In the leech, the command-like neuron called cell Tr2 is known to stop swimming, but the connections from cell Tr2 to the swim central pattern generator have not been identified. We used fluorescence resonance energy transfer voltage-sensitive dyes to identify three neurons that are synaptic targets of cell Tr2. We then used electrophysiological techniques to show that these connections are monosynaptic, chemical, and excitatory. Two of the novel targets, cell 256 and cell 54, terminate swimming when stimulated. These neurons are likely to mediate swim cessation caused by cell Tr2 activity, and thus play the role of intermediate control cells in the leech CNS.
- swim termination
- voltage-sensitive dye
- fluorescence resonance energy transfer
- command neuron
Hierarchical control is an organizational pattern found in both invertebrate and vertebrate nervous systems. In vertebrate nervous systems, the forebrain and brainstem are involved in the initiation, termination, and coordination of complex rhythmic behaviors, but the cycle-to-cycle control of these behaviors is mediated by pattern-generating networks in the spinal cord (Grillner et al., 1997). Similar hierarchical control is found in invertebrates, with higher-order neurons activating pattern generators, which then synapse on motor neurons (Orlovsky et al., 1999).
Single neurons that evoke or terminate whole behaviors (command-like neurons) have been found in many invertebrate preparations, whereas they seem to be rare in vertebrate preparations, presumably because of the larger number of neurons in these systems (Kupfermann and Weiss, 1978; Pearson, 1993; but see Roberts et al., 1997; Perrins et al., 2002). For instance, stimulating a single command-like neuron in the leech (Weeks and Kristan, 1978; Brodfuehrer and Friesen, 1986a) or in Tritonia (Frost and Katz, 1996) can elicit full-blown bouts of swimming, whereas it is necessary to stimulate many neurons in the brainstem of the lamprey (Orlovsky et al., 1999) to achieve a similar effect.
In the leech, in addition to swim-initiating neurons, there are command-like neurons that terminate an ongoing swim when stimulated (Brodfuehrer and Friesen 1986a; Brodfuehrer and Burns, 1995; O'Gara and Friesen 1995). One of these, cell Tr2, is found in the head brain of the leech (Brodfuehrer and Friesen 1986a; O'Gara and Friesen, 1995). Stimulation of this cell terminates an ongoing swim bout. It is not known, however, how cell Tr2 connects to the swim central pattern generator (CPG), and thus it is not known how it causes swim termination. One possibility is that cell Tr2 connects directly to the swim CPG cells. For instance, cell Tr2 could terminate swimming by inhibiting all or most of the swim CPG or by exciting part of the swim CPG in such a way as to disrupt rhythm generation. Such a scheme is used in the Xenopus tadpole, with mid-hindbrain reticulospinal neurons directly inhibiting the motor neurons involved in swimming (Perrins et al., 2002). A second possibility is that cell Tr2 inhibits the known swim-gating cells, thereby stopping swimming through these intermediaries. Because the swim-gating cells are necessary for the maintenance of swimming, this should be sufficient to stop an ongoing swim (Weeks, 1981; Nusbaum and Kristan, 1986). A third possibility is that cell Tr2 activates an intermediate network that then projects to the swim CPG cells and swim-gating cells. This intermediate network might be part of command-to-motor pathways for multiple behaviors, and together with the swim-gating cells could provide a neuronal locus in which pro-swimming and anti-swimming inputs are resolved. A precedent for this organization is provided by studies in which conflicting sensory inputs lead to a single, defined behavioral output (Kristan and Shaw, 1997).
As a first step toward resolving which of these schemes (if any) is used in the leech, we used fluorescence resonance energy transfer (FRET) voltage-sensitive dye imaging to search for synaptic targets of cell Tr2 (González and Tsien, 1995, 1997; Cacciatore et al., 1999; Zochowski et al., 2000). We found three neurons in the midbody ganglia that receive monosynaptic input from cell Tr2 and determined that two of these synaptic targets were able to terminate swimming when stimulated. None of these cells are either swim CPG cells or swim-gating cells, suggesting that they are part of an intermediate network by which Tr2 activity terminates swimming.
