HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS		QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) All Versions of this Article: 94/1/567    most recent 00896.2004v1 Alert me when this article is cited Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in Web of Science Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Web of Science (16) Citing Articles via Google Scholar Google Scholar Articles by Shinomoto, S. Articles by Fujita, I. Search for Related Content PubMed PubMed Citation Articles by Shinomoto, S. Articles by Fujita, I.

Regional and Laminar Differences in In Vivo Firing Patterns of Primate Cortical Neurons

¹Graduate School of Science, Kyoto University, Kyoto; ²Graduate School of Frontier Biosciences, Osaka University, Osaka; and ³Hitachi, Totsuka, Yokohama, Japan

Submitted 30 August 2004; accepted in final form 7 March 2005

	ABSTRACT

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX GRANTS ACKNOWLEDGMENTS REFERENCES

 
The firing rates of cortical neurons change in time; yet, some aspects of their in vivo firing characteristics remain unchanged and are specific to individual neurons. A recent study has shown that neurons in the monkey medial motor areas can be grouped into 2 firing types, "likely random" and "quasi-regular," according to a measure of local variation of interspike intervals. In the present study, we extended this analysis to area TE of the inferior temporal cortex and addressed whether this classification applies generally to different cortical areas and whether different types of neurons show different laminar distribution. We found that area TE did consist of 2 groups of neurons with different firing characteristics, one similar to the "likely random" type in the medial motor cortical areas, and the other exhibiting a "clumpy-bursty" firing pattern unique to TE. The quasi-regular type was rarely observed in area TE. The likely random firing type of neuron was more frequently found in layers V–VI than in layers II–III, whereas the opposite was true for the clumpy-bursty firing type. These results show that neocortical areas consist of heterogeneous neurons that differ from one area to another in their basic firing characteristics. Moreover, we show that spike trains obtained from a single cortical neuron can provide a clue that helps to identify its layer localization.

	INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX GRANTS ACKNOWLEDGMENTS REFERENCES

 
Neurophysiological studies have succeeded in correlating the firing activity of specific populations of neurons to animal behaviors, defining sites with neuronal activity in particular behavioral contexts as the functional areas corresponding to those behaviors. Although such observations are interesting in themselves, these studies do not necessarily examine neuronal activity in the context of circuitry. To address this question, information regarding not only neuronal activity but also cell type and location of individual neurons is required. Extracellularly recorded spike sequences can provide information that allows us to distinguish between cell types. In the hippocampus, for example, the firing rate of an interneuron is typically higher than that of a pyramidal neuron (Buzsáki et al. 1983; Ranck 1973) and, in the neocortex, there is a difference in the action potential waveform between pyramidal neurons and a specific class of interneurons (Constantinidis et al. 2002; Csicsvari et al. 1998; Mountcastle et al. 1969).

We have recently shown that the analysis of firing patterns can be used to classify neocortical neurons into distinct groups (Shinomoto et al. 2003). In that study, we compared the conventional coefficient of variation (Cv) and a newly introduced measure of time-local variation (Lv) of the recorded sequences of interspike intervals (ISIs). Cv reflects the global variability of an entire neuronal spike sequence and is sensitive to fluctuations in the firing rate. Lv, on the other hand, reflects the local stepwise variability of ISIs, depends less on the firing rate, and robustly represents the firing characteristics specific to individual neurons. Neurons in the medial motor areas (the presupplementary motor area, supplementary motor area, and rostral cingulate motor area) have been categorized into 2 distinct groups with different values of Lv. One group has an Lv of 0.81 on average and tends to fire in a random fashion and the other group has an Lv of 0.38 whose firing pattern is quasi-regular. The 2 groups of neurons classified according to their Lv values exhibit different responses to the same stimulus, with a significant difference in their onset latency. In contrast, no clear categorization is made on the basis of the Cv distribution.

Different cortical areas share many of the firing statistics, such as the dynamic range of responses and the spike count variance (Shadlen and Newsome 1998). It is unknown, however, whether the classification of neurons in the medial motor areas also applies to other cortices. In addition, it is unclear whether the 2 groups of neurons are intermingled or tend to cluster in separate locations in the gray matter, exhibiting separate localization in different layers, for instance. In the present study, we analyzed spike sequences recorded from area TE of the inferior temporal cortex of macaque monkeys. Histological markings made for each electrode penetration allowed for layer localization of the recorded position. Independently of this laminar location assignment, the recorded spike sequences were analyzed with respect to Cv and Lv. As in the medial motor areas, the distribution of Lv values was bimodal, whereas the distribution of Cv values was unimodal. The distribution peaks for TE neurons differed substantially from those of medial motor cortical neurons. The neurons exhibiting firing patterns with higher Lv values were mostly located in layers II–III, whereas those with lower Lv values were principally found in layers V–VI.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX GRANTS ACKNOWLEDGMENTS REFERENCES

 
Neurophysiological experiment

Here we briefly review the experiment described by Tamura et al. (2004). We applied our present analysis to these data.

Neuronal responses were recorded from area TE of the inferior temporal cortex in 4 anesthetized monkeys (Macaca fuscata; body weight, 5.2–7.5 kg). All the experimental procedures conformed to the guidelines of the National Institutes of Health (1996) and were approved by the animal experiment committee of Osaka University. General experimental procedures have been described elsewhere (Tamura et al. 2004). Monkeys were prepared for repeated recordings through an initial aseptic surgery under sodium pentobarbital surgical anesthesia.

