INTRODUCTION

Involvement of the Mauthner cell (M cell) in vertebrate escape behaviors was hypothesized many years ago (reviewed by Eaton et al. 2001;Zottoli and Faber 2000). Electrical recording of M cells in free-swimming goldfish (Zottoli 1977) showed that their firing was correlated with a C-start behavior that seems to be an important maneuver by which cyprinid fishes escape from predators. It was subsequently shown that firing of the M cell is sufficient to initiate a C bend and counter-turn (Nissanov et al. 1990). Other brain stem neurons also appear to be involved in the initiation and control of C-start type escapes as indicated by lesion studies of the M cell, which failed to eliminate the behavior (Eaton et al. 1982, 1984; Kimmel et al. 1980; Zottoli et al. 1999). Anatomical studies of neurons projecting into the spinal cord of larval zebrafish revealed a large array of neurons of which two cells (MiD2cm and MiD3cm) were suggested to be segmental homologs of the M cell (Metcalfe et al. 1986) and were proposed to provide directional control over the escape behavior (Foreman and Eaton 1993). Subsequent in vivo calcium imaging studies supported this hypothesis: caudal stimuli activated just the M cell, while rostral stimuli activated the “Mauthner array”, i.e., the M cell plus the two segmental homologs (O'Malley et al. 1996). Dramatic confirmation of this hypothesis came with laser ablation of the M cell and its homologs: lesioning just the M cell increased response latencies to caudal stimuli, while lesioning the entire Mauthner array sharply increased response latencies to rostral stimuli (Liu and Fetcho 1999). An unexpected result, however, was that after an abnormally long latency, fairly normal looking “escape” behaviors occurred that approached wild-type C starts in terms of bend angle and angular velocity (Budick and O'Malley 2000a; Liu and Fetcho 1999). Because the long latencies of these responses would bode poorly in terms of escape from predators, these abnormal behaviors might be referred to as “large, delayed” turns.

From the preceding data, we might infer that at least two additional classes of neurons contribute to teleost C-start escape behaviors. First, some neurons in the brain stem must mediate the large, delayed turns observed after ablation of the Mauthner array. Second, additional neurons are predicted to mediate the counter-turn that follows the initial c-bend (termed A2 neurons by Foreman and Eaton 1993). The identities of these postulated neurons are unknown, but this group is composed, at least in part, of neurons descending from brain into spinal cord. The descending neurons in the larval zebrafish are anatomically well defined, numbering about 220 in total and including reticulospinal, vestibulospinal, T-reticular, and IC neurons as well as cells of the nucleus of the medial longitudinal fasciculus (nMLF) (Gahtan and O'Malley 2001;Kimmel et al. 1985; Metcalfe et al. 1986). Using confocal calcium imaging, we set out to determine which neurons would respond to escape-eliciting stimuli. Unexpectedly, we found that a substantial majority of descending neurons, spanning the rostral-caudal extent of the brain stem, exhibit large calcium responses indicative of the firing of action potentials and the transmission of escape-related information to spinal cord.

METHODS

A 75% solution of the fluorescent calcium indicator Oregon-green dextran bis-(o-aminophenoxy)-N,N,N′,N′-tetraacetic acid (BAPTA, 10,000 molecular weight) was injected into the caudal spinal cord of 2 to 4 days posthatching larval zebrafish (Danio rerio) that had been anesthetized with 0.02% 3-aminobenzoic acid ethyl ester (MS222, Sigma). A glass microelectrode, broken back to approximately half the diameter of the spinal cord, was used for injections. Labeling of hindbrain neurons is believed to occur by the severing of their spinal axons (Gahtan and O'Malley 2001). A substantial but variable number of descending neurons are typically labeled after the injections, which are aimed at the ventral portion of the cord. After injection, larvae were placed individually in circular wells (1.7-cm diam) containing 10% Hanks solution (Westerfield 1995). They were housed in an incubator (approximately 27°C, 14 h/10 h light/dark cycle) overnight to allow for recovery and transport of the indicator from spinal axons to the soma of reticulospinal neurons. For confocal imaging, larvae were again anesthetized, placed on their backs on a glass coverslip, and restrained in agar, after which the anesthetic was replaced with 10% Hanks solution (see O'Malley and Fetcho 2000 for further details). A Zeiss Axiovert microscope with a ×40, 0.75 NA objective and BioRad MRC600 laser scanning confocal microscope were used for imaging reticulospinal neurons in intact larvae. Reticulospinal and other descending neurons were identified morphologically based on soma location and other anatomical features (Kimmel et al. 1985; Metcalfe et al. 1986).

