Hippocampal Feedforward Inhibition Focuses Excitatory Synaptic Signals into Distinct Dendritic Compartments

Feedforward inhibition controls the time window for synaptic integration and ensures temporal precision in cortical circuits. There is little information whether feedforward inhibition affects neurons uniformly, or whether it contributes to computational refinement within the dendritic tree. Here we demonstrate that feedforward inhibition crucially shapes the integration of synaptic signals in pyramidal cell dendrites. Using voltage-sensitive dye imaging we studied the transmembrane voltage patterns in CA1 pyramidal neurons after Schaffer collateral stimulation in acute brain slices from mice. We observed a high degree of variability in the excitation-inhibition ratio between different branches of the dendritic tree. Many dendritic segments showed no depolarizing signal at all, especially the basal dendrites that received predominantly inhibitory signals. Application of the GABAA receptor antagonist bicuculline resulted in the spread of depolarizing signals throughout the dendritic tree. Tetanic stimulation of Schaffer collateral inputs induced significant alterations in the patterns of excitation/inhibition, indicating that they are modified by synaptic plasticity. In summary, we show that feedforward inhibition restricts the occurrence of depolarizing signals within the dendritic tree of CA1 pyramidal neurons and thus refines signal integration spatially.


Introduction
Under physiological conditions principal cells in the central nervous system receive biphasic innervation patterns composed of excitatory postsynaptic potentials (EPSPs) from direct excitatory inputs and inhibitory postsynaptic potentials (IPSPs) from interneurons, activated by collaterals of the excitatory input [1,2]. The interval between the excitatory and inhibitory phases in such feedforward inhibition is short, in the order of a few milliseconds. This tight temporal coupling results in a significantly faster repolarisation than passive membrane properties would produce. In the hippocampus, Pouille and Scanziani [3] have shown that feedforward inhibition thus controls the time window for synaptic integration in pyramidal neurons, ensuring temporal precision in the hippocampal CA1 area.
While the temporal characteristics and impact of feedforward inhibition are well understood, we know much less about its spatial properties. Synaptic integration takes place in the geometrically complex dendritic tree, which receives synaptic contacts over its entire surface. Synapses are not randomly distributed, but show remarkable specificity in their localisation. Both excitatory and in particular inhibitory synapses show input-specific preference for particular compartments. In the hippocampus, excitatory inputs arrive in a layer-specific organization with entorhinal input arriving at the most distal dendritic locations in stratum lacunosum moleculare and Schaffer collaterals (SCs) from the CA3 area arriving more proximally [4]. SC synapses themselves also show spatial organization, with inputs originating close to the hilus terminating in the apical stratum radiatum and inputs from cells close to CA1 forming synapses predominantly on the proximal apical dendrite in the stratum radiatum and onto basal dendrites in the stratum pyramidale/oriens [5]. Even more specific is the spatial distribution of synapses from different interneuron types, which has been used for their classification [1,6]. The relative location of excitatory and inhibitory synapses will significantly affect how these signals will interact [7].
We have recently shown that small inhibitory synaptic signals can be studied using voltage sensitive dye (VSD) imaging at sufficient resolution to allow the study of their dendritic integration within a single CA1 pyramidal neuron undisturbed by whole-cell recordings [8].
Here we study the effect of feedforward inhibition on dendritic signal integration using VSD imaging. Dye-loaded cells were imaged after SC stimulation and the voltage transients in different dendritic subcompartments were then analysed.
We demonstrate a critical influence of γ-aminobutyric acid receptor-(GABA A R-) mediated inhibition on the subcellular membrane potential patterns in CA1 pyramidal neurons.

