Input/Output Relationships for the Primary Hippocampal Circuit
- 1Departments of Anatomy & Neurobiology, University of California-Irvine, Irvine, California 92697
- 2Neurobiology & Behavior, University of California-Irvine, Irvine, California 92697
- 3Psychiatry & Human Behavior, University of California-Irvine, Irvine, California 92697
-
Author contributions: B.G.G., C.M.G., and G.L. designed research; B.G.G. and B.S.P. performed research; B.S.P. contributed unpublished reagents/analytic tools; B.G.G. and G.L. analyzed data; B.G.G., C.M.G., and G.L. wrote the paper.
Abstract
The hippocampus is the most studied brain region, but little is known about signal throughput—the simplest, yet most essential of circuit operations—across its multiple stages from perforant path input to CA1 output. Using hippocampal slices derived from male mice, we have found that single-pulse lateral perforant path (LPP) stimulation produces a two-part CA1 response generated by LPP projections to CA3 (“direct path”) and the dentate gyrus (“indirect path”). The latter, indirect path was far more potent in driving CA1 but did so only after a lengthy delay. Rather than operating as expected from the much-discussed trisynaptic circuit argument, the indirect path used the massive CA3 recurrent collateral system to trigger a high-frequency sequence of fEPSPs and spikes. The latter events promoted reliable signal transfer to CA1, but the mobilization time for the stereotyped, CA3 response resulted in surprisingly slow throughput. The circuit transmitted theta (5 Hz) but not gamma (50 Hz) frequency input, thus acting as a low-pass filter. It reliably transmitted short bursts of gamma input separated by the period of a theta wave—CA1 spiking output under these conditions closely resembled the input signal. In all, the primary hippocampal circuit does not behave as a linear, three-part system but instead uses novel filtering and amplification steps to shape throughput and restrict effective input to select patterns. We suggest that the operations described here constitute a default mode for processing cortical inputs with other types of functions being enabled by projections from outside the extended hippocampus.
Significance Statement
Despite intense interest in hippocampal contributions to behavior, surprisingly little is known about how signals are processed across the network linking cortical input to CA1 output. Here, we describe the first input/output relationship for the system with results challenging the traditional trisynaptic circuit concept. Signal throughput requires mobilization of recurrent activity within CA3 to amplify sparse input from the dentate gyrus into an unexpectedly stereotyped composite response. Potent low-pass filters determine effective input patterns. These results open the way to new analyses of how variables such as aging affect the hippocampus and its contributions to behavior while providing the material needed for biologically realistic models of the structure.
Introduction
The hippocampus is critical for declarative and episodic memory and has been postulated to integrate information relating to the identity and location of an object, as well as the context in which it appears (Squire and Dede, 2015). Studies of rodents and humans demonstrated that inputs from the lateral and medial entorhinal cortices carry information about the identities and locations of sequential cues (episodic “what” and “where” information, respectively; Hafting et al., 2005; Hargreaves et al., 2005; Yoganarasimha et al., 2011; Hunsaker et al., 2013; Tsao et al., 2013; Reagh and Yassa, 2014; Fernandez-Ruiz et al., 2021). Experimental work on rodents suggests that the dense CA3 recurrent collateral system supports encoding of the sequence in which events are sampled, a third basic element of an episode (Cox et al., 2019). While these results identify pathways that are critical for specific elements of episodic memory, they provide little insight into the types of dynamic operations executed by the network. In this regard, the large literature describing neuronal activity (spiking and local field potentials) associated with different aspects of learning (Colgin, 2016) has identified effects that emerge from active circuits but does not provide insights into the operational steps (i.e., signal transformations) that generate the observed action potentials or rhythms.
The absence of information on these points is particularly surprising because the design of the hippocampal circuit has been described in detail (Amaral and Witter, 1989). The perforant path arising from layer II of the entorhinal cortex (EC) forms two branches that target the dentate gyrus (DG) and CA3, thereby initiating direct (two synapses: EC-CA3-CA1) and indirect (three synapses: EC-DG-CA3-CA1) routes to the CA1 output station (Amaral, 1993; Witter, 1993, 2007b). This configuration constitutes the primary hippocampal circuit. A sparser excitatory projection from layer III EC to the most distal apical dendrites of CA1 (Amaral, 1993) and long-range inhibitory projections originating in layers II/III (Melzer et al., 2012; Basu et al., 2016) add further complexity. There are detailed anatomical and physiological analyses of synaptic transmission and plasticity mechanisms for most of the links in these pathways (Andersen et al., 2006) and numerous studies of interneurons in circuit nodes (Freund and Buzsaki, 1996; Klausberger et al., 2003, 2004; Fuentealba et al., 2008; Klausberger and Somogyi, 2008; Pelkey et al., 2017).
Of particular relevance to circuit function, frequency-dependent operations occurring at synaptic links reveal a series of presynaptic amplifiers (i.e., synaptic facilitation) and filters (i.e., synaptic depression; Toth et al., 2000; Mori et al., 2004; Jackman et al., 2016; Aldahabi et al., 2022; Quintanilla et al., 2022, 2024); such mechanisms are believed to be critical in determining circuit functions (Abbott and Regehr, 2004; Jackman and Regehr, 2017). Although the majority of studies have relied upon monosynaptic recordings, a smaller body of work shows that signal transfer from the hippocampus back to the EC is frequency-dependent (Moreno et al., 2016). There is also evidence suggesting that the lateral entorhinal cortex can entrain CA1 in a frequency-specific manner (Johnson et al., 2023). Despite these advances, the critical question of how the functional properties of individual links and nodes of the circuit contribute to signal transfer from the entorhinal cortex across the multiple stages of the hippocampus has not been addressed. There have for example been no descriptions of the spiking response of CA1 to stimulation of the perforant path—accordingly, very little is known about the manner in which preceding steps in the circuit shape its output.
The present studies address the following basic question: How are cortical input signals to the primary hippocampal circuit converted into reliable CA1 output? To assess this, we prepared hippocampal slices that retain sufficient connections between the major subdivisions to analyze throughput, from entorhinal input to CA1 output. Single-pulse lateral perforant path (LPP) stimulation produced a complex CA1 response that was much slower than expected from the conventional “trisynaptic circuit” models of the hippocampus. Moreover, signal throughput was strongly dependent upon the frequency, and rhythmicity, of the LPP input. Subsequent experiments discovered that mobilization of local recurrent activity within the dense CA3 collateral system by the mossy fiber (MF) projection, and not the canonical DG-CA3-CA1 circuit, drives the CA1 spiking response. This elaborate arrangement ensures reliable signal throughput, in a frequency and/or pattern-dependent manner, across the low-order (i.e., nontopographic) hippocampal circuit. The discovery of these novel operations provides essential information for hypotheses about how the operation of the hippocampal circuit at physiologically relevant frequencies might contribute to the processing of complex memories.
Materials and Methods
Animals
All studies used male C57/BL6 mice (Charles River Laboratories) from 2–4 months of age. Animals were group housed (five per cage) with access to food and water ad libitum and were on a 12 h light/dark cycle with lights on at 6:30 A.M. Experiments were conducted in accordance with the Institutional Animal Care and Use Committee at the University of California, Irvine, and the National Institute of Health Guidelines for the Care and Use of Laboratory Animals. For all electrophysiology studies, mice were anesthetized with isoflurane and killed by decapitation.
Electrophysiological recordings
Hippocampal slices were prepared as previously described (Cox et al., 2019; Quintanilla et al., 2022). Experiments were initiated from 8–10 A.M. Upon removal from the cranium, brains were placed in ice-cold, oxygenated (95% O2/5% CO2) high-Mg2+, artificial cerebrospinal fluid (HM-aCSF) containing the following (in mM): 87 NaCl, 26 NaHCO3, 25 glucose, 75 sucrose, 2.5 KCl, 1.25 NaH2PO4, 0.5 CaCl2, 7 MgCl2 (320–335 mOsm). Horizontal sections (400 µm) were cut using a Leica Vibratome (model VT1000s) in ice-cold (4°C) HM-aCSF and rapidly transferred to an interface recording chamber containing a constant perfusion (60–70 ml/hr) of oxygenated (95% O2/ 5% CO2) aCSF containing the following (in mM): 124 NaCl, 3 KCl, 1.25 KH2PO4, 1.5 MgSO4, 26 NaHCO3, 2.5 CaCl2, and 10 glucose (300–310 mOsm, pH 7.4, 31 ± 1°C). Recordings began 1–1.5 h later. All extracellular recordings were digitized at 20 kHz using an AC amplifier (A-M Systems, model 1700) and collected using NacGather 2.0 (Theta Burst).
