Volume 529, Issue 1 p. 205-213
Free Access

Distinct frequency preferences of different types of rat hippocampal neurones in response to oscillatory input currents

Fenella G. Pike

Corresponding Author

Fenella G. Pike

Departments of Physiology and Pharmacology, and MRC Anatomical Neuropharmacology Unit, Oxford, UK

Corresponding author
F. G. Pike: Department of Physiology, Parks Road, Oxford OX1 3PT, UK. Email: [email protected]Search for more papers by this author
Ruth S. Goddard

Ruth S. Goddard

Departments of Physiology and Pharmacology, and MRC Anatomical Neuropharmacology Unit, Oxford, UK

Search for more papers by this author
Jillian M. Suckling

Jillian M. Suckling

Departments of Physiology and Pharmacology, and MRC Anatomical Neuropharmacology Unit, Oxford, UK

Search for more papers by this author
Paul Ganter

Paul Ganter

Departments of Physiology and Pharmacology, and MRC Anatomical Neuropharmacology Unit, Oxford, UK

Search for more papers by this author
Narayanan Kasthuri

Narayanan Kasthuri

Departments of Physiology and Pharmacology, and MRC Anatomical Neuropharmacology Unit, Oxford, UK

Search for more papers by this author
Ole Paulsen

Ole Paulsen

Departments of Physiology and Pharmacology, and MRC Anatomical Neuropharmacology Unit, Oxford, UK

Search for more papers by this author
First published: 01 November 2000
Citations: 277

Abstract

  • 1

    Coherent network oscillations in several distinct frequency bands are seen in the hippocampus of behaving animals. To investigate how different neuronal types within this network respond to oscillatory inputs we made whole-cell current clamp recordings from three different types of neurones in the CA1 region of rat hippocampal slices: pyramidal cells, fast-spiking interneurones and horizontal interneurones, and recorded their response to sinusoidal inputs at physiologically relevant frequencies (1-100 Hz).

  • 2

    Pyramidal neurones showed firing preference to inputs at theta frequencies (range 2-7 Hz; n= 30). They showed subthreshold resonance in the same frequency range (2-7 Hz; mean 4.1 ± 0.4 Hz; n= 19).

  • 2

    Interneurones differed in their firing properties. Horizontal interneurones in the stratum oriens showed firing preference to inputs at theta frequencies (range 1.5-10 Hz; n= 10). These interneurones also showed resonance at low frequencies (range 1-5 Hz; mean 2.4 ± 0.5 Hz; n= 7). In contrast, fast-spiking interneurones with cell bodies in the pyramidal cell layer fired preferentially at input frequencies in the gamma band (range 30-50 Hz; n= 10/12). These interneurones showed resonance at beta-gamma frequencies (10-50 Hz; mean 26 ± 5 Hz; n= 7/8).

  • 3

    Thus, in the hippocampus, different types of neurones have distinct frequency preferences. Therefore, in the CA1 layer of the hippocampal network, a compound oscillatory input may be segregated into distinct frequency components which are processed locally by distinct types of neurones.

Coherent network oscillations at several distinct frequencies occur during various behavioural states in animals (Buzsáki et al. 1983; Singer, 1993; Steriade et al. 1993), including humans (Berger, 1929; Kahana et al. 1999). This has led to the suggestion that these oscillations play a role in these behaviours. However, the function of this oscillatory activity in terms of neuronal signal processing remains unknown.

In the hippocampus, the CA3 subfield can intrinsically generate population oscillations at theta (4-7 Hz) (MacVicar & Tse, 1989; Cobb et al. 1999; Fellous & Sejnowski, 2000) and gamma frequencies (30-100 Hz) (Bragin et al. 1995; Fisahn et al. 1998) and such activity can propagate to the CA1 (Fisahn et al. 1998). Although pyramidal cells and some subpopulations of interneurones show intrinsic oscillations at theta frequencies (Leung & Yim, 1991; Cobb et al. 1995; Chapman & Lacaille, 1999) and other types of interneurones are known to play an important role in gamma oscillations (Whittington et al. 1995; Traub et al. 1996; Fisahn et al. 1998), the way in which network oscillations affect the activity of individual neurones in the network is unknown.

