Hippocampal Place Fields Emerge upon Single-Cell Manipulation of Excitability During Behavior

See allHide authors and affiliations

Science  17 Aug 2012:
Vol. 337, Issue 6096, pp. 849-853
DOI: 10.1126/science.1221489


The origin of the spatial receptive fields of hippocampal place cells has not been established. A hippocampal CA1 pyramidal cell receives thousands of synaptic inputs, mostly from other spatially tuned neurons; however, how the postsynaptic neuron’s cellular properties determine the response to these inputs during behavior is unknown. We discovered that, contrary to expectations from basic models of place cells and neuronal integration, a small, spatially uniform depolarization of the spatially untuned somatic membrane potential of a silent cell leads to the sudden and reversible emergence of a spatially tuned subthreshold response and place-field spiking. Such gating of inputs by postsynaptic neuronal excitability reveals a cellular mechanism for receptive field origin and may be critical for the formation of hippocampal memory representations.

The hippocampus plays a crucial role in the formation of long-term memories for facts and events in humans and for spatial learning in rodents (1). When a rodent explores an environment, spatial information is prominently represented in the spiking activity of a substantial, environment-specific subset of hippocampal pyramidal neurons. Each such “place cell” fires action potentials (APs) selectively whenever the animal is in a particular region—called the cell’s place field—within the environment (2), whereas the remaining neurons, called silent cells, fire few or no spikes (3). Similarly, subsets of human hippocampal neurons spike selectively for specific items or episodes over a background of low–firing rate cells (4). The establishment and stability of these stimulus-specific spiking responses are believed to be the neural basis of hippocampal-dependent learning and memory.

What is the origin of the spatially selective firing of place cells? Each pyramidal cell in hippocampal subregion CA1 receives excitatory inputs from other spatially tuned neurons (5, 6). Therefore, models have generally assumed that the critical factor is the environmental location where each input fires, with summation of these inputs (their firing rates times synaptic weights) followed by thresholding leading to various degrees of spatially selective output spiking or silence (711). Indeed, intracellular recordings have revealed that place cells have a region with clearly elevated somatic membrane potential (Vm)—i.e., a “hill”—under their place-field spiking (1214), whereas silent cells have a flat, spatially uniform Vm (14).

However, another critical aspect of neuronal integration is the interaction of synaptic inputs with the membrane properties of the postsynaptic cell. Inputs can be strongly filtered or amplified (15, 16) before determining output spiking, yet such features have not been included in place cell models. With intracellular recording, place cells were, unexpectedly, also found to be more excitable than silent cells—even before the animal encountered the environment (14)—which suggested that cellular properties strongly influence the response to inputs during behavior.

How can one separate the role of cellular properties from inputs in the origin of place cell firing? One approach would be to test whether increasing excitability without altering synaptic inputs can convert a previously spatially untuned silent cell into a place cell. A straightforward way to increase excitability is to depolarize a neuron’s baseline Vm by injecting a constant, i.e., spatially uniform, current into the soma and thus bias it toward spiking. Here, we used the whole-cell recording method to precisely manipulate the somatic Vm in single CA1 pyramidal neurons while rats explored a novel environment, then measured the spatial distribution of subthreshold and spiking responses to see the effect on the integration of inputs.

We recorded 10 silent cells (Fig. 1A) in 10 rats (14, 17, 18) as they moved around an oval track (mean recording duration during behavior was 28 min), in four cases in both directions. Because place cells can have fields in a single direction in such environments, we treated the 14 directions separately. For each direction, animals first explored each location in the maze at least twice with no applied current (Fig. 1, B and C). The flat mean Vm as a function of location (Fig. 1, C to E) (14) suggests that silent cells receive either few inputs or inputs whose overall tuning is spatially uniform. Alternatively, silent cells may receive substantial spatially tuned input, but their lower excitability severely limits or prevents propagation to the soma. It was surprising that the latter was the case. Depolarization of the somatic baseline Vm by constant current injection immediately caused not only the appearance of a spiking place field (Fig. 1, B and C) but also an underlying subthreshold (i.e., excluding APs and calcium spikes) Vm hill (Fig. 1, D and E, red), both stable across laps. Of five directions (from five cells) in which a single current level was applied, a single spiking place field and spatially tuned subthreshold hill emerged in two cases (Fig. 1 and fig. S1): multiple, stable hills in one case and firing covering most of the maze in the other cases. This suggested that the number and size of place fields vary over a narrow voltage range.

Fig. 1

Emergence of a spatially tuned subthreshold response and place field by a small depolarization of the somatic membrane potential (Vm). (A) Morphology of recorded CA1 pyramidal neuron. (B) Animal trajectory (gray), AP locations (red), and AP rate map in “O”-shaped maze (inner wall not shown) during periods when the animal faced in the counterclockwise direction before (bottom) and after (top) injecting constant 83 pA current into soma. (C) Vm (black) and AP rate (color) as a function of linearized location of animal around track in (B) for selected laps. Baseline Vm (left), peak AP rate (right) for each lap. Animal movement direction (arrow). Large gaps in laps reflect changes in animal movement direction during lap or periods when experimenter adjusted current injection value. (D) Vm (black) and mean subthreshold Vm (color). (Inset) Vm versus time corresponding to location marked by black bar. (E) Overlay of mean subthreshold Vm from all laps (thin lines; red: 83 pA, blue: 0 pA). Thick lines: averages for each current level. Colors in (C) to (E) are matched.

