|Home | About | Journals | Submit | Contact Us | Français|
Foraging animals have distinct exploration and exploitation behaviors that are organized into discrete behavioral states. Here we characterize a neuromodulatory circuit that generates long-lasting roaming and dwelling states in Caenorhabditis elegans. We find that two opposing neuromodulators, serotonin and the neuropeptide pigment dispersing factor (PDF), each initiate and extend one behavioral state. Serotonin promotes dwelling states through the MOD-1 serotonin-gated chloride channel. The spontaneous activity of serotonergic neurons correlates with dwelling behavior, and optogenetic modulation of the critical MOD-1-expressing targets induces prolonged dwelling states. PDF promotes roaming states through a Gαs-coupled PDF receptor; optogenetic activation of cAMP production in PDF receptor-expressing cells induces prolonged roaming states. The neurons that produce and respond to each neuromodulator form a distributed circuit orthogonal to the classical wiring diagram, with several essential neurons that express each molecule. The slow temporal dynamics of this neuromodulatory circuit supplement fast motor circuits to organize long-lasting behavioral states.
Animal behaviors are often organized into discrete, long-lasting behavioral states. For example, many foraging animals alternate between local feeding bouts and migrations between feeding sites, with each state lasting for minutes to hours (Owen-Smith et al., 2010). Local feeding and active migration differ in arousal levels, metabolic costs, risk of predation, and the potential to discover new resources, and therefore this bistable structure represents an example of an exploration-exploitation axis. Transitions between behavioral states can be induced by external stimuli, but also occur probabilistically in the absence of specific trigger stimuli, suggesting that they can be internally generated (Goulding et al., 2008; Martin et al., 1999). The molecular and circuit mechanisms that generate stable behavioral states and transitions between them are incompletely understood. As there are similarities between these easily observable locomotor states and emotional and cognitive states, it is possible that their underlying mechanisms have common elements.
The best-characterized behavioral states are those associated with sleep, waking, or arousal levels, each of which is regulated by biogenic amine and neuropeptide neuromodulators. In mammals, arousal and waking states are stimulated by the neuropeptide orexin/hypocretin, as well as serotonin, histamine and norepinephrine, whereas the neuropeptides galanin and MCH promote sleep states (Saper et al., 2010). In Drosophila, the neuropeptide pigment dispersing factor (PDF), octopamine (an invertebrate amine related to norepinephrine) and dopamine promote waking states, whereas serotonin promotes sleep states (Sehgal and Mignot, 2011). In each system, the neurons producing biogenic amines and neuropeptides are heterogeneous, with diffuse projections to numerous brain regions, and the effects of the neuromodulators differ depending on the target cell and receptor being regulated. For example, 5HT2c serotonin receptor knockout mice have decreased NREM sleep and normal REM sleep (Frank et al., 2002), whereas 5HT7 serotonin receptor knockouts have decreased REM sleep and normal NREM sleep (Hedlund et al., 2005). In Drosophila, dopamine can promote or suppress arousal in different contexts (Lebestky et al., 2009). These results support the importance of neuromodulators as behavioral regulators, but indicate that arousal is defined not by a uniform neurochemical state, but by a circuit state.
Neuromodulation is widespread in the nematode C. elegans, whose compact nervous system consists of only 302 uniquely-defined neurons. In addition to classical neurotransmitters, the C. elegans nervous system contains over 100 neuropeptides, and the biogenic amine modulators serotonin, dopamine, tyramine, and octopamine (Chase and Koelle, 2007; Li and Kim, 2008). Neuromodulators have long been known to drive behaviors associated with different C. elegans feeding states. For example, serotonin and dopamine promote food-related behaviors such as eating, egg-laying and slow locomotion, whereas octopamine antagonizes these effects, mimicking the absence of food (Horvitz et al., 1982; Sawin et al., 2000). Food-related egg-laying and locomotion behaviors are also positively and negatively regulated by neuropeptides, including insulin-related and FMRFamide peptides (Li and Kim, 2008). Despite these strong effects on behavior, relatively little is known about the acute relationships between neuromodulatory signaling and neural circuit activity. Here, we examine the effects of neuromodulation on the spontaneous generation and maintenance of stable behavioral states in the presence of food.
