Zebrafish hindbrain neurons group into distinct mono- and binocular clusters
To localize and functionally characterize hindbrain neurons active during oculomotor behaviour, we stimulated larvae with patterns of moving gratings to elicit optokinetic responses while measuring GCaMP6f calcium signals in individual neurons (Fig. 1a, b).
Zebrafish show a high degree of binocular coordination: most of the time, the eyes are moved in a conjugate fashion with the notable exception of convergence during prey capture and spontaneous monocular saccades ([33], own observations). In order to assess the binocular coordination within the oculomotor system and to identify the location of internuclear neurons (INNs) and other structures, we applied a stimulus protocol (Fig. 2a) geared to decouple both eyes and reduce the gain of the non-stimulated eye to < 0.1 by showing a moving grating to the stimulated and a stable grating to the non-stimulated eye ([27], Fig. 2a′). This enabled us to classify neurons according to their innervated eye(s) based on their response profile. The stimulus consisted of stimulus phases driving primarily monocular and conjugate eye movements, respectively. The strong decorrelation of left and right eye movements enabled us to classify the monocular or binocular coding of each neuron (Fig. 2). For the characterization of neuronal response types, we calculated the correlation of neural activity traces with each of 52 regressors formed to identify neurons primarily coding for different kinematic parameters (Fig. 2c, see the “Methods” section). These parameters included eye movement direction, ocular selectivity (which identifies the encoded eye muscle(s) when combined with eye movement direction), eye position tuning, and OKR slow-phase eye velocity tuning. We found that eye motion-correlated neurons were virtually always active during clockwise or counter-clockwise binocular stimulation (2380 out of 2508 neurons, from 15 larvae with each recording depth sampled 8-fold). They only differ from each other with regard to the extent of recruitment during monocular eye movements, while neurons exclusively active during monocular eye movements are virtually absent in the hindbrain.
We identified four primary response types in our hindbrain data: two monocular (M) types with activity for either the left or the right eye (LE, RE), which were also active during the binocular stimulus phase (types MLE and MRE, Fig. 2a′, Fig. 3a, b, Additional file 1: Figure S1a-b), and two binocular response types. The binocular response types (types BA and BP, Fig. 2b, b′ and Fig. 3c, d) were either active during all three (monocular and binocular) stimulus phases (‘binocular always’, BA, Fig. 2b), or showed a preference towards binocular eye movements (‘binocular preferred’, BP, Fig. 2b′).
Since the motor range for eye movements during the binocular stimulation phase was mostly larger than during the monocular phases, we excluded all neurons that did not reach their firing threshold during the monocular phase (Additional file 2: Figure S2, Additional file 3: Figure S3).
Ninety-eight percent of eye movement-correlated neurons, caudal to the Mauthner cells, responded in an ipsiversive manner (2110 vs. 37), though this hemispheric restriction was less prominent rostral to the Mauthner cells (63%, 228 vs. 133). Eye movement-correlated neurons on the right side of the hindbrain are thus increasingly active during rightward eye positions (of the left and/or right eye) and vice versa.