Brain slice preparation
All experiments were approved by Basel cantonal veterinary authorities. Recordings were performed in 300 µm thick brain slices of heterozygous knock-in mice P21 to P32 expressing GFP from the GAD67 gene locus [9]. After deep isoflurane anaesthesia mice were decapitated and transversal hippocampal slices were cut using a vibrating microtome (VT1200S, Leica, Switzerland). Slicing was performed in icecold solution containing (in mM) NaCl 87, Sucrose 75, Glucose 25, NaHCO 3
While staining cells with VSD, somatic recordings were performed using a Multiclamp 700A amplifier (Axon Instruments, Germany) and an upright microscope (Olympus BX51-WI, Olympus, Switzerland Background fluorescence increases due to dye spillage was avoided by tip-filling the electrode with dye-free solution. In addition, before reaching the cells, positive pressure in the pipettes was kept low in the bath and controlled with a manometer at ~ 5 mbar (Model 840081; Sper Scientific, Scottsdale, AZ). Staining time was determined by measuring the resting fluorescence from the cell body at reduced excitation light intensity (~0.1 % of the laser light). Cell loading with VSD did not cause pharmacological effects [8,11]. After enough dye had diffused into the cell (30 min), pipettes were gently removed by forming outside-out patches. Optical recordings were performed when dendrites were sufficiently filled with VSD (~ 20-30 min after electrode removal).

Electrophysiology
If relevant for the experiments, electrical signals were detected at the beginning of the experiments simultaneous to optical signal detection.
For paired recordings, potentially connected interneurons of CA1 stratum radiatum were searched, while pyramidal cells were being filled with the dye. To improve the detection of small unitary inhibitory postsynaptic potentials (uIPSPs), pyramidal cells were patched with intracellular solution containing 40 mM Cl -. Hence, the driving force for chloride increased, potentials were larger in amplitude and evoked potentials were positive due to a more depolarized reversal potential for Cl -.
Somatic electrode recordings were acquired at 16 kHz and filtered at 4 kHz by using the Redshirt imaging system or acquired at 20 kHz and filtered at 2 kHz by a separate A/D board (NI USB-6343, National Instruments, Switzerland).

Optical recordings
Excitation of the VSD was achieved using a 532 nm-300 mW solid-state laser (model MLL532; CNI, China). Hence, the dye was stimulated at the border of its absorption spectrum and the largest dynamic range in fluorescence could be reached.
Advantages of laser illumination compared to a conventional xenon arc lamp were described earlier [8].
GABA A -receptor-mediated uIPSPs and evoked EPSP/IPSP patterns were detected optically at a frame rate of 500 Hz and 6 % of the full laser intensity. Optical signals were captured with a high-speed, 80 x 80 pixels CCD camera (NeuroCCD-SM, RedShirtImaging LLC, China). The fluorescence image of the cell was projected via a 0.2 optical coupler onto the CCD camera. The imaged field in our measurements was ~125 μm x 125 μm. The excitation light was directed to the preparation using a 570 nm dichroic mirror and a water immersion objective (Olympus 60x/1.1 NA, Olympus, Switzerland). The emission light was filtered with a 610 nm long-pass filter.
Optical signals were averaged over 10-20 pixels along compartments of the stained cells, at a pixel size of 1.56 µm x 1.56 µm. To improve the signal-to-noise ratio, averages of 3-12 trials were taken.
For analysis, dendritic sub-compartments (20-50 pixels in size) were chosen by comparing VSD fluorescence images with two-photon reconstructions of the neurons. Direct comparison of DF/F signals between different cellular compartments have to take into account the possibly uneven dye distribution. We have therefore used the ratio between deand hyperpolarizing signals as our main measurement -these ratios are not affected by dye inhomogeneity.

Stimulation and Pharmacology
Extracellular stimulation was performed by using borosilicate patch pipettes filled with ACSF. Hydraulic manipulators (Narishige) were used to place pipettes. Feedforward inhibition was evoked by stimulation of Schaffer collaterals in stratum radiatum between CA3 and CA1. They innervate different types of interneurons located in stratum pyramidale and radiatum [6].
Stimulation pulses were 0.1 ms of duration and strength varied between 20 to 60 μA. Pulses were delivered by an IS4 stimulator (SC-Devices, Switzerland) and triggered by stimulation protocols written in IGOR Pro software (Wave Metrics, USA).
GABA A R-mediated potentials were tested by bath application of the competitive receptor antagonist bicuculline (20 μM).

