|Home | About | Journals | Submit | Contact Us | Français|
Neurons propagate information through circuits by integrating thousands of synaptic inputs to generate an action potential output. Inputs from different origins are often targeted to distinct regions of a neuron’s dendritic tree, with synapses on more distal dendrites normally having a weaker influence on cellular output compared to synapses on more proximal dendrites. Here we report that hippocampal CA2 pyramidal neurons, whose function has remained obscure for 75 years, have a reversed synaptic strength rule. Thus, CA2 neurons are strongly excited by their distal dendritic inputs from entorhinal cortex but only weakly activated by their more proximal dendritic inputs from hippocampal CA3 neurons. CA2 neurons in turn make strong excitatory synaptic contacts with CA1 neurons. In this manner CA2 neurons form the nexus of a highly plastic disynaptic circuit linking the cortical input to the hippocampus to its CA1 neuronal ouput. This circuit is likely to mediate key aspects of hippocampal-dependent spatial memory.
Recent findings raise fundamental questions about the circuit through which the hippocampus processes information from its entorhinal cortex (EC) inputs to store and propagate spatial memories. Most previous studies have focused on the trisynaptic path (van Strien et al., 2009), in which layer II (LII) neurons of the EC initially excite dentate gyrus (DG) granule cells, which then activate CA3 pyramidal neurons, which in turn excite CA1 pyramidal neurons, the major hippocampal output. Genetic approaches have demonstrated the importance of long-term potentiation of synaptic transmission (LTP) (Neves et al., 2008) at each stage of the trisynaptic circuit (McHugh et al., 2007; Otto et al., 2001; Tsien et al., 1996). However, it has recently been found that CA1 neurons show high levels of spiking in vivo and preserve key aspects of hippocampal-dependent spatial memory even when their CA3 inputs have been disrupted (Brun et al., 2002; Nakashiba et al., 2008).
The residual hippocampal function in the absence of CA3 inputs has been proposed to depend on the direct inputs from EC LIII neurons to CA1 neurons, which bypass the trisynaptic path (Brun et al., 2002; Nakashiba et al., 2008). However, whereas CA1 neurons are strongly excited by their CA3 inputs, which terminate on proximal CA1 dendrites, CA1 neurons are only weakly excited by their LIII inputs, which are targeted to distal CA1 dendrites and, thus, greatly attenuated by the dendritic cable (Golding et al., 2005; Spruston, 2008). As a result, LIII inputs are generally thought mainly to modulate CA1 function (Dudman et al., 2007; Golding et al., 2002; Judge and Hasselmo, 2004; Levy et al., 1998; Remondes and Schuman, 2002; Takahashi and Magee, 2009) and shape memory storage (Brun et al., 2008; Remondes and Schuman, 2004), rather than to trigger CA1 spiking. The difficulty in explaining how weak LIII inputs can drive CA1 output in the absence of CA3 input prompted us to reexamine key aspects of the hippocampal circuit, focusing on CA2 pyramidal neurons.
First characterized by (Lorente de No, 1934), the CA2 region comprises a small population of neurons between CA3 and CA1 whose function remains largely unexplored. Although CA2 neurons are generally assumed to form a minor pathway linking CA3 to CA1 (Sekino et al., 1997), in vivo stimulation of the inputs to EC can trigger the initial firing of spikes in the CA2 region, followed by CA1 and then CA3 (Bartesaghi and Gessi, 2004), suggesting that CA2 neurons may mediate a disynaptic link from EC to CA1 (Bartesaghi et al., 2006). However, it is difficult to understand how EC synapses could drive CA2 firing as these inputs terminate on distal CA2 dendrites, similar to CA1. Moreover, given the small number of CA2 neurons and the weak unitary connections between most hippocampal neurons, it is unclear how CA2 neuron input could elicit CA1 output. Finally, the finding that CA3–CA2 synapses fail to undergo normal LTP (Simons et al., 2009; Zhao et al., 2007) raises questions about the role of CA2 neurons in memory storage.