Signal throughput CA1
For analysis of signal throughput, slices were prepared using the temporal half of the hippocampus. This typically yielded four 400-µM-thick slices (for each hemisphere), but optimal responses were usually only obtained from the middle two sections. We assume that a degree of connectivity sufficient for throughput analyses is present in a minority of transverse sections through the ventral hippocampus. A stimulating electrode was placed in the dentate gyrus (DG) outer molecular layer toward the apex of the two granule cell blades targeting the direct and indirect LPP projections while two recording pipettes were positioned, the first in CA1c stratum (str.) radiatum and the second in str. pyramidale of the same subfield (Fig. 1a). Single-pulse stimulation produced a complex two-part field excitatory postsynaptic potential (fEPSP), and stimulation intensity was set such that single units were reliably observed on the second component in the CA1 pyramidal cell (PC) layer response. A 20 min baseline period of single-pulse stimulation (0.3 Hz) was recorded, with spontaneous activity [i.e., single units and sharp waves (SPW)] additionally collected over a 10 s period after each stimulation pulse. To test the effect that repetitive LPP stimulation had upon CA1 spike output, brief trains were delivered across a range of frequencies [i.e., theta (5 Hz), gamma (50 Hz)] or patterns (theta–gamma; five bursts). Stimulation trains were delivered in a random order and separated by a minimum of 10 min. To confirm stimulating electrode placement, fEPSPs in str. lacunosum of field CA3 were recorded in response to paired-pulse stimulation (40 ms interval): Only slices displaying robust paired-pulse facilitation, indicating that the activated fibers belong to the LPP rather than MPP (Berzhanskaya et al., 1998), were included in the analysis.
The direct and indirect LPP projections differentially engage the hippocampal circuit. a, Left, Nissl stain of hippocampal slice depicting the position of stimulating electrode (X) and the two recording pipettes in CA1 (red areas). The direct and indirect circuits are illustrated with the components of the trisynaptic circuit labeled (1–3). Right, Representative LPP-evoked two-part fEPSPs recorded from the CA1 str. pyramidale and str. radiatum with the ensemble average fEPSP shown in red (scale bars: y = 0.25 mV, x = 20 ms). b, Left, Nissl stain illustrating the location of knife cuts used to remove the indirect (Cut 1) and direct (Cut 2) LPP inputs to CA3. The position of stimulating electrodes (X) and the recording electrodes (red areas) are illustrated along with representative responses recorded from the intact circuit and following removal of the indirect (top, MF CUT) and direct (bottom, LPP CUT) LPP inputs (scale bars: MF-CA3, y = 0.5 mV, x = 10 ms; LPP-CA3, y = 0.5 mV, x = 20 ms). c, Representative spontaneous SPWs recorded from the CA1 cell body layer (top) and apical dendrites (bottom). The red dots mark the individual SPWs on the dendritic trace (scale bars: y = 100 µV, x = 500 ms). d, e, Bar graphs summarize the mean ± SEM frequency (d) and amplitude (e) of SPWs recorded from CA1 apical dendrites in each slice configuration [intact circuit; direct path (DIR) only, with LPP-DG cut; indirect path (INDIR) only, with LPP-CA3 cut; p = 0.016 unpaired Student’s t test vs intact]. f–h, Schematic illustrating the circuit configuration (top) with exemplar raw (middle) and filtered (bottom) CA1 responses for the intact circuit (f) and with direct (g) and indirect (h) paths only. The red dots mark the single units in the raw (top) and filtered (bottom) recording for each (scale bars: y = 0.5 mV and 100 µV; x = 10 ms). i, The bar graph shows the percent of CA1 responses with spikes and their distribution across the first (1) and second (2) components of the fEPSP in the intact circuit (left) and with intact direct (DIR) and indirect (INDIR) paths only (right; for i, mean ± SEM values shown). j, k, Bar graphs summarize the latency to the first spike (j) and associated “jitter” (k) and the mean output frequency (l) for each circuit configuration (*p < 0.05; ***p < 0.001 unpaired Student’s t test). m, The graph shows the mean correlation of the first spike amplitude with those of the second and third spikes in CA1 after LPP activation in intact slices (left) and in slices containing only indirect LPP input (right). n, o, Scatter plots with unity lines (red dashed) show the correlation of the first spike amplitude (n) with those of the second and third spikes in the intact circuit (from 30 successive trials) and (o) with that of the second spike in a representative slice containing only the indirect input.
Contribution of direct and indirect paths
To test the relative contributions of the direct and indirect paths to CA1 spike output, surgical cuts were made to the hippocampal slice to sever the mossy fiber (MF)-CA3 projection or the LPP-CA3 projection (Fig. 1b). Surgical cuts were made immediately prior to slices being placed in the interface chamber to recover for 1–1.5 h after slice preparation. CA1 responses to baseline single-pulse and repetitive stimulation (i.e., theta, gamma, and theta–gamma) were conducted as described above. Following each experiment, MF-CA3 cuts were confirmed by delivering a 10 pulse, 20 Hz train via a stimulating electrode repositioned in the MF projections and recording responses from a pipette placed in the PC layer of CA3b (Fig. 1b). LPP-CA3 cuts were confirmed by repositioning one of the recording pipettes to the distal apical dendrites (i.e., str. lacunosum moleculare) of CA3 and delivering paired-pulse stimulation (40 ms interval, Fig. 1b).
MF-CA3 responses
To directly activate MF projections, a stimulating electrode was positioned within the hilus proximal to the granule cell layer, and fEPSPs were recorded from field CA3c via two recording pipettes, the first in str. radiatum and the second on the apical edge of the PC layer. Stimulation intensity was set to evoke a modest monosynaptic fEPSP in the PC layer. A second set of experiments recorded responses across the proximodistal axis of CA3. As such recording pipettes were positioned on the apical edge of the PC layer in CA3c, CA3b, and CA3a.
LPP-CA3 responses
A stimulating electrode was positioned in the DG outer molecular layer (as above), and a single recording pipette was positioned within the PC layer of CA3a. The stimulation intensity was set to evoke a modest fEPSP (0.5–1 mV) that reliably contained single units. A 20 min period of single-pulse stimulation (0.3 Hz) was recorded, with spontaneous activity (i.e., single units and SPWs) additionally collected (10 s) after each stimulation pulse.
Analysis of data
All recordings were analyzed offline. The properties of the fEPSP waveform were analyzed using NacShow 2.0 (Theta Burst), while a custom code (Python version 3.8) was used to analyze spontaneous activity and evoked single units.
fEPSP analysis
LPP-evoked fEPSPs recorded from the apical dendrites (i.e., str. radiatum) of CA1c were analyzed with regard to peak amplitude (total), initial slope (20–80%) of the rising phase, area, and half width of the waveform. An ensemble average fEPSP was generated (20 events) and used to calculate the fEPSP decay τ and latency to onset. The decay τ was described using the monoexponential equation Y(t) = A ∗ exp(−1 / t). Responses recorded from the PC layer were analyzed with regard to peak amplitude and area. Total values as well as those for the initial and secondary components of the waveform were measured.
Evoked single units
LPP- and MF-evoked spikes recorded from CA1 or CA3 were analyzed offline using a custom-written computer code created with Python 3.8. Briefly, extracellular recordings (10 s sweeps) were fed through a bandpass filter (300–5,000 Hz). Spikes were detected using a combined amplitude threshold (−45 µV) and rate of rise threshold (−240 µV/ms). Such detection parameters were used to avoid the detection of noise. The spike output generated by single-pulse LPP activation was analyzed across the 50 ms period following the stimulation artifact. For each slice, 30 consecutive responses to single-pulse stimulation were analyzed for the number of spikes, interspike interval (ISI), latency to first spike, spike amplitude, and instantaneous frequency (i.e., 1/ISI) of spike output. The standard deviation of the first spike latency was used to describe the “jitter” within each slice. To confirm that the measured spikes were driven by LPP activation, we calculated the probability of such an output occurring spontaneously after any given spontaneous spike. For each slice, we took the mean ± SD of the spike output (number and frequency) along with latency to the first spike and measured the probability of such a pattern occurring after any single spontaneous spike across the 10 s for each of 30 consecutive sweeps. Individual trials were visually inspected, and any responses that were contaminated with spontaneous events or spurious noise were dropped from the analysis.