Feeding C. elegans spontaneously switch between two discrete foraging states called roaming and dwelling (Ben Arous et al., 2009; Fujiwara et al., 2002). Roaming animals move quickly across a bacterial lawn and turn infrequently to explore the bacterial lawn, while dwelling animals move slowly and turn more frequently, remaining in a small area. Both states include common motor patterns such as forward locomotion, reversals, and turns, but these motor patterns appear in roaming- and dwelling-specific combinations that last several minutes and change through abrupt transitions (Fig. 1A). The proportion of time spent roaming increases when food is limited or low in quality, suggesting that roaming behavior reflects an exploration-exploitation decision about the value of the current environment (Ben Arous et al., 2009; Shtonda and Avery, 2006). Accordingly, the relative occupancy of roaming and dwelling states is regulated partly by internal metabolic status, and partly by environmental cues detected by specific sensory neurons. The cGMP-dependent protein kinase EGL-4 promotes dwelling (Fujiwara et al., 2002), a result of particular interest because its insect homolog, foraging, is a regulator of active and inactive foraging states (rover and sitter behaviors) in Drosophila and other insects (Osborne et al., 1997). The circuits that integrate these disparate cues are unknown.
To understand how long-lasting behavioral states emerge from the interactions between neural circuits and modulatory cues, it is necessary to identify the cells that produce the modulators, the cells that detect them and mediate their effects, and the acute and long-term effects of the modulators in each behavioral state. Here we show that long-lasting roaming and dwelling behaviors emerge from neural circuit regulation by two opposing neuromodulators, serotonin and PDF.
To gain insight into circuit mechanisms important for roaming and dwelling, we examined 57 mutants lacking individual neurotransmitter receptors, neuropeptide receptors, and gap junction subunits in a simple exploration assay, by observing the tracks that individual animals left on a lawn of E. coli over a 16-hour period (Fig. 1B). This behavior will be referred to as “exploration” to distinguish it from quantitative roaming and dwelling assays. On average, wild-type animals explored about 70% of the lawn during a 16-hour interval (Fig. S1A). Over the same interval, known mutants with elevated dwelling behavior explored less than 20% of the lawn, and the egl-4 mutant with elevated roaming behavior explored over 90% of the lawn (Fig. S1A), indicating that the exploration assay correlated with roaming and dwelling. At a false discovery rate (FDR) of 5%, six of 57 tested mutants displayed a change in this exploration assay without any obvious uncoordinated movement (Fig. 1C; Table S1). Among four mutants with elevated exploration, two were deficient in feeding (eat-2 and inx-20), supporting previous observations that decreased food intake promotes roaming behavior (Ben Arous et al., 2009), and a third mutant, lgc-47, had a small effect. The fourth mutant, mod-1, had elevated exploration but is known to feed on E. coli as efficiently as wild-type animals (Li et al., 2012), so it was examined further.
mod-1 encodes a serotonin-gated chloride channel (Ranganathan et al., 2000), one of five known serotonin receptors in C. elegans. Increased exploration in mod-1 mutants was rescued by a mod-1 cDNA expressed under the mod-1 promoter, confirming the involvement of the gene (Fig. 1D). To gain a more general view of serotonin signaling in exploratory behavior, we examined mutations affecting serotonin synthesis, reuptake, and other serotonin receptors. The rate-limiting enzyme for serotonin synthesis is encoded by tph-1 (Sze et al., 2000). Like mod-1 mutants, tph-1 mutants displayed increased exploration, which was rescued by a tph-1 genomic transgene (Fig. 1E). mod-1 tph-1 double mutants resembled the stronger tph-1 single mutant, suggesting that these genes act in a common pathway (Fig. S1B). Conversely, mutants with enhanced serotonergic function due to a mutation in the serotonin reuptake transporter MOD-5 displayed reduced exploration (Fig. S1C). None of the four other known serotonin receptors in the C. elegans genome (ser-1,4,5,7) had an effect (Table S1). This result distinguishes dwelling from general slowing of locomotion, since exogenous serotonin can slow C. elegans locomotion via either mod-1 or ser-4 receptors (Gurel et al., 2012). The congruent effects of tph-1 and mod-1, and opposite effect of mod-5, indicate that endogenous serotonergic signaling through the MOD-1 receptor suppresses exploration.