Extracellular recording of such network activity reveals oscillations which appear sinusoidal (Leung & Yim, 1986) and occur at several distinct frequencies superposed on each other (Buzsáki et al. 1983; Bragin et al. 1995). The coherent network activity, controlled by, among others, the medial septal input, contributes to the synaptic inputs on individual hippocampal neurones, which is seen collectively during intracellular recordings in vivo as sinusoidal oscillations at several distinct frequencies (Soltesz & Deschenes, 1993; Fig. 1A). An understanding of how different types of neurones within a network respond to oscillatory input patterns may provide important clues as to the roles that different cell types play in the information transfer through the network. Neurones transfer information by transforming input signals into trains of discrete action potentials. We therefore investigated the effect of sinusoidal inputs at various distinct frequencies on action potential generation in different neuronal types, and specifically asked whether different neurones exhibit any frequency preference in their response to sinusoidal input current.

Details are in the caption following the image

Frequency preference of signal transfer in distinct types of hippocampal neurones

A, simplified diagram of the hippocampal CA1 circuitry analysed with one pyramidal cell and two distinct types of local-circuit interneurones illustrated. The pyramidal cell (P) and one of the interneurones (I) receive an oscillatory input from CA3 via the Schaffer collaterals (left). The input (top trace) is the superposition of two oscillations at 30 Hz (middle) and 5 Hz (lower trace) as illustrated. The horizontal cell (H) is a distinct class of interneurone which does not receive this input, but is activated by recurrent excitatory collaterals. B and C, small-amplitude oscillatory input currents (20 pA at 5 Hz (B) and 30 Hz (C)) were applied through the recording pipette (left). The voltage responses in an interneurone with soma in the pyramidal cell layer (middle left), a pyramidal cell (middle right), and an interneurone with cell body in the stratum oriens-alveus (far right) were recorded at membrane potentials just around threshold. Note in the pyramidal cell the generation of action potentials with an input oscillation at 5 Hz, with only small voltage fluctuations observed at 30 Hz. Conversely, in the fast-spiking interneurone, note the relatively small voltage fluctuations with an input of 5 Hz, and the larger voltage responses with clusters of spiking activity at 30 Hz. In the horizontal interneurone, action potentials were fired with an input oscillation at 5 Hz, while only modest voltage fluctuations were observed at 30 Hz. D, a small-amplitude oscillatory input current at stepwise decrementing frequencies (200 to 0.5 Hz; left), with corresponding voltage responses in a fast-spiking interneurone (middle left), pyramidal cell (middle right) and horizontal cell (far right). The fast-spiking interneurone generated action potentials at input frequencies between 20 and 40 Hz, whereas the pyramidal cell and horizontal interneurone fired at frequencies smaller than 5 and 7 Hz, respectively. Action potentials are truncated.

Frequency preference of action potential generation could be due to active membrane properties. An enhanced voltage response in a neuronal membrane to a narrow bandwidth of input frequencies is termed resonance (Hutcheon & Yarom, 2000). Neocortical neurones show resonant behaviour (Gutfreund et al. 1995; Hutcheon et al. 1996), and frequency preference has also been shown to exist at subthreshold membrane potentials in hippocampal pyramidal neurones (Leung & Yu, 1998). However, the frequency preferences of different types of hippocampal neurones at threshold have not been reported.

The aim of this study was to investigate how different types of neurones within the hippocampal CA1 network respond to oscillatory input patterns, by analysing their action potential discharge in response to intracellular sinusoidal current. The possible mechanisms underlying action potential transfer properties were studied by investigating the resonance properties of the neurones at membrane potentials close to the firing threshold.

METHODS

Slice preparation

Horizontal hippocampal slices (300 μm) were prepared from young Wistar rats (postnatal day (P)13-P20) of both sexes after decapitation under deep isoflurane-induced anaesthesia, in accordance with British Home Office regulations. Slices were maintained at room temperature in a submerged-style holding chamber until transferred one by one to the recording chamber and superfused with artificial cerebrospinal fluid (ACSF) containing (mm): NaCl 126; KCl 3; NaH2PO4 1.25; MgSO4 2; CaCl2 2; NaHCO3 24; glucose 10; pH 7.2-7.4; bubbled with carbogen gas (95% O2-5% CO2).

