeld-Tanenbaum Research Institute, Mount Sinai Hospital, Toronto, Canada
    2. Departments of Molecular Genetics, and Physiology, University of Toronto, Toronto, Canada
    Contribution
    Resources
    Competing interests
    No competing interests declared
  8. Mark J Alkema

    Department of Neurobiology, University of Massachusetts Medical School, Worcester, United States
    Contribution
    Resources, Supervision, Writing - original draft, Project administration, Writing - review and editing
    For correspondence
    mark.alkema@umassmed.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-1311-5179
  9. Mei Zhen

    1. Lunenfeld-Tanenbaum Research Institute, Mount Sinai Hospital, Toronto, Canada
    2. Departments of Molecular Genetics, and Physiology, University of Toronto, Toronto, Canada
    Contribution
    Resources, Supervision, Writing - original draft, Project administration, Writing - review and editing
    For correspondence
    meizhen@lunenfeld.ca
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0003-0086-9622
  10. Aravinthan DT Samuel

    Department of Physics and Center for Brain Science, Harvard University, Cambridge, United States
    Contribution
    Conceptualization, Supervision, Funding acquisition, Writing - original draft, Project administration, Writing - review and editing
    For correspondence
    samuel@physics.harvard.edu
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-1672-8720

Funding

National Institute of Neurological Disorders and Stroke (NS082525-01A1)

  • Mark J Alkema
  • Mei Zhen
  • Aravinthan DT Samuel

National Institute of General Medical Sciences (PO1 GM103770)

  • Aravinthan DT Samuel

National Institute of General Medical Sciences (RO1 GM084491)

  • Mark J Alkema

Burroughs Wellcome Fund

  • Vivek Venkatachalam

Canadian Institutes of Health Research (154274)

  • Mei Zhen

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

Acknowledgements

We thank Daniel Witvliet for insights on C. elegans connectomes and generating Figure 2B. We thank the laboratories of Yun Zhang, Daniel Colón-Ramos, Shawn Xu, and Cori Bargmann for strains. This work was supported by NIH P01 GM103770 (ADTS), NIH R01 NS082525-01A1 (ADTS, MZ, MJA), NIH R01 GM084491 (MJA), the Burroughs Wellcome (VV), and the CIHR foundation 154274 (MZ). We thank Steven Flavell and members of the Samuel, Alkema, and Zhen laboratories for constructive advice and help in completion of the study and manuscript preparation.

Senior Editor

  1. Ronald L Calabrese, Emory University, United States

Reviewing Editor

  1. Manuel Zimmer, Research Institute of Molecular Pathology, Vienna Biocenter and University of Vienna, Austria

Publication history

  1. Received: March 27, 2021
  2. Accepted: April 8, 2021
  3. Accepted Manuscript published: April 21, 2021 (version 1)
  4. Accepted Manuscript updated: April 22, 2021 (version 2)
  5. Version of Record published: May 21, 2021 (version 3)

Copyright

© 2021, Ji 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.

Metrics

  • 1,001
    Page views
  • 133
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)