Responses of cerebellar Purkinje cells during fictive optomotor behavior in larval zebrafish

Although the cerebellum is one of the simplest and most highly ordered circuits in the vertebrate brain, links between its structure and function remain elusive. Zebrafish may be an ideal model system for making such links because of the accessibility of their brains to optical imaging and manipulations of neural activity. Our study provides one of the first detailed electrophysiological descriptions in zebrafish of the responses of identified cerebellar neurons during behavior.

Abstract

Although most studies of the cerebellum have been conducted in mammals, cerebellar circuitry is highly conserved across vertebrates, suggesting that studies of simpler systems may be useful for understanding cerebellar function. The larval zebrafish is particularly promising in this regard because of its accessibility to optical monitoring and manipulations of neural activity. Although several studies suggest that the cerebellum plays a role in behavior at larval stages, little is known about the signals conveyed by particular classes of cerebellar neurons. Here we use electrophysiological recordings to characterize subthreshold, simple spike, and climbing fiber responses in larval zebrafish Purkinje cells in the context of the fictive optomotor response (OMR)—a paradigm in which fish adjust motor output to stabilize their virtual position relative to a visual stimulus. Although visual responses were prominent in Purkinje cells, they lacked the direction or velocity sensitivity that would be expected for controlling the OMR. On the other hand, Purkinje cells exhibited strong responses during fictive swim bouts. Temporal characteristics of these responses are suggestive of a general role for the larval zebrafish cerebellum in controlling swimming. Climbing fibers encoded both visual and motor signals but did not appear to encode signals that could be used to adjust OMR gain, such as retinal slip. Finally, the observation of diverse relationships between simple spikes and climbing fiber responses in individual Purkinje cells highlights the importance of distinguishing between these two types of activity in calcium imaging experiments.

NEW & NOTEWORTHY

Although the cerebellum is one of the simplest and most highly ordered circuits in the vertebrate brain, links between its structure and function remain elusive. Zebrafish may be an ideal model system for making such links because of the accessibility of their brains to optical imaging and manipulations of neural activity. Our study provides one of the first detailed electrophysiological descriptions in zebrafish of the responses of identified cerebellar neurons during behavior.

despite decades of intensive investigation, links between the highly ordered and relatively simple circuitry of the cerebellum and its function remain elusive. Although the majority of work on the cerebellum has focused on mammals, core features of cerebellar circuitry are conserved across vertebrate phylogeny (Finger 1983; Hibi and Shimizu 2012; Larsell 1967; Meek 1992; Nieuwenhuys 1967). Hence studies of simpler vertebrates may provide a useful perspective on cerebellar function. The larval zebrafish is particularly promising as a model organism for studying the cerebellum. First, the number of cerebellar neurons is far smaller than in other systems in which the cerebellum has traditionally been studied. There are roughly 300 Purkinje cells in the 7 days postfertilization (dpf) larval zebrafish cerebellum (Hamling et al. 2015) compared with 1–2 million in adult cat (Mwamengele et al. 1993; Palkovits et al. 1971) and roughly 100,000 in adult mice (Herrup and Trenkner 1987). These small numbers together with the optical transparency of the larval zebrafish offer the potential to monitor the activity of all of the neurons in the cerebellum (together with activity in other brain regions) simultaneously during behavior (Ahrens et al. 2012, 2013). This unique potential for large-scale activity monitoring along with rapidly emerging technologies for mapping circuits and manipulating genetically identified cell types make the larval zebrafish a uniquely attractive model organism for cerebellar studies (Okamoto 2014).

Core features of cerebellar circuitry are shared between mammals and larval zebrafish, including the presence in both of mossy fibers, granule cells, parallel fibers, Purkinje cells, climbing fibers, molecular layer interneurons, and Golgi cells (Bae et al. 2009; Takeuchi et al. 2015). A difference between the cerebellum in mammals and teleost fish, including zebrafish, is the location of the glutamatergic neurons that receive input from Purkinje cells and project to brain regions outside the cerebellum. Whereas in mammals such neurons are located in separate deep cerebellar or vestibular nuclei, in fish the large majority are located adjacent to Purkinje cells (Bae et al. 2009; Finger 1978; Heap et al. 2013). This proximity could, in fact, be an advantage for understanding how Purkinje cells shape cerebellar output—a question that has been extremely difficult to address in mammals.

Several lines of evidence suggest that the cerebellum is functional at larval stages. Developmental studies have shown that by 5 dpf Purkinje cell and granule cell layers have formed and that the two major input pathways to Purkinje cells—the mossy fiber-granule cell-parallel fiber pathway and the olivocerebellar climbing fiber pathway—are in place (Bae et al. 2009; Takeuchi et al. 2015). Electrophysiological studies have shown that larval zebrafish Purkinje cells exhibit both simple spikes and climbing fiber responses (CFRs), with firing patterns that change relatively little after 6 dpf (Hsieh et al. 2014; Sengupta and Thirumalai 2015). Optogenetic activation or silencing of larval Purkinje cells alters swimming movements during the optomotor response (OMR) (Matsui et al. 2014). Finally, it has been reported that lesioning of the olivocerebellar pathway prevents motor adaptation in a closed-loop, fictive OMR paradigm (Ahrens et al. 2012) and that lesioning of the cerebellum impairs classical conditioning (Aizenberg and Schuman 2011).

Although calcium imaging studies have revealed that cerebellar neurons are active during the OMR (Ahrens et al. 2012; Matsui et al. 2014), the nature of the signals they convey remains unclear. For example, the limited temporal resolution of calcium imaging has not allowed for a detailed analysis of how cerebellar activity relates to the structure of swimming behavior in larval zebrafish, which consists of rapid tail beats organized into discrete bouts. Moreover, calcium responses in Purkinje cells could be due to simple spikes, CFRs, or some combination of both. More information about the basic response properties of identified cerebellar cell types during behavior is needed before specific hypotheses regarding the function(s) of the zebrafish cerebellum can be formulated and tested.

The goal of this study was to begin to provide such information by recording from Purkinje cells in the context of the fictive OMR. We observed a variety of visual and motor-related responses during the fictive OMR at the level of Purkinje cell membrane potential, simple spike firing, and CFRs. Our results place some constraints on the possible role of the cerebellum in guiding the OMR and suggest specific directions for future investigations, including the need to better understand the nature of the signals conveyed by climbing fibers.