Here, we used whole cell recordings in acute hippocampal slices to overcome the limitations of extracellular recordings and directly probe the role of CA2 neurons in the hippocampal circuit. We find that CA2 neurons operate under an unusual, reversed rule of synaptic drive. Thus, CA2 neurons receive uniquely strong convergent excitatory inputs from LII and LIII EC neurons on their distal dendrites. These synapses undergo robust LTP and efficiently drive CA2 firing. In contrast, CA3 inputs onto proximal CA2 neuron dendrites are relatively weak, dominated by feed-forward inhibition, and fail to undergo LTP. Finally, CA2 neurons make unusually strong unitary connections with CA1 neurons, forming a potent disynaptic circuit, independent of the trisynaptic path, that is powerfully regulated by long-term synaptic plasticity. Thus, through the fine-tuning of synaptic weights, CA1, CA2 and CA3 pyramidal neurons mediate distinct patterns of information flow through the hippocampus that are likely to play specific roles in spatial learning and memory.
To date there have been few detailed electrophysiological characterizations of CA2 pyramidal neurons and the data that is available is based on studies of immature rats (Zhao et al., 2007). Therefore, we first characterized the electrophysiological properties of CA2 pyramidal neurons and compared their properties to those of CA1 and CA3 pyramidal neurons in slices from adult mouse hippocampus. Under DIC optics (Fig. S1) we visually identified the CA2 area as the region with the thickest somatic layer located between CA3, characterized by its distinct mossy fiber terminal layer, and CA1, characterized by its compact cell body layer.
CA2 pyramidal neurons identified in the above manner displayed a series of electrophysiological properties that were distinct from those of CA1 or CA3 pyramidal neurons. Thus, the CA2 neuron input resistance was lower and the membrane capacitance was higher compared to CA1 or CA3 neurons (see Table 1). Moreover, whereas CA3 neurons fired a burst of action potentials at the beginning of a depolarizing current step, CA1 and CA2 neurons fired during the entire depolarization (Fig. 1A). CA2 neurons were further distinguished from CA1 neurons because the prominent slow after-hyperpolarization present in CA1 neurons was absent in CA2 neurons (Fig. 1A). Finally, the depolarizing sag in response to a hyperpolarizing current pulse, which is caused by the hyperpolarization-activated cation current (Ih), was much larger in CA1 neurons than in CA2 neurons (Fig. 1B and Table 1). This difference is consistent with the higher levels of expression of the HCN1 subunit, which underlies Ih, in CA1 neurons compared to CA2 or CA3 neurons (Notomi and Shigemoto, 2004; Santoro et al., 2004). A small proportion of recorded neurons in CA2 (~10%) had different electrophysiological properties than those described above and were not included in this study. These cells were likely to be interneurons as they displayed similar firing properties to those described previously for CA2 interneurons (Mercer et al., 2007).
To confirm the identity of the putative CA2 pyramidal neurons from which we obtained the electrophysiological recordings, we included a fluorescent dye in the patch pipette to label cells and performed a post hoc immunolabeling for α-actinin 2, a protein enriched in CA2 pyramidal neurons (Wyszynski et al., 1998). All labeled neurons that displayed the distinctive electrophysiological properties described above (as well as the characteristics described in the following figures) had the typical dendritic branching pattern of CA2 pyramidal neurons described previously (Ishizuka et al., 1995). Thus, the apical dendrite bifurcated close to the soma (31.1 ± 3.6μm, n = 19) into two or three main branches that extended to stratum lacunosum moleculare (SLM) with few secondary, oblique branches in stratum radiatum (SR) but many branches in SLM (Fig. 1C and. Fig. S1). This is in contrast with the morphology of CA1 neurons, which have a main apical dendrite that either did not bifurcate within SR or bifurcated far from the soma (107.6 ± 11.1 μm, n = 14). Moreover, the main apical dendrite of CA1 neurons had a large number of secondary oblique dendrites in SR and few thin branches in SLM (Ishizuka et al., 1995). In addition, CA2 neurons had a much larger soma than CA1 neurons (surface area of 391 ± 22 μm2 in CA2 versus 179 ± 11 μm2 in CA1). CA2 neurons could be further distinguished from CA3 neurons in that CA2 (and CA1) neuron dendrites branch along the transverse axis of the hippocampus (parallel to the plane of the slice) whereas CA3 neurons mostly extend their dendrites along the longitudinal axis of the hippocampus (Ishizuka et al., 1995).
