Synaptic dynamics control the timing of neuronal excitation in the activated neocortical microcircuit

Abstract

It is well established that sensory stimulation results in the activity of multiple functional columns in the neocortex. The manner in which neurones within each column are active in relation to each other is, however, not known. Multiple whole-cell recordings in activated neocortical slices from rat revealed diverse correlation profiles of excitatory synaptic input to different types of neurones. The specific correlation profile between any two neurones could be predicted by the settings of synaptic depression and facilitation at the input synapses. Simulations further showed that patterned activity is essential for synaptic dynamics to impose the temporal dispersion of excitatory input. We propose that synaptic dynamics choreograph neuronal activity within the neocortical microcircuit in a context-dependent manner.

The neocortex is functionally divided into multiple modules, also referred to as cortical columns, that form complex feature maps such as somatosensory maps, auditory maps and various visual maps (Merzenich & Brugge, 1973; Hubel & Wiesel, 1977; Shmuel & Grinvald, 1996; Mountcastle, 1997). While the functional organization of modules to form these maps has been studied extensively, the nature of activity within such a single module is not known. These neocortical modules are all constructed from a rather stereotypical microcircuit of neurones (Douglas & Martin, 1991; Mountcastle, 1997; Kozloski et al. 2001; Silberberg et al. 2002), suggesting a generic operation subserving multiple tasks. A key question is therefore how the detailed organization of the neocortical microcircuit orchestrates the activity that emerges and what the role is of the different neurone types. We approached this question by studying the correlation of excitatory synaptic input to different neurones in the neocortical microcircuit during activation.

Methods

Slice preparation and electrophysiology

All experimental procedures were carried out according to the Swiss federation guidelines for animal experiments. Neocortical slices (Sagittal, 300μm thick) were obtained from Wistar rats (postnatal days 13–16 after rapid decapitation). Slices were incubated for 30 min at 33–35°C and then at room temperature (20–22°C) until transferred to the recording chamber (35±0.5°C). Neighbouring neurones in layer V of the somatosensory area were selected for recording according to the morphology of their somata and proximal dendrites. The slice was visualized by IR-DIC optics using a Zeiss Axioscope and Hamamatsu CCD camera. The bathing solution consisted of (mm): NaCl 125, NaHCO₃ 25, glucose 25, KCl 2.5, CaCl₂ 2, NaH₂PO₄ 1.25, MgCl₂ 1. Simultaneous whole-cell recordings from clusters of up to seven neurones were made using patch pipettes (5–10 MΩ), containing (mm): potassium gluconate 110, KCl 10, Hepes 10, phosphocreatine(Na) 10, MgATP 4, NaGTP 0.3 and biocytin 4 mg ml⁻¹. Somata of recorded neurones were located at least 40μm below the slice surface to enable reliable morphological identification and were separated from each other by less than 220μm (average Euclidean distance: 98μm; average lateral (parallel to pia) distance: 60μm). No correlation was observed between correlation lags and the somatic distance within this range (data not shown). Voltage recordings were obtained using Axopatch 200/B amplifiers (Axon Instruments). Data acquisition and analysis was performed using IgorPro (WaveMetrics, Inc.).

Cross-correlation

Normalized correlation functions were calculated according to:

where A and B are the correlated traces of length T, and barred characters represent mean values. The normalization factor is such that auto-correlations would result in a value of 1 at zero lag and any non-identical traces would have values between −1 and 1 according to the degree of similarity for different temporal lags. The same calculation for estimating correlated subthreshold activity between neurone pairs was used in vivo (Lampl et al. 1999). Voltage traces used to create cross-correlograms were 60 or 120s long, with sampling intervals of 250μs. Cross-correlograms were calculated only from subthreshold traces. In cases in which voltage traces contained action potentials, only subthreshold intervals longer than 10s were cross-correlated. Peak lags were extracted from the highest peak within an interval of ±1s of the cross-correlograms. Median lags were calculated from the mid-value point of the cross-correlation integral over the same interval of ±1 s. EPSP rise times were calculated between the points of 20% and 80% of EPSP amplitudes and EPSP decay times were calculated by fitting the initial phase of the EPSP decay with an exponential function.

Slice excitation

Activity in the slice was induced by altering the ionic composition of the extracellular solution. Changes in the concentration of K⁺ affect the resting potential by changing the reversal potential of the neurones' leak current and decreasing the concentration of the divalent ions (Mg²⁺ and Ca²⁺) lowers the threshold for firing and increases activation of NMDA synaptic transmission. The altered solution contained (mm): KCl 6.25, CaCl₂ 1.5 and MgCl₂ 0.5, compared to 2.5, 2 and 1, respectively, in the standard extracellular solution. Similar excitation procedures were used in recent studies in various slice preparations (Sanchez-Vives & McCormick, 2000). The excitant solution was perfused at a rate of 25μl s⁻¹ resulting in gradual solution change during several minutes. Recordings in excited slices were obtained in current-clamp configuration.

Statistics

Values for peak lag and median lag were not normally distributed, as tested by the Lilliefors goodness-of-fit normality test. We therefore used the Kolmogorov-Smirnov (K-S) test to evaluate the differences between latency distributions. This test does not assume that observations originated from normal or similar distributions and was therefore the most suitable for our data. Statistical tests were performed using the statistics toolbox provided by MATLAB (version 6.5.1, The MathWorks, Inc.).

Simulations

It has been previously shown that synapses with different dynamics operate optimally when driven by different activity patterns (Tsodyks & Markram, 1997; Natschlager & Maass, 2001). In our simulations, presynaptic neurones were connected to postsynaptic integrate-and-fire neurones by synapses with different dynamics. Postsynaptic neurones had the same membrane time constant (20ms) and received excitatory synapses modelled by an α function (α=2 ms). Dynamic properties of the synapse were implemented by using the model described in Tsodyks & Markram (1997), in which three dynamic parameters (U, D and F) are used to describe the synapse. When the presynaptic train was composed of a single rate Poisson train, the average rate was 5Hz, close to the average rate in experimental recordings (see Results). In simulations where the action-potential train was composed of alternating frequency epochs, the rates used were 1Hz and 200Hz, tuned to yield the average rate of 5Hz by setting the relative durations of the different frequency epochs. The postsynaptic responses of pairs of simulated target neurones were cross-correlated exactly in the same manner as experimental recordings. All simulations were performed by MATLAB (version 6.5.1, The MathWorks, Inc.).

Results

In order to isolate the key parameters that shape excitatory input to neurones in the microcircuit, we obtained simultaneous whole-cell recordings from clusters of neighbouring neurones in slices from the somatosensory cortex of young Wistar rats (Fig. 1A), activated the microcircuit by perfusion of an excitant extracellular solution and examined cross-correlations of subthreshold activity between different neurone pairs (see Methods). While the microcircuit was activated, the recorded neurones were hyperpolarized by negative current injection to below resting potential (−75±5 mV), thereby preventing discharge of action potentials and allowing isolation of excitatory synaptic input. The activity induced in the slice was assessed by recording the neurones without negative current injection (Fig. 1B), and was composed of a large range of discharge frequencies, from 0.03 to 179.82Hz (4.1±4.3Hz, n=125 neurones). The C.V. of interspike intervals was 139±96% and ranged between 2 and 540% as calculated for 1-min-long action-potential trains. High-frequency bursts were observed, although they were not rhythmic as described in ferret slices under similar conditions (Sanchez-Vives & McCormick, 2000). Under these conditions, subthreshold correlations emerged between all recorded neurone pairs, regardless of whether the neurones were synaptically connected or not (peak correlation coefficient: 0.25±0.075; 210 pairs; Fig. 1C). When the excitant solution contained fast synaptic blockers (50 μm APV, 10 μm CNQX and 10 μm bicuculine), discharge patterns of recorded neurones did not exhibit bursts or fast frequency transitions (Fig. 1B) but were more regular (C.V. of interspike intervals: 69±18%, n=5). Under control conditions (normal ACSF, see Methods), none of the recorded neurones generated spontaneous action-potentials, suggesting that slices were inactive in terms of neuronal discharge (Fig. 1B). The subthreshold voltages of any two neurones were uncorrelated when recorded in the control solution, as no peaks were observed in the cross-correlograms (Fig. 1C).

Figure 1. Cross-correlograms were obtained from multiple simultaneous voltage recordings from various neocortical neurone types under different recording conditions.

A, a cluster of six neurones, 2 pyramidal and 4 interneurones in neocortical layer V. Neurones were stained using biocytin filling through the recording patch-pipette. B, under control conditions (normal ACSF, see Methods) no discharge was observed (upper trace). When slices were activated by excitant solution (Methods) neurones discharged irregularly, displaying bursts and rate fluctuations (middle trace). When the excitant solution contained fast excitatory synaptic blockers, neurones discharged in a more regular way with lower ISI variability (bottom trace). C, simultaneously obtained subthreshold voltage recordings from two cortical neurones in activated slice conditions (left) and the resulting cross-correlogram (right, black trace). The cross-correlogram is presented in comparison to that obtained under control conditions (right, grey trace).

We noticed a wide diversity in the profiles of these cross-correlations for different pairs of neurones (Fig. 2). Correlation functions differed in the temporal lag of the peak and in the degree of asymmetry around zero, resulting in different lags of median values. We therefore examined these profiles more closely in terms of the delay to peak (peak lag) and the delay of the median (median lag) of the correlation function (see Methods). The cross-correlations for pairs of pyramidal neurones (P-P) differed significantly (Kolmogorov-Smirnov distribution test, see Methods) from those between pyramidal neurones and interneurones (P-I). Between pyramidal neurones, the cross-correlations were similar to auto-correlations, with a very small range in peak lag values (71 pairs; Fig. 2A) in comparison to the cross-correlations between pyramidal neurones and interneurones (96 pairs; Fig. 2B). Peak lag for P-P pairs was 2.3±2.4 ms and median lag was 14.2± 16.2 ms while for P-I pairs peak lag was 25.9±29.4 ms and median lag was 50.9±38.2 ms (K-S distribution test: P < 10⁻⁷). The asymmetry in individual cross-correlations, as reflected in median lag values, and the broad distribution of peak lag values for interneurones, indicate that excitatory synaptic activity in interneurones can be correlated with synaptic activity that occurred in pyramidal neurones more than 200 ms earlier as well as with activity more than 50 ms later.

Figure 2. Cross-correlations between pairs of pyramidal neurones (P-P) were different from those between pyramidal neurones and interneurones (P-I).

A, simultaneous subthreshold voltage recordings obtained from two layer V pyramidal neurones (left) and cross-correlogram obtained from a 1-min-long recording (right). Note the short delays between recorded events in the two neurones, and the reflection in the peak and symmetry of the cross-correlogram. B, the same as A, in this example the recorded neurones were a layer V pyramidal and a layer V interneurone. The delays observed in some of the recorded events contribute to the longer peak delays and asymmetry in the cross-correlogram. C, example of recorded voltage traces and cross-correlogram between two layer V interneurones.

The large heterogeneity in cross-correlation profiles for P-I pairs could mirror the heterogeneity in morphological and electrophysiological properties of interneurone types found in the neocortex as compared to pyramidal neurones (Peters & Jones, 1984; Kawaguchi & Kubota, 1997; Gupta et al. 2000). It may also reflect the heterogeneity in synaptic dynamics employed at synapses between pyramidal neurones and interneurones (Thomson et al. 1993; Markram et al. 1998; Reyes et al. 1998; Kozloski et al. 2001; Beierlein et al. 2003) as compared to the more homogeneous synapse type found between pyramidal neurones (Markram et al. 1997). We therefore compared the cross-correlations for P-I pairs in which interneurones received depressing synapses from pyramidal neurones (P-I_d; Fig. 3A) with those in which interneurones receive facilitating connections (P-I_f;Fig. 3B).

Figure 3. Cross-correlations of P-I pairs in which excitatory connections were depressing (P-I_d) had shorter peak lag and median lag values than those of P-I pairs with facilitating excitatory connections (P-I_f).

A, an action-potential train in a pyramidal neurone evokes a depressing synaptic response in an interneurone (top). Cross-correlation between a pyramidal neurone and an interneurone receiving depressing excitatory synaptic input (bottom). B, an action-potential train in a pyramidal neurone evokes a facilitating synaptic response in an interneurone (top). Cross-correlation between a pyramidal neurone and an interneurone receiving facilitating excitatory synaptic input (bottom). C, comparison of peak lag and median lag between P-P, P-I_d and P-I_f correlations. Histogram of all peak lag values of different pairs of simultaneously recorded neurones (left). Both peak lag and median lag values were significantly different for different neurone pairs (for all, P<10⁻⁵; K-S test, see Methods), as presented in the bar-graphs (right).

Cross-correlograms were significantly different (K-S test, P<10⁻⁵) with smaller peak lag and median lag values for P-I_d pairs than P-I_f pairs (Fig. 3C) suggesting that synaptic dynamics may indeed be instrumental in positioning the timing of peak synaptic input to different types of interneurones and that differences in synaptic dynamics are not simply averaged out during microcircuit activity. The distribution of peak lag values for P-I_d pairs was also broader than for P-P pairs (which are also interconnected by depressing synapses), consistent with the larger heterogeneity in the dynamics of these synaptic connections (Thomson et al. 1993; Markram et al. 1998; Reyes et al. 1998; Kozloski et al. 2001; Beierlein et al. 2003) as well as the variety of target cell properties (Peters & Jones, 1984; Kawaguchi & Kubota, 1997; Gupta et al. 2000). There was a substantial number of cases in which the peak lag in P-I correlations was negative (20 cases in P-I_d compared to one case in P-I_f correlations), meaning that in these cases the average excitatory input to the interneurones preceded the input to pyramidal neurones. Negative delays may occur for numerous reasons such as differential targeting of synapses (peri-somatic synapses will have shorter latencies than dendritic ones), faster synaptic kinetics as well as stronger synaptic depression in excitatory inputs to these interneurones (as they all occurred in P-I_d pairs except in one P-I_f pair; see Fig. 3C).

We further examined whether differences in cross-correlations might be related to differences in other neuronal and synaptic properties. Membrane time constants of recorded neurones were 23.9±3.1ms (n=30) for pyramidal neurones, 16.4±4.8ms (n=50) for interneurones receiving depressing excitatory connections, and 23.6±6.9ms (n=40) for interneurones facilitating excitatory connections. Differences in membrane time constants could therefore not account for the observed peak lags in cross-correlations since the range of these lags was more than 5 times larger than that of the membrane time constants (136.5 compared to 25.7ms). We then examined the dependence on the intrinsic properties of the neurones by comparing cross-correlations obtained for two different electrophysiological types of interneurones (according to Gupta et al. 2000), both of which received depressing excitatory synapses. The observed differences were non-significant (peak lag for stuttering type, 12.1±6.0 ms, n=14; non-accommodating type, 14.1±4.2 ms, n=21; in all cases, P > 0.05). More importantly, the peak lag of the cross-correlations for P-I_d pairs where most interneurones exhibited delayed discharge in response to near threshold step currents was actually much shorter than for P-I_f pairs where the interneurones responded rapidly, with an initial burst of two to three action-potentials. Another possible factor affecting the temporal properties of the cross-correlograms is the synaptic kinetics, determining the shape of the individual EPSPs. EPSP rise times were for P-P 2.4±0.7ms (n=30), for P-I_d 1.9±1.1ms (n=15), and for P-I_f 2.5±1.1ms (n=19). EPSP decay times were for P-P 22.2±5.4 ms, for P-I_d 13.0±5.8 ms, and for P-I_f 19.5±11.2 ms. Kinetics of the different EPSP types differed by only a few milliseconds (see also Thomson et al. 2002; Beierlein et al. 2003), suggesting that the differences in the kinetic properties of EPSPs also can not account for the observed correlation distribution.

In order to further explore the activity conditions generating the different cross-correlation profiles, we simulated integrate-and-fire neurones receiving common excitatory input conveyed by either static or dynamic synapses (Fig. 4A). The subthreshold voltage traces of these neurones were cross-correlated in the same way as for experimental data. Presynaptic activity was constructed from experimentally derived discharge statistics, containing epochs of low (1Hz) and high (200Hz) rates. When the synapses were static (without short-term plasticity), the resulting cross-correlograms were symmetrical and had small peak lags. However, when dynamic synapses were used, peak and median lags depended on the dynamic parameters values, increasing as synapses were less depressing and more facilitating (Fig. 4B and C). When the experimentally derived presynaptic discharge statistics (alternating frequencies, C.V.>200%) were replaced by regular action potential trains (C.V.=0%), cross-correlograms were symmetrical and independent of synaptic dynamics (Fig. 4D). When single-rate Poisson trains (C.V.=100%) were used, median lags reached much smaller values than in the experimentally derived discharge (Fig. 4D). These simulation results suggest that the importance of dynamic synapses in shaping the peak excitatory input is pattern dependent.

Figure 4. Delays in cross-correlations depend on synaptic dynamics and activity patterns.

Simulated integrate-and-fire neurones received common input conveyed by synapses with different dynamics. A, simulated integrate-and-fire neurones received common-input from presynaptic neurones that discharged with alternating frequencies, similar to discharge patterns observed in experiments. B, the input was conveyed by static or dynamic synapses. When the conveying synapses were static, peak lags were small and correlograms were symmetrical, whereas when dynamic synapses were used, cross-correlograms had larger peak and median lags that depended on the dynamics. C, a series of cross-correlograms of simulated voltage traces from pairs of neurones, one of which was a pyramidal neurone and the others were interneurones receiving input by synapses with incrementally increasing facilitation (increasing facilitation time constant and decreasing depression time constant), as depicted by the arrow on the left. D, temporal lags in the cross-correlation functions depend on the discharge pattern. Cross-correlation median lags are plotted against the degree of facilitation of the synapses, as characterized by the ratio between the time-constants of facilitation and depression (F/D ratio). Median lags were longer and strongly depended on the dynamics for patterned input (black trace) than when discharge was a steady-state Poisson train (grey trace) or regular (dashed trace).

Discussion

In summary, we found that characteristic correlation profiles of excitatory synaptic input emerge between different types of neurones when the neocortical microcircuit is activated. Further analysis revealed that the temporal properties of the correlation profiles can be predicted by the different dynamics of excitatory synapses which neurones receive. Simulations showed that: (a) synaptic dynamics are capable of inducing the temporal dispersion of excitatory input, (b) when activated by regular or single-rate Poisson discharge, synaptic dynamics do not induce temporal dispersion, but (c) when activated by patterned activity, as found experimentally in the excited microcircuit, synaptic dynamics do impose temporal dispersion.

In this study we focused mainly on the role of synaptic dynamics in shaping the temporal properties of the microcircuit activity although other network properties such as electrotonic distances of synaptic connections, differential effect of neuromodulators, and intrinsic ionic currents may also contribute to the temporal properties of excitatory input to different neurone populations. Other temporal properties such as the membrane time constants and EPSP kinetics differed between interneurone populations, such that interneurones receiving facilitating synapses also had longer membrane time constants and longer synaptic decay times, potentially contributing to the temporal dispersion imposed by synaptic dynamics.

The diversity of morphological and electrophysiological types of interneurones (Peters & Jones, 1984; Kawaguchi & Kubota, 1997; Gupta et al. 2000) suggests that interneurones have different roles in neocortical microcircuit operations. Anatomical studies have also shown that different types of interneurones innervate different domains of target neurones (Somogyi et al. 1998), further implying a differential function. Experiments have indeed demonstrated that interneurones targeting different postsynaptic domains have a differential impact on neuronal discharge (Klausberger et al. 2004). Interneurones are also differentially targeted by thalamocortical synapses (Beierlein et al. 2003), further suggesting that their different functions might be crucial in sensory processing. Our data show, for the first time, that various interneurone types are maximally excited at different times during information processing in the active microcircuit. We further show that this temporal dispersion, imposed by dynamic synapses, requires patterned activity in the network. We therefore conclude that synaptic dynamics can ‘choreograph’ excitatory input within the neocortical microcircuit, in a context-dependent manner.

Our findings are consistent with in vivo experiments showing that stimulus–response patterns of cortical neurones contain excitatory and inhibitory inputs occurring at various relative delays (Zhu & Connors, 1999; Anderson et al. 2000; Monier et al. 2003). Discharge patterns containing rate transitions and bursts have been reported both in vitro and in vivo and in different brain states (Bair et al. 1994; Stern et al. 1997; Sanchez-Vives & McCormick, 2000; Vinje & Gallant, 2000; Chiu & Weliky, 2001; Steriade, 2001; Monier et al. 2003), showing that such activity patterns are common in the neocortical microcircuit and would indeed enable synaptic dynamics to differentially shape neuronal excitation.