METHODS

All experiments performed in this study were approved by the Columbia University Institutional Animal Care and Use Committee. Most experiments were conducted in transgenic aldoca:gap43-Venus fish to allow visualization of Purkinje cells (Takeuchi et al. 2015; Tanabe et al. 2010) or in Nacre or Casper strains to facilitate visualized recordings and imaging.

Experimental preparation.

Six to ten days postfertilization (dpf) larva were anesthetized with 0.01% MS222 and then embedded in a small block of low-gelling-temperature agarose (Sigma-Aldrich no. A0701), which was subsequently glued to the glass bottom of a slice recording chamber. Agar was removed from above the head and adjacent to the right side of the trunk, from muscle segments 7–24, to allow placement of the neural and motor recording electrodes. Fish were paralyzed with 1 mg/ml α-bungarotoxin (Tocris) applied for 1 min locally to the exposed portion of the trunk, where a small nick in the skin around muscle segment 23 facilitated paralysis. The skin over the cerebellum was gently removed with a bent tungsten dissecting needle (Roboz Surgical Instrument no. RS-6063). During the experiment, the recording chamber was continuously perfused with aerated Evans solution containing (in mM) 134 NaCl, 2.9 KCl, 2.1 CaCl₂, 1.2 MgCl₂, and 10 HEPES (pH 7.8, 280–290 mosM).

Visual stimuli.

Visual stimuli were presented on a screen ∼1 cm beneath the fish. In luminance experiments (see Figs. 5, 9, and 10), movies showed alternating 3- to 6-s presentations of all- black and all-white screens. In the OMR open-loop experiments (see Figs. 4, 8, 9, and 10), movies showed a square-wave grating with spatial period of 20 mm that moved alternately in 10- to 25-s periods of “OMR-inducing drift” (tail-to-head motion) and “OMR-suppressing drift” (head-to-tail motion), with a 5-s period of no drift between each drift period. Each trial consisted of four rounds of this alternating drift with increasing speed of 0.4 cm/s to 1.2 cm/s. There were two to six trials per cell.

Fig. 5.

Luminance modulation of membrane potential. A: average membrane potential and simple spike luminance responses of cell with graded oscillation frequency response shown in Fig. 4D (n = 20 trials). Cell exhibits transient response to changes in luminance (both onset and offset). B: example cell with membrane potential depolarization and increased simple spikes in response to dark (n = 17 trials). C: example cell with membrane potential depolarization and increased simple spikes in response to light (n = 16 trials). Gray traces indicate 1 SE.

Fig. 9.

Visual modulations of CFR rate. A: example cells with transient increases in CFR rate at light offset. Top: time course of luminance states (black fill indicates light off). Bottom: raster and histogram of CFR rates. B: example cells with transient increases in CFR rate at onset of OMR-inducing or OMR-suppressing drift Top: time course of OMR-inducing and OMR-suppressing drift. Middle: raster and histogram of CFR rates for cell with transient increase in CFR rate in OMR-inducing direction only. Bottom: raster and histogram of CFR rates for cell with transient increase in CFR rate in OMR-suppressing direction only. C: timing of onset of OMR-suppressing drift-related increase in CFR rate (bottom) relative to time course of average bout recorded at motor nerve (top). Bouts are rapid compared with drift-related increase in CFR rate, consistent with the lack of observed CFR modulation during transient drift in the OMR-suppressing direction during playback of visual stimulus driven by bouts in closed loop (Fig. 8C).

Fig. 10.

Diverse relationships between CFR and simple spike (SS) responses in individual Purkinje cells. A: examples of diverse CFR and SS relationship during motor bouts. Left: cell 1 has opposite SS and CFR response polarity. Right: cell 2 has matching SS and CFR response polarity. B: example of diverse CFR and SS relationship during sustained grating drift. Top: SS rate is increased for drift in either direction. Bottom: CFR rate is only transiently modulated and only for 1 direction of drift. C: examples of diverse CFR and SS response relationship during luminance changes. Left: cell 1 has no SS response but strong CFR response to light off. Right: cell 2 has strong SS as well as CFR response to light off.

Fig. 4.

Visual modulation of membrane potential. A: example membrane potential traces (middle) at various speeds during 3 drift states (top): OMR inducing (negative velocities), no drift, and OMR suppressing (positive velocities). Drifting in either direction and at all speeds strongly drives activity compared with periods of no drift. This sustained depolarization could not be explained by motor-related activity (bottom: motor bout onsets) as fish swim only sporadically when presented with an open-loop OMR-inducing stimulus and rarely if ever during an OMR-suppressing stimulus. B: summary of intracellular membrane potential responses to graded OMR-inducing drift velocities. For each drift velocity, the average membrane potential (collected from a 100-ms window before every motor bout that occurred at that speed to minimize effects on our measurements of any motor-related activity) was compared to the average membrane potential at the −1.6 cm/s drift velocity. Only 3 cells recorded intracellularly (solid lines) and 1 cell recorded extracellularly (data not shown) exhibited even a moderate grading in response (3/24 intracellular, 1/5 extracellular, R > 0.3, P < 0.05) C: summary of intracellular membrane potential responses to OMR-inducing and -suppressing drift (irrespective of speed) compared with no drift. Many cells were significantly modulated by drift in both directions (solid lines) [13/24 intracellular, 1/5 extracellular (data not shown)] (1-way ANOVA P < 0.05, post hoc Tukey test P < 0.05), suggesting that the modulation is not simply an artifact caused by responses to the motor activity during OMR-inducing drift. D: oscillations in membrane potential (smoothed) of cell from A at different OMR-inducing drift speeds. Similar oscillations were also apparent in the OMR-suppressing direction (data not shown). Frequency of oscillation scales with magnitude of drift velocity. This would be consistent with activity driven by the luminance changes caused by the alternating black and white bars of the OMR stimulus.

Fig. 8.

Bout-related modulations of CFR rate. A: examples of bout-related modulations of CFR rate. Top: processed motor nerve recording. Bottom: CFR histogram triggered on bout onset recorded at motor nerve. B: probability of a CFR during bouts across graded OMR-inducing drift velocities. Drift velocity did not substantially affect CFR probability during bouts in most cells with CFR bout responses (9/11 cells, binary logistic regression, P < 0.05) and in the remaining cells (bold traces) explained <15% of the variance in CFR firing across bouts (Nagelkerke R²). C: comparison of CFR rate during bouts in closed loop (CL), motor bouts in playback (PB), and the visual consequences of bouts replayed during PB trials. Cell exhibits CFR modulation during bouts in CL (shaded histogram) and, in PB, shows similar modulation to motor bouts (solid line, top) but not to the bout-driven visual consequences replayed from CL (solid line, bottom).

