- Email: [email protected]
Identifying Purkinje cells using only their spontaneous simple spike activity
Journal of Neuroscience Methods 232 (2014) 173–180
Contents lists available at ScienceDirect
Journal of Neuroscience Methods journal homepage: www.elsevier.com/locate/jneumeth
Basic Neuroscience
Identifying Purkinje cells using only their spontaneous simple spike activity Robert A. Hensbroek a , Tim Belton a,1 , Boeke J. van Beugen a,2 , Jun Maruta a,3 , Tom J.H. Ruigrok b , John I. Simpson a,∗ a b
Department of Neuroscience and Physiology, New York University School of Medicine, New York, NY 10016, USA Department of Neuroscience, Erasmus MC Rotterdam, Rotterdam, The Netherlands
h i g h l i g h t s • • • •
Correct identification of cerebellar cortical neurons using only spontaneous spike activity. No need for anatomical data in order to classify. Applicable for both Purkinje cells and cerebellar cortical interneurons. Applicable for anesthetized and awake animals.
a r t i c l e
i n f o
Article history: Received 6 December 2013 Received in revised form 4 April 2014 Accepted 28 April 2014 Keywords: Flocculus Basket cell Stellate cell Golgi cell Granule cell Unipolar brush cell
a b s t r a c t Background: We have extended our cerebellar cortical interneuron classification algorithm that uses statistics of spontaneous activity (Ruigrok et al., 2011) to include Purkinje cells. Purkinje cells were added because they do not always show a detectable complex spike, which is the accepted identification. The statistical measures used in the present study were obtained from morphologically identified interneurons and complex spike identified Purkinje cells, recorded from ketamine–xylazine anesthetized rats and rabbits, and from awake rabbits. New method: The new algorithm has an added decision step that classifies Purkinje cells using a combination of the median absolute difference from the median interspike interval (MAD) and the mean of the relative differences of successive interspike intervals (CV2). These measures reflect the high firing rate and intermediate regularity of Purkinje cell simple spike activity. Results: Of 86 juxtacellularly labeled interneurons and 110 complex spike-identified Purkinje cells, 61 interneurons and 95 Purkinje cells were correctly classified, 22 interneurons and 13 Purkinje cells were deemed unclassifiable, and 3 interneurons and 2 Purkinje cells were incorrectly classified. Comparison with existing methods: The new algorithm improves on our previous algorithm because it includes Purkinje cells. This algorithm is the only one for the cerebellum that does not presume anatomical knowledge of whether the cells are in the molecular layer or the granular layer. Conclusions: These results strengthen the view that the new decision algorithm is useful for identifying neurons recorded at all cerebellar depths, particularly those neurons recorded in the rabbit vestibulocerebellum. © 2014 Elsevier B.V. All rights reserved.
Abbreviations: ISI, interspike interval; CVlog, coefficient of variation of ISIs expressed using the natural logarithm of the intervals in ms; ISIperc05, the value of the 5th percentile of ISIs; CV2, mean of two times the absolute difference of successive ISIs divided by the sum of both intervals; MAD, median absolute difference from the median ISI in seconds; AvgFreq, average firing frequency in Hertz; DP, D’Agostino-Pearson omnibus test statistic. ∗ Corresponding author. Tel.:.: +1 2122635428. E-mail address: [email protected] (J.I. Simpson). 1 Present address: Department of Biochemistry and Molecular Biophysics, Columbia University, New York, NY 10032, USA. 2 Present address: ENT Department, St Mary’s Hospital, London W2 1NY, UK. 3 Present address: Brain Trauma Foundation, Building 7, World Trade Center, New York, NY 10007, USA. http://dx.doi.org/10.1016/j.jneumeth.2014.04.031 0165-0270/© 2014 Elsevier B.V. All rights reserved.
174
R.A. Hensbroek et al. / Journal of Neuroscience Methods 232 (2014) 173–180
1. Introduction A recurring issue with in vivo extracellular recordings is determining the identity of particular neurons. We have previously pursued this issue for cerebellar cortical interneurons using juxtacellular labeling and the creation of a decision algorithm to identify these cells based on their spontaneous activity characteristics (Ruigrok et al., 2011). Purkinje cells have been considered an exception to the identification question because of their unique signature consisting of complex spikes followed by a pause in simple spike firing (Granit and Phillips, 1956). This signature is widely accepted as conclusive evidence for identifying Purkinje cells. Unfortunately, extracellularly recorded Purkinje cells do not always show a detectable complex spike, probably due to the electrode position in relation to the cell. Repositioning of the electrode can sometimes reveal the complex spike, but at other times such cells are identifiable as Purkinje cells only by penetrating the cell with the microelectrode and observing the intracellular reflection of the complex spikes (Hensbroek et al., 2011; Maruta et al., 2007). Consequently, Purkinje cells without detectable complex spikes can easily be mistaken for interneurons, prompting us to extend our decision algorithm so as to avoid misidentifying Purkinje cells. There have been previous attempts at discriminating Purkinje cells from interneurons using only simple spike activity (Holtzman et al., 2006; Van Dijck et al., 2013). These methods rely on analyzing subsets of cerebellar neurons according to their encounter in the molecular or the granular layer. Whereas layer identity can be determined at the cerebellar surface using recording electrode depth, beyond the first layers it is necessary to rely more and more on electrical signals from local field potentials, particularly signals from complex spikes (Van Dijck et al., 2013). Unfortunately, this is an undependable method because complex spikes are not always observed, especially when the local field potential signal is weakened by the use of the high impedance microelectrodes necessary for recording smaller cells like unipolar brush cells and granule cells. Thus, we set out to improve our previous decision algorithm to now include Purkinje cells by requiring only statistics of their spontaneous simple spike activity. We did this by combining data from interneurons that were identified by juxtacellular staining and simple spike data from Purkinje cells that were identified by their complex spike and resulting simple spike pause. A decision algorithm was constructed that reliably indicates whether spontaneous activity is derived from a Purkinje cell, a Golgi cell, a unipolar brush cell, a basket or stellate cell, or a granule cell. A preliminary report has been presented in short form (Hensbroek et al., 2011). 2. Material and methods 2.1. Animals Purkinje cell and unidentified interneuron recordings were obtained from acute experiments in 55 anesthetized Dutch belted rabbits of either sex and weighing 1.6–2.6 kg and from chronic awake experiments in two Dutch belted rabbits of either sex and weighing 1.8–2.0 kg. These experiments were conducted in New York, conformed to the Principles of Laboratory Animal Care, and were approved by the Institutional Animal Care and Use Committee of the New York University School of Medicine. 2.2. Juxtacellular labeling of interneurons in rats and rabbits Juxtacellular labeled interneurons (n = 86) were studied previously (Ruigrok et al., 2011) in 27 rats and 5 rabbits. Surgery,
neuronal recording, juxtacellular labeling and histochemical techniques for studying interneurons are described in Ruigrok et al. (2011). 2.2.1. Purkinje cell and interneuron recordings in acute anesthetized rabbits Of the 55 Dutch belted rabbits used in acute experiments, 53 rabbits were anesthetized with ketamine (45 mg/kg) and xylazine (5 mg/kg), and anesthesia was maintained by supplemental injections of ketamine (10 mg/kg) and xylazine (2 mg/kg) every 30–45 min. One rabbit was anesthetized with a mixture of urethane (500 mg/kg) and alpha chloralose (50 mg/kg) and one rabbit was anesthetized with urethane (1 g/kg). Normal body temperature was maintained with a heating pad. The head was held in the rabbit’s ‘freeze’ position with the nasal bone inclined at 57◦ to the horizontal. Therefore the horizontal canals were tilted approximately 7◦ backward and the vertical canals were essentially vertical (De Zeeuw et al., 1995). In order to allow recording of neuronal activity, an opening was made through a superior-lateral portion of the left occipital bone, and the dura mater overlying the cerebellum was removed. Neural recordings were made using glass microelectrodes advanced into the flocculus at an angle of 27–33◦ to the vertical. Electrodes were filled with 2–3 M NaCl and had diameters of 0.6–3 m and impedances of 2–8 M. Electrode sizes were varied because smaller cell types are more easily recorded with smaller electrodes and larger cell types with larger electrodes. Signals were amplified, bandpass filtered at 10 Hz/100 Hz to 10 kHz, captured, and stored for off-line analysis using a CED1401 data acquisition device and Spike2 software (Cambridge Electronics Design). 2.2.2. Purkinje cell and interneuron recordings in chronic awake rabbits Two Dutch belted rabbits were prepared for chronic awake recording with the use of sterile surgical techniques as described in detail in De Zeeuw et al. (1995). In short, animals were anesthetized with a mixture of acepromazine (0.1 mg/kg i.m.), ketamine (45 mg/kg i.m.) and xylazine (5 mg/kg i.m.); supplements were administered as required. An acrylic head fixation pedestal was molded and fixed on the skull. The pedestal was oriented so that the animal’s nasal bone made an angle of 57◦ to the horizontal. A craniotomy was performed over the left paramedian lobule of the cerebellum, and a metal recording chamber was fixed around the craniotomy by extending the head fixation pedestal. This cylindrical chamber was oriented with its axis in a sagittal plane at an angle of 27◦ to the vertical. The brain was covered by a Silastic sheet and the chamber was closed by a screw top. The animal was allowed a recovery period of at least 1 week during which it was habituated to the recording setup. Neural recordings were made with the same techniques and equipment as described above. When spontaneous activity was recorded the eyes were stationary and centered in the orbit. Recording sessions generally ran 4 h, but were terminated if the animal showed signs of agitation. Between recording sessions, the brain was covered by a Silastic sheet and the chamber was sealed. 2.3. Cell identification Juxtacellularly identified cells were the same sample presented in Ruigrok et al. (2011). The Purkinje cells used in constructing the new algorithm were identified by the presence of a pause in simple spike activity after the complex spike (Granit and Phillips, 1956). Complex spikes had their characteristic waveform and a low firing frequency of around 1 Hz (Eccles et al., 1966). On occasion, Purkinje cells that initially showed no complex spike were purposely penetrated by the microelectrode often resulting in several seconds of injury discharge, after which relatively large excitatory
R.A. Hensbroek et al. / Journal of Neuroscience Methods 232 (2014) 173–180
175
Fig. 1. Two examples of Purkinje cells that were first extracellularly recorded and then intracellularly recorded after penetration of the cell by the glass microelectrode. (A) Extracellular recording of a Purkinje cell that displayed both simple and complex spikes (marked with a dot). The complex spike has its typical low firing frequency (∼1 Hz). The firing statistics were: CVlog = 0.0560, AvgFreq = 29.8, MAD = 0.00410, CV2 = 0.220, ISIperc05 = 0.0241, ISImed = 0.0327. (B) Intracellular recording of the Purkinje cell from A showing 3 s of injury discharge and several complex spikes (marked with a dot). (C) Extracellular recording from a second Purkinje cell that initially showed only simple spikes. The firing statistics were: CVlog = 0.113, AvgFreq = 76.6, MAD = 0.00218, CV2 = 0.340, ISIperc05 = 0.00797, ISImed = 0.0125. (D) Intracellular recording of the Purkinje cell from C showing 3 s of injury discharge and several complex spikes (marked with a dot). Only at this point could this cell be positively identified as a Purkinje cell because of its characteristic response to climbing fiber input. Electrodes for these intracellular recordings typically had diameters of approximately 1 m and impedances of around 4 M. Signals were bandpass filtered between 10 Hz and 10 kHz. Time scales are the same for all traces.