Therefore, we asked whether, by applying different current levels across laps (five cells, nine directions), we could always create a single, narrow field, as expected in environments of this size. In four of five cells, narrow fields were produced in at least one direction, for a total of six such fields from five directions (one direction had two fields). For population analyses, we included these six fields plus the two from single-current–level experiments. Examples for which, at an intermediate level of depolarization, a single, narrow field was produced (red) are shown in Fig. 2 and fig. S2. Laps were ordered by baseline Vm because this best predicted resulting subthreshold and spiking activity. With more depolarization, subthreshold and spiking responses broadened (orange). Field emergence depended on baseline Vm, not current injection itself, as illustrated in fig. S3.

Fig. 2

Tuning of subthreshold response and place field as a function of the somatic baseline Vm level. (A) Vm (black) and AP rate (color) for individual laps ordered by baseline Vm (left). Peak AP rate (right). Current injected (rightmost). (B) Mean subthreshold Vm of each lap (thin lines), average for each group of laps (thick lines). Spatially tuned subthreshold response and place field emerge at intermediate (red) baseline Vm levels.

For four fields, we returned to the original baseline Vm (by returning to 0 pA) intermittently. In these cases, the field disappeared (Fig. 2A, laps 7 and 12, and fig. S4). Thus, field emergence was not simply experience-dependent. Also, the presence of the field for a few laps was apparently insufficient to induce plasticity that could then maintain the field without depolarization.

Within the (eventual) region of the created field, the amplitude of the hill (“peak – baseline” in Fig. 3A) as a function of baseline Vm showed a sharp rise above a certain baseline Vm value (Fig. 3B). For each field, we estimated the baseline Vm at the midpoint of the rise (Vm,gate = –56.6 ± 1.0 mV) then aligned the fields with respect to Vm,gate. The pooled result (Fig. 3C) showed a thresholdlike transition from a low to maximal amplitude hill within a ~1- to 3-mV range of baseline Vm values, revealing striking sensitivity of a neuron to inputs around its own “threshold” value. This suggests that nonlinear voltage-dependent mechanisms underlie the generation of the subthreshold place field.

Fig. 3

Sudden emergence of a spatially tuned subthreshold response with increasing somatic baseline Vm. (A) Two laps from Fig. 2A with similar baseline Vm. Peak (Vp) and baseline (Vb) of mean subthreshold Vm. (B) Amplitude (peak – baseline) of subthreshold response as function of baseline Vm for Fig. 2 neuron. Baseline “Vm,gate” (dashed line) giving maximal difference of mean amplitude between groups above and below that level. Laps in (A) (filled circles). (C) Normalized subthreshold response amplitude versus baseline Vm, aligned by Vm,gate and pooled across fields (n = 7, different symbols for each field). Sigmoid fit (red). (D) Examples of subthreshold hills without APs (horizontal bars: eventual field locations). (E) Difference between mean and standard deviation of Vm inside and outside location of eventual field in initial 0 pA laps (n = 8 fields).

We checked several possible sources for the emergence of the hill. First, hills did not require the generation of somatic APs. Laps with baseline Vm within the transition range displayed a clear subthreshold hill without APs (Fig. 3D). Also, the ascending slope of a hill could begin before any spiking in the field in the first lap with current injection (Fig. 1D, inset).

Was there a small, preexisting hill at the eventual location of the field that was amplified by depolarization? No, the mean in-field and out-of-field Vm in the initial 0 pA laps did not differ across the population (difference = 0.03 ± 0.05 mV, P = 0.49) (Fig. 3E) or for individual fields (fig. S5).

Alternatively, the field could emerge from a location with no difference in mean Vm but an elevated input resistance (RN) (due to decreased excitatory and inhibitory input), which would depolarize more than regions with lower RN in response to a constant current. In this case, the initial appearance of the hill could be independent of voltage-gated mechanisms. However, for two of the three fields we could test, there was no evidence of this (P = 0.24, P = 0.47).

We then checked whether the subthreshold response emerged from regions with larger fluctuations in Vm (but no difference in mean Vm). Neither the standard deviation (P = 0.25) (Fig. 3E), nor gamma (25 to 100 Hz) band power (P = 0.63), nor right tail of the distribution (fig. S6) of Vm were initially different inside versus outside the eventual field location. We examined the prominent theta (4 to 10 Hz) band in more detail. Theta power was higher within the field for laps with fields (P = 0.01) (Fig. 4, A and B) (13), but, again, not in the initial 0 pA laps (P = 0.89) (fig. S7 and Fig. 4B). Rather, theta power varied with Vm and not location per se (i.e., the power-Vm relation did not differ inside and outside the field) (Fig. 4C and fig. S8).