Monkeys were anesthetized with nitrous oxide and isoflurane during recordings. Body temperature was maintained at 37–38°C. End-tidal CO₂ was kept at 4.0–4.5%. Electrocardiogram and arterial oxygen saturation levels were continuously monitored throughout the experiment. Eyes were dilated, covered with preselected contact lenses, and irrigated with saline.

Multiple single-unit recordings were made from area TE using a single-shaft electrode with 7 recording probes (Heptode; Thomas, Esslingen, Germany) (Fig. 1, A and B). Monkeys were paralyzed with pancuronium bromide during electrophysiological recordings. Recordings were made at intervals of 300 µm along a penetration axis. In every penetration, sampling was made throughout the gray matter from layer 2 to layer 6. All neurons encountered were recorded and analyzed. Isolation and classification of spikes from recorded signals were carried out off-line by an automated method (Kaneko et al. 1999).

View larger version (107K): [in this window] [in a new window]   FIG. 1. A and B: recording sites are indicated as a shaded region in the lateral view of the right hemisphere (A) and as a box in a frontal section (B) taken at the level indicated by the vertical line in A. C: photograph of a Nissl-stained section of a portion of area TE indicated by the box in B. Arrows indicate the 3 electrolytic lesions made along a recording penetration. amts, anterior middle temporal sulcus; h, hippocampus; ios, inferior occipital sulcus; pmts, posterior middle temporal sulcus; sts, superior temporal sulcus.

For each penetration, 2 or 3 small electrolytic lesions were made by passing a cathodal current of 10 µA through the electrode tip for 10 s to allow later histological verification of the recording sites (Fig. 1C). After completion of the final recordings, monkeys were deeply anesthetized using sodium pentobarbital and perfused transcardially. The brain was cut into 100-µm-thick serial sections. All the electrolytic lesions were successfully recovered from Nissl-stained sections. We determined the laminar positions of the recording sites using a combination of the electrolytic lesions and the electrode manipulator readings noted for each recording site. Like other neocortical areas in monkeys, area TE exhibits a clear 6-layer organization (Fig. 1C). Identification of layers was based on the description of Fujita and Fujita (1996).

Visual stimuli

The stimulus set consisted of 64 visual stimuli (53 geometrical shapes and 11 photographs of natural or man-made objects; see Fig. 1 of Tamura et al. 2004). Images were 4° in visual angle. Visual stimuli were presented individually for 1 s with an interstimulus interval of 1 s at the center of the receptive field against a homogeneous gray background. The stimuli were presented once in a pseudorandom order within a stimulus presentation block. Ten blocks were presented for each recording site.

Lv as a measure for local variation of interspike intervals

To analyze firing patterns in a standardized manner, the recorded spike sequences were segmented into fragments consisting of 100 consecutive ISIs. For each neuron, if the total number of segments was 30, we took the first 30 sequences for analysis. Neurons whose spike rates were too low to provide 30 sequences were omitted. The minimum spike rate among the accepted neurons was 2.2 spikes/s. The population mean of the spike rate among accepted neurons was 8.5 spikes/s (SD 8.9; n = 288). Thus in the first of our analyses, we analyzed continuous spike sequences of 100 ISIs without regard to stimulus events; the sequences included spikes during stimulus presentations and spikes during interstimulus intervals.

For each set of 100 consecutive interspike intervals, the coefficient of variation Cv and the measure of local variation Lv were computed. Cv is defined as

(1)

where T and are, respectively, the SD and the mean of the 100 ISIs. For a series of intervals that are independently exponentially distributed, Cv = 1 in the limit of a large number of intervals. For a regular spike sequence in which T_i is constant, Cv = 0.

Lv is defined as

(2)

where T_i is the ith ISI, and n is the number of ISIs (in this study, n = 100) (Shinomoto et al. 2003). The summand is proportional to the square of Cv₂ 2|T₁ – T₂|/(T₁ + T₂), which has been introduced for the purpose of comparing the temporal ISI randomness of neurons (Holt et al. 1996). The factor 3 is included here so that for a series of intervals that are independently exponentially distributed, Lv = 1. For a regular spike sequence in which T_i is constant, Lv = 0. In this way, both Cv and Lv adopt a value of 1 for a sufficiently long Poisson (random) series of events, and a value of 0 for a sequence of perfectly regular intervals. Cv represents the global variability of an entire ISI sequence and is sensitive to firing rate fluctuations. Lv represents the local stepwise variability of ISIs and depends less on the firing rate modulation (Shinomoto et al. 2005).

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX GRANTS ACKNOWLEDGMENTS REFERENCES

 
A total of 288 neurons were analyzed, with 63, 124, 44, and 57 neurons analyzed for monkeys 1–4, respectively. From the values of Cv and Lv computed for 30 sequences of 100 ISIs, the medians were taken as the representative Cv and Lv values for each neuron. For the present data, the median values are almost identical to the mean values for individual neurons, and the following results change little if the median values are replaced with the mean values. The distributions of the median Cv and Lv values for the 288 neurons are plotted in Fig. 2. The distribution of Lv for the summed data from all monkeys exhibits 2 distinct peaks, whereas the Cv shows a unimodal distribution for the summed data (far right panels). The bimodality of the Lv distribution was observed individually for monkeys 1, 3, and 4. The Lv distribution for monkey 2 was not distinctly bimodal, but exhibited a tendency toward bimodality. The component distributions for the neurons in layers II–III, IV, and V–VI are shown as colored broken lines. The distribution of Cv was statistically distinguishable among different layers (Kruskall-Wallis, P < 10^–3) but substantially overlapped with each other. The distribution of Lv values also differed between layers (Kruskall-Wallis, P < 10^–7) and was clearly distinguishable between layer II–III neurons and layer V–VI neurons. The spike sequences recorded from layers II–III and layers V–VI mainly clustered in regions of larger and smaller Lv values, respectively. Layer IV neurons did not show this bias in Lv distribution. These results suggest that neurons in layers II–III and neurons in layers V–VI exhibit different firing patterns.