Recording conditions

Patch-pipette recordings of CA1 neurones were made under visual guidance by infrared differential interference contrast video microscopy (Sakmann & Stuart, 1995). Patch pipettes were pulled from standard-walled borosilicate tubing. The electrode solution contained (mm): potassium gluconate 110; Hepes 40; NaCl 4; ATP-Mg 4; GTP 0.3; with 5 mg ml−1 biocytin (pH 7.2-7.3; osmolarity 280-300 mosmol l−1). Biocytin was included to enable later processing and structural analysis but this was not used in the present study. All chemicals and drugs were purchased from BDH, Sigma and Tocris (TTX). Whole-cell current-clamp recordings were made with an Axoclamp-2B amplifier in bridge mode. Capacitance compensation was maximal and bridge balance adjusted (15-50 MΩ) during recording. Cells were identified by their location, shape and orientation as seen by video microscopy, and by their characteristic responses to square current pulses. All recordings were made at room temperature (20-25°C).

Oscillatory current protocols

Three different oscillatory current protocols were used: (1) 5 s long sinusoidal currents at fixed frequencies (0.5, 1, 2, 3, 5, 7, 10, 15, 20, 30, 50, 100, 200 Hz); (2) a 15 s long sinusoidal current at varying discrete frequencies (200, 150, 120, 100, 70, 50, 40, 30, 20, 15, 12, 10, 7, 5, 4, 3, 2, 1.5, 1 and 0.5 Hz), each for four cycles, except for the lower frequencies where two (2 and 1.5 Hz) or one cycle (1 and 0.5 Hz) was used; (3) a 20 s long sinusoidal current at linearly varying frequency from 0 to 30 or 100 Hz and from 30 or 100 Hz to 0 Hz (a ZAP function; Puil et al. 1986). Protocol 1 was used to determine both spike transfer properties and the reported values for subthreshold resonance. However, subthreshold properties were not investigated in all cells. To record responses at different membrane potentials, or to record responses at threshold, DC current was injected, in 10 pA steps. Data in 1, 3 were obtained with protocol 1. Protocol 2 was used to measure the frequency preferences for action potential firing; these data are reported in Fig. 1D. Protocol 3 was used only at subthreshold membrane potentials and to confirm the results found using protocol 1 (e.g. data reported in Fig 2). All protocols were written using Igor Pro software (Wavemetrics, Lake Oswega, OR, USA).

Details are in the caption following the image

Distinct action potential firing preferences and resonance properties of the three different types of hippocampal neurones

A, spike frequency plotted against input frequency for a representative fast-spiking interneurone (i), pyramidal cell (ii) and horizontal interneurone (iii). The spike rate for the fast-spiking interneurone shows a peak at an input frequency between 30 and 50 Hz, the pyramidal cell peaks at an input frequency of 5 Hz, while the horizontal interneurone peaks at an input frequency of 5 Hz. Aiv, plot of data from all cells (open symbols), showing that all pyramidal cells (triangles) have a preferred maximum firing frequency at input frequencies between 3 and 10 Hz, while most fast-spiking interneurones (left circles) fired maximally at input frequencies between 30 and 50 Hz, with one cell firing maximally at 10Hz. One fast-spiking interneurone did not fire at all in response to oscillatory inputs. Horizontal interneurones (right circles) fired maximally between 3 and 10 Hz. Filled symbols with error bars, mean ±s.d.B, impedance magnitude plotted against input frequency for typical cells compared to the theoretical impedance magnitude-frequency curve for RC equivalent circuits with the same time constant and input resistance (dotted line). The impedance estimates were made during periods devoid of action potentials. In the fast-spiking interneurone (i), note the resonance peak at an input frequency of 30 Hz, and the suppression at input frequencies below 10 Hz. In the pyramidal cell (ii), note the resonance peak at an input frequency of 3 Hz. The horizontal interneurone (iii) showed a resonance peak at input frequencies in the theta range. Biv, plot of data from all cells (open symbols), showing that all pyramidal cells (triangles) show resonance at input frequencies between 2 and 7 Hz, while fast-spiking interneurones (left circles) show resonance at input frequencies between 5 and 50 Hz and horizontal interneurones (right circles) show resonance between 2 and 7 Hz. Filled symbols with error bars, mean ±s.d. Note the different frequency axis in traces i vs. traces ii and iii.