Closed loop/playback experiments.

In the OMR closed loop/playback experiments (see Figs. 3 and 8), we based our methods on a published fictive swimming paradigm (Ahrens et al. 2012). Fish were again presented with trials containing four rounds of OMR-inducing and OMR-suppressing drift; however, this time the baseline drift was combined with a “virtual swimming” drift component, driven by the fish's recorded motor nerve output. The magnitude of the underlying drift in these experiments was kept constant at 1 cm/s and alternated directions. The virtual swimming drift component was always in the head-to-tail direction (i.e., the direction of visual drift that results from forward swimming), defined here as positive, and was added linearly to the underlying drift. The magnitude of the virtual swimming drift component was calculated based on the recorded motor nerve signal. Fish swim in discrete units of swimming called bouts, which are apparent in the recorded motor signal as transient increases in voltage variance. The motor signal was processed by first taking the standard deviation of the raw motor trace over a sliding window of 10 ms. During a bout (detected automatically when the processed motor signal crossed a baseline threshold set manually during periods of no swimming) the swim-related component of the grating's velocity, v_s, was calculated as the average of the processed motor signal since the last update, m_av, minus the baseline threshold, multiplied by a constant of proportionality, k, such that v_s = k·(m_av − b). The constant of proportionality was set experimentally for each fish such that swim bouts were roughly able to transiently stabilize the grating during OMR-inducing periods. After each bout, the swim-related component decayed back to zero at a rate of −15 cm/s². The total grating velocity experienced by the fish was equal to this swim velocity plus the baseline drift (±1 cm/s or 0 cm/s). This total grating velocity was updated on average at >200 Hz and was smoothed at each update so that (effective drift velocity) = α·(calculated velocity) + (1 − α)·(previous effective drift velocity), with α = 0.3. The change in grating position at each update was equal to the total grating velocity multiplied by the time since the last update.

Fig. 3.

Visual and motor activity during closed-loop OMR behavior. A: example cells with responses to motor bouts during closed-loop (CL) stimulus. In CL trials, fictive backward displacement (simulated by a grating drifting in the tail-to-head direction) can be transiently stabilized by the fish's motor nerve output. Under these conditions membrane potential responses during bouts (bottom) could be related to either motor nerve activity (middle) or its visual consequences (top). B: responses during playback (PB). In PB trials the visual stimulus that was generated by motor nerve activity during CL is played back independent of ongoing motor nerve activity. Therefore, motor bouts executed in PB (middle) are not yoked to the played-back visual consequences of bouts executed in CL (top), allowing us to separately probe the motor- and visual-related components of neural activity (bottom). C: comparison of the average membrane potential (±1 SE) during bouts in CL with the average membrane potential during motor bouts in PB (left) and with the average membrane potential during replay of the visual consequences of bouts initiated in CL trials (right). Both cells have modulations during bouts in CL and in PB show similar modulation to motor bouts (left) but not to the bout-driven visual consequences replayed from CL (right). Although the motor bout-driven responses were similar in shape between CL and PB, they could have somewhat different amplitudes (cell 1, solid line, left).

Each of these closed-loop trials generated a unique visual stimulus movie resulting from the combination of the preset underlying drift and the fish-controlled virtual swim component. After a closed-loop trial, a playback period was initiated in which this same visual stimulus movie was played again, now entirely unyoked from the fish's motor output, in order to be able to dissect the visual and motor components of any activity modulations seen in closed loop.

Electrophysiology.

Motor nerve recordings were made based on published methods (Ahrens et al. 2012; Masino and Fetcho 2005). Briefly, a glass microelectrode filled with Evans solution and beveled to lie flat against the fish's side was placed, with light suction, on a myotomal cleft between muscle segments 11 and 16.

Purkinje cells were targeted for cell-attached or whole cell recordings with Dodt contrast microscopy. Recordings were made with glass microelectrodes (8–17 MΩ) filled with internal solution for whole cell recordings (see below) or Evans solution for cell-attached recordings. Pipettes were wrapped in Parafilm to reduce capacitance. Internal solution contained (in mM) 122 K-gluconate, 7 KCl, 10 HEPES, 0.4 Na₂GTP, 4 MgATP, 0.5 EGTA, and 0.05 Alexa 594 (pH 7.2, 280–290 mosM). Motor nerve and brain recordings were digitized at 40 kHz and 20 kHz, respectively (CED Micro1401-3 hardware and Spike2 software; Cambridge Electronics Design, Cambridge, UK).

Recordings were included for analysis only if two distinct types of all-or-none events were clearly evident (see results), fish exhibited fictive swim bouts in response to OMR-inducing (i.e., tail to head) grating drift, and, for whole cell recordings, the resting membrane potential was less than −45 mV. For whole cell recordings the average seal resistance was 2.2 ± 1.1 GΩ and the average input resistance was 3.4 ± 2.3 GΩ. Resting membrane potentials ranged from −45.5 to −58.5 mV corrected for a calculated liquid junction potential of 15.5 mV. The average recording duration was 56 ± 23 min. Recordings were terminated if spike height changed abruptly. To facilitate analysis of subthreshold responses, data were typically collected with a small amount of hyperpolarizing current to hold the membrane potential at or below spike threshold. The average membrane potential during data collection was −65 ± 6 mV.

In a subset of cells, which were labeled with Alexa 594 in the recording pipette, morphology and location within the cerebellum were visualized on a two-photon microscope (Chameleon Ultra II Coherent laser at 850- and 920-nm wavelength and for data collection PrairieView software, Prairie Technologies). In these experiments, Purkinje cell identity was verified by visualization of Venus fluorescence around the soma of the recorded cell. In experiments in wild-type fish, Purkinje cell identity was established electrophysiologically by the presence of two distinct spike types (see results).

Data analysis and statistics.

Data were analyzed off-line with Spike2, MATLAB (MathWorks, Natick, MA), and SPSS (IBM, Armonk, NY).

