1. Neuroscience
Download icon

Direct translation of climbing fiber burst-mediated sensory coding into post-synaptic Purkinje cell dendritic calcium

  1. Seung-Eon Roh
  2. Seung Ha Kim
  3. Changhyeon Ryu
  4. Chang-Eop Kim
  5. Yong Gyu Kim
  6. Paul F Worley
  7. Sun Kwang Kim  Is a corresponding author
  8. Sang Jeong Kim  Is a corresponding author
  1. Department of Physiology, Seoul National University College of Medicine, Republic of Korea
  2. Department of Biomedical Sciences, Seoul National University College of Medicine, Republic of Korea
  3. Neuroscience Research Institute, Seoul National University College of Medicine, Republic of Korea
  4. Department of Physiology, College of Korean Medicine, Kyung Hee University, Republic of Korea
  5. Department of Neuroscience, School of Medicine, Johns Hopkins University, United States
  6. Department of Physiology, College of Korean Medicine, Gacheon University, Republic of Korea
Research Article
  • Cited 0
  • Views 844
  • Annotations
Cite this article as: eLife 2020;9:e61593 doi: 10.7554/eLife.61593

Abstract

Climbing fibers (CFs) generate complex spikes (CS) and Ca2+ transients in cerebellar Purkinje cells (PCs), serving as instructive signals. The so-called 'all-or-none' character of CSs has been questioned since the CF burst was described. Although recent studies have indicated a sensory-driven enhancement of PC Ca2+ signals, how CF responds to sensory events and contributes to PC dendritic Ca2+ and CS remains unexplored. Here, single or simultaneous Ca2+ imaging of CFs and PCs in awake mice revealed the presynaptic CF Ca2+ amplitude encoded the sensory input’s strength and directly influenced post-synaptic PC dendritic Ca2+ amplitude. The sensory-driven variability in CF Ca2+ amplitude depended on the number of spikes in the CF burst. Finally, the spike number of the CF burst determined the PC Ca2+ influx and CS properties. These results reveal the direct translation of sensory information-coding CF inputs into PC Ca2+, suggesting the sophisticated role of CFs as error signals.

Introduction

Each Purkinje cell (PC), the sole cerebellar output neuron, receives strong excitatory inputs from the inferior olive (IO) through a single climbing fiber (CF), which innervates several PCs (Eccles et al., 1966). During cerebellar learning, the CF fires in response to unexpected sensory events to provide instructive signals to the PC, turning on Ca2+ mediated plasticity mechanisms (Hansel and Linden, 2000; Rancz and Häusser, 2006). According to the Marr–Albus–Ito theory of learning, a CF-induced PC complex spike (CS) response is ‘all-or-none’ because IO stimulation generates seemingly binary responses in the PC (De Schutter and Maex, 1996; Marr, 1969). This notion has been prevailed as slice studies have show that a single CF stimulation induces similar EPSC (excitatory post-synaptic current) above a certain stimulus intensity (Konnerth et al., 1990) and is enough to induce parallel fiber (PF)-PC synapse long-term depression (LTD) (Ito and Kano, 1982), in which the level of learning is determined by the range of PF excitation (Reynolds and Hartell, 2000). This means that the CF’s impact has little variation in strength and does not carry quantitative information, although CF input, by controlling the Ca2+ transient in PC, could critically affect PF-PC LTD (Ohtsuki et al., 2009). Instead, it is generally accepted that the magnitude of learning is affected by how many invariant CFs synchronously fire (Squire, 2009).

However, the so-called ‘all-or-none’ property of CF error signal becomes questionable when it comes to in vivo conditions, since the CF’s bursting properties have been described (Mathy et al., 2009) and a single non-burst stimulation of CF failed to induce associative learning (Rasmussen et al., 2013). In line with this, the previously unknown complexity of the error signal has received attention (Najafi and Medina, 2013). The CF burst reflected the olivary oscillation frequency and the number of spikes ranged from 1 to 3 in anesthetized animals (Mathy et al., 2009). Also, burst activity seemed to be affected by certain types of visual stimulation (Maruta et al., 2007). Furthermore, longer CS duration, presumably resulting from the longer burst, was associated with enhanced cerebellar motor learning (Yang and Lisberger, 2014).

It is thus highly interesting to determine whether and how a CF itself encodes quantitative information of unexpected sensory events, such as differential stimulus intensities, and transmits the error signals in a graded manner to the PC in awake animals. One recent report described the sensory-driven enhancement of PC Ca2+ signals, in which the authors only observed post-synaptic PC Ca2+ transients, which they claimed non-CF components affect (Najafi et al., 2014b). Although CF Ca2+ activity was described in vitro (Nishiyama et al., 2007), so far, sensory coding by the CF can hardly be investigated because sensory inputs disappear in ex vivo preparation and the direct recording of the CF axon has been technically infeasible to obtain in vivo except for recent studies (Gaffield et al., 2018; Gaffield et al., 2019). Using genetically encoded Ca2+ indicators (GECIs) and in vivo 2-photon microscopy imaging of CF and/or PC populations in awake mice, here we show CF burst-mediated sensory coding and its direct representation in a post-synaptic PC. At rest, the CF Ca2+ activity exhibited highly variable strength and synchrony. Electrophysiological analysis indicated a direct correlation between CF Ca2+ amplitude and the number of spikes in the CF burst. Employing unexpected sensory stimuli, we revealed that CF Ca2+ signal encodes sensory stimulus intensity, just as PC Ca2+ activity does. Our dual-color simultaneous imaging of CF and PC Ca2+ signals also indicated a linear correlation between CF and PC responses in awake animals, suggesting that presynaptic CF inputs carrying sensory information strongly contribute to post-synaptic PC responses.

Results

CF Ca2+ activity is highly variable in strength and associated with the number of spikes in the burst

CF activities have been indirectly inferred from the signatures of CS in PCs by attached or field potential recordings, which are contaminated with local circuit activities such as those of the PF or molecular interneuron (MLI). In this study, 2-photon Ca2+ imaging with GECI expression enabled us to directly visualize in vivo CF activity. We injected an adeno-associated viral (AAV) vector into the IO, which expresses GCaMP6f using Camk2a promoter (Figure 1a). Histological characterization reveals strong green fluorescence in bilateral IOs (Figure 1a). The neuronal somas in the brain stem and the axon fibers in cerebellar white matter appeared green (Figure 1a2-3). CF axon buttons and shafts were also observed in the molecular layer (Figure 1a4).

The CF Ca2+ activity is highly variable in awake mice in vivo.

(a) Schematic diagram of the IO viral injection. A coronal section view of the brain stem region of a brain in which GCaMP6f was expressed for 3 weeks (a1). GCaMP6f expressed soma and projecting axons (yellow and white arrowhead, respectively) in a brain stem region including the IO (a2). GCaMP6f-expressed axons (white arrowhead) within white matter (a3). GCaMP expressed CF varicosities (white arrow) within the molecular layer of lobules IV/V (a4). (b) A schematic diagram of two-photon microscopy of awake mice on a disk treadmill. A coronal view of the z-stack projection image of CFs expressing GCaMP6f in the cerebellar cortex (b1). A dorsal view of the z-stack image (maximum projection image) of white box regions of b1, which represent the molecular layer (b2). (c) An example of ROI detection of CF varicosities using the Suite2p. The field of view is 512 × 64 pixels. (d) Resting-state GCaMP6f intensity traces the 16 ROIs over 60 s with event detection plot (grey lines). Intensities were expressed as standard deviation as signals were z-score normalized. (e) An example matrix of correlation coefficients among every pair of the 16 ROIs, with the Pearson correlations described by colored intensity. Right: A scale bar for correlation coefficient. (f) The correlation coefficient among the CFs in terms of the mediolateral distance between each pair of all ROIs. n = 397 pairs of CFs, R2 = 0.034. (g) Average frequency (0.85 Hz ± 0.24 SEM, n = 69 varicosities), amplitudes by cell (2.41 SD ± 0.045 SEM, n = 69 cells), and amplitudes by event (2.349 SD ± 1.49 SD, n = 3481 events, N = 5 mice).

To observe CF Ca2+ activity, we created a cranial window on the cerebellar cortex of the lobule IV/V vermis while injecting AAV-Camk2a-GCaMP6f in IO. Three weeks later, we performed 2-photon Ca2+ imaging in awake, head-fixed mice on a disk treadmill (Figure 1b). The coronal projection image of the z-stack indicates the strong fluorescence of CFs that terminate in the cerebellar cortex (Figure 1b1). The z-projection image of molecular layer clearly shows CF axonal varicosities in the CFs (Figure 1b2 from the box in 1b1). For accurate identification of simultaneously firing pixels of CF Ca2+ and reliable signal/event detection, we have utilized Suite2p, an open-source Ca2+ imaging tool, which allows detection of the Ca2+ signals from axonal varicosities (Marius Pachitariu et al., 2016). Suite2p efficiently identified CF varicosities where ones with correlation coefficient of 0.85 were merged in its graphical user interface, while ROIs with low signal-to-noise level are discarded (Figure 1c). Similar to a previous report about CF-evoked PC activity (Ozden et al., 2012), the resting state CF firing frequency was 0.85 Hz ±0.24 SEM and the CFs population also showed highly synchronous activity within an imaging field of 208 µm-wide (mediolaterally) (Figure 1e). As the Ca2+ synchrony of nearby PC dendrites is known to decline through mediolateral separation (Ozden et al., 2009; Schultz et al., 2009), the synchrony level gradually fell as a function of mediolateral distance (Figure 1f and Figure 1—source data 1).

Strikingly, CF amplitudes were substantially variable (2.349 ± 1.49 SD, Figure 1g and Figure 1—source data 1) with maximum being up to 8.366, while the average Ca2+ amplitudes per cell converged to 2.41 SD ±0.045 SEM. Mathy et al., 2009 recorded in vitro CF axonal activity and described its bursting property, which encodes olivary oscillation. Also, EPSC numbers in PC, presumably evoked by the spike number of the CF burst varied between 1 and 3 under anesthesia. Because the CF Ca2+ activity we observed here was also variable, we asked whether CF Ca2+ activity is related to the number of spikes in the burst. In cerebellar slices prepared from the mice expressing GCaMP6f in IO, CFs were directly stimulated at the granule cell layer with 1 to 9 spikes in 400 Hz bursts (Figure 2a,b). Our data indicate that increasing the number of spikes in the burst stimulations accordingly augments the CF Ca2+ amplitude, which is saturated with a burst of seven spikes (Figure 2c–d and Figure 2—source data 1), showing that the variability of CF Ca2+ activity is positively correlated with the number of spikes per CF burst. To support this notion, we performed amplitude histogram analysis with in vivo CF Ca2+ amplitudes of 3 min recordings. Only first-peak amplitudes within 0.5 s windows were gathered since closely following events can exert an additive effect on amplitudes (Figure 2d). Interestingly, the distribution appeared discrete, with several amplitude clusters existing (Figure 2e and Figure 2—source data 1), possibly driven by the number of spikes in the burst. Hence, the huge variability in CF Ca2+ amplitude may be caused by the variable number of CF bursts.

The variable CF Ca2+ activity encodes the spike number in the burst.

(a) The sagittal slice image of GCaMP6f-expressed CF in the cerebellar PC and molecular layer. Five responding axonal varicosities (<2 µm) were selected (as indicated by yellow circles) and averaged for each cell’s traces. The approximate PC soma was marked with a thick white dashed line. The patch pipette was depicted as a thin white dashed line. (b) Burst stimulation–evoked CF GCaMP6f intensity plots of the 30 s recordings of nine independent CFs from three mice. The thick red trace represents the averaged trace. (c) Quantification of the burst-evoked GCaMP6f amplitudes. One-way ANOVA followed by Bonferroni test: ***p<0.001, *p<0.05. (d) Example trace of in vivo CF Ca2+ imaging showing a sampling of events for amplitude distribution analysis in e. Only the first-peak amplitudes out of 0.5 s window were analyzed. The short lines below the trace represent the sampled events. (e) Amplitude distribution of 31 CFs from four independent 3 min recordings of 2-photon Ca2+ imaging in three animals. The first-peak amplitudes were normalized with the median values to display all data and were presented as a heat map histogram, in which x and y represent normalized amplitudes and cell numbers, respectively. The color map scale shows the number of events.