Together, these data show that CA2 neurons display unique electrophysiological and anatomical features, and suggest that they might be engaged differently by synaptic inputs compared to CA3 or CA1 neurons. Therefore, we next investigated the strength of cortical versus intrahippocampal inputs onto CA2 neurons, and compared these synaptic weights with those onto CA1 and CA3 neurons.
To compare the efficacy of the LIII EC synaptic inputs in depolarizing CA1 versus CA2 neurons, we placed a focal stimulating electrode in the SLM region of CA1 and obtained whole cell recordings of the somatic EPSPs from CA1 and CA2 neurons located approximately equidistant from the stimulating electrode (Figure 2A). Most CA1 neurons studied were located midway along the transverse axis of CA1, but similar results were obtained for CA1 neurons close to the subiculum or CA2 (see below). As previously reported (Empson and Heinemann, 1995b), focal stimulation of the LIII EC axons (perforant path) in SLM elicited only very small EPSPs in the CA1 neuron soma, even with inhibitory transmission blocked by GABAA and GABAB receptor antagonists (Fig. 2A). Surprisingly, the same stimulation had a much stronger impact on CA2 pyramidal neurons, generating somatic EPSPs up to 5 times larger than those in CA1 (Fig. 2A). Thus a moderately strong 20 V extracellular stimulus elicited a mean EPSP of only 1.9 ± 0.4 mV (n = 8) in CA1 neurons versus an EPSP of 9.4 ± 1.2 mV (n = 10) in CA2 neurons ( p<0.001).
One potential concern with such results is that the difference in excitatory drive is an artifact caused by differences in the extent to which the EC inputs to the CA1 and CA2 regions are preserved in the hippocampal slice preparation or activated by the focal stimulating electrode. However, three lines of evidence argue that the difference is genuine. First, the extracellular field potential (fEPSP) elicited by the distal focal stimulation, an index of the local synaptic current density, was identical in simultaneous recordings from the SLM region of CA1 and CA2 (Fig. 2B), indicated that we activated an equal synaptic input. Second, the extracellular presynaptic fiber volley isolated after blocking synaptic transmission (which reflects the number of activated axons) was also identical in CA1 and CA2 (data not shown). Third, to minimize any potential differences due to the location in the slice of the recorded cells, we examined CA1 neurons at the border with CA2. These CA1 neurons had the typical electrophysiological features and morphology of neurons in regions of CA1 more distal to CA2 (Fig. S2A). Despite their proximity to CA2, such CA1 neurons also displayed very small EPSPs in response to LIII EC stimulation, similar to the EPSPs seen in more distal regions along the transverse axis of CA1 (Fig. S2B). Although thalamic input from nucleus reuniens also runs in SLM (Wouterlood et al., 1990) and may contribute to the EPSP evoked by electrical stimulation in SLM, such inputs are unlikely to contribute to the large CA2 EPSP as the density of thalamic fibers diminishes abruptly at the transition from CA1 to CA2.
As stimulation of LIII inputs engages a strong feed-forward inhibition to CA1 neurons (Empson and Heinemann, 1995a), we examined whether the increased excitatory drive in CA2 might be offset by a large feed-forward inhibitory response. However, when we measured the net synaptic response with inhibition intact (no GABA blockers), the distinction between CA2 and CA1 became even more marked. Although LIII stimulation still evoked a large net depolarizing response in CA2 neurons, there was now often a net hyperpolarizing response observed in CA1 neurons (Fig. 2C). The hyperpolarization in CA1 was not due to a lack of excitatory inputs as all CA1 neurons exhibited a clear EPSP in response to LIII stimulation in the presence of GABA blockers (n=24 out of 24 neurons).
