Leech Behavioral Choice
Summary and Keywords
New techniques for recording the activity of many neurons simultaneously have given insights into how neuronal circuits make the decision to perform one of many possible behaviors. A long-standing hypothesis for how behavioral choices are made in any animal is that “command neurons” are responsible for selecting individual behaviors, and that these same neurons inhibit the command neurons that elicit other behaviors. In fact, this mechanism has turned out to be just one of several ways that such decision-making is accomplished. In particular, for some behavioral choices, the circuits appear to overlap, sometimes extensively, to perform two or more behaviors. Making decisions using such “multifunctional neurons” has been proposed for large neural networks, but this strategy appears to be used in relatively small nervous systems, too. These simpler nervous systems can serve as useful test systems to test hypotheses about how the dynamics of networks of neurons can be used to select among different behaviors, similar to the mechanisms used by leeches deciding to swim, shorten, crawl, or feed.
Based upon observations of the natural behaviors of animals—mostly fish and birds—the ethologist Nikolaas Tinbergen proposed that animals chose between competing behavioral choices, such as food-getting and nest-making, by “inhibitive interactions” among centers in their brains that are responsible for the different behaviors (Tinbergen, 1951). Adding in the sensory input to these decision makers and the output onto the motor patterns that are produced produces a flow diagram like those shown in Figure 1. The inputs from sensory receptor neurons—both intrinsic and extrinsic—is processed in sometimes very complex ways (Felleman & Van Essen, 1991). This processed sensory information provides the input onto the decision makers that, when the pattern of processed sensory input is appropriate, activates a decision maker, which in turn activates the proper central pattern generator (Delcomyn, 1980). Each pattern generator produces a spatial and temporal pattern of motor neuronal activity that constitutes a distinct behavior. The discovery, in the 1960s and 1970s, of many “command neurons” whose activity elicits complex behaviors in many animals (an example of one that elicits leech swimming is shown in Figure 2) suggested a site for Tinbergen’s inhibitive interactions: the same command neurons that elicit behaviors might also inhibit each other, so that only one behavior at a time can be expressed (Figure 1). In other words, these command neurons were postulated to be the decision makers. Evidence for such interactions was first presented for feeding vs. withdrawal to touch in a marine slug (Kovak & Davis, 1977).
A more extensive study in a fresh-water snail showed that the circuitry that produces touch-induced shortening does inhibit feeding command neurons (Pirger et al., 2014). Modeling studies, both in invertebrates (Edwards, 1991), and vertebrates (Wang, 2008) have shown that such reciprocal inhibition among command neurons can reproduce many of the features of behavioral choice. The leech central nervous system provides a useful test case to determine whether a similar mechanism—reciprocal inhibition between command neurons—is responsible for choosing among several different leech behaviors. For more than 50 years, the leech CNS has served well as a system to study such issues as membrane biophysics, synaptic physiology, mechanosensation, neuromuscular control, specificity in regeneration, mechanisms of development, and neuronal circuits that produce a variety of behaviors. These topics have been reviewed at a rate of about once a decade (Muller, Nicholls, & Stent, 1981; Drapeau, Kristan, & Shankland, 1995; Kristan, Calabrese, & Friesen, 2005; Wagenaar, 2015). These studies of the individual behaviors provided the basis for studying the mechanisms of decision-making in the leech.
Leech Behavioral Choice
The leech behaviors used for studying behavioral choice were feeding (Gaudry & Kristan, 2009) plus four different responses to tactile stimulation: shortening, local bending, crawling, and swimming (Figure 3). The studies of touch-induced behaviors showed that the leech body has three overlapping behavioral regions: approximately the front third responds to moderate touch by shortening, the middle third produces local bending, and the rear third produces locomotory responses, either crawling or swimming. Although not every neuron is known for each of the five behaviors, something is known about each of them, and a great deal is known about some of them (Kristan et al., 2005). Decision-making studies have been done primarily by presenting two stimuli simultaneously that individually produce different, mutually-exclusive behaviors. (For instance, it is mechanically impossible for a leech to both shorten and swim at the same time.) In principle, there are a variety of possible outcomes: some kind of combined behavior might occur, or no response at all, or there might even be a third behavior, one that is never produced by either stimulus on its own. In fact, for all the pairs of behaviors tested, one of the two behaviors always wins out. For instance, simultaneous presentation of stimuli that cause shortening and swimming always elicits shortening (Shaw & Kristan, 1997), and feeding inhibits all responses to mechanosensory input (Gaudry & Kristan, 2009). The interactions between swimming and crawling are complex (Briggman, Abarbanel, & Kristan, 2005). These three combinations (swimming inhibiting all behaviors, shortening inhibiting swimming, and swimming vs. crawling) will be treated separately because each of them uses a qualitatively different mechanism of choice. Interestingly, none of them use the hypothesized reciprocal inhibition between command neurons (Figure 1).