The spike number of CF bursts encodes graded sensory information

Considering the variability of CF Ca2+ by spike number (Figure 1g), we asked whether the spike number in the CF burst could encode the intensity of natural sensory stimuli by employing periocular air-puff stimulation (Figure 3a), which triggers trigeminal CFs projecting to the paravermal lobule V regions (Najafi et al., 2014a). We tested the two strengths of stimulation sets that were reported to differentially modulate PC dendritic Ca2+ responses (Najafi et al., 2014a). In this study, we only analyzed data at resting state and excluded images in which the animals are walking or running since locomotion produces complex Ca2+ transients. As for the strength, CF Ca2+ amplitude was significantly enhanced with lower pressure (P1) than with spontaneous responses and further increased in higher pressure (P2) (Figure 3b–e and Figure 3—source data 1). Here, neither P1 nor P2 air-puff stimulation caused motion artifacts in our experimental condition (Figure 3—figure supplement 1). These data indicate that the CF Ca2+ amplitude—the spike number of the burst, in other words—conveys information about sensory stimuli with differential strength.

Figure 3 with 2 supplements see all
Sensory coding by CF Ca2+ signals.

(a) A schematic showing the 2-photon imaging with periocular air-puff stimulation and animal motion monitoring by IR camera. (b-c) Representative CF Ca2+ traces for spontaneous and 30 ms periocular air puffs (orange column) of pressure 1 (P1, b) and 2 (P2, c). (d-e) Averaged CF Ca2+ traces (d) and amplitudes (e) of spontaneous and air-puff responses of P1 and P2. n = 448, 283 and 346 CF Ca2+ events in 4 (P1) and 6 (P2) independent 1 min imaging sessions from four mice, respectively. One-way ANOVA followed by Tukey’s test: **p<0.01, ***p<0.001, **** p<0.0001. (f) Correlation analysis between CF Ca2+ amplitudes versus the peak speed of forepaw movement at rest. n = 2926 pairs from four recordings in two mice, R2 <0.001. (g) Representative traces of the speed of forepaw twitch-like movement during periocular air-puff stimulation. Air puff–to–motion onset time = 23.2 ± 2.9 SEM ms, air puff–to–peak motion speed time = 57.6 ± 5.8 SEM ms. (h) Correlation analysis between CF Ca2+ amplitudes versus the peak motion speed during air-puff stimulation. n = 760 pairs from four recordings in two mice. R2 <0.001.

CF inputs to the cerebellum have also been known to be associated with body movements (Ozden et al., 2012) and could be evoked even by small movements in the paravermal area of lobule V (Rushmer et al., 1976). Since CF Ca2+ has movement-related activities in our experiments, we tested whether the spike numbers encoded in CF Ca2+ amplitudes are related to movements by correlation analysis between CF Ca2+ amplitudes and the strength of the small movement. The animal’s motion speed was acquired using high-speed IR camera by tracking the IR-reflective patch on the forepaw at rest and during air-puff stimulation (Figure 3a). At rest, CF Ca2+ events were not correlated with the peak motion speed (R2 <0.001, Figure 3f and Figure 3—source data 1). Air-puff stimulation generally induced small twitch-like movement (Figure 3g and Figure 3—source data 1), while it was not correlated with CF Ca2+ amplitudes either (R2 <0.001, Figure 3h). Although we did not examine any other movements related to air-puff stimulation, such as orofacial movement, we suggest that the CF Ca2+ activity could differentiate the graded sensory stimuli but not movement strength under the assumption that forepaw motion represented startle movement during the air-puff stimulation.

Next, we sought to determine whether our set of ipsilateral periocular air-puff stimuli induced graded responses in PC dendritic Ca2+, as previously reported (Najafi et al., 2014b). For the specific expression of GECI in PCs, we used Pcp2-cre transgenic (TG) mice, of which cerebellar vermis was targeted for cre-dependent jRGECO1a expression (Figure 3—figure supplement 2a–b and Figure 3—figure supplement 2—source data 1). The PC dendritic Ca2+ activity was similarly detected with Suite2p. Sensory stimuli were shown to enhance PC Ca2+ responses, as compared to spontaneous responses, and were graded with different pressure strengths (Figure 3—figure supplement 2c–d). The results collectively suggest that the spike number of CF bursts reflects sensory strength and may direct the PC-mediated strength-dependent sensory coding.

Direct translation of spike number in the CF burst by PC dendritic Ca2+ signals

Although the aforementioned data suggest that both the CF and PC process sensory information, these results do not ensure their direct correlation. Hence, we set out to simultaneously record the pre- and post-synaptic Ca2+ activity. In Pcp2-cre TG mice, AAV-Camk2a-GCaMP6f was injected into the IO (Kimpo et al., 2014), and the cre-dependent expression of jRGECO1a—the sensitive red color Ca2+ indicator (Dana et al., 2016)—was achieved (Figure 4a). Dual-color imaging was performed using GFP and RFP filters under 1000 nm two-photon laser excitation, which revealed the structure of CF axon varicosities and PC dendrites located adjacent to them (Figure 4b). The CF-PC pairs were readily identifiable by their proximity (Figure 4b). To check whether signal bleed-through existed between the two channels, dual-channel imaging was performed in the cerebellums expressing only CF-GCaMP6f and PC-jRGECO1a, respectively. Even though some very strong signals of CF-GCaMP6f varicosities were visible in the red channel, no obvious spectral overlaps existed in selected ROIs (Figure 4—figure supplement 1a). Also, the transmission of jRGECO1a signals into EGFP filters was negligible (Figure 4—figure supplement 1b). Furthermore, Ca2+ signal processing with Suite2p should remove such negligible bleed-throughs since Suite2p detects fluorescence traces with a high signal-to-noise ratio (Marius Pachitariu et al., 2016).

Figure 4 with 1 supplement see all
The direct translation of spike number for each CF burst by post-synaptic PC Ca2+ response.

(a) A schematic that shows the dual expression of Ca2+ indicator jRGECO1a and GCaMP6f in PC and CF, respectively. (b) ROI detection by Suite2p and average projection images of CF and PC dual-calcium imaging in the cerebellar cortex (50 µm from dura). Two examples of CF-PC pairs are indicated with number. (c) Traces of the two CF-PC pairs (1 and 2) from b. (d) Correlation coefficients of the signals are shown in four conditions which include ‘paired’, ‘unpaired’, ‘neighboring unpaired’ and ‘distant unpaired’. n = 13 (paired), 61 (unpaired), 37 (neighboring unpaired), and 24 (distant unpaired) CF-PC pairs. One-way ANOVA followed by post hoc Tukey multiple comparisons test. ****p<0.0001, ***p<0.001, *p<0.05. (e) A representative non-linear regression analysis of Ca2+ amplitudes in two pairs of CF-PC shown in c. n = 73 (pair 1) and 85 (pair 2) events. R2 = 0.756 (pair 1) and 0.658 (pair 2). (f) R2 values (left) and slope (right) for CF-PC pair Ca2+ amplitude-correlation analysis with 13 CF-PC pairs. Average R2 = 0.48 ± 0.06 SEM. Average slope = 0.61 ± 0.08 SEM. Data are from five independent recording sessions of three mice.

As shown in Figure 4c, signal traces and the detected events from the two CF-PC pairs revealed a highly synchronous activity (Video 1). To determine whether CF Ca2+ activity covaries with PC Ca2+ activity, we performed correlation analysis in terms of signal and paired amplitudes. For signal correlation analysis, the correlation coefficient between the CF and PC Ca2+ signals was computed. The CF and the identified PC dendrites spatially located in the proximity (<1 μm) with the CF were regarded as ‘paired’ CF-PC (Figure 4b,d), for which the correlation coefficient reveals a close relationship (R = 0.605 ± 0.027 SEM). But the correlation was significantly lower in ‘unpaired’ CF-PC, where PC dendrites were remotely located from the CF ROIs (R = 0.443 ± 0.032 SE, Figure 4d and Figure 4—source data 1). The ‘unpaired’ CF-PC pairs were further categorized as ‘neighboring unpaired’ and ‘distant unpaired’ if at least 30 µm apart from each other. The distantly located CF-PC showed significantly lower signal correlation (0.327 ± 0.057 SE) compared to neighboring ones (0.498 ± 0.032 SE, Figure 4d,e and Figure 4—source data 1). Next, we analyzed the correlation of CF-PC amplitude pairs that were collected where the CF Ca2+ events occurred. Linear regression analysis with the two CF-PC pairs revealed significant correlation (Figure 4e and Figure 4—source data 1). Such a prominent correlation was consistent with 13 pairs (Average R2 = 0.48 ± 0.06 SEM, Figure 4f and Figure 4—source data 1). Thus, the results suggest that CF Ca2+ signals are directly translated into post-synaptic PC Ca2+ signals in awake animals.

Video 1
CF-PC dual-calcium imaging.

60 s 32 Hz time-lapse movie for CF-PC dual imaging by two-photon microscopy using 1000 nm excitation. CF, PC, and merged images from the top to the bottom. Time scale is indicated in upper left and the scale bar is indicated lower right as 20 μm.

The spike number of CF bursts directly affects the amplitudes of PC dendritic Ca2+ response

Mathy et al., 2009 reported that varying the number of spikes in CF burst affects the number of spikes in post-synaptic PCs. Hence, we tested whether the different spike numbers of CF bursts cause graded Ca2+ spike amplitudes in post-synaptic PC with ex vivo Ca2+ imaging (Figure 5a). The low-affinity Ca2+ dye Fluo-5F was loaded after making the whole-cell at the PC soma. The paired CFs were stimulated at 400 Hz while whole-cell recording and Ca2+ imaging were performed. Interestingly, a higher number of burst stimuli induced greater amplitudes in the post-synaptic PC dendritic Ca2+ response (Figure 5b,e and Figure 5—source data 1). Also, the duration of CS and the spikelet number were significantly enhanced the number of CF stimuli increased (Figure 5b–d and Figure 5—source data 1). These results strongly indicate that the spike number of CF bursts directly affects the PC CS properties as well as Ca2+ amplitudes, suggesting powerful presynaptic governance in PC Ca2+-mediated sensory coding.

The CFs’ spike number-dependent Ca2+ influx in PC dendrites.

(a), Representative image of a PC filled with the low-affinity Ca2+ dye Fluo-5F taken with whole-cell recording. The schematics on the left describe the number of CF stimuli (1, 3, 5, 7, and 9) at 400 Hz. b, Representative aligned traces of CS recordings and PC Ca2+ traces measured by Fluo-5F in response to the indicated numbers of 400 Hz CF stimuli (1, 3, 5, 7, and 9). The length between the two red asterisks represents the duration of CS. c–e, CS duration (c), spikelet numbers, (d) and amplitudes of the post-synaptic PC Ca2+ transient (e) in response to different number of spikes in the CF burst stimuli. n = 7 recordings of seven independent cells from three mice. One-way ANOVA followed by Bonferroni test: *p<0.05, **p<0.01, ***p<0.001, and ****p<0.0001.

Discussion

In our study, we made a series of novel observations demonstrating the significant role of CF input in cerebellar sensory coding. First, Ca2+ imaging in CF axon varicosities in awake mice showed great variability in a resting-state activity. The ex vivo experiments revealed that Ca2+ activity directly reflects the number of spikes in CF bursts. Also, by applying air puffs as sensory stimuli, we found that CF bursts convey quantitative sensory information, just as PC Ca2+ signals do. Further, CF-PC dual-color Ca2+ imaging revealed a systemic correlation between pre- and post-synaptic activity during rest. Finally, the number of spikes in the CF burst linearly affected the CS properties and Ca2+ influx in PCs. These results suggest that PC dendritic Ca2+ activity and its sensory coding process are largely governed by the sophisticated control of presynaptic CF inputs.

The substantial variability of CF Ca2+ activity stems from its burst activity

The classical view of olivo-cerebellar transmission is that PC CS is an ‘all-or-none’ response (Eccles et al., 1966). The CF activity itself has also been regarded as having a binary property, but the poor signal-to-noise problem is compensated for by pooling the activity of multiple CFs (Najafi and Medina, 2013). This was further supported by observations that synchronous activity of PC CS is related to cerebellar information processing, such as movements and sensory stimuli (Ozden et al., 2012). Hence, it has been thought that more significant information is processed when more CFs are activated. However, it remains unresolved whether individual CFs encode parametric information of sensory input (Squire, 2009).

