Synaptic currents at individual connections among stellate cells in rat cerebellar slices

Abstract

1. Unitary inhibitory synaptic connections among stellate cells were studied in rat cerebellar slices. Presynaptic action potentials and inhibitory postsynaptic currents (IPSCs) were simultaneously recorded by loose cell-attached and tight-seal whole-cell recording, respectively.

2. Several types of synaptic connections were distinguished on the basis of the shape of the amplitude distribution of successfully evoked currents. For simple synapses, which presumably arise at single release sites, these histograms could be fitted to a single Gaussian (5 cases). In four additional cases a small amplitude component (< 50 pA) was superimposed to a single Gaussian peak. The small events had slow rise times and widely distributed amplitudes. Finally eleven histograms showed two or more Gaussian components and were classified as complex connections.

3. Failure rates ranged from 0.06 to 0.85 for unitary connections (n= 20) and from 0.59 to 0.78 for simple synapses (n= 5).

4. Coefficient of variation values derived from Gaussian fits to simple synapse histograms ranged between 0.20 and 0.38 (n= 5).

5. In simple synapses peak current amplitudes were positively correlated to both current rise time and decay half-width.

6. Intervals between presynaptic action potentials were widely distributed. During stationary periods there was no correlation between interspike interval and amplitude size, success rate or latency. In some experiments, episodes with shorter interspike intervals were observed. During these periods, amplitude and success rate decreased, and the latency increased. Thus, IPSC characteristics depend on the mean frequency of presynaptic spikes, but not on random fluctuations of interspike intervals during stationary periods.

Stellate cells together with the closely related basket cells, constitute the two classically distinguished types of inhibitory interneurones in the molecular layer of the cerebellar cortex. Stellate cell somata are located in the external two-thirds of the molecular layer while basket cell somata are in the internal third. For both cell types the dendritic and axonal arborizations are essentially contained in a single plane which is parallel to the sagittal plane for the part of the cerebellum which is close to the centre of the vermis. Interneurones receive excitatory inputs from parallel fibres as well as from climbing fibres, and they send inhibitory synapses to Purkinje cells and to other interneurones. Their axons tend to run parallel to the Purkinje cell layer (Palay & Chan-Palay, 1974).

In recent years, the elucidation of the electrical properties of stellate cells has made rapid progress. Stellate cells fire action potentials even in the absence of synaptic input (Llano & Gerschenfeld, 1993; Llano & Marty, 1995). Their voltage-dependent conductance mechanisms have been studied in detail (Midtgaard, 1992). Spontaneous unitary synaptic currents have been described under voltage-clamp conditions both for excitatory and inhibitory inputs (Llano & Gerschenfeld, 1993). Interestingly, miniature IPSCs (mIPSCs) recorded in the presence of tetrodotoxin have mean amplitudes as large as those of spontaneous tetrodotoxin-sensitive IPSCs. mIPSCs are also unusual in that they have a large mean amplitude (near 140 pA) and a very skewed amplitude distribution that occasionally extends up to 1 nA (data obtained at -60 mV under symmetrical high Cl⁻ concentrations). Yet the shape and mean value of the amplitude distribution of mIPSCs is not strongly modified by removal of extracellular Ca²⁺ ions, so that the hypothesis that large mIPSCs arise from synchronized vesicular release following depolarization-driven Ca²⁺ entry can be dismissed. Based on these observations it was proposed that synapses among stellate cells comprise a small number of release sites, and that each release site is associated with a large unitary current (Llano & Gerschenfeld, 1993). It is therefore likely that at least some of the synapses of stellate cells involve a single release site with a large quantal size. Such conditions could provide an opportunity to study a single release site synapse with an excellent signal-to-noise ratio. Anatomical data indicate that the main source of inhibitory synapses into interneurones are other interneurones (Palay & Chan-Palay, 1974). Therefore, we decided to study synaptic currents elicited in one stellate cell following spontaneous firing of action potentials in another identified stellate cell. The electrical activity in the presynaptic stellate cell was monitored by loose patch-clamp recording, while the postsynaptic currents were measured with tight-seal whole-cell recording.

METHODS

Slice preparation and identification of stellate cells

The basic procedures for the preparation and maintenance of slices were described in the paper by Llano & Gerschenfeld (1993). Rats (14-21 days old) were decapitated under anaesthesia with Metofane (Mallinckrodt Veterinary, Bray, Ireland), the cerebellum was removed and slices (180 μm thick) were cut from the vermis, parallel to the sagittal plane. Slices were incubated for 1-6 h at 33°C in oxygenated saline until transfered to the recording chamber. This saline contained (mm): 125 NaCl, 2.5 KCl, 1.3 NaH₂PO₄, 26 NaHCO₃, 2 CaCl₂, 1 MgCl₂ and 10 glucose.

During the experiments, the preparation was observed with an upright microscope (Axioskop, Zeiss, Göttingen, Germany; we used a × 63 water immersion and 0.9 numerical aperture objective) equipped with differential interference contrast. Stellate cells were unambiguously identified according to the following criteria. (1) Size. Stellate cells have diameters of about 8 μm, significantly larger than that of migrating granule cells. (2) Electrical activity. In the molecular layer, only basket and stellate cells show a significant number of spontaneous synaptic currents and large voltage-dependent inward currents. We routinely checked I-V plots elicited by depolarizing steps and these plots matched those reported by Llano & Gerschenfeld (1993). (3) Location. Among interneurones, stellate cells are distinguished from basket cells on the basis of their position in the molecular layer. Recordings were only taken from the external half of the molecular layer and this should guarantee that the recorded interneurones are stellate cells. (4) Morphology. Some of the cells recorded were stained by the biocytin-filling method and their morphology was investigated, later confirming their identification as stellate cells (data not shown).