To define the precise effects of mod-1 and tph-1, we used quantitative locomotion assays in which animals were filmed during 90 minutes of movement on large (600 cm2) homogeneous bacterial lawns, then analyzed for the speed and turning parameters typical of roaming and dwelling (Ben Arous et al., 2009). Movement patterns scored over 10 sec intervals fell into two classes: a high speed/low turning class (roaming) and a low speed/high turning class (dwelling; Fig. S2A), each of which typically persisted over several minutes (Fig. S2C). To segment behaviors into roaming and dwelling states in a standardized fashion, we employed a Hidden Markov Model (HMM), which proposes that hidden roaming and dwelling states generate the observed behavior. A two-state HMM was fit to tracking data from wild-type animals (n=350) and validated by simulations (Fig. S2B; Extended Experimental Procedures). When applied to the locomotion data from wild-type animals, the model yielded a distribution of dwelling state durations that fit a single exponential (τ =482 sec), suggesting that transitions from dwelling to roaming states follow Poisson statistics (Fig. 2A). Roaming state durations fit a double-exponential distribution, indicating that there might be two populations of roaming states that differ in their durations (τ =75 sec and 520 sec) (Fig. 2B).
Using these quantitative methods, we found that the proportion of time spent roaming was higher in mod-1 and tph-1 animals than in wild-type, in agreement with the exploration assay (Fig. 2C1, 2D1). A previous study reported enhanced roaming of mod-1, but enhanced dwelling of tph-1 (Ben Arous et al., 2009); the discrepancy with our results probably resulted from a second linked mutation in the original tph-1 strain (Omura, 2008). An increased proportion of time spent roaming in mod-1 and tph-1 mutants could be caused by different durations of dwelling, roaming, or both states. Analyzing each class of event separately showed that both were affected: dwelling durations were decreased, and roaming durations were increased, in mod-1 and tph-1 mutants (Fig. 2C-D). Close inspection indicated that mod-1 mutants had two classes of dwelling events: a near-normal class (31%, τ =517 sec), and a severely shortened class that was only about a minute long (69%, τ=63 sec) (Fig. S2D-G). Thus, serotonergic signaling through MOD-1 affects both roaming and dwelling: it extends dwelling states and truncates roaming states.
To define the neural circuit by which serotonin controls exploratory behavior, we first identified the essential serotonin-producing neurons. To match serotonin levels to the normal range for each neuron, a single floxed copy of the tph-1 gene (including promoter, exons and introns) was inserted into a defined location on chromosome IV using the mosSCI technique (Fig. 3A) (Frokjaer-Jensen et al., 2008), and shown to rescue the tph-1 mutant phenotype (Fig. 3B). Cell-specific transgenes expressing Cre recombinase in ADF, NSM, HSN or ASG, as confirmed with a Cre reporter strain, were then used to eliminate serotonin production by individual neurons. Expression of Cre in either NSM or HSN (or both; Fig. S3A), but not in ADF or ASG, suppressed tph-1 rescue (Fig. 3B), but had no effect in a wild-type background (Fig. S3B). These results suggest that serotonin production by both NSM and HSN is required for normal adult exploration behavior.
NSM is a serotonergic/glutamatergic neuron pair in the pharynx whose axons have abundant secretory vesicles near the nerve ring, the major C. elegans neuropil. HSN is a hermaphrodite-specific serotonergic/cholinergic motor neuron pair in the mid-body that controls egg-laying; its identification as a regulator of exploration was unexpected, although its axon also enters the nerve ring. To confirm the involvement of HSN, we examined egl-1(n986dm) mutants, in which HSN neurons die early in development (Conradt and Horvitz, 1999). egl-1 animals had elevated exploration, supporting the conclusion that tph-1 in HSN suppresses exploration (Fig. S3C).
Neurons that respond to serotonin were sought based on the expression of mod-1. A mod-1 promoter fragment that rescued the mod-1 exploration defect (Fig. 1D) drove expression in AIY, RME, RID, RIF, ASI, DD1-6 and PVN neurons. We used an intersectional promoter strategy to rescue mod-1 in subsets of these neurons (Fig. 3C). The floxed mod-1 cDNA was placed in an inverted, inactive orientation under one promoter (either ser-2b or odr-2b) and was activated by Cre expression under a second promoter (mod-1), which inverted the cDNA to allow functional expression in overlapping cells. This design tested sufficiency of mod-1 expression rather than necessity as for tph-1, a choice made primarily for practical reasons. The enhanced exploration in mod-1 mutants was rescued by restoring mod-1 expression in AIY, ASI and RID, but not by restoring expression in ASI alone or in AIY and RID (Fig. 3D and data not shown). mod-1 expression in RIF was also sufficient for rescue (Fig. 3E).