The description of CF burst activity in vivo has made the properties of individual CFs appear more complicated than previously thought (Mathy et al., 2009). In that study, the recorded number of spikes in the CF burst varied from 1 to 3 in anesthetized animals, making it possible that the spike number reflects the degree or types of certain information (Mathy et al., 2009). Yet, the authors concluded that the spike number may not carry the strength or intensity of sensory inputs, in support of a study in which cat IO neurons showed a unitary spike in response to the stimulation of afferent inputs to IO (Crill, 1970). However, such studies rely on results from anesthetized animals and the physiological role of axon burst has remained unanswered. In our study, we overcame the difficulty of direct CF recording by employing two-photon Ca2+ imaging and GECI expression in CFs. Also, since general anesthesia disrupts normal neuronal firing (Greenberg et al., 2008), we performed imaging in awake animals. The open-sources software, Suite2p, enabled us to successfully detect individual CF varicosities and their Ca2+ signals (Figure 1c–d). The paired correlation fell off with increased mediolateral distance (Figure 1f), which is similar with the PCs’ correlation (Schultz et al., 2009). The high variability of CF Ca2+ amplitudes in awake mice, even during the resting state, was very interesting (Figure 1d,g). Although small sounds or movement could induce CF firing (Ozden et al., 2012; Rushmer et al., 1976), the maximum amplitude was 8.366 SD, suggesting a highly variable range of fold-change (Figure 1g). This result is in line with a study that reported the number of excitatory post synaptic potentials (EPSPs) of PCs ranges from 1 to 5 during spontaneous activity (Maruta et al., 2007). Also, a recent study reported the variability of CF Ca2+ activity in the Crus II lobule (Gaffield et al., 2019). Importantly, we revealed that the CF Ca2+ amplitude was directly regulated by the number of spikes in the CF axon burst stimulation at 400 Hz (Figure 2a–c), which suggests that the Ca2+ amplitudes are a readout of the degree of bursts. The physiological number of spikes likely ranges from 1 to 6 or seven in an awake state, since the ex vivo CF Ca2+ responses are saturated in a burst of 7 spikes. The discrete amplitude distribution of in vivo signals further supports the dependency on the spike number in the burst (Figure 2e). Thus, the great variability of CF Ca2+ signals in awake mice tells us that the spike number in the CF burst also varies even in the resting condition.

CF bursts convey sensory information to post-synaptic PCs

Several reports have shown that PC dendritic Ca2+ spikes convey sensory information (Kitamura and Häusser, 2011; Najafi et al., 2014a; Najafi et al., 2014b). We also confirm that PC Ca2+ amplitudes differentiate the strength of periocular air-puff stimuli (Figure 3—figure supplement 2). If the sensory coding of a PC is derived from presynaptic CF activity, then the CF should differentiate the sensory stimuli and its strength. Here, we observed how sensory input enhanced CF Ca2+ response which was further enhanced with stronger stimuli (Figure 3). This is strong evidence for the critical role of presynaptic CF input in shaping PC Ca2+ response during unexpected sensory events. Also, the results from dual-color Ca2+ recording of CF axons and PC dendrites in resting-state revealed that the signal and amplitudes of CF-PC pairs have robust correlations (Figure 4), suggesting that PC Ca2+ activity is strongly driven by CF burst activity.

It seems that PF and CF inputs converge onto the same PC during sensory events (Apps and Garwicz, 2005). This allows some to argue that PF may contribute to the sensory coding of CF-evoked PC Ca2+ amplitudes by showing that small supralinear non-CF inputs were detectable in the absence of CF-induced response (Najafi et al., 2014b). However, a recent study reported that silencing CF activity completely abrogated PC Ca2+ activity (Gaffield et al., 2019), and it is also unlikely that PF-evoked Ca2+ activity is dendrite-wide as CF-evoked responses are (Kitamura and Häusser, 2011). Further, the so-called ‘sensory-evoked non-CF inputs’ do not account for the tight coupling of CF-PC activity even during the resting condition (Figure 4d). We presume that such non-CF contributions in CF-evoked PC Ca2+ response are scarce, and PC Ca2+ activity—during resting or sensory processing—is mostly determined by CF burst activity. On the other hand, however, CF and PC Ca2+ signals were not perfectly correlated. Hence, CF activity may serve as a trigger with qualitative information for post-synaptic response, and other factors, such as PC intrinsic properties (Kitamura and Häusser, 2011), PF activity-mediated depolarization (Wang et al., 2000) and noradrenergic pre-synaptic control (Carey and Regehr, 2009), may participate to form an even more complex shape of PC Ca2+ and CS responses.

Physiological impacts of CF pre-synaptic governance over PC activity

What is the meaning of CFs having such powerful and fine control over PCs? First, the strength of the cerebellar output to the deep cerebellar nucleus (DCN) can be regulated at the level of the CFs’ input magnitudes. It has been thought that an increased population of so-called binary CF activation will generate stronger PC-mediated outputs (Squire, 2009). However, our data suggest that individual CFs can provide a differential magnitude of inputs onto PCs, depending on the number of spikes in their burst (Figure 5). Considering the high synchrony of CF population activity (Figure 1g), they tend to fire simultaneously like PCs do (Kitamura and Häusser, 2011; Ozden et al., 2012; Tsutsumi et al., 2015). On top of that, the graded amplitudes of synchronously firing CFs can help generate a diverse range of cerebellar outputs. Second, the variable CF activities will present more sophisticated error signals to the PC during learning. Yang and Lisberger, 2014 presumed that the CS duration is the critical determinant for the degree of learning, as expected by recording CSs in monkey undergoing smooth pursuit learning. They found that the CS duration tends to be longer in the first 30 trials out of 100. This is also in line with a study showing that the degree of IO stimulation determines the direction of learning (Rasmussen et al., 2013). We provide direct evidence that CS duration, spikelet number, and PC dendritic Ca2+ are all critically affected by the number of CF bursts (Figure 5), suggesting that stronger or newer experiences will generate longer CSs with more spikelet number and Ca2+ influx (Figure 3). Thus, the CF Ca2+ activity may be strong at first and weaken over time during cerebellum-dependent learning such as during eye-blink conditioning. Also, limiting the spike number to one or two at the initial learning phase will interrupt the learning process. Finally, the degree of learning-associated motor control could also be determined at the IO activity level. Behavioral learning may weaken CF activity over time, thereby decreasing the PC output onto the DCN, which then produces stronger motor output signals (Low et al., 2018).

The CF burst depends on the oscillatory state of the IO neurons, which are electrically coupled by gap junctions (Lampl and Yarom, 1993; Mathy et al., 2009). This oscillatory property is heterogeneous (Hoge et al., 2011), and each olivary neuron has a distinctive and stable oscillatory property (Khosrovani et al., 2007). Interestingly, the cerebellar cortex is compartmentalized in terms of Zebrin II expression in the PC, and the CS pattern also differs by zones (Cerminara et al., 2015). We observed that PC and CF Ca2+ properties differ significantly by zones (Roh et al., 2017). Thus, we suspect that the heterogeneous property of oscillation of IO neurons, which determines CF burst, may shape the specific patterns of CSs across cortical zones, facilitating diversified control over CSs, and dendritic Ca2+ transients in the PC.

This study suggests that CF pre-synaptic activity conveys variable and graded sensory information to post-synaptic PC in vivo, which is in line with other recent studies that denied the long-held ‘all-or-none’ notion for the CF activity (Najafi et al., 2014b; Yang and Lisberger, 2014; Gaffield et al., 2019). Such graded signal transmission is governed by the state of olivary oscillation—the number of spikes in the burst. Hence, it calls for the investigation of how IO sophisticatedly controls PCs over the whole cerebellum, which will unveil crucial mechanisms for cerebellar learning.

Materials and methods

Key resources table
Reagent type
(species) or resource
DesignationSource or referenceIdentifiersAdditional
information
Strain, strain background Mus musculusB6.129-Tg(Pcp2-cre)2Mpin/JJackson LaboratoryRRID:IMSR_JAX:004146Stock no: 004146
Strain, strain background M. B6.Cg-Tg(Camk2a-cre)T29-1Stl/JJackson LaboratoryRRID:IMSR_JAX:005359Stock No: 005359
OtherAAV1.Camk2a.GCaMP6f.WPRE.SV40Upenn Vector Core
OtherAAV1.CAG.FLEX.jRGECO1a.WPRE.SV40Upenn Vector Core
OtherAAV1.CAG.FLEX.GFP.WPRE.SV40Upenn Vector Core
Chemical compound, drugZoletilVirvac
Chemical coumpound, drugRompunBayer
Chemical compound, drugDexamethasoneSamyang Phamaceutical
Chemical compound, drugMeloxicamBoehringer Ingelheim
Software, algorithmMATLABMathwroks IncRRID:SCR_002881
Software, algorithmPythonhttps://www.python.org/RRID:SCR_008394

Animals, craniotomy, and genetically encoded Ca2+ indicator (GECI)

Request a detailed protocol

The experimental processes were approved by the Seoul National University Institutional Animal Care and Use Committee and performed under the guidelines of the National Institutes of Health. Seven- to ten-week-old wild-type or B6.129-Tg(Pcp2-cre)2Mpin/J (Jackson Laboratory, ME, USA) mice were anesthetized with intraperitoneal injections of Zoletil/Rompun mixture (30 mg / 10 mg/kg). A small craniotomy was made over lobule IV/V of the cerebellar vermis/paravermis according to previous descriptions but with some modifications (Kim et al., 2016). In short, after placing the anesthetized mouse on a stereotaxic frame (Narishige, Tokyo, Japan), the skin was incised, and bone was removed with a no.11 surgical blade. To minimize edema and related inflammation, dexamethasone (0.2 mg/kg) and meloxicam (20 mg/kg) were administered by subcutaneous injection. A metal ring for head fixation was attached with Superbond dental cement (Sun Medical, Japan). For PC-specific GECI expression, 100–200 nl virus solution of 3–5 × 109 genome copies containing AAV1.CAG.FLEX.jRGECO1a.WPRE.SV40 (Upenn Vector Core, PA, USA) were injected at two or three sites at the cerebellar cortex of the Pcp2-cre TG mice with a beveled glass pipette (5 MΩ). Then, a 1.3 × 2.3 mm size glass coverslip (Matsunami, Japan) was tightly placed on the cortex and fixed by applying cyanoacrylate glue (Vetbond, 3M). For GCaMP6f expression in CF, the virus was injected into the IO 3–4 days before the creation of a chronic window, as previously described (Kimpo et al., 2014). Briefly, bilateral injections were made at the midpoint between the edge of the occipital bone and the C1 cervical vertebra. The glass pipette was set at a 55° angle from vertical and 7° from the midline. After approaching a 2.5 mm depth, virus solution containing 100–200 nl of AAV1.Camk2a.GCaMP6f.WPRE.SV40 was injected with a Picopump at 5 nl / sec. The pipettes were left in place for 10 min before they were removed to minimize backflowing. For GFP expression in CFs or PCs, AAV1.CAG.FLEX.GFP.WPRE.SV40 was injected into Pcp2-cre or Camk2a-cre mouse (B6.Cg-Tg(Camk2a-cre)T29-1Stl/J) (Jackson Laboratory, ME, USA).

Two-photon microscopy and chronic awake imaging

Request a detailed protocol

Confocal microscopy was performed with a laser scanning multiphoton microscope (Zeiss LSM 7 MP, Carl Zeiss, Jena, Germany) equipped with non-descanned detectors (NDD). Excitation was carried out with a mode-locked titanium:sapphire laser system (Chameleon, Coherent, Santa Clara, CA, USA) operating at a 900 nm wavelength for GCaMP6f using GFP filter and 1030 nm for jRGECO1a using RFP filter. Generally, objective W Plan-Apochromat 20×, 1.0 numerical aperture (Carl Zeiss) was used. Images were acquired using ZEN software (Zeiss Efficient Navigation, Carl Zeiss) and processed using a custom-written MATLAB (MathWorks) script. High-resolution of 512 × 512 pixel reference images were acquired at a rate of 8 s per frame in the PC layer (120–150 μm from the dura) and molecular layer (around 20–60 μm from the dura). For the 3D Z-stack images of the CF, 512 × 512 pixel images were acquired at every 2 µm from the dura to a 180 µm depth using the depth correction mode by ZEN. For PC and CF Ca2+ imaging, 32 Hz high-speed time-lapse scanning was performed with a 512 × 64 resolution window at 30–50 μm from the dura. For awake imaging, 10 days after chronic window surgery, the mice were subjected to handling until they show grooming (5–10 min) as well as acclimation on a custom-made disk treadmill with their head fixed using a clamp and custom-made metal rings (30–60 min) for 3 days. Imaging was performed 14 days after surgery.