Fig. 4

Theta frequency power depends on mean somatic Vm. (A) Vm (top), spectrogram (middle), and instantaneous power in 4 to 10 Hz band (bottom) versus time for Fig. 2A lap 5. Inside of place field (box). (Above) Expanded Vm dynamics inside and outside the field. (B) Ratios of mean theta power inside versus outside location of (eventual) field for initial 0 pA and place field laps (n = 8 fields). (C) Theta power as a function of mean Vm inside (filled) and outside (open) (eventual) field for Fig. 2 neuron. Linear regression for open circles (black line).

Therefore, there was no obvious indication of where the field would eventually emerge. When it did emerge, the subthreshold response could remain elevated for an extended period (4.8 ± 0.5 s) while the animal remained in the field (Fig. 1D, inset, and Fig. 4A).

In two recordings of place, as opposed to silent cells, we suppressed spiking with hyperpolarizing current. In one case, the original subthreshold hill disappeared (fig. S9), which provides further evidence that place fields may not originate from passive summation of synaptic input in novel environments.

Although any spatial information in the hippocampus must ultimately come from external inputs (5, 1921), these findings directly demonstrate that nonspatial, cellular factors can play a decisive role in how neurons respond to inputs during behavior. In particular, most CA1 pyramidal cells receive spatially tuned synaptic input that, in silent cells, does not propagate to the soma to yield place-field activity. However, a spatially uniform signal that slightly depolarizes the soma can reveal this input. This gating of inputs, as opposed to just outputs, by somatic Vm constitutes a novel mechanism for receptive fields. It also implies that integration of multiple inputs is gated by somatic Vm (fig. S10).

These results are the opposite of those expected from standard models of cortical receptive fields (22, 23) and place cells (711). In these models, a neuron receives more excitatory synaptic input in response to the preferred stimulus, and this input is passively summed then compared with the AP threshold. Silent cells would have spatially tuned somatic Vm hills of various amplitudes that do not reach threshold, and somatic depolarization would reduce the driving force and, thus, hill amplitude (23).

At the other extreme is a model in which synaptic inputs are uniformly distributed in stimulus space, and the neuron selectively amplifies a subset of them. This possibility is supported by recent in vivo work in visual cortex showing that input signals coding for multiple stimulus orientations are present in each neuron, regardless of its output tuning (24), and in vitro work showing that particular dendrites can propagate inputs to the soma more effectively than others (25) via dendritic spiking (15, 16, 25, 26).

Our results strongly support such nonlinear dendritic amplification mechanisms, regardless of whether synaptic inputs are uniformly distributed. In particular, in vitro work has shown that depolarizing the somatic (27, 28) or dendritic (29) Vm can gate inputs in a thresholdlike manner. The lack of a hill or increased fluctuations at resting Vm makes unlikely a purely somatic locus of amplification. Rather, somatic depolarization may (i) shift the dendritic Vm closer to the activation range of voltage-gated conductances or deactivate hyperpolarization-activated currents (increasing RN locally), and thus amplify spatially tuned responses and trigger dendritic spikes, or (ii) simply push the peak response in the dendrite above the dendritic spike threshold. Or instead, depolarization may act to let already existing dendritic spikes propagate to the soma (29).

These findings have important implications for place cells, spatial memory, and hippocampal-dependent memory in general. In novel environments, inhibitory neuron activity drops (30), which could depolarize the somatic baseline Vm of a silent cell and produce place fields (30, 31) without requiring synaptic plasticity. Neuromodulation could also cause depolarization or lower the threshold baseline Vm,gate described here. Furthermore, already existing place cells could be those neurons whose Vm,gate is below their baseline Vm. That is, (baseline VmVm,gate) could reflect intrinsic excitability, with depolarization compensating for lower excitability in silent cells. Alternatively, place cell inputs could differ or be arranged differently on the dendritic tree, although this would not easily explain preexploration differences in excitability (14). Finally, the explicit demonstration that most CA1 pyramidal neurons, even originally silent ones, can be place cells in a given maze suggests that the role of CA1 is not only to create spatial tuning, but to choose which subset of neurons should be active within a specific environment. The exclusively input-based models of place cells (711) could work as described but with an independent, excitability-based “AND” gate added. The distribution of excitability levels across the population would then determine which cells would represent a new item or environment in memory (14, 32). Moreover, the gating machinery would be a potential locus of plasticity for long-term storage (2529).

Supplementary Materials

Materials and Methods

Figs. S1 to S10


References and Notes

  1. Acknowledgments: We would like to thank G. Shtengel, J. Osborne, L. Ramasamy, D. Cabaniss, S. Bassin, C. Werner, B. Shields, A. Hu, M. J. Kim, H.-P. Liaw, G. Crosby, and D. Culley for technical advice and assistance and J. Dudman, N. Spruston, G. Murphy, J. Magee, K. Svoboda, J. Cohen, and D. Rich for valuable discussions about the manuscript. This work was supported by the Howard Hughes Medical Institute.
View Abstract

Navigate This Article