Anatomical reconstruction and analysis
Anatomical reconstruction of neurons was carried out from a stack of two-photon excitation fluorescence images obtained using a tunable, modelocked titan-sapphire laser (MaiTai HP, Spectra Physics) set to 880 nm and a laser scanning system (FV300, Olympus) with a high-aperture 20x water-immersion lens (Olympus LUMPLAN 20x).
Optical signals were analyzed as fractional changes of fluorescence (ΔF/F). Optical and electrophysiological recordings were analyzed with dedicated software written in MATLAB (The MathWorks). Optical signals were corrected for the bleach fraction. Stimulation artefact was cleared for clarify.
Statistics were calculated in Excel (Microsoft Office 2010) and averages are presented as the mean ± standard error of the mean (SEM).

Results
We first tested the suitability of VSD imaging to study feedforward inhibition by comparing classical electrode recordings with VSD imaging.
The staining procedure was described in more detail in an earlier publication [8] and is schematically shown in Figure 1 A.
First, we performed whole-cell patch recordings to load hippocampal CA1 pyramidal cells with the VSD JPW1114. We did not observe a significant change in either the amplitude or the time course of SC stimulation-evoked synaptic signals during dye filling ( Figure S1).
After cells were filled adequately, pipettes were slowly removed from the cell somata. The field of view of the camera allowed the optical recording of membrane potential distributions in several regions of interest covering multiple dendritic branches (rectangles, Figure 1 A).
Feedforward inhibition could be initiated in CA1 pyramidal cells by extracellular stimulation of projecting SC in stratum radiatum of the hippocampus (Figure 1 B). Excitatory potentials initiated by SC synapses projecting to CA1 neurons were followed by a disynaptically-induced hyperpolarisation (Figure 1  C). In a subset of cells, whole-cell patch-clamp recordings were established by somatic re-patching. As shown for five examples in Figure 1 C, both the somatic electrode recording and the average VSD imaging trace show the same de-and hyperpolarizing pattern. The VSD imaging signal over the whole dendritic field and the somatic recordings produced the same EPSP/IPSP ratio and the same slope of EPSP depolarization ( Figure 1D solid symbols, Table S1).The correlation coefficient between the two signals was >0.9 for each of these experiments ( Figure S2), indicating that feedforward inhibition can be faithfully recorded by VSD imaging.
Looking at the voltage transients in different dendritic branches and segments revealed significant inhomogeneity in EPSP/IPSP ratio and slope of EPSP depolarization between the different compartments ( The large variability in the excitation-inhibition ratio suggests an uneven and non-overlapping distribution of activated excitatory and inhibitory synapses. How much are the transients shaped simply by the distribution of activated synapses and how much does the interaction of the opposing signals contribute? To measure the influence of GABA A R-mediated inhibition in shaping dendritic voltage transients, we applied the specific blocker bicuculline (Figure 2), while cells were imaged as described before.