postsynaptic potentials due to the climbing fiber input could be recorded (Eccles et al., 1966; Maekawa and Simpson, 1973; Maruta et al., 2007). At times, we encountered other Purkinje cells that initially showed no complex spike, but repositioning the microelectrode by 30–60 m along the electrode track sometimes revealed the complex spike and the associated simple spike pause. In order to be considered as a successful repositioning revealing a complex spike the signal of the simple spike had to remain present throughout the electrode movement with no significant sign of another cell being present. Furthermore, the firing statistics, as described in the following section, had to remain essentially the same.
2.4. Statistical measures and analysis All the cell records included in the construction of the new decision algorithm had durations of spontaneous activity lasting for at least 30 s and usually 60 s. Our previous decision algorithm used the following statistical measures: CVlog which is the coefficient of variation of interspike intervals (ISI) expressed using the natural logarithm of the intervals in milliseconds; the average firing frequency; CV2 which is the mean of two times the absolute difference between two successive ISIs divided by the sum of both intervals (Holt et al., 1996); the value of the 5th percentile of ISIs; and the median ISI. All 48 measures described in Ruigrok et al. (2011), including the median absolute difference from the median ISI (MAD), were calculated and analyzed. The measures that best distinguished Purkinje cells from other cerebellar neurons were first determined by a classification program called CTree, a Microsoft Excel-based classification tree written by Angshuman Saha (available at https://www.sites.google.com/site/sayhello2angshu/dminexcel), that uses the C4.5 algorithm. This program also allowed us to determine the effectiveness of the simultaneous use of multiple measures. Then, visual inspection of scatter plots of all combinations of these best measures were used to determine the most selective for Purkinje cell identification and for identification of “border zones”, which allowed for avoiding false-positives. The D’Agostino–Pearson omnibus test was used to test for normal distribution. This test uses the statistic DP that approximates a Chi-squared (2 ) distribution with 2 degrees of freedom if the population is normally distributed. The formula for DP is DP = (skewness/SES)2 + (kurtosis/SEK)2 , with SES being the standard error of skewness and SEK being the standard error of kurtosis (D Agostino et al., 1990).
Recordings of Purkinje cells with complex spikes (n = 110) were split into three groups. The first group (n = 39) consisted of Purkinje cell recordings from 16 anesthetized rabbits and this group was used to investigate all statistical measures and create the new decision algorithm. A second group (n = 36) consisted also of Purkinje cell recordings from another set of 11 anesthetized rabbits and this group was used to independently determine the effectiveness of the new decision algorithm. A third group (n = 35), consisting of Purkinje cell recordings from two awake rabbits, was used to determine whether the new algorithm was effective for recordings from awake rabbits. Finally, unclassified cells that did not display complex spikes at any time during the recording were analyzed using the new algorithm. Of these cells, 182 were recorded from 29 ketamine–xylazine anesthetized rabbits of which 17 are not included above, and 170 were recorded from the two awake rabbits.
3. Results 3.1. Purkinje cells without detectable complex spikes Extracellular Purkinje cell recordings do not always show a detectable complex spike, presumably due to the relative positions of the recording electrode and the Purkinje cell. In some cases, such cells can be identified as Purkinje cells after penetration by the recording glass microelectrode and the subsequent appearance of the complex spike (Eccles et al., 1966; Hensbroek et al., 2011; Maekawa and Simpson, 1973; Maruta et al., 2007). Fig. 1 shows raw data from two penetrated Purkinje cells. The first Purkinje cell showed complex spikes (indicated by the black dot) and simple spikes both before (Fig. 1A) and after penetration (Fig. 1B). The second Purkinje cell showed only simple spikes in the extracellular record (Fig. 1C) and as a result, this cell could easily have been mistaken for an interneuron. However, after penetration complex spikes were revealed, thereby identifying this cell as a Purkinje cell (1D). After intracellular penetration the observed complex spike usually reflects only the postsynaptic activity due to the climbing fiber input. These responses constitute a unique signature due to the brief high frequency bursts of action potentials (up to seven action potentials at 500–800 Hz) that often comprise a single climbing fiber input (Maruta et al., 2007). Thus, in the intracellular configuration positive identification of Purkinje cells does not require a pause in simple spikes after the complex spike while the simple spikes are degraded to an injury discharge. In other cases, Purkinje cells that initially did not show a complex spike can sometimes be identified after making a small
176
R.A. Hensbroek et al. / Journal of Neuroscience Methods 232 (2014) 173–180
Fig. 2. Three examples of cells that initially showed no complex spikes but whose complex spikes were revealed after a small extracellular movement of the microelectrode. (A and B) Instantaneous frequencies of simple spikes (A) and complex spikes (B) isolated using spike sorting techniques. The microelectrode was moved at t = 22 s (arrowhead). (C) Extracellular record is from t = 7 to t = 10 s with a 60 ms section blowup. (D) Extracellular record is from t = 26 s to t = 29 s, also with a 60 ms section blowup. A suspected complex spike in the blowup of (C) and complex spikes in (D) are marked with a dot. The firing statistics before moving the electrode were: CVlog = 0.110, AvgFreq = 37.0, MAD = 0.00410, CV2 = 0.274, ISIperc05 = 0.0169, ISImed = 0.0256, and the firing statistics after movement were: CVlog = 0.0809, AvgFreq = 35.8, MAD = 0.00377, CV2 = 0.238, ISIperc05 = 0.0187, ISImed = 0.0264.(E) A second cell that initially did not show complex spikes. The rabbit from which this cell was obtained was anesthetized with urethane. (F) Record from the same cell as (E) after a small movement of the electrode that resulted in the appearance of complex spikes. The firing statistics before moving the electrode were: CVlog = 0.0779, AvgFreq = 31.5, MAD = 0.00509, CV2 = 0.310, ISIperc05 = 0.0208, ISImed = 0.0300, and the firing statistics after movement were: CVlog = 0.0808, AvgFreq = 28.4, MAD = 0.00559, CV2 = 0.323, ISIperc05 = 0.0226, ISImed = 0.0333. (G). A third cell that initially did not show complex spikes. (H) Record from the same cell as G after a small movement of the electrode that resulted in the appearance of complex spikes. The firing statistics before moving the electrode were: CVlog = 0.0623, AvgFreq = 65.8, MAD = 0.00166, CV2 = 0.194, ISIperc05 = 0.0115, ISImed = 0.0150, and the firing statistics after movement were: CVlog = 0.0703, AvgFreq = 65.5, MAD = 0.00170, CV2 = 0.221, ISIperc05 = 0.0110, ISImed = 0.0151. Signals were bandpass filtered between 10 Hz and 10KHz. Note that the time scales are not the same for all traces.