Recording and data acquisition

Pipettes for loose cell-attached and tight-seal whole-cell recordings were pulled from borosilicate glass, fire-polished and coated with dental wax. Pipettes were filled with a K⁺-based internal solution, which contained (mm): 120 KCl, 4.6 MgCl₂, 10 EGTA, 1 CaCl₂, 10 Hepes, 4 Na-ATP and 0.4 Na-GTP. With this internal solution the reversal potential for GABA-sensitive currents was close to 0 mV. The resistance of these pipettes in the bath solution was in the range 4-8 MΩ. Effective access resistance during postsynaptic whole-cell recording was in the range 10-30 MΩ. No series resistance compensation was employed. The holding potential for postsynaptic recordings was -60 mV. For presynaptic recordings, the pipette was gently placed in contact with a potential presynaptic stellate cell. No suction was applied during recording. The pipette potential was maintained at the bath potential. Pre- and postsynaptic signals were recorded using two patch-clamp amplifers (EPC7, List Electronic, Darmstadt, Germany; EPC9, HEKA Elektronik, Lambrecht, Germany). Signals were filtered at 2 or 3 kHz, digitized and sampled at 4 kHz, and stored onto the hard disk of a computer.

Interneurone dendrites for young rats are short (< 100 μm) and thick (> 1 μm; e.g. Llano, Tan & Caputo, 1997) suggesting that they are electrically tight. Indeed rise times of IPSCs are very short compared with the faster kinetic component of the decay phase, indicating that dendritic filtering is negligible (Llano & Gerschenfeld, 1993).

In the recording chamber slices were perfused with oxygenated saline at a rate of 2 ml min⁻¹ and were kept at room temperature (20-22°C).

Analysis of evoked IPSCs

Action potentials were identified in the presynaptic trace by using a detection routine (IGOR, Wavemetrics, Lake Oswego, USA) written by C. Pouzat. This routine allowed us to align sections of the postsynaptic trace with respect to the position of the presynaptic action potential. Figure 1 shows the method used to measure the parameters of evoked IPSCs (eIPSCs). The onset position was determined by comparison of a detection threshold with the value of the standard deviation of the current (s.d.) in a movable window period. The control s.d. before the peak of action potential was calculated from the mean s.d. values for five consecutive sequences of five data points. The movable window had also five data points. The search started from the peak of the presynaptic action potential. When the s.d. in the test region exceeded the threshold (1.5 × control s.d.), the search stopped and the fourth point in the window was considered as the onset point (point 2 in Fig. 1). The program then searched for the local minimum closest to the onset point. This was taken as the peak position (point 6 in Fig. 1). The latency (a) was calculated as the time between the peak of the presynaptic action potential and the onset point. The peak amplitude (c) was calculated by subtracting the amplitude value of the peak position from that of the onset point. No averaging was performed on the baseline points in order to minimize errors associated with sloping baselines resulting from previous spontaneous IPSCs. The rise time (b) was measured from the times corresponding to 10 and 90 % of the peak amplitude.

Figure 1. Measurement of peak amplitude and kinetic parameters of IPSCs.

Single pre- and postsynaptic traces from cell pair 1. The postsynaptic trace was smoothed using a 3-box filter. Arrow 1 is placed at the tip of the presynaptic action current, and arrow 2 indicates the event onset, determined as explained in Methods. The latency, a, is the difference between the positions of arrows 1 and 2. Arrow 6 indicates the position of the peak current. The event amplitude, c, is the difference between current amplitudes at cursors 2 and 6. Arrows 3, 4 and 5 indicate the positions at 10, 50 and 90 % of the full amplitude. The 10-90 % rise time is the difference between the positions of arrows 3 and 5, and d shows the half-width of the current decay.

For noise histograms a number of points equal to five times the number of successful events were randomly chosen on the raw traces. Current values were entered for each of these points as well as for points located after time intervals equal to those between onset and peak in eIPSCs. Noise amplitudes were calculated by subtracting the values for each set of two points. In the figures noise histograms are scaled to the same peak amplitude as the histograms of eIPSCs.

Modelling three component connections

Three sites models

Amplitude distributions for each site are assumed to be Gaussian with a common mean amplitude, a₀, and a common s.d. value, s₀. The sites are initially assumed to be totally independent. Then the mean and s.d. values for the second and third peak are 2a₀, 3a₀ and √2s₀, √3s₀, and the peak amplitudes follow the predictions of a binomial model with n= 3. The fit was performed in two steps, first adjusting a₀ and s₀ by eye, and then the probability of release p by using a least square minimization routine. This model gave clearly unsatisfactory fits. Therefore a degree of coupling was introduced by allowing the release probabilities to alternate between 0 and p. When the release probabilities were not 0 the events from the three sites were still combined together on the basis of independence. This could be treated using the same model by scaling the probability histogram in such a manner that the success rate could be artificially increased from the observed value up to 1.

Two sites models

Here, two release sites are assumed to have release probabilities p₁, p₂. Amplitude distributions for each site are assumed to be Gaussian with mean amplitudes a₁, a₂, and s.d. values s₁, s₂. The two sites are first assumed to be totally independent. Then the probabilities to observe events in the first and second peak are:

(1)

(2)

The third peak is generated by random superimposition of events from the first and second sites such that:

(3)

(4)

(5)

where a₃, s₃ and P₃ are, respectively, the mean and s.d. of the amplitudes, and the probability of occurrence of the third peak events.