Monocular neurons
Monocular position encoding neurons are primarily located in rhombomeres 5 and 6, forming two distinct columns in each rhombomere (Fig. 3a; Additional file 1: Figure S1a). A second cluster can be seen around 150 μm caudal to the Mauthner cells and 40 μm lateral to the medial longitudinal fasciculus (MLF). This region in rhombomere 7/8 partially overlaps with the areas previously described as the OI in zebrafish [15,16,17], extending caudal-ventrally into the inferior olive (IO), which we found is mostly monocular encoding. The putative OI region contains a high number of neurons encoding the position of the contralateral eye and only few neurons encoding the position of the ipsilateral eye. Within our imaged brain volume containing rhombomeres 5 and 6, position neurons coding for the ipsilateral eye span only a narrow band 30 to 70 μm ventral to the MLF (Fig. 3a left, Additional file 1: Figure S1a). This brain volume corresponds to the anatomical position of the abducens MNs, which we confirmed using a separate mnx1-transgenic line (Tg(mnx:TagRFP-T)vu504Tg, [35]) to label MNs (see overlapping grey shaded areas in Fig. 3a and Additional file 1: Figure S1a). The activity of neurons in this brain volume mostly matched the ipsilateral connections of these motoneurons to the abducting lateral rectus muscle (Fig. 1a, active for the ipsilateral eye during ipsiversive eye movements). Internuclear neurons carrying the information used to innervate the medial rectus should be located on the contralateral side and respond to contraversive positions. Such putative INNs are abundant and located more medially and dorsally than motoneurons, spanning a wider range from 60 μm ventral to around 30 μm dorsal to the MLF (see Additional file 1: Figure S1a for the labelled anatomical extents of INNs and MNs). These two clusters of putative moto- and INNs in the ABN are mirror-symmetrical between monocular left and right eye position encoding neurons (Fig. 4a). The cluster containing monocular neurons encoding movement of the contralateral eye forms again two (sub-) clusters in each hemisphere, one located dorsally, the other one located more ventrally. These two putative INN clusters were separated by a faint gap with fewer neurons 10 to 30 μm ventral to the MLF rotated roughly by 20° along the RC-axis (black arrows in Fig. 4a pointing towards the gap).
Monocular slow phase eye velocity neurons are mainly located ventrally to the MLF in rh7/8 and code for the contralateral eye. They are clustered slightly ventro-rostrally to the putative OI position neurons with some overlap between both clusters. As is the case for the monocular position neurons, the rh7/8 region also contains only few monocular velocity coding neurons for the ipsilateral eye. Rostral to these identified velocity neurons, some sparse, ungrouped neurons are located in both hemispheres, extending to the caudal end of rh6 (Fig. 3b; Additional file 1: Figure S1b).
Monocular neurons preferentially active during one monocular stimulation phase and silent during binocular movements (monocular exclusive) were heavily underrepresented for both position and velocity (159 of 2508, Additional file 4: Figure S4). Neurons exclusively active during both monocular stimulation phases were virtually absent (Additional file 5: Figure S5d).
Binocular neurons
We identified binocular neurons that were always active (BA) or were preferentially active during binocular eye movements (binocular preferred, BP). The vast majority of BP neurons encode eye position, not velocity (Fig. 3c). They overlap with monocular position coding neurons in rhombomere 7/8, but their centre of mass is shifted to a more lateral position. The rightward and leftward tuned BP neurons are distributed in the right and left hemispheres, respectively, as expected from the ipsiversive coding scheme. In the ABN, BP neurons are clustered more ventrally than neurons encoding eye movements monocularly. Furthermore, more BP neurons were found in the left hemisphere than in the right hemisphere (100 vs. 144; caudal to the Mauthner cells). We do not think that this discrepancy necessarily reflects an anatomical asymmetry/lateralization in the zebrafish, but rather was caused by sampling error or history effects from the stimulus presentation.
Binocular BA-type neurons, which are always active regardless of the stimulated eye or stimulus phase, are homogeneously distributed in the ABN and putative OI (Fig. 3d), following the pattern of their monocular counterpart, and no lateralization across hemispheres was observed. However, those BA neurons that encode velocity form a narrow band (Fig. 3d, black cells in the right panel) spanning from the dorsal end of rh6 (within our imaged region) to the location of monocular velocity coding neurons in rh7/8 and are absent from the remaining ABN and caudal rh7/8 regions.
While BA neurons responded during all stimulus phases, their responses during monocular stimulus phases were typically smaller than those during binocular stimulus phases, which can likely be attributed to the smaller explored motor range during monocular stimulation (for an assessment of response type classification see the “Methods” section, Additional file 1: Figure S1d).