Analysis of repetitive stimulation
The CA1 spike output to brief repetitive LPP stimulation at different frequencies (i.e., 5 Hz and 50 Hz) and patterns (i.e., theta–gamma) was analyzed using the same custom code described above. LPP-evoked spikes were analyzed for each pulse with regard to the number of spikes, latency to the first spike and associated “jitter,” spike amplitude, ISI, and instantaneous frequency of spike output. With 5 Hz stimulation, spiking was analyzed for 50 ms after the stimulation artifact associated with each pulse. With 50 Hz or theta–gamma burst stimulation, spiking was assessed (1) across the 20 ms interval between gamma frequency pulses and (2) over 50 ms periods following the initial pulse in the train (i.e., divided into four consecutive 50 ms epochs) or burst. Assessing LPP-evoked spiking over such 50 ms periods enabled quantitative comparison with single-pulse output and with responses to 5 Hz stimulation.
Statistical analysis
All results are presented as group means ± SEM. Statistical comparisons between different circuit configurations were made using unpaired Student’s t tests (GraphPad Prism version 6.0), unless otherwise stated in the text or figure legends. For all experiments, the group n refers to the number of slices used for electrophysiological recordings with a minimum number of three animals per group. A p-value of <0.05 was considered significant for the Student’s t test and two-way ANOVA, with a more stringent criterion for significance (p < 0.01) used for the Kolmogorov–Smirnov test.
Results
Single-pulse stimulation of the LPP produces a novel CA1 response
Using a novel brain slice preparation, we investigated the propagation and transformation of LPP-evoked signals across the hippocampal circuit (Fig. 1a and Materials and Methods). Single-pulse LPP stimulation elicited a surprisingly complex two-part fEPSP in CA1 that was not expected given the arrangements of the two LPP inputs (i.e., direct and indirect) and the well-described trisynaptic circuit (Fig. 1a). We had expected that in a slice retaining a significant percentage of LPP inputs and the connections between subfields, CA1 responses would occur with the latency predicted for a di- or trisynaptic system and would display the expected summation of responses that occur in close temporal proximity (i.e., not two distinct waveforms). While the onset of the initial broad fEPSP was consistent with a di- or even trisynaptic circuit (5.2 ± 0.3 ms; n = 11 slices; Table 1), the presence of a significantly delayed (∼20 ms to peak) larger secondary potential was surprising and could not be accounted for by the direct (LPP-CA3-CA1) or indirect (LPP-DG-CA3-CA1) paths to CA1. Despite this, both the initial and secondary potentials were negative in the proximal apical dendrites and positive in the cell body layer, indicating that they were indeed fEPSPs generated by the dense CA3 to CA1 projections (Fig. 1a). It thus appears that unidentified processing steps within the network produced a substantial secondary activation of CA1; a notion supported by the significantly slower slope of the delayed secondary response, in conjunction with the increased variance of this response in both dendritic and cell body layers (Table 1). Sharp waves and spontaneous events generated in CA3 (Kubota et al., 2003; Rex et al., 2009), and recorded in CA1, exhibited polarity reversals along the cell body–dendritic axis (Fig. 1c–e) as expected for slices with considerable preservation of the Schaffer collateral system. There was no evidence that LPP-targeted stimulation engaged a short latency (monosynaptic) projection to CA1.
Properties of dendritic (A) and somatic (B) LPP-evoked fEPSPs recorded from slices containing the intact hippocampal circuit as well as those lacking the indirect or direct components
Action potential spiking in CA1, denoting circuit output, was reliably evoked by single-pulse LPP activation, although surprisingly this spiking was largely restricted to the delayed secondary component of the two-part fEPSP response (Fig. 1f); this occurred in 92 ± 2% of LPP-evoked responses whereas spikes on the first fEPSP occurred in only 7.4 ± 3% of the cases (Fig. 1i). The latency to the first spike (19.6 ± 1.0 ms; n = 20 slices) was much longer than anticipated for the classic trisynaptic hippocampal circuit yet surprisingly consistent between successive pulses within and across slices (mean “jitter”: 4.2 ± 0.4 ms, n = 20 slices; Fig. 1j,k). The majority of delayed secondary responses (76 ± 5%) contained more than one spike with a mean spike frequency of 256 ± 12 Hz (Fig. 1L, Table 2). Two lines of evidence suggested that this high-frequency spiking was generated by the near synchronous activation of a small population of CA1 pyramidal cells. First, the amplitude of the first spike was variable across trials within the same slice [coefficient of variation (CV) = 37 ± 2; n = 20 slices], and second, the sizes of the first and subsequent spikes were not correlated, indicating that these spikes were likely generated by different cells (Fig. 1m,n, Table 2).
Properties of LPP-evoked CA1 spiking recorded from slices containing the intact hippocampal circuit as well as those lacking the indirect or direct components
Origins of the two-part CA1 response to LPP stimulation
In principle, the two-part response could reflect sequential activation of the CA3 to CA1 projection by the subcircuits initiated by the two branches of the LPP (to CA3 or DG). However, given that the primary difference between the two LPP projections is the addition of a single synapse (i.e., LPP-DG in the indirect path), such a hypothesis does not intuitively explain the production of the two temporally distinct waveforms and the protracted delay before the onset of spiking (i.e., these responses do not simply reflect the sum of the synaptic delays). We severed specific projections using knife cuts (Fig. 1b and Materials and Methods) to test the validity of such a two-pathway argument. Severing the LPP input to CA3 (LPP-CA3) or the mossy fibers (MF-CA3) did not reliably change the frequency of SPWs recorded in the apical dendrites of CA1, although a modest yet significant decrease in the amplitude was observed in slices lacking the direct LPP-CA3 input (Fig. 1d,e). This indicates that these transections only modestly disrupted the self-organizing events leading to large spontaneous depolarizations. In contrast, both cuts had major effects on the field CA1 fEPSP. The MF cut eliminated the second component of the CA1 response to LPP stimulation, leaving a short latency potential that corresponded to the first element of the response recorded in the intact slice (Fig. 1g, Table 1). Cutting the LPP-CA3 axons, thereby leaving intact the LPP-DG-CA3 indirect path, produced complementary results: The first component of the CA1 response was gone but the delayed secondary potential was intact (Fig. 1h, Table 1). The properties of fEPSPs recorded from slices lacking the indirect (i.e., MF-CA3 cut) or direct (i.e., LPP-CA3 cut) input were remarkably similar to the initial and secondary components of the two-part response of the intact circuit, respectively (Table 1).
The transection experiments also provided an opportunity to test for interactions between the two pathways at the level of CA1 spiking. Although cutting the MF-CA3 axons did produce a modest increase in the unreliable spiking response evoked by the direct path on the first of two CA1 fEPSPs (Fig. 1i), this reflected spikes occurring significantly later in the waveform (Fig. 1j, Table 2). Sectioning the LPP-CA3 axons had little if any effect on the probability, or the latency and associated “jitter,” of spikes following LPP activation (Fig. 1j,k). As in the intact slice, a high-frequency spike output was observed that appeared to be driven by the near synchronous discharge from a small number of CA1 pyramidal cells (Fig. 1m,o, Table 2).
Collectively, the results show that the LPP activates two intrahippocampal pathways that sequentially engage CA1 with an unexpected delay between them. The second, delayed input drives the CA1 output (i.e., firing) and interactions between it and the direct path are surprisingly subtle.