Feeding Overrides Other Behaviors by Inhibition
Medicinal leeches are attracted to a food source both by warmth and by chemicals on the host’s body surface (Dickinson & Lent, 1984). When a leech detects food and starts to feed, it cannot be deterred by mechanical stimuli of any intensity, even at levels that produced strong responses in nonfeeding leeches (Misell, Shaw, & Kristan, 1998). To study the neural basis of this response, a semi-intact preparation was used (Gaudry & Kristan, 2009), consisting of a fully intact front half and a back half with only the nerve cord left intact (Figure 4A). Both intracellular and extracellular recordings were made in the exposed ganglia before, during, and after the intact front end of the animal fed. A variety of command neurons for different behaviors were recorded and found to be completely unaffected by feeding. In addition, stimulating a swim command neuron induced the swimming motor pattern in the back end while the front end continued to feed unabated. In the neurons receiving mechanosensory input (i.e., “sensory processors” in Figure 1), however, a big difference was seen: The synaptic potentials diminished greatly in size during feeding (Figure 4B,C). To test whether this decrease was due to changes in the presynaptic (sensory receptor) or postsynaptic (sensory processor) neuron, pairs of closely spaced sensory action potentials were elicited in the sensory neuron, and the ratio of the amplitudes of the resulting EPSPs in the postsynaptic neurons were measured. Before feeding, this ratio did not differ from 1, but it increased by about 50% during feeding. Such a positive “paired-pulse ratio” (PPR) is taken as strong evidence for an increase in the amount of transmitter released in the presynaptic terminals by the second action potential (Schulz, Cook, & Johnston, 1995). Despite this facilitation of the second EPSP, the amplitudes of the synaptic potentials were significantly lowered by feeding. This same variation in synaptic potentials and PPR was seen in a number of interneurons involved in different leech behaviors, which lowered the response to below any behavioral threshold. These studies showed that feeding overrides all other touch-induced behaviors by presynaptic inhibition of the terminals of the mechanosensory neurons (Figure 5). Effectively, the animal becomes numb to mechanical stimulation as it feeds. This mechanism of behavioral choice differs from the hypothesized “mutual inhibition among command neurons” (Figure 1), but it does use inhibition to block behavioral responses to touch.
The source of the presynaptic inhibition of the sensory terminals is not known, but the effect can be mimicked by bath application of serotonin to the exposed nerve cord, and the effects of both feeding and bath application of serotonin can be negated by applying mianserin, a substance that blocks a serotonergic inhibition in C. elegans (Ranganathan, Cannon, & Horvitz, 2000). Interestingly, serotonin has an excitatory effect on feeding (Lent, 1985), so this transmitter appears both to aid in the production of feeding and to block the expression of other behaviors.
Feeding does not turn off touch-induced behaviors in all types of leeches. In a study of six different leech species, all of the sanguivorous (blood-sucking) leeches showed feeding-induced insensitivity to touch, but none of the carnivorous (meat-eating) leeches did (Gaudry, Ruiz, Huang, Kristan, & Kristan, 2010). This difference may reflect their metabolic life-style: sanguivorous leeches take a small number of huge meals—up to 10 times their pre-feeding body weight (Palmer, Barnett, Copado, Gardezy, & Kristan, 2014)—whereas carnivorous leeches take small meals every day. Because sanguivorous leeches may have limited opportunities for full blood meals, feeding appears to be their top behavioral priority when a meal is available.
Shortening Overrides Swimming by a Combinatorial Code
When its front end is touched, a leech produces a brisk shortening of its whole body, even if the leech is in the midst of a locomotory behavior (i.e., swimming or crawling) (Shaw & Kristan, 1995). Several command neurons were known that initiate swimming (Kristan et al., 2005), so they were good candidates for being inhibited by shortening (Shaw & Kristan, 1995, 1997). The command neurons were of two sorts, based on the duration of the intracellular stimulation needed to produce swimming (Figure 6A): Trigger neurons (e.g., cells Tr1 and SE1) require just a short stimulus duration—1 second or less (Brodfuehrer & Friesen, 1986)—whereas Gating neurons (like cell 204, Figure 2) must be constantly depolarized to keep the swim motor program active (Weeks & Kristan, 1978). The initial experiments on the choice between swimming and shortening (Shaw & Kristan, 1995) showed that a stimulus that elicits shortening does strongly inhibit cell 204 (Figure 6B,C). Recording from other Gating neurons (e.g., cell 61) or the Trigger Neurons (Tr1 and SE1) painted a quite different picture, however; instead of being inhibited, all of these neurons were excited by the stimulus that causes shortening (Figure 6C). This means that three of the four neuronal types that, when activated by themselves cause swimming, are excited by a stimulus that causes shortening, a behavior that is incompatible with swimming. These neurons may be part of a group of decision-making neurons that select behaviors by a combinatorial code, so that the activation of the pattern SE1+Tr1+61+204 produces swimming, but SE1+Tr1+61-204 produces shortening (Figure 7A). This circuitry is quite different from the hypothesized inhibition between command neurons (Figure 1) in that no single neuron involved in making the behavioral choice is activated in just a single behavior—every one of them is multifunctional. This multifunctionality means that, to determine which behavior is being selected, one needs to know which combination of neurons is active and which are inactive (Figure 7B).
Swimming and Crawling Interact by Dynamic Activity of Multifunctional Neurons
Leeches locomote by either swimming or crawling. Swimming covers distance faster, but leeches pay much more attention to sensory cues—such as tactile, vibratory, and visual—as they crawl (Harley, Rossi, Cienfuegos, & Wagenaar, 2013). Hungry leeches tend to swim rather than crawl (Lent & Dickinson, 1984), as do leeches that have been agitated (Willard, 1981), and fully-fed leeches crawl but do not swim (Gaudry & Kristan, 2010). In a “basal” state—neither hungry nor fully sated, and not provoked—leeches are about equally likely to swim or to crawl in response to moderate touch to their posterior (Palmer et al., 2014). This swim/crawl probability is the same in an isolated leech nerve cord (Figure 8A,B), using electrical stimulation of nerves to stimulate mechanoreceptors and extracellular recordings of motor neurons from nerves to record the motor programs for the two behaviors (Briggman et al., 2005).