(30–60 m) extracellular movement of the microelectrode to reveal the complex spike, as shown by the exemplary recordings from three Purkinje cells in Fig. 2. Initially, the first Purkinje cell did not show identifiable complex spikes (Fig. 2B and C) but after movement of the microelectrode at time t = 22 s (indicated by the arrowheads) the complex spikes appeared and could be clearly distinguished from simple spikes (Fig. 2B and D). The identity of this cell was confirmed by a simple spike pause after complex spike occurrence. An interesting observation in this particular record is that during the first 22 s several very short interspike intervals (ISIs) were observed, resulting in the high instantaneous frequencies seen initially in Fig. 2A. Such short ISIs are unusual for Purkinje cell simple spikes. After the complex spikes became visible these short ISIs abruptly disappeared. Fig. 2C shows a detail of such a short ISI. The second spike (marked with a dot) is similar to the other spikes and were it not for the very short ISI this particular spike would have remained unnoticed. Further analysis showed that these short-interval spikes were followed by a transient absence of spikes (e.g. Fig. 2C, top), similar to the simple spike pause after a complex spike. Therefore, we propose that these short-interval spikes were actually complex spikes which, due to the position of the microelectrode in relation to the cell, appear similar to
simple spikes. The occasional high instantaneous frequencies seen in Fig. 2A would then be the result of the independent occurrence of complex spikes among the simple spikes. This finding underscores the need for stable statistics for spike activity in a decision algorithm because simple spikes may be contaminated by complex spikes. A second example of a Purkinje cell that initially did not show complex spikes is shown in Fig. 2E and F. Before movement of the electrode only one type of spike was visible (2E), but after movement the complex spike and the associated pause in simple spike activity appeared in the extracellular record (2F). A third example is shown in Fig. 2G and H. Similar to the previous example, initially only one type of spike was visible (Fig. 2G), but after movement of the electrode the complex spike and the associated simple spike pause appeared (2H). Unlike the Purkinje cell in Fig. 2A–D, the two Purkinje cells in Fig. 2E–H did not display high frequency events before electrode movement, showing that undetectable complex spikes, appearing similar to simple spikes, are not always found. In three animals, we made 14 attempts to reveal the complex spikes for cells that initially had only a spike pattern closely resembling that of simple spike activity. In four attempts (29%) the
R.A. Hensbroek et al. / Journal of Neuroscience Methods 232 (2014) 173–180
complex spike was revealed extracellularly and in two attempts (14%) the complex spike was revealed intracellularly by penetration. In the other eight attempts (57%) we were unable to reveal the complex spike. In a related experiment in 14 animals (9 of which were included only in this experiment) we tried to estimate how common it is that Purkinje cell simple spikes can be recorded without an accompanying complex spike. To do this we first obtained Purkinje cell recordings with a clear signal of both the complex spike and simple spike and subsequently slowly moved the recording electrode along its approach track over distances up to 60 m. In 59% of those instances (13/22 Purkinje cells) the complex spike signal was selectively lost while the simple spike signal remained. The other 41% of Purkinje cells (9/22 cells) showed a simultaneous decline of both the simple and complex spike. This experiment included one rabbit that was anesthetized with urethane and one rabbit that was anesthetized with urethane and alpha chloralose. These two rabbits provided records of six Purkinje cells, two of which selectively lost their complex spike thereby demonstrating that this phenomenon is not restricted to rabbits anesthetized with ketamine–xylazine. These findings indicate the existence of Purkinje cells that have areas near them where only simple spikes are visible to the recording electrode. As mentioned in Section 1, our existing decision algorithm (Ruigrok et al., 2011) does not include Purkinje cells, and Figs. 1 and 2 demonstrate how one could confuse a Purkinje cell without a detectable complex spike for an interneuron. If such a recording is subsequently analyzed using our previously published algorithm, the result would be a “border” classification, or even a false-positive. This situation prompted us to construct a new decision algorithm that correctly identifies both interneurons and Purkinje cells. 3.2. Analysis of Purkinje cell spontaneous simple spike activity Purkinje cell recordings from anesthetized rabbits were analyzed in order to determine variables that could distinguish them from interneurons. Stability against small variations in spike discrimination and the ability to create a clear statistical separation between Purkinje cells and interneurons were key features desired for selecting discriminating variables. For the initial investigation in anesthetized rabbits, we used 39 Purkinje cells from 16 animals that were identified by complex spikes and the associated simple spike pause. Only simple spikes were used for calculating the variables. Visual inspection of the analysis with a C4.5-based decision tree and two-dimensional scatter plots of statistical measures revealed that the spontaneous firing characteristics of the Purkinje cells diverged from those of cerebellar interneurons most notably in the combination of values of MAD and CV2 (Fig. 3). The divergence found with these variables reflects the high firing rate and intermediate regularity of simple spike firing. The MAD and CV2 for the initial Purkinje cell group (n = 39) had averages of 0.00368 and 0.354 with standard deviations of 0.00178 and 0.102, respectively. Neither the MAD nor the CV2 had a normal distribution because both showed moderate to high positive skewness (0.991 for MAD and 0.918 for CV2), and low to moderate positive excess kurtosis (0.391 for the MAD and 1.701 for the CV2). In addition, the MAD and CV2 distributions were tested for deviation from the normal distribution using the D Agostino–Pearson omnibus test (D Agostino et al., 1990), and both measures showed significant differences from normality (MAD: DP = 7.14, p < 0.05; CV2: DP = 11.17, p < 0.005). 3.3. The new decision algorithm The ability of the combination of MAD and CV2 to discriminate Purkinje cell activity from that of virtually all interneurons (Fig. 3) suggests that Purkinje cells might be selectable at an early decision
177
Fig. 3. MAD and CV2 values calculated from spontaneous activity (see Section 2 for details). Shown are 86 morphologically identified interneurons from 27 anesthetized rats and 5 anesthetized rabbits (Ruigrok et al., 2011), and 39 Purkinje cells identified by their complex spike from 16 ketamine–xylazine anesthetized rabbits. Granule cells are not shown because they are easily classified in the first step in the analysis sequence (see text). The dashed lines indicate the decision boundaries between Purkinje cells and border cells and between border cells and interneurons in the new decision algorithm (see also Fig. 4).
point in our new decision algorithm. Further analysis with a C4.5-based decision tree and visual inspection supported this view and showed that our earlier algorithm could be successfully extended by including an additional early decision step in the analysis sequence for Purkinje cells based on CV2 and MAD (Fig. 4). For some cell recordings discrimination using these two measures was still difficult, which prompted us to create “border zones”, where cells that were deemed unclassifiable would be placed to avoid false-positives and thereby improve the reliability of the new decision tree. A brief description of the analysis sequence for the new tree: Step 1: Granule cells were found to be either nearly silent or fire very irregularly with both very short (∼1 ms) and very long (∼seconds) interspike intervals. Thus, recordings were classified as granule cells when they had a CVlog greater than 0.38 or an average frequency smaller than 0.5 Hz. Cells with a CVlog smaller than 0.34 and an average frequency larger than 0.6 Hz were continued in the algorithm. Cells not captured by these criteria were deemed unclassifiable and called border cells. Step 2: Purkinje cell simple spikes showed a combination of a high firing rate and an intermediate regularity. Of the cells that were continued in the algorithm from step 1, those classified as Purkinje cells had a MAD smaller than 0.008 and a CV2 larger than 0.2. Cells with a MAD larger than 0.01 or a CV2 smaller than 0.15 were continued in the algorithm. The rest were border cells. Step 3: Unipolar brush cells showed regular firing. If a cell that was continued from step 2 had a CV2 smaller than 0.24 then it was identified as a unipolar brush cell. Cells with a CV2 larger than 0.28 were continued in the algorithm and the rest were border cells. Step 4: Basket and stellate cells often fire irregularly. They can have a high proportion of short interspike intervals, but they do not fire in high frequency bursts. Cells that were continued from step 3 and had either a CVlog larger than 0.17 or a 5th percentile ISI smaller than 22 ms were identified as a basket or stellate cell. Cells that had a CVlog smaller than 0.15 and a 5th percentile ISI larger than 44 ms were continued in the analysis; the rest were deemed border cells. Step 5: The large majority of Golgi cells are rather irregular and fire at a relatively slow rate, but not slower than a median ISI of 300 ms.
178
R.A. Hensbroek et al. / Journal of Neuroscience Methods 232 (2014) 173–180
Fig. 4. New decision algorithm derived from a previously published algorithm (Ruigrok et al., 2011). We added an additional step to identify Purkinje cells using only their spontaneous simple spike activity. This new classification step uses a combination of the MAD and CV2 statistics, which are sensitive to the high firing rate and the intermediate regularity of simple spike activity.
However, a particular class of slowly firing basket and stellate cells had median ISIs larger than 320 ms. The cells with intermediate values of the median ISI between 300 and 320 ms became border cells. 3.4. Validation of the new decision algorithm When the new algorithm was applied to the initial group of interneurons (n = 86) and Purkinje cells (n = 39) it showed an accuracy of 78% (97 out of 125 cells), with 20% (25 out of 125) considered as border cells, and 2% (3 out of 125) misclassified (interneurons: 61 correct, 22 border cells and 3 misclassified; Purkinje cells: 36 correct, 3 border cells and 0 misclassified). The 2 cells in Fig. 1 were classified as Purkinje cells. Of the 3 cells in Fig. 2, the first (Fig. 2A–D) and second (Fig. 2E and F) were classified as Purkinje cells both before and after movement. The third cell (Fig. 2G and H) was only classified as a Purkinje cell after movement but was classified as a border cell before movement. That situation shows the need for a border zone as this cell would have been a false-positive for another cell type if no border zones were present. Next, the validity of the new decision algorithm was independently tested using two new groups of Purkinje cells identified by their complex spikes. One group came from 11 anesthetized rabbits and another from two awake rabbits (Table 1). Out of the 36 Purkinje cells from the anesthetized rabbits, 29 (81%) were correctly classified, 1 (3%) was misclassified and 6 (17%) were designated as border cells. For the awake rabbits, out of the 35 Purkinje cells, 30 (86%) were correctly classified, 1 (3%) was misclassified and 4 (11%) were designated border cells. These numbers are similar to those for the 39 Purkinje cells used in construction of the new decision algorithm above and strengthen our view that the new algorithm is unbiased and does
Table 1 Test of the new decision algorithm by using two additional groups of Purkinje cells. One group (n = 36) was recorded in anesthetized rabbits and another group (n = 35) was recorded in awake rabbits. All these Purkinje cells were first identified by the presence of their complex spikes and the associated pause in simple spike activity.