View larger version (34K): [in this window] [in a new window]   FIG. 2. Distributions of Cv (A) and Lv (B) values for area TE of monkeys 1–4 and the summed data for all monkeys. "Fraction of cells" is the number of cells counted in each bin of width = 0.2, divided by the total number of cells and (=0.2). Data from all layers, layers II–III, layer IV, and layers V–VI are depicted as black solid, blue dot-dashed, green dotted, and red dashed lines, respectively.

We fitted 2-component Gaussian mixture distribution functions to the set of Lv values obtained from electrophysiological recordings (Fig. 3A, APPENDIX). For monkeys 1–4, the centers of the 2 components resulting from this fit were {1.60, 1.07}, {1.50, 0.93}, {1.46, 0.92}, and {1.66, 0.96}, respectively. The 2 Gaussian components fitted to the summed data for all monkeys were centered at 1.56 and 0.98, with weights of 0.55 and 0.45, and SDs of 0.22 and 0.21. For the purpose of categorizing neurons, we used this 2-component Gaussian fit and determined a cutoff value of Lv that defines the classification boundary. Neurons with Lv values above and below this boundary are classified into separate groups. We assume that the classification boundary minimizes the total areas of the 2 Gaussian tails that are on the "wrong" sides, relative to the cutoff . This is mathematically identical to seeking the value of Lv at which the 2 component distributions meet. In the summed data for all monkeys, the classification boundary was 1.25, yielding a theoretical misclassification rate of 8.7%. The optimal cutoff determined through this procedure was nearly identical to the value of the midway point between the centers of the 2 components, (1.56 + 0.98)/2 = 1.27. The following results were obtained by classifying all data from all the monkeys with the cutoff value = 1.25. The results did not change when we adopted the cutoff value of 1.27.

View larger version (50K): [in this window] [in a new window]   FIG. 3. A: 2-component Gaussian distributions fitted to the set of Lv values for monkeys 1–4 and the summed data for all the monkeys. Original data were plotted as histograms, fitted 2 Gaussian distributions were depicted as dotted lines, and the sum of the 2 Gaussian distributions was depicted as a solid line. B: sample raster diagrams of 20 "likely random" firing neurons of 0.6 < Lv < 1.1 and 20 "clumpy-bursty" firing neurons of 1.3 < Lv < 1.9. Each vertical bar indicates time of occurrence of a spike. Each raster contains 20–50 spikes. Clumpy-bursty rasters appear to have considerably fewer than the real number of spikes because the spikes clump together in time.

Note here that Lv = 1 or Cv = 1 does not necessarily guarantee the intervals to be independently exponentially distributed. For example, even a deterministic spike sequence in which short intervals alternate with long intervals with no variation in the sequencing or the 2 interval lengths can yield Lv = 1 or Cv = 1. However, the spike sequences in TE neurons with Lv 1 appeared to be random (Fig. 3B, left). In addition, the spike sequences with Lv 1 tended to exhibit Cv significantly >1 (Fig. 2A). This observation is consistent with the previous analytical and simulation results that Cv is 1 for any time-dependent Poisson (random) processes with rate fluctuation over time (Shinomoto and Tsubo 2001; Shinomoto et al. 2005). The "likely random" firing patterns are also observed in the medial motor area and the prefrontal cortex, whose values of local variation are Lv 0.8 (Shinomoto et al. 2003). "Clumpy-bursty" patterns characterized by large Lv 1.6 are rare in the medial motor areas or the lateral prefrontal cortex. In the medial motor areas, there are neurons that exhibit spike sequences of smaller values of local variation (typically Lv 0.4). Such spike sequences can be distinguished from the "likely random", and were termed "quasi-regular." In striking contrast, few TE neurons showed the quasi-regular patterns characterized by such small values of Lv.

The experiments examining neurons in the medial motor and lateral prefrontal cortices were done on awake, behaving monkeys. We then addressed whether the difference in the anesthetized versus awake monkeys caused the difference in Lv between the present and previous results. We analyzed the spike sequences recorded in area TE of awake monkeys, which gazed at a fixed point on a display (Kumano et al. 2001). TE neurons were tested for binocular disparity embedded in circular patches of random-dot stereograms presented parafovealy at a 2° visual angle. The Lv calculations for the neurons from this experiment showed a bimodal distribution with peaks at 1.0 and 1.6, which is consistent with the results obtained from anesthetized monkeys in the present study.

Lv as a measure for neuron-specific firing characteristics

A measure can be considered specific to an individual neuron if the variation in its value over time for any given neuron is small in comparison with the variation in its value among different neurons. To determine the amount by which Lv, Cv, and the mean of interspike intervals vary in time, we used a scatter graph to plot pairs of values for each measurement obtained from 2 sequences of the 100 ISIs recorded from a single neuron (Fig. 4, A–C). We picked the 1st and 15th sequences from the 30 sequences of the 100 ISIs measured for each of the 288 neurons. The results are not significantly altered by choosing a different pair of sequences. For the present experimental data, the recorded spike rate was typically 8.5 spikes/s, meaning that the time period needed to obtain one sequence of the 100 ISIs is about 100/8.5 12 s, and the time interval between the 1st and 15th sequence is 180 s. Two values of Lv determined from different time periods exhibited a strong correlation (r = 0.85, n = 288). The observed variation of Lv among neurons was larger than the variation within neurons (F = 185.615, P < 0.001; ANOVA). In contrast, Cv as well as exhibited a weaker correlation (r = 0.52 and 0.48, n = 288).