Variability and consistency in the CA1 response
As described, CA1 output evoked by single-pulse LPP stimulation was largely restricted to the second, considerably delayed fEPSP. This was the case even when a sizeable number of spikes were triggered by single-pulse activation of the LPP. Trial-by-trial responses did not differ greatly for individual slices containing the intact circuit or those with the indirect path only (Fig. 2a,b). There was, however, considerable variability between slices in both slice configurations (Fig. 2c,d). There was a significant correlation between the number of evoked spikes and the frequency of spontaneous SPWs (Fig. 2e), indicating that the between-slice variations were not due simply to the location of the stimulating electrode. We measured ISIs for those slices containing the intact circuit where LPP stimulation elicited at least three spikes in 50% of trials and found minimal differences within and between slices. The mean ISI was 4.7 ± 0.4 ms between the first and second spike and 4.8 ± 0.6 ms between the second and third spike, with these values being correlated across slices (Fig. 2f. R2 = 0.7899). As noted, spikes within a response are generated by different neurons, and it is therefore likely that the low variance in spike spacing across trials and slices involves a yet to be identified interaction between neighboring cells. Although cutting the direct LPP projection (i.e., LPP-CA3) decreased the number of LPP-evoked spikes per trial across slices (Fig. 2c,d,g), there was no evident effect on ISI for responses triggered by the LPP (intact, 4.5 ± 0.3 ms; indirect, 5.1 ± 0.5 ms; Fig. 2g).
The two LPP branches interact to modestly increase the CA1 spike output to single-pulse stimulation. a, b, Example recordings from an intact slice (a) and a slice with an intact indirect LPP path only (b; configuration shown in Fig. 1h) show CA1 spike output (top) and the accompanying raster plots (bottom) illustrating the distribution of single units within the LPP-evoked responses (15 successive pulses: open circles). Individual spikes within the filtered trace (top) and the mean output in the raster plot (bottom) are indicated by red dots. Scale bars: y = 50 µV, x = 10 ms. c, d, The proportion (%) of LPP-evoked CA1 responses containing one to six spikes for each slice (circles) in the intact configuration (c) and in the indirect path-intact configuration (d; bars show mean ± SEM values for each number of spikes). e, f, Scatter plots with unity line (red dashed) show the relationships between (e) the mean number of LPP-evoked CA1 spikes and the mean frequency of spontaneous SPWs and (f) the intervals between the first and second spikes and the second and third spikes in intact slices. g, Graph summarizing the mean number of spikes and their temporal distribution within the LPP-evoked CA1 response in slices containing the intact circuit (open circles) and indirect path only (blue circles). h, i, Cumulative probability plots show (h) the difference in the distribution of spike numbers per response between the intact circuit and indirect path-only circuit (p < 0.001 Kolmogorov–Smirnov test) and (i) no difference in the temporal distribution of spikes within the waveform.
In a further test for the effect of severing the direct LPP input to CA3 on the responses generated by the indirect path (i.e., LPP-DG-CA3-CA1), we plotted the number of evoked spikes for all slices and all trials as a cumulative probability. The resultant curve for the indirect path-only group was clearly left shifted toward fewer spikes from that for the intact slices (Fig. 2h; p < 0.0001, Kolmogorov–Smirnov test). In contrast, curves for the onset time of evoked discharges were superimposed (Fig. 2i). This suggests that the direct path does enhance the spiking response to indirect path stimulation but a large sample size is needed to detect the effect.
The MF-CA3 link adds a complex response to the indirect path
The possibility that the indirect path, comprised of only three synapses (i.e., the “trisynaptic circuit”: LPP-DG-CA3-CA1), requires almost 20 ms to drive CA1 spiking seems unlikely. This observation strongly implies that additional levels of processing within the circuit are necessary to drive throughput, and the dense recurrent commissural–associational (C/A) system in field CA3 provides an attractive candidate for performing such operations. Indeed, prior studies have demonstrated that MF stimulation is capable of driving a secondary recurrent fEPSP in CA3 (Skucas et al., 2013), although little is known about the interaction(s) between these two responses and their respective contribution to hippocampal throughput. To re-examine this point, we first stimulated the MFs deep within the DG hilus and recorded from the PC bodies and proximal apical dendrites of field CA3b (Fig. 3a). Consistent with the earlier report (Skucas et al., 2013) single-pulse stimulation produced a small, short latency monosynaptic fEPSP, which was followed by a large, long-duration fEPSP that reversed in the proximal apical dendrites, indicating the presence of a large current sink at this site (Fig. 3b). The peak of the second potential, whether recorded from the pyramidal cell layer or apical dendrites, was delayed by ∼4 ms from the monosynaptic MF response (Fig. 3c, Table 3). This interval combined with the presence of a somadendrite dipole indicates that the secondary response was generated by the CA3 C/A projections (MF-CA3-CA3).
The mossy fiber synapse reliably engages the CA3 recurrent system. a, Nissl stain of hippocampal slice showing the location of the stimulating electrode (X) adjacent to the granule cell layer and the recording pipettes in str. radiatum and the pyramidal cell (PC) layer of CA3a (red shading). The positions of recording pipettes across the proximodistal axis of CA3 (i.e., CA3c and CA3b) are also illustrated. b, Representative MF-evoked fEPSPs recorded from CA3a str. pyramidale (top) and str. radiatum of CA3a across a range of stimulation intensities. The red dot marks the short latency, monosynaptic MF-evoked fEPSP recorded from the PC layer (y = 2 mV, x = 10 ms) c, Graph shows the latency to fEPSP peak for mono- and disynaptic MF-evoked responses. d–f, Graphs show the latency to onset (d) and peak (e) of the monosynaptic MF-CA3 fEPSP and the latency to the peak of the disynaptic fEPSP (f) across the CA3 proximodistal axis: CA3c, CA3b, and CA3a. For c–f, bars denote mean ± SEM values. g, Representative MF-evoked responses recorded from CA3c, CA3b, and CA3a, with responses to increasing stimulation intensities shown in gray. The peak mono- and disynaptic responses used to calculate the ratio are identified with closed and open red circles, respectively (scale bars: y = 2 mV, x = 10 ms). h, Exemplar raw (top) and filtered (bottom) MF-evoked CA3 responses, with single units identified (red circles; scale bars: y = 1 mV and 50 µV, x = 10 ms). i, Raster plot from a representative slice showing the distribution of single units within the MF-evoked CA3 responses across 15 successive pulses (open circles) and with the mean slice output indicated (red circles).
Properties of mono- and disynaptic MF-evoked fEPSPs recorded from the cell body (somatic) and dendritic layers of field CA3a
The nature of the MF projections, being thin and poorly myelinated (Claiborne et al., 1986), suggests that signal conduction along the proximodistal axis of CA3 will be slow. This is relevant to the question of the long delay in the indirect path response because the sampled CA1 region (CA1c) receives most of its input from the more distal regions of CA3 [i.e., CA3a (Witter, 2007a)]. To evaluate this, we measured the amplitudes and onset times for the monosynaptic MF response and the subsequent C/A potential at three proximodistal points in CA3 (i.e., CA3c-CA3b-CA3a). As expected, the monosynaptic MF response was considerably delayed in distal CA3 relative to values recorded proximal to the DG (p = 0.0001, one-way ANOVA; Fig. 3d,e). In contrast, a tendency toward greater delay with distance from the DG for the peak of the secondary potential did not approach statistical significance (p = 0.1104; Fig. 3f). The amplitude of the monosynaptic MF response underwent a marked decrease from the middle to the distal end of CA3 (Table 3) presumably due to (1) a decreasing percentage of stimulated fibers that remain in the plane of the slice between the stimulation and recording sites and (2) the decline in MF terminals within the basal dendritic field along the CA3 proximodistal axis (Blackstad et al., 1970; Henze et al., 2000). The results for the C/A fEPSP did not parallel those for the monosynaptic MF responses. Peak amplitude was greatest at the midpoint of the CA3 axis (CA3b) and was not nearly as different between the distal and proximal fields as was the case for the monosynaptic MF potential (Fig. 3g). The recurrent collateral system is comprised of local and long-range fibers arising from sites throughout the field (Sik et al., 1993; Li et al., 1994; Guzman et al., 2016). The gradient of secondary potentials just described could reflect a forward bias (toward CA1 rather than the DG) in the long collateral branches.
Spiking initiated by MF stimulation typically began with the initial monosynaptic response and continued through the rising phase and up to the peak of the secondary potential (Fig. 3h). A trial-by-trial record for a typical slice showed that pyramidal cell discharges typically continue for ∼10 ms (Fig. 3i). This block of time, when added to delays required by processing in and the subsequent transmission through the LPP-DG-MF pathway, provides a plausible explanation as to why almost 20 ms is required for the indirect path to engage CA1. We therefore propose that CA1 is not directly triggered by the monosynaptic response of CA3 to MF input (i.e., the trisynaptic circuit) but rather by the local cycling of the associational CA3–CA3 connections set in motion by the MFs. These features combined with the conduction velocity of the MFs account for the slow throughput that is a singular feature of the hippocampal circuit.