To confirm the importance of mod-1-expressing cells in exploration, we killed neurons individually using a laser microbeam. Ablation of RID or PVN had no effect, but ablation of AIY, ASI or RIF reduced exploration (Fig. 3F). Thus AIY, ASI, and RIF neurons promote exploration, and as MOD-1 is an inhibitory serotonin receptor, these three neurons might be hyperactive in mod-1 mutants. Together, these results suggest that serotonin produced by NSM and HSN inhibits AIY, ASI and RIF through the MOD-1 serotonin-gated chloride channel to suppress exploration (Fig. 3G).
The genetic requirement for serotonin is consistent with two general models: the serotonergic circuit (Fig. 3G) could be tonically active in the presence of bacteria, providing permissive input onto a bistable circuit for roaming and dwelling, or dynamic changes in neurons that produce and detect serotonin might directly underlie bistability during constant exposure to bacteria. To distinguish between these possibilities, we monitored calcium levels as a proxy for neuronal activity in NSM and AIY, as examples of serotonin-producing and -detecting neurons. Calcium was monitored in freely-moving animals expressing the genetically-encoded calcium indicator GCaMP5 (Akerboom et al., 2012) using a custom-designed imaging system (Fig. 4A; Albrecht, Larsch, & Bargmann, in preparation). Although both dwelling and roaming behavior were observed using this imaging system, the constraints of the small viewing field truncated roaming states, which we will call “runs” rather than roams to respect this difference.
Locomotion parameters and NSM calcium signals were obtained for 20 wild-type animals over 384 minutes; a representative trace is shown in Fig. 4B. Across the dataset, NSM calcium levels were inversely correlated with locomotion speed (Fig. S4A). In individual animals, NSM showed sporadic, sustained (~60 sec) calcium peaks that were not associated with any apparent external event such as an encounter with the lawn edge (Fig. 4B). To understand the significance of these calcium peaks, event-triggered averages were used to align all traces based on NSM peaks (Fig. 4C) or behavioral transitions (Fig. 4D), allowing other parameters to follow passively. NSM calcium peaks correlated with a rapid decrease in locomotion speed and termination of forward runs (Fig. 4C; NSM::GFP controls are in Fig. S4B-C). Conversely, NSM calcium levels reached a local minimum when runs were initiated, falling for at least one minute before a run (Fig. 4D). These results indicate that NSM calcium peaks correlate with acute decreases in speed and suppression of forward runs.
In mod-1 mutants, NSM neurons had normal calcium peaks, but these peaks were less strongly associated with locomotion speed and forward run probability (Fig. 4E). Forward runs were initiated without a preceding reduction in NSM calcium levels, although calcium levels were still reduced during runs (Fig. 4F). These results indicate that mod-1 helps couple NSM calcium levels to behavior.
Calcium levels in the MOD-1-expressing AIY neuron were reciprocally correlated with behaviors compared to NSM. AIY calcium levels were consistently highest during forward runs (Fig. S4D, S4F), with gradual increases preceding forward run initiation (Fig. 4G; AIY::GFP controls in Fig. S4D-E). Together, these results indicate that a serotonin-producing neuron (NSM) and a serotonin-detecting neuron (AIY) both show dynamic changes in calcium levels correlated with behavioral states and transitions over a minutes-long time-scale.
Serotonin mutants have long roaming and short dwelling states (Fig. 2). To ask whether serotonin signaling directly drives roaming animals into the dwelling state, we performed optogenetic experiments to manipulate the serotonergic circuit in roaming animals. To depolarize serotonergic neurons, we used channelrhodopsin-2*(ChR2-C128S), an increased sensitivity variant that is activated by blue light at levels that do not induce endogenous behavioral responses in C. elegans (Berndt et al., 2009; Schultheis et al., 2011). One minute of ChR2-mediated depolarization of the serotonergic neurons (Fig. 5A, S5A) caused 70% of roaming animals to transition into dwelling states (vs 19% of controls). Activation of NSM alone had a similar but weaker effect (Fig. S5B). The induced dwelling states continued after ChR2-C128S inactivation by green light, with a long duration similar to endogenous dwelling states (Fig. S5C). tph-1 and mod-1 mutants were less responsive to ChR2 (Fig. S5D-E), supporting a role for serotonin in dwelling. In wild-type animals, hyperpolarizing serotonergic neurons with the green light-activated proton pump archaerhodopsin-3 (ARCH) (Chow et al., 2010) caused 24% of dwelling animals to initiate roaming states (vs 3% of controls); these roaming states ended when the green light was extinguished (Fig. 5B). Thus, acute activation of serotonergic neurons in roaming animals increases the probability of roaming-to-dwelling transitions, and acute inhibition of serotonergic signaling in dwelling animals transiently increases roaming behavior.