We isolated the underlying inhibitory postsynaptic potential (IPSP) from the net synaptic response by subtracting the synaptic response in the presence of GABA blockers (pure EPSP) from the response in the absence of blockers (combined EPSP and IPSP). The deduced IPSP in CA2 neurons was similar in size to that in CA1 neurons. Thus, the IPSP (elicited by a 20 V stimulus) was −2.3 ± 0.4 mV in CA1 (n = 8) versus −3.7 ± 1.2 mV in CA2 (n = 8; p = 0.26; Fig. 2C). However, the large CA2 EPSPs ensured the presence of a net depolarizing response even when inhibition was intact, with a 3–4 fold ratio of excitation to inhibition (Fig. 2D). In contrast, the small size of the EPSPs in CA1 neurons resulted in an EPSP/IPSP ratio close to (or slightly less than) one.
CA1 neurons receive their major excitatory input from the CA3 Schaffer collateral inputs (SC), whose synapses terminate on proximal dendrites of both CA2 and CA1 neurons in SR. As the SC synapses provide a much stronger excitatory drive onto CA1 neurons than do the LIII inputs, we expected that the SC inputs would potently excite CA2 neurons. Instead, we found that the SC EPSPs (with inhibition blocked) were significantly smaller in CA2 neurons than in CA1 neurons. Thus the SC EPSP (elicited by a 16 V stimulus) was equal to 17.5 ± 2.6 mV in CA1 (n = 8) versus 10.8 ± 2.2 mV in CA2 (n = 7; p = 0.008, Fig. 3A). The larger SC intracellular EPSP in CA1 versus CA2 neurons is likely to result from a larger excitatory synaptic current as the local fEPSP in the SR region of CA1 was larger than that in CA2 (Fig. 3B), despite similar fiber volleys in both areas (not shown). The larger SC synaptic response in CA1 versus CA2 is consistent with the greater number of oblique dendritic branches in the SR of CA1 compared to CA2 neurons and the more extensive branching of CA3 axons in CA1 versus CA2, suggestive of a greater number of synaptic contacts.
In contrast to the difference in EPSP size, IPSPs evoked by SC stimulation were similar in CA1 and CA2 neurons (Fig. 3C). Thus, activation of SC inputs (by a 16 V stimulus in SR) elicited an IPSP of −12.6 ± 2.9 mV in CA1 (n = 8) versus an IPSP of −13.8 ± 2.4 mV in CA2 (n = 7; p > 0.5). As a result, SC stimulation produced a large net depolarization in CA1 neurons (EPSP/IPSP ratio ~2), whereas inhibition prevailed over excitation in CA2 (EPSP/IPSP < 1) (Fig. 3D). Altogether, these data show that CA1 and CA2 pyramidal neurons display a reversed net synaptic drive along their apical dendrites in response to stimulation of their EC versus SC inputs that is caused by differences in the strength of the underlying EPSPs.
What are the consequences of the differences in somatic EPSPs following stimulation of the LIII or SC inputs in driving the firing of CA1 and CA2 neurons? To address this question, we activated these two inputs with a brief burst of stimuli (5 at 100 Hz) and measured action potential firing in CA1 and CA2 neurons. With inhibition intact, LIII stimulation produced a net hyperpolarization and failed to elicit a spike in CA1 neurons. In contrast, LIII stimulation produced a net depolarization in CA2 neurons that was sufficient to trigger one or more spikes in 20% of cells tested (n = 3/15, Fig. 4A,B). Following block of inhibition, the burst of LIII stimuli caused a larger depolarization that triggered spikes in 80% of CA2 neurons (n = 12/15) but remained subthreshold for nearly all CA1 neurons (only 1 in 12 fired a spike; Fig. 4C). We obtained similar results using extracellular field recordings of the population spike (PS) in the CA1 and CA2 somatic layers: single or paired stimulation of LIII inputs consistently evoked a PS in CA2 but not in CA1, where only a small field EPSP was present. Blocking inhibition greatly increased the PS amplitude in CA2 but failed to lead to a PS in CA1 (not shown). These field recording results indicate that the differential ability of LIII inputs to drive firing in CA2 but not CA1 is genuine and not an artifact of the whole-cell recordings.