Voltage-sensitive dyes were used to identify neurons that influenced the swim/crawl decision, by recording the activity of many of the neurons on one surface of a ganglion (Figure 8B) in an isolated leech nerve cord (Briggman et al., 2005). Surprisingly, most of the neurons in the ganglion were activated by the stimulus (Figure 8C,D) and, even more surprisingly, the patterns of activity in most of the neurons were quite similar whether the nerve cord produced the swimming or the crawling motor pattern. The motor neuronal activity patterns (traces below the dye recordings in Figure 8C,D) show that there is a “preparatory period”—seen most clearly in the Swim data—which begins during the nerve stimulation period (indicated by the pair of vertical black lines) and continues for several seconds. In intact animals, there is a co-contraction of antagonistic muscles that stiffens the animal’s body, thereby creating its skeleton (called a muscular hydrostat), which allows the patterned activity to produce the swimming and crawling movements (Wilson, Skierczynski, Blackwood, Skalak, & Kristan, 1996; Kristan et al., 2000). At some point during the preparatory period, the nervous system activates the pattern generator for either swimming or crawling. To find the neurons that might be making the swim/crawl decision, the activity patterns of all the recorded neurons were compared between the onset of stimulation and before the motor patterns for swimming and crawling were expressed (Figure 9, bottom recording).
The top three traces in Figure 9 gives examples of the three different kinds of neuronal activity patterns found: those whose membrane potential trajectories were the same during swimming and crawling (“individual nondiscriminators”), those whose trajectories differed only after the behavior began (“late individual discriminators”), and those whose trajectories differed before the behavior began (“early individual discriminators”). For each neuron, 10 trajectories are overlain that produced either swimming (blue traces) or crawling (red traces) in response to a single presentation of a stimulus (gray box on the time axis). In this example, the motor neuronal recording is for swimming, indicated by the four repeated bursts of motor neuronal spikes in the bottom trace (Kristan et al., 2005). Note that the red and blue trajectories of the “late individual discriminator” separate only after swimming motor pattern is expressed (area marked “Make rhythmic movements”), whereas the “early individual discriminator” traces separate during the prelocomotory period (“Elongate, flatten”) when an intact animal would be establishing its hydrostatic skeleton and getting its body into shape to produce coordinated locomotory movements. The area marked “Swim!” represents the time in which the leech nervous system makes the decision to act (in this case, to swim). The existence of an even earlier event (“Get ready!”) is signaled by increased motor neuronal activity (bottom trace) before the decision is made to make any particular behavior. In effect, this activity at an early time prepares the body for subsequent behaviors. The interneurons in this pathway respond quickly to a tactile stimulus that produces any behavior (shortening, swimming, crawling) and help to prepare the body to perform one of these behaviors (Frady, Kapoor, Horvitz, & Kristan, 2016). Interestingly, the previously identified command neurons turned out to be “early individual discriminators,” i.e., they were activated only after the behavioral choice was complete. These neurons, therefore, seem to help in activating the pattern generators and in maintaining their activity, but are not part of decision-making per se.
More than half of the recorded neurons showed no significant difference in their activity patterns in the preparatory period (Briggman et al., 2005). Of the ones that did show a difference, most of them were “late individual discriminators” (Figure 9), differing in their swim/crawl trajectories only after the pattern generators were activated. On average, 19 neurons of the nearly 150 recorded neurons were “early individual discriminators,” showing a statistically different activity pattern before either swimming or crawling motor patterns were detectable. The recordings of the earliest discriminators diverged about 340 msec. before the first motor neuronal burst in swimming motor activity. Disappointingly, activating or inactivating any of these neurons individually did not bias the behavioral choice toward swimming or crawling. This result opened the possibility that a more subtle mechanism was responsible for the choice to swim or to crawl.
One possible mechanism is that small differences in the massed activity of many neurons might be responsible for this behavioral choice. To test this possibility, the joint dynamics of all the recorded neurons were plotted (Figure 10). To understand how this discrimination was made, consider first a simple example: plotting the trajectory of the activity of just three cells, C1, C2, and C3 (Figure 10A). The activity (i.e., spike rate or membrane polarization changes) of each cell is plotted on three axes. At the first unit of time (leftmost point) the activity of C2 and C3 are both in their midrange and C1 is at the bottom of its range. At the next time point, the activity of C3 is decreased, but the activity of C1 and C2 does not change very much. A line is drawn to connect these two points. The plot continues for three additional time periods, in which the C1 activity increases, and both C2 and C3 decrease. Note that the length of the line varies between the points, reflecting the amount of variation in each of the cell’s activity in the particular time period.
