Plastic corollary discharge predicts sensory consequences of movements in a cerebellum-like circuit

SUMMARY

The capacity to predict the sensory consequences of movements is critical for sensory, motor, and cognitive function. Though it is hypothesized that internal signals related to motor commands, known as corollary discharge, serve to generate such predictions, this process remains poorly understood at the neural circuit level. Here we demonstrate that neurons in the electrosensory lobe (ELL) of weakly electric mormyrid fish generate negative images of the sensory consequences of the fish’s own movements based on ascending spinal corollary discharge signals. These results generalize previous findings describing mechanisms for generating negative images of the effects of the fish’s specialized electric organ discharge (EOD) and suggest that a cerebellum-like circuit endowed with associative synaptic plasticity acting on corollary discharge can solve the complex and ubiquitous problem of predicting sensory consequences of movements.

INTRODUCTION

Predicting the sensory consequences of an animal’s own behavior is a critical function of the nervous system. In the sensory domain, predicting and cancelling sensory input caused by an animal’s own movements allows for more effective processing of behaviorally relevant stimuli(Cullen, 2004;Sperry, 1950;von Holst and Mittelstaedt, 1950). Though many sensory regions, including sensory areas of cerebral cortex, receive input from motor systems, the functions of such inputs remain largely unknown(Crapse and Sommer, 2008;Poulet and Hedwig, 2007). According to theoretical accounts of motor control, online predictions of the sensory consequences of motor commands, known as forward models, are critical for generating fast and accurate movements despite noise and delays in sensory feedback(Miall and Wolpert, 1996;Shadmehr and Krakauer, 2008). Though converging lines of evidence suggest that the mammalian cerebellum is involved in predicting sensory consequences of motor commands(Anderson et al., 2012;Bastian, 2006;Brooks and Cullen, 2013;Ebner and Pasalar, 2008;Wolpert et al., 1998), detailed knowledge of the underlying circuit mechanisms is lacking. Finally, numerous lines of evidence suggest that failures of corollary discharge-based predictions contribute to psychotic symptoms in schizophrenia(Ford and Mathalon, 2012), though here as well the underlying mechanisms are unknown.

Studies of weakly electric mormyrid fish have provided unique insights into the cellular and circuit mechanisms for predicting the sensory consequences of a simple electromotor behavior — the EOD. Mormyrid fish emit brief, highly stereotyped EOD pulses for communication and active electrolocation. However, the fish’s own EOD also affects passive electroreceptors tuned to detect external fields(Bell and Russell, 1978). This problem is solved at the level of ELL principal cells, where input from electroreceptors is integrated with input from a mossy fiber-granule cell-parallel fiber system conveying timing signals related to the EOD, known as electric organ corollary discharge (EOCD). Anti-Hebbian plasticity at parallel fiber synapses onto principal cells sculpts patterns of activity that are temporally-specific negative images of principal cell response to the EOD(Bell, 1981;Bell et al., 1993;Bell et al., 1997b;Roberts and Bell, 2000). Negative images serve to cancel out responses to the fish’s own EOD, allowing responses to external fields to be processed more effectively.

The circuitry of the mormyrid ELL is similar in numerous respects to that of the mammalian cerebellum, including the presence of granule cells that provide plastic input to Purkinje-like cells via a system of parallel fibers, as well as Golgi cells, unipolar brush cells, and inhibitory molecular layer interneurons(Bell et al., 2008). ELL neurons also receive electrosensory input, which, although clearly different from climbing fiber input to Purkinje cells, may function analogously insofar as both serve to instruct plasticity at parallel fiber synapses. Indeed, roles for granule cells and parallel fiber plasticity established in previous experimental and theoretical studies of ELL(Bell, 1981;Bell et al., 1997b;Kennedy et al., 2014;Roberts and Bell, 2000) closely resemble longstanding Marr-Albus(Albus, 1971;Marr, 1969) and adaptive filter models(Dean et al., 2010;Fujita, 1982) of mammalian cerebellar cortex. Given these similarities, studies of ELL may shed light on the more complex problem of understanding adaptive functions of the mammalian cerebellum(Boyden et al., 2004;Gao et al., 2012;Ke et al., 2009;Schonewille et al., 2011).

Are mechanisms described previously for generating negative images of the effects of the EOD powerful and flexible enough to solve the more difficult problem of generating negative images of the sensory consequences of movements (Figure 1)? Whereas the EOD motor command is a completely stereotyped event generated by a small number of neurons in a dedicated command nucleus(Bennett et al., 1967;Grant et al., 1986), movement motor commands are numerous, diverse and generated by a far more complex and distributed motor system. Here we show that in addition to EOCD signals, ELL neurons also receive movement-related corollary discharge signals from the spinal cord. Despite the major differences between the EOD and movements, ELL neurons form flexible and accurate negative images based on this movement-related corollary discharge. These results provide direct neurophysiological evidence for predictions of the sensory consequences of movements based on plastic corollary discharge. These results also suggest that the fairly complete and well-tested model of corollary discharge function established for the mormyrid ELL in the context of specialized electromotor behavior may be broadly relevant for understanding how corollary discharge signals operate in other systems in the context of movements.

Figure 1. Simple Scheme for Predicting Movement Consequences in ELL.

(A) Changes in the electric field due to movements of the electric organ in the tail (filled arrow) are typically proportional to tail displacement from the midline, with tail movements towards the side of the receptive field resulting in an increase in the local electric field amplitude and an increase in electroreceptor activation. The electrosensory consequences of two different tail movements as a function of time are schematized in the bottom panel. (B) ELL principal cells integrate electrosensory input with parallel fiber input from granule cells. Granule cells are located in an external cell mass, known as the eminentia granularis posterior (EGp) and receive excitatory mossy fiber input from a variety of sources. Previous studies have described mossy fibers conveying EOD motor command timing information and proprioceptive input, but whether mossy fibers convey corollary discharge (CD) signals related to movements is unknown. Anti-Hebbian plasticity at synapses between granule cells and ELL principal cells underlies the cancellation of predictable patterns of electrosensory input. In order to effectively predict the sensory consequences of movements, principal cells must be able to store multiple negative images related to different movement commands.