In each larva, a reticulospinal neuron was selected for recording and a glass “tapper” was positioned near the ear on the side opposite to the spinal projection of that neuron's axon. An electrical pulse to a piezoelectric crystal attached to the tapper caused the glass tip to gently contact the head of the larvae; this stimulus reliably induces an escape turn contralateral to the tap (Fetcho and O'Malley 1995; O'Malley and Fetcho 2000;O'Malley et al. 1996). On each trial, a tap was delivered during collection of a sequence of images of the cell, and analysis of fluorescence values before (baseline) and after the tap was made off-line (as described in Fetcho and O'Malley 1995). At least 2 min were allowed to elapse between trials as this frequency of stimulation tends to minimize any habituation of the escape response (Fetcho and O'Malley 1995;Fetcho et al. 1998). Cells were imaged using either frame scans (24 scans at ∼500 ms/scan), slow line scans (512 scans at 6 ms/scan), or fast line scans (512 scans at 2 ms/scan). Tap-evoked behaviors caused cells to move briefly out of the plane of focus. This “behavioral” movement artifact was clearly distinguishable from the much smaller movement artifact produced by the tapper alone and served as a rough marker of the presence and duration of behaviors. However, these movement artifacts precluded imaging of cellular calcium levels during the behavior. In some fish, after collecting a series of trials, the fish was paralyzed by bath application of curare (3 mg/ml), and further trials were collected. Such trials allowed better estimation of the latency of tap-evoked fluorescence changes because of a reduction in the duration of the movement artifact. A reticulospinal cell type was classified “responsive” during escapes if a fluorescence increase (posttap peak/pretap baseline) of ≥10% occurred on at least three of five trials (in noncurarized fish) for four different cells in at least three different larvae. Cell types were classified “nonresponsive” only if they failed to meet the response criteria and other cell types in the same larvae showed responses to head taps. Cells that were nonresponsive to head taps contralateral to their spinal axon projection were also tested for their responsiveness to ipsilateral taps.

RESULTS

Escape-eliciting stimuli generated calcium responses in neurons in both the medulla and midbrain. Figure 1shows examples of responses in three specific cell types (right) and their location in the brain stem (left). Cells examined in the nMLF, the most rostral group of neurons that projects from brain to spinal cord in larval zebrafish, responded to taps to the head. A medial-lateral (cMeL) cell of the nMLF (top) shows a large calcium response just after the tap (*), which decays nearly to baseline by the end of the imaging sequence. Examples of calcium responses in a rostral and a caudal medullary neuron are also shown (middle and bottom). In each case, the calcium dynamics are typical of somatic calcium signals (insets: note rapid rise, gradual recovery) observed in vertebrate neurons both in culture and in vivo (Fetcho and O'Malley 1995, 1997; O'Donovan et al. 1993;O'Malley 1994; O'Malley et al. 1996,1999). These prior studies demonstrated that somatic calcium signals of this size occur only after the firing of one or more action potentials: in no case have we observed somatic calcium signals in response to subthreshold stimuli.

To examine calcium responses with higher temporal resolution, line-scans (i.e., rapid scanning of a single line across a cell) were used to measure calcium dynamics with 2-ms resolution. A line-scan of a medullary cell type called RoM3L is shown in Fig.2. A and B show the anatomical location of the cell and the position of a line that was repetitively scanned across the soma (rotated 90° in 2B). In the line-scan image in Fig. 2 C, the green band is the cell body, and it shows a rapid fluorescence increase just after the end of the movement elicited by the head-tap (tap indicated by ↓). The underlying plot shows that fluorescence was near maximal by the time the fish stopped moving, indicating that calcium had begun entering the cell before the end of the movement. Because fluorescence cannot be measured when the larva is moving, we also examined tap-elicited responses in curare-paralyzed fish (Fig. 2 D). Here a similar calcium response is seen, but at a shorter latency, in this instance occurring 12 ms after the onset of the stimulus artifact produced by movement of the tapper itself.

Calcium responses were examined in 15 cell types in 22 different paralyzed fish. In the great majority of cases, fluorescence increases were detected at latencies ranging between 5 and 40 ms after the onset of the tapper movement (Fig. 2 E). The variability in latency determination was due primarily to differences in the duration of the tapper-movement artifact, which varied from preparation to preparation. Thus we cannot state the actual latency of the onset of calcium increase but can only provide an upper limit before which the response may have begun. The histogram is plotted in conjunction with an example escape behavior, and it can be seen that most calcium responses occurred by either the end of the C bend or during the counter-turn. The specific cells in which paralyzed-latencies were measured are noted in Fig. 3.