CF varicosity and PC dendritic Ca2+ signal analysis

Request a detailed protocol

CF Ca2+ imaging data were processed and analyzed with the open-source analysis tool Suite2p which efficiently detects Ca2+ signals both at individual axonal varicocities (Marius Pachitariu et al., 2016). After motion correction and source extraction by Suite2p, the ROIs with low signal-to-noise ratio were discarded and ones with proper size/morphology were selected for analysis in the Suite2p’s graphic user interface (Figure 1c). If ROIs are located in the similar parasagittal plane (~20 µm) and their correlation coefficient of signals are above 0.85, the pairs were merged in Suite2p. Then fluorescence signals (F) were subtracted with neurophil (background) signals and the signals were z-score normalized for further analysis (https://github.com/NeuRoh1/Calcium_signal_processingRoh, 2020; copy archived at swh:1:rev:b0732bd900ccf0e198f52818366f93715a49cea6). The event detection was performed using Suite2p’s built-in deconvolution (tau: 0.4, window for maximum: 60, smoothing constant for gaussian filter: 25), and the deconvolved signals were scaled to match the amplitude of the z-scored F signals (Figure 1d). The amplitudes of CF Ca2+ transients were obtained from peak values of detected events. To ensure the event detection quality, the events with amplitudes lower than 0.5 SD were discarded. To analyze the synchrony of the Ca2+ spikes, Pearson’s correlation coefficients for the firing patterns between every pair of ROIs were computed (Figure 1e). For ‘synchrony by mediolateral distance’ analysis, the mediolateral distance between pairs of all ROIs was obtained with ImageJ and paired with the corresponding calculated R values for correlation analysis (Figure 1f). PC Ca2+ signals imaged with jRGECO1a were similarly processed with Suite2p. The ROIs of PC dendrites were selected based on their structural identity that appears elongated along anterior-posterior axis (Kitamura and Häusser, 2011; Ozden et al., 2009; Schultz et al., 2009) and signal quality. Over-segmented ROIs were merged in Suite2p. For all analyses of CF and PC Ca2+ signals, we only included resting-state data and excluded the images during locomotion as we are interested in sensory processing, and it generates many complex Ca2+ transients.

Two-photon dual-color Ca2+ imaging and analysis

Request a detailed protocol

For simultaneous imaging of the CF and PC, AAV1.Camk2.GCaMP6f.WPRE.SV40 was first bilaterally injected into the IO; after 1–3 days, AAV1.CAG.FLEX.jRGECO1a.WPRE.SV40 was then injected into the cerebellar cortex in Pcp2-cre TG mice while creating a chronic imaging window. The cerebellar cortex was excited with a 1000 nm wavelength, and the signals for GFP and RFP filters were simultaneously acquired. The CF varicosity and the PC dendrite were regarded as a pair when they spatially overlap and their signals are significantly similar (Figure 4b,c). To examine the relationship between CF and PC, we performed correlation analysis of the signals and of paired amplitudes. For the signal-correlation analysis, we computed the correlation coefficient between signals of each pair with MATLAB. If PC dendrites were spatially overlapped with CF, they were considered ‘paired’ CF-PC. The CF and the PC dendrites that were not overlapped were regarded as ‘unpaired’ CF-PC, which were further categorized as ‘neighboring unpaired’ if their boarders meet within 30 µm and ‘distant unpaired’ if the borders were separated by more than 30 µm. The signal correlations were compared in the four conditions (Figure 4d). For the amplitude-correlation analysis, the paired CF and PC amplitudes were collected where CF Ca2+ events occurred. All paired amplitudes from 1 min recordings were subjected to a non-linear fit analysis using GraphPad Prism (GraphPad Software Inc, CA, USA), and R2 values and slope of the fit was presented (Figure 4e–g).

Air-puff sensory stimulation, motion tracking, and correlation with Ca2+

Request a detailed protocol

Animal sensory stimulation and motion tracking were controlled by a custom-written program in LabVIEW (National Instruments, USA). Briefly, the sensory stimulation and motion tracking were synchronized with two-photon imaging through a trigger generated by the program. Periocular air puffs lasting 30 ms were delivered with a Pneumatic Picopump (WPI, USA) at 5 s intervals, at 20 (P1) or 50 (P2) psi. Animal motion was acquired at 64 Hz using a high-speed CCD camera (IPX-VGA210, IMPERX, USA) under infrared (IR) illumination (DR4-56R-IR85, LVS, S. Korea). Motion tracking was achieved by tracking the IR-reflective patches (4 mm diameter) attached to the mice’s forepaws. The velocity of the patch calculated on two dimensions (X and Y axes) was considered to be the animals’ velocity of motion. Small forepaw movements induced by air-puff stimulation were successfully tracked (Figure 3g). To analyze the correlation between motion strength and CF Ca2+ amplitude, the peak speeds were acquired within 200 ms after the onset of spontaneous CF Ca2+ signal (for correlation during rest) or air-puff stimulation (for correlation during air puffs) (Figure 3f–h). The peak motion speed and CF Ca2+ amplitudes were subjected to linear regression analysis using GraphPad Prism (GraphPad Software Inc, CA, USA), and R2 values were presented.

Ex vivo slice electrophysiology and Ca2+ imaging

Request a detailed protocol

Acute preparation and an electrophysiological experiment were carried out as previously described (Ryu et al., 2017). Briefly, 5- to 9-week-old mice were anesthetized by isoflurane and decapitated. Then, 250 μm thick sagittal slices of the cerebellar vermis were obtained from mice using a vibratome (VT1200, Leica). The ice-cold cutting solution contained 75 mM sucrose, 75 mM NaCl, 2.5 mM KCl, 7 mM MgCl2, 0.5 mM CaCl2, 1.25 mM NaH2PO4, 26 mM NaHCO3, and 25 mM glucose with bubbled 95% O2 and 5% CO2. The slices were immediately moved to artificial cerebrospinal fluid (ACSF) containing 125 mM NaCl, 2.5 mM KCl, 1 mM MgCl2, 2 mM CaCl2, 1.25 mM NaH2PO4, 26 mM NaHCO3, and 10 mM glucose with bubbled 95% O2 and 5% CO2. Then, they were recovered at 32°C for 30 min and at room temperature for 1 hr. All of the recordings were performed within 8 hr of recovery.

The brain slices were placed in a submerged chamber with perfusion of ACSF for at least 10 min before recording. Whole-cell recordings were made at 29.5–30°C. We used recording pipettes (3–4 MΩ) filled with (in mM): 9 KCl, 10 KOH, 120 K-gluconate, 3.48 MgCl2, 10 HEPES, 4 NaCl, 4 Na2ATP, 0.4 Na3GTP, and 17.5 sucrose (pH 7.25). Electrophysiological data were acquired using an EPC9 patch-clamp amplifier (HEKA Elektronik) and PatchMaster software (HEKA Elektronik) with a sampling frequency of 20 kHz, and the signals were filtered at 2 kHz. All of the electrophysiological recordings were acquired in lobule III–V of the cerebellar central vermis. The CFs were stimulated by ACSF-containing glass pipettes placed onto the granule cell layer.

For CF Ca2+ imaging, slices were prepared from mice that had AAV1.Camk2a.GCaMP6f.WPRE.SV40 injected into their IO 3–4 weeks before the preparation. For PC Ca2+ imaging with CS recording, the low-affinity Ca2+ dye Fluo-5F (0.5 mM, F14221, Molecular Probes) was loaded from recording pipettes into the PC. A microscope (BX50W, Olympus) was equipped with a 40X objective lens (LUMPlanFLN, Olympus). Images were acquired at 5 Hz with a scientific CMOS camera (Prime, Photometrics). All of the experiments were performed in duplicate or triplicate, and randomly selected traces of each cell were analyzed.

Statistics

Graph plotting and statistical analysis were carried out with GraphPad Prism (GraphPad Software Inc, CA). The hypothesis was tested by one-way ANOVA followed by post hoc Tukey’s test, for multiple comparisons using either GraphPad Prism or MATLAB. Unpaired t-tests between sample pairs were carried out. The correlation between Ca2+ signals and motion was analyzed by linear regression using GraphPad Prism. Results were considered significant if the P-value was below 0.05. Asterisks denoted in the graph indicate the statistical significance. * means p-value<0.05, **<0.01, ***<0.001, and ****<0.0001. The test name and statistical values are presented in each figure legend.

References

  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
  7. 7
  8. 8
  9. 9
  10. 10
  11. 11
  12. 12
  13. 13
  14. 14
  15. 15
  16. 16
  17. 17
  18. 18
  19. 19
  20. 20
  21. 21
  22. 22
  23. 23
  24. 24
  25. 25
  26. 26
  27. 27
  28. 28
  29. 29
  30. 30
  31. 31
  32. 32
  33. 33
  34. 34
  35. 35
    Differential regulation of purkinje cell output between cerebellar compartments
    1. SE Roh
    2. SH Kim
    3. YG Kim
    4. CE Kim
    5. SK Kim
    6. SJ Kim
    (2017)
    Japan Neuroscience Society 10:a2–a3.
  36. 36
  37. 37
  38. 38
  39. 39
  40. 40
  41. 41
  42. 42
  43. 43

Decision letter

  1. Megan R Carey
    Reviewing Editor; Champalimaud Foundation, Portugal
  2. Ronald L Calabrese
    Senior Editor; Emory University, United States

In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.

Acceptance summary:

In this study, Roh et al. investigate graded signaling between climbing fibers and cerebellar Purkinje cells. While classical theories of cerebellar learning treat climbing fiber input as an all-or-none binary signal, recent work has questioned this doctrine by demonstrating graded climbing fiber responses recorded postsynaptically at Purkinje cells. Such graded responses would change current thinking about how climbing fibers drive both synaptic plasticity and motor learning. Currently it remains unresolved how such graded responses are generated. Here the authors present evidence that the number of spikes in high frequency burst firing of olivary cells could be the main determinant of postsynaptic calcium influx amplitude in Purkinje cells. The authors perform simultaneous calcium recordings of CF varicosities and of their postsynaptic Purkinje cells in vivo during behavior and describe a strong correlation in amplitude between presynaptic and postsynaptic signals, suggesting that presynaptic burst size may control the postsynaptic response. Further, as has been reported for postsynaptic PC responses, the authors report that climbing fiber inputs represent levels of sensory stimulation in a graded manner.

Decision letter after peer review:

Thank you for submitting your article "Direct translation of climbing fiber burst-mediated sensory coding into post-synaptic purkinje cell dendritic calcium" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Ronald Calabrese as the Senior Editor. The reviewers have opted to remain anonymous.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

In this study, Roh et al. investigate whether and how cerebellar climbing fibers can encode sensory input in a graded manner. This question is crucial to understanding the mechanisms of cerebellar learning, because climbing fibers produce an instructional signal that can drive both heterosynaptic plasticity at cerebellar Purkinje cell synapses as well as some forms of motor learning. While classical theories of cerebellar learning treat climbing fiber input as an all-or-none binary signal, recent work has questioned this doctrine by demonstrating graded climbing fiber responses recorded postsynaptically at Purkinje cells, and currently it remains unresolved how such graded responses are generated.

Here the authors propose that the number of spikes in high frequency burst firing of olivary cells could be the main determinant of postsynaptic calcium influx amplitude in Purkinje cells. If demonstrated, this would constitute a new and exciting result with deep consequence on our understanding of cerebellar supervised learning. The authors perform simultaneous calcium recordings of CF varicosities and of their postsynaptic Purkinje cells in vivo during behavior and describe a strong correlation in amplitude between presynaptic and postsynaptic signals, suggesting that presynaptic burst size may control the postsynaptic response.

The authors show nicely that, as has been reported for postsynaptic PC responses, climbing fiber inputs represent levels of sensory stimulation in a graded manner. The major conclusion of this study is that the variability in Purkinje cell dendritic calcium signal amplitude and magnitude/duration of complex spikes are a direct result of variability in the size of climbing fiber input in vivo during sensory processing. While none of the data presented in this paper contradict this claim, the data presented are all either partial or circumstantial in support of this conclusion. Moreover, there are substantial technical concerns regarding the imaging analysis and data interpretation. If the authors can address these concerns to convincingly demonstrate that the CF Ca2+ transients are governed by the number of spikes in CF, and that these signals truly co-vary with the postsynaptic response, the reviewers believe the paper would be of significant interest.