Granule cells Purkinje cells Unipolar brush cells Basket or stellate cells Golgi cells Border cells
Anesthetized (n = 36)
Awake (n = 35)
– 81% (29) – 3% (1) – 17% (6)
– 86% (30) – 3% (1) – 11% (4)
not result in over-fitting. Furthermore, the high accuracy of identification for Purkinje cells recorded in awake rabbits highlights the wider applicability of the algorithm. Finally, the new decision algorithm was applied to extracellular recordings of cells that did not show a complex spike. One set (n = 182) was from 29 ketamine–xylazine anesthetized rabbits and the other set (n = 170) was from 2 awake rabbits (Table 2). In Table 2 Application of the new decision algorithm to two groups of unidentified neurons. One group (n = 182) was recorded in anesthetized rabbits and the other group (n = 170) was recorded in awake rabbits.
Granule cells Purkinje cells Unipolar brush cells Basket or stellate cells Golgi cells Border cells
Anesthetized (n = 182)
Awake (n = 170)
23% (42) 19% (34) 13% (24) 14% (25) 11% (20) 20% (37)
22% (38) 15% (25) 14% (24) 13% (22) 10% (17) 26% (44)
R.A. Hensbroek et al. / Journal of Neuroscience Methods 232 (2014) 173–180
both sets of recordings a substantial number of cells (19% and 15%, respectively) were classified as Purkinje cells even though they did not show a complex spike, which underscores the need for a Purkinje cell identification step in the decision tree. 4. Discussion Using only statistics derived from spontaneous spike activity, we devised an algorithm consisting of five sequential steps that enabled correct identification of interneurons and Purkinje cells recorded from the rabbit vestibulocerebellum. The ability to identify Purkinje cells simply from their spontaneous simple spike activity is an improvement over our previous decision algorithm because before Purkinje cells could only be identified by the presence of their complex spikes. 4.1. Alternative identification methods Identification of some types of cerebellar cortical interneurons on the basis of their spike activity has been proposed before (Edgley and Lidierth, 1987; Van Kan et al., 1993; Vos et al., 1999). Only more recently with the advent of juxtacellular and intracellular labeling techniques and subsequent morphological identification of cerebellar interneurons (Barmack and Yakhnitsa, 2008; Holtzman et al., 2006; Jorntell and Ekerot, 2003; Simpson et al., 2005), has it become feasible to try to create spontaneous spike activity-based identification methods (Ruigrok et al., 2011; Van Dijck et al., 2013). Such methods are of importance especially in situations where juxtacellular labeling is not routinely feasible such as in awake animals. This is not to say that information unrelated to spontaneous activity is not valuable. In our opinion, the presence of a complex spike with the associated pause in simple spike firing remains the “gold” standard for Purkinje cell identification (Granit and Phillips, 1956). Background complex spikes may also be used to support identification of basket and stellate cells because the polarity of the complex spike is predominantly negative when recorded in the molecular layer (Eccles et al., 1967). In fact, a recently published method (Van Dijck et al., 2013) utilizes such information from complex spikes to distinguish between molecular and granular layer neurons. Subsequently, this method uses a probabilistic model to identify neurons using measures of firing frequency and irregularity that are not unlike those presented here. The key difference from our method is the initial sorting by the experimenter into either molecular or granular layer recordings, which is necessary because the probabilistic method is unable to discriminate molecular layer basket/stellate cell recordings when mixed with granular layer Golgi cell and granule cell recordings. As mentioned above, one can probably be quite certain about recording from the molecular layer when a negative complex spike is visible in the local field potential. Unfortunately, complex spikes in the local field potentials are not always visible especially when using the small-tipped electrodes necessary for recording small interneurons such as granule cells. Certainly, the absence of a negative complex spike is no guarantee that the electrode is in the granular layer. Regrettably, Van Dijck et al. (2013) offer no systematic method or validity analysis for identifying the recording site layer, which is problematic given that the overall success rate of their method is the product of the reliability of selecting the correct anatomical layer and the efficacy of the probabilistic method itself. This issue limits the practical application of their method for recordings deep to the cerebellar surface. 4.2. Applicability Various factors such as species, types of anesthesia and cerebellar region may all influence the firing behavior of neurons,
179
but signature patterns derived from spontaneous activity may still show enough stability for identification analysis. For instance, while it cannot be excluded that ketamine–xylazine anesthesia may have reduced spontaneous granule cell activity (Hevers et al., 2008), activity patterns in the granular layer have been reported to be similar in awake and ketamine anesthetized rats (Hartmann and Bower, 2001; Santamaria et al., 2007). In addition, Van Dijck et al. (2013) used neurons from different cerebellar regions in ketamine–xylazine anesthetized mice, urethane anesthetized rats and decerebrate cats and did not report marked differences in the distributions of their statistics. Similarly, the recorded activity signatures of spontaneously firing morphologically identified interneurons in the vestibulocerebellum of ketamine–xylazine anesthetized rats and rabbits were both shown to adhere to the same decision algorithm (Ruigrok et al., 2011). Furthermore, recordings from unidentified neurons (Table 2) showed similar distributions whether they were from anesthetized or awake rabbits, indicating that the statistical measures and cell populations from anesthetized and awake rabbits did not differ markedly. Finally, Purkinje cells identified by complex spikes recorded from awake rabbits were correctly discriminated by the new algorithm with a similar percentage as Purkinje cells from anesthetized rabbits. Taken together, these findings strengthen our view that the new decision algorithm may have a wide applicability. 5. Conclusion We have improved upon our previous cerebellar interneuron classification algorithm by adding measures for classification of Purkinje cells using only their spontaneous simple spike activity recorded in the awake and anesthetized rabbit. We suggest that our measures are valid not only for neurons in the rabbit vestibulocerebellum, but also for neurons in the vestibulocerebellum of other species. Acknowledgements This work was supported by grants from the NIH (NS-13742) and the Dutch Ministry of Health, Welfare and Sport (TJHR). References Barmack NH, Yakhnitsa V. Functions of interneurons in mouse cerebellum. J Neurosci 2008;28:1140–52. D Agostino RB, Belanger A, D Agostino RB Jr. A suggestion for using powerful and informative tests of normality. Am Stat 1990;44(4):316–21. De Zeeuw CI, Wylie DR, Stahl JS, Simpson JI. Phase relations of Purkinje cells in the rabbit flocculus during compensatory eye movements. J Neurophysiol 1995;74:2051–64. Eccles JC, Ito M, Szentagothai J. The cerebellum as a neuronal machine. Berlin: Springer-Verlag; 1967. Eccles JC, Llinas R, Sasaki K. The excitatory synaptic action of climbing fibers on the Purkinje cells of the cerebellum. J Physiol 1966;182:268–96. Edgley SA, Lidierth M. The discharges of cerebellar Golgi cells during locomotion in the cat. J Physiol 1987;392:315–32. Granit R, Phillips CG. Excitatory and inhibitory processes acting upon individual Purkinje cells of the cerebellum in cats. J Physiol 1956;133(3):520–47. Hartmann MJ, Bower JM. Tactile responses in the granule cell layer of cerebellar folium crus IIa of freely behaving rats. J Neurosci 2001;21:3549–63. Hensbroek RA, Van Beugen BJ, Maruta J, Ruigrok TJH, Simpson JI. Discriminating Purkinje cells without using complex spikes. Soc for Neurosci 41st Annu Meet 2011, 922.15. Hevers W, Hadley SH, Luddens H, Amin J. Ketamine, but not phencyclidine, selectively modulates cerebellar GABAA receptors containing ␣6 and ␦ subunits. J Neurosci 2008;28:5383–93. Holt GR, Softky WR, Koch C, Douglas RJ. Comparison of discharge variability in vitro and in vivo in cat visual cortex neurons. J Neurophysiol 1996;75:1806–14. Holtzman T, Rajapaksa T, Mostofi A, Edgley SA. Different responses of rat cerebellar Purkinje cells and Golgi cells evoked by widespread convergent sensory inputs. J Physiol 2006;574(2):491–507.
180
R.A. Hensbroek et al. / Journal of Neuroscience Methods 232 (2014) 173–180
Jorntell H, Ekerot CF. Receptive field plasticity profoundly alters the cutaneous parallel fiber synaptic input to cerebellar interneurons in vivo. J Neurosci 2003;23(29):9620–31. Maekawa M, Simpson JI. Climbing fiber responses evoked in vestibulo-cerebellum of rabbit from visual system. J Neurophysiol 1973;36:649–66. Maruta J, Hensbroek RA, Simpson JI. Intraburst and interburst signaling by climbing fibers. J Neurosci 2007;27(42):11263–70. Ruigrok TJH, Hensbroek RA, Simpson JI. Spontaneous activity signatures of morphologically identified interneurons in the vestibulocerebellum. J Neurosci 2011;31(2):712–24. Santamaria F, Tripp PG, Bower JM. Feedforward inhibition controls the spread of granule cell-induced Purkinje cell activity in the cerebellar cortex. J Neurophysiol 2007;97:248–63.
Simpson JI, Hulscher HC, Sabel-Goedknegt E, Ruigrok TJ. Between in and out: linking morphology and physiology of cerebellar cortical interneurons. Prog Brain Res 2005;148:329–40. Van Dijck G, Van Hulle MM, Heiney SA, Blazquez PM, Meng H, Angelaki DE, et al. Probabilistic identification of cerebellar cortical neurones across species. PLoS One 2013;8(3):e57669. Van Kan PL, Gibson AR, Houk JC. Movement-related inputs to intermediate cerebellum of the monkey. J Neurophysiol 1993;69(1):74–94. Vos BP, Volny-Luraghi A, De Schutter E. Cerebellar Golgi cells in the rat: receptive fields and timing of responses to facial stimulation. Eur J Neurosci 1999;11(8):2621–34.