Materials and Methods
Preparation. Subjects were adult Hirudo medicinalis (4-8 gm), the European medicinal leech, obtained from Leeches USA (Westbury, NY) and maintained in artificial pond water at 15°C. We dissected out the full nerve cord, including the head brain, all 21 midbody ganglia, and the tail brain. We removed the connective tissue sheath from ganglia in which we planned to impale cells or image activity. Usually we desheathed the ventral surface of the head brain (subesophageal ganglion only) and the ventral side of one of the midbody ganglia from segments 10-13. We prepared at least one dorsal posterior (DP) nerve, usually chosen from ganglia 14 through 17, for extracellular recording. A motor neuron in this nerve bursts during the dorsal contractile phase of swimming, so the nerve serves as a monitor of swimming activity (Ort et al., 1974). In some experiments, we stabilized the imaged ganglion by pinning small strips of sausage casing across them to minimize motion artifact (Cacciatore et al., 1999). The preparation was maintained in a chamber filled with ∼5 ml of leech saline, consisting of (in mm): 115 NaCl, 4 KCl, 1.8 CaCl2, 1.5 MgCl2, 10 dextrose, 4.6 Tris maleate, and 5.4 Tris base, pH 7.4.
Staining with FRET dyes. We first stained the ganglion to be imaged with the FRET donor, N-(6-chloro-7-hydroxycoumarin-3-carbonyl)-dimyristoylphosphatidylethanolamine (CC2-DMPE) (Panvera LLC, Madison, WI), a coumarin-labeled phospholipid (González et al., 1999). We made a 30 μm staining solution from 3 μl of 5 mm CC2-DMPE in DMSO, 1 μl of 20 mg/ml pluronic F-127 in DMSO (Molecular Probes, Eugene OR), and 500 μl of saline. We pinned out the nerve cord in a Sylgard-coated dish, placed a small plastic cylinder over it, and then sealed it with petroleum jelly to make a watertight chamber. We then replaced the saline in the chamber with the staining solution and stained the ganglion for 30 min. During staining, we constantly recirculated the staining solution using a peristaltic pump (model RP-1; Rainin, Oakland, CA). After 30 min, the staining solution was taken off, and the tissue was washed several times with fresh saline.
We next stained the whole nerve cord with the FRET acceptor, bis(1,3-diethyl-thiobarbiturate)-trimethine oxonol [DiSBAC2(3)] (Panvera LLC), an oxonol dye (Tsien, 1976; González and Tsien, 1995). We made an 8 μm staining solution from 8 μl of 12 mm DiSBAC2(3) in DMSO and 12 ml saline. This solution was then bath-sonicated for at least 1 min. We replaced the bathing solution with this solution and left it on for at least 30 min. In many experiments, we left the oxonol solution on for as long as imaging was done, replacing it with fresh staining solution every few hours. After taking off the staining solution, we replaced it with fresh saline that was sometimes continuously perfused using a gravity perfusion system. Leaving the oxonol in the bath did not increase background fluorescence because its extinction coefficient is small at 405 nm, the wavelength of the excitation light, and oxonol has 20-fold less fluorescent yield in aqueous solution than in the membrane (Rink et al., 1980).
Electrophysiology and cell fills. We recorded intracellularly from cells using 40-60 MΩ glass microelectrodes filled with 1 m potassium acetate, using an Axoclamp 2A amplifier (Axon Instruments, Foster City, CA). We recorded extracellularly using suction electrodes and a four-channel differential amplifier (model 1600; A-M Systems, Sequim, WA). We digitized all electrical data at 1-2 kHz using a 12-bit analog-to-digital (A-to-D) board (PCI-MIO-16E-4; National Instruments, Austin, TX) and custom LabVIEW (National Instruments) software.
We filled individual neurons using either tetramethylrhodamine dextran [3000 molecular weight (MW); Molecular Probes] or Alexa 488 dextran (10,000 MW; Molecular Probes). Microelectrodes were backfilled using solutions with concentrations of 50 mg/ml (rhodamine) or 25 mg/ml (Alexa). Cells were then impaled, and dye was iontophoretically injected. Dye was allowed to diffuse for ∼1 hr, and tissue was fixed overnight, dehydrated, cleared, and mounted on a slide for viewing. Fills were imaged using a confocal microscope (1024ES; Bio-Rad, Hercules, CA). Dyes were excited by the 488 nm (Alexa) or 568 nm (rhodamine) emission line of a KrAr laser (60-WL-DZ; Bio-Rad). Images shown in Figure 3 were edited manually to remove background fluorescence using Corel Photo-Paint (Corel Corporation, Ottawa, Ontario, Canada).