A cell was determined to have a simple spike (extracellular) or membrane potential (intracellular recordings) motor bout response if the extreme in a response window 0–150 ms after bout onset was more than three standard deviations larger than the extreme in the same direction in a baseline window 75–225 ms before the bout onset (see Figs. 6 and 7). This criterion was used for assessing closed-loop and playback (motor and visual) responses as well (see Fig. 3). Student's t-tests were used to evaluate statistical significance (α = 0.01). Spike train data used in this analysis were first transformed into a smoothed waveform by convolving with a triangular kernel of width 50 ms. Onset of bout response was defined as the time point at which the average response crossed a threshold of 1 standard deviation above baseline (defined as the average response over the interval −300 ms to −50 ms before motor bout onset) and stayed there at least until the response extreme.

Fig. 6.

Bout-related increases in membrane potential and simple spike activity. A: heat map of normalized bout onset-triggered responses showing time course of increased membrane potential and simple spike responses (sorted by time of peak response). Black vertical bars, detected onset of response for each cell. Response onset preceded recorded motor onset in most cells. Open circles, detected peak of response for each cell; asterisks, extracellularly recorded cells. B: examples of bout response activity triggered off of termination of motor nerve bout activity. Top: average motor nerve traces processed with a 5-ms RMS sliding window. Bottom: average membrane potential (gray traces indicate 1 SE). Many bout responses extended well beyond the termination of motor nerve bout activity. C: 2 examples of tail-beat frequency substructure in bout responses. Top: average motor nerve traces processed with a 5-ms RMS sliding window, which is necessary for resolving bursts in multibout averages. Bursts correspond to individual tail beats. Bottom: simple spike histogram triggered on final burst in a bout. Simple spike frequency exhibits modulation at same frequency as motor bursts and continues beyond last recorded motor burst.

Fig. 7.

Bout-related decreases in membrane potential and simple spike activity. A: processed motor nerve recording (top) and membrane potential (bottom) triggered on final burst in a bout in 2 example cells. Membrane potential recordings (shown as average ± 1 SE) exhibit troughs coincident with bout terminations. B: motor bout duration (binned by number of bursts per individual bout) vs. average time of membrane potential trough from bout onset in the example cells from A. As bout duration increases, time to trough increases linearly in both cells.

When evaluating correlation of activity with speed of OMR-inducing drift, measurements of simple spike rates and membrane potential were taken from a 0.1-s period before each swim bout executed during OMR-inducing drift (see Fig. 4). This ensured that measurements were taken during the same conditions across all drift speeds and were relatively uncontaminated by motor responses. During OMR-suppressing drift, fish rarely swam. Therefore, when evaluating correlation with speed of OMR-suppressing drift, measurements of simple spike rates and membrane potential were taken at regular 2-s intervals for all drift speeds (see Fig. 4).

Assessments of simple spike rate or membrane potential during drift compared with no drift were based on a one-way ANOVA (P < 0.05) for each cell comparing the simple spike rate or membrane potential (average over a 0.1-s window) measured at 2-s intervals during drift periods (OMR inducing or OMR suppressing) to the same measurements during pauses in grating drift (see Fig. 4). To contend with the possibility that significant differences during OMR-inducing drift compared with pause periods could be due to differences in motor activity, a significant increase (based on a post hoc Tukey test P < 0.05) in activity during both OMR-inducing and OMR-suppressing drift was required for a cell to be considered to have a drift response.

Cells were evaluated for CFR motor bout responses by evaluating whether the extreme in the average response window 0–150 ms after bout onset was more than 3 standard deviations larger than the extreme in the same direction in the baseline window 75–225 ms before the bout onset (see Fig. 8). This criterion was used for assessing closed-loop and playback (motor and visual) responses as well (see Fig. 8). Student's t-tests were used to evaluate statistical significance (α = 0.01). Spike train data used in this analysis were first transformed into a smoothed waveform by convolving with a triangular kernel of width 150 ms.

The effect of OMR-inducing drift speed on identified CFR bout responses was evaluated by counting the number of bouts at each drift speed in which a CFR did or did not occur (see Fig. 8). Binary logistic regression was then used to assess whether the likelihood of a CFR occurring during a bout was affected by drift speed.

A cell was determined to have a transient CFR drift response based on a one-way ANOVA comparing the peak response in a 1-s window after drift (OMR inducing or OMR suppressing) to the peak response in a 1-s window after pause onset. Cells with responses that were significantly different (based on a post hoc Tukey test P < 0.01) for one or both drift directions, compared with the pause periods, were considered to have CFR responses (see Fig. 9).

Independence of CFR and membrane potential or simple spike signals was assessed for responses to luminance, drifting vs. non-drifting grating, and motor bouts. To maximize the likelihood of identifying a dependent relationship, for each cell the existence of a response was evaluated in a binary manner (either yes or no) regardless of whether that response could be further broken down into response subtypes (i.e., a significant membrane potential deflection during motor bouts was considered a motor response regardless of whether it was a hyperpolarized or depolarized deflection). We conducted Fisher's exact tests for independence between CFR and simple spike or membrane potential responses for each response type (P < 0.05).

RESULTS

We obtained visualized recordings (n = 28 whole cell; n = 10 cell attached) from Purkinje cells in the corpus cerebelli of 6–10 dpf zebrafish. Consistent with recently published recordings of larval zebrafish Purkinje cells (Hsieh et al. 2014; Sengupta and Thirumalai 2015), Purkinje cell recordings exhibited two types of all-or-none events that differed in their waveforms, rate of occurrence, and dependence on the underlying membrane potential (Fig. 1). The smaller, more frequent events likely correspond to simple spikes and the larger, infrequent events to responses to climbing fiber input from the inferior olive, referred to here as CFRs (for cell-attached recordings event rates were 10.4 ± 7.7 Hz for simple spikes vs. 0.55 ± 0.50 Hz for CFRs; n = 10) (Fig. 1C). In whole cell recordings we observed that simple spike firing could be completely abolished by small hyperpolarizing current injections without affecting CFR rates (0.26 ± 0.51 Hz at rest vs. 0.25 ± 0.23 Hz with hyperpolarizing bias current; n = 23; P > 0.05, Student's t-test; Fig. 1A). This is expected because simple spikes are generated within the Purkinje cell while CFRs are due to powerful synaptic input from the inferior olive. The observation that CFR rates were unchanged at hyperpolarized membrane potentials also suggests that the large, infrequent events were not dendritic calcium spikes, which have been shown to be evoked by strong membrane potential depolarization in teleost Purkinje cells (Alvina and Sawtell 2014; Han and Bell 2003; Sengupta and Thirumalai 2015; Zhang and Han 2007). Though different from the complex spikes evoked by climbing fiber activation in mammalian Purkinje cells, the appearance of CFRs in our in vivo recordings is consistent with that reported previously for fish, including larval zebrafish (Hsieh et al. 2014; Sengupta and Thirumalai 2015) and mormyrid fish (Alvina and Sawtell 2014; de Ruiter et al. 2006; Han and Bell 2003; Zhang and Han 2007).