We next compared the ability of a burst of SC stimulation to drive firing of CA1 versus CA2 neurons. As expected, the SCs exerted a powerful excitatory drive in CA1, triggering spikes in over 1/3 of neurons tested (n = 4/11, Fig. 4D,E). In contrast, SC stimulation never led to firing in CA2 (n = 0/8) but rather elicited a large net hyperpolarizing response (Fig. 4D,E). Blocking inhibition greatly increased the efficiency of spike firing in CA1 neurons, with 100% of neurons firing spikes (11/11), and also enabled the SC inputs to trigger spikes in the majority of CA2 neurons (75%, n = 6/8, Fig. 4F). This suggests that SC inputs on their own are strong enough to drive firing in CA2 pyramidal neurons, but their impact is tightly controlled by feed-forward inhibition.
LII EC axons form excitatory synapses on granule cells in the molecular layer of dentate gyrus (DG), then project onto the distal dendrites of CA3 neurons, and finally terminate in a U shaped-area near the distal dendrites of CA2 (Niimi et al., 2007; Nishimura-Akiyoshi et al., 2007; Tamamaki and Nojyo, 1995). To examine whether LII axons also form synapses on CA2 neurons, we excited LII inputs with a focal stimulating electrode in the middle molecular layer of DG. Activation of LII inputs with a moderate stimulus produced surprisingly large EPSPs in CA2 neurons with little synaptic delay, indicating they were monosynaptic responses. Stronger stimulation of LII inputs evoked a large inhibitory response, and addition of GABA blockers unmasked polysynaptic EPSPs (Fig. 5A). This polysynaptic response likely resulted from the recruitment of the DG→CA3→CA2 pathway either by activation of LII inputs to granule cells or by their direct activation as it was blocked by the mGluR-II agonist DCG-IV, which produces potent presynaptic inhibition at the mossy fiber synapses from DG to CA3 neurons but does not alter SC EPSPs (Kamiya et al., 1996) (Fig. 5A).
Extracellular recordings of field potentials in the SLM of CA2 provided further support that the field stimulation in DG was able to recruit both monosynaptic LII→CA2 excitatory inputs as well as the DG→CA3→CA2 pathway. Evidence for the presence of a monosynaptic response from LII inputs is based on the fast negative voltage response in SLM (fEPSP), consistent with the presence of a local inward excitatory synaptic current (current sink). With stronger LII stimulation, this negative field response was followed by a late positive voltage response (Fig. 5A). The reversed direction of the late response indicates that it does not originate at the site of the recording but likely reflects a local outward current in the distal dendrites (current source) due to an inward excitatory synaptic current sink generated by the SC inputs onto more proximal regions of the CA2 dendrite. Again, only the late response was abolished by application of DCG-IV (Fig. 5A). The LII inputs also elicited direct EPSPs in CA3 neurons. However, CA3 EPSPs were less than half the size of CA2 responses. A 20 V stimulus intensity elicited an EPSP of 2.8 ± 0.8 mV in CA3 neurons (n = 6) versus 7.4 ± 0.9 mV in CA2 (n = 8; p<0.003; Fig. 5B).
To confirm the strong connection between the cortex and CA2, as well as the selectivity of the stimulation of layer II and layer III EC inputs, we obtained extracellular field potential recordings from different regions along the transverse axis of the cell body layer of the hippocampus in the CA1, CA2 and CA3 regions. Upon focal stimulation in either SLM of CA1 or the molecular layer of DG, the largest field potential response was observed in the CA2 cell body area (Fig. 5C). DG stimulation produced a smaller but detectable field response in CA3 with little or no response in CA1. In contrast, stimulation in SLM of CA1 produced a sizable response in CA1 with little response in CA3. These data are consistent with the known anatomical projections of LII and LIII EC axons and confirm that our stimulating electrodes in DG and CA1 specifically recruited LII and LIII inputs, respectively. Moreover, these inputs produced their largest synaptic response in CA2.
As both LII and LIII neurons fire during theta activity (Hafting et al., 2008), we wondered whether these inputs converge on the same CA2 pyramidal neuron. Indeed, all CA2 neurons exhibited EPSPs in response to independent stimulation of both LII and LIII inputs (n = 20). When activated together, LII and LIII EPSPs showed linear summation for small amplitude EPSPs, whereas larger amplitude EPSPs displayed sublinear summation (Fig. 6A). This likely resulted from saturation as the distal EPSPs approached their reversal potential, and not from recruitment of an inhibitory component, as similar results were observed with inhibition blocked (data not shown).