Contents lists available at ScienceDirect
Journal of Neuroscience Methods journal homepage: www.elsevier.com/locate/jneumeth
Basic Neuroscience
Identifying Purkinje cells using only their spontaneous simple spike activity Robert A. Hensbroek a , Tim Belton a,1 , Boeke J. van Beugen a,2 , Jun Maruta a,3 , Tom J.H. Ruigrok b , John I. Simpson a,∗ a b
Department of Neuroscience and Physiology, New York University School of Medicine, New York, NY 10016, USA Department of Neuroscience, Erasmus MC Rotterdam, Rotterdam, The Netherlands
h i g h l i g h t s • • • •
Correct identification of cerebellar cortical neurons using only spontaneous spike activity. No need for anatomical data in order to classify. Applicable for both Purkinje cells and cerebellar cortical interneurons. Applicable for anesthetized and awake animals.
a r t i c l e
i n f o
Article history: Received 6 December 2013 Received in revised form 4 April 2014 Accepted 28 April 2014 Keywords: Flocculus Basket cell Stellate cell Golgi cell Granule cell Unipolar brush cell
a b s t r a c t Background: We have extended our cerebellar cortical interneuron classification algorithm that uses statistics of spontaneous activity (Ruigrok et al., 2011) to include Purkinje cells. Purkinje cells were added because they do not always show a detectable complex spike, which is the accepted identification. The statistical measures used in the present study were obtained from morphologically identified interneurons and complex spike identified Purkinje cells, recorded from ketamine–xylazine anesthetized rats and rabbits, and from awake rabbits. New method: The new algorithm has an added decision step that classifies Purkinje cells using a combination of the median absolute difference from the median interspike interval (MAD) and the mean of the relative differences of successive interspike intervals (CV2). These measures reflect the high firing rate and intermediate regularity of Purkinje cell simple spike activity. Results: Of 86 juxtacellularly labeled interneurons and 110 complex spike-identified Purkinje cells, 61 interneurons and 95 Purkinje cells were correctly classified, 22 interneurons and 13 Purkinje cells were deemed unclassifiable, and 3 interneurons and 2 Purkinje cells were incorrectly classified. Comparison with existing methods: The new algorithm improves on our previous algorithm because it includes Purkinje cells. This algorithm is the only one for the cerebellum that does not presume anatomical knowledge of whether the cells are in the molecular layer or the granular layer. Conclusions: These results strengthen the view that the new decision algorithm is useful for identifying neurons recorded at all cerebellar depths, particularly those neurons recorded in the rabbit vestibulocerebellum. © 2014 Elsevier B.V. All rights reserved.
Abbreviations: ISI, interspike interval; CVlog, coefficient of variation of ISIs expressed using the natural logarithm of the intervals in ms; ISIperc05, the value of the 5th percentile of ISIs; CV2, mean of two times the absolute difference of successive ISIs divided by the sum of both intervals; MAD, median absolute difference from the median ISI in seconds; AvgFreq, average firing frequency in Hertz; DP, D’Agostino-Pearson omnibus test statistic. ∗ Corresponding author. Tel.:.: +1 2122635428. E-mail address: [email protected] (J.I. Simpson). 1 Present address: Department of Biochemistry and Molecular Biophysics, Columbia University, New York, NY 10032, USA. 2 Present address: ENT Department, St Mary’s Hospital, London W2 1NY, UK. 3 Present address: Brain Trauma Foundation, Building 7, World Trade Center, New York, NY 10007, USA. http://dx.doi.org/10.1016/j.jneumeth.2014.04.031 0165-0270/© 2014 Elsevier B.V. All rights reserved.
174
R.A. Hensbroek et al. / Journal of Neuroscience Methods 232 (2014) 173–180
1. Introduction A recurring issue with in vivo extracellular recordings is determining the identity of particular neurons. We have previously pursued this issue for cerebellar cortical interneurons using juxtacellular labeling and the creation of a decision algorithm to identify these cells based on their spontaneous activity characteristics (Ruigrok et al., 2011). Purkinje cells have been considered an exception to the identification question because of their unique signature consisting of complex spikes followed by a pause in simple spike firing (Granit and Phillips, 1956). This signature is widely accepted as conclusive evidence for identifying Purkinje cells. Unfortunately, extracellularly recorded Purkinje cells do not always show a detectable complex spike, probably due to the electrode position in relation to the cell. Repositioning of the electrode can sometimes reveal the complex spike, but at other times such cells are identifiable as Purkinje cells only by penetrating the cell with the microelectrode and observing the intracellular reflection of the complex spikes (Hensbroek et al., 2011; Maruta et al., 2007). Consequently, Purkinje cells without detectable complex spikes can easily be mistaken for interneurons, prompting us to extend our decision algorithm so as to avoid misidentifying Purkinje cells. There have been previous attempts at discriminating Purkinje cells from interneurons using only simple spike activity (Holtzman et al., 2006; Van Dijck et al., 2013). These methods rely on analyzing subsets of cerebellar neurons according to their encounter in the molecular or the granular layer. Whereas layer identity can be determined at the cerebellar surface using recording electrode depth, beyond the first layers it is necessary to rely more and more on electrical signals from local field potentials, particularly signals from complex spikes (Van Dijck et al., 2013). Unfortunately, this is an undependable method because complex spikes are not always observed, especially when the local field potential signal is weakened by the use of the high impedance microelectrodes necessary for recording smaller cells like unipolar brush cells and granule cells. Thus, we set out to improve our previous decision algorithm to now include Purkinje cells by requiring only statistics of their spontaneous simple spike activity. We did this by combining data from interneurons that were identified by juxtacellular staining and simple spike data from Purkinje cells that were identified by their complex spike and resulting simple spike pause. A decision algorithm was constructed that reliably indicates whether spontaneous activity is derived from a Purkinje cell, a Golgi cell, a unipolar brush cell, a basket or stellate cell, or a granule cell. A preliminary report has been presented in short form (Hensbroek et al., 2011). 2. Material and methods 2.1. Animals Purkinje cell and unidentified interneuron recordings were obtained from acute experiments in 55 anesthetized Dutch belted rabbits of either sex and weighing 1.6–2.6 kg and from chronic awake experiments in two Dutch belted rabbits of either sex and weighing 1.8–2.0 kg. These experiments were conducted in New York, conformed to the Principles of Laboratory Animal Care, and were approved by the Institutional Animal Care and Use Committee of the New York University School of Medicine. 2.2. Juxtacellular labeling of interneurons in rats and rabbits Juxtacellular labeled interneurons (n = 86) were studied previously (Ruigrok et al., 2011) in 27 rats and 5 rabbits. Surgery,
neuronal recording, juxtacellular labeling and histochemical techniques for studying interneurons are described in Ruigrok et al. (2011). 2.2.1. Purkinje cell and interneuron recordings in acute anesthetized rabbits Of the 55 Dutch belted rabbits used in acute experiments, 53 rabbits were anesthetized with ketamine (45 mg/kg) and xylazine (5 mg/kg), and anesthesia was maintained by supplemental injections of ketamine (10 mg/kg) and xylazine (2 mg/kg) every 30–45 min. One rabbit was anesthetized with a mixture of urethane (500 mg/kg) and alpha chloralose (50 mg/kg) and one rabbit was anesthetized with urethane (1 g/kg). Normal body temperature was maintained with a heating pad. The head was held in the rabbit’s ‘freeze’ position with the nasal bone inclined at 57◦ to the horizontal. Therefore the horizontal canals were tilted approximately 7◦ backward and the vertical canals were essentially vertical (De Zeeuw et al., 1995). In order to allow recording of neuronal activity, an opening was made through a superior-lateral portion of the left occipital bone, and the dura mater overlying the cerebellum was removed. Neural recordings were made using glass microelectrodes advanced into the flocculus at an angle of 27–33◦ to the vertical. Electrodes were filled with 2–3 M NaCl and had diameters of 0.6–3 m and impedances of 2–8 M. Electrode sizes were varied because smaller cell types are more easily recorded with smaller electrodes and larger cell types with larger electrodes. Signals were amplified, bandpass filtered at 10 Hz/100 Hz to 10 kHz, captured, and stored for off-line analysis using a CED1401 data acquisition device and Spike2 software (Cambridge Electronics Design). 2.2.2. Purkinje cell and interneuron recordings in chronic awake rabbits Two Dutch belted rabbits were prepared for chronic awake recording with the use of sterile surgical techniques as described in detail in De Zeeuw et al. (1995). In short, animals were anesthetized with a mixture of acepromazine (0.1 mg/kg i.m.), ketamine (45 mg/kg i.m.) and xylazine (5 mg/kg i.m.); supplements were administered as required. An acrylic head fixation pedestal was molded and fixed on the skull. The pedestal was oriented so that the animal’s nasal bone made an angle of 57◦ to the horizontal. A craniotomy was performed over the left paramedian lobule of the cerebellum, and a metal recording chamber was fixed around the craniotomy by extending the head fixation pedestal. This cylindrical chamber was oriented with its axis in a sagittal plane at an angle of 27◦ to the vertical. The brain was covered by a Silastic sheet and the chamber was closed by a screw top. The animal was allowed a recovery period of at least 1 week during which it was habituated to the recording setup. Neural recordings were made with the same techniques and equipment as described above. When spontaneous activity was recorded the eyes were stationary and centered in the orbit. Recording sessions generally ran 4 h, but were terminated if the animal showed signs of agitation. Between recording sessions, the brain was covered by a Silastic sheet and the chamber was sealed. 2.3. Cell identification Juxtacellularly identified cells were the same sample presented in Ruigrok et al. (2011). The Purkinje cells used in constructing the new algorithm were identified by the presence of a pause in simple spike activity after the complex spike (Granit and Phillips, 1956). Complex spikes had their characteristic waveform and a low firing frequency of around 1 Hz (Eccles et al., 1966). On occasion, Purkinje cells that initially showed no complex spike were purposely penetrated by the microelectrode often resulting in several seconds of injury discharge, after which relatively large excitatory
R.A. Hensbroek et al. / Journal of Neuroscience Methods 232 (2014) 173–180
175
Fig. 1. Two examples of Purkinje cells that were first extracellularly recorded and then intracellularly recorded after penetration of the cell by the glass microelectrode. (A) Extracellular recording of a Purkinje cell that displayed both simple and complex spikes (marked with a dot). The complex spike has its typical low firing frequency (∼1 Hz). The firing statistics were: CVlog = 0.0560, AvgFreq = 29.8, MAD = 0.00410, CV2 = 0.220, ISIperc05 = 0.0241, ISImed = 0.0327. (B) Intracellular recording of the Purkinje cell from A showing 3 s of injury discharge and several complex spikes (marked with a dot). (C) Extracellular recording from a second Purkinje cell that initially showed only simple spikes. The firing statistics were: CVlog = 0.113, AvgFreq = 76.6, MAD = 0.00218, CV2 = 0.340, ISIperc05 = 0.00797, ISImed = 0.0125. (D) Intracellular recording of the Purkinje cell from C showing 3 s of injury discharge and several complex spikes (marked with a dot). Only at this point could this cell be positively identified as a Purkinje cell because of its characteristic response to climbing fiber input. Electrodes for these intracellular recordings typically had diameters of approximately 1 m and impedances of around 4 M. Signals were bandpass filtered between 10 Hz and 10 kHz. Time scales are the same for all traces.
