Propagating Activation during Oscillations and Evoked Responses in Neocortical Slices

Optical signals of population oscillations

The oscillations appear in cortical slices from the somatosensory areas as part of the spontaneous epochs that occur when the preparation is bathed in low [Mg] ACSF (Silva et al., 1991; Flint and Connors, 1996) (Fig. 2A). Each epoch started with a synchronous spike, and oscillations emerged afterward (Fig. 2B). In each epoch there was a variable number (usually 3–200) of oscillation cycles. The low-amplitude peaks of oscillations appeared to emerge from low-amplitude, asynchronized activities. As the epoch progressed, the oscillation frequency gradually decreased while the amplitude increased (Fig. 2B). At 26°C, the frequency was 5–8 Hz.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Epoch oscillations measured with an LFP electrode.A, When a cortical slice is perfused in low [Mg] ACSF, epochs of oscillations spontaneously occur. This recording is made by an electrode placed in cortical layer II, after 2 hr of perfusion in low [Mg] ACSF. B, An expanded display shows that an epoch started with a spontaneous synchronous spike and 7–10 Hz oscillations gradually emerged a few seconds later. In different preparations the epochs consisted of 3–200 cycles of oscillation. In all preparations tested (n > 50), all oscillations appeared in spontaneous epochs, and each epoch started with a synchronous spike. However, in seven preparations, there were only one or two oscillation cycles in each epoch. The number of oscillation cycles in each epoch appeared to increase the longer the interepoch intervals were.

We found that these LFP signals were highly correlated with the voltage-sensitive dye signals. When LFP and voltage-sensitive dye signals were simultaneously recorded, the waveform of LFP and optical detectors at nearby areas was similar and highly correlated (Fig.3). The optical signals were wavelength-dependent, with a maximum signal (dI/I) at 705 nm and no signal at 670 nm. These are major features of the signal from RH 479, the oxonol voltage-sensitive dye. The amplitude of an optical signal is proportional to the depolarization of all the stained membranes under each detector (for review, see Wu and Cohen, 1993). These signals had a rapid time course and thus could not have come from other optical sources, such as activity-related intrinsic optical signals (Hochman et al., 1995; Malonek and Grinvald, 1996; Yuste et al., 1997; Peterson et al., 1998). These results indicate that the optical signals faithfully represent the population neuronal activity, which allows us to examine this oscillation from many neighboring locations optically.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Comparison of electrical (LFP) and voltage-sensitive dye signals. A, During oscillations, the LFP signals and optical signals from neighboring sites were highly correlated. The recording sites were ∼0.5 mm apart, and there was a small phase difference between the different locations.B, The cross-correlation between LFP signals and the two optical detectors was high during the oscillations.

Spatial distribution of the oscillations

In the measurement shown in Figure4, a 4-mm-wide cortical slice was imaged (Fig. 4A), in which ∼50 neighboring locations were recorded simultaneously (Fig.4A,B). In these measurements we began an optical recording trial when an epoch occurred. In most of the preparations examined (n > 20), all locations oscillated at the same overall frequency (Fig. 4C). To our surprise, the correlations between two locations did not decrease significantly with distance, although there were large phase shifts at a large distance (Fig. 5A). The correlation was better between two points within one vertical “beam” than between two points separated by the same distance but in a tangential direction (Fig. 5B), suggesting that cortical columns might serve as synchronized units during oscillations.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 4.

Multiple site recording of the oscillations.A, A section of cortical slice (Sm Iarea) imaged by ∼50 neighboring photodetectors and one LFP electrode (in layer II). The imaging field was ∼4.5 mm in diameter.RF, Rhinal fissure; WM, white matter.B, Selected recording traces from the sites (solid squares in A). Both electrical and optical signals were low-pass filtered at 300 Hz, and the amplitude of the traces has been normalized to emphasize the phase relationships. C, The power spectra of the data in B. The gray trace is the average of all individual FFTs, in which the major frequency peak was 6.2 Hz (at 26°C). A minor frequency peak (∼10.8 Hz) does not have a systematic representation in the raw data shown in B. This minor peak was also seen in in vivo LFP recordings (Nicolelis et al., 1995). This recording trial contained 16,000 frames of voltage-sensitive dye images. Subsequent figures (see Figs. 5-8) were made from selected sections of these same data.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 5.

Auto- and cross-correlations. A, Cross-correlations. I, The amplitude of the correlations between the signals did not decrease significantly with distance. Thesolid and dotted traces are correlations between two detectors 800 and 3200 μm apart, respectively.II, Signals from the same column (detectors 2, 3) have a somewhat larger correlation (dotted trace) than do the signals from the same layer but different columns (detectors 1, 2;solid trace). B, Autocorrelations of oscillation signals at five different locations (filled rectangles). At different locations, phase variations appear in outside peaks. The overall frequency was 6.2 Hz.

Autocorrelations of the signals from five locations (Fig.5B) suggested that the entire preparation was dominated by one frequency. However, the autocorrelation peaks were no longer superimposed after two cycles in either direction from 0 time, indicating that there were slow variations in frequency at different locations (Fig. 5B). A plot of the time interval between sequential cycles at several locations (Fig.6) indicated that for most cycles (cycles 1–6, 11–17) the frequency variations were highly correlated at different locations. However, in addition to a general tendency to slow down at all locations, in 72 out of 105 of the epochs we found sudden changes in the cycle-to-cycle period that occurred simultaneously at different locations (as illustrated in Fig. 6, cycles 7–10). The correlations between different locations during this sudden change in frequency followed fixed patterns. The change appeared to occur first at one side of the preparation and moved to the other side in approximately two cycles. Although it is difficult to understand what is happening from the plot in Figure 6, the pseudocolor images from the same data set (Fig. 7C, II; see next section) offer a simple and elegant explanation for the sudden variation in frequency.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 6.

Variations in the cycle period at different locations. The cycle-to-cycle periodicity has sudden changes in some cycles (at cycles 7–10 in this epoch). At different locations the changes are in different directions. These kinds of sudden changes were very common. We examined 105 epochs from four preparations, and the sudden changes were found in 72 epochs.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 7.

Propagating activation during the oscillations.A, The recording arrangement for the data inB and C. RF, Rhinal fissure. B, Raw data from the LFP electrode and six optical detectors. Three sections from these data (I,II, III) are displayed in thepseudocolor images in C.C, Pseudocolor images from the three selected sections I, II, andIII in B displayed at 10 msec/frame. In the images warm colors represent a larger depolarization. I, During each cycle of oscillation, one dynamic ensemble emerged and traveled across the field of view. Three dynamic ensembles appeared in this section and movedupward across the field of view. When they moved past the LFP electrode, a cycle of oscillation was detected on the electrode (black trace in B).II, Later (0.5 sec), two dynamic ensembles appeared simultaneously (at the top and bottom of the field of view) and collided near the center. Two sequential collision events occurred during period II. These collisions generated the dramatic change in oscillation frequency seen in Figure 6. Activity was quickly annihilated after the collisions. III, After the collisions in periodII, the propagation direction of the ensembles was reversed. The dynamic ensembles emerged from the top of the field of view and traveled downward. In all eight preparations imaged, the oscillations appeared as propagating activation—dynamic ensembles. The physical size of dynamic ensembles was estimated by the isopotential contour lines in thepseudocolor display and was ∼0.5 mm wide in the tangential direction.

Propagating activations during the oscillations

The images derived from all the optical detectors revealed that the neurons activated during oscillations were not uniformly distributed over the slice. Instead, each oscillation peak appeared as a localized hot spot propagating across the cortex. In Figure 7we made pseudocolor images based on the amplitude of the voltage-sensitive dye signals, from three time sections (I–III) within one epoch. When the LFP recordings in Figure 7B are compared with the pseudocolor images in Figure7C, it is obvious that each oscillation peak correlated to a propagating hot spot. In each oscillation cycle, when a hot spot passed by an LFP electrode (or an optical detector), a depolarization peak was detected. When many hot spots repeatedly passed a detector, a periodic depolarization was recorded that appeared as oscillations (Fig.7B). Thus at each individual recording point, the activity could be defined as an oscillation by its periodicity in FFT and autocorrelations (Figs. 4B, 5B). However, the images of the same data set revealed propagating hot spots (Fig.7C), suggesting that the periodicity is in part determined by the propagation velocity.

The shape of the hot spots varied somewhat from cycle to cycle of the oscillations, and no particular pattern was correlated with each individual preparation. It is difficult to determine accurately the physical size of a hot spot from the amplitude pseudocolor image. However, using the 50% isopotential lines of the contour map, we determined that the size of the spots was ∼0.5 mm wide in a tangential direction and encompassed all cortical layers. The spots were relatively similar in size but somewhat irregular in shape (Fig.7). The cycle-to-cycle variability of the shape and size suggested that these propagating hot spots were most likely dynamically organized. The hot spots were seen in the slices with different slicing angles from the barrel cortex. The isopotential contour did not appear to be correlated to the barrel patterns, suggesting that the activity is likely to be organized by distributed intracortical connections. Because the hot spots are likely to be dynamically organized and they are not a classical stationary oscillation, we give them the name “dynamic ensembles” to describe this propagating activation and to distinguish it from other kinds of population activity.

Propagation, collisions, and jumps of dynamic ensembles

For each cycle of an oscillation, one and occasionally two dynamic ensembles emerged and propagated away. The overall propagation velocity was 30 ± 10 mm/sec (mean ± SD; n = 15; at 26°C), and the distance traveled was usually larger than the field that was imaged (i.e., >5 mm). In a brain slice only two propagation directions are possible. A consecutive series of dynamic ensembles usually propagated in the same direction; 97% (n = 200) kept the same direction as their predecessor. However, the propagating direction was not fixed. The direction could reverse in the middle of an oscillation epoch (Fig.7B,C, I,III). When two dynamic ensembles emerged simultaneously at different locations, they sometimes propagated toward each other, resulting in collision and annihilation of the activity (Fig. 7B,C, II). These collisions and changes in propagation direction explain why the oscillation frequencies varied drastically in the middle of some epochs (Fig. 6, cycles 7–10).

The propagation of excitation in cortical tissue often contains abrupt changes or “jumps” (Chervin et al., 1988; Wadman and Gutnick, 1993). We used the isopotential contour lines of the dynamic ensembles to examine variations in the propagation velocity. We found that the traveling course of each dynamic ensemble often had several jumps (Fig. 8B,arrows) that appeared as a rapid progression of the front edge of the isopotential contour line within a few milliseconds (Fig.8A), followed by rapid withdrawal of the back edge (data not shown). Although the pattern of horizontal intracortical connections may be important in the propagation velocity of an excitation wave (Chervin et al., 1988; Wadman and Gutnick, 1993;Golomb and Amitai, 1997), in our data the locations of the jumps appeared to vary for each dynamic ensemble (Fig. 8A,I–III). It is thus more likely that the jumps in our image are dynamically distributed and not determined by the intracortical connectivity patterns.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 8.

Details of the propagation of dynamic ensembles.A, The front edge contours of three dynamic ensembles are shown every 2 msec. The wide gaps between thecontour lines suggest a sudden increase in propagating velocity. I, II, Two consecutive dynamic ensembles jumped at similar (but not identical) locations.III, A dynamic ensemble that emerged 1 sec later had quite different jump locations. B, The propagation velocity of the three dynamic ensembles (measured at cortical layer II) is shown. The overall velocities are 36, 29, and 21 mm/sec, composed of jumps (arrows) and slow propagation. We have examined the front edge contours from 10 dynamic ensembles from three preparations; the dynamically distributed jumps were seen in all 10.

Neuronal activity within a dynamic ensemble

The density of active neurons in a dynamic ensemble can be estimated by the amplitude of voltage-sensitive dye signals. A maximum response evoked by an electrical shock around the tip of a stimulating electrode was used as a standard of “maximum activation” of a neuron population. We assume that the majority of neurons near the electrode were simultaneously (within 1 msec) activated, orthodromically or antidromically, by the shock. The amplitude of the optical signal of a dynamic ensemble was measured from the same field of view and was 5–7% of the amplitude of maximum activation. This suggested a low-firing probability within a dynamic ensemble.

The low-firing probability in a dynamic ensemble is further confirmed by extracellular, multiunit recordings that directly measure the spiking density within a dynamic ensemble. One electrode can usually distinguish spikes from two to three neurons near its tip by the amplitude of the spikes. When a dynamic ensemble was passing by the region of the electrode, the neurons sampled by the electrode only fired one to three spikes [from two preparations, 50+ dynamic ensembles (J.-Y. Wu and L. Guan, unpublished observation)]. The same electrode could record many more spikes during the synchronous spike occurring at the beginning of an oscillation epoch. This showed that the electrode was able to detect the spikes from many neurons but only a few fired during each dynamic ensemble.

From limited multiunit recordings, we did not find spikes correlated in a millisecond time domain even though the two neurons were physically close to each other. This is consistent with the optical images because correlated firing of a large number of neurons in the millisecond domain would have generated a sharp spike in the optical signal.

Distinguishing dynamic ensembles from epileptiform events

We attempted to use voltage-sensitive dye imaging to distinguish dynamic ensembles from epileptiform events. The synchronous spike at the beginning of each oscillation epoch (Fig. 2B) had characteristics of epileptiform events in brain slices bathed in low [Mg] (Anderson et al., 1986; Zhang et al., 1996). This allowed us a direct comparison of the two different events in the same preparation. The synchronous spikes and dynamic ensembles appeared different in several ways. First, the synchronous spike appeared in the imaging as a spreading of an excitation wave, starting from a small initiation site and expanding to the entire preparation (Fig.9B, top four images). This pattern is similar to the imaging of epileptiform events described previously (Sutor et al., 1994; Albowitz and Kuhnt, 1995; Jackson and Scharfman, 1996; Tsau et al., 1998, 1999). In contrast, the activation of a dynamic ensemble was cohesive (not expanding), its physical size remaining relatively constant during propagation (Figs. 7C, 9B, bottom row). Second, the amplitude of the dye signal of the synchronous spike was 5–20 times larger than that of dynamic ensembles (Fig.9A), suggesting that the former had a higher density of active neurons. This is consistent with the extracellular electrode recordings, which showed that a synchronous spike had many more spikes than did a dynamic ensemble (data not shown). Third, epileptiform events generated a large peak in the intrinsic optical signal that was slow and wavelength independent (Fig. 9A, right) (see also Grinvald et al., 1986; Hochman et al., 1995; Yuste et al., 1997; Pazdalski et al., 1998). With dynamic ensembles this intrinsic signal was much smaller. Finally, epileptiform events and dynamic ensembles had different refractory periods. The former could not be evoked by an electrical stimulus applied at a rate higher than four times per minute, and the latter occurred at 5–8 Hz at 26°C.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 9.

Dynamic ensembles are different from epileptiform events. A, Left, Middle, The epileptiform event was 10 times larger in dye signal amplitude (705 nm absorption) than were the dynamic ensembles (which are seen as an oscillation in one detector). Intrinsic optical signal also appears at 705 nm, but with a much slower temporal course. We used high-pass filters (both analog and digital, with the RC time constant of 0.1 sec) to emphasize the voltage-sensitive dye signals in these twopanels. Right, The intrinsic optical signal of an epoch is shown. The signal is much larger and slower than are those on the left and in the middle. These were the same data shown on the left but displayed with a minimum high-pass filter and different time and gain scales. We examined three preparations in white light in which voltage-sensitive dye signals were undetectable. In these conditions the intrinsic optical signals were the same as that shown on theright. B, Top row, An epileptiform event was evoked by an electrical stimulus (Stim) at the center of the field of view (see diagram on the right). This event expanded to the entire field of view in ∼30 msec. Bottom Row, A dynamic ensemble emerged from the same cortical area (same field of view, 7 sec later) and remained cohesive for at least 80 msec. The two rows are displayed with differentpseudocolor scales to show the propagation of both events.

Evoking dynamic ensembles in normal ACSF

We attempted to show that a dynamic ensemble is an activity related to the intrinsic cortical circuits rather than a phenomenon only appearing during low [Mg] oscillations. We attempted to evoke dynamic ensembles when the cortical slice was bathed in normal ACSF, in which the neuronal network is not in a hyperexcitable condition. It is well known that in normal ACSF, the duration of a “passive” evoked response in the cortical slice is short [5–20 msec (Chagnac-Amitai and Connors, 1989)] and not self-sustaining. We propose, however, that with different ways of activating the cortical network, the excitation and inhibition balance may be altered, which would sustain the long-lasting activations of dynamic ensembles.

We used two kinds of electrical stimulation to evoke population activation in slices bathed in normal ACSF: (1) a conventional thin coaxial electrode, which evoked a high density of coactivation in a small area and which almost always elicited passive short-lasting evoked responses, and (2) a large tip single-pole electrode placed 100–300 μm above the tissue surface, which produced an electrical field over a large area, including all cortical layers (Fig.10A). With this single-pole electrode, the area of activation and the density of activated neurons can be independently adjusted by changing the distance between the electrode and the tissue surface and by using different stimulus intensities. We can thus obtain a low level of activation (0.3–3 times the LFP response threshold) in a large area (1–2 mm in diameter). With this electrode and proper stimulus intensities, the evoked activity developed into long-lasting, propagating activations. In four preparations we delivered weak electrical stimuli at an interstimulus interval of 30 sec, and an LFP recording was used to detect the long-lasting population events (Fig.10B, traces II, III). The optimum stimulus intensity was ∼0.5–0.7 of the passive LFP response threshold. With an optimum stimulation intensity, >90% of the stimuli evoked these long-lasting events. Increasing the stimulus intensity dramatically reduced the probability of inducing dynamic ensembles. When the intensity was increased by 20%, only ∼40% of evoked activations became long-lasting; other responses were brief (Fig. 10B, trace I). Although the duration of the events varied from trial to trial, it was obvious that the activity was an all-or-none event. These data suggested that the long-lasting events are self-sustained population activations, which can be evoked in a cortical network with unaltered excitability.

• Download figure
• Open in new tab
• Download powerpoint

Fig. 10.

Dynamic ensembles evoked in normal ACSF. A, Schematic diagram of a single-pole electrode placed above the slice is shown. This stimulation method can evoke a low density of activated neurons distributed in a large area including several cortical layers. The intensity and the size of the area can be independently controlled by adjusting stimulation voltage and the distance between the electrode and the slice. B, Long-lasting response in local field potential evoked by weak electrical shocks is shown. In this preparation an electrode was placed on top of layer IV, ∼80 μm above the preparation. The interstimulus interval was 30 sec. Trace I, With stimulus intensity of 7 V (duration = 0.05 msec, for all stimuli), just above the threshold of the passive LFP response, a passive response (short and more synchronized) was evoked. Traces II,III, With intensity reduced to 3 V, long-lasting activity was evoked. The long-lasting responses were all-or-none events. In this preparation, 98% (n = 120) of 3 V stimuli evoked a long-lasting response. When the stimulus was increased to 5.5 V, ∼60% of the stimuli failed to evoke a long-lasting response. At 7 V, all the stimuli (n = 20) evoked the passive response (trace I). Traces II and III are consecutive trials, illustrating the variation of the long-lasting response. C, The long-lasting response was spatially confined (not expanding) and propagating, similar to dynamic ensembles (Fig. 7). A weak electrical shock (∼10 V × 0.05 msec) from an electrode placed 100 μm above the preparation [at the upper right corner of the image field (see diagram on the left)] induced activity over a large area including all layers (Stim image on the left). Seven milliseconds later, a long-lasting, propagating activation emerged next to the area activated by the stimulus (second image from theleft). This activity had an amplitude that was similar to that of the dynamic ensembles observed during 7–10 Hz oscillations (e.g., Fig. 7). The activity remained cohesive for ∼40 msec and propagated downward in the image field. The slice was perfused in normal ACSF, and the optical signal had a small amplitude compared with that of an epileptiform spike. Note the propagation velocity was ∼6–10 mm/sec (n = 10, measured from 2 preparations), significantly slower than the propagation during oscillations. Stim, Stimulus.

Optical imaging confirmed that these long-lasting events were similar to the dynamic ensembles in their size, shape, and active neuron density (Fig. 10C). Optical imaging also showed that the long-lasting activity always emerged in a location 1–2 mm away from the center of the stimulation point and in most of the cases 20–50 msec after the stimulus. This further suggested that the long-lasting events were dynamically organized by the cortical network and not a passive activation resulting from the electrical stimulus. On the basis of these observations, we extend our definition of dynamic ensembles to include both spontaneous and evoked population events sustained by the intrinsic cortical circuits that are dynamically organized and self-sustaining and have a low density of active neurons.

Dynamic ensembles and population oscillations

We have demonstrated that the same population events can be described either as oscillations at an individual location (measured optically or with electrodes) or as propagating dynamic ensembles (measured by imaging) (Figs. 7, 10B). The ubiquitous existence of oscillations (Tank et al., 1994; Singer and Gray, 1995) and the link between population oscillations and dynamic ensembles raise the possibility that dynamic ensembles might be a common form of population event in the neocortical neuronal network.

Slice preparations from different parts of the brain often have an oscillation frequency that is similar to those found in vivo. These oscillations include 6–10 Hz spindles in the thalamus (Bal et al., 1995; Kim et al., 1995), 7–10 Hz in somatosensory cortex (Silva et al., 1991) (this paper), 20–80 Hz in auditory cortex (Metherate and Cruikshank, 1999), 10–50 Hz in other parts of the neocortex (Llinas et al., 1991), and 40 Hz in the hippocampus (Whittington et al., 1995). In a few slice preparations, the spatiotemporal patterns of oscillations have been examined. Field potential measurements from ferret thalamus slices indicated that the 5–10 Hz spindle oscillations also appear as propagating waves (Kim et al., 1995). These thalamic spindle waves had a similar frequency but propagated 10 times slower than did the dynamic ensembles we have seen in the cortex. Because propagation velocity determines the wavelength or distance between dynamic ensembles, it would be interesting to compare the images of oscillation waves generated by different mechanisms, e.g., “lurching” thalamic spindle versus “continuous” cortical waves (Golomb et al., 1996; Rinzel et al., 1998). This would allow us to examine whether dynamic ensembles are of more general occurrence in different local circuits.

Imaging of in vivo oscillations has shown that they also appeared as propagating excitation waves (Delaney et al., 1994;Kleinfeld et al., 1994; Lam, 1997; Prechtl et al., 1997). The rostral oscillation in turtle olfactory bulb has a simple propagating pattern (Y.-W. Lam, L. B. Cohen, M. Wachowiak, and M. R. Zochowski , unpublished observations), which is strikingly similar to our in vitro observations. At each cycle of the rostral oscillation, a “blob” of activation emerged from one location and propagated with a certain velocity and direction in the bulb. The physical size, active neuron density, and propagation velocity of the blob were very similar to those of the dynamic ensembles that we observed in cortical slices. Visually induced oscillations in turtle cortex have multiple frequency domains, with propagating blobs representing each frequency domain (Prechtl et al., 1997). Oscillations in the somatosensory cortex of behaving rats (7–10 Hz) are also organized in epochs during whisking, and the oscillations also appear as traveling waves (Nicolelis et al., 1995). Unfortunately, descriptions of the spatial pattern of in vivo oscillations have been limited because no imaging data from mammalian cortex have been available.

Going from an intact cortex to a cortical slice is a significant reduction from a large three-dimensional cortical network to a relatively small and virtually two-dimensional network. This reduces the number of possible initiators of oscillations and the connections between them. Thus a propagating wave would encounter fewer interactions with activation initiated by other initiators. The propagation of spindle waves in thalamic slices (Kim et al., 1995) might have much fewer collisions than in the intact thalamus (Contreras et al., 1997). Thus our images of dynamic ensembles might only present a simplified situation of oscillatory events in vivo.

Population oscillations have been proposed to serve as a timing signal (von der Malsburg, 1985; Singer, 1993), as reference-carrying waves for temporal coding (Hopfield, 1995) or for arousal or attention (Steriade et al., 1991; MacKay and Mendonca, 1995; Munk et al., 1996). Our data showed that during oscillations different locations were activated sequentially as dynamic ensembles propagated. This presents questions about the definition of synchrony in the hundred millisecond time domain and how such synchrony is used for cortical processing.

Dynamic ensembles and cortical neuronal organization

The interaction between cortical cellular organization (e.g., columns) and propagating dynamic ensembles remains to be explored. Our data suggested that the correlation within a cortical column is better than that between neighboring columns (Fig. 5). It might be speculated that a dynamic ensemble consists of activity in one or several cortical columns and that this activity jumps from one column to another. However, we noticed that the propagation patterns of the dynamic ensembles (Fig. 8) were not consistent with the speculation that dynamic ensembles jump from barrel to barrel. Also, in tangential slices from the somatosensory areas, we could not find a correlation between the activity of dynamic ensembles and the whisker barrels (Wu and Guan, unpublished observation). [Chervin et al. (1988) also found no correlation between barrel patterns and propagation jumps.] These observations suggested that dynamic ensembles were more likely to be organized by intrinsic (local) circuits rather than by afferent-related cellular organizations.

The organization of intrinsic circuits in the neocortex is relatively similar in different areas (Shepherd, 1988; Douglas and Martin, 1990). This similarity [the “canonical” circuit (Douglas and Martin, 1990)] suggests universal rules for organizing local population activation in different cortices. As a continuation of this work, it would be interesting to compare oscillations and evoked population events in different cortical areas, to test whether dynamic ensembles can be evoked in all cortical tissues.

The nature of activation in a dynamic ensemble

We estimate that a dynamic ensemble has only ∼5–7% of the maximum activation of the cortical network. The voltage-sensitive dye signals come from a combination of spikes and synaptic potentials. If we assume that our signals were a linear summation of spikes and synaptic potentials in the population (Wu and Cohen, 1993), then one spike (100 mV) of 1 neuron would contribute equally with 10 neurons that have a 10 mV depolarization. At one extreme, we could assume that all signals were contributed by spikes, so that 5–7% of the total population was firing simultaneously in each sampling period. Thus during the 200 msec duration of a dynamic ensemble, each activated neuron would fire ∼10 times. However, extracellular unit recordings revealed that each neuron only fired one or two spikes. Therefore it seems likely that subthreshold depolarizations of many cells are an important contributor to the dynamic ensemble signals. We thus speculate that a dynamic ensemble is organized as a domain of moderately depolarized neurons in which some neurons will increase their firing probability to reach one to two spikes per event. This kind of organization is similar to the spontaneous population coactivation observed in developing cortex (Yuste et al., 1992) and retina (Wong et al., 1995; Feller et al., 1997).

Self-sustainability: functional significance

The activation of dynamic ensembles lasted much longer than an ordinary evoked response. We considered this activation to be self-sustaining, i.e., sustaining a certain level of activity in the cortex without afferent input. We hypothesize that self-sustainability is a population property of a reciprocally activating neuronal network. The development of a dynamic ensemble might follow certain population rules, e.g., depend on the size of the activation, the density of active neurons in the area, and the temporal pattern of the input (cf.Golomb and Amitai, 1997). In this report we evoked long-lasting activity by activating a low density of neurons in a large area involving all cortical layers (Fig. 10). The all-or-none nature of a population response supports our speculation that the activity is self-sustained. This kind of long-lasting evoked response can also be evoked by stimulating thalamocortical afferents (Metherate and Cruikshank, 1999).

How dynamic ensembles interact with evoked events remains to be explored. In intact brain, dynamic ensembles may be evoked by sensory input or emerge during evoked oscillations. This would allow spatial interactions between different ensembles and temporal interactions between concurrent input and dynamic ensembles evoked by previous input. Self-sustained activity in dynamic ensembles may provide an uneven local excitability, in which the same input may elicit a higher (or lower) population firing rate within an ensemble than in the regions outside of it. This suggests a mechanism for cortical reverberation (Hebb, 1949; Amit, 1995), a process in which a local network can amplify a weak input with a compatible spatiotemporal pattern and reject a strong input with incompatible patterns.

The existence of dynamically organized population modules has been postulated as a “cell assembly” (Hebb, 1949; Amit, 1995; Nicolelis et al., 1997) that reverberates with afferent input and may play a role in sensory discrimination (Zhang, 1996), temporary information storage, attentional modulation of sensory processing, and transition from short- to long-term plasticity (Rauschecker, 1995). Our study of dynamic ensembles suggests that cortical intrinsic circuits can serve as a host for self-sustained and dynamically organized population activity. Such a form of activity could be used to “bind” afferent input distributed in time and space.