Under control conditions, optical recordings in dendritic subregions (ROI 1-8) along the apical dendrite revealed an inhomogeneous EPSP/IPSP distribution (Figure 2 B, left). Note the complete absence of EPSPs in two side branches (ROI 7 and 8). Bath application of bicuculline (20 μM) completely suppressed IPSPs in all regions imaged (Figure 2 B, right). Moreover, robust depolarizing potentials throughout the dendrite were observed in the absence of GABA A R mediated inhibition -notably in areas in which no EPSP was visible before (ROI 7 and 8). A robust increase in depolarization was observed in all 26 ROIs in four cells imaged ( Figure 2D), which was accompanied by a loss of the hyperpolarizing signal (data not shown). Comparison of the signals from the different ROIs in a given cell showed a clear increase in their correlation after the addition of bicuculline in all four cells ( Figure S3, Table S2). This indicates that GABAergic signals substantially contribute to the variability of the dendritic signals. Animated data is available in the movie (Movie S1). In accordance with previous electrode recordings [3] the excitatory transients also lasted substantially longer. This was particularly visible in average scans from apical and basal dendrites (Figure 2 C).
All synaptic signals evoked by SC stimulation were sensitive to the addition of the AMPA receptor antagonist NBQX (10 μM;   Dendritic propagation of depolarizing signals within the dendritic tree is actively controlled by feedforward inhibition in CA1 pyramidal neurons. To quantify these effects we analysed 182 of ROIs in 21 imaged cells. We subdivided the dendritic tree into broad anatomical categories, apical main trunk (ad, blue), apical side branches (sb, green) and expanded our analysis in particular to basal dendrites (bd, red) ( Figure 3). Imaging dendritic responses to SC stimulation revealed the familiar inhomogeneous de-and hyperpolarizing pattern in ad and sb segments (Figure 3. B, ROI 1-4), while bd segments showed nearly uniform hyperpolarization (Figure 3 Depolarization occurs to a variable degree and only in a subset of dendritic compartments, it is notably absent from many basal dendrites after SC stimulation. Are these patterns static or can they be altered by synaptic plasticity? To study this issue, we applied simple tetanic stimuli in the SC pathway (100 pulses by 100 Hz; Figure 4 B), commonly used to induce synaptic plasticity. Optical recordings in CA1 pyramidal cell dendrites were taken shortly before and after the tetanic stimulus in several ROIs of four different neurons (N=29 ROIs). EPSPs and IPSPs showed stable baselines within a time interval of 150-250 s before tetanisation (p(EPSPs) = 0.175; p(IPSPs ) = 0.347; Figure S4). In The pronounced inhomogeneity in the dendritic excitationinhibition patterns is particularly visible for the excitatory side. What is the reason for the more homogeneous inhibitory component? To address this question we investigated the contribution of single interneurons to the inhibitory response in CA1 pyramidal neurons.
We produced paired recordings between neighbouring GABAergic neurons and dye-filled principal cells (N=7; Figure  4 E). We focused on interneurons located in the stratum radiatum -which have a high probability to receive input from SCs and to contribute to feedforward inhibition [2]. Interneurons were stimulated by eliciting action potentials (black trace, Figure 4 F) through somatic current injection. To enhance the chances of detecting connected pairs we increased the size of GABAergic postsynaptic potentials by dye-loading pyramidal cells with an intracellular solution containing 40 mM chloride. A simultaneous electrical (black, somatic electrode) and optical (grey, average over the apical dendrite) recording of a pair is shown in Figure 4 F (bottom). Retraction of the loading pipette resulted the re-establishment of physiological, hyperpolarizing chloride gradients in the pyramidal cell, as shown previously [8]. We now studied the responses in different apical (ROI 1 and 2) and basal (ROI 3-5) compartments (Figure 4 G) of the target pyramidal cell. We did not detect major inhomogeneities in the dendritic responses in any of the cells imaged.