Optical recording. We acquired fluorescence images using an upright microscope (Axioskop 2FS; Zeiss). We usually used a 40×, 0.8 numerical aperture (NA) water-immersion objective (Acroplan; Zeiss), but occasionally used a 20×, 0.5 NA water-immersion objective (Acroplan; Zeiss) when a wider field of view was desired. For epi-illumination we used a tungsten halogen lamp (64625 HLX; Osram Sylvania, Danvers, MA) in a standard housing (HAL 100; Zeiss), powered by a low-ripple power supply (JQE 15-12M; Kepco, Flushing, NY). For all imaging, we used only the coumarin emission, because it provided brighter fluorescence and higher voltage sensitivity. The filter set consisted of a 405 ± 15 nm bandpass excitation filter, a 430 nm dichroic mirror, and a 460 ± 25 nm bandpass emission filter (Chroma Technology Corporation, Brattleboro, VT). We used a water-cooled CCD camera (MicroMax 512 BFT; Roper Scientific, Tucson, AZ) operated in frame-transfer mode to acquire the optical data, at a frame rate of 20 Hz. The CCD chip in this camera has 512 × 512 pixels, but we normally binned at 4 × 4 pixels, yielding a 128 × 128 image. The quantum efficiency of the camera at the coumarin emission peak was 80%. The CCD chip was maintained at -25°C during imaging. Imaging data were acquired using the software package WinView/32 (Roper Scientific, Trenton, NJ). We synchronized the optical and electrical recordings by feeding the frame timing signals emitted by the camera into the A-to-D board, along with all the electrophysiology signals.
The combination of CC2-DMPE and DiSBAC2(3) yielded sensitivities in the range of 2-8%/100 mV for 1 Hz sinusoidal voltage signals with a 10 mV amplitude, centered around a baseline voltage of -50 mV. Much of the variation in sensitivity seemed to be attributable to differences in soma size, with larger somata giving higher sensitivity.
Analysis. After acquiring the data, we analyzed them using Matlab (The Mathworks, Natick, MA). We outlined the images of individual somata manually using a custom-made graphic user interface. All pixels within each cellular outline were then averaged in each frame, yielding a raw fluorescence signal for each cell, which we denote by F(t). The noise in these single-cell signals was normally shot-noise dominated. Raw fluorescence signals were usually corrupted by a slow downward drift attributable to bleaching, but this was eliminated by fitting it with a mathematical function and dividing it out of the signal. For each cell outline, bleaching was fit by computing a moving average of F(t) with a Gaussian window (σ = 500 msec). This slow signal we denote by Fslow(t). Because this slow signal also includes the DC component of F(t), we have: We estimated coherence using multi-taper methods (Thomson, 1982; Percival and Walden, 1993; Cacciatore et al., 1999). Only the differences from the procedures used in Cacciatore et al. (1999) will be described. For our coherence estimates, we fixed the frequency resolution, Δf, defined as the half-width of the spectral bands, at Δf = 2/3 Hz, and the number of tapers was adjusted according to the duration of the data set. For the data presented here, the trial duration was always T = 9.5 sec; thus the number of tapers with good leakage properties was: To calculate standard errors for the coherence estimates, we used the jackknife (Thomson and Chave, 1991). For the error in the coherence phase, we used the same procedure as described by Cacciatore et al. (1999). For the coherence magnitude, we used a slightly different procedure. After calculating the coherence magnitude estimate, C, and the take-away-one coherence magnitude estimates, Ci, the estimate and the take-away-one estimates were replaced with the transformed values given by y = f(C) = ln[C2/(1 - C2)], as suggested in Thomson and Chave (1991). The SE of the transformed estimate is given by the jackknife expression: where . Thus a one SE bar on y would be the interval (y - σy, y + σy). The error bar used for C is the interval (f-1(y - σy), f-1(y + σy)), or: Because the distribution of y is roughly Gaussian, this interval provides an ∼68% confidence interval. This interval also has the advantage that it is guaranteed to be a subset of [0,1], which is not the case if the above transformation is not used.
To test whether the coherence magnitude of a given cell was significantly greater than zero, i.e., larger than would be expected by chance from a signal with zero coherence, we compared the estimate of C with the null distribution for coherence magnitude. It can be shown that for the null distribution the coherence magnitude will exceed the value only in 100α% of trials (Hannan, 1970; Jarvis and Mitra, 2001). We used a low α level of 0.001 in all experiments, to avoid false positives. We also calculated the multiple comparisons α level for each trial, given by αmulti = 1 - (1 - α)n, where n is the number of cells, and verified that it did not exceed 0.05 on any trial.