RESULTS

Corollary Discharge Responses during Fictive Swimming

Though previous studies have thoroughly characterized EOCD inputs to ELL(Bell et al., 1992;Bell et al., 1983;Kennedy et al., 2014), it is not known whether ELL also receives corollary discharge signals related to motor commands for swimming. To address this, we developed a fictive swimming preparation that allowed us to isolate putative movement-related corollary discharge from somatosensory and electrosensory signals. First, we evoked rhythmic swimming movements by continuous microstimulation (40 or 100 Hz) of the mesencepahlic locomotor region (MLR)(Fetcho and Svoboda, 1993;Le Ray et al., 2011;McClellan and Grillner, 1984;Uematsu and Todo, 1997). Movements ranged from 1–6 Hz and ceased upon termination of microstimulation. The frequency of the movements was always far below the frequency of continuous microstimulation, and, clearly graded with stimulus intensity at relatively low current strengths, although the absolute values of the current strength varied across fish (Figure 2A). After characterizing the movements evoked by microstimulation we paralyzed the fish (eliminating movement-related somatosensory and electrosensory signals) and monitored motor commands directly by recording from motor nerves exiting the spinal cord in the dorsal ramus of the ventral root. Nerve recordings revealed rhythmic bursts of activity, the frequency of which graded with microstimulation current intensity, comparable to swimming frequencies prior to paralysis (Figure 2B, top traces). As expected, simultaneously recorded motor commands to discharge the electric organ were entirely distinct from activity recorded from spinal motor nerves (Figure 2B, bottom traces) in that the timing of their occurrences was unrelated to motor nerve activity. Hence, motor nerve activity reflects commands related to swimming movements.

Figure 2. Motor Patterns and Mossy Fiber Responses Evoked by Fictive Swimming.

(A) Tail position measured by laser displacement sensor in response to three intensities of 100 Hz microstimulation in the MLR. (B) Recordings of dorsal ramus of the ventral root show motor patterns evoked by three intensities of 100 Hz MLR microstimulation. Bottom rows (blue) depict the spinal electromotoneuron volley (EMN) that in an unparalyzed fish would cause an EOD. EMN was measured simultaneously via an electrode near the electric organ. Scale bars, inset: 10uV, 3 ms. (C) Smoothed spike rate (20 ms Gaussian window) from an extracellular recording of an EGp mossy fiber (MF) in response to microstimulation-evoked fictive swimming at two frequencies. Black trace is the rectified and smoothed motor command signal recorded in the dorsal ramus of the ventral root, scaled for ease of comparison to mossy fiber firing rate. (D) Frequency at power spectral density (PSD) peak from smoothed spike rate trace vs. frequency at PSD peak from rectified, smoothed motor nerve burst trace for all recorded mossy fibers at all tested microstimulation-evoked frequencies. Each color corresponds to the same mossy fiber. Open circles indicate spontaneous fictive swimming. Gray dotted line is the regression line. (E) Frequency of firing rate at PSD peak for all mossy fibers in which microstimulation evoked two frequencies of fictive swimming movements. Black line indicates the mean. mossy fibers were analyzed regardless of whether motor nerve signals were obtained. See also Figures S1 and S2.

Previous studies have shown that mossy fiber inputs to the eminentia granularis posterior (EGp)—a cell mass that contains the granule cells that project to ELL (Figure 1B)—originate from several sources in the brain and spinal cord(Bell et al., 1981;Szabo et al., 1990;Szabo et al., 1979) and convey a variety of information including EOCD, proprioceptive, and electrosensory signals(Bell et al., 1992;Kennedy et al., 2014;Sawtell, 2010). To test whether some mossy fibers convey movement-related corollary discharge, we combined the fictive preparation described above with extracellular recordings from putative mossy fiber axons in EGp(Bell et al., 1992;Kennedy et al., 2014;Sawtell, 2010). A subset of tonically-active mossy fibers exhibited firing rate modulation (greater than three S.D. from baseline) during spontaneous (Figure S1) or microstimulation-evoked motor nerve activity (n=23 of 48 fibers; Figure 2C). Further analysis was performed on those fibers that included periods of rhythmic motor nerve activity (19 of 23 fibers had such periods). For these fibers, rhythmic firing rate modulations were correlated with motor nerve activity (cross-correlation between firing rate and smoothed motor nerve bursts, r=0.40±0.18; n=19 fibers). We also observed a strong correlation between the frequency of mossy fiber firing rate modulation (as measured by the frequency at which the peak occurred in the power spectral density (PSD); see Experimental Procedures) and frequency (at PSD peak) of smoothed motor nerve bursts (r=0.79, n=19 fibers, 1–3 frequencies per fiber for n=29 total observations; Figure 2D), including for spontaneous bursts (n=2 fibers; open circles on Figure 2D). For a subset of mossy fibers tested with two microstimulation amplitudes, we found that the frequency of firing rate modulation (at PSD peak) increased with microstimulation intensity (amp1: 2.37±1.17 Hz; amp2: 3.71±1.19 Hz, n=9 fibers, p=0.0039, sign test; Figure 2E). Hence, a subset of mossy fiber conveys graded motor information related to the frequency of rhythmic swimming movements.