Discussion
We have revealed a significant degree of inhomogeneity in dendritic membrane potential transients in CA1 pyramidal cells after SC stimulation and have shown this to be largely the product of the interaction of excitatory and inhibitory synapses. While NBQX-sensitive feedforward hyperpolarization was detected in almost all dendritic subregions, many dendritic segments, particularly in basal dendrites showed no discernible depolarizing responses. The small fraction of NBQX resistant inhibition cannot explain the profound influence of feedforward inhibition on dendritic membrane potential dynamics, evidenced by the large effect of bicuculline on dendritic excitation. In the absence of GABA A R mediated feedforward inhibition, excitatory potentials did spread throughout the dendritic arborization, indicating that these patterns are actively shaped by the network of interneurons surrounding CA1 pyramidal cells [2,12]. Active shaping of spatial membrane potential profiles by feedforward inhibition were also observed by Tominaga et al. [13], who found a selective perisomatic inhibition after SC stimulation. As we clearly show, feedforward inhibition is temporally so tightly coupled to excitation that it reaches a substantial fraction of the dendritic tree before excitatory signals spread there. Thus, processes that rely on dendritic depolarization, such as the opening of voltage-gated conductances or NMDA receptor activation, are restricted to specific dendritic regions. Antagonists of GABA A receptors have long been shown to dramatically facilitate long-term potentiation [14]. As we now confirm, they not only affect the overall excitability of the postsynaptic neurons, but profoundly alter synaptic integration -facilitating the interaction between synapses on distant dendritic branches.
The control of excitation in time and space by GABAergic inhibition is crucial in shaping the propagation of information throughout the central nervous system. This is particularly evident in sensory processing. In the spatial domain, GABAergic neurons reduce the activity of nearby pyramidal cells, sharpening and controlling their excitatory profiles. Such activity has been observed in several areas, e.g. the whisker barrel cortex or the visual system [15][16][17]. Blocking GABA A R mediated inhibition broadens the tuning curves of pyramidal neurons in olfactory [18], auditory [19] and visual cortex [20]. In the temporal domain feedforward inhibition in the hippocampus  [3] and cortex [21] was shown to shorten and thus control the integration time window for excitatory synaptic input.
Both excitatory and inhibitory inputs to CA1 pyramidal cells show layer-specific topographic organization consistent with the patterns we have observed. SCs axons from CA3 neurons close to the hilus contact chiefly apical dendrites and their branches, while CA3 pyramidal cells close to CA1 send their axons mostly to basal dendrites [5]. The extracellular stimulation of the classical SC pathway at the border of CA3 to CA1 used in our experiments preferentially activates fibers, which terminate in the apical stratum radiatum [22], consistent with the excitatory focus observed in the apical dendrite and the near absence of excitation in basal dendrites. However, as we show, this selective excitatory innervation results in a focused dendritic depolarization only in the presence of feedforward inhibition.
The precise subtype of the involved interneurons could not be characterized. The strong evidence of compartment-specific innervation by different subtypes of interneurons [6] can explain the selective silencing of dendritic branches.
We thus postulate a dynamic network of excitatory and inhibitory connections innervating pyramidal neurons such that excitatory inputs are surrounded and controlled by inhibitory synapses. For example, basket cells, which innervate the soma, have been shown to participate in feedforward inhibition [1,23] and are likely to control the spread of apical EPSPs to basal dendrites.
VSD imaging is particularly suited to the study of dendritic signals in the extremely thin basal dendrites. To date, their integrative properties remain elusive due to the difficulties in recording from them with classical electrode techniques. Recently, it has been shown in layer V pyramidal neurons that they exhibit strong synaptic scaling and can generate active signals [24,25]. Given their high input resistance they are particularly susceptible to GABAergic shunting, which may control these active properties [26,27].
Paired recordings between individual interneurons and CA1 pyramidal cells showed relatively uniform hyperpolarizing signals in both basal and apical dendrites. The influence of GABA A Rs on dendritic EPSPs can therefore be best explained by their shunting effect, which is significantly more localized than the hyperpolarization they cause [23]. Thus, to understand the subcellular signalling of different interneuron subclasses it is important to design experiments that reveal shunting inhibition. Dendritic membrane potential patterns by themselves are not fixed, but can be altered by stimuli that induce synaptic plasticity. Interestingly, this plasticity is non-uniform in itselfproximal portions of the apical dendrite showed small changes in the EPSP-IPSP ratio, while distal portions showed a more pronounced increase. The exact locations of the synaptic changes will need further investigation.
Intrinsically, synaptic integration in neurons is controlled in time by the membrane time constant and in space by the dendritic length constants. As has been shown previously for the temporal domain [3] and here for the spatial domain, feedforward inhibition dramatically improves on these limits and allows a much more precise control over both when and where excitatory synaptic signals can interact. (TIF) Figure S3. Loss of inhomogeneity by blocking GABA A receptors. Correlation plots for optical signals from different ROIs within one example cell under control (top, black) and bicuculline (bottom, red) conditions. Each plot shows the correlation between the optical signals from two different regions. x-axis components originate from region i, with i being the row number of the plot. y-axis components originate from region j, with j being the column number of the plot. For example, plot in row 2, column 3 shows the signal correlation between regions 2 and 3. Signal amplitude was normalized to [0,1] for all data.