The parameters a₁, a₂, s₁ and s₂ were first adjusted by eye in order to fit the position and width of the two first peaks. p₁ and p₂ were then optimized by using a least square routine in order to fit the entire amplitude histogram. In a second model variant, coupling was introduced between the two sites by assuming that the release probabilities could take either the values (0, 0) or (p₁, p₂). This variant often gave better results than the original model based on total intersite independence.

RESULTS

This paper presents results obtained with paired recordings of synaptically connected stellate cells in rat cerebellar slices (2-3 postnatal weeks old). We found in preliminary experiments that if the presynaptic cell was submitted to whole-cell recording, evoked responses gradually decreased with time and were abolished after a period of 15 min (3 cells, data not shown). This phenomenon is presumably related to dialysis of presynaptic factors required for vesicle release, since in the same cells IPSCs, which were not elicited by firing of the presynaptic cell, were not significantly modified. Therefore, in subsequent experiments we combined tight-seal whole-cell recording of evoked IPSCs (eIPSCs) with loose cell-attached recording of presynaptic action potentials. Under these circumstances we obtained stable synaptic connections for periods extending to more than 1 h. Another advantage of this approach was that several potential presynaptic neurones could be assayed in quick succession using the same recording pipette. Overall using this method we found twenty synaptically connected pairs out of one hundred trials (20 %). In the same series two electrical connections were found (2 %; data not shown; in neither case did the electrically coupled cells have chemical synaptic connections). These results were obtained with animals aged 2-3 weeks. Out of forty trials in 4- to 5-week-old animals, we found only one synaptically connected pair (results not included in this manuscript), indicating that the rate of success of paired recordings is greatly influenced by the age of the experimental animal.

An example of a simple synapse connection

Figure 2 illustrates data from a ‘simple synapse’ (called cell pair 11 in Table 1). As shown in Fig. 2B, spontaneous IPSCs occurred at random intervals, representing the contribution of various presynaptic neurones. Presynaptic action potentials that elicited a postsynaptic response are identified by arrowheads in Fig. 2B. In 6 min of recording, 285 action potentials out of 774 elicited postsynaptic responses, so that the failure rate was 0.63. Interspike intervals were broadly distributed and had a stable average near 460 ms during the entire time period of recording (Fig. 2C). Over the same period of time, the mean amplitude of events excluding failures was stable (Fig. 2C). Figure 2D shows some examples of pre- and postsynaptic traces, including one failure. Latency values ranged from 1 to 2.25 ms in this synapse. The distribution of peak IPSC amplitudes (excluding failures) could be fitted by a single Gaussian with a mean of 75.5 pA and a s.d. of 21.4 pA, so that the coefficient of variation (c.v. =s.d./mean) was 0.28 (Fig. 2E). The noise histogram from the same recording had a s.d. of 3.2 pA, much smaller than that of successful responses (Fig. 2E). Finally Fig. 2F shows the amplitude histogram for all IPSCs recorded in the postsynaptic cell together with the histogram of Fig. 2E corresponding to the input of the recorded presynaptic neurone. This connection accounted for a small part (6 %) of the total IPSC amplitude distribution. The means of the two distributions were markedly different (75.5 vs. 256.3 pA), and the IPSC amplitudes corresponding to the recorded presynaptic cell had a much narrower distribution than that of all spontaneous IPSCs.

Figure 2. A paired-recording experiment from a simple connection.

A, schematic drawing of the recording configuration. B, a section of pre- and postsynaptic traces at slow time scale. Arrowheads indicate presynaptic action potentials that elicited a postsynaptic response. Here the presynaptic action potentials are recorded as vertical lines of various amplitudes. The variability in the corresponding amplitudes is due to the limited sampling rate of the figure; when plotted with higher time resolution the action potential waveforms appear very reproducible (D). C, interspike intervals are distributed between ≈0.1 and 2 s. Mean amplitudes and mean interspike intervals are stable over the duration of the experiment that was used for the analysis (6 min). D, examples of pre- and postsynaptic traces at a faster time scale. Only one failure is shown, but the failure rate was 0.63. E, amplitude distributions for successful events and for the recording noise. The eIPSC distribution can be fitted to a single Gaussian with mean value of 75.5 pA and a s.d. value of 21.4 pA, whereas the s.d. of the noise histogram was 3.2 pA. The noise histogram is normalized to the same maximum as the histogram of successful responses. F, amplitude distribution for all IPSCs recorded in the examined period (6 min). The histogram from E is also shown superimposed.

Table 1.