Details are in the caption following the image

Calculation of impedance magnitude

A, example of a ZAP function (protocol 3), with linearly increasing frequency from 0 to 30 Hz over a period of 10 s, which was applied through the recording pipette. The current amplitude was 60 pA peak-to-peak. Simultaneously with AC current injection, DC current injection of -60 pA was used to ensure that the voltage responses were subthreshold. B, subthreshold voltage response of a pyramidal cell to the ZAP input current shown in A. C, FFT magnitude of the current input (FFT(I)). D, the FFT magnitude of the voltage response (FFT(V)). E, impedance magnitude as a function of frequency for the same cell. Data in C and D were used to calculate FFT(V)/FFT(I) to yield the impedance magnitude profile for a representative pyramidal cell. Note that the peak in the impedance magnitude profile is at 3 Hz, consistent with that seen for this pyramidal cell using protocol 1.

Data acquisition and analysis

Data were acquired on-line and analysed with custom-made procedures using Igor Pro. Membrane time constants were obtained by fitting a single exponential function to the initial part of an averaged voltage response to small hyperpolarising current pulses. Input resistances were derived from Ohm's law by dividing the maximal averaged voltage deflection to small hyperpolarising current pulses by the applied amplitude of the current pulse. Spike duration was measured at half-peak amplitude.

In order to estimate the impedance magnitude the current (I) and voltage (V) recordings were converted from the time domain to the frequency domain using fast Fourier transforms (FFTs) to enable analysis of the power of I and V at different frequencies (Fig. 2). A minimum of two periods were analysed for calculation of FFTs. Impedance (Z) was calculated as the ratio of the FFTs:
urn:x-wiley:00223751:media:TJP205:tjp205-math-0001u

The magnitude of the complex-valued impedance was plotted against frequency to give an impedance-magnitude (IM) profile. Resonance is expressed as a frequency-specific enhancement of the voltage responses (Fig 2). A membrane without any frequency-specific enhancement of voltage responses, i.e. without resonance, would result in a smooth IM curve, with greatest impedance at the lowest input frequency and monotonously decreasing with increasing frequency.

In order to compare the subthreshold voltage response of individual neurones to sinusoidal inputs to that expected from a passive membrane circuit, a theoretical IM profile was calculated for an RC equivalent circuit, using the same input resistance and time constant as those of the recorded neurone. In such a case the impedance magnitude |Z| as a function of frequency (f) is given by:
urn:x-wiley:00223751:media:TJP205:tjp205-math-0002u
where Ri= input resistance and τ= membrane time constant (Gutfreund et al. 1995).

All data in the text are presented as means ±s.e.m. Student's t test was used for statistical analysis.

RESULTS

Basic electrophysiological properties of different neuronal types

The different types of neurones showed distinct electrophysiological properties at resting membrane potentials. In particular, the time constants were significantly different between pyramidal cells and interneurones (P < 0.05), and also between the two types of interneurones investigated (P < 0.05). Input resistances were significantly different between pyramidal cells and interneurones (P < 0.05). Details of the electrophysiological properties (means ±s.e.m.) measured from the three different cell types are given in Table 1.

Table 1. Electrophysiological properties of different types of hippocampal neurones
Pyramidal cell Fast-spiking interneurone Horizontal interneurone
No. of cells (n) 30 12 10
RMP (mV) −62 ± 1 −58 ± 1 −65 ± 1
Input resistance at RMP (MΩ) 113 ± 9 148 ± 34 230 ± 28
Time constant at RMP (ms) 22 ± 2 9 ± 1 17 ± 1
Spike width (ms) 1.1 ± 0.1 0.7 ± 0.1 1.0 ± 0.1
  • RMP, resting membrane potential. Values are means ± S.E.M.

Action potential transfer properties

In order to investigate how single neurones process oscillatory inputs, we first recorded the responses to sinusoidal input currents at two physiologically relevant frequencies (5 and 30 Hz) in the three different cell types (Fig. 1). To enable direct comparison between the neurones all recordings were made at membrane potentials close to the threshold for action potential generation. The action potential transfer properties were studied by superposing an oscillatory current on the maximum DC current which itself did not elicit action potentials.

Injection of 20 or 40 pA peak-to-peak sinusoidal current at 5 Hz superposed on this maximum DC current into pyramidal cells resulted in reliable spiking activity (n= 30, Fig. 1B). In contrast, pyramidal cells showed small voltage deflections and did not fire action potentials in response to sinusoidal current injection at 30 Hz with the same peak-to-peak amplitude (n= 30), and even with increased depolarisation, action potentials occurred infrequently and irregularly. In response to an oscillatory input at a wider range of frequencies (0.5-200 Hz, Fig. 1D), pyramidal neurones fired primarily at theta frequencies (range 2-7 Hz; n= 18; Fig. 1D). Estimated using protocol 1, the maximum spiking activity occurred at an input frequency of 4.5 ± 0.4 Hz (n= 18; Fig. 3Aii).