When plotting the coherence in polar plots (see Figs. 1E, H, 2B), we corrected for the phase shift caused by the slow response of the dye signal by shifting the coherence phase by +54°. This is the phase angle by which the fluorescence signal lags the voltage when the voltage trajectory is a 1 Hz sinusoid. It is close to tan-1(2πτf) = 66°, the theoretical phase lag for f = 1 Hz, given the reported time constant of the dye, τ = 360 msec (Cacciatore et al., 1999). (The difference arises because the reported time constant was based on data for a range of frequencies, and the +54° shift was based solely on data for 1 Hz signals.) The phase shift was not adjusted in any way on a per-trial basis to achieve a better match between coherence estimates for optical versus electrical signals.
Cell identification. We identified the novel neurons described here (cells 252, 256, and the pair of cells 54) on the basis of their soma position, soma size, amplitude of action potential as recorded in the soma, and presence of one-for-one EPSPs from cell Tr2. All three cells are found on the ventral aspect of the ganglion. Cells 256 and 252 are found in the posterior medial packet, and cells 54 are found in the posterior lateral packets. In a few cases, the identity of a cell was established purely on the basis of soma position, size, and strong optical coherence with cell Tr2 at a phase lag between 60° and 120°. This was done only after a cell had been singled out optically and verified electrically in many animals and only in unambiguous cases. Cells that were reliably identified in many animals were assigned numeric “names” (e.g., cell 252) by matching up their usual soma position and size with those on the canonical map of the leech ganglion (Muller et al., 1981).
Cell Tr2 was identified on the basis of the description given in Brodfuehrer and Friesen (1986a). Cell Tr1 was typically found just medial to the Retzius cell in segment R1, and cell Tr2 was typically just medial to cell Tr1. The soma of cell Tr2 was typically slightly larger than that of Tr1 (∼60 vs ∼50 μm), and both cells displayed large (20-40 mV) action potentials that rose without prepotential. In all preparations in which we could elicit swimming (67 of 96), we verified that cell Tr2 stopped or strongly slowed swimming when stimulated. In all such cases but one, the cell initially identified as cell Tr2 on the basis of position, soma size, and spike shape was found to stop or strongly slow swimming. In seven preparations, we filled cell Tr2 with fluorescent tracer and verified that the morphology of the cell was consistent with that described in Brodfuehrer and Friesen (1986a).
Using imaging, we identified three putative synaptic targets of cell Tr2 in the segmental ganglia
To identify candidate synaptic targets of cell Tr2, we imaged many neurons simultaneously while stimulating cell Tr2. A midbody ganglion (usually ganglion 10) was stained with both components of the FRET system (see Materials and Methods), and one cell Tr2 was impaled and stimulated via current injection (Fig. 1A). In each trial, we injected a series of pulse trains into cell Tr2, using a level of current sufficient to elicit one spike per pulse. Each train was 500 msec long, with an intertrain interval of 500 msec, yielding a train frequency of 1 Hz. Within each train, we used a pulse frequency of 20 Hz. During stimulation, we imaged a region of the stained ganglion using a CCD camera, producing a time series of fluorescence images (Fig. 1B). We typically viewed approximately one-third of the ganglionic surface (20-80 cells) at a time. For each trial, we manually drew ellipses around each of the visible cell bodies (Fig. 1C) and assigned an alphanumeric label to each ellipse. We then averaged the pixel values inside the ellipse of each cell to provide a single time-varying optical signal for that cell (a subset of these signals is shown in Fig. 1D).
We quantified how strongly each cell responded to cell Tr2 activity by estimating the coherence between each optical signal and the electrical recording of the cell Tr2 membrane potential (Cacciatore et al., 1999) (see Materials and Methods). The coherence magnitude gives a measure of how well two signals correlate with one another at each frequency, with a value of 1 implying perfect correlation and a value of zero implying no correlation. The coherence phase gives the phase difference between the two signals at each frequency. We focused on the coherence at 1 Hz, the stimulation frequency, where most of the power in the cell Tr2 voltage signal is concentrated.