To mimic serotonin’s effect on the mod-1-expressing neurons, we hyperpolarized them with ARCH. The majority of roaming animals rapidly entered a dwelling state that outlasted the light stimulus (Fig. 5C). Conversely, ChR2-mediated activation of the mod-1-expressing neurons (Fig. 5D, S5F) or a subset of these neurons (RIF and AIY; Fig. S5G) caused animals that were dwelling to enter roaming states that were similar in duration to endogenously generated roaming states (Fig. S5H). The reciprocal effects of ARCH and ChR2 on roaming and dwelling transitions demonstrate a key role for the mod-1-expressing neurons in these behavioral states.
mod-1::ARCH also prolonged dwelling states. Two minutes of light exposure caused existing dwelling states to extend eight minutes longer than control dwelling states (Fig. 5E), indicating that inhibition of mod-1-expressing neurons maintains dwelling states as well as inducing them. Serotonin contributes to these long-lasting states, as dwelling states induced by mod-1::ARCH activation in mod-1 or tph-1 mutants were shorter than those induced in a wild-type background (Fig. S5C, S5I, S5J). Together, these experiments indicate that serotonergic signaling can initiate and maintain dwelling states.
Among the candidates tested in the initial exploration screen, effects reciprocal to those of the serotonin pathway were observed in mutants with disrupted pigment dispersal factor (PDF) neuropeptide signaling (Fig. 1C). C. elegans has two genes encoding PDF neuropeptides, and one gene encoding a PDF receptor; defects in PDF signaling cause uncharacterized locomotion defects in hermaphrodites and eliminate mate search behaviors in males (Barrios et al., 2012; Janssen et al., 2008). We found that pdf-1 mutants, pdf-1; pdf-2 double mutants, and mutants for the one known C. elegans PDF receptor, pdfr-1, had greatly reduced exploration behavior (Fig. 6A), but near-normal sinusoidal locomotion and responses to touch (Fig. S6A, Video S1-2). Quantitative analysis of pdfr-1 and pdf-1; pdf-2 mutants demonstrated that both strains had prolonged dwelling states, shortened roaming states, and reduced speed during roaming (Fig. 6B-C; Fig. S6B). The truncated roaming states in pdfr-1 mutants were reasonably well fit by a single exponential that resembled the shorter of the two wild-type roaming states (τ=101 sec, Fig. S6C-D).
The neural circuit for PDF neuropeptide signaling was analyzed by neuron-specific pdf-1 depletion using a floxed pdf-1 gene and Cre recombination (Fig. 6D). A pdf-1 cDNA under its own promoter rescued roaming and was expressed in AVB, SIAD, SIAV, PVP and AIM neurons (Fig. 6E, S6E-F). Cre expression in AVB, PVP and SIAV neurons eliminated rescue (Fig. 6E), and expression in subsets of these neurons had intermediate effects, identifying AVB, PVP, and SIAV as potential PDF-1 sources for roaming behavior.
The pdfr-1 gene is predicted to have two alternative promoters (Fig. S6G). Expression of pdfr-1 from the distal promoter fully rescued the pdfr-1 phenotype, as did a genomic fragment spanning both promoters; a genomic fragment with the proximal promoter did not rescue (Fig. S6H-I). The distal promoter served as a starting point to identify relevant pdfr-1-expressing neurons using an inverted, floxed pdfr-1 cDNA and intersectional promoters (Fig. 6F). Partial rescue of roaming was observed upon pdfr-1 expression in AIY, RIM and RIA neurons, and complete rescue with broader neuronal expression (Fig. 6G-H). These experiments suggest that pdfr-1 expression in AIY, RIM, RIA, and other neurons promotes roaming behavior.