View larger version (15K): [in this window] [in a new window]   FIG. 4. Scattergrams of the local variation Lv (A), the coefficient of variation Cv (B), and the mean interspike interval (C), for 2 sequences of 100 ISIs from different time periods, each sampled from the same neuron. Each dot denotes data from a single neuron. n = 288.

Effects of stimulus events on Lv

The results in Fig. 4 suggest that Lv can be an indicator of an inherent spiking property of neurons. This analysis, however, was performed on spike sequences spanning both stimulus-present and stimulus-absent periods without any consideration of effects of visually driven responses on Lv. We next addressed to what extent Lv depends on stimulus events and how it relates to various aspects of visual responses.

We first asked whether Lv in individual neurons differs between stimulus-present periods and stimulus-absent periods. We computed Lv separately for stimulus periods and interstimulus intervals in each neuron. Because the number of spikes in a single 1-s stimulus period or 1-s interstimulus interval was too small to obtain a reliable estimate of these coefficients, we concatenated spike sequences of stimulus periods and interstimulus intervals, respectively, to construct the stimulus-present and stimulus-absent spike sequences (Fig. 5A).

View larger version (36K): [in this window] [in a new window]   FIG. 5. A: schema for concatenating spike sequences of stimulus periods and interstimulus periods, respectively, to construct stimulus-present and stimulus-absent spike sequences. B: scattergrams of the local variation Lv (left), the coefficient of variation Cv (center), and the mean interspike interval (right), for 2 sequences of 100 ISIs from the stimulus-present and stimulus-absent spike sequences, each derived from the same neuron.

On average, only 15% of the 64 stimuli evoked statistically significant responses, and the other stimuli were ineffective in our TE neurons (Tamura et al. 2004), raising the possibility that most of the "stimulus-present" periods may really be "no-stimulus" periods for the neurons. We then computed Lv separately for effective-stimulus periods (P < 0.05, Wilcoxon signed-rank test) and for interstimulus intervals. This analysis was performed for 194 of the 288 neurons that provided a sufficient number of spikes (30 sequences of 10 ISIs) for calculation of Lv for effective-stimulus periods and for interstimulus intervals. Figure 6 plots Lv and mean ISI () for visual responses during effective stimulus periods and interstimulus intervals of each neuron. Lv measured during the 2 periods showed a strong correlation (r = 0.76 for n = 154 layer II–III and layer V–VI neurons, r = 0.89 for n = 154 + 40 layer IV neurons; both P < 0.001); whether the measurement of Lv is from visual responses or ongoing discharges, individual neurons show a consistent Lv. Lv during effective-stimulus periods was slightly smaller than that during ineffective-stimulus interstimulus intervals in each neuron. Across the population of neurons, neurons with a stronger firing rate (i.e., a smaller ) showed a smaller Lv (r = –0.61).

View larger version (51K): [in this window] [in a new window]   FIG. 6. Scatter plot of Lv and for effective-stimulus periods and interstimulus periods. Each line shows data from each of the 194 neurons analyzed. Blue lines indicate data from layer II–III neurons, green lines those from layer IV neurons, and red lines those from layer V–VI neurons.

In example neurons shown in Fig. 3B, Lv is unlikely related to spike rate transitions at stimulus onsets and offsets. We now addressed the relationship between Lv and sharpness of spike rate transients at stimulus onsets and offsets for a population of neurons. In this analysis, sharpness of the response onset and offset is defined as the slope of linear regression line fitted to portions of peristimulus time histograms spanning 80 to 200 ms after the stimulus onset or offset, respectively. Only responses with significant regression (P < 0.05) were subjected to further analysis. Because the slope was correlated with the firing rate (r = 0.224, P = 0.019 for onset slope; r = –0.323, P = 0.007 for offset slope), and as we have described above, Lv was correlated with the firing rate, we calculated partial correlation coefficient between Lv and the slope by removing the effects of the firing rate (r_xy–z). Analysis of partial correlation coefficients allowed us to examine the net correlation between Lv and the slopes by removing the effect of the firing rate. This analysis indicates that Lv was not correlated with the slopes of the response onset or offset (Fig. 7, A and B; r_xy–z = 0.12, P = 0.200, n = 110 for onset slope; r_xy–z = 0.24, P = 0.055, n = 68 for offset slope).

View larger version (18K): [in this window] [in a new window]   FIG. 7. A: relationship between Lv and the slope of response onset. Three points with a large slope are not included in the graph. B: relationship between Lv and the slope of response offset. See text for definition of the slope of response onset and offset. C: relationship between Lv and the number of stimuli that evoked responses >25% of the maximal response. Correlation coefficient or partial correlation coefficient is provided as r or rxy–z with corresponding P values.

The results indicate that neurons retain the firing characteristics captured by Lv to a large extent regardless of stimulus events and associated transient changes in firing rates. However, Lv did negatively correlate with the firing rate both within a neuron and across a population of neurons. This may be partly because ISIs inevitably take a smaller range of variation with an increase in the overall firing rate.