The same sort of analysis can be done for any number of neurons, with an axis for each of them. In the data used to make the remainder of this figure, there were 143 neurons, so the graph would have 143 axes. A 143-dimensional space is impossible to visualize, but these data can be readily handled mathematically. To make the data easier to conceptualize and visualize, Principal Component Analysis (PCA) was used, which is an increasingly standard technique for reducing the number of dimensions (Briggman, Abarbanel, & Kristan, 2006; Shenoy, Sahani, & Churchland, 2013). PCA components are generated by drawing a line through the complex space (143-D, in this case) that minimizes the distance of all the points to the line. (If there were only two dimensions, this would be the linear regression line.) This first component, called PC1 (Figure 10B), is visualized as the average activity of each of the 143 neurons during 100 msec. intervals, using data from each of the neurons during 14 response periods, each of which was 10 seconds long, i.e., the average value of 1400 points (10 samples/sec X 10 sec X 14 iterations) for each of the 143 neurons. The PC1values for all the neurons are positive, reflecting the fact that all the neurons were excited by the stimuli. The second component, PC2, is the distance from a line that, for mathematical tractability, must be perpendicular to the first line, and is oriented to minimize the remaining variability. Note that some of the neurons in PC2 are positively directed relative to this line and others are negatively oriented, reflecting the fact that some were, on average, excited and others were inhibited during the 10-second period. In the same way, PC3 was calculated (as a line perpendicular to both the PC1 and PC2 lines), which has its own characteristic activity patterns. Because of the way they are calculated, PC1 inherently accounts for more variability in the data than any other component, PC2 accounts for the next most, and so on. For each experiment, the number of PCs is equal to the number of neurons. If all 143 PCs were calculated for this data set, no information would be lost: The abstracted data contains the same information as the original data set. Ideally, however, enough variation is accounted for by the first three PCs so that a 3-D graph will show the most significant features of the data. This situation proved to be true for these data (Figure 10C).
Figure 10C shows PC1, PC2, and PC3 plotted along the three axes for 10 different responses to identical nerve stimulation. In effect, this plot is similar to plotting the activity of three neurons (Figure 10A), except that the location of each point on the trajectory is a measure of how closely the activity of all the neuronal activity matches each of the PCs (Figure 10B). On each plot, time proceeds from lower left to upper right. Three kinds of time points are marked on each plot: the start of the recording (“1st Frame,” indicated by a star), the time when the stimulus was delivered (filled circle), and the time when the first burst of action potentials was seen in those cases when swimming resulted (filled square). On the far right of Figure 10C are the extracellular recordings for each of the 10 recordings in this data set. The top four traces show cases in which the stimulus elicited swimming, and the bottom six elicited crawling. The trajectories of the 10 responses cluster into two groups, one in which PC1 predominates and the other in which PC2 and PC3 predominate. The three PC1-dominated trajectories (blue, magenta, yellow) were generated when swimming occurred, and most of the PC2/PC3-dominated trajectories led to crawling. (The one exception is the green trace, which—both by its trajectory and by its motor neuronal pattern—suggests that the initial response was to crawl, but then it switched to swimming. This could be the leech equivalent of “changing its mind,” but it did not occur sufficiently frequently to study in any detail.)
To determine the earliest time at which the trajectories for swimming and crawling diverged (Figure 10C), we used Linear Discriminate Analysis (LDA). (Geometrically, this amounts to sliding a 2-D plane down into the trajectories to determine the earliest time at which the two clusters of trajectories diverge, i.e., until most of the points leading to swimming are on one side of the plane and the points leading to crawling are on the other side.) Typically, for all the data sets, this time was about 90 msec. earlier than any single “early individual discriminator” reliably predicted the behavioral choice. The LDA also made it possible to find the neurons whose activity contributed most to the inclination of the 2-D plane, i.e., those neurons that contributed most strongly to the distinction between behaviors. Surprisingly, none of the “individual early discriminators” were among the potential decision makers (Figure 9) identified by the combination of PCA and LDA.
To test more strongly whether the PCA-identified decision makers were responsible for behavioral choice, individual neurons were either depolarized or hyperpolarized during alternating presentations of the stimulus. These manipulations did affect the choice. The most powerful effect was shown by cell 208: Depolarizing it during the presentation of the nerve stimulation strongly biased the subsequent behavior toward crawling, and hyperpolarizing cell 208 made swimming much more likely (Briggman et al., 2005). These results strongly suggest that the leech’s decision to swim or to crawl is made by a small number of neurons (around a dozen) in each segmental ganglion. Surprisingly, the activity trajectories of none of these neurons ever discriminated between the two behaviors, either early or late, i.e., they were “individual nondiscriminators” (Figure 9). This means that any difference in the activity that was responsible for the decision-making was below our ability to detect in single neurons, but could be ferreted out using PCA and LDA.
In addition to the decision makers in the segmental ganglia, there are neurons in the anterior brain that influence the swim/crawl behavioral choice. One of them—named R3b1 based upon the location of its soma in the brain—can activate locomotion behaviors (Esch, Mesce, & Kristan, 2002; Esch & Kristan, 2002). Like cell 204 (Figure 2), R3b1 can turn on swimming, but remarkably, it can also turn on crawling. In an isolated nerve cord, which behavior is activated is random, but in a semi-intact preparation, which behavior occurs depends upon conditions. For instance, if the intact part of the body is in shallow water (i.e., it just covers the leech’s body), the leech crawls in response to R3b1 activation, whereas if it is in deep water (≥5X the thickness of the body), activating R3b1 produces swimming. This neuron has been shown to be both necessary and sufficient to activate crawling behavior in the isolated nerve cord (Puhl, Masino, & Mesce, 2012). In fact, the swim trigger neurons Tr1 and SE1, although initially defined as swim trigger neurons (Brodfuehrer & Friesen, 1986), can also sometimes trigger crawling in an isolated nerve cord (Brodfuehrer et al., 2008). The nature of the interactions between these neurons in the anterior brain and the group discriminators in the segmental ganglia has yet to be determined.