During theta activity, LII neurons fire earlier in successive cycles (phase precession), while LIII neurons fire randomly or at a fixed phase during the cycle (Hafting et al., 2008). Thus, both layers rarely fire synchronously, but are activated within tens of milliseconds of each other. When stimulated at different intervals, the two inputs summated over a symmetrical time window that lasted around 20 ms and showed a maximum summation when activated simultaneously (Fig. 6B). Blockade of inhibition increased summation when LII was activated first, likely because LII stimulation also recruited CA3 inputs to CA2 (Fig. 6B). These results show that both LII and LIII converge onto CA2 neurons where they interact and exert their maximal effect throughout the hippocampus.
If CA2 pyramidal neurons are part of a disynaptic cortico-hippocampal loop important for memory formation, then we might expect to observe robust LTP at the EC-CA2 synapses. However, previous studies using whole cell recordings reported that LTP could not be induced at the SC-CA2 synapses under normal conditions (Simons et al., 2009; Zhao et al., 2007). As it is possible that the whole cell recording configuration led to a washout of LTP, we investigated whether LTP could be induced at SC–CA2 synapses using extracellular recordings of CA2 fEPSPs. However, our results confirmed the previous whole-cell findings as high-frequency tetanic stimulation (HFS) of the SC inputs led to only a small and transient potentiation of the CA2 fEPSPs (Fig. 7A). In contrast, and as expected, a large LTP was observed for the SC fEPSPs in CA1 neurons (Fig. 7A. 153.5 ± 13.0% [n = 9] in CA1; 108.9 ± 4.2 % [n = 9] in CA2; p = 0.006).
In contrast to the lack of LTP at CA3–CA2 synapses, we observed a surprisingly large and robust LTP of LIII CA2 fEPSPs (162.5 ± 9.1%, n = 9). Moreover, this LTP was also much greater than the weak LTP observed at the LIII inputs to CA1 (127.1 ± 5.4%, n = 7; p = 0.01; Fig. 7B). The induction of LTP at LIII synapses in both CA1 and CA2 was blocked by the NMDA receptor antagonist APV, indicating these two processes share certain molecular mechanisms (Fig. S3A). We confirmed that the large LTP of the fEPSP was actually due to potentiation of synaptic responses in CA2 pyramidal neurons using whole cell recordings (Fig. S3B, 158.7 ± 5.2%, n = 6). Finally, induction of LTP at the LIII inputs produced a large, ~250% increase in the amplitude of the CA2 population spike elicited by strong stimulation (Fig. S4A), indicating an enhancement in the efficiency with which cortical inputs drive CA2 firing.
The LII EC synapses with CA2 neurons generated a correspondingly large LTP (194.3 ± 9.8% increase in the fEPSP, n = 6) that was more than two-fold greater than the amount of LTP at LII synapses in CA3 (135.1 ± 9.3%, n = 6, p = 0.0008, Fig. 7C). LTP at the LII –CA2 synapses was not altered in slices in which the mossy fiber inputs were cut, or with DCG-IV present to block glutamate release from the mossy fiber terminals, indicating that the LTP did not require activation of CA3 neurons (Fig. S4B). Although LII and LIII EC inputs to CA2 terminate in close proximity in SLM, LTP was pathway specific because LIII EC inputs displayed only a small and transient increase in efficacy when the HFS was delivered to LII EC inputs (Fig. S4C).
If CA2 pyramidal neurons are indeed part of a powerful disynaptic pathway that drives hippocampal output, the CA2 neurons might be expected to make strong unitary excitatory connections with CA1 pyramidal neurons. Anatomical reconstructions have shown that CA2 pyramidal neurons send their axons to both SR and stratum oriens (SO) of CA1 (Tamamaki et al., 1988), whereas ipsilateral CA3 neurons only send their axons to SR. However, the postsynaptic target of CA2 neurons has not been verified and a synaptic connection between CA2 and CA1 pyramidal neurons has never been reported. To address this issue, we first determined whether CA2 axons reached area CA1 in the mouse hippocampal slices using a functional approach. Local stimulation of fibers in either the SR or SO regions of CA1 routinely was able to elicit antidromic spikes in the somatic layer of CA2. In contrast, an antidromic spike was only elicited in CA3 neurons when the stimulating electrode was placed in SR, but not in SO, of CA1. These results are consistent with the known anatomical innervation patterns of CA2 and CA3 axons and confirm that CA2 axons do indeed project to both SR and SO of CA1 (Fig. 8A).