Since any movement that alters the position of the electric organ relative to electroreceptors has electrosensory consequences, corollary discharge signals should be engaged for different types of movements. To test this, we used microstimulation of the optic tectum, homologue of the mammalian superior colliculus, to evoke rapid, isolated tail and trunk movements characteristic of orienting or escape behavior(Herrero et al., 1998;Saitoh et al., 2007). A brief microstimulation train (500 Hz, 10–15 pulses) applied to the tectum evoked a single, rapid unilateral tail movement, the speed and amplitude of which graded smoothly with stimulus intensity (Figure 3A). We next paralyzed the fish and recorded extracellularly from mossy fibers (Figure 3B). In a subset of mossy fibers we observed activity that graded with stimulus intensity, typically in the form of brief bursts or pauses (Figure 3C). A summary of these responses is shown in Figure 3D (bursts, difference in integrated modulation: amp2-amp1, 342±281 spikes, n=10 fibers, p=0.002, sign test; pauses, difference in integrated modulation: amp2-amp1, −240±193 spikes, n=7 fibers, p=0.0156, sign test).

Figure 3. Mossy Fiber Responses Evoked by Microstimulation of the Optic Tectum.

(A) Microstimulation-evoked tail movements in response to four stimulus intensities as measured by a laser displacement sensor. (B) Extracellular traces from two representative mossy fibers in response to tectal microstimulation. Left trace also appears in 3F, bottom panels; right trace also appears in 3C, top panels. (C) Smoothed spike rate histograms from two mossy fibers in response to two amplitudes of tectal microstimulation. Green traces represent tail position. Note graded bursts in top example and graded pauses in bottom example. (D) Box plots of integrated firing rate modulations of mossy fibers, segregated by response pattern. Integrated modulations were calculated by summing spikes over a small window following microstimulation (~100 ms). Black dots correspond to individual data points. (E) Tail movements evoked by microstimulation at two different sites in the optic tectum. (F) Smoothed spike rate histograms from two mossy fibers in response to tectal microstimulation at two sites. (G) Summary of mossy fiber responses to tectal microstimulation at two sites, segregated by recording location. (H) Smoothed spike rate histograms from a single mossy fiber responding to both MLR and tectal microstimulation. Top, green traces represent microstimulation-evoked tail displacement prior to paralysis. See also Figures S2 and S3.

From the perspective of an ELL neuron, an ipsilateral tail movement (one that brings the electric organ closer to the neuron’s receptive field) will have an electrosensory consequence opposite to that of a contralateral tail movement(Sawtell and Williams, 2008). Hence, in order to be useful for cancelling the electrosensory consequences of movements, corollary discharge signals must distinguish between different motor commands. To test this, we microstimulated at two sites in the tectum which reliably evoked tail movements in opposite directions (Figure 3E). We found that a majority of mossy fibers responded either selectively to one stimulation site (Figure 3F, top) or oppositely to stimulation of the two sites (Figure 3F, bottom), consistent with corollary discharge signals providing a basis for distinguishing between motor commands related to different movements (opposite: n=11; same: 26; one side: n=59).

Finally, mossy fibers modulated by MLR or tectal microstimulation did not exhibit responses to the EOD motor command (Figure S2). Hence, as expected based on previous anatomical studies(Bell et al., 1995;Bell et al., 1992;Bell et al., 1983;Bell et al., 1981), corollary discharge signals related to movements are conveyed via separate pathways from EOD motor command signals. We obtained additional insight into the origins of movement-related corollary discharge signals by recording in a superficial fiber tract at the anterior margin of EGp which contains mossy fiber axons originating from the spinal cord (Figure S3). Previous anatomical studies have shown that this pathway shares a number of similarities with the ventral spinocerebellar tract in higher vertebrates(Szabo et al., 1979;Szabo et al., 1990). Mossy fibers recorded in the spinal tract exhibited bursts and pauses in response to tectal microstimulation similar to those recorded in EGp. A summary of mossy fiber responses to tectal microstimulation at two sites, segregated by recording location, is shown in Figure 3G (n=22 spinal tract fibers; n=74 EGp fibers). These results suggest that corollary discharge signals reach EGp via the spinal cord, instead of being relayed via branches of a descending motor pathway originating from the tectum. Consistent with this, anatomical studies have failed to reveal direct connections between EGp and the tectum or other brain centers involved in controlling movement are absent. If corollary discharge signals return to EGp from the spinal cord, it might be expected that the same fibers would relay motor command-related signals irrespective of the central origin of the commands, as has been shown for some mammalian spinocerebellar pathways, see e.g. (Jankowska et al., 2011). To test this, we recorded from EGp mossy fibers in two paralyzed fish in which we evoked both rhythmic movements via MLR stimulation and rapid, isolated tail movements via tectal stimulation. Fourteen of twenty-one mossy fibers modulated by MLR microstimulation were also modulated by tectal microstimulation (example in Figure 3H).

Corollary Discharge Inputs to ELL Neurons Are Plastic

Negative images of the electrosensory consequences of the EOD described previously depend critically on the capacity to shape the effects of EOCD signals on ELL neurons via mechanisms of associative plasticity. Direct in vivo evidence for such plastic shaping of EOCD signals has been obtained by pairing EOD motor commands with dendritic spikes evoked intracellularly in medium ganglion (MG) cells(Bell et al., 1993;Kennedy et al., 2014;Sawtell et al., 2007). MG cells inhibit glutamatergic output neurons of ELL, occupying a position in ELL circuitry analogous to Purkinje cells in the cerebellum(Bell, 2002;Bell et al., 2008). Such experiments have revealed temporally-specific depression of EOCD responses, consistent with anti-Hebbian spike timing dependent plasticity at parallel fiber-MG cell synapses documented in vitro(Bell et al., 1997b;Han et al., 2000). We conducted similar experiments but using tectal microstimulation at two sites to evoke motor commands related to two different movements (Figures 3E and 4A). We obtained whole-cell recordings from MG cells and paired a dendritic spike evoked by intracellular current injection at a fixed delay (50 or 100 ms) after microstimulation of one of the tectal sites. Subthreshold responses to microstimulation before pairing were modest, but after pairing for 5–10 minutes, we observed a response depression that was both specific to the paired microstimulation site and greatest at the paired delay (Figures 4B and 4C; 50 ms, n=20 pairings from 14 cells; 100 ms, n=9 pairings from 8 cells). Because pairing is restricted to the recorded cell, changes observed in these experiments likely reflect plasticity at synapses conveying movement-related corollary discharge signals to MG cells. Previous in vitro, in vivo, and modeling studies suggest that the observed response depression can be explained by removal of excitation mediated by selective weakening of parallel fiber synapses active before the dendritic spike(Bell et al., 1997b;Kennedy et al., 2014;Roberts and Bell, 2000). Both the temporal- and site-specificity of the response depression are notable. The former suggests that individual MG cells may possess the capacity to generate negative images of the electrosensory consequences of movements that are extended in time relative to the motor command that evokes them, while the latter suggests a capacity to generate and store multiple negative images related to different movements.