While monocular and binocular position neurons share roughly the same anatomical locations in the zebrafish hindbrain, an anatomical response type gradient exists for velocity neurons caudal to rh6 (Fig. 4b): binocular velocity neurons are located more rostro-dorsally while monocular velocity neurons form a cluster in the ventral part of rh7/8.
Having identified four primary response types, we next sorted all occurring response types according to the number of identified neurons for each response type and grouped them according to the encoded eye direction (CW, CCW), controlled eye muscles (lateral rectus, medial rectus, or both), and kinematic parameter (eye position or OKR slow-phase velocity). This analysis (Fig. 4c) revealed that position neurons are more frequent in the hindbrain than slow-phase eye velocity neurons (1938 position vs. 570 velocity). We found more monocular neurons coding for the medial rectus than monocular neurons coding for the lateral rectus eye muscle (1043 medial vs. 618 lateral). Also, using our stimulus protocol, we found more neurons coding for the position of the right eye than for the left eye position (779 right vs. 582 left; this might have been caused by a history dependence, as in 90% of the recordings the left eye was monocularly stimulated before the right eye). For all mono- and binocular response types, we found neurons dorsal to the MLF and rostral to the Mauthner cells which show an intermingled anatomical distribution of ipsiversive and contraversive response types. This cluster corresponds to the caudal end of the previously described “hindbrain oscillator” (also termed ARTR, [3, 5, 6], Fig. 3, Additional file 1: Figure S1).
To reveal the here reported coding properties of neurons in the hindbrain, we made use of response type classification (Figs. 2, 3, and 4a–c). While this approach is useful to get an overview of the anatomical distributions of the different functionally identified neurons, such classification approach is rather ignorant to the possibility that neuronal responses might form a continuum in-between classified response types. We looked into this issue by first checking the per-neuron difference of correlation to the left versus the right eye (see the “Methods” section). As expected, binocular neurons were located in the centre and had a unimodal distribution, while monocular neurons were distributed more towards the sides caused by the left and right coding population [Fig. 4d, Index running from − 1 (more monocular left eye position coding) to 1 (more monocular right eye position coding)]. The results presented in Fig. 4d, and other publications [36], indicate that the responsivity of neurons is graded. The oculomotor neuron population forms gradients within the parameter space spanned by the regressors used in our response type classification. Thus, our binary analysis—while providing a useful simplification for understanding the oculomotor processing repertoire—disregards the existing functional gradients. Since oculomotor neurons can code for many parameters in parallel, the response type classification can in addition be biased by existing correlations. For example, eye movements during the binocular stimulus phase were faster than during the monocular stimulus phases, which could have resulted in some of the BP position neurons being classified as BP (and not BA) due to a weak encoding of eye velocity in these BP neurons. Furthermore during binocular stimulation, more eccentric eye positions were reached than during monocular stimulation. We checked for these issues by comparing the velocity influence of BA (n = 206) and BP (n = 306) position coding neurons (see the “Methods” section, Fig. 4e). We found that both groups showed similar velocity-position distributions, with BA position neurons having a slightly stronger position component than BP position neurons (two-sided Wilcoxon rank-sum test, p = 5.7*10–7, Index running from − 1 (velocity) to 1 (position)). The firing thresholds (from the firing threshold analysis, Additional file 2: Figure S2, Additional file 3: Figure S3) of BP position neurons were shifted towards the ON direction compared to BA and monocular position neurons, and for the right eye, BA neurons showed significantly earlier thresholds than MRE neurons (Fig. 4f, g). These observed threshold differences likely result from experimental intricacies such as sampling bias. This control analysis shows that the BP position classification was likely slightly affected by velocity components and a larger dynamic range of eye positions during the binocular stimulation phase, and furthermore, some BP neurons were also active during the monocular stimulation phases, albeit at low activity levels preventing their classification as BA or monocular. Taken together, this suggests that BA and BP neurons might not be two distinctively separate groups but that they exist along a continuum, with the extreme cases being BA and BP.