The LPP triggers a prolonged but stereotyped CA3 response
We compared the LPP-driven activation of CA3 with that obtained following MF stimulation (above). Single LPP pulses produced a striking response. Recordings in the CA3 pyramidal cell layer were characterized by an initial large potential followed by a series of smaller fEPSPs most of which were accompanied by an action potential (Fig. 4a). Indeed, the LPP pulse triggered more spikes than did MF stimulation, as can be seen in trial-by-trial measures for representative cases (Fig. 4b) and group summaries (Fig. 4c,d, Table 4). Note that although LPP stimulation triggered more action potentials in CA3 than it did in CA1 (Fig. 1f), the high-frequency nature of the spiking along with the lack of correlation between the amplitude of the first and subsequent spikes suggests that they are similarly generated by a small population of CA3 pyramidal cells (Fig. 4e,f). As expected, the onset of the CA3 response to LPP stimulation was delayed and more variable than that for the direct MF activation (Fig. 4g,h, Table 4).
LPP activation produces a prolonged spike output from CA3 that is highly stereotyped in nature. a, Exemplar raw (top) and filtered (bottom) LPP-evoked responses recorded from CA3a: The single units are identified (red circles) for each (scale bars: y = 200 µV and 50 µV, x = 10 ms). b, Raster plots show the distribution of single units within the LPP-evoked (top) and MF-evoked (bottom) CA3 responses (15 successive pulses: open circles) derived from representative slices. The mean output for each slice is shown (red circles). c, d, Bar graphs summarize (c) the mean number of spikes per response and (d) the proportion of responses with >2 spikes for LPP (open circles) and MF-evoked (gray circles) CA3 responses; mean ± SEM values shown in bar graphs. e, Graph summarizing the mean correlation of the first spike amplitude with those of the subsequent six spikes in CA3 following LPP activation. f, Scatter plot with unity line (red dashed) summarizing the correlation of the first spike amplitude with those of the second (open circles) and seventh spike (gray circles) recorded from CA3a following LPP activation. g, h, Bar graphs summarizing (g) the mean first spike latency and (h) the associated jitter of LPP- (open circles) and MF-evoked (gray circles) CA3 responses. i, j, Bar graphs show the proportion (%) of CA3 responses containing one to nine spikes for each slice following LPP (i; open circles) and MF (j; gray circles) stimulation: The bars show the mean ± SEM values for each number of spikes. k, Graph summarizing the mean number of spikes and their temporal distribution within the LPP-evoked (open circles) and MF-evoked (gray circles) CA3 response. l, Graph summarizing the mean interspike intervals (ISI) between successive spikes (Spikes 1–7) across the LPP-evoked waveform for each slice. The red bars indicate the mean values. m, Exemplar raw (top) and filtered (bottom) spontaneous SPWs recorded from CA3a. The single units are identified (red circles) for each (scale bars: y = 100 µV, x = 10 ms). n, Raster plots show the distribution of single units within the waveform of LPP-evoked CA3 responses (top) and SPWs (bottom; 15 successive responses for each: open circles) derived from representative slices. The mean output for each slice is indicated (red circles). o, Graph of the within-slice comparison between the mean number of LPP-evoked spikes (open circles) with the mean of those associated with SPWs (green circles). The red circle identifies the case where SPW spiking was elevated. p, Bar graph summarizing the proportion (%) of CA3 SPWs containing one to nine spikes for each slice, with bars showing the mean ± SEM value for each number of spikes: Note the similarity to the distribution of the LPP-evoked response (panel i). q, Graph shows the mean number of spikes and their temporal distribution within the waveform of LPP-evoked responses (open circles) and spontaneous SPWs (green circles) recorded from CA3a. r. The graph shows the mean ISI between successive spikes (Spikes 1 to 5) associated with SPWs for each slice; the red lines indicate the mean ± SEM values.
Properties of LPP- and MF-evoked spiking recorded from CA3a
The CA3 spike output in response to LPP and MF stimulation exhibited the same low within-slice variance evident in CA1 (Fig. 4b). As anticipated, given the lower mean number of MF-evoked spikes, the number of spikes elicited for each trial was typically lower than seen following LPP stimulation (Fig. 4i,j). Indeed, following LPP activation, at least 50% of trials elicited seven spikes or more in at least half of the slices tested, whereas MF stimulation only evoked three spikes (Fig. 4i–k). Despite these differences, the ISI was remarkably similar in responses evoked by the two inputs (Fig. 4k). The mean ISI for LPP stimulation was 5.0 ± 0.3 ms, and this was consistent across successive interspike intervals (Fig. 4l). The intervals between the first and second spikes were modestly, yet significantly different (LPP, 4.5 ± 0.3 ms; MF, 3.3 ± 0.3 ms; p = 0.0228 unpaired t test), whereas no differences were evident for the interval between the second and third spike (LPP, 4.2 ± 0.2 ms; MF, 5.2 ± 1.1 ms; p = 0.1606 unpaired t test). These findings suggest that the recurrent C/A system may be predisposed to self-organization and subsequent generation of stereotyped responses. If so, then it is possible that the large spontaneous SPWs in CA3 would display a similarly stereotyped spiking profile. We tested this using the same filtering procedures employed in the stimulation experiments and estimating the start of the SPW from the field potential trace (Fig. 4m). Within-slice comparisons confirmed that fewer spikes were present during SPWs than were elicited by LPP stimulation (Fig. 4m–o; p = 0.0002 paired t test). As with the LPP-evoked responses, the variability in spikes associated with SPWs was low within each slice (Fig. 4n), but considerably larger across slices (Fig. 4p). In general, the ISI values for spikes within a SPW were remarkably similar to those obtained following LPP stimulation (Fig. 4q). However, in contrast to LPP-evoked spikes the variance in the ISIs between successive spikes (i.e., first, second, and third) associated with SPWs actually decreased, with a tendency for the mean ISI to be slightly shorter than was the case with LPP stimulation (Fig. 4q,r).
Throughput from the LPP to field CA1 is frequency-dependent
We next investigated the output of the circuit in response to LPP stimulation frequencies that are both prominent during behavior and widely utilized in models of cortical and hippocampal computations (Colgin, 2016). Activation at the theta frequency (5 Hz) produced spiking by CA1 pyramidal cells that were maintained throughout a train of 10 pulses with or without the direct input to CA3 (Fig. 5a,b). Results for all slices in both groups confirmed that throughput (i.e., CA1 cell firing) was almost always present for each pulse in the train (Fig. 5b), with the number and patterning of spikes elicited by the 10th pulse not being reliably different than those for the 1st pulse (Fig. 5c,d). The similarity between the first and last responses extended to the onset time and its associated variance for the spiking response, which were virtually identical for the 1st and 10th pulses in the intact slices (Fig. 5e, left). Interestingly, the transection of the direct LPP input shortened the delay to the 10th pulse relative to the 1st pulse (Fig. 5e, right). The number of action potentials in the 50 ms following the first and last pulses correlated across slices with the responses collected during baseline recording (Fig. 5f), indicating again that repetitive stimulation at 5 Hz does not significantly change the resting properties of the system.