In addition to the multifunctionality of the decision-making neurons (Figures 7, 9), there is a great deal of overlap between the pattern-generating circuitry for swimming and crawling (Briggman & Kristan, 2006). Optical recordings from nearly all the neurons (average of 350) in a ganglion showed that 188 neurons oscillated in some phase with the crawling rhythm, 90 oscillated in phase with swimming, and the swim-oscillating neurons are mostly (except for six of the 90) a subset of the crawl-oscillating neurons (Figure 11A). As an example, the intracellularly recorded membrane potentials of two such neurons are nearly overlapping during three cycles of crawling (Figure 11B), but become antiphasic when the motor pattern switches to swimming (Figure 11C). Some of the neurons recorded in this study were motor neurons that do not contribute to the generation of either motor pattern (Ort, Kristan, & Stent, 1974), but the fact that a high percentage of neurons that oscillate in some phase with both behaviors indicates that the two behaviors must share a large number of pattern-generating neurons. In addition, stimulating some newly identified interneurons showed that they could modify the timing of both the swimming and the crawling motor pattern (Briggman & Kristan, 2006), which qualifies these neurons as members of the pattern generators for both locomotory behaviors.
The large extent of overlap in the functions of the neurons involved in leech crawling and swimming suggests a conceptual diagram (Figure 12A) that is quite different from previous ones (Figures 1, 5, 7A). The circuitry of the swim/crawl behaviors is identical from sensory neurons through decision makers, and even the pattern generators for the two behaviors overlap a great deal. Some of the motor neurons (e.g., those to the longitudinal muscles) are shared in the two behaviors, although their patterns of activity differ in swimming and crawling (Figure 11B,C), and other motor neurons (e.g., those to the circular muscles) are active in one behavior (crawling) but are silent in the other one (swimming). The major difference in the circuitry for deciding and producing the two behaviors, therefore, is not in the neurons used (i.e., not the connectome) but in the activity (i.e., the dynamics) of largely the same circuit.
A way to conceptualize how the same circuitry can produce two different behaviors is shown in Figure 12B. In this diagram, the axes could be measures of the activity of three individual neurons (as in Figure 10A) or PCA values representing activity levels in many neurons (Figure 10B). In Figure 12B, the activity level at rest is low in most neurons and varies randomly. When a stimulus is given (represented by the thunderbolt) the activity level increases along one of two trajectories, one of which ends in a stable activity pattern that produces one recognizable behavior (e.g., swimming) and the other trajectory leads to a different stable pattern (e.g., crawling). As drawn, the two trajectories follow a similar path for some time before diverging into one of the two stable states. During this period of similar trajectories, telling which pathway will be selected is difficult; it may require monitoring the activity patterns of many neurons at once, as was true in the data of Figure 10C. Another possibility is that, during the time marked as “Decision-making” on the diagram, the nervous system has not yet committed to one behavior or another; i.e., it has committed to performing a locomotory act but has not yet decided which one to produce. In this scenario, the choice to swim or to crawl is not made until the activity patterns diverge, heading toward the stable activity patterns recognizable as swimming or crawling.
A more accurate representation of swimming and crawling might be an elaboration of Figure 12B, in which the activity patterns are represented as being widely separated. Instead, the trajectories might share overlapping locations in space, with interlocking trajectories (Figure 12C) (Briggman & Kristan, 2008). Since both swimming and crawling are rhythmic behaviors, each cycle of their activity patterns in this diagram is represented by a closed loop (e.g., the green loop during crawling and the red loop during swimming). Successive cycles would follow one of the two loops, with some variability. (Such stable loops of activity are called “dynamic attractors” or just “attractors.”) Either spontaneously, or by some external stimulation, the activity pattern could get knocked out of its stable path into a region of activity that is unstable. The activity state might fall back onto the same trajectory, which would be seen as a glitch in an ongoing behavioral rhythm. Alternatively, the activity might—as represented by the arrow in Figure 12C—fall into the other stable activity pattern, which would be seen as a switch from one behavior (in this case, crawling) into the other one (swimming).
Multiple Mechanisms of Behavioral Choice
The neuronal that mechanisms have been characterized for how leeches choose between three different pairs of behaviors (shortening vs. swimming, crawling vs. swimming, and feeding vs. all other touch-elicited behaviors) found qualitatively different mechanisms of behavioral choice for each of them. Not only were they different from one another, they were also different from the hypothesized “inhibition between command neurons” that has been a recurring proposal since its inception (Tinbergen, 1951). In fact, the presynaptic inhibitory mechanism by which feeding turns off all other behaviors (Gaudry & Kristan, 2012) shares many of the features of the Tinbergen model (it uses inhibition to keep any other behavior from interfering with feeding) but it differs in the site of the inhibition: Rather than inhibiting command neurons for all the behaviors, feeding behavior accomplishes the same goal by simply inhibiting the presynaptic terminals of one type of neuron, the mechanosensory neurons.
The “combinatorial code” by which leeches choose shortening over swimming (Shaw & Kristan, 1997) shows that neurons can be multifunctional. One indication of this multifunctionality is that three types of neurons that individually command swimming are also activated when shortening occurs (Figure 6). One might argue that one of the command neurons (cell 204) is inhibited during shortening, so that it is not so different from the “inhibition between command neurons” model. Two additional observations argue against this interpretation, however. First, two of the command neurons that are active in both behaviors (the trigger cells SE1 and Tr1 [Brodfuehrer & Friesen, 1986]) excite cell 204 rather than inhibiting it, and the other such neuron (cell 61, another gating neuron [Nusbaum & Kristan, 1986]) has no connection to cell 204 at all. Hence the inhibition of cell 204 during shortening is generated by some other, as yet unidentified, neuron. Second, although it is turned off during shortening, cell 204 is rhythmically activated in phase with crawling. Hence, a combinatorial coding scheme (Figure 7B) seems to be a more accurate description for how shortening overrides swimming than does inhibition among the command neurons. In fact, the extensive overlap in circuitry between swimming and crawling (Figure 11A) means that multifunctional neurons may be the rule, rather than the exception, in the leech nervous system. The third scheme—different dynamics among the same sets of neurons as a way to select between swimming and crawling—was wholly unexpected. This kind of mechanism was proposed for large assemblies of neurons that might lock into different stable modes of activity (e.g., Sussillo & Abbott, 2009; Cannon, Kopell, Gardner, & Markowitz, 2015), but not for a nervous system as simple as that of the leech.