Differential encoding of velocity and position in individual neurons
Our first experiment was geared towards identifying monocular versus binocular tuning. We also classified neurons as either mainly position or mainly velocity encoding (Fig. 3) in this experiment, although intermediate “multi-dimensional” responsivity likely occurs as well. ABN neurons should receive slow-phase velocity signals during optokinetic stimulation, e.g., via the pretectum, vestibular nuclei, cerebellum, and the OI (Fig. 1a′ [8, 23, 37,38,39]) since a muscle force step is needed to overcome the dampened, viscous kinetics of the oculomotor plant [40, 41]. In order to investigate the differential coding of oculomotor neurons and to visualize the anatomical distribution of position and velocity coding within rhombomeres 7/8, we developed a binocular closed-loop stimulation protocol to disentangle eye position from eye velocity correlations by eliciting different eye velocities at different eye positions (Fig. 5a–a″, see the “Methods” section). This allowed us to consistently evoke combinations of eye position and velocity which would only occur sporadically during optokinetic responses to fixed stimulus sequences. At the same time, the stimulus protocol minimized the occurrence of fast phase eye movements (saccades) in order to improve our ability to relate neuronal activity to slow phase behaviour in this correlative experiment, i.e., the experiment was not designed to identify or characterize the burst system responsible for generating saccades [3, 42]. From the whole recording, we constructed two-dimensional tuning curves covering the activity for almost all different eye position and slow phase eye velocity combinations within a certain range (eye position: − 15° to + 15°, eye velocity: − 7 to + 7°/s, Fig. 5b–d, Additional file 6: Figure S6a-c). Using this protocol, we analysed 889 neurons, which exhibited different combinations of eye position and slow-phase eye velocity tuning. To classify the differences in position and velocity coding for each of these neurons, we calculated a Position-Velocity index (PVIndex) based on the correlation of the neuronal response to behavioural regressors (see the “Methods” section). This index runs from − 1 (pure velocity coding) to + 1 (pure position coding). Both neurons tuned exclusively to position (neuron 1) or velocity (neuron 3) exist, as well as intermediate cases (neuron 2, Fig. 5b–d). For neurons with an intermingled position and velocity component (− 0.5 < PVIndex < 0.5), the preferred direction was almost always the same for position and velocity (94%, 440/470).
Firing thresholds of position neurons are distributed across a broad range of eye positions while velocity neurons mainly get activated at velocities close to 0°/s
To quantify the neuronal recruitment, we used the two-dimensional tuning curves and analysed the activation thresholds for position and velocity in the position and velocity planes intersecting with the origin. This procedure results in one-dimensional eye position tuning curves around eye velocities of 0°/s (black and red line in Fig. 5b–d middle panel) and eye velocity tuning curves around eye positions of 0° (right panel) for the same neurons. Since it is difficult to detect the true onset of action potential firing (firing threshold) using our measured calcium indicator fluorescence signals (see the “Methods” section), the identified activation thresholds were likely positioned slightly in the ON direction relative to the true firing threshold in each neuron. For position encoding neurons (PVIndex > 0, n = 533 neurons with identified position threshold), we found that the eye position thresholds are distributed across a broad motor range (roughly − 10° to + 10°, Fig. 5e). Leftward and rightward eye position encoding neurons had slightly different eye position thresholds in our dataset [Wilcoxon rank sum p = 0.000016, median for rightward coding neurons pooled on ON direction (n = 250): 5.5°, leftward 4.5° (n = 283)]. Given the small difference, we are not convinced that this discrepancy represents actual asymmetries in the zebrafish larvae, but rather stems from history-dependent effects or the optical setup. For the velocity-encoding neurons (PVIndex < 0, n = 279), the activation thresholds for velocity mostly span a range between ± 2°/s, so that the calcium signals started to increase at eye velocities close to 0°/s. Some of the neurons were already active at velocities below 0°/s and thus were tuned to both negative and positive velocities. No difference was observed between velocity neurons coding for leftward vs. rightward velocities (Fig. 5f, Wilcoxon rank sum p = 0.24; rightward n = 104, leftward n = 175). The strongest fluorescence increases were usually observed after crossing a velocity of 0°/s. However, as mentioned above, the true firing thresholds may start further into the OFF direction (≤ 0°/s) as (i) we likely could not reliably detect single action potentials using GCaMP6f in our preparation [43] and (ii) our statistical test used to detect thresholds was quite conservative (see the “Methods” section, Additional file 3: Figure S3).