Lv and burst firing

We next analyzed the frequency of burst firings. We defined 2 consecutive short intervals <5 ms each as a "unit burst," and calculated the rate of bursts R_b as the number of such "unit bursts" divided by the total number of 2 consecutive intervals n – 1, for a single sequence of 100 ISIs. A Poisson process with a mean ISI of 120 ms (i.e., 8.5 spikes/s) is expected to yield a very low rate of bursts

However, our recorded spike sequences exhibited R_b values significantly larger than this (Fig. 8A; range 0–0.67, mean 0.034, median 0.010, n = 8,640 sequences = 30 sequences x 288 neurons). A pair of R_b values determined for 2 sequences of the 100 ISIs obtained at different time periods from a single neuron were correlated with each other (r = 0.67, P < 0.0001, n = 288; Fig. 8B), indicating that bursting is also a fairly specific character of individual neurons. Lv was also correlated with the R_b (r = 0.25, P < 0.0001; Fig. 8C); neurons with larger Lv more often show the burst firing defined above than neurons with smaller Lv.

View larger version (13K): [in this window] [in a new window]   FIG. 8. A: histogram of the rates of bursts Rb for all recorded sequences of 100 ISIs. Although the data ranged from 0 to 0.67, the number of 100 ISI sequences of Rb >0.3 was very small (114 out of 8526, 1.3%) and is not pictured here. B: scattergram of Rb for 2 sequences of 100 ISIs in different time periods, each sampled from the same neuron. C: scattergram of Rb–Lv for the same sequences of 100 ISIs individually sampled from the 288 neurons.

We next examined how GABAergic inhibitory interneurons, which constitute 20–25% of the total population of cortical neurons (Gabbott and Somogyi 1986; Hendry et al. 1987), are distributed over "clumpy-bursty" and "likely random" firing neurons. It is known that neurons producing short-duration action potentials are likely to be interneurons, although the opposite is not necessarily true (Kawaguchi 1995; Tamura et al. 2004). Figure 9A depicts sample action potentials recorded from area TE. We define the spike width as the duration of time from the initial negative trough to the subsequent positive peak. Figure 9B shows the joint distribution of spike widths and the values of Lv for the 288 neurons. A small hump in the spike width distribution with a short-duration of about 0.3 ms may correspond to fast-spiking interneurons. Most neurons with such thin action potentials exhibited "likely random" firing patterns. However, the opposite is not necessarily true; "likely random" firing neurons did not necessarily produce thin action potentials. Figure 9C depicts the spike width distribution for different cortical layers. All layers contain neurons of thin and thick action potentials, but the layers V–VI contain relatively larger fraction of neurons of thin action potentials than other layers.

View larger version (31K): [in this window] [in a new window]   FIG. 9. A: 2 examples of spike shapes with short and long widths. B: distribution of the spike widths and the Lv values of the 288 neurons. C: distributions of the spike widths from all layers, layers II–III, layer IV, and layers V–VI, depicted as solid, dot-dashed, dotted, and dashed lines, respectively.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX GRANTS ACKNOWLEDGMENTS REFERENCES

Comparison of Lv with the mean and the conventional coefficient of variation (Cv) of ISIs indicates that Lv represents the spiking characteristics specific to individual neurons more reliably than the other measures we tested. When we determined Lv values for 2 sequences of 100 ISIs, each selected from recordings of individual TE neurons, they were strongly correlated (r = 0.85). Similar correlations of Lv values have been seen in the medial frontal cortex, or medial motor areas (r = 0.78–0.85), although lateral frontal cortex neurons show a weaker correlation (r = 0.59) (Shinomoto et al. 2003). More important, the bimodal distribution of Lv with peaks at Lv = 1.0 and 1.6 was preserved under drastically different experimental conditions, one in which responses of TE neurons were tested with a variety of shapes and photographs in anesthetized monkeys (Tamura et al. 2004) and the other in which they were tested using a range of disparities embedded in dynamic random dot stereograms in awake, fixating monkeys (Kumano et al. 2001). These results suggest that Lv reflects an aspect of the in vivo firing pattern of these neurons that is largely unaffected by the type of visual stimuli presented and the experimental conditions.

The neuron-specific firing patterns may be intrinsically determined by the membrane properties of individual cell types. Cortical neurons produce specific firing patterns in response to intracellularly applied depolarizing current pulses, indicating that this aspect of the firing characteristic is intrinsic to cells (e.g., Kawaguchi 1995; McCormick et al. 1985; for reviews, see Amitai and Connors 1995; Connors and Gutnick 1990). However, it is also possible that the input-feeding circuit differs among TE neurons, and that this contributes to the differences in firing patterns captured by Lv measurements. The microorganization of afferent input pathways differs between cells within a cortical area or even cells residing in the same layer of the same cortex (Callaway 2002; Thomson and Deuchars 1997; Thomson and Morris 2002).