Essential revisions:

1) The quantification of calcium imaging data

a) The legend of Figure 1 states that "independent ROIs" were selected. It is unclear how ROI selection was made for CF boutons – was this manual? And how were ROI's clustered into independent CFs, i.e., how was independence determined? A single climbing fiber makes many varicosities and innervates multiple Purkinje cells. This figure quantifies both climbing fiber firing frequency and synchrony, measurements that rely on measurements from unique climbing fibers. A simple spatial segregation of ROIs may not be sufficient to establish such independence. Observation of Figure 1C-E suggests that several of the ROIs actually correspond to the same climbing fiber, given their near perfect correlation. In particular, it looks like some ROIs that were analyzed separately may correspond to a single CF e.g. ROIs 2-3 and 4-5 in Figure 1. Related to this point, it is unclear why the correlation matrix heatmap in Figure 1E is smoothed. This is both confusing and misleading.

b) They state "Hence the results indicate that the variability of CF Ca2+ activity arises from its differential spike number of burst". I assume that the authors mean by this that the variability in CF Ca2+ amplitude is mainly driven by the number of spikes in the olivary response underlying it. Figure 1G is where they show the distribution of CF Ca2+ amplitudes in vivo. One presumes that this plot is obtained by aggregating all the data from the 60 cells they recorded. Therefore it stands to reason that a lot of the variability in the amplitudes is actually driven by cell to cell variation rather than the number of spikes in the burst. Indeed, this suspicion is supported by Figure 2D, where it is obvious that CF calcium amplitudes vary widely from cell to cell for a fixed number of spikes in the burst. To strengthen their conclusions, they should analyse their in-vivo data differently: rather than lumping all the Ca2+ responses from all the cells together, have they tried plotting amplitude histograms for each individual recording? If they could show separate peaks in such a histogram, that would represent quite convincing evidence that the amplitude of their CF Ca2+ responses are being driven by the number of spikes in the olivary burst. They could consider getting a population histogram by normalizing each cell's data to the first peak in its histogram.

c) The authors report correlation coefficient values for CF boutons in Figure 1H, but it is unclear over what spatial extent these were calculated. It would be more useful to report these as a function of mediolateral distance. An important, and easily testable prediction would be that these correlations should fall off at a similar spatial scale as Purkinje cell dendritic calcium signals (as in Ozden et al., J. Neuroscience, 2009, Figure 1D or Gaffield et al., J. Neurophysiology, 2016, Figure 4B). This sort of analysis may also provide a useful metric for defining CF ROIs that correspond to the same olivary neuron, as these are likely to exhibit higher correlations that may be represented as outliers to distribution that decays smoothly as a function of space (as observed for Purkinje cell calcium imaging).

d) There are now a variety of freely available software suites, such as CaImAn (https://elifesciences.org/articles/38173) or Suite2p (https://www.biorxiv.org/content/10.1101/061507v2), that allow for rigorous and standardized selection, curation, and analysis of imaging data. It may be advisable to use such tools for data analysis.

e) There are also no examples of how event detection was performed on the imaging data. It would be useful to show example traces with detected events. This is particularly important in Figure 4, where correspondence between CF imaging and PC imaging is necessary. Similar to the point above about image segmentation, there are also a variety of freely-available software suites that allow for event detection. It may be useful to use these. The reported methods state that signals were detected using a single threshold, but event detection algorithms take indicator kinetics into account to detect events with a particular shape (i.e. fast rise, slow decay) that match the likely waveforms of real events. In particular, I would recommend MLspike, which is well-suited for detecting sparse events that may overlap in time (https://github.com/MLspike/spikes).

2) Throughout the manuscript, there is a systematic lack of reporting of how many animals were used, how many repetitions were performed for different sets of experiments, and how group statistics were calculated:

a) Figure 1F-H: please report how many ROIs, how many fields of view, and how many mice were used in these analyses (and if any ROIs were resampled. The only mention of the number of samples here is regarding Figure 1H, and it states that the data come from 10 recordings (without mentioning the number of ROIs per recording, the number of recordings per animal, or the number of animals).

b) Figure 2: the data appear to be pooled from 9 CF imaged in 3 mice and the mean(?) response for each CF is shown as a thin line in Figure 2C, while the mean response across CFs is shown as a thick red line. It is unclear how many repetitions were performed per imaged CF. Furthermore, it is unclear whether a single, small, round ROI (i.e. a single bouton) was chosen as an ROI for each CF or whether multiple boutons were averaged per CF. Finally, it appears that a statistical comparison was only made between the single stimulation group and the other conditions. Are the other groups (i.e. 3, 5, 7, and 9 stimulations) different from each other?

c) Figure 3F-H: it is unclear how many CF boutons were in to the reported number of events. Given that there are several “event data points” (y-axis) for each "speed data point” (x-axis), it appears that multiple CFs were imaged simultaneously. This should be made explicit.

d) Figure 4: it is unclear how many CF-PC pairs went into the 184 “events” that are reported in panel D. It is also not clear if CF or PC transients were used to define these “events.” The example Purkinje cell ROI in panel B is also inconsistent in size and shape with the criteria reported in the manuscript's methods. Analyzing these data carefully is crucial, since the main conceptual advance of the paper is to show that CF signal amplitudes co-vary systematically with postsynaptic PC signal amplitudes.

e) The main observation of the paper, the correlation in Figure 4D, raises many questions. First, only 184 events are plotted, corresponding to about 3 minutes of recording overall. This is a very small number of observations for 3 mice, as many varicosity/dendrite appositions should be visible in a single field of view. Most importantly, the number of independent pre/post pairs is not stated and all the data are pooled. Thus co-variation in signal intensity between pairs could be mistaken for co-variation within a pair. Indeed, Figure 2C shows a ten-fold variation in DF/F from one CF to the other, which should obscure the correlation in Figure 4D. Astonishingly, in vitro transients saturate at about 50 % DF/F (Figure 1C) for long bursts, while in vivo calcium transients in varicosities range from 80 % to 700 % (Figure 1D,G and 4C,D). All these issues should be addressed and in particular correlation should be established pair by pair, with many more events, not for the whole set of data.

f) Figure 5: There is no mention of the number of repetitions that were performed per cell here.

3) Small structures such as the varicosities imaged here can easily move into and out of the focal plane when there is subtle brain movement, producing artificially larger DF/F responses. There is no mention in the methods or figures of how motion correction was applied to bouton imaging data, which is likely to contain motion artifacts because the animals were awake and locomoting and the imaged structures are quite small. Was such correction done? If so, how? This is particularly striking, given how stable the baseline appears in Figures 1D, 3B-C If any transient movement (not necessarily the paw movement) is correlated with the intensity of stimulation, this could produce the “graded” responses in Figure 3. Movement artifacts could be assessed by imaging GFP in climbing fiber varicosities during stimulation, or by demonstrating that any other stimulus-induced movement is correlated in amplitude with paw movement. Related to this point, it is not clear to me what is plotted in Figure 3 F and H.

4) The contribution of movement to “sensory driven” responses is not sufficiently characterized.

a) The authors use IR tracking of a forepaw to reveal animal movement, but a periocular air puff is likely to produce orofacial movements not captured by paw tracking. The authors show no correlation between the amplitude of paw movements and peak calcium transients, but paw motion may not be the relevant movement that enhances climbing fiber responses in this recording location. The authors should demonstrate that the amplitude of paw movements correlates with the amplitude of any other movements generated by the air puff, or if that is not possible, at the very least, allow for this possible confound in their data.

b) If I understand Figure 3 G correctly, a transient paw movement is triggered by air puff stimulation. Is this distinct from the movement plotted in Figure 3 F and H, which appears to be a peak paw speed associated with running at the time of stimulation in H, and in absence of stimulation in F? Clearly in F, there is no stimulation, so paw movements must be related to running? In any case, it is not evident that running speed in panel F and the transient (unexpected) paw movement generated by puffs are the same. If not, there may be a correlation between the transient paw movement and climbing fiber activity independent of running.

5) The literature demonstrating the influence of postsynaptic factors on the CF calcium transient in Purkinje cells (Kitamura and Hausser 2011, Otsu et al. 2014, Gaffield et al. 2018 to cite only a few recent) is mostly overlooked. Similarly, presynaptic modulations able to change presynaptic calcium influx in CF varicosities and therefore glutamate release should be discussed, as a possible additional source of variability on top of the number of spikes in bursts.

There are several assertions made in the paper regarding the relationship between CF bursts (action potentials) and CF axonal calcium responses. While portions of the statements made by the authors are true, they sometimes apply faulty logic to arrive at their conclusions. Specifically:

a) “ Our data indicate that increasing the number of burst stimulation

significantly augments the CF Ca2+ amplitude, which is saturated at 7 bursts (Figure 2C,D).”: – While it is true (as the authors show) that increasing the stim number in a burst increases the CF Ca2+ signal amplitude, this does not definitively prove their next sentence “…the variability CF Ca2+ activity arises from its differential spike number of burst.” Just because the first statement is true, does not mean the second one (its converse) is also true.

b) Results related to Figure 5 – The title of this section is “The spike number of CF burst determines the amplitudes of PC dendritic Ca2+ response.” While their data show that these events are correlated, this statement is not uniquely true. For example, Wang et al. (Nature Neuro, 2000), Kitamura and Hausser (J. Neuroscience, 2011), and others have demonstrated that PC dendritic calcium signals are related to the interactions between PF inputs, CF inputs, and the intrinsic properties of PCs. Thus, the authors' statement is not uniquely true.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for submitting your work entitled "Direct Translation of Climbing Fiber Burst-Mediated Sensory Coding into Post-Synaptic Purkinje Cell Dendritic Calcium" for consideration by eLife. Your article has been reviewed by two peer reviewers, and the evaluation has been overseen by a Reviewing Editor and a Senior Editor. The reviewers have opted to remain anonymous.

Our decision has been reached after consultation between the reviewers. Based on these discussions and the individual reviews below, we regret to inform you that your work will not be considered further for publication in eLife.

Although the reviewers were appreciative of your efforts to improve the manuscript with this revision, it was agreed in the consultation phase that the original technical concerns persist and have not been sufficiently overcome. In particular, the issues raised by reviewer 1 about ROI detection, quantification of event amplitudes, and the identification of CF-PC pairs are seen as critical and were not perceived to be adequately defended in the manuscript. Given these concerns, together with the lack of significant differences between the unpaired neighbors and the paired neighbors, the reviewers do not believe that the conclusion that climbing fiber (CF) high frequency bursts are responsible for the graded amplitude of the postsynaptic calcium transients in the Purkinje cells (PCs) is adequately supported by the data.

Reviewer #1:

In the revised version of their paper, the authors have performed an extensive new analysis of their dataset using state of the art methods. Data are quantified and displayed in a more detailed way and the paper is overall substantially improved. However, the results still fail to support convincingly the main conclusion of the paper, namely that climbing fiber (CF) high frequency bursts are responsible for the graded amplitude of the postsynaptic calcium transients in the Purkinje cells (PCs). The identification of individual CFs is problematic, and equating CF calcium transient amplitude with CF firing bursts remains correlative. Furthermore, the claim that CF-PC pairs are recorded does not seem warranted based on the data provided. In conclusion the correlations between CF and PC may result from other factors than the proposed burst coding.

1) The main issue still arises from the identification of CF ROIs and CF-PC pairs, a problem raised in the first round of reviews. The additional data provided in the revised version strongly suggest that CF ROIs defined by the authors encompass multiple CFs.

a) Morphological arguments

All data from the literature that traced the projections of single olivary cells point to the fact that the target Purkinje cells are not clustered at a given antero-posterior position but randomly spread in a thin parasagittal band. Therefore, each climbing fiber should only encompass a few varicosities in strict parasagittal alignment within a single PC dendrite, much thinner than the average 30 µm extension described in the paper.

b) Functional arguments

Functional studies on PC calcium transients indicate that PC cells displaying calcium transients with close to 100 % synchrony (hence receiving the same CF) are the exception rather than the rule (Ozden et al. 2009). Careful observation of the movies in the present dataset indicates that the only varicosities that display 100 % synchrony in their calcium transients are indeed strictly in parasagital alignment, hence contacting a single Purkinje cell.

c) In conclusion, the authors should detect events on single varicosities, which are easily identified manually and then check for 100% correlation to group varicosities in a single CF. It is hard to understand why the automatic algorithm defines such contorted ROIs covering large regions with no detectable signal. High level of correlation between nearby CF and poor signal to noise levels (Figure 1—figure supplement 1B) may explain the pooling of multiple CFs in a single ROI.