To determine whether CA1 pyramidal neurons are innervated by CA2 neuron axons and to measure the strength of such connections, we performed paired recordings between single CA2 and CA1 neurons (that is, unitary synaptic connections). As reported for CA3–CA1 connections (Bolshakov and Siegelbaum, 1995), the probability of connectivity between CA2 and CA1 neurons was very low in transverse slices (~5%, 5 out of ~100 pairs tested), limiting our ability to obtain detailed data. Nonetheless, we were able to characterize the synaptic properties of three of the five connected pairs using dual whole-cell recordings and two of the connections using whole-cell recordings from a CA1 neuron and extracellular stimulation of a single CA2 neuron through a cell-attached patch pipette. To improve the signal-to-noise ratio, we recorded excitatory postsynaptic currents (EPSCs) from CA1 neurons under whole-cell voltage clamp conditions.
As expected for quantal synaptic transmission, successive presynaptic stimuli either elicited unitary EPSCs of variable amplitude (successes) or failed to trigger an EPSC (failures). Histograms of the EPSC amplitude distribution showed a prominent peak at 0 pA, representing the failures, and a broad distribution of EPSC successes clearly resolvable from the peak of failures. The probability of a CA2–CA1 EPSC success ranged between 0.4 and 0.6 (Fig. 8D), similar to the value at CA3–CA1 synapses (Bolshakov and Siegelbaum, 1995). However, the potency (average amplitude of EPSC successes) was about two-fold larger at CA2–CA1 synapses, around −8 pA (Fig. 8D), than previously reported for CA3–CA1 synapses, around−4 pA (see (Bolshakov and Siegelbaum, 1995; McMahon et al., 1996). In addition, whereas a single peak of EPSC successes was reported for CA3–CA1 synapses, the CA2–CA1 EPSCs always showed multiple peaks of successes, which occurred at equally spaced intervals of around −5 pA (Fig. 8B). In one pair of CA2 and CA1 neurons, we observed five distinct peaks of equally spaced EPSC successes (Fig. 8C). The multiple peaks of successes may reflect the simultaneous release of multiple quanta of transmitter, either at a single synaptic contact or at independent synapses. In this case, the release probability at an individual synapse will be lower than the typical success probability.
Given the strong unitary connection between individual CA2 and CA1 neurons, we next asked whether CA2 neurons, despite their small number compared to CA3 neurons, might provide a sufficient synaptic drive to fire CA1 neurons, thus completing a putative cortico CA2 CA1 loop. We applied a brief train of stimuli (2–5 pulses) to LIII axons to elicit spikes in CA2 neurons (Fig. 9A). As described above, moderate intensity stimulation of LIII axons ( 15 V) evoked only a small, direct EPSP in the CA1 neurons. However, when we increased the stimulus strength to 20–40 V, a large, late polysynaptic EPSP was elicited in the CA1 neurons that eventually reached threshold for firing an action potential (Fig. 9A). This polysynaptic response was only observed when the LIII stimulation elicited spikes in CA2 neurons, as indicated by simultaneous extracellular recordings of the PS in the CA2 somatic layer (Fig. 9A). Furthermore, the polysynaptic response in CA1 was present when slices were cut between CA3 and CA2 in 4/6 experiments, indicating that CA3 was not involved in this process. However, polysynaptic responses and action potentials were never evoked in CA1 neurons when slices were cut between CA2 and CA1, even with stronger LIII stimulation (up to 60 V, n = 8, Fig. 9B). Thus, our results in hippocampal slices demonstrate that CA2 neurons can be driven by their cortical inputs to elicit the firing of CA1 neurons, enabling cortical input to drive hippocampal CA1 output in a disynaptic loop.