PDFR-1, a secretin-receptor family G-protein coupled receptor (GPCR), stimulates Gαs to increase cAMP levels when expressed in heterologous cells (Janssen et al., 2008), like the Drosophila PDF receptor. To ask if cAMP mimics endogenous PDFR-1 signaling, we expressed a constitutively active version of the C. elegans adenylyl cyclase, ACY-1(P260S) (Saifee et al., 2011), in PDFR-1-expressing neurons. This transgene caused a dramatic increase in roaming behavior, prolonging roaming states and truncating dwelling states (Fig. 7A). Expression restricted to the AIY, RIM, and RIA neurons also increased exploration (Fig. S7A). pdfr-1::acy-1(P260S) rescued exploration in a pdfr-1 mutant (Fig. 7B), consistent with the possibility that the adenylyl cyclase acts downstream of pdfr-1.
As an optogenetic approach to mimic acute activation of PDFR-1, we expressed a blue light-activated adenylyl cyclase, BlaC, under the distal pdfr-1 promoter (Fig. 7C) (Ryu et al., 2010). One minute of blue light illumination caused 77% of dwelling pdfr-1::BlaC animals to switch to roaming states (vs 11% of controls) (Fig. 7D), which lasted at least as long as endogenous roaming states (Fig. S7B). This behavioral effect was abolished by a point mutation (D265K) that inactivates the adenylyl cyclase activity of BlaC (Fig. 7D) (Ryu et al., 2010).
PDF prolongs roaming and shortens dwelling states, whereas serotonin has reciprocal effects. To ask how these systems interact, we generated mod-1; pdfr-1 double mutants and compared their locomotion to wild-type, mod-1, and pdfr-1 mutants. mod-1; pdfr-1 double mutants had short dwelling states that resembled mod-1 single mutants, and short roaming states that resembled pdfr-1 single mutants (Fig. 7E). These results indicate that the prolonged roaming states observed in mod-1 mutants require pdfr-1, and the prolonged dwelling states observed in pdfr-1 mutants require mod-1 (Fig. S7C); in animals lacking both modulators, both dwelling and roaming states are short.
To further characterize the interaction between these two circuits, we examined the effects of optogenetic manipulations in mod-1; pdfr-1 double mutants. Imitating PDF signaling with pdfr-1::BlaC initiated roaming states in double mutants, and imitating serotonin signaling with mod-1::ARCH initiated dwelling states (Fig. 7F-G). Thus each modulatory circuit can independently regulate the initiation of its corresponding foraging state.
Our results show that serotonergic signaling through mod-1 initiates and extends dwelling states, whereas PDF signaling through pdfr-1 initiates and extends roaming states.
What does a complete neuromodulatory circuit look like, in its organization and its relationship to fast circuits? Despite the compact size of the C. elegans nervous system, the serotonin and PDF that regulate roaming and dwelling each have several important sources, and their receptors each act in several target neurons. These modulatory circuits include sensory neurons, interneurons, and motor neurons, but their organization does not follow the dominant sensory-to-motor hierarchy of the classical synaptic wiring diagram (Fig. 6I). The serotonin sources, NSM and HSN, are both motor neurons (although NSM may have sensory functions as well), and the PDF sources are interneurons; the targets include sensory neurons (ASI) and multiple interneurons. The serotonin and PDF circuits are mostly non-overlapping, intersecting only at the AIY neurons. Moreover, although serotonin- and PDF-expressing neurons have chemical and electrical synapses in the C. elegans wiring diagram (black arrows in Fig. 6I), these synapses do not overlap with the neuromodulatory connections inferred from our genetic mapping experiments (red and green arrows in Fig. 6I). Thus the neuromodulatory circuit for long-lasting behavioral states is essentially orthogonal to the synaptic connectivity diagram.
Extrasynaptic function of neuromodulators is well established in C. elegans and other animals (Chase and Koelle, 2007), but this need not imply that serotonin and PDF act as systemic hormones. All of the relevant neurons in this neuromodulatory circuit have processes in or near the C. elegans nerve ring, suggesting that diffusion over ~100-200 μm would be sufficient for their communication. The graded rescue of each neuromodulator by expression in specific neurons or groups of neurons suggests that both the quantity and the source of neuromodulators contribute to their function.