The hippocampal circuit operates as a low-pass filter in response to repetitive stimulation. a, Representative filtered LPP-evoked CA1 responses to the 1st (top) and 10th (bottom) pulse of a 5 Hz (10 pulses) stimulation train in the intact circuit (left) and with the indirect path only (right; scale bars: y = 50 µV, x = 10 ms). The single units are identified (red circles) for each. b, Raster plot of LPP-evoked CA1 spiking across the 5 Hz stimulation train for each slice (open circles) recorded with the intact circuit (top) and indirect path only (bottom). c, The distribution of CA1 spiking during the 1st and 10th response for each circuit configuration is illustrated on an expanded time scale; the red circles indicate the mean spike distribution for all slices in both. d, e, Bar graphs show the mean number of CA1 spikes (d) and first spike latency (e) on the 1st and 10th stimulation pulse of a 5 Hz train in the intact (white) and indirect path-only (blue) slice preparation. f, Scatter plots with unity line (red dashed) show the mean number of CA1 spikes evoked during the single-pulse baseline period relative to the same-slice response to the 1st and 10th pulse of a 5 Hz stimulation train for the intact circuit (left) and the indirect path-only preparation (right). g, Representative filtered LPP-evoked CA1 baseline response (top) and 10th pulse (bottom) of a 50 Hz (10 pulse) stimulation train in the intact circuit. h, Representative filtered LPP-evoked CA1 baseline response (top) and responses over the first 50 ms (3 pulses) of a 50 Hz (10 pulses) stimulation train in the intact circuit (scale bars for both: y = 50 µV; x = 10 ms). The single units are identified (red circles) for both. i, Raster plot of LPP-evoked CA1 spiking across the 50 Hz stimulation train for each slice (open circles) recorded from the intact circuit (top) and from the indirect path-only preparation (bottom). j, The distribution of CA1 spiking during the first and last 50 ms periods of the stimulation train for each slice (both circuit configurations) is illustrated on an expanded time scale; the red circles denote the mean spike distribution for all slices in both. k, Graph summarizing the number of evoked CA1 spikes in each 50 ms epoch across the 50 Hz stimulation train in the intact slice (open circles) and the indirect path-only preparation (blue circles). l, The mean number of spikes evoked in the first and last 50 ms of the 50 Hz train for the intact circuit (white) and indirect path only (blue). m, Scatter plots with unity line (red dashed) show the mean number of CA1 spikes evoked during the single-pulse baseline period and during the first and last 50 ms epochs of a 50 Hz stimulation train for each slice in the intact circuit (left) and with indirect path only (right). Bar graphs show mean ± SEM values.
Repeating the experiments with LPP stimulation at 50 Hz (gamma) yielded a startling result: There was typically no spike response to the 10th pulse in the train (Fig. 5g, bottom). Combined with the theta results, this outcome indicates that the hippocampal circuit operates as a potent low-pass filter (i.e., 5 Hz activity drove continued throughput to CA1 whereas 50 Hz activity did not). Representative traces showed that the filtering was not present in the response to the second pulse (Fig. 5h, bottom), whereas pulse-by-pulse analyses for all slices in the group indicated that suppression emerged by the third pulse (Fig. 5i). While the effects of transection appeared minimal (Fig. 5i), efforts to quantitatively assess possible effects were complicated by the similarity between the latency to spike (19.6 ± 1.0 ms) and the 20 ms intervals in a 50 Hz train. To circumvent this problem, we divided the 200 ms gamma train into 50 ms epochs, which was the length of the sampling period used to assess CA1 spike output to baseline and 5 Hz stimulation. The results showed that there was a ∼85% reduction in the number of evoked spikes between the first and fourth stimulation blocks (Fig. 5k,l). The number of evoked spikes in the first 50 ms epoch was tightly correlated with, and greater than, the number elicited by single pulses during baseline stimulation. The last block of responses was predictably reduced from pretrain values (Fig. 5m, left). A similar pattern held for slices in which the direct path had been severed (Fig. 5m, right). Note that, consistent with single-pulse stimulation, activation of the direct LPP input alone (i.e., MF cut) did not reliably drive CA1 spikes at either of the two frequencies.
Theta–gamma LPP stimulation obviates low-pass filtering by the circuit
It is puzzling that the cognitively critical gamma rhythm is not relayed across the multiple stages of the hippocampal circuit. Possibly relevant to this, gamma activity often occurs in brief bursts separated by the period of the theta wave (Colgin, 2015), and, as described (Fig. 5h), two pulses applied to the LPP at 50 Hz elicit a robust response in CA1. The question then arises as to whether a series of short gamma frequency bursts produce reliable throughput as opposed to engaging potent filtering mechanisms. We tested this by applying three pulses to the LPP at 50 Hz and then repeated this at 200 ms intervals for a total of five gamma bursts. LPP activation with such theta–gamma stimulation produced reliable CA1 spike output in response to each individual burst with the response to the fifth burst being equal to or greater than that of the first (Fig. 6a). A burst-by-burst analysis across all slices confirmed that each set of three pulses elicited a robust response in CA1 (Fig. 6b). These data also suggested that while removing the direct path reduced the amount of spiking to a given burst, there was an increase in bursting across successive bursts (Fig. 6b,c). An analysis of individual burst responses strengthened this impression. While the difference in the mean number of spikes across the 50 ms period following initiation of each gamma burst approached significance (intact, 3.8 ± 0.1; indirect, 3.1 ± 0.3, p = 0.0680 unpaired t test), the absolute change in spike number across successive bursts differed significantly between the two groups (Fig. 6d, F(4,100) = 2.478, p = 0.0488 two-way ANOVA). These results constitute a clear instance in which the direct path influences an operation of the indirect path.
Theta–gamma stimulation obviates the low-pass filter. a, Representative filtered CA1 responses to theta–gamma (5 bursts) stimulation of the LPP input (top) in an intact slice preparation. Responses to bursts 1 and 5 are shown on an expanded time scale (bottom) with the single units identified (red circles) for each burst (scale bars: y = 50 µV, x = 100 ms and 10 ms). b, Raster plots of LPP-evoked CA1 spiking across the theta–gamma stimulation train for each slice (open circles) recorded with the intact circuit (top) and indirect path-only preparation (bottom); the red dots indicate the mean spike distribution for all slices. c, The distribution of CA1 spiking during the first and fifth burst for each slice (shown in panel b) is illustrated on an expanded time scale. d, Graph summarizing the mean change in the number of evoked spikes for each burst recorded from intact slices (open circles) and those containing only the indirect path (blue circles). Note the stable response in the intact slice and the significant increase in the number of spikes across bursts in the absence of the direct input. e, f, Graphs illustrating the mean CA1 spike number evoked (e) by successive bursts with theta–gamma activation of the LPP in the intact circuit (open circles) and the indirect path-only preparation (blue circles) and (f) by the first and fifth bursts in a theta–gamma train in each slice configuration. g, Scatter plot with unity line (red dashed) shows the mean number of CA1 spikes evoked per burst relative to same-slice single-pulse baseline period for intact circuit (open circles) and with indirect path only (blue circles). h, i, Similar scatter plots show the mean number of CA1 spikes elicited by the first and fifth burst of a theta–gamma train relative to the same-slice single-pulse baseline response in the intact circuit (h) and indirect path-only (i) preparations.
As anticipated, the number of evoked spikes was extremely stable over the course of five theta–gamma bursts in the intact slices, a feature that was less evident in slices lacking the direct LPP-CA3 input (Fig. 6e,f). Notably, between-slice variations in spike responses were correlated with the magnitude of the responses to single-pulse stimulation during baseline recording (Fig. 6g–i). As with single-pulse stimulation, 5 and 50 Hz LPP stimulation to the direct input alone was ineffective at driving CA1 spiking. We conclude from these results that theta–gamma signals from the lateral entorhinal cortex are reliably transferred across the circuit with minimal distortion and that the direct path promotes this indirect path function. Indeed, the strongest influence of the direct path on the throughput of LPP signals emerges during theta–gamma input.
Discussion
The goal of this project was to obtain a first description of how simple inputs are transformed across the multiple stages of the primary hippocampal circuit. Such information is not only critical for constraining arguments about hippocampal function but is also essential for the development of realistic simulations and the related bottom-up hypotheses of memory-related functions of this structure. Results were unexpected and largely inconsistent with usual assumptions about the operation of intrahippocampal networks. Single-pulse stimulation of the LPP, one of two primary EC inputs, evoked two distinct fEPSPs in CA1. Cell firing occurred almost exclusively on the second of these waveforms, displaying surprising consistency with regard to the first spike latency and the intervals between events. Sectioning studies revealed that the direct LPP branch (i.e., LPP-CA3) initiated the first potential whereas the second waveform and associated spiking were driven by the LPP-DG-CA3 subcircuit. However, these experiments did not explain why the inclusion of one additional synapse (i.e., DG-CA3) produces uncommonly slow yet temporally consistent throughput.