Visual inspection of all strong velocity neurons (PVIndex < − 0.5) revealed that some of the identified velocity neurons showed firing saturation at higher velocities (29%; 40 of 139; Fig. 5g). Calcium indicator saturation, which occurs at high calcium concentrations ([Ca]2+> > Kd), is unlikely to account for the observed fluorescence saturation, since the dynamic range of fluorescence values (FMax/FMin) was (i) much smaller (~ 2.5) than the published range of the GCaMP6f indicator (51.7) [43] and (ii) similar for non-saturating position neurons and saturating velocity neurons (Fig. 5g).
For the two velocity tuning types (saturating vs. non-saturating), no clear anatomical clustering is visible (Additional file 7: Figure S7) and we therefore merged the corresponding neurons into one group (potentially the non-saturating neurons could still saturate at higher eye velocities not reached in our experimental protocol).
No anatomical gradients of oculomotor tuning thresholds in the hindbrain
In order to investigate topographical arrangements of tuning thresholds in the hindbrain, we generated anatomical maps of firing thresholds for position (PThres) and velocity (VThres) for position neurons with an identified threshold (PVIndex > 0, n = 533, Additional file 8: Figure S8a) and for velocity neurons (PVIndex < 0, n = 279, Additional file 8: Figure S8b). Position thresholds do not appear to be anatomically grouped, and no clear anatomical gradient within any of the neuronal clusters could be identified (Kruskal-Wallis test for position threshold differences p = 0.07; rh5: 214; rh6: 249; rh7/8: 27). We investigated whether MNs (based on anatomical location) are distributed topographically according to position firing threshold, but were unable to identify a significant gradient [Kruskal-Wallis p = 0.22, Additional file 8: Figure S8a].
Eye velocity thresholds (VThres) also did not show any spatial clustering, and no gradient could be observed within the hindbrain. No statistical difference was observed (Kruskal-Wallis p = 0.79; rh5: 11; rh6: 10; rh7/8: 184).
Neurons in rhombomere 7/8 exhibit a velocity-to-position gradient
The anatomical clusters of position and velocity coding neurons that we identified using the PVIndex from the closed-loop experiment were generally in agreement with those obtained from the separate experiment described above (compare Fig. 6a–c to Fig. 3 and Additional file 1: Figure S1). Neurons in the ABN (rh5/rh6) displayed an average PVIndex of 0.44 (± 0.23 STD; n = 521) indicating position tuning with some minor velocity sensitivity. Within the ABN, the velocity component is strongest around a gap (described above in the “Monocular neurons” section, see Fig. 4a, black arrows) in-between two clusters of neurons 20 μm ventral to the MLF. The velocity neurons identified using the velocity-position stimulus reside in the ventral part of rh7/8 and extend into the area caudal to rh6, overlapping with the volumes containing the BA, MLE, and MRE velocity neurons (Fig. 3b–d, Additional file 1: Figure S1b). In the caudal part of rhombomeres 7/8, we found neurons with more position coding dependence than in the rostral part, especially laterally (Fig. 6a–c). Following the anterior-posterior and ventro-dorsal axes in the caudal hindbrain (rh7/8), our analysis therefore reveals a prominent PVIndex gradient, shifting from velocity towards an intermingled velocity/position tuning with neurons exhibiting a stronger position coding at the dorso-caudal end.