Fig. 1.

Electrophysiological properties of zebrafish Purkinje cells. A: intracellular trace from a Purkinje cell held at 2 different resting membrane potentials (left: −9 pA; right: −7 pA). Simple spike (SS) rate is strongly dependent on resting membrane potential, while climbing fiber response (CFR) rate is not (filled circles, example CFR events). B: average CFR (left) and SS (right) waveforms from recording in A (10 events each; gray traces indicate 1 SE) C: extracellular cell-attached recording from a Purkinje cell exhibiting spontaneous SSs and CFRs (filled circles, example CFR events). D: average CFR (left) and SS (right) waveforms from recording in C (10 events each; gray traces indicate 1 SE).

In a subset of recordings (n = 20) performed in transgenic fish (aldoca:gap43-Venus) in which the fluorescent protein Venus was selectively expressed in Purkinje cells (Tanabe et al. 2010), we were able to confirm Purkinje cell identity based on a halo of Venus fluorescence around the cell body (Fig. 2A, inset). An additional seven cells exhibited only a single type of spike and were not Venus positive (data not shown) and were thus not considered Purkinje cells. During recording, we filled cells with a fluorescent dye to visualize their morphology and position within the cerebellum. Purkinje cells had extensive dendritic arbors, which appeared to be densely studded with spines (Fig. 2B). In some cases a thinner beaded process was also visible, likely the Purkinje cell's axon. To map the relative location of Purkinje cells within the cerebellum across fish, we used the Venus expression in the Purkinje cell population to make a standard image of the shape and extent of the labeled Purkinje cell region and then used a point transformation to map individual recorded cells onto it (Fig. 2C). For cells whose locations could not be visualized with fluorescence (e.g., cells recorded extracellularly or not in transgenic aldoca:gap43-Venus fish), the location of the recording pipette tip was noted manually under Dodt visualization when possible (Fig. 2C, dotted outlines). A majority of Purkinje cells were recorded from the corpus cerebelli, though one or two cells may have been recorded at the edge of the valvula (Fig. 2C).

Fig. 2.

Morphological properties of zebrafish Purkinje cells A: Purkinje cell filled with Alexa 594, visualized against Venus-labeled Purkinje cell population. z Stack (inset) through center of nucleus shows Venus localized to cell membrane (scale bar, 5 μm). B: morphology of Purkinje cells from 6 dpf larvae filled with Alexa 594. C: locations of recorded cells mapped onto a standardized hemisphere of Venus-labeled cerebellum. Solid-outlined cells were mapped with a point transformation of each filled cell relative to Venus background. Dotted-outlined cells could not be fluorescence visualized for point transformation and were instead plotted based on manually noted coordinates of pipette tip under Dodt visualization.

Subthreshold and simple spike responses during closed-loop optomotor behavior.

A previous study using whole brain calcium imaging revealed responses in the cerebellum during a closed-loop fictive OMR paradigm in which motor commands related to swimming (monitored in paralyzed fish by recordings from trunk motor nerves) are used to control the motion of a visual display (Ahrens et al. 2012). Under these conditions, fish can transiently stabilize the position of a tail-to-head drifting grating by emitting swim commands. However, this study could not determine which cerebellar cell types were responsible for the calcium responses or whether calcium responses were due to simple spikes, CFRs, or both. Furthermore, the temporal resolution of whole brain calcium imaging in this study was not sufficient to relate activity to the detailed structure of larval zebrafish motor behavior, which is composed of rapid tail beats (∼30 Hz) organized into discrete bouts of swimming lasting on the order of a few hundred milliseconds. We took a complementary approach by recording subthreshold, simple spike, and CFR activity from individual Purkinje cells in the context of a similar closed-loop fictive OMR paradigm.

Purkinje cell activity during the OMR could relate to the visual stimulus, the fish's swim commands, or both. To differentiate between these possibilities we compared Purkinje cell activity in closed-loop conditions in which fictive swim bouts controlled the position of a grating stimulus (closed loop, Fig. 3A) to activity recorded when the same visual stimulus was played back independent of the fish's motor commands (playback, Fig. 3B). We focus initially on subthreshold and simple spike responses. CFR responses are discussed in another section below. Under closed-loop conditions, most Purkinje cells exhibited strong subthreshold (6 of 7, intracellular) and simple spike (3 of 4, extracellular) modulations with onsets similar to individual fictive swim bouts (Fig. 3A). Responses to bouts under playback conditions were similar to those observed under closed-loop conditions (motor triggered, Fig. 3C). On the other hand, not a single cell exhibited clear visual responses (i.e., membrane potential responses triggered on the onset of visual acceleration) during playback (visual triggered, Fig. 3C). These observations strongly suggest that the bout responses observed under closed-loop conditions were largely motor (as opposed to visual) responses.

In some cases, small differences in bout responses were observed between closed-loop and playback (Fig. 3C, cell 1). Such differences could reflect interactions between visual and motor signals, e.g., a component of Purkinje cell motor responses related to a mismatch between actual and expected visual consequences. However, given our limited readout (i.e., a motor nerve recording at a single site), we cannot rule out the possibility that differences in neural responses are due to differences in the fish's fictive swim behavior under closed-loop vs. playback conditions.

Subthreshold and simple spike responses to sustained visual motion.

OMR behavior allows larval zebrafish to maintain their position relative to a visual stimulus; however, such stabilization is not instantaneous but rather is achieved on a timescale substantially longer than that of individual swim bouts. Hence the presence of sustained visual motion might be a more relevant signal for controlling the OMR than instantaneous visual motion. To test whether such signals were present, we examined 29 Purkinje cells in the context of a simple open-loop visual stimulus in which drift velocity was held constant at different values. Specifically, we presented fish with alternate periods of tail-to-head (OMR inducing) and head-to-tail (OMR suppressing) drift. Drift periods were separated by periods in which the grating was stationary (Fig. 4A). There were four rounds of forward-backward drift pairs per trial, with drift speed increasing successively across rounds (0.4 cm/s, 0.8 cm/s, 1.2 cm/s, 1.4 cm/s), and two to six trials presented per cell. Consistent with previous observations, fictive swim bouts were most frequent at the onset of tail-to-head drift and were quite rare during pauses in drift or head-to-tail drift.