Figure 4. Site- and Temporally-specific Plasticity of Corollary Discharge Responses in MG Cells.

(A) Tail position measured by laser displacement sensor in response to microstimulation at two sites in the optic tectum. (B) Traces from a representative MG cell recorded in same experiment as laser traces in A, showing average microstimulation-evoked synaptic responses before pairing (pre, top row), during pairing (middle row), and after pairing (post, third row). In the post and pre conditions, narrow spikes were digitally removed and membrane potentials interpolated before averaging. The middle panel shows five overlaid traces taken during the pairing period. For this cell, current injections were delivered to evoke a dendritic spike, paired at a fixed delay to microstimulation at site 1 (dotted line). The bottom panels show thedifference in the average microstimulation-evoked responses before and after pairing. Note that the depression is restricted to the paired microstimulation site and greatest around the delay at which the spike was paired. (C) Average difference traces across cells, pooled independently of microstimulation site. Current injections were delivered at two delays relative to microstimulation (arrows) and restricted to one microstimulation site, showing both site-and temporal-specificity of the response depression. Gray outlines indicate standard error of the mean (SEM) and gray boxes obscure microstimulation artifacts.

Negative Images Based on Corollary Discharge Signals

The most important result at the core of previous models of ELL adaptive function is that ELL neurons are capable of generating highly-specific negative images of the electrosensory consequences of the fish’s own EOD. We performed experiments in order to determine whether ELL neurons could likewise generate negative images of the electrosensory consequences of movements evoked by MLR or tectal microstimulation. We simulated natural patterns of activation for the electrosensory system by delivering a brief electrical pulse following each EOD motor command, mimicking the duration and timing of the fish’s own EOD. We also used microstimulation of the electromotor command pathway (see Experimental Procedures) to achieve EOD command rates within the range of those observed in swimming fish (13Hz in our experiments). Because of the additional demands of the electrosensory stimulation, we did not monitor motor nerve activity in these experiments. Instead, we delivered brief pulses to the MLR (500 Hz, 10–15 pulses) to evoke rapid ipsilateral tail movements, similar to movements evoked by tectal stimulation.

We then paralyzed the fish and made extracellular single-unit recordings from ELL principal cells, including both putative MG cells and ELL output cells (Figure 5B)(Bell et al., 1997a;Bell and Grant, 1992). Responses to MLR stimulation were compared before and after a pairing period (10–15 min) during which the amplitude of a local electrosensory stimulus (ES) applied to the neuron’s receptive field was smoothly graded as a function of the temporal profile of the microstimulation-evoked tail movement (Figure 5A), as measured before paralysis using a laser displacement sensor (dashed lines in Figures 5–7). Before pairing, responses to microstimulation were small or absent. After pairing, responses to microstimulation resembled smoothly graded negative images of the response to the ES during pairing (Figures 5C and 5D; n=4). Similarly, smoothly graded negative images were obtained for tectal microstimulation in separate experiments (Figures 5E and 5F;n=14). These results suggest that ELL neurons possess the capacity to transform brief motor commands into much longer-lasting patterns of activity that are temporally-aligned with and appropriate to cancel the electrosensory consequences of movements. Indeed, the timing of the peak modulation of mossy fibers in response to tectal microstimulation (46.98±7.63 ms, n=106 observations from n=70 fibers) is far more restricted than the peak modulation of negative images (121.21±21.85 ms, n=14 cells). Finally, the fact that negative images were similar in these two sets of experiments also suggests a general capacity to cancel the electrosensory consequences of motor commands regardless of the origins of the motor commands.

Figure 5. ELL Principal Cells Exhibit Negative Images Based on Corollary Discharge.

(A) Top row, green trace: representative microstimulation-evoked tail movement measured by laser displacement sensor prior to paralysis. Top row, dotted gray trace: Waveform envelope used as look-up table for delivering local ES. Note that waveform is based on the pre-paralysis movements but is not an exact match. For purposes of pooling across fish in which movements varied in their exact time course, we created an idealized waveform closely modeled on the typical time-course of movements. Middlerow: timing of EOD motor commands. Note constant rate of 13Hz due to microstimulation of the EOD command pathway. Bottom row: A local ES, placed in the receptive field of the recorded cell, was varied in amplitude according to the timing of the command and the relative value of the look-up table waveform. A constant-value global ES was delivered in conjunction with the local one, allowing for bidirectional modulation of the local EOD mimic’s amplitude around a constant mean. (B) Representative extracellular trace from cell depicted in 5C. Large gray box obscures tectal microstimulation artifact. Small gray boxes obscure electromotor command pathway microstimulation artifact (first small gray box of each pair) and global electrosensory stimulus artifact (second small gray box of each pair). (C) Smoothed (20 ms boxcar filter) spike rate histograms from an extracellular recording of an ELL principal cell illustrating typical response patterns as a function of time relative to microstimulation of the MLR before, (pre, top row), during (pairing, second row), and after (post, third row) pairing with a local ES (dotted gray line). Non-smoothed histograms shown in gray (1 ms bins). The difference traces in the bottom panels show the effects of pairing (after pairing minus before pairing). (D) Average of difference traces pooled across cells. Difference traces were constructed by subtracting the smoothed spike rate histograms. (E) Smoothed spike rate histogram from an extracellular recording of an ELL principal cell illustrating typical response patterns as a function of time relative to microstimulation of the tectum before, (pre, top row), during (pairing, second row), and after (post, third row) pairing with a local ES (dotted gray line). Non-smoothed histograms shown in gray (1 ms bins). The difference traces in the bottom panels show the effects of pairing (after pairing minus before pairing). (F) Average of difference traces pooled across cells. For averages shown in D and F cells were pooled to match polarity of difference trace irrespective of site or cell type. Each cell is normalized to its pre-pairing baseline firing rate. In all panels, gray outlines indicate SEM across cells and gray boxes obscure microstimulation artifacts.