postsynaptic potentials due to the climbing fiber input could be recorded (Eccles et al., 1966; Maekawa and Simpson, 1973; Maruta et al., 2007). At times, we encountered other Purkinje cells that initially showed no complex spike, but repositioning the microelectrode by 30–60 m along the electrode track sometimes revealed the complex spike and the associated simple spike pause. In order to be considered as a successful repositioning revealing a complex spike the signal of the simple spike had to remain present throughout the electrode movement with no significant sign of another cell being present. Furthermore, the firing statistics, as described in the following section, had to remain essentially the same.
2.4. Statistical measures and analysis All the cell records included in the construction of the new decision algorithm had durations of spontaneous activity lasting for at least 30 s and usually 60 s. Our previous decision algorithm used the following statistical measures: CVlog which is the coefficient of variation of interspike intervals (ISI) expressed using the natural logarithm of the intervals in milliseconds; the average firing frequency; CV2 which is the mean of two times the absolute difference between two successive ISIs divided by the sum of both intervals (Holt et al., 1996); the value of the 5th percentile of ISIs; and the median ISI. All 48 measures described in Ruigrok et al. (2011), including the median absolute difference from the median ISI (MAD), were calculated and analyzed. The measures that best distinguished Purkinje cells from other cerebellar neurons were first determined by a classification program called CTree, a Microsoft Excel-based classification tree written by Angshuman Saha (available at https://www.sites.google.com/site/sayhello2angshu/dminexcel), that uses the C4.5 algorithm. This program also allowed us to determine the effectiveness of the simultaneous use of multiple measures. Then, visual inspection of scatter plots of all combinations of these best measures were used to determine the most selective for Purkinje cell identification and for identification of “border zones”, which allowed for avoiding false-positives. The D’Agostino–Pearson omnibus test was used to test for normal distribution. This test uses the statistic DP that approximates a Chi-squared (2 ) distribution with 2 degrees of freedom if the population is normally distributed. The formula for DP is DP = (skewness/SES)2 + (kurtosis/SEK)2 , with SES being the standard error of skewness and SEK being the standard error of kurtosis (D Agostino et al., 1990).
Recordings of Purkinje cells with complex spikes (n = 110) were split into three groups. The first group (n = 39) consisted of Purkinje cell recordings from 16 anesthetized rabbits and this group was used to investigate all statistical measures and create the new decision algorithm. A second group (n = 36) consisted also of Purkinje cell recordings from another set of 11 anesthetized rabbits and this group was used to independently determine the effectiveness of the new decision algorithm. A third group (n = 35), consisting of Purkinje cell recordings from two awake rabbits, was used to determine whether the new algorithm was effective for recordings from awake rabbits. Finally, unclassified cells that did not display complex spikes at any time during the recording were analyzed using the new algorithm. Of these cells, 182 were recorded from 29 ketamine–xylazine anesthetized rabbits of which 17 are not included above, and 170 were recorded from the two awake rabbits.
3. Results 3.1. Purkinje cells without detectable complex spikes Extracellular Purkinje cell recordings do not always show a detectable complex spike, presumably due to the relative positions of the recording electrode and the Purkinje cell. In some cases, such cells can be identified as Purkinje cells after penetration by the recording glass microelectrode and the subsequent appearance of the complex spike (Eccles et al., 1966; Hensbroek et al., 2011; Maekawa and Simpson, 1973; Maruta et al., 2007). Fig. 1 shows raw data from two penetrated Purkinje cells. The first Purkinje cell showed complex spikes (indicated by the black dot) and simple spikes both before (Fig. 1A) and after penetration (Fig. 1B). The second Purkinje cell showed only simple spikes in the extracellular record (Fig. 1C) and as a result, this cell could easily have been mistaken for an interneuron. However, after penetration complex spikes were revealed, thereby identifying this cell as a Purkinje cell (1D). After intracellular penetration the observed complex spike usually reflects only the postsynaptic activity due to the climbing fiber input. These responses constitute a unique signature due to the brief high frequency bursts of action potentials (up to seven action potentials at 500–800 Hz) that often comprise a single climbing fiber input (Maruta et al., 2007). Thus, in the intracellular configuration positive identification of Purkinje cells does not require a pause in simple spikes after the complex spike while the simple spikes are degraded to an injury discharge. In other cases, Purkinje cells that initially did not show a complex spike can sometimes be identified after making a small
176
R.A. Hensbroek et al. / Journal of Neuroscience Methods 232 (2014) 173–180
Fig. 2. Three examples of cells that initially showed no complex spikes but whose complex spikes were revealed after a small extracellular movement of the microelectrode. (A and B) Instantaneous frequencies of simple spikes (A) and complex spikes (B) isolated using spike sorting techniques. The microelectrode was moved at t = 22 s (arrowhead). (C) Extracellular record is from t = 7 to t = 10 s with a 60 ms section blowup. (D) Extracellular record is from t = 26 s to t = 29 s, also with a 60 ms section blowup. A suspected complex spike in the blowup of (C) and complex spikes in (D) are marked with a dot. The firing statistics before moving the electrode were: CVlog = 0.110, AvgFreq = 37.0, MAD = 0.00410, CV2 = 0.274, ISIperc05 = 0.0169, ISImed = 0.0256, and the firing statistics after movement were: CVlog = 0.0809, AvgFreq = 35.8, MAD = 0.00377, CV2 = 0.238, ISIperc05 = 0.0187, ISImed = 0.0264.(E) A second cell that initially did not show complex spikes. The rabbit from which this cell was obtained was anesthetized with urethane. (F) Record from the same cell as (E) after a small movement of the electrode that resulted in the appearance of complex spikes. The firing statistics before moving the electrode were: CVlog = 0.0779, AvgFreq = 31.5, MAD = 0.00509, CV2 = 0.310, ISIperc05 = 0.0208, ISImed = 0.0300, and the firing statistics after movement were: CVlog = 0.0808, AvgFreq = 28.4, MAD = 0.00559, CV2 = 0.323, ISIperc05 = 0.0226, ISImed = 0.0333. (G). A third cell that initially did not show complex spikes. (H) Record from the same cell as G after a small movement of the electrode that resulted in the appearance of complex spikes. The firing statistics before moving the electrode were: CVlog = 0.0623, AvgFreq = 65.8, MAD = 0.00166, CV2 = 0.194, ISIperc05 = 0.0115, ISImed = 0.0150, and the firing statistics after movement were: CVlog = 0.0703, AvgFreq = 65.5, MAD = 0.00170, CV2 = 0.221, ISIperc05 = 0.0110, ISImed = 0.0151. Signals were bandpass filtered between 10 Hz and 10KHz. Note that the time scales are not the same for all traces.