Surprisingly, we did not observe marked sensitivity to the velocity or direction of grating drift in Purkinje cells. Only 4 of 29 cells exhibited even a modest grading of membrane potential or simple spike responses to OMR-inducing drift velocity (3/24 intracellular, 1/5 extracellular, R > 0.3, P < 0.05) (Fig. 4, A and B) and even fewer (2/24 intracellular, 0/5 extracellular, R > 0.3, P < 0.05) to OMR-suppressing drift velocity (data not shown). In some cells, however, we noted a dramatic shift in the membrane potential at transitions between periods of drifting vs. stationary gratings (Fig. 4, A and C). When we evaluated this effect quantitatively, we found that 14 of 29 Purkinje cells exhibited sustained membrane potential depolarization and simple spike rate increases in response to drift in both directions compared with a stable grating (13/24 intracellular, 1/5 extracellular, 1-way ANOVA P < 0.05, post hoc Tukey test P < 0.05) (Fig. 4C). These responses were present in the OMR-inhibiting as well as -inducing directions of grating drift, indicating they were not a simple result of the motor responses driven by the OMR-inducing grating drift. In a few cells with particularly strong drift responses, we noted oscillations in the membrane potential that increased in frequency with increasing speed of visual motion (Fig. 4D). Given that the frequency of the oscillations was similar to the spatial frequency of the grating stimulus, such oscillations could simply be due to changes in luminance. Indeed, responses to luminance change have been reported previously in larval zebrafish Purkinje cells (Hsieh et al. 2014).

Luminance responses were tested directly with a full-field visual stimulus that alternated from black to white with a period of 3–5 s (Fig. 5). Eighteen of twenty-three Purkinje cells showed clear modulations of simple spikes or membrane potential related to luminance changes. Only luminance changes during which fish did not swim were used for this analysis in order to avoid any confounding effects of motor-related activity. Responses varied across cells in terms of both their polarity (whether cells were excited by light onset and/or offset) and their temporal profiles (whether they were relatively sustained or quite transient) (Fig. 5). In addition to explaining the observed frequency-dependent membrane potential oscillations, these luminance responses likely account for some portion of the drift responses we observed due to increases in activity as the grating's darker or lighter regions moved past. In theory, a large enough luminance response with an appropriate decay constant could result in speed modulation as well as a general sensitivity to motion, but any such effect did not appear to be strong in our data. Taken together, these results suggest that although individual Purkinje cells receive strong visual input, most Purkinje cells do not encode drift direction or speed signals that would be appropriate for controlling the OMR.

Subthreshold and simple spike responses to fictive swim bouts.

Larval zebrafish control swim speed by adjusting a small number of swim bout parameters, namely, bout frequency, bout duration, and tail-beat frequency (Budick and O'Malley 2000; Buss and Drapeau 2001; Masino and Fetcho 2005; Severi et al. 2014). The high temporal resolution afforded by electrophysiological recording allowed us to look in more detail at how Purkinje cell membrane potential and simple spiking responses related to parameters of swim bouts. We examined the bout responses of 28 Purkinje cells (including the 11 cells initially examined in the closed-loop OMR paradigm) and found that a majority of Purkinje cells had significant modulations of membrane potential (16 of 19 whole cell recordings) or simple spikes (7 of 9 cell-attached recordings) (Student's t-test, P < 0.01). Eighteen of the twenty-three bout-responding cells had membrane potential depolarizations or increases in simple spike firing responses during bouts (Fig. 6A). The onsets of these responses were closely tied to the onset of the bout itself, recorded at the motor nerve. The average response onset preceded the first recorded motor burst by 39.8 ± 4.27 ms, consistent with the possibility that Purkinje cell activity participates in initiating and/or shaping swim bouts. Bout responses had a variety of time courses (Fig. 6, A and B). In some instances, responses did not peak until after bout termination (Fig. 6B, left). Even in instances where the peak occurred early after bout onset, responses tended not to return all the way to baseline until after bout termination (Fig. 6B, right). Finally, three Purkinje cells exhibited bout-related simple spike firing with fluctuations that were clearly time-locked to the simultaneously recorded motor nerve bout's burst substructure (Fig. 6C). Interestingly, the fluctuations appeared to extend even beyond the end of the recorded swim bouts. These Purkinje cells could participate in signaling or controlling tail-beat frequency, which relates directly to swim power.

A smaller subset (n = 5) of the bout-responsive cells exhibited membrane potential hyperpolarization or simple spike rate decreases during motor bouts (Fig. 7). Unlike the varied timing of the peaks of the bout responses with opposite polarity, the troughs of these cells appeared near bout termination (Fig. 7A). In two cells we recorded enough bouts to subdivide subthreshold bout responses according to bout duration as measured by the number of tail beats per bout. When we compared response timing across these groups we found a strong correlation between bout duration and the time of the response trough (Fig. 7B; R = 0.95 and R = 0.96, P < 0.005). Hence these cells could participate in signaling or controlling the timing of bout termination.

Climbing fiber responses during OMR behavior.

Theories of cerebellar function posit that climbing fiber input to Purkinje cells conveys a teaching or error signal that serves to sculpt appropriate patterns of simple spike firing via plasticity at parallel fiber synapses (Albus 1971; Ito 1972; Marr 1969). If motor command signals related to swim bouts are subject to this type of error-driven correction, CFRs should preferentially occur during or after swim bouts. Indeed, we found that most Purkinje cells (17/28) had CFRs that were modulated during fictive swim bouts (Student's t-test, P < .01) (Fig. 8A).

A potentially relevant teaching signal for adjusting swimming in the context of the OMR would be one that grades based on the stabilization success of a swim bout. If this were the case, the probability of a CFR occurring during a bout would be expected to grade with the magnitude of tail-to-head drift (lack of stabilization) that occurred over its duration. However, when we examined the probability of a CFR during bouts in the context of graded open-loop drift velocities, we saw little evidence of corresponding CFR grading. A binary logistic regression was performed to ascertain the effect of velocity on the likelihood that a CFR would occur during a bout. In only 2 of 11 cells with CFR bout responses was the logistic regression model statistically significant (P < 0.05), and even in those cases the model explained only 9% and 13% (Nagelkerke R²) of the variance in CFR firing across bouts (Fig. 8B).