To assess the extent of activation of descending neurons, we imaged calcium responses in most of the distinct cell types projecting from brain into spinal cord. In attempting this survey, our goal was simply to determine whether or not these cells were responding to escape-eliciting stimuli rather than to extensively evaluate the response properties of one or just a few cell types. For this purpose, the criterion chosen was to examine at least four individual cells of each type, distributed among at least three different fish. Cell types that consistently exhibited calcium responses at the end of the movement artifact are color-coded green, while cells that usually showed no detectable calcium response are coded red (Fig.3 A). All of these cell types are present in bilateral pairs but are only shown unilaterally in this illustration. A small fraction of cell types with inconsistent responses or incomplete results are coded yellow; cells in the image that were not examined were left white. Only a few cell types failed to respond to contralateral head-taps, so ipsilateral stimuli were also tested in these cells to confirm that they would not respond to head taps. The most striking observation is that a majority of cells, and cell types, consistently responded to head taps.

Based on earlier categorizations, we distinguish 27 types of reticulospinal cell (Metcalfe et al. 1986), plus 4 large nMLF cells, IC cells, T-reticular neurons, and vestibulospinal cells (Kimmel et al. 1985). Of these cell types, 17 or 55% consistently responded to head taps (Fig. 3 B, green rows). An additional seven cell types showed calcium responses but did not reach criterion (yellow rows): in some cases cells were encountered only infrequently (e.g., RoL2) and in other cases, they exhibited calcium responses but not often enough to reach the criterion established for inclusion in the responding category (e.g.,MiV1). Nonetheless, all six of these equivocal cell types showed at least some clear calcium responses, and if we add them to the responding classes, it would amount to a total of 24 responding cell types. This would constitute 77% of all distinguishable cell types that project from brain into spinal cord, but more significantly, it amounts to 92% of the cell types examined in this study. Only two cell types (6%) failed to respond consistently, and five reticulospinal cell types were not examined. Cell types that responded included: cells in all seven hindbrain segments (ranging from RoL1 toCaD), all four of the larger cell types in nMLF (smaller nMLF cells were not examined) and T-reticular cells. The IC cells were the largest group of cells that did not show tap-evoked calcium responses (vestibulospinal cells were not examined in this study). The large majority of cells studied exhibited no spontaneous calcium activity when the larva was restrained on the stage of the confocal microscope, so the observed calcium responses were absolutely correlated with the escape-eliciting stimulus.

DISCUSSION

Previously, there was no physiological data on any of the 21 cell types in the brain stem of larval zebrafish that we observed here to respond to escape-eliciting stimuli. This number includes the seven “equivocal” cell types but does not include the three Mauthner-array cell types previously linked to C-start escape behaviors. In addition to constituting most of the anatomically distinct cell types that project from the brain into spinal cord, this group also includes a large majority of the total number of descending neurons. All cell types are present in bilateral pairs and there are multiple “copies” of many of them, as noted in Fig. 3 B. We estimate that in larval zebrafish there are, in total, ∼220 neurons that project from brain into spinal cord. If we assume that our recordings of an individual cell type are representative of all cells of that type, then we can estimate, for purposes of discussion, that we have sampled responses of 140 neurons of the total population. Of these 140 neurons, 94 (or 63%) consistently respond to escape eliciting stimuli and an additional 28 neurons (19%) exhibit calcium responses at least some of the time. To summarize, our best estimate is that, in larval zebrafish, escape-related activity can be observed in 82% of the neurons projecting from brain into spinal cord. This result has several ramifications.

The brain stem escape network (BEN), as originally envisioned by Eaton and colleagues (Eaton et al. 2001; Foreman and Eaton 1993), included a small number of both ipsilaterally and contralaterally projecting neurons. Liu and Fetcho (1999)'s results indicated that additional neurons in brain stem are capable of generating escape-like turns, i.e., turns that are much faster and larger than the other main type of larval turning behavior, referred to as “routine” turns (Budick and O'Malley 2000b). From the Liu and Fetcho study, it seemed certain that brain stem neurons in array-lesioned fish were firing after an abnormally long delay (i.e., after the C bend and counter-turn would normally be over), but it was not known if these putative escape-controlling neurons would normally fire concurrently with the Mauthner array in nonlesioned fish. Our results show that many brain stem neurons exhibit escape-related calcium responses during either the C bend, or at latest, during the counter-turn. The large numbers of neurons activated adds considerably to the possible anatomical foundation of the BEN. While it is not necessarily the case thatall of these neurons are directly involved with the generation and control of escape locomotion, they all appear to be sending information to spinal cord that is time-locked to the escape behavior. Because the neuroanatomical substrate underlying the escape behavior appears to be quite similar for both larval and adult zebrafish and goldfish as well (Eaton et al. 2001;Lee and Eaton 1991; Lee et al. 1993;Metcalfe et al. 1986), the calcium-imaging data may be generally applicable to teleost locomotion.

While we do not know the specific functional roles of the escape-responsive neurons, lesion studies suggest that they are indeed important for escapes. We have made repeated efforts to eliminate swimming and turning behaviors by laser-ablating either the Mauthner array + nMLF (Budick and O'Malley 2000a) or larger groupings of neurons (Gahtan and O'Malley 2000) and have not been able to eliminate any behavior, although we have observed Mauthner cell ablations to delay the escape behavior (unpublished data in agreement with Eaton et al. 1982; Liu and Fetcho 1999). In addition, severing the axons of yet larger numbers of reticulospinal neurons disrupts the normal kinematics of the C start, producing an initially S-shaped bend (Gahtan and O'Malley 2001), which is consistent with the present imaging data indicating the involvement of a large fraction of descending neurons. One possibility is that controls for different components of the escape behavior are distributed throughout the brain stem. Some fraction of these neurons may contribute to directional control of the escape trajectory, which is quite variable (Budick and O'Malley 2000b; Foreman and Eaton 1993). Other neurons may help coordinate the sequencing of different components of the behavior, i.e., C bend, counter-turn, and burst swim. Simultaneous high-speed behavioral imaging and confocal calcium imaging of neurons, in partially restrained larval zebrafish, may prove useful for relating neuronal activity to these specific locomotive components (Bhatt et al. 2000). The fact that these neurons are distributed throughout all segments of the hindbrain has additional significance.

Vertebrate brains appear to develop by a common genetic plan that involves segmentation of the hindbrain (Keynes and Lumsden 1990; Prince et al. 1998). This segmental organization appears important for a variety of neural operations that underlie rhythmic behaviors (Bass and Baker 1997). Perhaps the clearest example of a segmentally organized system concerns the escape behavior, where the M cell and its two segmental homologs appear to provide directional control over the behavior (Foreman and Eaton 1993; Liu and Fetcho 1999;O'Malley et al. 1996). Because the much larger population of calcium-responsive neurons identified in the current study are distributed across all hindbrain segments, these neurons might be viewed as part of a multisegmental network that collectively coordinates the larva's escape behavior. This idea is consistent with the proposal that there are multiple sets of segmental homologs (Metcalfe et al. 1986), which might generate segmental codes to shape different aspects of the behavior. The results of the laser-ablation and axon-severing experiments also support the idea that multisegmental networks of neurons coordinate motor control. The extent to which this occurs in higher vertebrate behaviors is currently not well understood in large part because of the difficulties in establishing the cellular-level organization of motor control in such animals (Kiehn et al. 1998).

A detailed cellular-level description is important not only for addressing issues such as segmental coding but perhaps more importantly for defining the type or class of neural architecture being implemented. Aside from the Mauthner array's control of escape latency, the descending motor control system does not appear to be of the “dedicated” type, where specific neurons are exclusively involved in the control of specific behaviors. Other types of neural architectures, e.g., “distributed” and “reorganizing” (reviewed in Morton and Chiel 1994; see also Eaton and DiDomenico 1985; Leonard 2000), seem more likely to fit with our behavioral and confocal observations. Indeed, our results are reminiscent of the widespread activation of hindbrain neurons by vestibular stimuli in lamprey (Deliagina et al. 1992; Orlovsky et al. 1992) and the widespread activation of abdominal ganglion neurons in Aplysia(Wu et al. 1994; Zochowski et al. 2000). The resiliency of the larval behaviors to extensive laser-ablations might also suggest a redundant type of neural architecture. The BEN is not a simple system. It is capable not only of coordinating the different components of the escape behavior but also of executing a sensorimotor transformation to produce correctly sized and timed C bends and counter-turns so as to generate an appropriate escape trajectory (Eaton et al. 1991, 2001). More recently evolved vertebrates have far greater numbers and types of descending neurons which are often intermingled (Brodal 1981;Siegel and Tomaszewski 1983) and so, experimentally, pose a far greater challenge (see e.g., Kiehn et al. 1998). We believe that to accurately determine the type of neural architecture mediating a behavior, it is essential to identify most (if not all) of the cell types involved in that behavior and to learn what each cell type contributes.

FOOTNOTES

• Address for reprint requests: D. M. O'Malley, Dept. of Biology, 414 Mugar Hall, Northeastern University, Boston, MA 02115 (E-mail: d.omalley@neu.edu).