The circuits that drive short roaming and dwelling states in the absence of serotonin and PDF remain to be defined, as do the neurons required for the individual motor patterns of roaming and dwelling. Importantly, each neuron in the roaming and dwelling circuit has other behavioral functions, which will provide challenges for mapping the circuits. For example, the PDF-producing AVB neuron is essential for coordinated forward locomotion, which it controls through chemical and electrical synapses onto motor neurons (Chalfie et al., 1985). Thus AVB is required for coordinated forward movement at rapid timescales, and releases PDF-1 peptides to regulate locomotion over longer durations. Because neurons like AVB are multifunctional, precise manipulation of neuronal signaling molecules may be needed to disentangle the activities of neurons, synapses, and modulators on behavior.
Multifunctionality is also a property of the PDF and serotonin neuromodulators, which each affect a variety of behaviors by mobilizing different sets of neurons. For example, in males, PDF-1 produced by AIM neurons signals to the PDFR-1-expressing neurons URY, PQR, and PHA to stimulate mate search (Barrios et al., 2012). Although these neurons are all present in the hermaphrodite, they appear unimportant in roaming and dwelling behavior. Similarly, serotonin regulates feeding and egg-laying using receptors and neurons distinct from those defined here (Gurel et al., 2012; Li et al., 2012). In summary, neither neurons, nor neuromodulators, nor behaviors, are subsets of one another: each represents a separate functional organization within the nervous system.
Dwelling and roaming are persistent behaviors that last for several minutes. The endogenous calcium signals in serotonergic NSM neurons are long-lasting, but not identical to the behavioral states: NSM calcium transients last about one minute, but predict dwelling states for several minutes thereafter. Either optogenetic excitation of NSM or optogenetic inhibition of its MOD-1-expressing target neurons with ARCH was sufficient for persistent dwelling states. As ARCH should have a direct hyperpolarizing effect on the MOD-1-expressing neurons, a persistent circuit state for dwelling may be induced by transient neuronal inhibition. The failure of mod-1::ARCH to induce long-lasting dwelling states in mod-1 and tph-1 mutants suggests that continued serotonergic signaling maintains dwelling states.
The origin of the endogenous NSM calcium signals is unknown. NSM’s position in the pharynx suggests that it could detect cues associated with feeding; in addition, NSM calcium levels are indirectly regulated by attractive and repulsive odors and the biogenic amine tyramine (Li et al., 2012). Thus NSM could detect both nutrients and sensory cues relevant to roaming and dwelling. However, nutrients and sensory cues also regulate ASI and AIY, so there are many neurons in the serotonin and PDF circuits that provide possible entry points for regulation. Understanding the relationship between egl-4, which functions in sensory neurons to promote dwelling (Fujiwara et al., 2002), and the circuit described here might clarify how nutrient and sensory cues are coupled to behavioral transitions.
The C. elegans PDF receptor is coupled to Gαs and cAMP production, so for optogenetic imitation of PDFR-1 activation, we employed the bacterial light-activated adenylyl cyclase BlaC (Ryu et al., 2010). BlaC activation in PDFR-1-expressing neurons triggered roaming states that lasted for several minutes after light cessation should terminate its catalytic activity (Ryu et al., 2010). Persistent roaming could result from sustained cAMP levels, sustained activation of the cAMP-dependent protein kinase PKA, or sustained phosphorylation of PKA targets. Taking a broader view, the slow time course of neuromodulatory G-protein signaling is well-suited to convert short-lasting electrical signals into longer-lasting biochemical and behavioral states.
A role for PDF signaling in promoting roaming, an arousal state in C. elegans, is reminiscent of the ability of PDF to promote arousal during waking states in Drosophila. In flies, PDF-expressing neurons integrate the circadian cycle, light levels, and modulatory octopamine and dopamine inputs to regulate PDF release and arousal (Sehgal and Mignot, 2011). A mammalian neuropeptide, vasoactive intestinal peptide (VIP), has a similar role in stimulating arousal downstream of light inputs and neuromodulation in the suprachiasmatic nucleus (Vosko et al., 2007). PDF receptors and VIP receptors are similar in sequence, suggesting that these neuropeptide systems may have similar or even conserved roles in arousal states.