(30–60 m) extracellular movement of the microelectrode to reveal the complex spike, as shown by the exemplary recordings from three Purkinje cells in Fig. 2. Initially, the first Purkinje cell did not show identifiable complex spikes (Fig. 2B and C) but after movement of the microelectrode at time t = 22 s (indicated by the arrowheads) the complex spikes appeared and could be clearly distinguished from simple spikes (Fig. 2B and D). The identity of this cell was confirmed by a simple spike pause after complex spike occurrence. An interesting observation in this particular record is that during the first 22 s several very short interspike intervals (ISIs) were observed, resulting in the high instantaneous frequencies seen initially in Fig. 2A. Such short ISIs are unusual for Purkinje cell simple spikes. After the complex spikes became visible these short ISIs abruptly disappeared. Fig. 2C shows a detail of such a short ISI. The second spike (marked with a dot) is similar to the other spikes and were it not for the very short ISI this particular spike would have remained unnoticed. Further analysis showed that these short-interval spikes were followed by a transient absence of spikes (e.g. Fig. 2C, top), similar to the simple spike pause after a complex spike. Therefore, we propose that these short-interval spikes were actually complex spikes which, due to the position of the microelectrode in relation to the cell, appear similar to
simple spikes. The occasional high instantaneous frequencies seen in Fig. 2A would then be the result of the independent occurrence of complex spikes among the simple spikes. This finding underscores the need for stable statistics for spike activity in a decision algorithm because simple spikes may be contaminated by complex spikes. A second example of a Purkinje cell that initially did not show complex spikes is shown in Fig. 2E and F. Before movement of the electrode only one type of spike was visible (2E), but after movement the complex spike and the associated pause in simple spike activity appeared in the extracellular record (2F). A third example is shown in Fig. 2G and H. Similar to the previous example, initially only one type of spike was visible (Fig. 2G), but after movement of the electrode the complex spike and the associated simple spike pause appeared (2H). Unlike the Purkinje cell in Fig. 2A–D, the two Purkinje cells in Fig. 2E–H did not display high frequency events before electrode movement, showing that undetectable complex spikes, appearing similar to simple spikes, are not always found. In three animals, we made 14 attempts to reveal the complex spikes for cells that initially had only a spike pattern closely resembling that of simple spike activity. In four attempts (29%) the
R.A. Hensbroek et al. / Journal of Neuroscience Methods 232 (2014) 173–180
complex spike was revealed extracellularly and in two attempts (14%) the complex spike was revealed intracellularly by penetration. In the other eight attempts (57%) we were unable to reveal the complex spike. In a related experiment in 14 animals (9 of which were included only in this experiment) we tried to estimate how common it is that Purkinje cell simple spikes can be recorded without an accompanying complex spike. To do this we first obtained Purkinje cell recordings with a clear signal of both the complex spike and simple spike and subsequently slowly moved the recording electrode along its approach track over distances up to 60 m. In 59% of those instances (13/22 Purkinje cells) the complex spike signal was selectively lost while the simple spike signal remained. The other 41% of Purkinje cells (9/22 cells) showed a simultaneous decline of both the simple and complex spike. This experiment included one rabbit that was anesthetized with urethane and one rabbit that was anesthetized with urethane and alpha chloralose. These two rabbits provided records of six Purkinje cells, two of which selectively lost their complex spike thereby demonstrating that this phenomenon is not restricted to rabbits anesthetized with ketamine–xylazine. These findings indicate the existence of Purkinje cells that have areas near them where only simple spikes are visible to the recording electrode. As mentioned in Section 1, our existing decision algorithm (Ruigrok et al., 2011) does not include Purkinje cells, and Figs. 1 and 2 demonstrate how one could confuse a Purkinje cell without a detectable complex spike for an interneuron. If such a recording is subsequently analyzed using our previously published algorithm, the result would be a “border” classification, or even a false-positive. This situation prompted us to construct a new decision algorithm that correctly identifies both interneurons and Purkinje cells. 3.2. Analysis of Purkinje cell spontaneous simple spike activity Purkinje cell recordings from anesthetized rabbits were analyzed in order to determine variables that could distinguish them from interneurons. Stability against small variations in spike discrimination and the ability to create a clear statistical separation between Purkinje cells and interneurons were key features desired for selecting discriminating variables. For the initial investigation in anesthetized rabbits, we used 39 Purkinje cells from 16 animals that were identified by complex spikes and the associated simple spike pause. Only simple spikes were used for calculating the variables. Visual inspection of the analysis with a C4.5-based decision tree and two-dimensional scatter plots of statistical measures revealed that the spontaneous firing characteristics of the Purkinje cells diverged from those of cerebellar interneurons most notably in the combination of values of MAD and CV2 (Fig. 3). The divergence found with these variables reflects the high firing rate and intermediate regularity of simple spike firing. The MAD and CV2 for the initial Purkinje cell group (n = 39) had averages of 0.00368 and 0.354 with standard deviations of 0.00178 and 0.102, respectively. Neither the MAD nor the CV2 had a normal distribution because both showed moderate to high positive skewness (0.991 for MAD and 0.918 for CV2), and low to moderate positive excess kurtosis (0.391 for the MAD and 1.701 for the CV2). In addition, the MAD and CV2 distributions were tested for deviation from the normal distribution using the D Agostino–Pearson omnibus test (D Agostino et al., 1990), and both measures showed significant differences from normality (MAD: DP = 7.14, p < 0.05; CV2: DP = 11.17, p < 0.005). 3.3. The new decision algorithm The ability of the combination of MAD and CV2 to discriminate Purkinje cell activity from that of virtually all interneurons (Fig. 3) suggests that Purkinje cells might be selectable at an early decision
177
Fig. 3. MAD and CV2 values calculated from spontaneous activity (see Section 2 for details). Shown are 86 morphologically identified interneurons from 27 anesthetized rats and 5 anesthetized rabbits (Ruigrok et al., 2011), and 39 Purkinje cells identified by their complex spike from 16 ketamine–xylazine anesthetized rabbits. Granule cells are not shown because they are easily classified in the first step in the analysis sequence (see text). The dashed lines indicate the decision boundaries between Purkinje cells and border cells and between border cells and interneurons in the new decision algorithm (see also Fig. 4).
point in our new decision algorithm. Further analysis with a C4.5-based decision tree and visual inspection supported this view and showed that our earlier algorithm could be successfully extended by including an additional early decision step in the analysis sequence for Purkinje cells based on CV2 and MAD (Fig. 4). For some cell recordings discrimination using these two measures was still difficult, which prompted us to create “border zones”, where cells that were deemed unclassifiable would be placed to avoid false-positives and thereby improve the reliability of the new decision tree. A brief description of the analysis sequence for the new tree: Step 1: Granule cells were found to be either nearly silent or fire very irregularly with both very short (∼1 ms) and very long (∼seconds) interspike intervals. Thus, recordings were classified as granule cells when they had a CVlog greater than 0.38 or an average frequency smaller than 0.5 Hz. Cells with a CVlog smaller than 0.34 and an average frequency larger than 0.6 Hz were continued in the algorithm. Cells not captured by these criteria were deemed unclassifiable and called border cells. Step 2: Purkinje cell simple spikes showed a combination of a high firing rate and an intermediate regularity. Of the cells that were continued in the algorithm from step 1, those classified as Purkinje cells had a MAD smaller than 0.008 and a CV2 larger than 0.2. Cells with a MAD larger than 0.01 or a CV2 smaller than 0.15 were continued in the algorithm. The rest were border cells. Step 3: Unipolar brush cells showed regular firing. If a cell that was continued from step 2 had a CV2 smaller than 0.24 then it was identified as a unipolar brush cell. Cells with a CV2 larger than 0.28 were continued in the algorithm and the rest were border cells. Step 4: Basket and stellate cells often fire irregularly. They can have a high proportion of short interspike intervals, but they do not fire in high frequency bursts. Cells that were continued from step 3 and had either a CVlog larger than 0.17 or a 5th percentile ISI smaller than 22 ms were identified as a basket or stellate cell. Cells that had a CVlog smaller than 0.15 and a 5th percentile ISI larger than 44 ms were continued in the analysis; the rest were deemed border cells. Step 5: The large majority of Golgi cells are rather irregular and fire at a relatively slow rate, but not slower than a median ISI of 300 ms.
178
R.A. Hensbroek et al. / Journal of Neuroscience Methods 232 (2014) 173–180
Fig. 4. New decision algorithm derived from a previously published algorithm (Ruigrok et al., 2011). We added an additional step to identify Purkinje cells using only their spontaneous simple spike activity. This new classification step uses a combination of the MAD and CV2 statistics, which are sensitive to the high firing rate and the intermediate regularity of simple spike activity.
However, a particular class of slowly firing basket and stellate cells had median ISIs larger than 320 ms. The cells with intermediate values of the median ISI between 300 and 320 ms became border cells. 3.4. Validation of the new decision algorithm When the new algorithm was applied to the initial group of interneurons (n = 86) and Purkinje cells (n = 39) it showed an accuracy of 78% (97 out of 125 cells), with 20% (25 out of 125) considered as border cells, and 2% (3 out of 125) misclassified (interneurons: 61 correct, 22 border cells and 3 misclassified; Purkinje cells: 36 correct, 3 border cells and 0 misclassified). The 2 cells in Fig. 1 were classified as Purkinje cells. Of the 3 cells in Fig. 2, the first (Fig. 2A–D) and second (Fig. 2E and F) were classified as Purkinje cells both before and after movement. The third cell (Fig. 2G and H) was only classified as a Purkinje cell after movement but was classified as a border cell before movement. That situation shows the need for a border zone as this cell would have been a false-positive for another cell type if no border zones were present. Next, the validity of the new decision algorithm was independently tested using two new groups of Purkinje cells identified by their complex spikes. One group came from 11 anesthetized rabbits and another from two awake rabbits (Table 1). Out of the 36 Purkinje cells from the anesthetized rabbits, 29 (81%) were correctly classified, 1 (3%) was misclassified and 6 (17%) were designated as border cells. For the awake rabbits, out of the 35 Purkinje cells, 30 (86%) were correctly classified, 1 (3%) was misclassified and 4 (11%) were designated border cells. These numbers are similar to those for the 39 Purkinje cells used in construction of the new decision algorithm above and strengthen our view that the new algorithm is unbiased and does
Table 1 Test of the new decision algorithm by using two additional groups of Purkinje cells. One group (n = 36) was recorded in anesthetized rabbits and another group (n = 35) was recorded in awake rabbits. All these Purkinje cells were first identified by the presence of their complex spikes and the associated pause in simple spike activity.