Using different dynamics among the same neurons to select and produce different behaviors also allows for modification of the choice. For instance, when its nervous system receives a barrage of serotonin, a leech—or its isolated nervous systems—is much more likely to swim (Willard, 1981), and when bathed in dopamine, it predominately crawls (Crisp & Mesce, 2006; Puhl & Mesce, 2008). In addition, leeches in very shallow water (i.e., a depth less than their body diameter) will readily crawl but never attempt to swim (Palmer et al., 2014). Also, hungry leeches swim at the slightest provocation (Lent & Dickinson, 1984), and well-fed leeches, in their extremely bloated state, crawl but never swim (Gaudry & Kristan, 2010; Palmer et al., 2014). Altogether, these studies show that neuromodulators and sensory stimuli of various sorts can bias the choice between swimming and crawling. The diagrams of the neuronal dynamics (Figure 12B,C) suggest that these various influences could be shifting the trajectory curves in one direction or another, or simply decreasing the size of the stable attractor areas for one of the behaviors. Neurons like R3b1, the neuron in the leech subesophageal ganglion that can activate both swimming and crawling (Esch et al., 2002), could be involved in such biasing of attractors to one behavior or the other.
What about other animals—do they use the same neuronal circuitry for making behavioral choices? The three schemes for behavioral choice in the leech (inhibition between competing circuits, combinatorial codes, and differential dynamics) are, in fact, found in other animals. There is good evidence that decision-making neurons for one behavior inhibit those for other behaviors in a variety of animals, including slugs (Kovac & Davis, 1977), snails (Pirger et al., 2014), Drosophila larvae (Jovanic et al., 2016), the mammalian basal ganglia (Mink, 1996), the owl optic tectum (Mysore & Knudsen, 2012), and the monkey visual system (Wang, 2008; Doya & Shadlen, 2012). Combinatorial codes are seen in some of the molluscan circuits (Kristan & Gillette, 2007). For instance, a group of cells added to a network that produces ingestion of food can be turned into one that produces egestion (Jing & Weiss, 2001). This study also provides evidence not just for a combinatorial code but also a way to read out the code in implementing two different behaviors. Differential dynamics is a concept that is being used to study visual (Harvey, Coen, & Tank, 2012; Mante, Sussillo, Shenoy, & Newsome, 2013) and tactile (Carnavale et al., 2013) sensations, as well as and arm/hand movement (Shenoy et al., 2013), all in monkeys, as well as behavioral patterning in C. elegans (Roberts et al., 2016).
The three mechanisms for behavioral choice found in leeches, alone or in combination, can produce behavioral choices in both simple and complex nervous systems. Which mechanism is used to implement a particular behavioral choice may have more to do with evolutionary history than with some optimization of function. One beneficial consequence of using alternative stable states of largely the same neuronal circuitry may be to allow the leech nervous system to switch quickly between swimming and crawling (Figure 11B,C), because having to stop one behavior and initiate a second one could require passing through the “preparatory states” previously shown (Figures 8C,D and 9). So, in addition to evolutionary constraints, the presence of different mechanisms for making choices may reflect functional constraints or benefits that will become apparent only with further, clever experiments.
Briggman, K. L., Abarbanel, H. D. I., & Kristan, W. B., Jr. (2005). Optical imaging of neuronal populations during decision-making. Science, 307, 896–901.Find this resource:
Briggman, K. L., Abarbanel, H. D. I., & Kristan, W. B., Jr. (2006). From crawling to cognition: Analyzing the dynamical interactions among populations of neurons. Current Opinion in Neurobiology, 16, 135–144.Find this resource:
Briggman, K. L., & Kristan, W. B., Jr. (2006). Imaging dedicated and multifunctional neural circuits generating distinct behaviors. Journal of Neuroscience, 26, 10925–10933.Find this resource:
Briggman, K. L., & Kristan, W. B., Jr. (2008). Multifunctional pattern generating circuits. Annual Review of Neuroscience, 31, 271–294.Find this resource:
Brodfuehrer, P. D., & Friesen, W. O. (1986). From stimulation to undulation: A neuronal pathway for the control of swimming in the leech. Science, 234, 1002–1004.Find this resource:
Brodfuehrer, P. D., McCormick, K., Tapyrik, L., Albano, A. M., & Graybeal, C. (2008). Activation of two forms of locomotion by a previously identified trigger interneuron for swimming in the medicinal leech. Invertebrate Neuroscience, 8, 31–39.Find this resource:
Cannon, J., Kopell, N., Gardner, T., & Markowitz, J. (2015). Neural sequence generation using spatiotemporal patterns of inhibition. PLoS Computational Biology, 11, e1004581.Find this resource:
Carnevale, F., de Lafuente, V., Romo, R., & Parga, N. (2013). An optimal decision population code that accounts for correlated variability unambiguously predicts a subject’s choice. Neuron, 80, 1532–1543.Find this resource:
Crisp, K. M., & Mesce, K. A. (2006). Beyond the central pattern generator: Amine modulation of decision-making neural pathways descending from the brain of the medicinal leech. Journal of Experimental Biology, 209, 1746–1756.Find this resource:
Delcomyn, F. (1980). Neural basis of rhythmic behavior in animals. Science, 210, 492–498.Find this resource:
Dickinson, M. H., & Lent, C. M. (1984). Feeding behavior of the medicinal leech, Hirudo medicinalis L. Journal of Comparative Physiology A: Neuroethology, Sensory, Neural, and Behavioral Physiology, 154, 449–455.Find this resource:
Doya, K., & Shadlen, M. N. (2012). Decision making. Current Opinion in Neurobiology, 22, 911–913.Find this resource:
Drapeau, P., Kristan, W. B., & Shankland, M. S. (1995). Neurobiology and development of the leech. Journal of Neurobiology, 27, 267–445.Find this resource:
Edwards, D. H. (1991). Mutual inhibition among neural command systems as a possible mechanism for behavioral choice in crayfish. Journal of Neuroscience, 11, 1210–1223.Find this resource:
Esch, T., & Kristan, W. B., Jr. (2002). Decision-making in the leech nervous system. Integrative and Comparative Biology, 42, 716–724.Find this resource:
Esch, T., Mesce, K. A., & Kristan, W. B. (2002). Evidence for sequential decision-making in the medicinal leech. Journal of Neuroscience, 22, 11045–11054.Find this resource:
Felleman, D. J., & Van Essen, D. C. (1991). Distributed hierarchical processing in the primate cerebral cortex. Cerebral Cortex, 1, 1–47.Find this resource:
Frady, E. P., Kapoor, A., Horvitz, E., & Kristan, W. B., Jr. (2016). Scalable semisupervised functional neurocartography reveals canonical neurons in behavioral networks. Neural Computation, 27, 1–45.Find this resource:
Friesen, W. O., & Kristan, W. B., Jr. (2007). Leech locomotion: Swimming, crawling, and decisions. Current Opinion in Neurobiology, 17, 704–711.Find this resource:
Gaudry, Q., & Kristan, W. B., Jr. (2009). Behavioral choice by presynaptic inhibition of tactile sensory terminals. Nature Neuroscience, 12, 1450–1457.Find this resource:
Gaudry, Q., & Kristan, W. B., Jr. (2010). Feeding-mediated distention inhibits swimming in the medicinal leech. Journal of Neuroscience, 30, 9753–9761.Find this resource:
Gaudry, Q., & Kristan, W. B., Jr. (2012). Decision points: The factors influencing the decision to feed in the medicinal leech. Frontiers in Neuroscience, 6, 1–10.Find this resource:
Gaudry, Q., Ruiz, N., Huang, T., Kristan, W. B., III, & Kristan, W. B., Jr. (2010). Behavioral choice across leech species: Chacun à son goût. Journal of Experimental Biology, 213, 1356–1365.Find this resource:
Harley, C. M., Rossi, M., Cienfuegos, J., & Wagenaar, D. (2013). Discontinuous locomotion and prey sensing in the leech. Journal of Experimental Biology, 216, 1890–1897.Find this resource:
Harvey, C. D., Coen, P., & Tank, D. W. (2012). Choice-specific sequences in parietal cortex during a virtual-navigation decision task. Nature, 484, 62–68.Find this resource:
Ito, M., & Doya, K. (2015). Parallel representation of value-based and finite state-based strategies in the ventral and dorsal striatum. PLoS Computional Biology, 3(11), e1004540.Find this resource:
Jing, J., & Weiss, K. R. (2001). Neural mechanisms of motor program switching in Aplysia. Journal of Neuroscience, 21, 7349–7362.Find this resource:
Jovanic, T., Schneider-Mizell, C. M., Shao, M., Masson, J. B., Denisov, G., Fetter, R. D., … Zlatic, M. (2016). Competitive inhibition mediates behavioral choice and sequences in Drosophila. Cell, 167, 858–870.Find this resource:
Kovac, M. P., & Davis, W. J. (1977). Behavioral choice: Neurophysiological mechanism in Pleurobranchaea. Science, 198, 632–634.Find this resource:
Kristan, W. B., & Gillette, R. (2007). Decision-making in small neuronal networks. In G. North & R. Greenspan (Eds.), Invertebrate neurobiology (pp. 533–554). New York: Cold Spring Harbor Laboratory Press.Find this resource:
Kristan, W. B., Jr., Calabrese, R. L., & Friesen, W. O. (2005). Neuronal basis of leech behaviors. Progress in Neurobiology, 76, 279–327.Find this resource:
Kristan, W. B., Jr., McGirr, S. J., & Simpson, G. V. (1982). Behavioural and mechanosensory neurone responses to skin stimulation in leeches. Journal of Experimental Biology, 9, 143–160.Find this resource:
Kristan, W. B., Jr., Skalak, R., Wilson, R. J. A., Skierczynski, B. A., Murray, J. A., Eisenhart, F. J., & Cacciatore, T. W. (2000). Biomechanics of hydroskeletons: Studies of crawling in the medicinal leech. In J. Winters & P. Crago (Eds.), Biomechanics and neural control of movement (pp. 206–220). Berlin: Springer-Verlag.Find this resource:
Lent, C. M. (1985). Serotonin modulation of the feeding behavior of the medicinal leech. Brain Research Bulletin, 14, 643–655.Find this resource:
Lent, C. M., & Dickinson, M. H. (1984). Serotonin integrates the feeding behavior of the medicinal leech. Journal of Comparative Physiology A: Sensory, Neural, and Behavioral Physiology, 154, 457–471.Find this resource:
Lockery, S. R., & Kristan, W. B., Jr. (1990). Distributed processing of sensory information in the leech. II. Identification of interneurons contributing to the local bending reflex. Journal of Neuroscience, 10, 1816–1829.Find this resource:
Mante, V., Sussillo, D., Shenoy, K. V., & Newsome, W. T. (2013). Context-dependent computation by recurrent dynamics in prefrontal cortex. Nature, 503, 78–84.Find this resource:
Mink, J. W. (1996). The basal ganglia: focused selection and inhibition of competing motor programs. Progress in Neurobiology, 50, 381–425.Find this resource:
Misell, L. M., Shaw, B. K., & Kristan, W. B., Jr. (1998). Behavioral hierarchy in the medicinal leech, Hirudo medicinalis: Feeding as a dominant behavior. Behavioral Brain Research, 90, 13–21.Find this resource:
Muller, K. J., Nicholls, J. G., & Stent, G. S. (1981). Neurobiology of the leech. Cold Spring Harbor, NY: Cold Spring Harbor Press.Find this resource:
Mysore, S. P., & Knudsen, E. I. (2012). Reciprocal inhibition of inhibition: A circuit motif for flexible categorization in stimulus selection. Neuron, 73, 193–205.Find this resource:
Nusbaum, M. P., & Kristan, W. B., Jr. (1986). Swim initiation in the leech by serotonin-containing interneurons, cells 21 and 61. Journal of Experimental Biology, 122, 277–302.Find this resource:
Ort, C. A., Kristan, W. B., Jr., & Stent, G. S. (1974). Neuronal control of swimming in the medicinal leech. II. Identification and connections of motor neurons. Journal of Comparative Physiology A: Sensory, Neural, and Behavioral Physiology, 94, 121–154.Find this resource:
Palmer, C. R., Barnett, M. N., Copado, S., Gardezy, F., & Kristan, W. B., Jr. (2014). Multiplexed modulation of behavioral choice. Journal of Experimental Biology, 217, 2963–2973.Find this resource:
Pirger, Z., Crossley, M., László, Z., Naskar, S., Kemenes, G., O’Shea, M., … Kemenes, I. (2014). Interneuronal mechanism for Tinbergen’s hierarchical model of behavioral choice. Current Biology, 24, 2018–2024.Find this resource:
Puhl, J. G., Masino, M. A., & Mesce, K. A. (2012). Necessary, sufficient and permissive: A single locomotor command neuron important for intersegmental coordination. Journal of Neuroscience, 32, 17646–17657.Find this resource:
Puhl, J. G., & Mesce, K. A. (2008). Dopamine activates the motor pattern for crawling in the medicinal leech. Journal of Neuroscience, 28, 4192–4200.Find this resource:
Ranganathan, R., Cannon, S. C., & Horvitz, H. R. (2000). MOD-1 is a serotonin-gated chloride channel that modulates locomotory behavior in C. elegans. Nature, 408, 470–475.Find this resource:
Roberts, W. M., Augustine, S. B., Lawton, K. J., Lindsay, T. H., Thiele, T. R., Izquierdo, E. J., … Lockery, S. R. (2016). A stochastic neuronal model predicts random search behaviors at multiple spatial scales in C. elegans. Elife, 5, e12572.Find this resource:
Schulz, P. E., Cook, E. P., & Johnston, D. (1995). Using paired-pulse facilitation to probe the mechanisms for long-term potentiation (LTP). Journal of Physiology (Paris), 89, 3–9.Find this resource:
Shaw, B. K., & Kristan, W. B., Jr. (1995). The whole-body shortening reflex of the medicinal leech: motor pattern, sensory basis, and interneuronal pathways. Journal of Comparative Physiology A: Neuroethology, Sensory, Neural, and Behavioral Physiology, 177, 667–681.Find this resource:
Shaw, B. K., & Kristan, W. B., Jr. (1997). The neuronal basis of the behavioral choice between swimming and shortening in the leech: Control is not selectively exercised at higher circuit levels. Journal of Neuroscience, 17, 786–795.Find this resource:
Shenoy, K. V., Sahani, M., & Churchland, M. M. (2013). Cortical control of arm movements: A dynamical systems perspective. Annual Review of Neuroscience, 36, 337–359.Find this resource:
Sussillo, D., & Abbott, L. F. (2009). Generating coherent patterns of activity from chaotic neural networks. Neuron, 63, 544–557.Find this resource:
Tinbergen, N. (1951). The study of instinct. Oxford: Clarendon Press.Find this resource:
Wagenaar, D. A. (2015). A classic model animal in the 21st century: Recent lessons from the leech nervous system. Journal of Experimental Biology, 218, 3353–3359.Find this resource:
Wang, X. J. (2008). Decision making in recurrent neuronal circuits. Neuron, 60, 215–234.Find this resource:
Weeks, J. C., & Kristan, W. B., Jr. (1978). Initiation, maintenance, and modulation of swimming in the medicinal leech by the activity of a single neuron. Journal of Experimental Biology, 77, 71–88.Find this resource:
Willard, A. L. (1981). Effects of serotonin on the generation of the motor pattern for swimming by the medicinal leech. Journal of Neuroscience, 1, 936–944.Find this resource:
Wilson, R. J. A., Skierczynski, B. A., Blackwood, S., Skalak, R., & Kristan, W. B., Jr. (1996). Mapping motor neurone activity to overt behaviour in the leech. Internal pressures produced during locomotion. Journal of Experimental Biology, 199, 1415–1428.Find this resource: