Descending control of swim posture by a midbrain nucleus in zebrafish

Abstract

The reticular formation in the brainstem controls motor output via axonal projections to the hindbrain and spinal cord. It remains unclear how individual groups of brainstem neurons contribute to specific motor functions. Here, we investigate the behavioral role of the nucleus of the medial longitudinal fasciculus (nMLF), a small group of reticulospinal neurons in the zebrafish midbrain. Calcium imaging revealed that nMLF activity is correlated with bouts of swimming. Optogenetic stimulation of neurons in the left or right nMLF activates the posterior hypaxial muscle and produces a graded ipsilateral tail deflection. Unilateral ablation of a subset of nMLF cells biases the tail position to the intact side during visually evoked swims, while sparing other locomotor maneuvers. We conclude that activity in the nMLF provides postural control of tail orientation and thus steers the direction of swimming. Our studies provide an example of fine-grained modularity of descending motor control in vertebrates.

Introduction

Elucidating the neural architecture of sensorimotor circuits is fundamental to the broad goal of understanding the neural basis of behavior. Two opposing views concerning the functional organization of such circuits are that they operate in a “distributed vs. modular fashion”. In the case of a distributed locomotor circuit, it is difficult to assign specific behavioral functions to individual neurons or even small groups of neurons given that global changes in circuit activity determine behavioral outputs. In a modular circuit design, the activity of discrete pools of neurons is dedicated to discrete kinematics, which are combined at the level of the musculature resulting in a complete behavioral program. Such neuronal modules could be used in varying combinations, giving rise to a diverse, seemingly continuous locomotor repertoire. Instead of being purely modular or distributed, it is likely that many behavioral circuits employ a mixture of these architectures.

Much of our knowledge of the organization of premotor circuitry has come from the investigation of invertebrate behaviors. Distributed neural coding schemes have been identified for the gill withdrawal reflex of Aplysia californica (Wu et al., 1994) and the local bending reflex in leech (Lockery and Kristan, 1990). The discovery of “command neurons” underlying escape behavior, including the tail-flip response in crayfish, on the other hand, support an extreme version of the module hypothesis (Wiersma, 1947; Boyan et al., 1986). In addition to reflexive behaviors, studies in the nematode C. elegans have uncovered pools of forward and backward command neurons that promote opposing directions of rhythmic locomotion (Chalfie et al., 1985).

In vertebrates, perhaps the best example of modular organization are the central pattern generators (CPGs) in the spinal cord. CPGs produce locomotion by coordinated, rhythmic activity of interneurons and motor neurons (Grillner, 2006; Kiehn, 2006; Tresch et al., 2002; Stein and Daniels-McQueen, 2002). Separate “unit CPGs” control antagonist limb movements, and the interaction between these circuit modules can be recombined to produce variations on a behavior such as changes in gait (Grillner, 2006). Additionally, an apparently modular organization has been identified in the descending reticulospinal system (RS) for 3D body orientation/orienting in lampreys, and for control of neck and back musculature in cats (Pavlova and Deliagina, 2002; Peterson et al., 1979).

The RS system in larval zebrafish is an attractive model for studies of descending motor control. There are relatively few RS neurons (about 150 on each side of the brain), many of which are individually identifiable from animal to animal (Kimmel et al., 1982). Functional studies of the RS system in zebrafish have been interpreted to support either modular (Huang et al., 2013; Orger et al., 2008) or distributed circuit organization (Gahtan et al., 2002; Liu and Fetcho, 1999). To further address this fundamental question, we investigated the behavioral role of the midbrain nucleus of the medial longitudinal fasciculus (nMLF). Neurons in the nMLF are the most rostral components of the RS system in larval zebrafish and possess dendrites that contact visual recipient regions as well as axonal projections that innervate circuits in the hindbrain and along the length of the spinal cord (Gahtan and O’Malley, 2003). Activity in the nMLF has been broadly correlated with multiple sensory stimuli and behaviors, however its exact function remained undefined (Gahtan et al., 2002, 2005; Orger et al., 2008; Sankrithi and O’Malley, 2010). We show, by calcium imaging, that activity in nMLF cells is highly correlated with swimming behavior. Unilateral optogenetic activation evokes smooth ipsilateral steering movements, driven by posterior hypaxial musculature, whose amplitude increased roughly linearly with stimulation frequency. In agreement with these activation experiments, unilateral nMLF ablations biases the position of the tail during swims, while leaving other behaviors intact. Together these findings suggest that one central function of the nMLF is postural control of tail orientation during swimming and provide evidence for modular locomotor control emanating from the midbrain of a vertebrate.

Results

Reticulospinal neurons in the nMLF are labeled in the Gal4^s1171t transgenic line

To begin to dissect the behavioral role of the nMLF, we searched a library of Gal4 drivers generated by enhancer trapping for lines that allow us to genetically target cells in the nMLF (Baier and Scott, 2009; Scott et al., 2007a). This “shelf screen” identified the Gal4^s1171t line, which drives expression of UAS-linked transgenes in the midbrain tegmentum including many axons within the MLF (Figure 1A and 1B). Critical for optogenetic approaches, cell populations directly dorsal or ventral to the tegmentum in Gal4^s1171t are not labeled although other brain areas express the transgene (Figure S1). In the tegmentum, approximately 600 neurons are labeled in Gal4^s1171t with the vast majority localized to medial regions. Lateral portions of the expression pattern have sparser labeling and also contain a dense neuropil.

Figure 1. Enhancer trap line Gal4^s1171t drives expression in the nMLF.

A. Confocal projection of the midbrain in a 6-day-old Gal4^s1171t/UAS:GFP fish. Optical sections (100μm) were collapsed to yield a maximum intensity projection. Reticulospinal neurons were backfilled from the spinal cord with Texas Red dextran. Right panel is an expanded view of the nMLF region indicated by the red box in the left panel. Green cells are GFP-labeled by Gal4^s1171t. Magenta cells are RS neurons in the rostral hindbrain and midbrain, labeled by Texas Red. White or pale magenta cells within the white circle are left nMLF neurons, which are double-labeled. B. Confocal projection of a slice (80 μm) of the GFP dorsal expression in Gal4^s1171t/UAS:GFP. The axon tract of the MLF (green arrowhead) and dendrites exiting the nMLF (orange arrowhead) are highlighted. C. Two-photon image of a single plane highlighting four identified nMLF neurons, MeLr and MeLc, plus the newly identified MeS1 and MeS2. D. Confocal image projection of a cryostat section (25 μm) from Gal4^s1171t/UAS:GFP stained with antibodies to GFP (green) and ChAT (red). Cell nuclei are labeled with a DNA dye (blue). See also Figure S1.

The nMLF is defined in zebrafish as a bilateral cluster of midbrain RS neurons, which are backfilled by neuronal tracers injected into the spinal cord (Kimmel et al., 1982). Spinal cord backfills with a dextran-conjugated fluorescent dye revealed that Gal4^s1171t labels > 80% of neurons within the nMLF, including the four canonical large identified cells MeLr, MeLc, MeLm and MeM, as well as a population of smaller cells, most of which reside in the dorsal nMLF (Kimmel et al., 1982) (Figure 1A, 1C, S1, and Movie S1). The small nMLF cells, which we have named MeS (Mesencephalic Small), are more numerous and vary in position making individual identification by current methods challenging (Figure S1B). While most MeS cells are located in dorsolateral positions, a few of these neurons could also be found ventrally, in the same planes as the MeL neurons. We were able to consistently identify a pair of MeS neurons (named MeS1 and MeS2 here) in ventral regions of Gal4^s1171t. These cells were positioned just lateral to MeLr and were also routinely backfilled with Texas Red dextran (Figure 1C and S1).

Antibody staining for choline acetyltransferase coupled with the Gal4^s1171t expression pattern revealed that the nMLF lies just rostral to the oculomotor nucleus (Figure 1D). This staining provided evidence that the nMLF is likely not homologous to the midbrain locomotor region (MLR), which has been described in other vertebrates. The MLR is thought to reside more rostrally, is partially cholinergic in non-mammalian vertebrates and has no direct spinal projections (Dubuc et al., 2008; Thankachan et al., 2012).

Both MeS and MeL cells show distinct, cell-type specific projection patterns in hindbrain and spinal cord

To label individual nMLF neurons we performed single cell electroporations of GFP-labeled cells with tetramethylrhodamine dextran in Gal4^s1171t/UAS:GFP fish. Our focus was on MeS cells, for which no morphological data existed. We recorded two distinct morphologies for the 15 individual MeS cells that were successfully labeled (Figure 2 and S2). One cell type possessed dendrites that project ventrally and also dendrites that project within the posterior commissure and cross the midline, ramifying in the neuropil of the contralateral nMLF (Figure 2B). Axons of these cells innervate the hindbrain and rostral spinal cord ipsilaterally and terminate roughly halfway down the spinal cord (Figure 2E, 2G and S2; average termination: at the level of myotome 11 ± 4.3). The second MeS cell type extends dendrites that run ventrally and axons that innervate the ipsilateral caudal hindbrain and rostral spinal cord (Figure 2C, 2F, 2G and S2; average termination: myotome 7 ± 3.1).

Figure 2. MeS neuron morphologies revealed by single-cell electroporations.

A-C. Examples of confocal image projections of the midbrain in 6-day-old Gal4^s1171t/UAS:GFP fish. MeLm (A), MeS type 1 (B) and MeS type 2 (C) are shown. Cells were electroporated with tetramethylrhodamine (TMR) dextran (magenta). GFP (green) is driven by Gal4^s1171t. TMR is brighter than GFP in cellular processes. The dendrites therefore appear magenta. The somas of electroporated cells are double-labeled by GFP and TMR and appear as white. D-F. Drawings of the nMLF cells in A-C showing their axonal projections. The corresponding photomicrographs are provided in Figure S2. Axon termination points are indicated by red arrowheads. Rostral collaterals in the hindbrain are indicated by blue arrowheads. G. Summary table for all electroporations performed. See also Figure S2.

The two MeLc neurons we labeled had similar dendritic patterns as previously described (Gahtan and O’Malley, 2003), although their axons terminated more caudally than previously reported (Figure 2G; myotome 20 and 23, respectively). In addition, we labeled an MeLm neuron whose anatomy has not previously been described. This cell had elaborate ventral dendrites and an axon with extensive ipsilateral innervation in the hindbrain and rostral spinal cord and terminated in myotome 19 (Figure 2A, 2D, 2G and S2). This axon also contained a prominent contralateral collateral in the hindbrain (Figure 2D). In summary, our data extend previous observations of the connectional complexity of nMLF cells. Of particular note, all nMLF cells investigated showed extensive axon collaterals in the hindbrain (in addition to their terminal arbors in the spinal cord), suggesting that they may coordinate the activity of several premotor cell groups and participate in multiple behaviors.

Calcium imaging shows that nMLF cells are broadly and bilaterally active during behavior

To image activity in the nMLF, we expressed the high-signal-to-noise, genetically encoded calcium sensor GCaMP6s (Chen et al., 2013) by crossing Gal4^s1171t fish to a newly constructed Tg(UAS:GCaMP6s)mpn101 line. We performed two-photon calcium imaging while simultaneously monitoring tail kinematics during spontaneous behaviors with a high speed camera (Movie S2). Animals were head-embedded upright, such that similar populations of neurons in the left and right nMLF were present in the same z-plane. Fish exhibited a variety of spontaneous behaviors under our imaging conditions. During individual trials (20 sec and 40 sec long) we observed slow swims (tail beat frequencies of 18-25 Hz), asymmetric tail flips and struggles.

In accordance with previous imaging of the MeL neurons, we observed global activation of MeLc, MeLr and MeLm during swims, flips and struggling (Figure 3A, 3C and S3). We found that swims had the highest probability (98.7%) of evoking responses followed by flips, (78.2%) and struggles (58.3%) (Figure 3D). During all behaviors, including single asymmetric tail flips we always observed activity in both the left and right nMLF (Figure 3A and 3C). In trials where multiple closely timed behaviors were present, we observed multiple activations in individual neurons that manifested as compound calcium responses (Figure 3C and S3). While we successfully recorded from each of the large MeL neurons, we could not reliably monitor activity in MeM neurons, as the high density of labeled neurons in medial regions hindered unambiguous identification of these cells.

Figure 3. Calcium imaging of nMLF activity during spontaneous behavioral bouts.

Fish larvae (6dpf) with genotype Gal4^s1171t/UAS:GCaMP6s were used. A. Examples of calcium responses in MeL neurons, accompanying a single swim (upper panel) and to swims and struggles (lower panel). B. Examples of calcium responses in MeS neurons that display transient responses to swims (upper panel) and swims and tail flips (lower panels). C. Calcium responses in MeS neurons that have slow response properties and turn off at the onset of tail flips. Responses in MeLc and MeLr neurons during tail flips are also shown. D. Summary table indicating the probability of calcium responses across neurons and spontaneous behaviors. Probability is represented by the color scale to the right of the table. Calcium traces were low-pass filtered for display purposes. Activity in nMLF cells is highly correlated (> 0.9) with swims and less correlated (0.4 – 0.9) with tail flips and struggles. See also Figure S3.

Imaging of the MeS cells revealed that they possess similar response properties to the MeL neurons, with swims again producing the highest probability of responses (97.6%) followed by tail flips (80.7%) and struggles (50%) (Figure 3B and 3D). We observed two distinct populations of MeS neurons based on their response properties. The majority of cells, predominantly located in dorsal regions of the nMLF, had transient responses that mirrored the activity of the MeL cells (Figure 3B). Based on their location, these cells fall into the two morphological types described above. In all animals that were imaged, we also observed a very small number of neurons (3-4) in both the left and right ventral nMLF that had slow dynamics. These neurons ramped up activity during intervals between behaviors and had a pronounced decrease in activity following the onset of all behaviors (Figure 3C). In summary, neurons in the nMLF are bilaterally active during swim maneuvers, including those that have highly dissimilar kinematics.

Optogenetic activation of the nMLF causes smooth tail deflections

To determine the behavioral consequences of nMLF activation, we generated fish in which Gal4^s1171t drove expression of the light-activated cation channel Channelrhodopsin-2 (ChR2) (Boyden et al., 2005). For photostimulation experiments, animals were head-embedded in agarose with their tails unrestrained (Figure 4A). A vertically oriented optic fiber (10 μm tip diameter) delivered 473 nm laser light into the fish’s brain, while the tail was imaged at 250 Hz. Stimulation of the caudal midbrain in Gal4^s1171t/UAS:ChR2 transgenic fish larvae with a laterally moving fiber resulted in smooth tail deflections that closely followed the position of the fiber (Figure 4B and Movie S3, which provides a dramatic example of this behavior). Stationary fiber stimulation of a small region within 50 μm of the midline on either the left or right side produced a sustained ipsilateral tail deflection that lasted for the duration of the stimulus (Figure S4). Tail deflection amplitudes were equivalent for left and right stimulation sites, with both having a pivot point caudal to the swim bladder in the vicinity of myotome 6 (Figure 4A; Movie S3). Light-induced movements were not observed in Gal4^s1171t/UAS:GFP transgenic animals (Figure S4).

Figure 4. ChR2 stimulation of nMLF neurons and their localization using photoconversion.

A. Dorsal view of the experimental setup used for ChR2 experiments. Two different tail positions are shown to illustrate the tail angle measurement used below and in subsequent figures. Optic fiber is pseudocolored for clarity. B. Change in tail angle (red) in a Gal4^s1171t/UAS:ChR2 fish, elicited by nMLF stimulation with a laterally moving optic fiber (green; 10μm fiber diameter, continuous beam 0.8 mW/mm²). Blue shaded region depicts the epoch of ChR2 stimulation. C. Confocal image projection of the midbrain in a Gal4^s1171t/UAS:ChR2/UAS:Kaede fish, which underwent bilateral Kaede conversions at sites that produced left and right steering, respectively. D. Confocal image projection, detailing the positions of two large MeL and the MeS neurons in the lateral nMLF. See also Figures S4 and S5.

We identified the stimulation site within the Gal4^s1171t expression pattern using fish that expressed both ChR2 and the photoconvertible fluorescent protein Kaede (Ando et al., 2002). Upon localizing a stimulation position that evoked ipsilateral tail deflections in Gal4^s1171t/UAS:ChR2/UAS:Kaede fish, the optic fiber was parked and the laser line switched to 405 nm to photoconvert Kaede at that position. In all 11 fish treated in this manner, a small converted region with a diameter of 10-50 μm, or two mirror-symmetric regions for bilaterally treated fish, was observed (Figure 4C and Figure S5). Converted neurons localized to portions of the Gal4^s1171t expression pattern that overlap with laterally situated neurons of the nMLF. Cells in this region included all of the large MeL neurons and also the MeS cells (Figure 4D). From these stimulation and conversion experiments, we conclude that activity in a lateral subregion of the nMLF drives smooth tail deflections, which resemble postural ‘steering’ movements. These movements were striking for their reproducibility within and across animals and because they have not, to our knowledge, been previously observed in response to reticulospinal neuron stimulation.

Steering and swimming are triggered differentially by dosage and location of nMLF photo-activation

To characterize the behavioral output of the nMLF, we stimulated 30 fish with three different stimulation frequencies (10 Hz, 20 Hz, 30 Hz) on both the left and right sides (Figure 5A). In nearly all fish (95%), 10 Hz stimulation resulted in a detectable ipsilateral steering movement in the tail (Figure 5D). In a minority of cases steering as well as swims or ipsilateral turns were present (13% and 1.6% respectively, Figure 5D). When stimulation frequency was increased to 20 Hz and then again to 30 Hz, we observed an increase in steer amplitude with a concomitant decrease in steer rise time (Figure 5B, 5C and S6). Stimulations at 20 Hz and 30 Hz always resulted in detectable steering movements. The probability of eliciting swims that accompanied steering was similar for all conditions (10 Hz p=0.13, 20 Hz p=0.14 & 30 Hz p=0.15, Figure 5D). When they occurred the tail beat frequency and duration of swims were not strongly dependent on stimulation frequency (Figure 5E and S6). These properties of the nMLF are different from those reported for the MLR, where increased activity produces a large increase in gait and swim frequencies in other vertebrates (Shik et al., 1966; Sirota et al., 2000). In contrast to swims, the probability of ipsilateral and contralateral turns increased with stimulation frequency, with ipsilateral turns being more common (Figure 5D). Turns consisted of either one large tail flip or multiple, laterally biased tail undulations. Increasing the light dose also resulted in a greater probability for trials where all three behaviors were present (steer, swim and turn) (Figure 5D).

Figure 5. Behavioral effects of ChR2 stimulation depend on light dose and fiber location.

A. (Top) Schematic of five stimulation sites within the midbrain expression pattern of Gal4^s1171t. (Bottom) Traces depict tail angle as a function of time in a single fish at five stimulus locations with 10 Hz, 20 Hz and 30 Hz light pulses (10 ms pulse duration, 1 mW/mm²). Blue shaded region depicts the epoch of ChR2 stimulation. B. Probability density histogram for maximum tail angles at left, middle and right stimulation sites (n ≥ 30 for left and right positions, n ≥ 20 middle; color code is the same as in A). C. Mean values for maximum tail angle at left, middle and right stimulation locations (effect of frequency and location: P < 0.01; color code is the same as in A). D. Probability of observing different tail kinematics at each stimulation frequency. Left and right trials were combined. E. Tail beat frequencies for swim bouts at each stimulation frequency (effect of frequency: P < 0.05; n ≥ 9). F. Two trials (blue and yellow) where long swimming bouts were evoked by bilateral stimulation with a large (105 μm) optic fiber (n = 2; 20 Hz pulse frequency). n values indicate number of fish. Error bars indicate s.e.m. See also Figure S6.

To determine if more medial regions of the nMLF produce ChR2-driven behavior we positioned the optic fiber above the midline, equidistant to verified left and right stimulation sites. For the majority of medial stimulation trials, we could not detect a change in tail kinematics (17 of 20 fish) (Figure 5A, 5B and S6). These results suggest that, when stimulated in isolation, medial regions of the nMLF are not directly involved in moving the tail. We next determined which tail kinematics are produced when the nMLF is bilaterally activated. In these experiments we used a large-diameter (105 μm) optic fiber to excite the entire nMLF region. Here, stimulation using 20 Hz or 30 Hz pulses was capable of producing long bouts of swimming that often lasted the duration of illumination (Figure 5F). However, for the majority of fish stimulated in this fashion, uncoordinated movements were elicited (8 of 12). When long bouts of swimming were present, tail beat frequencies were similar to those for unilateral stimulation (20.5 ±2.8 Hz), however bout durations far exceeded the length of swims observed in freely swimming or head-restrained fish (Budick and O’Malley, 2000; Portugues and Engert, 2011).

Both MeL and MeS neuronal populations contribute to steering behavior

Our Kaede conversion experiments determined that the region within the lateral nMLF where steering is evoked contains the MeL neurons as well as MeS neurons. To resolve the relative contributions of these two populations of neurons to steering, we measured ChR2 evoked steering in Gal4^s1171t/UAS:GFP/UAS:ChR2 animals before and after cell-targeted two-photon ablations (Figure 6A). To isolate steering behavior, relatively low stimulation parameters were used to reduce the likelihood of evoking swims or turns. Unilateral ablation of both MeL and MeS neurons (6-8 dorsal MeS neurons, MeS1, MeS2, MeLc, MeLr and MeLm) dramatically reduced steering on the ipsilateral side while leaving contralateral steering intact (Figure 6B). Targeting just the MeS cells produced a similar reduction, whereas removal of the MeL cells had an intermediate effect (Figure 6C). From these ablation results, we conclude that both neuron types contribute to steering, with a larger contribution coming from the MeS population.

Figure 6. Contributions of large MeL neurons and MeS neurons to steering.

A. Two-photon image projection of the midbrain in a Gal4^s1171t/UAS:GFP fish, in which the lateral region of the right nMLF was ablated 12 h prior. Scale bar, 50 μm. B. (left) Tail angle as a function of time after stimulating left (bottom traces) and right (top traces) ChR2 sites before and after lateral right nMLF ablation. (right) Reduction in tail angle is observed on the ablated side (n = 14, paired t-test, P < 10⁻⁵; t(13) = 6.91), but not on the control side (n = 14, paired t-test, P = 0.12; t(13) = 1.62). C. (left) Tail angle as a function of time for left and right ChR2 stimulation sites before and after ablation of large MeL neurons on the left side (bottom traces) and MeS on the right side (top traces). (right) Reduction in tail angle is observed for MeL ablations (n = 7, paired t-test, P < 0.02; t(6) = -3.49) and superficial MeS ablations (n = 7, paired t-test, P < 0.001; t(6) = -5.91). n values indicate number of fish. Error bars indicate s.e.m.

The nMLF is required for maintaining the direction of forward swims during optomotor behavior

Neurons in the nMLF are active when stimuli that drive forward optomotor behavior are presented (Orger et al., 2008). To determine the nMLF’s function in this visually guided behavior we performed optogenetic and lesion experiments using a head-fixed optomotor preparation (Figure 7A). In this assay, caudal-to-rostral moving gratings presented on vertically oriented LCD screens elicited forward swims, whereas rotating gratings produced tail flips toward the direction of rotation. Unilateral ChR2 activation of the nMLF introduced an ipsilateral bias to swims evoked by forward moving gratings presented to both eyes (Figure 7B and 7C). This increase in tail angle was most prominent during the first few tail oscillations. In line with these activation experiments, unilateral ablation of neurons in the lateral nMLF (6-8 dorsal MeS neurons, MeS1, MeS2, MeLc, MeLr and MeLm) caused forward swims to be biased toward the intact side (Figure 7D-G, Movie S4 and S5). In three animals we observed a striking tail deflection toward the intact side prior to the onset of tail oscillations (Figure 7F and Movie S5). These deflections resembled steering movements evoked by unilateral ChR2 stimulation. Ablations to isolate the behavioral contribution of the MeL and MeS cell types only uncovered a significant effect on swimming tail angle when MeS neurons were removed (Figure 7H). Bilateral nMLF ablations did not alter tail orientation during swims (Figure S7). Strikingly, unilateral nMLF ablations only affected forward swims, having no detectable effect on the amplitude of turns evoked by ipsiversive or contraversive OMR stimuli (Figure 7D, 7G and 7H). Bout frequency and bout duration for all three OMR behaviors were unaffected by unilateral removal of nMLF neurons (Figure S7), underpinning the specificity of our ablations.

Figure 7. Role of lateral nMLF neurons in the optomotor response.

A. Schematic of the optomotor assay. Animal is head-embedded in agarose with the tail free. Two LCD screens on each side of the fish display caudal to rostral moving gratings for forward OMR stimulation. A third screen (not shown) is added in front of the animal to create rotating gratings for contraversive and ipsiversive OMR stimulation. B. Change in tail angle in a Gal4^s1171t/UAS:ChR2 fish elicited by continuous forward OMR stimulation (black) or by both OMR stimulation and ChR2 activation of the right nMLF (red). Blue shaded region depicts the epoch of ChR2 stimulation. C. Increase in tail angle during forward OMR swim bouts induced by ChR2 stimulation (n = 21; P < 0.10⁻⁶; t(20) = -5.26). D. Change in tail angle in a Gal4^s1171t/UAS:GFP fish evoked by three different OMR stimuli before (black) and after (red) ablation of lateral neurons in the left nMLF. All traces are from the same fish. E. Projection of video frames displaying tail position of three fish (right ablated, control and left ablated) for an entire forward OMR trial. F. Change in tail angle evoked by a forward OMR stimulus before (grey) and after (red) ablation of lateral neurons in the right nMLF. Black arrowhead denotes a left tail deflection that precedes swimming. Traces are from the same fish. G. In ablated fish (red bars) swims are biased toward the intact side for forward OMR (n = 31; paired t-test, P < 10⁻⁶; t(30) = 6.96). Turns evoked by contraversive (n = 28; P = 0.076; t(27) = 1.84) or ipsiversive (n = 28; P = 0.55; t(27) = 0.59) grating stimuli are unchanged. In control fish (grey bars), no biases were observed for forward (n = 10; P = 0.70; t(9) = 0.39), contraversive (n = 8; P = 0.52; t(7) = -0.67) or ipsiversive (n = 8; P = 0.86; t(7) = -0.17) stimuli. H. Unilateral superficial small neuron ablations (blue bars) generated a bias in OMR forward swims toward the intact side (n = 13; paired t-test, P < 0.02; t(12) = 2.82). Responses to contraversive (n = 11; P = 0.63; t(10) = 0.48) or ipsiversive (n = 11; P = 0.88; t(10) = -0.14) stimuli were unchanged. Unilateral MeL neuron ablations (red bars) did not affect forward swims (n = 8; P = 0.62; t(7) = 0.51), or turns to contraversive (n = 8; P = 0.97; t(7) = 0.03) or ipsiversive (n = 8; P = 0.62; t(7) = 0.51) stimuli. n values indicate number of fish. Error bars indicate s.e.m. See also Figure S7.

Posterior hypaxial musculature drives steering movements

Given the unique and stereotyped steering produced by unilateral nMLF stimulation, we hypothesized that a particular muscle (or group of muscles) may underlie these movements. We took advantage of background muscle expression in Gal4^s1171t to conduct ChR2 muscle activation experiments, as well as calcium imaging, in muscle groups following unilateral nMLF stimulation. We found that ipsilateral stimulation of the posterior hypaxial muscles (PHM, Haines et al., 2004), a muscle group with previously unknown function, produced robust tail deflections that resembled unilateral activation of the nMLF (Figure 8A). Activation of medial trunk hypaxial muscles (THM) produced only modest tail deflections (Figure 8A). Kaede conversion experiments confirmed the identity of these muscle groups (Figure 8B and 8C). To demonstrate a functional connection between the nMLF and PHM, we imaged muscle activity using GCaMP6s before and after unilateral ChR2 stimulation of the nMLF. Here, we observed robust activity in the ipsilateral PHM and only modest activity in the THM (Figure 8D). The configuration of our optical setup did not allow for muscle imaging during nMLF stimulation. Stimulation of sites just caudal to the nMLF did not evoke muscle activity (Figure S8). Based on the position of the PHM, we expect its motor unit to be located in the hindbrain, in close proximity to axon collaterals from nMLF neurons.

Figure 8. Specific muscle activation generates tail deflections.

A. Change in tail angle in Gal4^s1171t/UAS:ChR2 fish elicited by ipsilateral hypaxial muscle stimulation with a stationary vertically oriented optic fiber (105μm fiber diameter, continuous beam 2 mW/mm²). Blue shaded region depicts the epoch of ChR2 stimulation. The blue trace represents the average change in tail angle when lateral muscle regions in the vicinity of myotomes 4-6 were stimulated (n = 6). The red trace represents the average change in tail angle when muscle regions just lateral to the midline were stimulated (n = 6). The green trace represents the average change in tail angle in Gal4^s1171t/UAS:GFP fish (n = 3). B. Confocal image projection from the side and top of a Gal4^s1171t/UAS:ChR2/UAS:Kaede fish that underwent Kaede conversion at a lateral muscle site, which produced a substantial tail deflection angle. This conversion labeled the posterior hypaxial muscle group (PHM) outlined in white. C. Confocal image projection from the side and top of a Gal4^s1171t/UAS:ChR2/UAS:Kaede fish that underwent Kaede conversion at a medial muscle site, which produced a small, but detectable tail deflection. This conversion labeled the trunk hypaxial muscle group (THM) outlined in white. D. Average calcium responses in ipsilateral PHM (blue), ipsilateral THM (red), contralateral PHM (blue dash) and contralateral THM (red dash) following unilateral nMLF optical stimulation in Gal4^s1171t/UAS:ChR2/UAS:GCaMP6 fish (n = 5). Blue shaded region depicts a 200 ms epoch of ChR2 stimulation. ChR2 was stimulated using the minimum laser power required to produce a tail deflection. n values indicate number of fish. PF denotes pectoral fin. See also Figure S8.

Discussion

The functional organization of the premotor brainstem is an area of intense investigation. Here, we sought to characterize the organization of descending motor commands emanating from the nMLF, the most rostral component of the larval zebrafish RS system. Previous results have shown that the nMLF is broadly active during a variety of behaviors including swimming, escape, optomotor responses and prey capture (Gahtan et al., 2002, 2005; Orger et al., 2008; Sankrithi and O’Malley, 2010). Using a combination of calcium imaging, optogenetic activation and laser ablations we showed that a central function of the nMLF is to provide postural positioning of the tail during forward swims, likely via activation of the posterior hypaxial musculature. Furthermore, our results indicate that this function is primarily carried out by the MeS neuron population, with contributions from MeL cells. This function is likely to be carried out in concert with separate descending motor commands for rhythmic tail oscillations during swimming (Arrenberg et al., 2009; Kimura et al., 2013).

From a library of Gal4 enhancer trap lines (Baier and Scott, 2009; Scott et al., 2007b), we identified the line Gal4^s1171t, which drives expression in the midbrain tegmentum. Retrograde tracing from spinal cord revealed that the vast majority of nMLF neurons are labeled in Gal4^s1171t, including the large MeL and MeM neurons, as well as the smaller MeS neurons. MeS neurons fall into two broad types based on their dendritic morphology, and have ipsilaterally innervating axons that terminate at the level of the caudal hindbrain and rostral spinal cord. MeL and MeM axons, on the other hand, project all the way to the caudal spinal cord. Collateral arbors from both MeL and MeS axons are found in the hindbrain, near the attachment site of the posterior hypaxial musculature and the probable location of the associated motor pool. Consistent with a functional connection between nMLF neurons and this muscle, their optogenetic activation each produced very similar tail deflections.

As a first step toward understanding how these neurons contribute to behavior, we used Gal4^s1171t to drive expression of GCaMP6s (Chen et al., 2013) and monitored population activity in the nMLF during spontaneous behaviors. Imaging revealed that neurons in nMLF are broadly active during slow swims, tail flips and struggles. We found that swims had the highest probability of evoking calcium responses, followed by tail flips and struggles. In addition, we observed that nMLF neurons on both sides were synchronously active during all behaviors including highly asymmetric tail flips. Our results agree with previous imaging results (Sankrithi and O’Malley, 2010) and extend them to a greater number of neurons and the MeS population.

While our imaging data suggested that the nMLF carried out an apparently nonspecific function during locomotor maneuvers, the optogenetic activation experiments indicated a greater functional differentiation within the nucleus. Unilateral activation of the nMLF with ChR2 elicited smooth ipsilateral steering movements in 95% of the animals tested. Higher light doses drove larger tail bends, often followed by full-blown swim bouts. Bilateral stimulation of the nMLF typically evoked uncoordinated movements, but was also capable of producing robust symmetric swimming. A potential neural basis for stimulus dependent recruitment of steering and swims is soma size of nMLF neurons. MeS neurons, which we have shown to be primarily responsible for steering evoked by low optical stimulation doses, are likely to reach threshold faster than the large MeL neurons given their compactness. Under this model, the larger MeL neurons would only reach threshold with higher stimulation and thus their activity may underlie the swim kinematics we observed with high frequency stimulation. This hypothesis should become further testable with improved methods for optical or genetic targeting of individual cells.

Our optogenetic results are at least partially consistent with electrical stimulation experiments in goldfish (Kobayashi et al., 2009; Uematsu et al., 2007), another cyprinid teleost. Here, unilateral injection of current in the vicinity of the nMLF evoked ipsilateral turns, whereas bilateral stimulation evoked swims. These earlier studies did not report isolated tail bends resulting from nMLF stimulation. A caveat of electrical stimulation is its limited spatial resolution; excitation of neurons outside the target site may confound interpretations of the circuitry underlying evoked behaviors. Furthermore, the close proximity of the nMLF to descending and ascending fibers in the MLF makes precise electrical targeting of the nMLF especially challenging. The specificity of optogenetic approaches compared to electrical stimulation techniques could explain the observed differences.

In loss-of-function experiments, we removed nMLF neurons using cell-targeted two-photon laser ablations and assessed behavior using a head-fixed optomotor assay. Previous imaging in fully embedded animals showed that both the MeL and MeS neuronal populations are tightly tuned for forward moving gratings that evoke swims and are mostly unresponsive to laterally moving gratings that promote turns (Orger et al., 2008). In line with these findings, nMLF ablations specifically affected OMR forward swims while leaving turns intact. Unilateral ablation of either the left or right nMLF caused swims to be biased to the intact side, however did not have an effect on the number of swim bouts per trial or the duration of bouts. Furthermore, in several fish, we observed a pronounced deflection of the tail to the intact side prior to the onset of swimming. This result indicates that activity in at least part of the nMLF precedes the initiation of locomotion. A potential source of this activity is the slow MeS neurons whose activity increases prior to the onset of locomotion.

Our results suggest that neurons in the left and right nMLF activate trunk musculature equally when fish swim straight and asymmetrically when fish swim with a biased trajectory. This model is consistent with previous findings where ablation of nMLF neurons reduced prey capture behavior by increasing the yaw angle between the head and the tail during capture swims (Gahtan et al., 2005). In light of the present results, it is likely that these altered capture swim kinematics result from a reduction in postural steering control by the nMLF.

Orientation of the body in three dimensions is critical for successful navigation of the environment. Studies in various organisms have revealed neural strategies for transforming sensory inputs into appropriate body orientation adjustments (Deliagina and Orlovsky, 2002; Deliagina et al., 1998). In the lamprey, unequal RS activation of spinal cord networks controls roll, pitch and turn orienting movements (Deliagina et al., 2000; Grillner and Wallén, 2002; Wannier et al., 1998; Zelenin et al., 2003). The steering mechanism we describe here is likely to act in concert with RS neuron modulation of spinal cord circuits to generate biased swims. The behavioral role we have identified for the nMLF has the distinct signature of postural control. Given the positioning of the nMLF, these circuits could function similarly to those in the mammalian ventral tegmental field that underlie postural control during ambulation (Iwahara et al., 1991; Mori, 1987).

Our results are most consistent with a large degree of functional specialization of premotor circuits. Recent studies of the RS system in larval zebrafish have also supported. such a modular architecture. Investigation of the vSPNs (ventromedial Spinal Projection Neurons), has provided strong evidence that these neurons are solely responsible for large tail deflections during turn behaviors. Imaging of the entire RS system while presenting OMR stimuli found that the vSPNs comprise a small population of neurons that preferentially respond to turn-inducing stimuli (Orger et al., 2008). Turns in response to OMR stimuli manifest as a series of tail oscillations where the first undulation cycle is highly biased to one side. Ablation of these vSPNs was shown to completely eliminate OMR turning while leaving swims intact (Orger et al., 2008). Subsequent studies have shown that vSPNs are also necessary for phototaxic turning, spontaneous turns and turns evoked by dark flashes (Huang et al., 2013). In the absence of the vSPNs, there is an increase in swim bouts, suggesting that in ablated animals turn events were transformed into additional swims bouts. The simplest explanation for this result is that activity in the vSPNs adds a lateral bias on top of an independent oscillatory motor pattern (Huang et al., 2013). In this model, the activity of a yet to be identified population of swim-inducing neurons, perhaps in the caudal hindbrain (Arrenberg et al., 2009), combined with the activity of the vSPN turn module results in the complete turn motor program.

Locomotor patterns have been proposed to result from the linear combination of muscle synergies (Bizzi et al., 2008; Roh et al., 2011). Under this scheme, dedicated modules – as opposed to distributed neural circuits – are responsible for commanding discrete groups of muscles (Briggman and Kristan, 2008). In the context of zebrafish behavior, specific motor components appear to be driven by distinct clusters of neurons, including the speed of the escape-associated C-bend (Liu and Fetcho, 1999; O’Malley et al., 1996; Prugh et al., 1982), spontaneous and stimulus-evoked turning (Huang et al., 2013; Orger et al., 2008), forward swimming (Arrenberg et al., 2009; Kimura et al., 2013) and ballistic eye movements (Schoonheim et al., 2010). We have shown here that activity in a small population of anatomically defined midbrain neurons is necessary and sufficient for controlling swim orientation as part of the optomotor response. Together, these data support a framework of modular neuronal control underlying vertebrate locomotion.

Experimental Procedures

Zebrafish lines

Zebrafish were raised and bred at 28°C on a 14 h light / 10 h dark cycle using standard techniques (Westerfield, 1994). All animal procedures conformed to the guidelines of the University of California, San Francisco and the Max Planck Society. Transgenic lines were made in the TLN background, which is based on the Tüpfel long-fin (TL) wildtype strain carrying mutations in mitfa (nacre, N). We used Tg(UAS-E1b:Kaede)s1999t, Tg(UAS:ChR2(H134R)-mCherry)s1986t, Tg(UAS:GFP)mpn100, Tg(UAS:GCaMP6s)mpn101, and Et(-0.6hsp70l:Gal4-VP16)^s1171t. The Gal4^s1171t line was established from a Tol2 enhancer-trap screen (Scott et al., 2007). Linker-mediated cloning established that Gal4^s1171t is inserted in the first intron of the sim1a gene (T. Thiele, unpublished result). Tg(UAS:GCaMP6s)mpn101 was constructed by first cutting the GCaMP6s open reading frame out of pGP-CMVGCaMP6s (Addgene no. 40753) and cloning it into a pTol2-14xUAS vector. This construct was then injected with transposase mRNA into one-cell-stage Gal4^s1171t embryos. Transgenic lines were maintained in either the TL or the TLN background. Designations of mutant and transgenic lines adhered to nomenclature rules set according to http://zfin.org.

Immunohistochemistry, backfills and confocal imaging

Gal4^s1171t/UAS:GFP larvae (5dpf) were fixed in 4% PFA and processed for antibody staining according to published protocols (Xiao and Baier, 2007). A mouse anti-GFP antibody (GTX13970, Genetex) was used at a concentration of 1:1000 and a goat anti-choline acetyltransferase antibody (AB144p, Millipore) was used at a concentration of 1:200. Alexa dye conjugated secondary antibodies (Invitrogen) were used at 1:1000 dilutions. Backfills were performed as described previously (Gahtan et al., 2005). Briefly, a 50% (w/v) solution of Texas Red dextran (10,000 MW, Invitrogen) was pressure injected into the spinal cord of 5-7dpf larvae anesthetized with 0.02% tricaine in Danieau’s solution. Confocal imaging was performed on either a Zeiss LSM700 or LSM780 microscope. Image processing was done using FIJI (Schindelin et al., 2012).

Electroporations

Gal4^s1171t/UAS:GFP fish (5dpf) were embedded in 2% low melting point agarose (Invitrogen) and immersed in extracellular physiological saline containing 0.02% tricaine. Patch pipettes (8-9 MΩ) were filled with intracellular saline containing 15% tetramethylrhodamine dextran (3000 MW). For MeS labeling, small GFP-positive somas dorsal to the MeL neurons were visually targeted using a 40x water immersion objective (Olympus, 0.8NA). Upon cell contact, light suction was applied, and a voltage train (1.5 second duration, 150 Hz, 1.5 ms pulse width, 2-7 volts) was applied using an Axon Axioporator (Molecular Devices). Cell morphologies were then imaged on a Zeiss LSM780 confocal microscope.

Calcium imaging

nMLF

Gal4^s1171t/UAS:GCaMP6s zebrafish (6dpf) were head-embedded in 2% low melting point agarose (Invitrogen). Agarose around the tail was dissected away using a scalpel blade. Fish were allowed to recover from the mounting procedure for several hours. Calcium responses were imaged using a customized moveable objective microscope (MOM, Sutter Instruments) and a 20x objective (Olympus XLUMP, 0.95NA). Scan control and image acquisition were controlled using ScanImage software (Pologruto et al., 2003). GCaMP6s was excited by 920 nm light (Chameleon Ultra, Coherent). Scan rates were 5.92 frames/second (256 × 256 pixels). Tail kinematics were simultaneously imaged at 100 Hz using an infrared ring light and an IR sensitive high speed CMOS camera (Photonfocus, MV1-D1312l-160-CL-12). Frame acquisition was controlled using StreamPix software (Norpix Inc). Data streams were synched using a custom Python script. Tail kinematics were scored manually and confirmed by independent observers. Data were analyzed using Igor Pro software (Wavemetrics). A threshold of 0.2 delta F/ F was used to define a calcium response.

Hypaxial muscle

Gal4^s1171t/UAS:GCaMP6s/UAS:ChR2(H134R)-mCherry larval zebrafish (6dpf) were head-embedded in 2% low melting point agarose. Imaging was performed using a water immersion objective (10x, 0.3NA) on a LSM780 Zeiss confocal microscope, controlled with ZEN software. Scan rates were 10 frames/second (128 × 128 pixels). Muscles were imaged using an Argon 488 nm laser with 0.9% power before and after unilateral nMLF stimulation. The nMLF was stimulated using a region bleaching scan mode with 40-50% laser power for 200 ms. The minimum laser power required to produce a repeatable tail deflection was used. Muscles were not imaged during the stimulation period.

ChR2 stimulation

Gal4^s1171t/UAS

ChR2(H134R)-mCherry larval zebrafish (7dpf) were head-embedded in 2% low melting point agarose. Agarose around the tail was dissected away using a scalpel blade. Fish were allowed to recover from the mounting procedure for several hours. Laser light (473 nm) was delivered to the fish’s head, using low numerical aperture multimode optic fibers (10 μm or 105 μm; HPSC10 or AFS105/125Y, Thorlabs). Optic fibers were prepared as described previously (Arrenberg et al., 2009). The position of the optic fiber was controlled using a micromanipulator (MC1000e, Siskiyou Corporation). A 473 nm direct diode laser (LuxX 80mW, Omicron) and 405 nm direct diode laser (LuxX 60mW, Omicron) were mounted within a laser beam combiner (Lighthub, Omicron) and coupled to the optic fiber. Light intensities were controlled by sending an analog voltage signal to the laser. Light intensities between 0.5 and 2 mW/mm² measured at the fiber tip were used for ChR2 activation. Tail kinematics were imaged at 250 frames per second (390×390 pixels) using a high-speed camera (Pike F032B, Allied Vision Technologies) and StreamPix software (Norpix Inc). The camera was coupled to a boom-mounted stereomicroscope (SMZ800, Nikon) with a C-mount adapter. Stimulation and imaging were synchronized using custom scripts written in LabVIEW. Muscle stimulation experiments were conducted in the same manner except a 105 μm diameter fiber was used.

For Kaede conversion experiments, the laser line was switched to 405 nm (1.8 mW) for 2 min. The light dosage for conversion experiments was therefore 60-90 times greater (intensity x duration) than that used for ChR2 experiments. Given this large difference in light exposure and the increased scattering at 405nm, our conversion experiments are likely to be an upperbound size estimate for ChR2 stimulation sites (Arrenberg et al., 2009).

Neuronal ablations

Neurons were located by position and GFP expression in Gal4^s1171t/UAS:GFP fish. Imaging and ablations were performed using the same two-photon microscope used for calcium imaging. Neurons were killed by scanning a focused 850 nm femtosecond pulsed laser beam for ~200 ms over a ~1 μm square in the center of soma. Laser power after the objective was ~270 mW/mm². Behaviors were assessed before ablations and 8-12 hours after surgery. For complete unilateral nMLF ablations, we targeted all visible MeS and MeL neurons (6-8 dorsal MeS neurons, MeS1, MeS2, MeLc, MeLr and MeLm in each animal). For MeS-only ablations, we targeted 6-8 neurons MeS neurons and MeS1 and MeS2. For MeL-only ablations, we targeted MeLc, MeLr and MeLm. The completeness of ablations was determined by imaging the ablated brain region after behavioral experiments. Data from animals with incomplete ablations were discarded.

Optomotor assay

Fish larvae (7-8dpf) were head-embedded in the same manner as for ChR2 stimulation except they were allowed to recover from mounting for 8-12 hours. We found this time-delay improved the responsiveness of mounted larvae to visual stimulation. The experimental arena and control software used for the optomotor assay were described previously (Schoonheim et al., 2010). Briefly, fish were placed in the middle of an arena surrounded by three LCD screens (5.5 × 7.5 cm), one in front of the fish and one on each side. Caudal-to-rostral drifting gratings were displayed on the two side screens to evoke a forward optomotor response. Rotating gratings were displayed on all three screens to evoke turn responses. Fish tail kinematics were imaged at 250 frames per second (390×390 pixels) using a high-speed camera (Pike F032B, Allied Vision Technologies) and StreamPix software (Norpix Inc). Grating presentation and speed were controlled using a custom script written in LabVIEW. At the start of an experiment, an ideal grating speed was determined for each fish. These speeds ranged from 16-28^o/second for forward gratings and 20-30^o/second for rotating gratings. The same grating speeds were used before and after ablations. In the text ipsiversive refers to gratings rotating toward the ablation site whereas contraversive refers to gratings drifting away from the ablation site.

Data analysis

Tail motions were tracked using custom software written in Python. Tail tracking software used OpenCV to load videos and then implemented a tracking algorithm, which returns a series of midpoints along the tail in each frame. The algorithm is seeded by a user-selected point near the base of the zebrafish larvae tail and then iterates towards the end of the tail. At each point, the tail’s lateral midpoint was located by taking a cross-section of the tail and convolving with a function representing the luminosity of a prototypical tail cross-section. The maximum of this convolution was used as the tail midpoint. This procedure is then repeated along the length of the tail returning ~40 points for each video frame. Tail angle was calculated by measuring the angle between the first midpoint near the tail base and the mean position of three midpoints at the end of the tail (to reduce noise). The detected tail midpoints were normalized to correct for small variations in the baseline position of the tail. Another custom Python script was used to segment the tracked tail movements into separate behavioral bouts. The bout detection algorithm operates by comparing the smoothed absolute value of the first derivative of the tail angle over time to a threshold. Data were further compiled and visualized using Igor Pro (Wavemetrics).

Statistics

Statistical analysis was performed in Python using the following libraries: Pandas for data structures, Scipy and Statsmodels for statistics. Welch’s t-test was used for pairwise comparisons. The family wise error rate was controlled with the Bonferonni correction. Normality and homoscedasticity were inspected visually (Q-Q plots) and using tests: Shapiro-Wilk for normality, and Bartlett’s and Levene’s for equality of variance. Individual tests were as follows: Steer angles (Figure 5C) ANOVA, tail beat frequency (Figure 5E) Box Cox transform and ANOVA, steer rise time (Figure S6) linear regression, swim duration (Figure S6) Box Cox transform and ANOVA, ablations (Figure 6, 7 and S7) paired t-test.