Summary of the analysis of 20 cell pair experiments

Cell pairs	Action potential frequency (Hz)	Success rate	Mean amp. (pA)	Latency (ms)	Rise time (ms)	Half-width (ms)	eIPSC/sIPSC	Category
1	0.94	0.55	98.1	1.9 ± 0.5	0.4 ± 0.2	14.6 ± 3.4	0.24	Simple + slow
2	0.37	0.36	120.4	3.1 ± 0.5	0.7 ± 0.2	8.3 ± 1.9	0.02	2 component + slow
3	6.02	0.41	77.5	1.0 ± 0.4	0.6 ± 0.1	10.5 ± 2.2	0.43	3 component
4	2.15	0.91	131.6	0.9 ± 0.3	0.8 ± 0.2	10.9 ± 2.7	0.21	Multiple
5	1.96	0.16	48.5	0.7 ± 0.2	0.9 ± 0.1	7.1 ± 2.6	0.08	3 component
6	6.07	0.23	29.1	2.2 ± 0.3	0.6 ± 0.1	5.9 ± 2.6	0.41	Simple
7	2.91	0.57	95.9	1.6 ± 0.4	0.7 ± 0.2	14.1 ± 3.7	0.64	Multiple
8	4.96	0.22	67.6	1.7 ± 0.5	0.7 ± 0.1	6.9 ± 1.6	0.14	Simple + slow
9	1.62	0.20	30.2	1.9 ± 0.4	0.7 ± 0.2	6.7 ± 2.6	0.03	Simple + slow
10	5.42	0.22	45.8	2.0 ± 0.5	0.6 ± 0.2	11.3 ± 4.3	0.25	Simple
11	2.15	0.37	75.5	1.5 ± 0.3	0.6 ± 0.1	8.9 ± 2.8	0.06	Simple
12	6.00	0.28	160.3	2.4 ± 0.6	0.8 ± 0.2	7.7 ± 1.7	0.23	Multiple
13	2.41	0.94	475.8	1.2 ± 0.2	0.6 ± 0.2	5.3 ± 1.4	0.20	Multiple
14	1.27	0.41	85.2	1.4 ± 0.4	0.4 ± 0.1	7.4 ± 2.3	0.09	Simple
15	4.93	0.47	189.2	1.2 ± 0.4	0.6 ± 0.2	9.1 ± 1.9	0.79	Multiple
16	4.02	0.24	159.0	1.6 ± 0.4	0.6 ± 0.1	8.1 ± 1.8	n.d.	3 component
17	2.88	0.15	60.4	1.6 ± 0.4	0.5 ± 0.1	7.2 ± 4.7	0.15	Simple + slow
18	1.78	0.29	319.1	1.5 ± 0.3	0.5 ± 0.1	7.3 ± 1.1	0.34	3 component
19	3.98	0.23	63.2	1.6 ± 0.3	0.6 ± 0.1	8.2 ± 2.1	0.38	Simple
20	2.62	0.40	118.8	2.0 ± 0.4	0.5 ± 0.1	15.2 ± 4.6	0.57	3 component

Fourth column, mean amplitudes of IPSCs excluding failures. Eighth column, ratio between frequencies of eIPSCs and sIPSCs. Values are means ±s.d. n.d., not determined.

The Gaussian amplitude distribution of eIPSCs illustrated in Fig. 2E is the hallmark of what will be called a simple connection hereafter.

Correlation of rise time and half-decay time with peak amplitude at simple connections

We noticed when analysing the results of simple connections that the rise and decay kinetics of individual traces were correlated with their peak amplitudes. This effect can be conveniently shown by grouping eIPSCs into three categories according to amplitude, as illustrated in Fig. 3. Marked differences were apparent between the normalized means from the three groups. The mean from the largest amplitude group had the longest rise time and the slowest decay kinetics while the smallest amplitudes were associated with the shortest rise time and fastest decay (Fig. 3A-B). Since this analysis involves an arbitrary choice of boundaries between groups, rise times and half-widths were also plotted as a function of amplitudes for individual responses. Highly significant correlations were found in both cases (Fig. 3C-D; P < 0.001 for both plots), showing that the correlations illustrated in Fig. 3A-B do not depend on the choice of group boundaries. The implications of the results of Fig. 3 will be discussed below.

Figure 3. Rise time and decay kinetics at a simple synapse.

Same data as in Fig. 2. A-B, means were calculated for three amplitude ranges, as indicated by black areas in the histogram (inset: 30-50 pA (n= 9); 70-80 pA (n= 40); and 100-110 pA (n= 17)). After normalization, the trace from the 30-50 pA group (dotted line) is found to have the fastest rise and shortest decay, while that from the 100-110 pA group (continuous line) has the slowest rise and longest decay. C-D, both the 10-90 % rise time and the half-width are positively and significantly (P < 0.001) correlated with the peak amplitude.

Out of a total of five pairs which behaved as simple connections, three gave a significant correlation between amplitude and kinetics. In all three cases the correlation was similar to Fig. 3: large amplitudes were associated with slow onset and decay kinetics, and small amplitudes with fast kinetics.

Classification of unitary interneurone-interneurone connections on the basis of amplitude histograms

Simple connections were only found in 25 % of the pairs. The shape of the amplitude distributions of eIPSCs varied greatly from connection to connection. Mean amplitudes excluding failures ranged from 30 to 500 pA, and failure rates ranged from 0.06 to 0.85. Figure 4 shows examples from several categories that were distinguished on the basis of the shape of amplitude histograms in the twenty pairs that were examined. The complexity of amplitude distributions grows from A to D in Fig. 4.

Figure 4. Classification of paired recordings according to the shape of the amplitude histograms.

Representative examples for four categories of paired recording results are shown. In each case only amplitudes for successful stimulations are displayed. The histogram Aa illustrates a simple synapse (cell pair 19) similar to that exemplified in Figs 2 and 3. Ba represents the superimposition of a simple synapse histogram with an additional component of events smaller than 50 pA (cell pair 17). These events had slow rise times of the order of several milliseconds. C represents a 3 component synapse which is further illustrated in Fig. 5 below (cell pair 20). D illustrates a complex connection with more than 3 components (cell pair 13). Traces Ab, Bb and Cb are scaled means for three sections of the corresponding amplitude histograms. Dotted, dashed and continuous traces correspond to the smallest, intermediate and largest responses, respectively. The traces Ab follow the same pattern as that shown in Fig. 3, where the mean amplitude is positively correlated to both rise time and decay half-width, but those in Bb and Cb deviate from this pattern.