Figure 7. Negative Images Depend on Corollary Discharge and Proprioceptive Signals Conveyed by an Ascending Spinal Mossy Fiber Pathway.

Negative images were induced in ELL principal cells under the conditions that mimic real movements (simultaneous rapid tail movements driven by a computer-controlled stage and tectal microstimulation), and then probed under the fictive (tectal microstimulation alone) and passive conditions (tail movements alone). (A) Left column: Smoothed spike rates from an extracellular recording of an ELL principal cell illustrating typical response patterns as a function of time relative to microstimulation of the optic tectum before, (pre, top row), during (during, second row), and after (post, third row) pairing with a local ES (dotted gray line). Right column: Same as left column except histograms are triggered by onset of tail movement. Non-smoothed histograms shown in gray (1 ms bins). (B) Top row: Average of difference traces pooled across cells probed under fictive (left) and passive (right) conditions. Bottom row: Following injection of 2% lidocaine into the spinal cord, negative images were abolished under both conditions, suggesting a spinal origin for both proprioceptive and corollary discharge signals. Error bars represent SEM across cells. Gray boxes obscure microstimulation artifacts.

The number and variety of sensory patterns evoked by movements is far greater than the sensory patterns resulting from the EOD, raising the question of the capacity of ELL neurons to generate and simultaneously store multiple negative images appropriate to cancel sensory consequences of different movements. Though testing many different movements was impractical, our preparation allowed us to ask whether ELL neurons were capable of forming two different negative images. We alternated microstimulation of the two tectal sites, and paired each with smoothly graded but opposite polarity changes in ES amplitude (Figure 6A). This mimics the natural situation in which ipsilateral versus contralateral tail movements have opposite electrosensory consequences. As in the previous results, before pairing, responses in ELL principal cells to microstimulation were small or absent. After pairing, responses to microstimulation resembled smoothly graded negative images of the response to the ES during pairing (Figure 6B). Notably, negative images were specific to the site of tectal microstimulation and bi-directional — i.e., the same neuron had the capacity to store two, opposite negative images consisting of either graded increases or decreases in firing depending on the effects of the ES during pairing (Figure 6C; n=9). Decreases in firing observed after pairing with an excitatory ES can be explained by previously described anti-Hebbian spike timing-dependent depression at parallel fiber synapses, while increases in firing observed after pairing with an inhibitory ES can be explained by previously described non-associative potentiation at parallel fiber synapses(Bell et al., 1997b;Han et al., 2000).

Figure 6. ELL Principal Cells Can Store Two Different Negative Images in Relation to Different Motor Commands.

(A) Top row, green trace: representative microstimulation-evoked tail movements measured by laser displacement sensor prior to paralysis. Top row, dotted gray trace: Waveform envelope used as look-up table for delivering local ES. Middlerow: timing of EOD motor commands. Bottom row: A local ES, placed in the receptive field of the recorded cell, was varied in amplitude according to the timing of the command and the relative value of the look-up table waveform. (B) Smoothed spike rate histograms from an extracellular recording of an ELL principal cell illustrating typical response patterns as a function of time relative to microstimulation of the tectum before, (pre, top row), during (during, second row), and after (post, third row) pairing with a local ES (dotted gray line). Non-smoothed histograms shown in gray (1 ms bins). The left and right-hand columns correspond to the microstimulation sites that, before paralysis, evoked ipsilateral and contralateral tail movements, respectively. Microstimulation was delivered alternately to each site throughout the experiment. In the example shown here, site 1 was paired with an ES that excited the cell while site 2 was paired with an ES that inhibited the cell. The difference traces in the bottom panels show the effects of pairing (after pairing minus before pairing). (C) Average of difference traces pooled across cells showing site-specificity of pairing at site 1 and site 2. Cells were pooled to match polarity of difference trace irrespective of site or cell type. Each cell is normalized to its pre-pairing baseline firing rate. In all panels, gray outlines indicate SEM across cells and gray boxes obscure microstimulation artifacts. See also Figure S4.

Finally, the capacity of ELL neurons to form negative images appeared to be highly flexible and robust. Negative image magnitude did not depend strongly on whether the effect of the paired ES on the recorded neuron was the same or opposite to that which would be caused by the evoked movement under natural conditions, on cell type, or on whether the effect of the paired ES was excitatory or inhibitory (Figure S4).

Negative Images Based on Corollary Discharge and Proprioception Require Spinal Input

Are negative images based on corollary discharge signals still formed under more natural conditions in which proprioceptive information related to movements is also available? To address this question, we induced negative images by pairing under conditions in which both corollary discharge and proprioception were activated. Tectal stimulation was delivered as before but paired with a passive displacement of the tail that matched the onset relative to tectal stimulation and time-course of the evoked movement measured prior to paralysis. Hence, in these experiments, the ES was, in principle, predictable based on both corollary discharge signals and proprioceptive signals. Comparing ELL principal cell responses to tectal stimulation alone or passive tail displacement alone before such pairings revealed that negative images were formed based on both corollary discharge and proprioceptive signals (Figures 7A and 7B; n=6). These results demonstrate that motor corollary discharge signals are still used even when sensory information related to movements is available, as would normally be the case.