Other classification schemes of cortical firing patterns

Previous work has classified cortical neurons according to firing type under both in vitro and in vivo conditions (e.g., Gray and McCormick 1996; Kawaguchi 1995; McCormick et al. 1985; Mountcastle et al. 1969; Nowak et al. 2003; Simons 1978). By applying cluster analysis to intracellular responses to pulses of electric current, Nowak et al. (2003) classified neurons in the cat primary visual cortex into regular spiking, intrinsic bursting, chattering, and fast spiking types. The first 3 represent spiny pyramidal and stellate cells and are presumed to be excitatory neurons, whereas the last type represents aspinous or poorly spinous neurons and is presumed to be inhibitory. In addition, Kawaguchi (1995) classified nonpyramidal neurons in layer II/III of the rat frontal cortex into fast-spiking cells, late-spiking cells, low-threshold spike cells, and regular-spiking nonpyramidal cells. These physiological types correspond to specific classes of inhibitory interneurons with distinct morphological and neurochemical characteristics. It is not readily obvious if any of these cell types correspond to any of the cell types defined by Lv for extracellularly recorded spike trains. At present, we do not see any simple correlation between the 2 classification schemes. For example, chattering cells share some features with "clumpy-bursty" firing neurons, in that they fire in bursty fashion and are distributed predominantly in layer II/III (Nowak et al. 2003). Chattering cells, however, have shorter-duration action potentials, whereas "clumpy-bursty" firing neurons have broader action potentials (Fig. 9).

Layer distribution of "clumpy-bursty" and "likely random" firing neurons

Layer localization of neurons recorded in physiological experiments provides an important clue to understanding how information is processed in the cortex. However, except for the primary visual cortex where the laminar distribution of receptive field properties is well documented (Snodderly and Gur 1995), it is usually difficult to determine during such experiments from which layer the neuronal recording is being made. The differential, although not perfectly distinct, distribution of "clumpy-bursty" and "likely random" firing neurons in layers II–III versus layers V–VI can help in identification of the recorded layer. In particular, simultaneous recordings from multiple neurons at a site would make this laminar estimation more reliable. Making a linkage between spiking patterns based on Lv and anatomically defined neuronal type, by using juxtacellular labeling techniques for instance, will be an important research direction in future studies.

Heterogeneity of cortical areas

The distribution patterns of Lv for neurons in area TE and the medial motor areas are both bimodal, but the peak positions differ markedly between them (Fig. 10). We showed that "clumpy-bursty" and "likely random" firing neurons in area TE are characterized by Lv = 1.56 and Lv = 0.98, respectively, whereas previous work showed that 2 types of neurons in the medial motor areas are characterized by Lv = 0.81 (likely random) and Lv = 0.38 (quasi-regular) (Shinomoto et al. 2003). The Lv distribution of the lateral prefrontal cortex is not distinctly bimodal, but it is also best fit by 2 Gaussian distributions of the mean Lv = 0.83 and Lv = 0.58 (Shinomoto et al. 2003). All 3 areas thus contain a neuronal population with Lv close to 1, suggesting that the random-firing type is common across cortical areas. In contrast, the "clumpy-bursty" firing neurons observed in area TE are rarely observed in the medial motor and lateral prefrontal cortices. Conversely, the "quasi-regular" firing patterns observed in the medial motor areas are rarely observed in area TE.

View larger version (20K): [in this window] [in a new window]   FIG. 10. Various types of spike sequences categorized according to Lv values. Centers of the fitted 2-component Gaussian distributions of Lv for the medial motor areas MM, the prefrontal cortex PF, and area TE of the inferior temporal cortex are located at {0.8 (likely random), 0.4 (quasi-regular)}, {0.8 (likely random), 0.6 (quasi-regular)}, and {1.6 (clumpy-bursty), 1.0 (likely random)}, respectively. Data for MM and PF are from Shinomoto et al. (2003).

According to morphological criteria, the same basic types of neurons are found in all cortical areas. In addition, the proportion of -aminobutyric acid (GABA) neurons to pyramidal neurons is preserved across different areas of the cerebral cortex (Hendry et al. 1987), and no differences in electrophysiological membrane properties have been detected between sensorimotor and anterior cingulate cortices (McCormick et al. 1985). Furthermore, different cortical areas share many of the basic statistical features of firing (Shadlen and Newsome 1998). However, in spite of these similarities, recent studies have begun to elucidate differences in the structural and physiological organization across different areas of the cortex. Morphological characteristics of pyramidal neurons (such as the size and complexity of dendritic arbors; the number of dendritic spines; and the size, shape, and distribution of clusters of horizontal axon arborization) differ between cortical areas, suggesting that the strategy of input sampling and dissemination differ among them (Amir et al. 1993; Benavides-Piccione et al. 2002; Lund et al. 1993; Tanigawa et al. 2005; for reviews, see Elston 2002; Fujita 2002). It remains to be determined whether some of these differences are related to the differences in Lv between the temporal cortex and the medial motor and lateral prefrontal cortices. What roles these different firing patterns might play in different kinds of processing in different cortical areas should also be addressed in future studies. If we could obtain a map of the entire brain with respect to firing characteristics, we would be able to have a better insight into the relationship between firing properties and the functional roles of neurons.

	APPENDIX

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX GRANTS ACKNOWLEDGMENTS REFERENCES

 
Gaussian mixture distribution

Because the distribution of Lv values was bimodal, we fitted a 2-component Gaussian mixture distribution function to this set of values. The methods and the criteria used for 2-component Gaussian distributions were the same as those used by Shinomoto et al. (2003).

The 2-component Gaussian mixture distribution is defined as

(A1)

where N(x|µ, ²) is the Gaussian (normal) distribution of mean µ and variance ², and w₁ > 0 and w₂ = 1 – w₁ > 0 are the weights of the 1st and 2nd component distributions. This distribution is fitted to a data set {x₁, x₂, ..., x_p} by locally maximizing the log-likelihood

(A2)

with respect to 5 parameters, = {µ₁, ₁², µ₂, ₂², w₁}. Fitting of the Gaussian mixture distribution can also be carried out by the expectation maximization (EM) algorithm (Dempster et al. 1977). The distribution functions fitted to the individual data sets are shown in Fig. 3A.

	GRANTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX GRANTS ACKNOWLEDGMENTS REFERENCES

 
This work was supported in part by grants from the Japanese Ministry of Education, Culture, Sports, Science and Technology to S. Shinomoto, I. Fujita, and H. Tamura; a Core Research for the Evolutional Science and Technology Program grant from the Japan Science and Technology Corporation to I. Fujita; and a Grant-in-Aid for the 21st Century COE, "Center for Diversity and Universality in Physics" to S. Shinomoto.

	ACKNOWLEDGMENTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX GRANTS ACKNOWLEDGMENTS REFERENCES

 
Thanks to H. Kaneko, M. Murayama, and K. Miyata for technical assistance in the neurophysiological experiments and to S. Tate for help in data analysis. We also thank J. Tanji and K. Shima for kind help with the recorded area map in Fig. 7.

	FOOTNOTES

 
The costs of publication of this article were defrayed in part by the payment of page charges. The article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

Address for reprint requests and other correspondence: S. Shinomoto, Department of Physics, Graduate School of Science, Kyoto University, Sakyo-ku, Kyoto 606-8502, Japan (E-mail: shinomoto{at}scphys.kyoto-u.ac.jp)

	REFERENCES

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION APPENDIX GRANTS ACKNOWLEDGMENTS REFERENCES

 
Amir Y, Harel M, and Malach R. Cortical hierarchy reflected in the organization of intrinsic connections in macaque monkey visual cortex. J Comp Neurol 334: 19–46, 1993.[CrossRef][Web of Science][Medline]

Amitai Y and Connors BW. Intrinsic physiology and morphology of single neurons in neocortex. In: Cerebral Cortex, edited by Jones EG and Diamond IT. New York: Plenum, 1995, vol. 11.

Benavides-Piccione R, Ballesteros-Yánez I, DeFelipe J, and Yuste R. Cortical area and species differences in dendritic spine morphology. J Neurocytol 31: 337–343, 2002.[CrossRef][Web of Science][Medline]

Buzsáki G, Leung LS, and Vanderwolf CH. Cellular bases of hippocampal EEG in the behaving rat. Brain Res Rev 6: 139–171, 1983.[CrossRef]

Callaway EM. Cell type specificity of local cortical connection. J Neurocytol 31: 231–237, 2002.[CrossRef][Web of Science][Medline]

Conners BW and Gutnick MJ. Intrinsic firing patterns of diverse neocortical neurons. Trends Neurosci 13: 99–104, 1990.[CrossRef][Web of Science][Medline]

Constantinidis C, Williams GV, and Goldman-Rakic PS. A role for inhibition in shaping the temporal flow of information in prefrontal cortex. Nat Neurosci 5: 175–180, 2002.[CrossRef][Web of Science][Medline]

Cox DR and Lewis PAW. The Statistical Analysis of Series of Events. London: Methuen, 1966.

Csicsvari J, Hirase H, Czurkó A, and Buzsáki G. Reliability and state dependence of pyramidal cell–interneuron synapses in the hippocampus: and ensemble approach in the behaving rat. Neuron 21: 179–189, 1998.[CrossRef][Web of Science][Medline]

Dempster A, Laird N, and Rubin D. Maximum likelihood from incomplete data via the EM algorithm. J R Stat Soc Ser B 39: 1–38, 1977.

Elston GN. Cortical heterogeneity: implication for visual processing and polysensory integration. J Neurocytol 31: 317–335, 2002.[CrossRef][Web of Science][Medline]

Fujita I. The inferior temporal cortex: architecture, computation, and representation. J Neurocytol 31: 373–385, 2002.[CrossRef][Web of Science][Medline]

Fujita I and Fujita T. Intrinsic connections in the macaque inferior temporal cortex. J Comp Neurol 368: 467–486, 1996.[CrossRef][Web of Science][Medline]

Gabbot PL and Somogyi P. Quantitative distribution of GABA-immunoreactive neurons in the visual cortex (area 17) of the cat. Exp Brain Res 61: 323–331, 1986.[Web of Science][Medline]

Gray CM and McCormick DA. Chattering cells: superficial pyramidal neurons contributing to the generation of synchronous oscillations in the visual cortex. Science 274: 109–113, 1996.[Abstract/Free Full Text]

Hendry SHC, Schwark HD, Jones EG, and Yan J. Numbers and proportions of GABA-immunoreactive neurons in different areas of monkey cerebral cortex. J Neurosci 7: 1503–1529, 1987.[Abstract]

Holt GR, Softky WR, Koch C, and Douglas RJ. Comparison of discharge variability in vitro and in vivo in cat visual cortex neurons. J Neurophysiol 75: 1806–1814, 1996.[Abstract/Free Full Text]

Kaneko H, Suzuki SS, Okada J, and Akamatsu M. Multineuronal spike classification based on multisite electrode recordings, whole-waveform analysis, and hierarchical clustering. IEEE Trans Biomed Eng 46: 280–290, 1999.[CrossRef][Web of Science][Medline]

Kawaguchi Y. Physiological subgroups of nonpyramidal cells with specific morphological characteristics in layer II/III of rat frontal cortex. J Neurosci 15: 2638–2655, 1995.[Abstract]

Kumano H, Tanaka H, Uka H, and Fujita I. Disparity selectivity of monkey inferior temporal neurons as examined by dynamic random dot stereograms. Neurosci Res Suppl 25: s161, 2001.

Lund JS, Yoshioka T, and Levitt JB. Comparison of intrinsic connectivity in different areas of macaque monkey cerebral cortex. Cereb Cortex 3: 148–162, 1993.[Abstract/Free Full Text]

McCormick DA, Connors BW, Lighthall JW, and Prince DA. Comparative electrophysiology of pyramidal and sparsely spiny stellate neurons of the neocortex. J Neurophysiol 54: 782–806, 1985.[Abstract/Free Full Text]

Mountcastle VB, Talbot WH, Sakata H, and Hyvärinen J. Cortical neuronal mechanisms in flutter-vibration studied in unanesthetized monkeys. Neuronal periodicity and frequency discrimination. J Neurophysiol 32: 452–484, 1969.[Free Full Text]

Nowak LG, Azouz R, Sanchez-Vives MV, Gray CM, and McCormick DA. Electrophysiological classes of cat primary visual cortical neurons in vivo as revealed by quantitative analyses. J Neurophysiol 89: 1541–1566, 2003.[Abstract/Free Full Text]

Ranck JB Jr. Studies on single neurons in dorsal hippocampal formation and septum in unrestrained rats. Part I. Behavioral correlates and firing repertoires. Exp Neurol 41: 461–531, 1973.[CrossRef][Medline]

Sakai Y, Funahashi S, and Shinomoto S. Temporally correlated inputs to leaky integrate-and-fire models can reproduce spiking statistics of cortical neurons. Neural Networks 12: 1181–1190, 1999.[CrossRef][Web of Science][Medline]

Shadlen MN and Newsome WT. The variable discharge of cortical neurons: implications for connectivity, computation, and information coding. J Neurosci 18: 3870–3896, 1998.[Abstract/Free Full Text]

Shinomoto S, Miura K, and Koyama S. A measure of local variation of inter-spike intervals. BioSystems 79: 67–72, 2005.[CrossRef][Web of Science][Medline]

Shinomoto S, Sakai Y, and Funahashi S. The Ornstein–Uhlenbeck process does not reproduce spiking statistics of neurons in prefrontal cortex. Neural Comput 11: 935–951, 1999.[CrossRef][Web of Science][Medline]

Shinomoto S, Shima K, and Tanji J. Differences in spiking patterns among cortical neurons. Neural Comput 15: 2823–2842, 2003.[CrossRef][Web of Science][Medline]

Shinomoto S and Tsubo Y. Modeling spiking behavior of neurons with time-dependent Poisson processes. Phys Rev E 64: 041910(1–13), 2001.[CrossRef]

Simons DJ. Response properties of vibrissa units in rat S1 somatosensory neocortex. J Neurophysiol 41: 798–820, 1978.[Abstract/Free Full Text]

Snodderly DM and Gur M. Organization of striate cortex of alert, trained monkeys (Macaca fascicularis): ongoing activity, stimulus selectivity, and widths of receptive field activating regions. J Neurophysiol 74: 2100–2125, 1995.[Abstract/Free Full Text]

Tamura H, Kaneko H, Kawasaki K, and Fujita I. Presumed inhibitory neurons in the macaque inferior temporal cortex: visual response properties and functional interactions with adjacent neurons. J Neurophysiol 92: 2782–2796, 2004.

Tanigawa H, Wang QX, and Fujita I. Organization of horizontal axons in the inferior temporal cortex and primary visual cortex of the macaque monkey. Cereb Cortex In press.

Thomson AM and Deuchars J. Synaptic interactions in neocortical local circuits: dual intracellular recordings in vitro. Cereb Cortex 7: 510–522, 1997.[Abstract/Free Full Text]

Thomson AM and Morris OT. Selectivity in the inter-laminar connections made by neocortical neurones. J Neurocytol 31: 239–246, 2002.[CrossRef][Web of Science][Medline]

This article has been cited by other articles:

				A. Bollimunta, Y. Chen, C. E. Schroeder, and M. Ding Neuronal Mechanisms of Cortical Alpha Oscillations in Awake-Behaving Macaques J. Neurosci., October 1, 2008; 28(40): 9976 - 9988. [Abstract] [Full Text] [PDF]

				K. Miura, Y. Tsubo, M. Okada, and T. Fukai Balanced Excitatory and Inhibitory Inputs to Cortical Neurons Decouple Firing Irregularity from Rate Modulations J. Neurosci., December 12, 2007; 27(50): 13802 - 13812. [Abstract] [Full Text] [PDF]

				D. Durstewitz and T. Gabriel Dynamical Basis of Irregular Spiking in NMDA-Driven Prefrontal Cortex Neurons Cereb Cortex, April 1, 2007; 17(4): 894 - 908. [Abstract] [Full Text] [PDF]

				G. La Camera, A. Rauch, D. Thurbon, H.-R. Luscher, W. Senn, and S. Fusi Multiple Time Scales of Temporal Response in Pyramidal and Fast Spiking Cortical Neurons J Neurophysiol, December 1, 2006; 96(6): 3448 - 3464. [Abstract] [Full Text] [PDF]

This Article Abstract Full Text (PDF) All Versions of this Article: 94/1/567    most recent 00896.2004v1 Alert me when this article is cited Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in Web of Science Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Web of Science (16) Citing Articles via Google Scholar Google Scholar Articles by Shinomoto, S. Articles by Fujita, I. Search for Related Content PubMed PubMed Citation Articles by Shinomoto, S. Articles by Fujita, I.