Figure 4A illustrates another example of a simple connection histogram. In this case the histogram could be fitted to a single Gaussian with a mean of 62 pA and a s.d. value of 12 pA (c.v. 0.20). Mean current kinetics for three amplitude groups are shown in panel Ab. As in the previous example shown in Fig. 3 the kinetics are slowest for the largest amplitude group, and fastest for the smallest amplitude group.

The amplitude distribution represented in Fig. 4B contains a Gaussian peak reminiscent of simple connections and in addition a component ranging from the threshold of detection (10 pA) to 50 pA. The small IPSCs displayed much slower rising phases than the larger amplitude events. These currents appear to be mediated by GABA_A receptors, because they are abolished by bicuculline (C. Auger, personal communication), but otherwise the underlying mechanisms are unclear. The middle and large amplitude groups follow the same rules as in simple connections: the largest amplitude group has a slower rise time and decay than the middle amplitude group (Fig. 4Bb). However due to the presence of the very slow events in the lowest amplitude group, this group displays markedly slower rise and decay kinetics than the other two groups (Fig. 4Bb). The pattern of amplitude histogram of Fig. 4B, called ‘simple + slow’, was found in three other pairs. The contribution of the slow currents varied markedly among ‘simple + slow’ paired recordings; Fig. 4B illustrates the case where this contribution was largest.

In a third category of connections (5/20), the main part of the amplitude distribution (excluding failures) could be approximated with the sum of three Gaussian curves (Fig. 4C). They were therefore classified as ‘3 component’ connections. As shown below, these histograms can be accounted for on the basis of the superimposition of two or three independent simple connections.

In a fourth category of connections (5/20) the amplitude histogram was quite broad and could not be unambiguously separated into Gaussian peaks (Fig. 4D). These connections were called ‘multiple’.

There was no significant correlation between amplitudes and rise time or decay kinetics for 3 component or for multiple connections (cases C and D in Fig. 4; see example of Fig. 4Cb). Thus the behaviour illustrated in Fig. 3 and in Fig. 4Ab only applies for simple connections.

Finally one connection gave an amplitude histogram that could be fitted with two Gaussians and which also contained some small and slow events. The mean amplitudes of the two components were near 105 and 150 pA. Since there was a single example of a 2 component histogram it was unclear whether this corresponds to a separate category of synapses or to an odd observation, and the experiment was not further analysed.

Modelling 3 component connections

Figure 5 illustrates fits to two examples of 3 component histograms. As the peaks are roughly equidistant, they can be fitted either using models with three sites having the same quantal size or with two sites having quantal sizes in a ratio of 2 to 1.

Figure 5. Fit of 3 component amplitude histograms with multiple Gaussian curves.

Two examples of complex amplitude histograms are illustrated where three roughly equally spaced peaks can be distinguished. The histograms can be fitted using two different classes of models. The fits in the upper panels were generated using the 3 sites models where each site contributes events with amplitudes close to the location of the first peak. The fits in the lower panels assume two independent release sites with peak amplitudes corresponding to the locations of the first and second peak. The third peak is generated by random superimpositions of events contributed by each of the two sites. A, failure rate, 0.59. Total number of entries for successful stimulations, 445. Upper panel: 3 sites models. First peak amplitude: 41 pA; s.d. 9 pA. A 3 sites binomial model cannot fit the data (dashed line). A better fit is obtained by assuming that release probabilities alternate between 0 and a common value p (continuous line; p= 0.46; effective failure rate 0.12 when p > 0; see Methods). Error bars show plus (continuous line) or minus (dashed line) the s.d. of the values obtained from Monte Carlo simulations based on the two theoretical curves. Lower panel: 2 sites models. Peak amplitudes of 39 and 82 pA, and corresponding s.d. values of 9 and 13 pA. Dashed line: an acceptable fit is obtained assuming an effective failure rate of 0.59, as observed. p₁= 0.24, p₂= 0.21. Continuous line: the fit can be slightly improved by assuming alternations between release probabilities of (p₁, p₂) and (0, 0) and a failure rate of 0.54 when the release probabilities are > 0. B, another complex amplitude histogram with three peaks (same histogram as in Fig. 4C). Failure frequency, 0.60. Number of successful stimulations, 379. Upper panel: 3 sites models. First peak at 80 pA, corresponding s.d. 15 pA. Dashed line: 3 sites binomial model. Continuous line: binomial model with probability fluctuations. p= 0.26; effective failure rate = 0.34 when p > 0. While the latter model correctly fits the second peak of the histogram, it fails to fit the third one properly. Error bars show plus (continuous line) or minus (dashed line) the s.d. of the values obtained from Monte Carlo simulations based on the two theoretical curves. Lower panel: 2 sites models. Peaks at 80 and 157 pA, and corresponding s.d. values of 15 and 20 pA. Dashed line: an acceptable fit is assuming an effective failure rate of 0.60, as observed. p₁= 0.30, p₂= 0.13. Continuous line: the fit can be slightly improved by assuming an artificial failure rate of 0.51. Normalized noise histograms are shown for A and B.

The upper panels of Fig. 5 illustrate fits with three sites models. A binomial model with equal probabilities at each site was unsuccessful (dashed lines). Better results were obtained if the probability of failure was left as a free parameter to account for the possibility of variations of the release probability from one trial to the next, for instance because of propagation failures (continuous lines). Nevertheless the fits were still not optimal, particularly in the example shown in Fig. 5B. Independent three sites models assuming different release probabilities at each site did not give significantly better fits than those of binomial distributions (not shown).