More generally, the circuit logic of roaming and dwelling resembles the logic of hypothalamic and brainstem circuits that control discrete mammalian sleep and wake behaviors (Saper et al., 2010). The transitions between wake, REM, and non-REM sleep are controlled by neuropeptides and biogenic amines produced in hypothalamic and brainstem nuclei. As is seen in roaming and dwelling, each state inhibits the others in a switchlike fashion, and loss of the neuromodulators leads to destabilized and truncated behavioral states. We suggest that these features may be signatures of a variety of discrete behavioral states.
Information about strains and plasmids used in this study is available in Extended Experimental Procedures and Table S2. Nematodes were grown on agar with Nematode Growth Medium (NGM) and OP50 bacteria. Mutant strains were backcrossed to a common strain (N2) to remove unlinked mutations prior to analysis (Table S1).
To measure exploration behavior, individual L4 animals were picked to a 35 mm agar plate uniformly seeded with E. coli strain OP50. After ~16 hours, plates were superimposed on a grid containing 3.5 mm squares and the number of squares entered by the worm tracks was manually counted. Tracks could enter a maximum of 86 squares. In Figs. S1B and S3A, assays were performed on 60 mm plates for 13 hours to distinguish between mutants with high levels of roaming (maximum number squares entered, 175). Transgenic and mutant strains were always compared to control animals assayed in parallel. For the candidate gene screen (Fig. 1C), five animals were tested per data point, and the false discovery rate was calculated using the Benjamini-Hochberg method. For other genotypes, 10-25 animals were tested per data point. All plates were scored by an investigator blind to the genotype of the animals.
High-resolution analysis of C. elegans locomotion was performed on one-day-old adult animals exploring 600 cm2 agar plates seeded with OP50. Animals were recorded for 90 minutes at 3 frames per second (fps). Worm trajectories were extracted from videos using custom Matlab scripts that calculated the speed and angular speed of each animal. Measurements were averaged over 10 second intervals, which easily distinguished roaming and dwelling intervals (Fig. S2A) (Ben Arous et al., 2009), allowing us to describe the trajectory of each animal as a sequence of discrete roaming and dwelling intervals. For each experiment, sequences from control animals were used to optimize a two-state HMM. After optimization, the most probable state path of each animal (for all genotypes) was calculated by applying the Viterbi algorithm to the sequence of discrete roaming and dwelling intervals for that animal. Additional information is available in Extended Experimental Procedures.
L4 experimental and control animals were picked to OP50-seeded NGM plates containing 50 μm all-trans-retinal. On the next day, adult animals’ locomotion was recorded using the setup described above. During video recordings, LED illumination was used to expose animals to blue (455 nm) or green (530 nm) light. For ChR2*(C128S), blue light delivery was always followed by 60 sec of green light delivery to ensure full inactivation of ChR2*(C128S). We determined the behavioral states of animals using the procedure above and then aligned behavioral measurements to periods of light exposure. Additional information is available in Extended Experimental Procedures.
Transgenic animals expressing GCaMP5 (Akerboom et al., 2012) in NSM or AIY neurons were assayed as one-day-old adults. Transgenic lines were generated in a lite-1 background to prevent behavioral responses to blue light (Edwards et al., 2008). Flat NGM agar pads (~1 mm thick) on a glass microscope slide were seeded with a ~5 mm diameter OP50 lawn. Animals were then picked to the pad, which was sealed in a small chamber to prevent evaporation. 30-minute videos of the animals were recorded at 10 fps using 10 msec light exposures. In this arena, animals sometimes left the region being recorded and returned; all available data were used in subsequent analyses. Additional information is available in Extended Experimental Procedures.
We thank H.R. Horvitz, S. Mitani, the National BioResource Project (NBRP), and the Caenorhabditis Genetics Center (CGC) for strains; L. Looger for GCaMP5; A. Gordus and A. Katsov for help with data analysis; I. Shachrai for initiating optogenetic experiments; Y. Saheki, A. Bendesky, J. Garrison, J. Greene, and other members of our laboratory for helpful discussions; and S. Kuehn and R. Axel for comments on the manuscript. This work was supported by the G. Harold and Leila Y. Mathers Foundation and the Ellison Medical Foundation (C.I.B.), a Helen Hay Whitney postdoctoral fellowship (S.W.F.), NIH grant GM07739 (E.Z.M.), and a Boehringer Ingelheim Fonds PhD Fellowship (J.L.). C.I.B. is an Investigator of the Howard Hughes Medical Institute.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.