We used the coherence magnitude at 1 Hz to rank the optical signals in order of how strongly they responded to Tr2 stimulation. The six optical signals shown in Figure 1D are those that had the highest coherence magnitude at 1 Hz. They are shown in descending order of magnitude. Because the coherence estimate is a statistic, we could calculate the coherence magnitude required to reach statistical significance. Only the top three signals in Figure 1D (shown in color) responded strongly enough to conclude that their coherence with cell Tr2 was not attributable to chance. We refer to these cells as “followers” of cell Tr2, after Peterlin et al. (2000). These cells were considered to be good candidates for being synaptic targets of cell Tr2. The coherence phase at the stimulation frequency served as a measure of the latency of the response of a neuron to cell Tr2 stimulation (Fig. 1E). The phase can also distinguish excitatory from inhibitory effects (Cacciatore et al., 1999).
Using optical recording, we identified three followers of cell Tr2 that appeared in similar positions across several preparations. The most coherent of these was an unpaired cell in the posterior medial packet, cell 256. A cell in the approximate position of cell 256, which had strong coherence with cell Tr2, was seen in 45 preparations (Figs. 1C-E). The optical signal of cell 256 often displayed a near-sinusoidal modulation at the stimulation frequency. A second strongly coherent cell, cell 252, was unpaired and located in the same packet. A highly coherent cell in the approximate position of cell 252 was seen in 21 preparations (Figs. 1C-E). The most coherent cell in the posterior medial packet was usually in cell 256 position, with a cell in cell 252 position the next most coherent. A third strongly coherent cell was also identified, cell 54 (Figs. 1F-H). Unlike the others, it was a paired cell, with one homolog in each of the posterior lateral packets. We observed a strongly coherent cell 30 times, in 22 preparations, in the approximate position of cell 54. In addition to displaying large coherence magnitude, all four neurons typically had coherence phase lags between 60° and 120° (after correcting for the dye time constant; see Materials and Methods). Because the coherence is evaluated at 1 Hz, this corresponds to a time lag of between 167 and 333 msec. Given that a typical conduction velocity for a Tr2 spike is ∼15 msec per segment (Brodfuehrer and Friesen, 1986a; our data not shown) and that these recordings were done in ganglion 10, these latencies suggest a direct excitatory connection from cell Tr2 (Fig. 1E,H).
Cell Tr2 is monosynaptically connected to candidate targets
After we found a strong follower of cell Tr2 using optical recording, we impaled the follower (in the same preparation) to determine the nature of its connection to cell Tr2. For all three cell types described above, we found that the strong coherence with cell Tr2 activity was attributable to summating volleys of EPSPs in the target cell [Fig. 2A (data for cell 252 is shown)]. As expected, the optical signal was a low-pass-filtered version of the electrical signal (because of the time constant of the dye), with a lower signal-to-noise ratio and a lower sampling rate. The strong agreement between the optical and electrical recordings produced a similarly strong agreement between the coherence estimates derived from the two recordings (Fig. 2B): the two estimates are within the error bars of one another.
When individual cell Tr2 spikes were elicited, they produced single EPSPs in all three cell types, at a constant latency, typically between 125 and 175 msec (Fig. 2C). The latency seen is consistent with the conduction delay for the cell Tr2 spikes to travel from the head brain to ganglion 10 (Brodfuehrer and Friesen, 1986a). We checked for EPSPs in 30 of the 45 optically identified cells 256, in 4 of the 21 cells 252, and in 6 of the 30 cells 54. In every case, we found that single cell Tr2 spikes caused one-for-one EPSPs in the observed follower (Fig. 3A-C). To test whether these connections were monosynaptic, we changed the concentration of divalent cations in the saline and examined the responses of the three cell Tr2 followers (Figs. 3A-C) (Nicholls and Purves, 1970; Berry and Pentreath 1976; Byrne et al., 1978; Liao and Walters, 2002). In all cases, the EPSP was preserved in 20 mm Ca2+/20 mm Mg2+ saline, which blocks polysynaptic PSPs, and disappeared in 0 mm Ca2+/20 mm Mg2+ saline, which blocks chemical synaptic transmission. Thus the connections from cell Tr2 to all three neurons appear to be monosynaptic and chemically mediated.
For all cells, the latency of the EPSPs increased when the preparation was in 20 mm Ca2+/20 mm Mg2+ saline. This was presumably caused by a reduction in the conduction velocity of action potentials attributable to elevated spike threshold. We verified that conduction velocity was slower in this saline by recording en passant from the connective between ganglia 8 and 9. We found that the spikes did indeed take longer to propagate to the recording site in 20 mm Ca2+/20 mm Mg2+ saline and that the increased EPSP latency scaled with the increased delay (data not shown).