Though our mossy fiber recordings suggest a spinal origin for corollary discharge signals, we wished to directly test whether negative images depended on ascending spinal input. In the same set of experiments described above, we injected a small volume (~1 uL) of lidocaine into the spinal cord after inducing negative images. Injections were done in a separate surgical site several centimeters from the recording site, and effectiveness was judged by disappearance of the electromotor command signal. We found that lidocaine completely abolished negative images (Figures 7A and 7B; n=6), without a significant change in baseline firing rate (pre-lidocaine: 14.66+8.47 Hz; post-lidocaine: 16.46+14.47 Hz, n=6, p=0.688, sign test). These results suggest that under natural conditions negative images are formed based on both corollary discharge and proprioceptive signals conveyed to ELL via the spinal cord.

DISCUSSION

Here we use an advantageous model system to demonstrate that a spinal corollary discharge pathway is used to form flexible and highly-specific negative images of the sensory consequences of motor commands at the level of individual neurons. Though the capacity to generate learned predictions about the sensory consequences of movements based on plastic corollary discharge is likely critical for sensory, motor, and cognitive functions in many species, neural correlates for such predictions, as shown here, have not been well characterized in other systems.

The first major finding of the present study is that, in addition to previously described EOCD signals related to highly-specialized electromotor behavior(Bell et al., 1983;Bell et al., 1992;Kennedy et al., 2014), ELL also receives corollary discharge signals related to movements. In contrast to EOCD signals, which merely relay the time of occurrence of the EOD, we show that movement-related corollary discharge signals convey graded information about the parameters of different types of movements (frequency in the context of rhythmic swimming and movement vigor and direction in the context of movements evoked by tectal microstimulation). The presence of varied and graded corollary discharge signals is consistent with the possibility, suggested by previous models of ELL function, that information relayed via mossy fibers and granule cells acts as a basis for generating negative images of the sensory consequences of movements via anti-Hebbian plasticity(Bell, 1981;Bell et al., 1997b;Roberts and Bell, 2000). Several lines of evidence presented here, together with previous anatomical studies(Bell et al., 1981;Szabo et al., 1990;Szabo et al., 1979), strongly suggest that corollary discharge inputs to ELL related to tail and trunk movements originate largely, if not exclusively, from the spinal cord. Spinocerebellar pathways conveying motor signals have been extensively studied in mammals(Arshavsky et al., 1978;Fedirchuk et al., 2013;Hantman and Jessell, 2010;Jankowska et al., 2011;Oscarsson, 1965;Spanne and Jorntell, 2013). The possible utility of such an ascending spinal corollary discharge in relation to negative image formation in ELL will be discussed below. Though not studied here, movements of the flexible chin appendage may also be associated with a corollary discharge(Engelmann et al., 2009). If such signals exist they would be expected to be relayed via a separate brainstem pathway(Bell et al., 1981;Maler et al., 1973;Szabo et al., 1979).

Though previous studies of cerebellum-like structures have provided evidence for predictions based on corollary discharge signals, these accounts have been limited to simple, highly-stereotyped behaviors, i.e. ventilation in elasmobranch fish(Bodznick et al., 1999) and the EOD in weakly electric mormyrid fish(Bell, 1981). Whereas the EOD motor command is simple and completely stereotyped, movement motor commands are obviously more complex and diverse. Hence, a key question is whether mechanisms described previously for predicting effects of the EOD, i.e. anti-Hebbian plasticity acting on corollary discharge inputs to principal cells, are sufficient for predicting the much greater variety of sensory patterns generated by movements. Two observations suggest that the capacity for forming and storing negative images related to movement motor commands exceeds that described previously in the context of electromotor behavior and may indeed be sufficient for predicting sensory consequences of movements. First, ELL neurons form negative images based on either MLR or tectal microstimulation; i.e., predictions are formed for different movements initiated by different brain structures. Such a capacity differs from the electromotor system in which the same motor command is initiated in a stereotyped fashion from a single brain structure. Second, individual ELL neurons are capable of simultaneously generating and storing two different negative images related to microstimulation of two distinct sites in the tectum. Though technical limitations prevented us from probing this capacity further, given the diversity of graded motor signals observed in mossy fibers together with the fact that each MG cells receives ~20,000 parallel fiber inputs and that ~30 MG cells converge onto each output cell(Bell et al., 2005;Meek et al., 1996), we expect that many more negative images could be stored.

Our results imply that ELL circuitry solves the complex problem of transforming copies of movement motor commands into a format appropriate to cancel their sensory consequences. Whereas bursts in pauses in mossy fibers evoked by tectal stimulation were stereotyped and brief, negative images in ELL neurons accurately match the temporal profiles of the fictive movements, which were substantially delayed relative to mossy fiber responses. Cancelling the effects of the fish’s own EOD poses a similar problem: EOD motor command signals conveyed by mossy fibers are much briefer in duration than the effects of the EOD on passive electroreceptors(Bell and Russell, 1978). EGp circuitry solves this problem by transforming stereotyped and minimally delayed EOD motor command signals conveyed by mossy fibers into granule cell responses that are more delayed and diverse(Kennedy et al., 2014). Such granule cell responses provide a basis for sculpting temporally-specific negative images via anti-Hebbian plasticity at parallel fiber synapses onto ELL neurons. A class of excitatory interneuron, the unipolar brush cell, appears to play a key role in generating delayed responses in granule cells. The accessibility of granule cells to in vivo recordings will allow us to test whether similar mechanisms could account for the ability to predict movement consequences that are extended in time relative to motor commands.