Granule cells Purkinje cells Unipolar brush cells Basket or stellate cells Golgi cells Border cells
Anesthetized (n = 36)
Awake (n = 35)
– 81% (29) – 3% (1) – 17% (6)
– 86% (30) – 3% (1) – 11% (4)
not result in over-fitting. Furthermore, the high accuracy of identification for Purkinje cells recorded in awake rabbits highlights the wider applicability of the algorithm. Finally, the new decision algorithm was applied to extracellular recordings of cells that did not show a complex spike. One set (n = 182) was from 29 ketamine–xylazine anesthetized rabbits and the other set (n = 170) was from 2 awake rabbits (Table 2). In Table 2 Application of the new decision algorithm to two groups of unidentified neurons. One group (n = 182) was recorded in anesthetized rabbits and the other group (n = 170) was recorded in awake rabbits.
Granule cells Purkinje cells Unipolar brush cells Basket or stellate cells Golgi cells Border cells
Anesthetized (n = 182)
Awake (n = 170)
23% (42) 19% (34) 13% (24) 14% (25) 11% (20) 20% (37)
22% (38) 15% (25) 14% (24) 13% (22) 10% (17) 26% (44)
R.A. Hensbroek et al. / Journal of Neuroscience Methods 232 (2014) 173–180
both sets of recordings a substantial number of cells (19% and 15%, respectively) were classified as Purkinje cells even though they did not show a complex spike, which underscores the need for a Purkinje cell identification step in the decision tree. 4. Discussion Using only statistics derived from spontaneous spike activity, we devised an algorithm consisting of five sequential steps that enabled correct identification of interneurons and Purkinje cells recorded from the rabbit vestibulocerebellum. The ability to identify Purkinje cells simply from their spontaneous simple spike activity is an improvement over our previous decision algorithm because before Purkinje cells could only be identified by the presence of their complex spikes. 4.1. Alternative identification methods Identification of some types of cerebellar cortical interneurons on the basis of their spike activity has been proposed before (Edgley and Lidierth, 1987; Van Kan et al., 1993; Vos et al., 1999). Only more recently with the advent of juxtacellular and intracellular labeling techniques and subsequent morphological identification of cerebellar interneurons (Barmack and Yakhnitsa, 2008; Holtzman et al., 2006; Jorntell and Ekerot, 2003; Simpson et al., 2005), has it become feasible to try to create spontaneous spike activity-based identification methods (Ruigrok et al., 2011; Van Dijck et al., 2013). Such methods are of importance especially in situations where juxtacellular labeling is not routinely feasible such as in awake animals. This is not to say that information unrelated to spontaneous activity is not valuable. In our opinion, the presence of a complex spike with the associated pause in simple spike firing remains the “gold” standard for Purkinje cell identification (Granit and Phillips, 1956). Background complex spikes may also be used to support identification of basket and stellate cells because the polarity of the complex spike is predominantly negative when recorded in the molecular layer (Eccles et al., 1967). In fact, a recently published method (Van Dijck et al., 2013) utilizes such information from complex spikes to distinguish between molecular and granular layer neurons. Subsequently, this method uses a probabilistic model to identify neurons using measures of firing frequency and irregularity that are not unlike those presented here. The key difference from our method is the initial sorting by the experimenter into either molecular or granular layer recordings, which is necessary because the probabilistic method is unable to discriminate molecular layer basket/stellate cell recordings when mixed with granular layer Golgi cell and granule cell recordings. As mentioned above, one can probably be quite certain about recording from the molecular layer when a negative complex spike is visible in the local field potential. Unfortunately, complex spikes in the local field potentials are not always visible especially when using the small-tipped electrodes necessary for recording small interneurons such as granule cells. Certainly, the absence of a negative complex spike is no guarantee that the electrode is in the granular layer. Regrettably, Van Dijck et al. (2013) offer no systematic method or validity analysis for identifying the recording site layer, which is problematic given that the overall success rate of their method is the product of the reliability of selecting the correct anatomical layer and the efficacy of the probabilistic method itself. This issue limits the practical application of their method for recordings deep to the cerebellar surface. 4.2. Applicability Various factors such as species, types of anesthesia and cerebellar region may all influence the firing behavior of neurons,
179
but signature patterns derived from spontaneous activity may still show enough stability for identification analysis. For instance, while it cannot be excluded that ketamine–xylazine anesthesia may have reduced spontaneous granule cell activity (Hevers et al., 2008), activity patterns in the granular layer have been reported to be similar in awake and ketamine anesthetized rats (Hartmann and Bower, 2001; Santamaria et al., 2007). In addition, Van Dijck et al. (2013) used neurons from different cerebellar regions in ketamine–xylazine anesthetized mice, urethane anesthetized rats and decerebrate cats and did not report marked differences in the distributions of their statistics. Similarly, the recorded activity signatures of spontaneously firing morphologically identified interneurons in the vestibulocerebellum of ketamine–xylazine anesthetized rats and rabbits were both shown to adhere to the same decision algorithm (Ruigrok et al., 2011). Furthermore, recordings from unidentified neurons (Table 2) showed similar distributions whether they were from anesthetized or awake rabbits, indicating that the statistical measures and cell populations from anesthetized and awake rabbits did not differ markedly. Finally, Purkinje cells identified by complex spikes recorded from awake rabbits were correctly discriminated by the new algorithm with a similar percentage as Purkinje cells from anesthetized rabbits. Taken together, these findings strengthen our view that the new decision algorithm may have a wide applicability. 5. Conclusion We have improved upon our previous cerebellar interneuron classification algorithm by adding measures for classification of Purkinje cells using only their spontaneous simple spike activity recorded in the awake and anesthetized rabbit. We suggest that our measures are valid not only for neurons in the rabbit vestibulocerebellum, but also for neurons in the vestibulocerebellum of other species. Acknowledgements This work was supported by grants from the NIH (NS-13742) and the Dutch Ministry of Health, Welfare and Sport (TJHR). References Barmack NH, Yakhnitsa V. Functions of interneurons in mouse cerebellum. J Neurosci 2008;28:1140–52. D Agostino RB, Belanger A, D Agostino RB Jr. A suggestion for using powerful and informative tests of normality. Am Stat 1990;44(4):316–21. De Zeeuw CI, Wylie DR, Stahl JS, Simpson JI. Phase relations of Purkinje cells in the rabbit flocculus during compensatory eye movements. J Neurophysiol 1995;74:2051–64. Eccles JC, Ito M, Szentagothai J. The cerebellum as a neuronal machine. Berlin: Springer-Verlag; 1967. Eccles JC, Llinas R, Sasaki K. The excitatory synaptic action of climbing fibers on the Purkinje cells of the cerebellum. J Physiol 1966;182:268–96. Edgley SA, Lidierth M. The discharges of cerebellar Golgi cells during locomotion in the cat. J Physiol 1987;392:315–32. Granit R, Phillips CG. Excitatory and inhibitory processes acting upon individual Purkinje cells of the cerebellum in cats. J Physiol 1956;133(3):520–47. Hartmann MJ, Bower JM. Tactile responses in the granule cell layer of cerebellar folium crus IIa of freely behaving rats. J Neurosci 2001;21:3549–63. Hensbroek RA, Van Beugen BJ, Maruta J, Ruigrok TJH, Simpson JI. Discriminating Purkinje cells without using complex spikes. Soc for Neurosci 41st Annu Meet 2011, 922.15. Hevers W, Hadley SH, Luddens H, Amin J. Ketamine, but not phencyclidine, selectively modulates cerebellar GABAA receptors containing ␣6 and ␦ subunits. J Neurosci 2008;28:5383–93. Holt GR, Softky WR, Koch C, Douglas RJ. Comparison of discharge variability in vitro and in vivo in cat visual cortex neurons. J Neurophysiol 1996;75:1806–14. Holtzman T, Rajapaksa T, Mostofi A, Edgley SA. Different responses of rat cerebellar Purkinje cells and Golgi cells evoked by widespread convergent sensory inputs. J Physiol 2006;574(2):491–507.
180
R.A. Hensbroek et al. / Journal of Neuroscience Methods 232 (2014) 173–180
Jorntell H, Ekerot CF. Receptive field plasticity profoundly alters the cutaneous parallel fiber synaptic input to cerebellar interneurons in vivo. J Neurosci 2003;23(29):9620–31. Maekawa M, Simpson JI. Climbing fiber responses evoked in vestibulo-cerebellum of rabbit from visual system. J Neurophysiol 1973;36:649–66. Maruta J, Hensbroek RA, Simpson JI. Intraburst and interburst signaling by climbing fibers. J Neurosci 2007;27(42):11263–70. Ruigrok TJH, Hensbroek RA, Simpson JI. Spontaneous activity signatures of morphologically identified interneurons in the vestibulocerebellum. J Neurosci 2011;31(2):712–24. Santamaria F, Tripp PG, Bower JM. Feedforward inhibition controls the spread of granule cell-induced Purkinje cell activity in the cerebellar cortex. J Neurophysiol 2007;97:248–63.
Simpson JI, Hulscher HC, Sabel-Goedknegt E, Ruigrok TJ. Between in and out: linking morphology and physiology of cerebellar cortical interneurons. Prog Brain Res 2005;148:329–40. Van Dijck G, Van Hulle MM, Heiney SA, Blazquez PM, Meng H, Angelaki DE, et al. Probabilistic identification of cerebellar cortical neurones across species. PLoS One 2013;8(3):e57669. Van Kan PL, Gibson AR, Houk JC. Movement-related inputs to intermediate cerebellum of the monkey. J Neurophysiol 1993;69(1):74–94. Vos BP, Volny-Luraghi A, De Schutter E. Cerebellar Golgi cells in the rat: receptive fields and timing of responses to facial stimulation. Eur J Neurosci 1999;11(8):2621–34.