Consistent with this apparent lack of effect of drift velocity on CFR bout responses, examination of CFRs of cells recorded under closed-loop/playback conditions showed that all cells with increased CFR activity during motor bouts in closed loop (7 of 11) had similar responses to motor bouts in playback when visual consequences were unrelated to motor output (motor triggered, Fig. 8C). No cells had CFR responses to the closed-loop visual stimulus played back independent of the fish's motor commands (visual triggered, Fig. 8C).

Although we detected no image velocity-related modulation of CFR bout responses or CFR responses to visual motion on the timescale of motor bouts, we did observe a number of non-bout-related visual signals in the CFR, indicating that it is not purely a motor signal. Consistent with a previous study (Hsieh et al. 2014), we observed increased CFR activity in response to luminance changes in many cells (15/23) (Fig. 9A). CFRs were also strongly modulated by onset of visual drift in most cells (1-way ANOVA P < 0.05, post hoc Tukey test P < 0.01; 21/29) (Fig. 9B). These drift-related CFR modulations were most often (20/21 cells) selectively driven by one of the two drift directions (12 responded to tail-to-head drift and 8 to head-to-tail drift) and were transient compared with the drift responses observed in the membrane potential and simple spike firing rates (Fig. 4A). Since fish rarely swam at the onset of head-to-tail motion, CFR responses to drift, in this direction at least, are unlikely to be due to motor signals.

The observation of transient CFR modulation in response to drift onset raises the question of why bout timescale visual responses were not observed in our closed-loop data (Fig. 8C). Closer inspection reveals that most swim bouts reach completion before the average onset of CFR modulation due to drift (Fig. 9C). This suggests that these drift-related CFRs would not be triggered by drift on the timescale of a bout, consistent with our earlier observations.

Relationship between simple spikes and climbing fibers in individual Purkinje cells.

Simple spikes and CFRs have distinct origins and likely play different functional roles. Although it has been reported in larval zebrafish that CFRs are capable of evoking membrane depolarization and simple spike firing (Sengupta and Thirumalai 2015), longer timescale relationships between stimulus-evoked modulations of simple spikes and CFRs have not been thoroughly examined (Hsieh et al. 2014). For example, an antiphasic relationship between simple spikes and complex spikes has often been reported for mammalian Purkinje cells (see e.g., Barmack and Shojaku 1992 and Stone and Lisberger 1990). Although we cannot rule out systematic relationships between stimulus-evoked modulations of simple spikes and CFRs, such relationships were not obvious in our data. No significant dependence was found between the existence of a CFR response and a membrane potential or simple spike response for any of the three types of responses we identified: luminance, drifting vs. nondrifting grating, and motor bouts (Fisher's exact test for independence, P < 0.05). The lack of a consistent, predictable relationship between CFRs and simple spikes across recorded Purkinje cells can also be seen in comparisons of sets of cells with similar response properties for one of these signals and contrasting response properties for the other. Similar bout-related increases in CFR rates were associated with simple spike decreases in some cells and increases in others (Fig. 10A). Similar non-direction-selective drift-induced increases in simple spikes could be accompanied by highly direction-selective CFR responses (Fig. 10B). Finally, similar changes in CFR rate in response to a luminance decrease could coincide with change or no change in simple spike firing (Fig. 10C). These results underscore the need for studies aimed at directly determining how calcium responses relate to simple spike vs. CFR activity in larval zebrafish Purkinje cells. It is notable in this regard that calcium imaging studies of mammalian Purkinje cells have focused almost exclusively on calcium transients related to complex spikes and, as far as we are aware, there have been no reports that simple spike firing rates can be recovered from calcium imaging data (Gaffield et al. 2016; Kitamura and Hausser 2011; Ozden et al. 2009; Schultz et al. 2009; Sullivan et al. 2005).

DISCUSSION

During OMR behavior zebrafish adjust the direction and speed of their swimming based on the direction and speed of retinal image motion (Neuhauss et al. 1999; Orger et al. 2008). Hence one possible role for the zebrafish cerebellum would be to transform image velocity signals into swim commands appropriate to stabilize that motion. Such a role would be analogous to that played by the cerebellum in the vestibulo-ocular reflex, during which Purkinje cells participate in transforming head velocity information (conveyed by mossy fibers) into motor commands to counterrotate the eye (Ito 1982). Direction-selective visual responses have been observed in areas of the larval zebrafish brain, including the pretectum, and have been suggested to play a role in the OMR (Kubo et al. 2014; Portugues et al. 2014). However, although we found that a drifting grating stimulus evoked sustained membrane potential depolarization and increases in simple spike firing in Purkinje cells, such responses did not grade with image speed. Direction-selective subthreshold and simple spike responses were also uncommon in Purkinje cells.

While these results clearly do not support a role for the zebrafish cerebellum in adjusting the OMR based on visual signals, they also by no means rule one out. We sampled a relatively small number of cells, only within the corpus cerebelli, within a limited developmental time window, and using a restricted set of visual stimuli. If velocity signals exist in a quite specific region of the cerebellum they could have been missed in our study. Consistent with possible regionalization of function within the larval zebrafish cerebellum, Matsui et al. (2014) found that effects of optogenetic manipulations on OMR-induced swimming were restricted to rostromedial regions of the cerebellum.

In contrast to the paucity of direction- or velocity-sensitive responses, a majority of recorded Purkinje cells exhibited subthreshold and simple spike responses to fictive swim bouts under both closed- and open-loop conditions. Our playback paradigm showed that responses in closed loop are mainly or entirely motor responses, while also demonstrating that the types of rapid changes in visual input that would normally accompany swim bouts are not by themselves a potent driver of simple spike or subthreshold activity. These results are consistent with a previous whole brain calcium imaging study that found that more cerebellar neurons showed activity that was strongly correlated with motor output than visual input during the OMR (Ahrens et al. 2012). They are also consistent with a recent electrophysiological study reporting Purkinje cell depolarizations related to spontaneous fictive swim bouts (Sengupta and Thirumalai 2015).