Morphology and segmental extent of cell Tr2 targets
The three cell Tr2 targets had distinct morphologies, which we visualized by filling the cells with fluorescent tracers (Fig. 3D-F). Cell 256 (Fig. 3D) is an unpaired cell, with a roughly symmetric pattern of neuritic branching. It sends a single process into the anterior medial connective. Cell 252 (Fig. 3E) is also unpaired, with a roughly symmetric pattern of neuritic branches. Unlike cell 256, it sends two processes out of the ganglion, one in the anterior medial connective and one in the posterior medial connective. Cell 54 (Fig. 3F) is a paired cell. Each member of the pair has a large tuft of neurites ipsilateral to the soma and sends its primary neurite contralateral. The primary neurite divides into two processes that then exit the ganglion, one via the contralateral (to the soma) anterior lateral connective and the other via the contralateral posterior lateral connective.
We found segmental homologs of cells 256, 252, and 54 in all midbody ganglia investigated (segments 9 through 12). Using imaging or electrical recordings, or both, we identified a total of 36 cells 256(10) in as many animals, 20 cells 252(10) in as many animals, and 36 cells 54(10) in 24 animals. The homologs in other segments were identified one to four times each per segment examined. Given the extensive segmental homology of the leech body plan, it is likely that cells 256, 252, and 54 are present in all or most midbody ganglia (Muller et al., 1981).
We found that both cell Tr2 homologs connected to all targets within a given midbody ganglion. That is, cells Tr2(L) and Tr2(R) both connected to cells 256(n), 252(n), 54(L,n), and 54(R,n), where n is the segment number (data not shown).
Cells 256 and 54 terminate swimming
We tested the ability of cells 256, 252, and 54 to terminate swimming by activating them individually during ongoing swimming episodes. We found that cells 256 and 54 stopped swimming episodes when driven strongly (Fig. 4) (data shown for cell 256 only). Consistent with this, cells 256 and 54 showed an increased firing rate during swim stops caused by cell Tr2 stimulation.
We found that driving cell 256 stopped swimming at least once in 67% of animals in which it was attempted (n = 9) (Fig. 4A) and that such stimulation stopped swimming in 50% of attempts (n = 26). We typically drove cell 256 at ∼30 Hz for 3-4 sec to obtain effective swim termination. This rate is substantially higher than the rate at which cell 256 spikes during cell Tr2-induced swim stops. In the example shown in Figure 4B, cell 256 spiked at 7.3 Hz during a Tr2-induced swim stop (n = 1).
To verify that the apparent effect of cell 256 was not attributable to chance, we performed a more controlled experiment on two nerve cords. Each trial in this experiment consisted of a nerve shock, which initiated swimming, followed at a set latency by stimulation of cell 256 at a high rate for 4 sec. A trial was counted as a “success” if swimming stopped during the stimulus and a “failure” if it stopped after the stimulus. (We discarded trials in which swimming stopped before the onset of stimulation.) We then compared the success rate in this condition with the success rate in a control condition in which cell 256 was not stimulated. A control trial was counted as a success if the swim stopped during the “stimulation window,” although no stimulus was delivered during this time. Swimming stopped during the stimulation window in 100% of the six stimulated trials, as compared with 20% of the five control trials. This difference is significant (p < 0.02; one-sided Fisher-Irwin test).
A single cell 54 was also capable of stopping swimming when stimulated strongly, although cells 54 stopped swimming more effectively when both homologs were stimulated simultaneously (data not shown). In some trials, swimming slowed (burst frequency decreased) for the duration of cell 54 stimulation but did not stop. Driving both cells 54 slowed or stopped swimming at least once in 100% of animals (n = 5) and in 93% of trials overall (n = 14). Driving both cells 54 stopped swimming in 60% of individuals and in 29% of trials. Driving a single cell 54 slowed or stopped swimming at least once in 100% of animals (n = 4) and in 73% of trials overall (n = 22). Driving a single cell 54 stopped swimming in 50% of individuals in 9% of trials.