In response to a sinusoidal current at 30 Hz, the interneurones with cell bodies in the pyramidal cell layer showed large voltage deflections and clusters of action potentials at 30 Hz (n= 10/12; Fig. 1C). Injection of 5 Hz frequency current into interneurones whose cell bodies lay in the pyramidal cell layer resulted in action potential firing in only two cells. The majority of cells showed smaller voltage deflections but no action potential firing (n= 10/12). Even with increased depolarisation with current injection up to 80 pA, few action potentials were elicited in the interneurones in response to a 5 Hz sinusoidal input, and the cells did not fire regularly. Given an oscillatory input at a wider range of frequencies (0.5-200 Hz, Fig. 1D), 10 out of 12 interneurones with cell bodies in the pyramidal cell layer fired preferentially at input frequencies in the beta-gamma band (range 15-100 Hz). In these cells the maximum spiking activity occurred at an input frequency of 42 ± 3 Hz (n= 10; Fig. 3Ai).

Finally, horizontal cells, with cell bodies in the stratum oriens-alveus, showed enhanced firing in response to low frequencies, similar to pyramidal neurones (n= 10). In response to an oscillatory input at a wider range of frequencies (0.5-200 Hz), these neurones fired primarily at or close to theta frequencies (range 1.5-10 Hz; Fig. 1D) with the maximum spiking activity occurring at an input frequency of 3.9 ± 1.0 Hz (n= 10; Fig. 3Aiii). The results suggest that different frequency components of an input signal could be transferred by different types of neurones in the network.

Resonance properties

Frequency preference of action potential generation in neurones could emerge if the collection of their active conductances produces a frequency-selective enhancement of the voltage response to oscillatory inputs (Fig. 3A). This phenomenon, termed resonance, has been demonstrated in several types of neurones subthreshold for generation of action potentials (e.g. thalamic neurones: Puil et al. 1994; cortical neurones: Gutfreund et al. 1995; Hutcheon et al. 1996; hippocampal pyramidal cells: Leung & Yu, 1998; and Fig. 2). In order to investigate whether subthreshold resonance could account for the observed differences in spike frequency preference (Fig. 3A), the subthreshold voltage response of individual neurones to sinusoidal inputs at several different frequencies was compared with that predicted from an RC equivalent circuit (see Methods) using the same input resistance and time constant as those of the recorded neurone (Fig. 3B).

In pyramidal neurones, the equivalent circuit closely predicted the experimental data for all frequencies except theta frequencies, for which the experimental voltage response was enhanced, as previously reported (Leung & Yu, 1998). The maximal enhancement was seen in the range of 2-7 Hz (mean = 4.1 ± 0.4 Hz; n= 19; Fig. 3Bii) and it was 50 ± 6 % above the values predicted by the passive model.

In contrast, only one of the eight fast-spiking interneurones with cell bodies in the pyramidal cell layer showed resonance in the theta frequency range. The majority of these neurones showed resonance at beta-gamma frequencies, with maximal enhancement seen in the range 10-50 Hz (mean = 26 ± 5 Hz; n= 7/8; Fig. 3Bi), and the enhancement averaged 200 ± 40% of that predicted from a passive model. Moreover, in these interneurones, not only was the voltage response at beta-gamma frequencies enhanced, but in seven out of eight interneurones, the voltage response at theta frequencies was suppressed compared to a passive membrane model (Fig. 3Bi). The suppression was maximal at 2-10 Hz (mean 4.1 ± 1.2 Hz; n= 7/8), and the suppression was 65 ± 8% of the predicted response.

Horizontal interneurones showed resonance at low frequencies. The maximal enhancement was seen in the range 1-5 Hz (mean = 2.4 ± 0.5 Hz; n= 7; Fig. 3Biii). Thus, distinct types of neurones could process different frequencies of an input signal by selectively enhancing their voltage response to certain frequencies whilst, in the case of some neurones, suppressing others.