Results of fits with two sites models are shown in the lower panels of Fig. 5. The sites were assumed to be independent. If the failure rate was held at the measured value an acceptable fit was obtained in both cases (dashed lines; eqns (1)–(5)). Slightly better fits could be obtained by leaving the success rate as a free parameter (continuous lines).

A similar analysis was carried out in three further paired recordings. In all cases the histogram could be fitted by the sum of three roughly equidistant components. In one of these three additional experiments the first peak was the largest, as in Fig. 5, and in the two other cases it was the second peak that was largest. Models using three independent release sites were totally unsuccessful in two of these experiments; however, they could fit the data in the last case provided that the success rate was allowed to be raised above the measured value. Models with two release sites could fit the data in all cases. However, only in one case was it possible to fit the data by using the measured success rate. In the other two cases it was necessary to raise the success rate above the measured value to obtain an acceptable fit.

Overall, models with three independent sites were usually not satisfactory. Models with two release sites worked better, but in several cases the fit was only possible when assuming fluctuations in the underlying release probabilities from trial to trial.

Lack of correlation between latency or amplitude and interspike interval

At most synapses, repeating presynaptic stimuli with intervals up to a few hundred milliseconds causes significant modifications of the postsynaptic response (paired-pulse facilitation or paired-pulse depression). We therefore considered the possibility that the variations in the amplitudes, kinetics and latencies of individual eIPSCs evident in Fig. 2–Fig. 5 could be related to variations in the interspike intervals of the presynaptic cell. The analysis of paired recordings was restricted to periods where presynaptic cell firing was stable. In most experiments, neither the amplitudes nor latencies of individual IPSCs were correlated to the intervals from the previous presynaptic spikes (Fig. 6A). Likewise the failure rate was not correlated to the intervals from the previous presynaptic spikes (results not shown). If only presynaptic spikes that elicited postsynaptic responses were considered, again no correlation was found between intervals from previous events and mean amplitude, mean latency, or mean failure rate of the postsynaptic responses (results not shown). Thus in most experiments there was no link between the characteristics of eIPSCs and the preceding electrical or synaptic activity. However, a slightly different result was found in a few experiments where the interspike-interval distributions extended to less than 100 ms. In these cases a correlation between interspike intervals and amplitude was found for intervals less than 100 ms. Overall, the results indicate that fluctuations in presynaptic firing have very little affect on the strength of the connections. Thus, the variations in amplitude, latency and kinetics of individual eIPSCs cannot be explained on the basis of fluctuations in the interspike intervals.

Figure 6. Effect of interspike intervals on peak amplitudes, latencies and failure rates.

A, lack of correlation between interspike intervals and IPSC amplitudes (Aa) or latencies (Ab) during a stationary presynaptic firing period. The same experiment as in Fig. 4A Each dot shows an individual IPSC. The same results are also presented as histograms with bin lengths of 50 ms. B, in this experiment the mean frequency of presynaptic firing changed during a short period (about 12 s) and then returned to its original value. Ba, plot of interspike intervals over a 4 min period. Continuous line links mean interspike interval values for episodes delimited by vertical dashed lines. Bb, c and d, mean peak amplitude, success rate and latency values for each episode. Error bars represent ±s.e.m. Amplitude and success rate decrease, and the latency increases, during the period with short interspike intervals.

Depression of evoked IPSC during high frequency episodes

As already stated, data were analysed from periods of 5 to 20 min where the frequency of presynaptic action potentials was stable. In three pairs spontaneous transitions were observed for short periods (1-2 min) where interspike intervals assumed a mean value different from that in the rest of the recording. We took advantage of these transitions to see whether mean amplitude, success rate and mean latency would be influenced by changes in mean interspike interval. In every case all three factors changed in parallel with the mean interspike interval. Figure 6B illustrates one of these experiments. After an initial stable period, there was an episode of around 1 min with a higher mean action potential frequency, followed by recovery to the initial frequency. Following the change of interspike interval (from 170 to 94 ms), mean amplitude and success rate decreased, respectively, from 147 to 118 pA and from 0.26 to 0.16, while the mean latency slightly increased from 2.1 to 2.3 ms. When presynaptic firing recovered its initial rate, all three parameters returned to their basal level. In the other two cases, the interspike interval decreased from 256 to 189 ms and increased from 345 to 625 ms, respectively. In both cases mean amplitude, success rate and latency were linked to the frequency change as illustrated in Fig. 6B Thus even though random fluctuations of interspike intervals during stationary firing periods did not affect the synaptic transmission, persisting changes in the mean firing rate induced complex alterations in the IPSC characteristics.

DISCUSSION

Stellate-stellate connections have large unitary currrents and few release sites

Mean amplitudes of eIPSCs (excluding failures) ranged between 30 and 500 pA, corresponding to conductances of 500-8300 pS. These are very large values compared with the resting conductance of stellate cells (150-500 pS, Llano & Gerschenfeld, 1993), even if the conductance increase resulting from the artificially high intracellular Cl⁻ concentration in our experiments is taken into consideration. Therefore, it seems likely that, as it was earlier suggested by Midtgaard (1992; see Fig. 15 of that publication), a single successful IPSC should prevent postsynaptic firing for a large part of the duration of the synaptic current. In all experiments failures were observed, and in most cases one or a few peaks could be distinguished in the amplitude histogram. These results indicate, in conformity with predictions drawn from the study of spontaneous and miniature IPSCs in this preparation (Llano & Gerschenfeld, 1993; Auger & Marty, 1997), that stellate-stellate synapses include only a few release sites.