The need to transform motor signals into a format appropriate to predict sensory input may also provide a rationale for an ascending spinal corollary discharge. Previous studies in other fish species have suggested that tonic locomotor drive from the MLR is transformed into phasic motor commands for swimming within spinal circuitry(Deliagina et al., 2002;Kyriakatos et al., 2011;Uematsu et al., 2007). Given that opposite tail movements will typically have opposite electrosensory consequences, a phasic signal returning from the spinal cord would be expected to provide a better basis for negative image formation than a tonic signal from the MLR itself. Since all tail and trunk commands are ultimately issued via the spinal cord, this suggests the possibility that, in this system at least, ascending spinal corollary discharge pathways may be sufficient for predicting the sensory consequences of a wide range of movements. These results may have implications for the functions of spinocerebellar pathways in mammals. Though it is well-established that mammalian spinocerebellar pathways convey motor signals(Arshavsky et al., 1978;Fedirchuk et al., 2013;Hantman and Jessell, 2010;Jankowska et al., 2011;Oscarsson, 1965;Spanne and Jorntell, 2013), roles for such pathways in predicting sensory consequences of motor commands have, to the best of our knowledge, not been clearly defined.

Real movements activate proprioception which provides an additional source of information that might be sufficient to cancel self-generated electrosensory input in the absence of corollary discharge (Bastian, 1995;Bastian et al., 2004;Bell et al., 1992;Sawtell, 2010;Sawtell and Williams, 2008). Our results show that both corollary discharge and proprioceptive signals are used under conditions that simulate real movements. An interesting question for future studies is how corollary discharge and proprioceptive feedback interact at the levels of mossy fibers, granule cells, and ELL principal cells, under natural conditions in which both signals are available.

Converging lines of evidence from theoretical(Anderson et al., 2012;Wolpert et al., 1998), human behavioral(Bastian, 2006;Izawa et al., 2012), and electrophysiological investigations(Brooks and Cullen, 2013;Ebner and Pasalar, 2008;Pasalar et al., 2006) suggest that the mammalian cerebellum is involved in generating internal models that predict the sensory consequences of motor commands. Internal models may have a variety of functions, from cancelling effects of self-generated sensory inputs(Angelaki and Cullen, 2008;Cullen, 2004) to online correction of rapid movements(Miall and Wolpert, 1996). Though the existence of such internal models is widely accepted, how they are implemented in cerebellar circuitry remains largely unknown. Established roles for granule cells and parallel fiber plasticity in generating negative images in ELL(Bell et al., 1997b;Bell, 1981;Bol et al., 2011;Kennedy et al., 2014;Roberts and Bell, 2000) closely resemble those proposed by leading theories of mammalian cerebellar function(Albus, 1971;Dean et al., 2010;Fujita, 1982;Marr, 1969;Medina et al., 2000). In light of this correspondence, mechanisms for predicting sensory consequences of movements revealed here for the cerebellum-like circuitry of the mormyrid ELL may be expected to closely resemble those at work in the cerebellum itself.

EXPERIMENTAL PROCEDURES

Experimental Preparation

All experiments performed in this study adhere to the American Physiological Society’s Guiding Principles in the Care and Use of Animals and were approved by the Institutional Animal Care and Use Committee of Columbia University. Approximately 60 mormyrid fish (7–14 cm in length) of the species Gnathonemus petersii were used in these experiments. Surgical procedures to expose EGp for recording were similar to those described previously(Sawtell, 2010). In a subset of experiments an additional anterior portion of the skull was removed to expose the optic tectum. The anesthetic (MS-222, 1:25,000) was then removed. To evoke tail movements, we targeted tungsten microelectrodes to either the MLR, the optic tectum, or, in some experiments, to both. Continuous (40 or 100 Hz) microstimulation (50–100 uA) of the MLR evoked slow slow rhythmic (1–6 hz) swimming movements. Brief, high-frequency (10–15 pulses at 500 Hz) microstimulation of either the MLR or the tectum evoked rapid, isolated tail movements. When we microstimulated at two sites within the tectum, the anterior site evoked ipsilateral movements while the posterior site evoked contralateral movements, consistent with previous reports in goldfish(Herrero et al., 1998). The fish rarely moved outside of microstimulation protocols. A laser displacement sensor (LK-503, Keyence Corporation, Woodcliff Lake, NJ) measured tail displacement from the midline (spatial precision: 50 μm; measurement delay: 2 ms). After tail movements were measured, gallamine triethiodide (Flaxedil) was given (~20 ug/cm of body length) to paralyze the fish. Paralysis blocks the effect of motor neurons, including the electric organ, which prevents the EOD. The motor command signal that would normally elicit an EOD continues to be generated by the fish at a variable rate of 2 to 5 Hz. The EOD motor command can be measured precisely (see below). In a subset of experiments we records from spinal nerves to observe the motor command that would normally elicit swimming in an unparalyzed fish (see below). This preparation allows us to observe the central effects of movement-related corollary discharge in isolation from electrosensory or somatosensory effects.

Electrophysiology

The EOD motor command signal was recorded with an electrode placed over the electric organ. Spinal nerve recordings were performed as described previously for goldfish(Fetcho and Svoboda, 1993). Briefly, the spinal nerves were exposed at a point rostral to the tail, but in the caudal half of the fish. A fire-polished, glass suction electrode was used to record extracellularly from the dorsal ramus of the ventral root, a nerve that innervates epaxial white musculature. Signals from the recording electrode were filtered (100 Hz high pass and 300 Hz low pass) and amplified (Warner Instruments, Hamden, CT, Model DP-311). Root burst frequency showed a clear dependence on current intensity at low current amplitudes (50–100uA).

Extracellular recordings from mossy fibers were made with glass microelectrodes filled with 2M NaCl (40–100 MOhm). Criteria for distinguishing mossy fiber recordings from other EGp units were the same as those described previously(Bell et al., 1992;Sawtell, 2010). Extracellular recordings from the medial zone of ELL were made with glass microelectrodes filled with 2M NaCl (8–10 MOhm). ELL cells can be broadly classified as E- or I-cells: E-cells are excited by an increase in local EOD amplitude in the center of their receptive fields, and I-cells are inhibited by such a stimulus (Bell et al., 1997a;Bell and Grant, 1992;Mohr et al., 2003). We recorded from I-cells located in or just above the ganglion layer, which likely included both interneurons (MG1 cells), and efferent neurons known as large ganglion (LG) cells. We also recorded from E-cells located below the ganglion layer, which were probably efferent neurons known as large fusiform (LF) cells.