Effect of tetrodotoxin

Thus, both pyramidal neurones and interneurones show resonance at membrane potentials close to the threshold for generation of action potentials. In order to investigate the resonance properties in the absence of action potentials, the experiments were repeated in the presence of the Na+ channel blocker tetrodotoxin (TTX; 1 μm). Under these conditions, the resonance at gamma frequencies was completely blocked in fast-spiking interneurones (no resonance peak observed; n= 4; Fig. 4B). However, in pyramidal cells the application of 1 μm TTX had little effect on the frequency of the observed resonance peak (n= 6; Fig. 4D), suggesting that other voltage-gated ion channels are important for resonance in these cells.

Details are in the caption following the image

The effect of tetrodotoxin on resonance

A, control plot for an interneurone. Impedance magnitude plotted against input frequency for a typical interneurone with soma in the pyramidal cell layer compared to a theoretical impedance-frequency curve for a passive cell with the same time constant and input resistance (dotted line). The impedance estimates were made during periods devoid of action potentials. Note the resonance peak at an input frequency of 30 Hz, and the suppression at input frequencies below 20 Hz. B, results from the same cell after addition of TTX. The impedance magnitude closely follows the theoretical impedance-frequency curve for an RC equivalent circuit with the same time constant and input resistance (dotted line), demonstrating that addition of TTX blocked both the resonant peak at 30 Hz and the suppression below 20 Hz. C, control plot for a pyramidal cell. Impedance magnitude plotted against input frequency for a representative pyramidal cell compared to a theoretical impedance-frequency curve for an RC equivalent circuit with the same time constant and input resistance (dotted line). The impedance estimates were made during periods without the occurrence of action potentials. In this cell, note the resonance peak at an input frequency of 3 Hz. D, result from the same cell after addition of TTX. The peak at 3Hz remains in the impedance-frequency curve, showing that TTX does not block the subthreshold resonance in pyramidal cells.

DISCUSSION

These results demonstrate that distinct classes of neurones within the hippocampal network show distinct input frequency preference for action potential generation. Pyramidal neurones show a firing preference at theta frequencies (2-7 Hz) as do horizontal interneurones, while another subpopulation of interneurones (fast-spiking interneurones) shows firing preference for inputs at beta-gamma frequencies (30-50 Hz). Thus an input signal containing different frequency components might be transferred by different types of neurones in the network, with pyramidal cells and horizontal interneurones responding preferentially to theta frequencies, while fast-spiking interneurones respond to inputs at beta-gamma frequencies.

Hippocampal pyramidal neurones and both types of interneurone show enhanced responses to a bandwidth of input frequencies. In all three types of cell, this frequency preference was expressed as an increased rate of firing for a relatively narrow band of input frequencies. All three types of neurones showed resonance at corresponding frequency ranges just subthreshold for the generation of action potentials (pyramidal cells between 2 and 7 Hz, horizontal interneurones between 1 and 5 Hz and fast-spiking interneurones between 10 and 50 Hz), suggesting that the resonance frequency of these neurones underlies their firing frequency preference. The firing frequency preference was at a slightly higher frequency than the resonance frequency. This might arise from an increase in resonance frequency with increasing depolarisation, or might be due to an effect of the spike firing per se. Resonance has been demonstrated in several types of neurones subthreshold for the generation of action potentials (e.g. trigeminal root ganglion neurones: Puil et al. 1986; thalamic neurones: Puil et al. 1994; and neocortical neurones: Llinás et al. 1991; Gutfreund et al. 1995; Hutcheon et al. 1996). The estimated resonance frequency for pyramidal cells in our study (≈4 Hz) is within the range reported by Leung & Yu (1998) for hippocampal pyramidal neurones (3-10 Hz).

There are several subpopulations of interneurones in the hippocampus with cell bodies in the pyramidal cell layer, including basket cells, axo-axonic cells, and bistratified cells (Buhl et al. 1994). The majority of these interneurones target pyramidal neurones close to the cell body (basket and axo-axonic interneurones; Buhl et al. 1994). It is likely that the fast-spiking interneurones recorded in this study from the pyramidal cell layer are members of these classes. The stratum oriens-alveus horizontal interneurones, many of which project to the stratum lacunosum-moleculare (Freund & Buzsáki, 1996), do not receive direct excitatory input from the CA3, but are exclusively activated by CA1 pyramidal neurones (Fig. 1A; Blasco-Ibá-ez & Freund, 1995). Furthermore, they are the only CA1 interneurones reported to show facilitation at their excitatory input synapses (Ali & Thompson, 1998). These interneurones appear to have different frequency-selective properties from the fast-spiking interneurones recorded in this study, suggesting that the frequency preference of distinct classes of interneurones may be related to their location in the network and possibly to their different processing functions within that network. The range of resonance frequencies recorded in the stratum oriens-alveus interneurones in this study (1-5 Hz) is similar to the frequency of subthreshold oscillations in interneurones situated in stratum lacunosum-moleculare (2-5 Hz, Chapman & Lacaille, 1999). These two classes of interneurone mostly target pyramidal cell dendrites. Thus it might be that dendritic-targeting interneurones support 5 Hz oscillatory activity, whereas fast oscillations might be mediated via interneurones targeting the soma and perisomatic regions of pyramidal cells.