Simple stellate-stellate connections

Several categories of connections could be distinguished on the basis of the shape of the amplitude histograms of successful events. ‘Simple’ or ‘simple + slow’ connections had a single peak; Gaussian fits to these peaks had mean amplitudes ranging between 26 and 98 pA. The average of these mean amplitude values was 60.7 pA (n= 9).

At some mossy fibre-granule cell synapses, single peak amplitude histograms result from very high release probabilities at separate release sites, because in the absence of failures all release sites add their signals together for each trial (Silver, Cull-Candy & Takahashi, 1996). But at simple connections among stellate cells, the position of the main peak is insensitive to manipulations that lead to marked decreases in the release probability (C. Auger, S. Kondo & A. Marty, unpublished observations). Therefore, simple connections do not reflect multiple release sites with high release probabilities. Another possibility would be that two or several release sites have similar quantal sizes and a very low release probability such that double events would be too rare to be detected. But this hypothesis is not consistent with the fact that events were not observed at twice the amplitude of the main peak. This point can be made more explicitly by examining the example of Fig. 2, where the failure rate is 0.63, and the total number of trials is 774. For two independent sites with equal release probabilities, the probability of obtaining two simultaneous release events can be calculated as 0.04 from the value of the success rate (0.37), corresponding to a number of double events of 0.04 × 774 = 31. This model therefore predicts a peak with thirty-one events centred at twice the mean amplitude of the first peak, that is at 2 × 75.5 = 151 pA. The histogram of Fig. 2E is clearly incompatible with such a prediction. Models assuming more release sites with equal release probabilites predict an even larger size of the second peak. Thus for a Poisson processs (large number of independent presynaptic sites) the predicted number of events in the second peak would be 48. Similar considerations can be made if two or several sites with different release probabilities are considered. If for example two sites are considered with very different release probabilities (one main site and one contaminating site) it can be estimated that the contaminating site should contribute less than 2 % of the main peak in the example of Fig. 2, or else a second peak would be visible. By this reasoning, at least 98 % of the events of Fig. 2E arise from a single site. In summary then, the absence of a second peak in amplitude histograms, together with the relatively high value of success rates, constitutes firm evidence that simple connections involve a single functional release site.

It is interesting to compare the above value for the mean peak amplitude at simple synapses with two other sets of results. Following treatment with low doses of α-latrotoxin, bursts of mIPSCs were observed. Some of these bursts, called simple bursts, were interpreted as arising at single release sites (Auger & Marty, 1997). These bursts have a Gaussian amplitude distribution with peak values that vary from burst to burst. The average of these is 54.8 pA (n= 24), in good agreement with the present value of 60.7 pA. The second set of data which is relevant to the present results is the mean value of amplitude distributions of mIPSCs obtained in tetrodotoxin. Mean amplitudes of mIPSCs in stellate cells within the age group 14-21 days was determined at 141 pA by Llano & Gerschenfeld (1993), a value that was confirmed by additional measurements during the course of the present study. Here the difference with the present data seems too large to be explained on the basis of sampling error. One possible explanation for the discrepancy could be that single release sites are preferentially associated with smaller quantal sizes, as suggested earlier (Auger & Marty, 1997). Alternatively, some of the larger mIPSCs could involve simultaneous release from several release sites.

There are conflicting views on the shape of the distribution of quantal amplitudes at a single release site in a central nervous synapse. Until recently the notion prevailed that such distributions are highly skewed (reviewed by Bekkers, 1994). However, recently several examples of Gaussian distributions at single sites have been reported. One example is provided by excitatory synapses between presynaptic pyramidal cells and postsynaptic interneurones in the CA3 region of the hippocampus. In these synapses roughy 90 % of the contacts are due to single sites as determined by electron microscopy, and amplitude histograms of evoked EPSCs are described by a single Gaussian (Gulyas, Miles, Toth, Sik & Freund, 1993; Arancio, Korn, Gulyas, Freund & Miles, 1994). Likewise single Gaussian histograms describe the distribution of evoked EPSCs at presumed single site synapses in mossy fibre-granule cell connections in the cerebellum (Silver et al. 1996), as well as the distribution of miniature EPSCs at single synaptic boutons in hippocampal cultures (Forti, Bossi, Bergamaschi, Villa & Malgaroli, 1997). Thus, despite previous claims to the contrary, new evidence is accumulating suggesting that the distribution of quanta at single synaptic sites in the central nervous system is described by a single Gaussian. The finding of Gaussian distributions at single functional release sites in interneurone-interneurone synapses (Auger & Marty, 1997; and present results) gives further support to this notion and suggests that it also applies to inhibitory synapses.

3 component connections

The analysis illustrated in Fig. 5 suggests that 3 component connections result from the combination of two or three elements, where each element behaves as a simple connection. The mean value for the first peak amplitude of the 3 component connections was 77.4 pA (n= 5), a value reasonably close to that obtained with simple connections (60.7 pA). The corresponding c.v. values ranged between 19 and 26 %, with a mean of 21 %. For comparison c.v. values at simple connections averaged 31 %. The broad similarity between mean amplitude and c.v. values for the first peak of 3 component synapses and for the main peak of simple synapses gives support to the notion that both types of connections employ the same building blocks.