As with cell 256, the spike rates needed to stop swimming by driving cells 54 were higher than the rates observed when we stopped swimming by stimulating cell Tr2. The average spike rate required for cell 54 to stop swimming was 42 ± 11 Hz (n = 5; in three animals), whereas its average spike rate when a cell Tr2 stopped swimming was only 2.8 ± 0.9 Hz (n = 3; in two animals).
The role of cell 252 in terminating swimming is not clear. Cell 252 did depolarize when Tr2 stimulation stopped swimming (n = 1), but driving cell 252 during swimming in three animals had weak and variable effects (data not shown).
Finally, none of the identified cell Tr2 targets seemed to function as either a swim CPG cell or as a swim-initiating cell. None underwent membrane potential oscillations during swimming, and none initiated swimming reliably when stimulated in the quiescent animal (data not shown).
We identified three novel synaptic targets of cell Tr2 using FRET voltage-sensitive dye imaging. Previously, no monosynaptic targets of cell Tr2 were known. All target neurons were found reliably across animals, in multiple midbody ganglia. Two of the targets, cells 256 and 54, stopped swimming when driven to fire at relatively high rates. These results suggest that cells 256 and 54, and possibly cell 252, provide a segmental neuronal layer that relays the activity of cell Tr2 to the swim-gating cells or to the swim CPG (Fig. 5). Thus we have identified what is likely to be the first step of the inhibitory neuronal pathway from cell Tr2 to the swim CPG.
What is the role of cells 256 and 54 in behavior?
We have shown that stimulating individual cells 256 and 54 can stop swimming but that the rates required are higher than those observed in the cells when driving cell Tr2 to stop swimming (Fig. 4). This is not surprising. Cells 256 and 54 were present in all four segmental ganglia searched, and it is probable that homologs exist in most of the 21 midbody ganglia. When all of these homologs are activated by cell Tr2 activity, the summed effect is likely to be quite strong. In this light, it is perhaps surprising that stimulating a single cell 256 or cell 54 is able to stop swimming. It seems reasonable that the combined activity of perhaps 20 cells 256 and 40 cells 54, all firing at modest rates, would be sufficient to explain the swim-terminating effects of cell Tr2.
If the entire functional role of cells 256 and 54 were to relay the activity of cell Tr2, it would seem simpler for cell Tr2 to synapse directly onto the swim CPG cells to stop swimming. In Xenopus tadpoles, for instance, reticulospinal neurons that stop an ongoing swim when stimulated connect directly onto a class of spinal motor neurons that help generate the swim rhythm (Roberts et al., 1997; Perrins et al., 2002). What does having cells 256 and 54 in the circuit contribute that could not be achieved by having cell Tr2 synapse directly onto the swim-gating cells or the swim CPG cells?
One possibility is that cells 256 and 54 act to distribute the output of cell Tr2. That is, each cell Tr2 could synapse on some 20-40 cells 256 or 54, or both, and each of these could then synapse on some number of swim CPG cells, call it n. In this way, the activity of cell Tr2 would effect 20n-40n downstream cells, although making only 20-40 synapses itself. This might be metabolically more efficient than having cell Tr2 make synapses onto all 20n-40n downstream cells and would effectively be a form of amplification.
Integration of anti-swimming inputs
Another possibility is that cells 256 and 54 have a broader role in the leech nervous system. It seems likely that their activity is not simply a function of the activity of cell Tr2 but reflects inputs from other neurons as well. For instance, cells 256 and 54 might act as integrators of various anti-swimming inputs. Previous authors have shown that there are parallel swim-activating and swim-inactivating systems in the leech (Brodfuehrer and Burns, 1995), and it is possible that cells 256 and 54 serve as part of an anti-swim-gating network, parallel to the swim-gating cells (cells 204, 21, and 61). These two networks would be expected to inhibit one another, thus forming a distributed “decision system” for swimming. Various pro-swimming and anti-swimming inputs would converge on the swim-gating and anti-swim-gating systems, and whichever system received the greater net input would prevail. Such a scheme has been proposed for neurons in the medial temporal and lateral intraparietal areas of monkey cortex that are thought to mediate judgments about motion directionality (Shadlen and Newsome, 1996, 2001; Shadlen et al., 1996; Gold and Shadlen, 2000).