Previous studies have led to various ionic mechanisms being suggested as possible candidates underlying resonance. Currents proposed include calcium currents (IT) in the thalamus (Puil et al. 1994), and a persistent sodium current (INaP), a potassium current (IIR) and a non-specific cation current (IH) in the neocortex (Hutcheon et al. 1996). In our study, TTX completely abolished the resonance at gamma frequencies in interneurones, suggesting that fast voltage-gated Na+ channels are involved in the resonance at gamma frequencies in these interneurones. However, it seems from our results that other voltage-gated ion channels are important for the resonance at theta frequencies in pyramidal cells. This is in contrast to sustained spontaneous oscillations in pyramidal cells which have been shown to depend on depolarising drive from TTX-sensitive sodium channels (Leung & Yim, 1991). Further investigation is necessary to identify the currents underlying resonance in hippocampal pyramidal cells. Both pharmacological investigations and modelling studies will be important here.

The findings from this series of experiments indicate that individual neurones within a network can act as bandpass filters that selectively transmit information that arrives within a specific frequency range. Thus, transmission of frequency-dependent information through a layer of a network may be modulated by controlling which subsets of neurones are in a resonant state. It has been suggested that networks of interneurones can generate coherent gamma frequency oscillations independent of pyramidal cells (Whittington et al. 1995). Since the voltage response to oscillatory inputs in the theta frequency range in most fast-spiking interneurones was actually suppressed, pyramidal cells might process information encoded on theta frequency oscillations without extraneous activity from these interneurones. However, whereas the information carried on the theta frequency oscillations can be transferred directly by the pyramidal neurones through the network, the local-circuit interneurones must somehow mediate their effect through pyramidal neurones if the information is going to reach the next layer of the network. Perhaps the information carried by theta oscillations is encoded through the rate of firing of those pyramidal neurones which are active, while the information in the faster oscillations is encoded through fast-spiking interneurones determining when these pyramidal neurones fire within each theta cycle. Thus, these results offer a mechanism for how information could be encoded in 40 Hz oscillatory subcycles within the main theta cycles, as has been suggested (Lisman & Idiart, 1995). This idea is compatible with anatomical observations that the majority of interneurones with cell bodies in the pyramidal cell layer target pyramidal neurones at the soma or in the perisomatic region (basket and axo-axonic interneurones; Buhl et al. 1994).

More generally, since these results suggest that neuronal networks could process complex signals with several layers of information encoded within them through frequency multiplexing, examining the resonance properties of different types of neurones within that network might indicate what role these neurones subserve in the information transfer of the network. As these resonance properties could be tuned to different frequencies or even selectively enabled or disabled, they suggest possible mechanisms for how external inputs can influence the information processing within the network. Finally, these findings indicate that characterising cortical neurones with physiologically plausible inputs can reveal fundamental characteristics of different neuronal classes sometimes hidden using the conventional square-current-pulse protocols.

Acknowledgements

We are grateful to Professor A. David Smith and Professor Peter Somogyi for support and useful comments on a previous version of the manuscript. We thank Dr John Bekkers for introducing us to programming in Igor Pro software and Mrs Neidja Gould for technical assistance. This research was supported by a Wellcome Trust project grant. Additional financial support from The Royal Society and the Medical Research Council is gratefully acknowledged. F.G.P. and J.M.S. are MRC Research Students in the Department of Pharmacology, Oxford, P.G. is a Wellcome Trust Research Student, N.K. holds a Rhodes Scholarship, and O.P. was also the Christopher Welch Junior Research Fellow in Biological Sciences at Wadham College, Oxford.