Amplitude-kinetics correlations

At simple connections, the kinetics of evoked currents were correlated with the peak amplitude: large currents were associated with slow rise times and slow decay times. These observations suggest that large amplitudes correspond to long agonist concentration transients. In another publication, we argue that single site synapses at stellate-stellate junctions are able to release several synaptic vesicles in response to a single action potential (C. Auger, S. Kondo & A. Marty, unpublished observations). As receptors are largely bound following the release of a single vesicle (Auger & Marty, 1997) the release of several vesicles leads to a modest increase in peak amplitude. Slight time differences between multiple vesicular release events account for the longer rise time of larger events. The prolonged presence of GABA resulting from asynchronous release can also be responsible for the prolongation of the decay phase for the large events.

Amplitude-kinetics correlations are consistently observed only in simple connections. In complex connections, individual release sites are likely to have different latency profiles and decay kinetics (see Auger & Marty, 1997, for the latter). The relatively small effects apparent at simple synapses are therefore expected to be obscured by the random superimposition of events drawn from different pools.

Use-dependent depression of evoked IPSCs

All results were gathered during stationary periods of presynaptic firing. During such periods the success rate and eIPSC amplitude were not correlated to the interval from the preceding presynaptic spike. This held true independently of whether all spikes were considered or only those that led to a successful synaptic response. These results indicate that a single presynaptic spike, or a single release event and associated postsynaptic response, fails to influence the IPSC characteristics. However it has to be noted that for most experiments > 95 % of interspike intervals were more than 100 ms (see Fig. 6); in the few cases where a significant fraction of interspike intervals was less than 100 ms, paired-pulse depression was obtained.

When interspike intervals were stationary, synaptic parameters such as the mean eIPSC amplitude and failure rate were stable. However, systematic changes were observed whenever spontaneous changes of firing frequency occurred. Following a frequency increase, the amplitude of eIPSC amplitude decreased, while the failure rate and the latency increased. The increase in latency and failure rate indicate a presynaptic modification involving a decrease in release probability. This presynaptic effect could be responsible for the decrease in eIPSC amplitude because of a lower occurrence of multiple release events (see discussion in Kondo & Marty, 1998).

High connectivity among stellate cells

A relatively high level of connectivity (20 %) was found in our preparation compared with values reported in other brain regions. In the cortex the probability of finding connections was 1-10 % between pyramidal cells, 1-3 % for interneurones to pyramidal cell synapses, and 8 % for pyramidal cell to interneurone synapses (Mason, Nicoll & Stratford, 1991; Deuchars & Thomson, 1995; Thomson, West, Hahn & Deuchars, 1996; Markram, Lübke, Frotscher, Roth & Sakmann, 1997). The high yield of connected pairs in our case could be the reflection of a high incidence of stellate-stellate connections. As all stellate cells are mainly contained in one plane parallel to the sagittal plane, the probability of synaptic contacts for two cells which have their somata in such a plane is particularly high. A similar geometrical restriction occurs at interneurone-Purkinje cell synapses, where the proportion of connected pairs can be as high as 70 % (Vincent & Marty, 1996). Additional factors could contribute to the high percentage of connections in the present study compared with other brain regions. (1) Some of the other groups used microelectrode recording and had accordingly a rather low signal-to-noise ratio. Low amplitude connections could be missed in these circumstances. (2) The age of the animals could have an influence on the connectivity (Pouzat & Hestrin, 1997). Our preliminary results on older rat preparations (more than 4 weeks postnatal) showed a lower level of connectivity (2.5 %).

Estimation of the number of presynaptic cells for one postsynaptic stellate cell

For each paired recording we analysed the frequency and amplitude distribution of all recorded spontaneous IPSCs, as illustrated in Fig. 2F. In all cases the total IPSC distribution was broader than that of IPSCs elicited by the presynaptic recorded cell. In many cases the mean of the former distribution was higher than that of the latter, but examples of the contrary were also found. The ratio between the number of eIPSCs that were evoked by the recorded presynaptic stellate cell and the total number of sIPSCs varied among experiments from 0.02 to 0.79 (Table 1). These ratios give an estimate of the proportion of sIPSCs that originated in one presynaptic neurone. They were corrected by a factor of 1.235 to account for the fact that an estimated 19 % of sIPSCs were mIPSCs (Kondo & Marty, 1998). To estimate the number of presynaptic cells to one postsynaptic cell, we took the inverses of the corrected ratios. From the geometric mean of these inverses the mean number of presynaptic cells was calculated as 4.25.

Physiological role of interneurone-interneurone synapses

What is the physiological role of interneurone-interneurone connections? In most of our paired recordings the distance between two stellate cells was less than 100 μm. The projections of stellate cells onto Purkinje cells and interneurones are largely within a distance of 100 μm from the somata (Pouzat & Kondo, 1996). The width of Purkinje cell dendritic trees is of the order of 200 μm, corresponding to one ‘beam’ of parallel fibres. From these numbers it appears that Purkinje cells which are excited by parallel fibre input are inhibited by interneurones receiving the same parallel fibre input. Upon stimulation of the parallel fibre input, interneurones fire up to six times in close succession (Eccles, Llinas & Sasaki, 1966). The role of interneurone-interneurone connections could be to curtail this discharge at high stimulation levels. At low levels of parallel fibre inputs, a small fraction of interneurones get stimulated, and those interneurones are unlikely to receive inhibition from other interneurones until their discharge is over. The inhibition onto Purkinje cells is then long lasting. At a high level of parallel fibre input, in contrast, interneurone-interneurone signals shorten the discharge, effectively quenching the inhibitory activity immediately following the inhibitory action on Purkinje cells. This quenching effect accelerates the return of the interneurone firing rate to the basal level and allows rapid reset of the system to a state where it can effectively process a new parallel fibre signal.