2) The improper identification of single climbing fibers affects the interpretation of the supposed paired recordings of Figure 4. CF1 in Figure 4B and D seems to be two fibers, one contacting PC1 and the other PC3. However, correlation with PC2, on which there is no contact appears as good, probably because PC1 and PC2 are themselves highly correlated. More intriguing, some calcium events which are obvious in PCs 1-3 of the color raster of Figure 4D do not show in the CF1. This may indicate that this(ese) CF do not actually contact these PCs or that detection of CF events is improper. All calcium transients seen in the PC should be found in the CF and conversely.

Panels 4G-H provide a much improved analysis of the paired data, based on individual CF-PC pairs. However, considering that CF ROIs cover multiple CFs, they can be interpreted in terms of CF population synchrony. Larger CF transients would correspond to multiple CF being active at the same time, which would in turn translate in larger PC transients, as the link between population synchrony and PC calcium transients has already been described by Najafi et al. (2014). This does not preclude burst coding to be participating to the graded encoding of synchrony, as larger errors may both stimulate more olivary cells and trigger bursts, but the data do not convincingly demonstrate this point. In particular the CF-PC correlation is not significantly different between paired and neighbor unpaired, in line with the interpretation that correlation results from populational synchrony at short distance and not from direct pairing.

3) Quantification of the transients is also a problem. It is entirely unclear how events are detected after non-negative deconvolution and denoising (which may be subject to caution given the signal to noise of the recordings and the possibility of overfitting, see Figure 1—figure supplement 1B). Failure to detect close-by events may arise in the report of artificially large event amplitudes. When multiple events are detected does the amplitude correspond to the summed peak of the DF/F trace or to the actual individual events amplitudes?

The very large amplitude of the CF transients (more than 2000 % in Figure 1) is surprising, as it may exceed the dynamic range of GCaMP6, particularly considering that ROIs cover large regions without any signal. Baseline subtraction, which may explain these large numbers if excess subtraction is applied, appears to be a recurrent issue in the paper with some panels displaying very low background noise (RFP in Figure 4 supplement 1a) and other very high (GFP in the same figure and in panel 1B).

Furthermore, the signals appear saturated in many of the figures as well as in the movies, which does not really allow to judge the dynamic range and shape of the activated elements. The two movies have compression issues which do not allow clear visualization on a frame by frame mode.

Finally, data of Figure 4 and the two movies originate from the same field of view and the two movies represent the same 20s but with very different levels of background subtraction and saturation. It is hard to understand why the same data are presented twice, and why not provide a more extensive sample of the data. Is it on purpose? If so, it should be explicitly mentioned.

Reviewer #2:

Roh et al. have revised their manuscript to sufficiently account for my previous concerns, most importantly by re-analyzing all of their calcium imaging data and enhancing the transparency of their data and methodological reporting.

Notably, I still have some concerns over the role of movement in the amplitude of the reported signals. The authors correlate DF/F with paw speed, but not the amplitude of paw movement, or any other movements the animals make in response to airpuff stimulation. However, they now perform additional controls suggesting that imaging artifacts from bouton movement do not contribute to their results, and they have tempered their conclusions regarding the role of movement appropriately. Given the challenging nature of this problem, the level with which the authors have now addressed the role of movement seems reasonable. Likewise, the enhanced analysis of “independent” CF ROIs remains less than airtight, but in my opinion now meets a reasonable standard.

Finally, there remain some typos, and some odd writing choices. Examples include statements like "this notion has been prevailed", and "sensory inputs disappear ex vivo". It is also atypical to end a manuscript with a final sentence telling the reader what the lab is working on next, rather than an overall statement about the significance of this study.

Overall, however, the manuscript now convincingly makes an important link between the presynaptic activity of climbing fibers and the postsynaptic calcium transient in Purkinje cells in awake animals. It thus provides a significant advance, in parallel with recent work from the Christie lab, to our understanding of cerebellar learning mechanisms.

https://doi.org/10.7554/eLife.61593.sa1

Author response

Essential revisions:

1) The quantification of calcium imaging data

We agree with the reviewers’ point that our data analysis and interpretation should maintain objectivity in terms of selecting ROIs. Here we have utilized CaImAn, the open source software tool to detect single CFs (Giovannucci et al., 2019). It includes non-rigid motion correction, constrained nonnegative matrix factorization (CNMF) algorithm for source extraction and sparse non-negative deconvolution for event detection.

a) The legend of Figure 1 states that "independent ROIs" were selected. It is unclear how ROI selection was made for CF boutons – was this manual? And how were ROI's clustered into independent CFs, i.e., how was independence determined? A single climbing fiber makes many varicosities and innervates multiple Purkinje cells. This figure quantifies both climbing fiber firing frequency and synchrony, measurements that rely on measurements from unique climbing fibers. A simple spatial segregation of ROIs may not be sufficient to establish such independence. Observation of Figure 1C-E suggests that several of the ROIs actually correspond to the same climbing fiber, given their near perfect correlation. In particular, it looks like some ROIs that were analyzed separately may correspond to a single CF e.g. ROIs 2-3 and 4-5 in Figure 1. Related to this point, it is unclear why the correlation matrix heatmap in Figure 1E is smoothed. This is both confusing and misleading.

As CaImAn was designed to be suitable for soma calcium signal detection, we slightly modified the analysis protocol by selecting the initialization method as “sparse NMF,” a method that is appropriate for detecting arbitrary structures such as axons rather than circular soma. Processing CF calcium imaging data with CaImAn gave many ROIs that are overlapping each other, so we performed manual curation of ROIs (4 out of 12 ROIs) and the ones overlapping or with bad signal were excluded by a human expert. This process was depicted in Figure 1—figure supplement 1. Independent CFs were well identified as axonal arborization structures that have mediolateral size of about 30.2 µm in average (Figure 1C, D, G) and their synchrony are moderately reduced with mediolateral separation (Figure 1E, F). Thus, we have successfully detected independent CFs and their calcium signals using the modified CaImAn.

b) They state "Hence the results indicate that the variability of CF Ca2+ activity arises from its differential spike number of burst". I assume that the authors mean by this that the variability in CF Ca2+ amplitude is mainly driven by the number of spikes in the olivary response underlying it. Figure 1G is where they show the distribution of CF Ca2+ amplitudes in vivo. One presumes that this plot is obtained by aggregating all the data from the 60 cells they recorded. Therefore it stands to reason that a lot of the variability in the amplitudes is actually driven by cell to cell variation rather than the number of spikes in the burst. Indeed, this suspicion is supported by Figure 2D, where it is obvious that CF calcium amplitudes vary widely from cell to cell for a fixed number of spikes in the burst. To strengthen their conclusions, they should analyse their in-vivo data differently: rather than lumping all the Ca2+ responses from all the cells together, have they tried plotting amplitude histograms for each individual recording? If they could show separate peaks in such a histogram, that would represent quite convincing evidence that the amplitude of their CF Ca2+ responses are being driven by the number of spikes in the olivary burst. They could consider getting a population histogram by normalizing each cell's data to the first peak in its histogram.

According to the reviewer’s suggestion, we have plotted amplitude histograms for each individual 3 minute recording for 11 cells (Figure 2D, E). As calcium spikes closely following the 1st event will present an additive effect on amplitudes, only events within the first 0.5 seconds were used for analysis, as shown in Figure 2D. The amplitude histogram for the 11 cells shows a “discrete” distribution, suggesting several clusters of events with similar amplitudes (Figure 2E). These data would suggest that CF calcium responses are driven by the number of spikes in the olivary burst.

c) The authors report correlation coefficient values for CF boutons in Figure 1H, but it is unclear over what spatial extent these were calculated. It would be more useful to report these as a function of mediolateral distance. An important, and easily testable prediction would be that these correlations should fall off at a similar spatial scale as Purkinje cell dendritic calcium signals (as in Ozden et al., J. Neuroscience, 2009, Figure 1D or Gaffield et al., J. Neurophysiology, 2016, Figure 4B). This sort of analysis may also provide a useful metric for defining CF ROIs that correspond to the same olivary neuron, as these are likely to exhibit higher correlations that may be represented as outliers to distribution that decays smoothly as a function of space (as observed for Purkinje cell calcium imaging).

We have computed synchrony values between CFs in terms of mediolateral distance (Figure 1F). Similar to the previous results from the Purkinje cell calcium imaging, the correlation decreased as a function of distance. The identification of CF ROIs was carried out by clustering pixels that fire together using the CaImAn CNMF algorithm. The average mediolateral width of CF ROIs was 30.2 ± 2.9 (SEM) µm with the largest at 75.8 µm.

d) There are now a variety of freely available software suites, such as CaImAn (https://elifesciences.org/articles/38173) or Suite2p (https://www.biorxiv.org/content/10.1101/061507v2), that allow for rigorous and standardized selection, curation, and analysis of imaging data. It may be advisable to use such tools for data analysis.

As mentioned above, we have successfully adapted CaImAn to CF calcium signal analysis in combination with human expert curation for selecting ROIs out of a suggested ROI set. We have uploaded the code that runs CaImAn for CF Ca2+ signal analysis including the parameters as well as the code that performs curation to

GitHub repository (https://github.com/NeuRoh1/Calcium-analysis-tools).

e) There are also no examples of how event detection was performed on the imaging data. It would be useful to show example traces with detected events. This is particularly important in Figure 4, where correspondence between CF imaging and PC imaging is necessary. Similar to the point above about image segmentation, there are also a variety of freely-available software suites that allow for event detection. It may be useful to use these. The reported methods state that signals were detected using a single threshold, but event detection algorithms take indicator kinetics into account to detect events with a particular shape (i.e. fast rise, slow decay) that match the likely waveforms of real events. In particular, I would recommend MLspike, which is well-suited for detecting sparse events that may overlap in time (https://github.com/MLspike/spikes).

In this revision, we have utilized the non-negative deconvolution method (Vogelstein et al., 2010), which has been used for PC Ca2+ event detection (Gaffield et al., 2016), as well as CF and PC Ca2+ event detection (CaImAn also utilizes this algorithm). As shown in Figure 1—figure supplement 1b and Figure 3—figure supplement 2c, event detection was successful for each cell type. The event detection of CF-PC pairs from dual imaging also reveals a high degree of correspondence (Figure 4C).

As for PC Ca2+ analysis, we first performed non-rigid motion correction (NoRMCorre), and manually selected PC dendrites based on the structural uniformity and signal properties (human expert). The event detection was performed using sparse non-negative deconvolution algorithm (Vogelstein et al., 2010). For the source extraction, we’ve tried principle component analysis followed by independent component analysis (PCA/ICA) (Mukamel et al., 2009). However, the signal quality has no advantages against human expert analysis supported by NoRMCorre and deconvolution (Author response image 1). Also, as PCA/ICA processing misses some ROIs, we proceeded with human expert ROI selection (Figure 3—figure supplement 2).

Author response image 1

2) Throughout the manuscript, there is a systematic lack of reporting of how many animals were used, how many repetitions were performed for different sets of experiments, and how group statistics were calculated:

a) Figure 1F-H: please report how many ROIs, how many fields of view, and how many mice were used in these analyses (and if any ROIs were resampled. The only mention of the number of samples here is regarding Figure 1H, and it states that the data come from 10 recordings (without mentioning the number of ROIs per recording, the number of recordings per animal, or the number of animals).

The number of ROIs, FOV, and recordings used for analysis are now all clearly stated in the figure legend (Figure 1F,G).

b) Figure 2: the data appear to be pooled from 9 CF imaged in 3 mice and the mean(?) response for each CF is shown as a thin line in Figure 2C, while the mean response across CFs is shown as a thick red line. It is unclear how many repetitions were performed per imaged CF. Furthermore, it is unclear whether a single, small, round ROI (i.e. a single bouton) was chosen as an ROI for each CF or whether multiple boutons were averaged per CF. Finally, it appears that a statistical comparison was only made between the single stimulation group and the other conditions. Are the other groups (i.e. 3, 5, 7, and 9 stimulations) different from each other?

Five round CF axonal varicosities of < 2 µm that respond to stimuli were chosen as ROIs and then averaged for a single CF trace. 1 or 2 repetitions were performed and traces were randomly selected for analysis. This method is described in the Materials and methods section. In addition, all the significant comparisons are presented in the figure and figure legend (between 1 vs 5, 1 vs 7, 1 vs 9 and 3 vs 9 stim).

c) Figure 3F-H: it is unclear how many CF boutons were in to the reported number of events. Given that there are several “event data points” (y-axis) for each "speed data point” (x-axis), it appears that multiple CFs were imaged simultaneously. This should be made explicit.