Some support for such an anti-swim-gating hypothesis comes from the observation that the organization of the cell Tr2 swim-terminating circuitry seems to grossly parallel elements of the swim-initiation circuitry. In part of the swim-initiation circuitry, cell Tr1, a head brain cell pair, synapses on segmental swim-gating cells. These cells then connect to the swim CPG (Brodfuehrer and Friesen, 1986a,b; Nusbaum and Kristan, 1986). Similarly, cell Tr2, a head brain cell pair, synapses on cells 256 and 54, which then (directly or indirectly) connect to the swim CPG. In both systems, however, it seems that the connections of the segmental cells are not strictly intrasegmental: cell 204 makes all its known connections with CPG cells intersegmentally (although intrasegmental connections may exist), and the morphology of cells 256 and 54 suggests that they too are likely to make many of their synapses intersegmentally. Thus the organization of the cell Tr2 swim-terminating circuitry seems roughly to parallel the cell Tr1 circuitry for swim initiation.
In the anti-swim-gating scheme described above, all of the cells 256 and 54 in the animal are collectively “representing” a single thing: the extent to which the current situation of the animal favors or disfavors swimming. Another possibility is that the cells 256 and 54 in different segments are representing different things. For instance, cell 256 or 54 (or both) might play the role of a “segmental swim slowing” neuron, which is used by the leech CNS to slow down or phase delay the swim oscillation locally (in the same segment and perhaps a few neighboring segments). This might be useful as part of a sensory feedback loop acting to maintain the proper phase in each segment in the face of local perturbations. Such local feedback, although not mediated by non-CPG interneurons, has been identified in the leech and lamprey swim circuits (Viana Di Prisco et al., 1990; Wallen, 1997; Cang and Friesen, 2000; Cang et al., 2001). If cells 256 and 54 play such a role, it would presumably complement the direct feedback identified previously (Cang et al., 2001).
A similar possibility is that cell 256 or 54 plays the role of a “segmental shortening” neuron. Such a neuron could be used during crawling to generate the coordinated wave of local shortening that comprises the contraction phase of a crawl step. Such cells have been posited to exist in a model of the leech crawl CPG (Cacciatore et al., 2000). In both this and the segmental-swim-slowing scheme, cells 256 and 54 would effectively serve two related but distinct roles: to mediate swim termination by Tr2 and coordinate other behaviors.
Using imaging to find synaptic targets
We found imaging to be an essential tool in identifying synaptic targets of cell Tr2. Even having established the morphology and electrophysiological properties of cell 256, we relied on imaging to find this cell in each preparation. The exact position of the soma of cell 256 varies and is in an area of the ganglion with many somata of similar size. Therefore, our primary means of identifying the cell is to drive cell Tr2 and image all the neurons in its general location. In contrast, we can reliably identify cells 252 and 54 solely on the basis of soma position and spike shape, although we used imaging for their initial discovery.
It should be noted that other targets of cell Tr2 may exist. Although the optical method greatly facilitates the search for synaptic targets, it does not guarantee that one will find all of them. In particular, the dorsal side of the midbody ganglia was not examined as extensively as the ventral side, so there may be additional Tr2 targets with their somata on the dorsal surface.
Our work is related to previous work using imaging to explore connectivity. Kozloski et al. (2001) recently used calcium imaging to explore cortical connectivity in the mouse, building on earlier work developing the technique (Peterlin et al., 2000). Those authors used imaging to find neurons in mouse cortex that are targets of an impaled neuron. This enabled them to examine the cell-type specificity of cortical microcircuitry. The ability to record the electrical activity of a number of individual neurons, using various optical techniques, should open the way for establishing the connectivity of neuronal circuits in both simple and complex brains.
This work was supported by a La Jolla Interfaces in Science Predoctoral Fellowship, funded by the Burroughs Wellcome Fund (A.L.T.); National Institutes of Health (NIH) Training Grant GM08107 (A.L.T.); and NIH Research Grants MH43396 (G.W.C., D.K., W.B.K.), RR13419 (D.K.), and NS35336 (W.B.K.). We thank J. E. Gonzalez and R. Y. Tsien for assistance with the FRET voltage-sensitive dyes; Panvera LLC for supplying the dyes gratis; K. Briggman, S. B. Mehta, T. M. Esch, the Kristan Laboratory, and GURU (Gary's Unbelievable Research Unit) for valuable discussions; and the two anonymous reviewers.
Correspondence should be addressed to Adam L. Taylor, Volen Center, Mail Stop 013, Brandeis University, 415 South Street, Waltham, MA 02454-9110. E-mail:.
Copyright © 2003 Society for Neuroscience 0270-6474/03/2311402-09$15.00/0