The high output frequency and lack of correlation between spike amplitudes prompted the conclusion that the CA1 response was produced by the near synchronous activation of a local population of neurons. This pattern undoubtedly reflects interactions between dendritic properties, local circuit operations, and the spatiotemporal organization of input from CA3. Pyramidal cells are commonly described as integrators, processing temporally dispersed inputs, or as coincidence detectors, preferentially responding to near synchronous signals (Abeles, 1982; Ratte et al., 2013). The functional mode the cells display strongly influences spike initiation and the degree of synchrony between the input and target cell output. Although dendritic integration is complex (Magee, 2000; Magee and Johnston, 2005), the difference between an integrator and coincidence detector can be viewed simplistically as the presence in the latter of a slow hyperpolarizing current at perithreshold potentials (Ratte et al., 2013). Typically, these currents are mediated by dendritic voltage- and Ca2+-dependent K+ channels that are activated by subthreshold depolarization evident during high levels of stochastic synaptic input (Destexhe et al., 2003). Given the considerable spontaneous activity in CA3, it can be assumed that these pyramidal cells receive a significant degree of continuous excitatory input and accordingly operate with coincidence detector traits. Such a scenario provides a plausible explanation for the low variability in the latency to CA1 spiking as well as the observation that the output is mediated by individual cells emitting single spikes (Ratte et al., 2013). The direct temporoammonic inputs to CA1, both excitatory and inhibitory (Amaral, 1993; Melzer et al., 2012), provide an attractive mechanism to modulate the operational mode of CA1 pyramidal cells, in a situation-specific manner, to influence throughput.
A prolonged and stereotyped CA3 output is critical for signal throughput
The above argument suggests that CA1 pyramidal cell activation requires temporally precise afferent input. In accord with this, single-pulse LPP stimulation produced a high-frequency (>200 Hz) spike output from CA3a (which innervates the CA1c subfield used for output measurements) lasting ∼20 ms. Spiking coincided with a slow wave exhibiting a dipole corresponding to that generated by direct activation of the recurrent collateral system. We conclude that the relay of signals from the cortex through CA3 is not mediated by the direct MF response but instead by a rapid build-up of local spiking driven by recurrent excitation. The amplification provided by this system offsets the step-down action associated with the sparse firing of DG granule cells (Jung and McNaughton, 1993; Leutgeb et al., 2007) and provides a mechanism for reliable throughput in the nontopographic hippocampal circuit. Such arrangements do, however, result in unexpectedly slow throughput across the circuit.
LPP activation elicited a more robust CA3 spike output than direct MF stimulation or spiking during spontaneous SPWs. The near synchronous activation of a small number of pyramidal cells across the CA3 proximodistal axis by the direct LPP (i.e., LPP-CA3) input may bias these neurons to preferentially respond to the delayed MF activation (i.e., indirect path) and mobilize the local recurrent activity necessary for CA1 spiking (Fig. 7a,b). The anatomical organization of the CA3 circuitry, and the unique spatiotemporal features this introduces, creates a complex series of parallel subcircuits that we propose drive throughput. Despite the slow MF conduction time, there were no reliable differences in the time to peak for recurrent responses recorded along the CA3 proximodistal axis. Associational fibers are comprised of dense local ramifications and rapidly conducting branches traveling across the full extent of CA3. The latter represents a serial circuit operating parallel to, but faster than, the MF. We suggest that long associational projections from those CA3 segments first activated by the DG “catch-up” with conduction along the unmyelinated MF as the latter travels along the proximodistal axis, shortening the delay to the onset of the recurrent response. Such arrangements will result in a near synchronous output from the full extent of CA3 to the different segments of CA1 (Fig. 7a).
Activation of field CA3 and hippocampal throughput mode. a, Schematic illustration (top) of CA3 pyramidal cells positioned across the proximodistal axis (i.e., CA3c-CA3b-CA3a) with the number and distribution of respective LPP (green), MF (blue), and C/A (gray) synapses depicted. Synapses are activated in a temporally distinct manner across the proximodistal axis of CA3 (bottom). Innervation of each subfield by the LPP terminals occurs in a near synchronous manner, while the parallel MF and C/A circuits produce a robust output across successive CA3 regions. b, Schematic illustrating the activation of CA3 pyramidal cells during throughput mode (top) and episodic mode (bottom). Concurrent activation of pyramidal cells by LPP (dark green) and MF (dark blue) inputs is required to promote local recurrent activation of neighboring cells (red). During throughput mode, a reliable output to CA1 is generated (red cells) although the likelihood of the same local population being activated across successive inputs is low. In contrast, in episodic mode, significant overlap in the population of cells spiking is observed across different pulses to support active encoding. Inputs from lower brain regions provide a likely candidate to drive the switch in operational modes. c, Schematic illustration of the proposed hippocampal throughput mode. Categorization of the relative familiarity of an experience or event is determined at the level of the neocortex and that signal relayed to the primary hippocampal circuit via the LEC. When exposed to a familiar experience or event LPP input arrives at theta or theta–gamma frequencies (top) driving a reliable signal back to deep layers of EC (i.e., layer V). In contrast, a novel environment produces gamma frequency (bottom) LPP inputs, which do not propagate through the circuit. A number of subfield-specific inputs from lower brain regions operate as gain control, having the capacity to switch the operational state of the circuit and modulate throughput. Note only the most prominent projections from the lower brain regions are depicted. Abbreviations: DA, dopamine; ACh, acetylcholine; OXT, oxytocin; AVP, vasopressin; LC, locus ceruleus.
The LPP-evoked CA3 response displayed unusually low variability in the intervals separating spikes within and between slices, a feature also observed following direct MF stimulation and during spontaneous SPWs. Such stereotypic CA3 responses could reflect the cycling time for the feedback collateral system connecting a local group of pyramidal neurons (Sik et al., 1993; Li et al., 1994; Witter, 2007a; Le Duigou et al., 2014). Indeed, the associational projection contains a high proportion of disynaptic motifs with high-fidelity transmission (Guzman et al., 2016). The ISI (5 ms) appears a realistic value for the time needed for a pyramidal cell to spike in response to input and then activate another cell. Stereotypy implies that signals are transmitted as packets with information encoded by a specific collection of active CA3 cells, not as spike patterns. Sparsification in the DG followed by amplification in CA3 recurrent collaterals reduces the likelihood of generating the same spatial pattern of responding cells to multiple applications of the same cortical input. These arrangements place considerable functional restrictions upon the circuit and its capabilities for performing complex operations.
The striking similarities between the LPP-evoked response in CA3 and spontaneously occurring SPWs are intriguing given the large literature on possible roles played by the latter in memory processing (Buzsaki, 2015). The evoked response reflects the activation of a small population of neighboring MF terminals, as indicated by a short latency field potential, and subsequent mobilization of recurrent activity by the associational system. Prior experimental and modeling work indicates that the great majority of SPWs occur when stochastic release from a local population of MF terminals occurs within a narrow time window and thereby drives a sufficient number of pyramidal cells to initiate recurrent activity (Rex et al., 2009). Such coordination of pyramidal cell spiking and subsequent mobilization of CA3 recurrent activity requires the recruitment of inhibitory circuits (Ellender et al., 2010; Schlingloff et al., 2014; Evangelista et al., 2020). In all, the singular arrangement of field CA3—a primary input composed of a very small population of potent terminals coupled to a massive feedback network—leads to the production of large spontaneous events that closely resemble evoked potentials. There are however pronounced differences at the circuit level. SPWs forming at any given point within CA3 propagate along the associational system throughout the subfield and into field CA1 as expected from the organization of CA3 projections (Kubota et al., 2003; Maier et al., 2003; Witter, 2007a; Schieferstein et al., 2024). In contrast, perforant path stimulation causes sequential activation of terminals formed by the slowly conducting MFs and thus serial activation of cells along the proximodistal axis of CA3 (Fig. 7a). We can also assume that the LPP projections to the distal dendrites of the pyramidal neurons will in some manner shape the processing of the evoked responses. While the response to LPP activation is stereotyped, the CA3 arrangements allow for diverse forms of transformation to signals arriving from the cortex and provide the capacity to generate remarkably long spiking responses to afferent inputs (Cox et al., 2019). On the other hand, SPWs, the incidence of which is influenced by behavioral state (Buzsaki, 2015), allow the hippocampus to propagate an aperiodic but standardized signal to downstream targets for prolonged periods (i.e., minutes long). As has been suggested, they could also serve to reactivate and thus potentially strengthen recently formed CA3 traces (Buzsaki, 2015; Colgin, 2016) as well modulating the function of lower brain structures (Tingley et al., 2021).