We have reanalyzed the CF calcium imaging data with CaImAn and plotted the correlation graphs again. The number of events, recordings, animals are clearly presented in Figure 3 legend. The number of CFs is not reported since the number varies (3 – 6 CFs).

d) Figure 4: it is unclear how many CF-PC pairs went into the 184 “events” that are reported in panel D. It is also not clear if CF or PC transients were used to define these “events.” The example Purkinje cell ROI in panel B is also inconsistent in size and shape with the criteria reported in the manuscript's methods. Analyzing these data carefully is crucial, since the main conceptual advance of the paper is to show that CF signal amplitudes co-vary systematically with postsynaptic PC signal amplitudes.

We take this comment very seriously and as such performed an in-depth correlation analysis between CF and PC Ca2+ signals: this included signal correlation as well as amplitude correlation analysis. We first identified CF signals by CaImAn and selected PC dendrites within the established CF territories. These pairs are regarded as “paired” CF-PC. When the PC dendrites are out of the CF territories, they are regarded as “unpaired” CF-PC (Figure 4B-D). Such “unpaired” CF-PC were further categorized as “neighboring unpaired” if CFs’ borders are touching each other and “distant unpaired” if the borders are at least 30 µm apart. Here, we report highest signal correlation in paired CF-PC with reduced correlation depending on mediolateral distance (Figure 4C,D). For the amplitude correlation analysis, CF and PC amplitudes were used where CF Ca2+ transients take place. The number of events, pairs, recording session, and animal numbers are clearly stated in the figure legend (Figure 4).

e) The main observation of the paper, the correlation in Figure 4D, raises many questions. First, only 184 events are plotted, corresponding to about 3 minutes of recording overall. This is a very small number of observations for 3 mice, as many varicosity/dendrite appositions should be visible in a single field of view. Most importantly, the number of independent pre/post pairs is not stated and all the data are pooled. Thus co-variation in signal intensity between pairs could be mistaken for co-variation within a pair. Indeed, Figure 2C shows a ten-fold variation in DF/F from one CF to the other, which should obscure the correlation in Figure 4D. Astonishingly, in vitro transients saturate at about 50 % DF/F (Figure 1c) for long bursts, while in vivo calcium transients in varicosities range from 80 % to 700 % (Figure 1D,G and 4C,D). All these issues should be addressed and in particular correlation should be established pair by pair, with many more events, not for the whole set of data.

As the reviewer suggested, we have established the correlation pair by pair with a larger number of event pairs (1661 pairs) in this revision (Figure 4E-G). We provide pair by pair correlation for 19 CF-PC pairs (Figure 4G), while presenting the example of correlation of single pair (Figure 4E) and 1CF-3PC pairs (Figure 4F). Although there is a discrepancy in amplitude saturation between in vitro and in vivo transients, there are apparently many differences between them, such as imaging condition (widefield imaging vs 2-photon imaging), health of sample (sliced brain tissue vs intact brain), and baseline intensity (already increased in slice vs very low baseline). We also provide a video for the CF-PC dual imaging (Video 1).

f) Figure 5: There is no mention of the number of repetitions that were performed per cell here.

Now we state that the number of experiments is “7 recordings of 7 independent cells from 3 mice” (Figure 5). 1 or 2 repetitions were performed, but only the representative data were analyzed.

3) Small structures such as the varicosities imaged here can easily move into and out of the focal plane when there is subtle brain movement, producing artificially larger DF/F responses. There is no mention in the methods or figures of how motion correction was applied to bouton imaging data, which is likely to contain motion artifacts because the animals were awake and locomoting and the imaged structures are quite small. Was such correction done? If so, how? This is particularly striking, given how stable the baseline appears in Figures 1D, 3B-C If any transient movement (not necessarily the paw movement) is correlated with the intensity of stimulation, this could produce the “graded” responses in Figure 3. Movement artifacts could be assessed by imaging GFP in climbing fiber varicosities during stimulation, or by demonstrating that any other stimulus-induced movement is correlated in amplitude with paw movement. Related to this point, it is not clear to me what is plotted in Figure 3 F and H.

In this revision, we processed and analyzed all CF data with CaImAn, which includes motion correction by “non-rigid motion correction (NoRMCorre)” (Pnevmatikakis and Giovannucci, 2017). It gives a very stable video, which is uploaded as a video file for the CF-PC dual color imaging (Video 2). Moreover, CaImAn employs a template-matching algorithm to detect calcium events with fast rise and slow decay. Also, here we present 2-photon in vivo GFP imaging results under periocular air puff stimulation. These results show that air puff-induced movement is not deforming to CF-GFP and PC-GFP for both P1 and P2 stimulation (Figure 3—figure supplement 1). We also present a video showing before and after motion correction using NoRMCorre for both CF and PC during air puff stimulation (Video 2).

Previously we did not clearly describe that the motion we observed was not locomotion or walking/running but rather a subtle twitch-like movement of the forepaw induced by air puff stimulation. We apologize for any confusion the omission of this detail may have caused. We exclude the recordings where animals walk or run because we are interested in sensory coding at rest and locomotion produces a lot of complex Ca2+ signals. Figure 3F,H presents the correlation of CF calcium amplitude and forepaw twitch movement at rest. We claim that air puff stimulation-induced CF calcium transient is not related to the peak speed of air puff-induced twitch movement. These additional details and clarifications are now included in the Results and Materials and methods sections.

4) The contribution of movement to “sensory driven” responses is not sufficiently characterized.

a) The authors use IR tracking of a forepaw to reveal animal movement, but a periocular air puff is likely to produce orofacial movements not captured by paw tracking. The authors show no correlation between the amplitude of paw movements and peak calcium transients, but paw motion may not be the relevant movement that enhances climbing fiber responses in this recording location. The authors should demonstrate that the amplitude of paw movements correlates with the amplitude of any other movements generated by the air puff, or if that is not possible, at the very least, allow for this possible confound in their data.

The movement we observed is a general movement for an unexpected stimulus. Although we have observed that eyeblink and forepaw movement occur simultaneously upon stimulation, we have now clearly stated the possibility of a confounding link to any other movements generated by air puff: “Although we did not examine any other movements related to air-puff stimulation, such as orofacial movement, we suggest that the CF Ca2+ activity could differentiate the graded sensory stimuli but not movement strength in an assumption that fore-paw motion is representing startle movement during air-puff stimulation.”

b) If I understand Figure 3G correctly, a transient paw movement is triggered by air puff stimulation. Is this distinct from the movement plotted in Figure 3F and H, which appears to be a peak paw speed associated with running at the time of stimulation in H, and in absence of stimulation in F? Clearly in F, there is no stimulation, so paw movements must be related to running? In any case, it is not evident that running speed in panel F and the transient (unexpected) paw movement generated by puffs are the same. If not, there may be a correlation between the transient paw movement and climbing fiber activity independent of running.

The confusion the reviewer indicated is because we previously did not state that we excluded the data with locomotion (see also our response to the above comment #3). The “spontaneous response” data in Figure 3F are limited to small and transient movements with CF Ca2+ events but not during walking or running. The peak speed is defined by the peak speed of fore-paw speed within 200 ms after the movement onset for spontaneous response (Figure 3F) or stimulation onset for air puff-induced movement (Figure 3H). These definitions are now described in the manuscript.

5) The literature demonstrating the influence of postsynaptic factors on the CF calcium transient in Purkinje cells (Kitamura and Hausser 2011, Otsu et al. 2014, Gaffield et al. 2018 to cite only a few recent) is mostly overlooked. Similarly, presynaptic modulations able to change presynaptic calcium influx in CF varicosities and therefore glutamate release should be discussed, as a possible additional source of variability on top of the number of spikes in bursts.

There are several assertions made in the paper regarding the relationship between CF bursts (action potentials) and CF axonal calcium responses. While portions of the statements made by the authors are true, they sometimes apply faulty logic to arrive at their conclusions. Specifically:

a) “ Our data indicate that increasing the number of burst stimulation

significantly augments the CF Ca2+ amplitude, which is saturated at 7 bursts (Figure 2C,D).”: While it is true (as the authors show) that increasing the stim number in a burst increases the CF Ca2+ signal amplitude, this does not definitively prove their next sentence “…the variability CF Ca2+ activity arises from its differential spike number of burst.” Just because the first statement is true, does not mean the second one (its converse) is also true.

In this revision, we provide additional evidence that in vivo CF Ca2+ amplitude distribution appears discrete, which could further support the CF spike number-driven variability of CF Ca2+ amplitudes. However, we toned down the conclusion sentence as “Hence, the huge variability in CF Ca2+ amplitude may be formed by the variable number of CF burst.”

b) Results related to Figure 5 – The title of this section is “The spike number of CF burst determines the amplitudes of PC dendritic Ca2+ response.” While their data show that these events are correlated, this statement is not uniquely true. For example, Wang et al. (Nature Neuro, 2000), Kitamura and Hausser (J. Neuroscience, 2011), and others have demonstrated that PC dendritic calcium signals are related to the interactions between PF inputs, CF inputs, and the intrinsic properties of PCs. Thus, the authors' statement is not uniquely true.

According to the reviewer’s recommendation, the section title was changed to “The spike number of CF burst directly affects the amplitudes of PC dendritic Ca2+ response.” Also, the other contributions (intrinsic property of PC, PF activity, and noradrenergic CF control) to PC Ca2+ response were discussed in the Discussion section.

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Reviewer #1:

In the revised version of their paper, the authors have performed an extensive new analysis of their dataset using state of the art methods. Data are quantified and displayed in a more detailed way and the paper is overall substantially improved. However, the results still fail to support convincingly the main conclusion of the paper, namely that climbing fiber (CF) high frequency bursts are responsible for the graded amplitude of the postsynaptic calcium transients in the Purkinje cells (PCs). The identification of individual CFs is problematic, and equating CF calcium transient amplitude with CF firing bursts remains correlative. Furthermore, the claim that CF-PC pairs are recorded does not seem warranted based on the data provided. In conclusion the correlations between CF and PC may result from other factors than the proposed burst coding.

1) The main issue still arises from the identification of CF ROIs and CF-PC pairs, a problem raised in the first round of reviews. The additional data provided in the revised version strongly suggest that CF ROIs defined by the authors encompass multiple CFs.

a) Morphological arguments

All data from the literature that traced the projections of single olivary cells point to the fact that the target Purkinje cells are not clustered at a given antero-posterior position but randomly spread in a thin parasagittal band. Therefore, each climbing fiber should only encompass a few varicosities in strict parasagittal alignment within a single PC dendrite, much thinner than the average 30 µm extension described in the paper.

b) Functional arguments

Functional studies on PC calcium transients indicate that PC cells displaying calcium transients with close to 100 % synchrony (hence receiving the same CF) are the exception rather than the rule (Ozden et al. 2009). Careful observation of the movies in the present dataset indicates that the only varicosities that display 100 % synchrony in their calcium transients are indeed strictly in parasagital alignment, hence contacting a single Purkinje cell.

c) In conclusion, the authors should detect events on single varicosities, which are easily identified manually and then check for 100% correlation to group varicosities in a single CF. It is hard to understand why the automatic algorithm defines such contorted ROIs covering large regions with no detectable signal. High level of correlation between nearby CF and poor signal to noise levels (Figure 1—figure supplement 1B) may explain the pooling of multiple CFs in a single ROI.

We admit that the identification of CF ROIs and consequently CF-PC pairs was problematic, compounding the interpretation of the results. Reviewing the literatures and data we conclude that CaImAn is not appropriate for CF Ca2+ analysis because it detects too large area (~30um) as indicated by the reviewer #1. Here, we tested and verified that another open source tool Suite2p effectively detects individual CF varicosities (Figure 1). If ROIs are located in the similar parasagittal plane (~20 um) and their correlation coefficient of signals are above 0.85, the pairs were merged in Suite2p GUI, resulting in less than 10 μm size of each ROI mediolaterally. Some CF ROIs appeared small which are less than 2-5 um, while some portion of bigger ROIs appeared parasagittaly aligned (i.e. ROI #1, 2, 7, 9, 13 in Figure 1C). Confirming that Suite2p analysis ensures proper interpretation of the results, we reanalyzed whole data set with the new method established and dispelled the concerns about CF ROI identification.