Our results provide some initial insights into how these motor signals related to swim bouts are encoded in zebrafish Purkinje cells. Although some Purkinje cells responded before motor nerve bouts, others responded at the same time or shortly after bout initiation. While it is not clear from these results whether Purkinje cells are active early enough to participate in initiating bouts, they certainly could participate in shaping their amplitude or duration. Consistent with this latter possibility, a recent study has shown that optogenetic activation or silencing of larval Purkinje cells could modify, but not initiate, swimming movements during the OMR (Matsui et al. 2014). A number of cells we recorded from had striking features when considered from this vantage point. One subset of cells exhibited a hyperpolarization and/or reduction in simple spike firing that was aligned with the termination of motor bouts. These responses clearly graded with bout duration in the cases in which it could be examined. Another small subset of cells exhibited structured simple spike firing within each bout that appeared to track tail-beat frequency, an important contributor to swim speed in larval zebrafish.

Overall, the responses we observed in individual Purkinje cells were heterogeneous and generally did not closely follow the time course of individual bouts. These findings are intriguing in light of a recent study of Purkinje cell encoding of another type of ballistic movement—saccades (Herzfeld et al. 2015). The authors showed that while responses of individual Purkinje cells did not accurately encode saccade kinematics, appropriately chosen populations of Purkinje cell responses did. For example, pooling cells with increases and decreases in simple spike firing related to saccades resulted in a temporal profile of activity that matched saccade kinematics. Bout responses reported here for Purkinje cells are similar to those observed for saccades in that both increases and decreases in simple spike firing were observed in different cells and that simple spike modulations often outlasted bouts. Future studies could examine this issue in more detail in zebrafish with the added possibility of using voltage-clamp recordings from efferent cells to directly isolate the summed response of a local population of Purkinje cells.

Aside from any specific role in controlling the OMR, the strong responses we observed to fictive swim bouts suggest that Purkinje cells play a general role in encoding, controlling, and/or adjusting parameters of larval zebrafish swimming (Sengupta and Thirumalai 2015). Such a generalized role for the cerebellum in adapting motor output could be desirable if the need for changes in swim strength was not task- or context specific (as in the case of the OMR) but due to changes in the motor plant, for example, growth or injury of the organism. In the larval zebrafish such a capacity would seem to be particularly important, as this developing creature's body changes dramatically on a rapid timescale. For example, larval zebrafish increase in length by 60% from 15 to 30 dpf and undergo dramatic changes in tail, anal, and dorsal fin development over this same time period (Singleman and Holtzman 2014). Studies of “natural” motor adaptation in developing zebrafish may provide a valuable complement to the extensive literature on controlled laboratory studies of motor adaptation conducted mainly in humans and monkeys (Shmuelof and Krakauer 2011).

Finally, our electrophysiological recordings provide some of the first information about the signals conveyed by climbing fibers during behavior in larval zebrafish. Studies of mammalian cerebellar involvement in behaviors such as the vestibulo-ocular reflex and smooth pursuit eye movements have suggested that climbing fibers drive adaptive modification of motor gain and timing by instructing synaptic plasticity in cerebellar circuitry, e.g., at synapses between parallel fibers and Purkinje cells (Ito 1993; Ito and Kano 1982; Medina and Lisberger 2008; Simpson et al. 1996). In these contexts complex spikes have been shown to encode “error” signals relevant for improving performance, such as unexpected image motion or retinal slip (Simpson et al. 1996). During the zebrafish OMR, image motion information provides one indication of whether a bout was successful in stabilizing the fish's position. Net tail-to-head image motion indicates that motor output is on average too weak, while net head-to-tail image motion indicates that motor output is on average too strong.

We did find that CFRs occur with significantly greater frequency during swim bouts in most cells, positioning them appropriately to provide a teaching signal about the sensory consequences of motor bouts. However, CFRs did not appear to encode the amount of net drift that occurred during a bout in open loop at different velocities, nor did we observe a clear difference in bout-related CFR activity depending on whether the visual information was predictable based on motor output (i.e., closed-loop vs. playback conditions). Here it is important to acknowledge that although we attempted to create a predictable relationship between the fish's motor output and visual input in closed loop, the precise relationship between swim commands and visual input almost assuredly deviated from that which the fish would normally experience. It is therefore possible that the bout-related CFR activity we observed was in fact in all cases an error signal relating to unexpected visual motion over which we simply did not have sufficient experimental control or readout.

Perhaps even more importantly, given the lack of velocity signals observed in subthreshold and simple spike activity, our data do not rule out that CFRs encode other types of “error” signals about the sensory consequences of motor bouts, e.g., vestibular or mechanosensory lateral line signals. Numerous aspects of the fictive OMR paradigm are unnatural, including the absence of vestibular and mechanosensory lateral line signals. The development of new methods for monitoring neural activity in freely moving zebrafish is needed to circumvent these issues (Naumann et al. 2010).

CFRs could also be involved in adjusting or controlling any number of behaviors we did not examine. We frequently observed that CFRs were modulated by changes in luminance. Such responses could be related to the known involvement of the teleost cerebellum in the dorsal light response (Yanagihara et al. 1993). That CFRs likely encode both sensory and motor signals was supported both by luminance responses as well as by the direction-selective responses to the onset of visual grating drift that we observed (Fig. 9B).

Conclusions.

Understanding the cerebellum will likely require a variety of approaches applied to a variety of systems. The larval zebrafish cerebellum is an attractive candidate for study given its small size and accessibility to population imaging, visualized electrophysiological recordings, and the potential for cell-type specific manipulations of neural activity. Key to such efforts will be defining the inputs to the cerebellum and understanding how they are transformed within cerebellar circuitry in the context of cerebellum-dependent behavior. The present description of the responses of Purkinje cells to visual and motor signals during the OMR provides an initial step toward this long-term goal.

GRANTS

This work was supported by grants from the Alfred P. Sloan Foundation and the McKnight Endowment Fund for Neuroscience to N. B. Sawtell and by grants from the Japan Society for the Promotion of Science (JSPS) (Strategic Young Researcher Overseas Visits Program for Vitalizing Brain Circulation and KAKENHI) to T. Shimizu and M. Hibi.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

K.S., T.S., M.H., and N.B.S. conception and design of research; K.S. and T.S. performed experiments; K.S. analyzed data; K.S. and N.B.S. interpreted results of experiments; K.S. prepared figures; K.S. and N.B.S. drafted manuscript; K.S., T.S., M.H., and N.B.S. edited and revised manuscript; K.S., T.S., M.H., and N.B.S. approved final version of manuscript.