Hippocampal throughput is dependent upon the frequency and pattern of LPP activation
The propagation of theta frequency activity through the entire network, with minimal distortion, aligns with the idea that the rhythm serves as a carrier wave in the cortical telencephalon (Colgin, 2016). In contrast, the circuit rejected gamma frequency trains. Given that mobilization of the recurrent CA3 system by the indirect path is critical for driving throughput, the low-pass filter observed at the circuit level may be at least partially attributable to the operation of the LPP-DG synapse (Quintanilla et al., 2022). However, the observation that brief gamma bursts separated by 200 ms (theta–gamma) triggered a reliable CA1 output indicates that circuit-level operations cannot simply be extrapolated from the properties of any single synapse. This outcome was not predicted from monosynaptic activation of the LPP-DG terminal (Quintanilla et al., 2022) and highlights the importance of performing circuit-level analyses to determine how the functional properties of a given synapse (i.e., filter and amplifier) manifest within the context of a complex circuit.
The recruitment of local inhibitory interneuron circuits provides powerful input-specific and frequency-dependent mechanisms to further modulate signal processing within and between subfields (Mori et al., 2004; Witter, 2007a; Ewell and Jones, 2010; Boehringer et al., 2017; Geiller et al., 2023; Adaikkan et al., 2024). A feature of this hetrogeneous population of GABAergic cells is that each subtype innervates discrete subcellular domains on their respective pyramidal cell or interneuron targets (Freund and Buzsaki, 1996; Klausberger and Somogyi, 2008). Among other effects, these arrangements create a situation in which interneuron subtypes suppress target cell excitability via inhibition at sites proximal to the spike initiation zone or modulate dendritic signal processing (e.g., dendritic filtering and gain). These operations are known to be input-specific and frequency-dependent (Miles et al., 1996; Topolnik and Tamboli, 2022). Although the types and locations of the inhibition engaged during circuit throughput remain to be determined, it is reasonable to assume that the low-pass filtering described here is due in part to within-train recruitment of interneurons to rapidly suppress pyramidal cell spiking.
Relatedly, controlled mobilization of the CA3 recurrent circuit following MF activation and subsequent generation of the stereotyped CA3 spiking are likely dependent upon distinct interneuron circuits (Lawrence and McBain, 2003; Mori et al., 2004; Schlingloff et al., 2014). Feedforward cells in the dendritic lamina activated by feedback collaterals would shape the response [i.e., integrator and coincidence detector (Ratte et al., 2013)] of the pyramidal cell dendrites to the collaterals while basket cells would ensure that pyramidal cells do not emit strings of spikes. The combination of dense feedback and interneuron subtypes could in this manner produce a system that cycles at a steady rate and thereby generates an output signal appropriate for activating CA1. The direct LPP projection to CA3 triggers direct and indirect (i.e., feedforward) inhibition (Cosgrove et al., 2010; Melzer et al., 2012) as well as providing excitatory drive in the distal apical dendrites of the pyramidal cells. It remains for future work to determine if such actions bias the target cell response to subsequent input from the indirect path. These mechanisms may contribute to the relatively subtle alterations in spiking observed following theta and theta–gamma stimulation in the absence of the direct LPP pathway.
The finely balanced nature of the many and disparate circuit elements suggests that subtle changes in the operation(s) of specific circuit elements may have profound effects on the nature of signal transformations across the circuit. In reality, the primary hippocampal circuit will on occasion be activated by inputs from layer II lateral and medial EC, conveying semantic and spatial information, respectively, and likely occur at different frequencies (Igarashi et al., 2014). The nature of the interaction between these two projections will be critical for determining throughput. Additionally, direct excitatory and inhibitory inputs to CA1 (Melzer et al., 2012; Kitamura et al., 2014; Basu et al., 2016; Grienberger and Magee, 2022) provide a subfield-specific mechanism of potentially modulating throughput. While studies have primarily focused on the respective role(s) of these inputs during learning, it is tempting to speculate that the direct cortical inputs to CA1 may act to bias the likelihood of pyramidal cells responding to subsequent input from CA3. Lower brain structures also provide inputs to specific circuit elements that are engaged in a situation-specific manner. Inputs from the supramammillary nucleus (SuM) to DG and CA2 convey contextual and social novelty, respectively (Chen et al., 2020); how these inputs shape the propagation of cortical signals across the hippocampal circuit remains to be established. The spatiotemporal features of these additional inputs, in terms of the somatodendritic site(s) innervated, the nature of the signaling (i.e., ionotropic and metabotropic), and the relative timing and frequency of inputs, add a considerable level of complexity. Future studies will be required to test how select inputs shape the frequency-dependent signal transformations for the primary hippocampal circuit described here.
Implications for hippocampal memory processing
Results indicate that the circuit, in the stripped-down state found in a slice, operates as a low-pass filter such that cortical inputs arriving at theta or theta–gamma frequencies are reliably transmitted to downstream targets, while trains of gamma frequency input are rejected by the hippocampus producing a sparse output to downstream sites (Fig. 7c). The presence or absence of a robust output signal may represent a command such as “continue” or “stop,” respectively. We postulate that such basic operations constitute a “default” mode that would be employed in familiar environments (i.e., no active encoding) where the default signal serves to confirm the anticipated environment or events (Fig. 7c). Activation of previously encoded place fields, which are disrupted by suppression of the CA3 associational system (Davoudi and Foster, 2019), is an example of such a scenario. However, this does suggest that the reactivation of cortical ensembles (i.e., engrams) encoding the information in question occurs at the appropriate frequency to convey the required signal (i.e., “continue” or “stop”) across the circuit.
The stereotyped CA3 response elicited by the LPP activation ensures a reliable signal in the low-order hippocampal circuit, but does not intuitively support the more complex operations (e.g., plasticity) required for episodic memory. Indeed, the variability in the spatial pattern of activated CA3 cells across successive LPP pulses suggests innervation of CA1 pyramidal cells will be unlikely to meet the requirements for encoding [i.e., long-term potentiation (LTP)]. Given the indirect path (i.e., LPP-DG-CA3) is critical for episodic memory in rodents (Cox et al., 2019), it would appear that the circuit has the capacity to operate across multiple modes as required (Fig. 7b,c). The direct LPP-CA3 input provides a more reliable cortical signal to CA3 that may shift the response to indirect activation (by the same cortical input) toward a single population of pyramidal neurons. However, given that the threshold for LTP at the LPP-CA3 terminal, despite being lower than the LPP-DG, is markedly higher than the SC-CA1 synapse (Le et al., 2022; Quintanilla et al., 2024), it remains to be determined if this mechanism is physiologically plausible in the context of episodic (i.e., one shot) learning. Alternatively, the operational state of the CA3 recurrent system itself may be a critical factor. Prior work indicates that activation of the CA3 associational projections can trigger prolonged periods of reverberating activity (Cox et al., 2019), approximating recurrent networks hypothesized to address the problem of associating temporally distinct cues. Such activity is critical for the acquisition of the temporal order of serial cues (Cox et al., 2019), yet how such a change in CA3 state influences throughput is unknown. It would seem plausible that increasing depolarization within the recurrent CA3 circuit would not only shift the spike initiation dynamics governing signal throughput but may also induce plasticity-related mechanisms within this local circuit. However, as activation of reverberating activity occurred in <40% of trials with no suggestion that it can be initiated by cortical input, the minority inputs from the lower brain may be needed to shift the operational mode of the circuit such that cues elicit prolonged firing and sequences are encoded.
Footnotes
- Received January 17, 2024.
- Revision received October 18, 2024.
- Accepted October 28, 2024.
-
This work was supported in part by NIH NICHD grant HD101642 and ONR grants N00014-18-1-2114, N00014-21-1-2940, and BCS-1941216.
-
The authors declare no competing financial interests.
- Correspondence should be addressed to Benjamin G. Gunn at bggunn{at}uci.edu or Gary Lynch at ga.s.lynch{at}gmail.com.
- Copyright © 2024 the authors