2) The improper identification of single climbing fibers affects the interpretation of the supposed paired recordings of Figure 4. CF1 in Figure 4B and D seems to be two fibers, one contacting PC1 and the other PC3. However, correlation with PC2, on which there is no contact appears as good, probably because PC1 and PC2 are themselves highly correlated. More intriguing, some calcium events which are obvious in PCs 1-3 of the color raster of Figure 4D do not show in the CF1. This may indicate that this(ese) CF do not actually contact these PCs or that detection of CF events is improper. All calcium transients seen in the PC should be found in the CF and conversely.

Panels 4G-H provide a much improved analysis of the paired data, based on individual CF-PC pairs. However, considering that CF ROIs cover multiple CFs, they can be interpreted in terms of CF population synchrony. Larger CF transients would correspond to multiple CF being active at the same time, which would in turn translate in larger PC transients, as the link between population synchrony and PC calcium transients has already been described by Najafi et al. (2014). This does not preclude burst coding to be participating to the graded encoding of synchrony, as larger errors may both stimulate more olivary cells and trigger bursts, but the data do not convincingly demonstrate this point. In particular the CF-PC correlation is not significantly different between paired and neighbor unpaired, in line with the interpretation that correlation results from populational synchrony at short distance and not from direct pairing.

The data analysis and its interpretation related to simultaneous CF-PC Ca2+ imaging were fundamentally improved by obtaining well segmented CF varicosities (less than 10um) using Suite2p. The CF and the PC were regarded as a pair when they spatially overlap and their signals are significantly similar. The pair 1 and 2 in Figure 4 present excellent examples of CF-PC pair definition as the CF is surrounding the PC dendrite and their Ca2+ activity also show robust correlation. Employing the strategy, we re-analyzed the entire data set, showing that the signal correlation and the amplitude correlation are significantly high between the CF-PC pairs. In response to the argument pointing out that CF-PC correlation is not significantly different between paired and neighbor unpaired, they now show significant difference with the data analyzed with Suite2p (Figure 4D) and it is in line with the population synchrony result that CF correlation falls as a function of mediolateral separation (Figure 1F).

3) Quantification of the transients is also a problem. It is entirely unclear how events are detected after non-negative deconvolution and denoising (which may be subject to caution given the signal to noise of the recordings and the possibility of overfitting, see Figure 1—figure supplement 1B). Failure to detect close-by events may arise in the report of artificially large event amplitudes. When multiple events are detected does the amplitude correspond to the summed peak of the DF/F trace or to the actual individual events amplitudes?

The very large amplitude of the CF transients (more than 2000 % in Figure 1) is surprising, as it may exceed the dynamic range of GCaMP6, particularly considering that ROIs cover large regions without any signal. Baseline subtraction, which may explain these large numbers if excess subtraction is applied, appears to be a recurrent issue in the paper with some panels displaying very low background noise (RFP in Figure 4—figure supplement 1A) and other very high (GFP in the same figure and in panel 1B).

Furthermore, the signals appear saturated in many of the figures as well as in the movies, which does not really allow to judge the dynamic range and shape of the activated elements. The two movies have compression issues which do not allow clear visualization on a frame by frame mode.

Finally, data of Figure 4 and the two movies originate from the same field of view and the two movies represent the same 20s but with very different levels of background subtraction and saturation. It is hard to understand why the same data are presented twice, and why not provide a more extensive sample of the data. Is it on purpose? If so, it should be explicitly mentioned.

In the new manuscript the Ca2+ event detection was performed with the state-of-the-art built-in deconvolution of Suite2p. The extracted Ca2+ signals were subtracted with background signals and then z-score normalized to avoid bias in amplitude quantification. The issue related to surprisingly large amplitudes of CF transients when CaImAn is employed was resolved with this approach (Suite2p and z-score normalization). Although the reviewer raised concerns regarding the saturation of figures and video, the data acquisition was carried out with high signal-to-noise level by minimizing laser power and properly adjusting gains. However, the images and movies were adjusted with contrast and brightness for clearer presentation of representative data. Please note that all analysis with Suite2p have been done with the original file. At this version of manuscript, we only present Video 1 for dual calcium imaging.

Reviewer #2:

Roh et al. have revised their manuscript to sufficiently account for my previous concerns, most importantly by re-analyzing all of their calcium imaging data and enhancing the transparency of their data and methodological reporting.

Notably, I still have some concerns over the role of movement in the amplitude of the reported signals. The authors correlate DF/F with paw speed, but not the amplitude of paw movement, or any other movements the animals make in response to airpuff stimulation. However, they now perform additional controls suggesting that imaging artifacts from bouton movement do not contribute to their results, and they have tempered their conclusions regarding the role of movement appropriately. Given the challenging nature of this problem, the level with which the authors have now addressed the role of movement seems reasonable. Likewise, the enhanced analysis of “independent” CF ROIs remains less than airtight, but in my opinion now meets a reasonable standard.

Finally, there remain some typos, and some odd writing choices. Examples include statements like "this notion has been prevailed", and "sensory inputs disappear ex vivo". It is also atypical to end a manuscript with a final sentence telling the reader what the lab is working on next, rather than an overall statement about the significance of this study.

The typos were corrected and the manuscript was edited by native speakers. The final sentence was edited to finish by overall statement of the significance of this study.

Overall, however, the manuscript now convincingly makes an important link between the presynaptic activity of climbing fibers and the postsynaptic calcium transient in Purkinje cells in awake animals. It thus provides a significant advance, in parallel with recent work from the Christie lab, to our understanding of cerebellar learning mechanisms.

https://doi.org/10.7554/eLife.61593.sa2

Article and author information

Author details

  1. Seung-Eon Roh

    1. Department of Physiology, Seoul National University College of Medicine, Seoul, Republic of Korea
    2. Department of Biomedical Sciences, Seoul National University College of Medicine, Seoul, Republic of Korea
    3. Neuroscience Research Institute, Seoul National University College of Medicine, Seoul, Republic of Korea
    4. Department of Physiology, College of Korean Medicine, Kyung Hee University, Seoul, Republic of Korea
    5. Department of Neuroscience, School of Medicine, Johns Hopkins University, Baltimore, United States
    Contribution
    Conceptualization, Resources, Data curation, Software, Formal analysis, Funding acquisition, Validation, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing
    Contributed equally with
    Seung Ha Kim
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-1222-4365
  2. Seung Ha Kim

    1. Department of Physiology, Seoul National University College of Medicine, Seoul, Republic of Korea
    2. Department of Biomedical Sciences, Seoul National University College of Medicine, Seoul, Republic of Korea
    Contribution
    Data curation, Investigation
    Contributed equally with
    Seung-Eon Roh
    Competing interests
    No competing interests declared
  3. Changhyeon Ryu

    1. Department of Physiology, Seoul National University College of Medicine, Seoul, Republic of Korea
    2. Department of Biomedical Sciences, Seoul National University College of Medicine, Seoul, Republic of Korea
    3. Neuroscience Research Institute, Seoul National University College of Medicine, Seoul, Republic of Korea
    Contribution
    Data curation, Validation, Investigation
    Contributed equally with
    Chang-Eop Kim and Yong Gyu Kim
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-5207-9142
  4. Chang-Eop Kim

    1. Department of Physiology, Seoul National University College of Medicine, Seoul, Republic of Korea
    2. Department of Physiology, College of Korean Medicine, Gacheon University, Seongnam, Republic of Korea
    Contribution
    Resources, Data curation, Software, Methodology
    Contributed equally with
    Changhyeon Ryu and Yong Gyu Kim
    Competing interests
    No competing interests declared
  5. Yong Gyu Kim

    1. Department of Physiology, Seoul National University College of Medicine, Seoul, Republic of Korea
    2. Department of Biomedical Sciences, Seoul National University College of Medicine, Seoul, Republic of Korea
    3. Neuroscience Research Institute, Seoul National University College of Medicine, Seoul, Republic of Korea
    Contribution
    Resources, Software, Methodology
    Contributed equally with
    Changhyeon Ryu and Chang-Eop Kim
    Competing interests
    No competing interests declared
  6. Paul F Worley

    Department of Neuroscience, School of Medicine, Johns Hopkins University, Baltimore, United States
    Contribution
    Supervision, Writing - review and editing
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-5086-614X
  7. Sun Kwang Kim

    Department of Physiology, College of Korean Medicine, Kyung Hee University, Seoul, Republic of Korea
    Contribution
    Conceptualization, Supervision, Funding acquisition, Writing - review and editing
    For correspondence
    skkim77@khu.ac.kr
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-2649-6652
  8. Sang Jeong Kim

    1. Department of Physiology, Seoul National University College of Medicine, Seoul, Republic of Korea
    2. Department of Biomedical Sciences, Seoul National University College of Medicine, Seoul, Republic of Korea
    3. Neuroscience Research Institute, Seoul National University College of Medicine, Seoul, Republic of Korea
    Contribution
    Conceptualization, Supervision, Funding acquisition, Writing - review and editing
    For correspondence
    sangjkim@snu.ac.kr
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0001-8931-3713

Funding

National Research Foundation of Korea (2018R1A5A2025964)

  • Sang Jeong Kim

National Research Foundation of Korea (2017M3C7A1029611)

  • Sang Jeong Kim

National Research Foundation of Korea (2016R1D1A1A02937329)

  • Sun Kwang Kim

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Acknowledgements

We thank Jeong-Kyu Han, Gee-Hoon Chung, Suk-Chan Lee, and Yong Seok Lee for their helpful discussions as well as Dong-Cheol Jang for the graphical illustrations. This work was supported by National Research Foundation of Korea (NRF) grants funded by the South Korean government (2018R1A5A2025964 and 2017M3C7A1029611 to SJ Kim; NRF-2016R1D1A1A02937329 to SK Kim).

Ethics

Animal experimentation: This study was performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. All of the animals were handled according to approved institutional animal care and use committee (IACUC) protocols (#SNU-111214-6-3) of the Seoul National University. The protocol was approved by the Committee on the Ethics of Animal Experiments of the Seoul National Universtiy. All surgery was performed under intraperitoneal injections of Zoletil/Rompun mixture (30 mg / 10 mg/kg), and every effort was made to minimize suffering.

Senior Editor

  1. Ronald L Calabrese, Emory University, United States

Reviewing Editor

  1. Megan R Carey, Champalimaud Foundation, Portugal

Publication history

  1. Received: July 30, 2020
  2. Accepted: September 17, 2020
  3. Accepted Manuscript published: September 28, 2020 (version 1)
  4. Version of Record published: October 22, 2020 (version 2)

Copyright

© 2020, Roh et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 844
    Page views
  • 146
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Download citations (links to download the citations from this article in formats compatible with various reference manager tools)

Open citations (links to open the citations from this article in various online reference manager services)

Further reading

    1. Neuroscience
    Hiromi Tamada et al.
    Research Advance

    Previously we showed that cryo fixation of adult mouse brain tissue gave a truer representation of brain ultrastructure in comparison with a standard chemical fixation method (Korogod et al 2005). Extracellular space matched physiological measurements, there were larger numbers of docked vesicles and less glial coverage of synapses and blood capillaries. Here, using the same preservation approaches we compared the morphology of dendritic spines. We show that the length of the spine and the volume of its head is unchanged, however, the spine neck width is thinner by more than 30 % after cryo fixation. In addition, the weak correlation between spine neck width and head volume seen after chemical fixation was not present in cryo-fixed spines. Our data suggest that spine neck geometry is independent of the spine head volume, with cryo fixation showing enhanced spine head compartmentalization and a higher predicted electrical resistance between spine head and parent dendrite.

    1. Neuroscience
    Opeyemi O Alabi et al.
    Research Article

    Goal-directed behaviors are essential for normal function and significantly impaired in neuropsychiatric disorders. Despite extensive associations between genetic mutations and these disorders, the molecular contributions to goal-directed dysfunction remain unclear. We examined mice with constitutive and brain region-specific mutations in Neurexin1α, a neuropsychiatric disease-associated synaptic molecule, in value-based choice paradigms. We found Neurexin1α knockouts exhibited reduced selection of beneficial outcomes and impaired avoidance of costlier options. Reinforcement modeling suggested this was driven by deficits in updating and representation of value. Disruption of Neurexin1α within telencephalic excitatory projection neurons, but not thalamic neurons, recapitulated choice abnormalities of global Neurexin1α knockouts. Furthermore, this selective forebrain excitatory knockout of Neurexin1α perturbed value-modulated neural signals within striatum, a central node in feedback-based reinforcement learning. By relating deficits in value-based decision-making to region-specific Nrxn1α disruption and changes in value-modulated neural activity, we reveal potential neural substrates for the pathophysiology of neuropsychiatric disease-associated cognitive dysfunction.