Whole cell recordings from MG cells in ELL were made using methods described previously(Sawtell, 2010). Electrodes (9–12 MOhm) were filled with an internal solution containing K-gluconate (122 mM), KCl (7 mM), HEPES (10 mM), Na₂ATP (0.5 mM), MgATP (2 mM), EGTA (0.5 mM), and 0.5% biocytin (pH 7.2, 280–290 mOsm). No correction was made for liquid junction potentials. Only cells with stable membrane potentials more hyperpolarized than −50 mV and access resistance < 100 MOhm were analyzed. All experiments were performed without holding current, unless otherwise noted. Membrane potentials were filtered at 3–10 kHz and digitized at 20 kHz (CED Power1401 hardware and Spike2 software; Cambridge Electronics Design, Cambridge, UK).

Dendritic Spike Pairing Experiments

Dendritic spike pairing experiments were conducted using intracellular recordings from MG cells using methods described previously(Bell et al., 1993;Sawtell, 2010;Sawtell et al., 2007). In these experiments, we paired a brief intracellular current injection to evoke a single dendritic spike (12–15 ms; 100–600 pA) at a fixed delay to one microstimulation site that evoked tail movements prior to paralysis. The neural response to microstimulation alone was compared immediately before and after the pairing period (1–2 minutes of data were used for analysis). Cells in which resting membrane potential, access resistance, or spike height changed substantially over the course of the experiment were excluded from the analysis.

Electrosensory Stimulus (ES) Pairing Experiments

ES pairing experiments were conducted using extracellular recordings from ELL principal cells. Cells that did not show plasticity (~15% of all recorded cells) under any condition were excluded from analysis. This is not unexpected, as there are known non-plastic cell types in ELL(Mohr et al., 2003).

Electrosensory responses were evoked by simultaneous global stimulation of the entire fish and local stimulation restricted to small area of the skin. Global stimuli were delivered by passing current between a small chloride silver ball inserted through the mouth in to the stomach of the fish and a second electrode placed in the water near the tail of the fish in an outside-positive configuration. The ES referred to in the paper is the modulation of the local field. Local stimuli were delivered with a bipolar stimulating electrode consisting of two small Ag-AgCl balls 5 mm apart. The electrode was held perpendicular to the skin at a distance of ~2 mm. For both global and local stimuli, brief pulses of current were delivered 4.5 ms after EOD command through the electrodes to activate electroreceptors. Absolute current strength for local stimuli ranged from 5uA-10uA while current strength for global stimuli ranged from 200–400mA. These values were chosen such that the amplitude of electrosensory-evoked field potentials could be both increased and decreased by the local stimulus, roughly mimicking the changes in EOD-evoked field potentials measured in response to tail movements in a previous study in which the natural EOD was left intact(Sawtell and Williams, 2008). Small adjustments to the local amplitude were made on a cell-by-cell basis to strongly excite or inhibit the cell with minimal current. In a subset of experiments, we controlled EOD motor command rate by lowering a concentric bipolar stimulating electrode (FHC, Bowdoin, ME) into the brain along the midline in or near the axons of the precommand nucleus, which course close to the midline along the ventral surface of the brainstem to the command nucleus (~ 5mm depth). Brief, single pulses of 0.2 ms reliably evoked an EOD motor command at low current strengths (10–20 uA). We typically microstimulated the EOD command at ~13 Hz. In experiments in which spinal inactivation was performed, electrosensory stimuli were delivered at a fixed delay from the microstimulation pulse as the EOD motor command could not be measured.

In ES pairing experiments, we varied the amplitude of the local ES in the center of the recorded cell’s receptive field to deliver a time-varying pattern of electrosensory stimulation based on the waveform of the tail movement recorded prior to paralysis. Since amplitude in local EOD amplitude is proportional to tail displacement for small angles, such a protocol approximates the electrosensory consequences induced by real tail movements. Microstimulation was always separated by at least 1.25 seconds. Pairing was conducted for 10–15 minutes. The neural response to microstimulation alone was compared immediately before and after the pairing period.

For the experiments conducted in Figure 7, we only microstimulated the anterior tectal site that evoked ipsilateral movements. Tail displacement was measured by a laser and then that signal was fed back into the servomotor to deliver passive tail movements that mimicked the microstimulation-evoked movement prior to paralysis. The fish’s tail was lightly held between two glass rods positions posterior to the electric organ. The rods were held by a manipulator mounted to a computer-controlled servomotor (Pacific Laser Equipment, Santa Ana, CA). A partition was placed between the tail and the rest of the fish to prevent water waves from activating lateral line receptors.

In these experiments, we examined neural responses before and after pairing under two conditions: (1) fictive, in which we delivered microstimulation alone; (2) passive, in which we delivered the tail movement alone. To simulate real movements, pairing was conducted under conditions in which we delivered both the microstimulation and the tail movement simultaneously. Prior to recording, a pipette containing a solution of 2% lidocaine HCl (Sigma-Aldrich, St. Louis, MO) was lowered into the spinal cord at a separate surgical site several centimeters from the recording site. After inducing plasticity, a small volume of lidocaine solution (~1 uL) was injected by manual pressure into the spinal cord. The neural response was recorded continuously and data collected starting at approximately 2 min following injection. Spinal inactivation was confirmed by the disappearance of the EOD motor command.

Data Analysis and Statistics

Data analysis was performed offline in MATLAB (MathWorks, Natick, MA) and Spike2 (Cambridge Electronic Design). Data are expressed as mean ± standard deviation (SD), unless otherwise noted. Tests for statistical significance are noted in the text. Differences were judged to be significant at p<0.05.

Power spectral density functions were constructed in Spike2 by first smoothing spikes with a small Gaussian window (20 ms), then using a Fast Fourier Transform (FFT) to convert the waveform data into a power spectrum, implemented with a Hanning window.