Our results in hippocampal slices firmly integrate CA2 pyramidal neurons within the cortico-hippocampal circuit and provide four main conclusions. First, CA2 neurons receive uniquely strong, convergent excitatory input from LII and LIII EC neurons, the only site of such convergence in the hippocampus. Second, CA2 neurons strongly excite CA1 pyramidal neurons through potent excitatory synaptic connections. Third, both the LII and LIII synapses with CA2 neurons undergo strong LTP, enhancing the efficacy with which these inputs fire CA2 neurons. Fourth, the inputs to CA2 neurons from CA3 neurons recruit strong feed-forward inhibition. Altogether, our findings in acute slices reveal that CA2 neurons are a major target of excitatory cortical input to the hippocampus and can potently excite their CA1 neuron targets. In contrast, CA2 neurons are largely inhibited by their intra-hippocampal input from CA3 pyramidal neurons. The view that CA1 neurons are only weakly excited by their direct EC inputs but can be driven by a disynaptic loop through CA2 has been previously inferred from in vivo recordings of hippocampal field potentials in response to stimulation of the commissural inputs to EC (Bartesaghi and Gessi, 2004; Bartesaghi et al., 2006). The good agreement between these in vivo recordings and our findings indicates that the differential synaptic responses we observe among CA1, CA2 and CA3 neurons are genuine and not an artifact of the hippocampal slice preparation.
In addition to the unexpectedly large excitatory drive that the distal EC inputs exert on CA2 neurons, we find that the SC inputs from CA3 neurons recruit a very large feedforward inhibition onto the CA2 neurons. This inhibition is likely to be important in gating information flow between CA3 and CA2 and may explain why CA2 neurons do not normally mediate a significant quadrisynaptic loop from EC to DG to CA3 to CA2 to CA1. Because this strong inhibition is initially engaged by the same input that directly activates CA2 (layer II input), it will preserve independence between the di- and the triynaptic loops. The strong inhibition might also terminate CA2 neuron firing after its initial activation by the disynaptic loop, and thus prevent excessive activation of CA1 neurons. The relevance of this inhibitory drive is suggested by the findings that selective decreases in CA2 inhibition are associated with schizophrenia (Benes et al., 1998) and epilepsy (Williamson and Spencer, 1994), two pathological conditions in which hippocampal excitability is increased.
Our findings also demonstrate how the synaptic drive of the same class of LIII EC inputs onto neighboring CA1 and CA2 neurons can be tuned to differentially route information through the hippocampal circuit. The weak excitatory drive of the cortical inputs to CA1 neurons likely restricts these inputs to a modulatory role in regulating CA1 output, for example, providing instructive signals for the induction of plasticity at the more potent CA3 inputs (Dudman et al., 2007; Judge and Hasselmo, 2004; Levy et al., 1998; Remondes and Schuman, 2002). In contrast, the strong excitatory drive of the cortical inputs to CA2 neurons, combined with the unusually powerful unitary synaptic connection between CA2 and CA1 neurons, forms a robust disynaptic circuit that may be sufficient to drive CA1 neuron output to mediate key hippocampal functions in spatial learning and memory.
Because of the cable properties of neuronal dendrites, excitatory synaptic inputs terminating on the distal regions of the dendritic tree usually provide a much weaker excitatory drive at the soma compared to inputs that terminate on more proximal regions of the dendritic tree. This normal pattern of synaptic drive is apparent in CA1 pyramidal neurons, where activation of the distal synaptic inputs from EC produces a much smaller somatic EPSP compared to activation of the more proximal SC inputs. In contrast, CA2 neurons display a reversed pattern of synaptic strength, with the distal EC inputs providing a much larger excitatory drive compared to the proximal SC inputs. This reversed excitatory drive in CA2 versus CA1 is due to two factors. First, the somatic EPSPs elicited by distal (EC) stimulation are five-fold larger in CA2 versus CA1 neurons. Second, the somatic EPSPs elicited by proximal (SC) stimulation are smaller in CA2 than CA1 neurons.
The difference in size of somatic EPSPs in CA1 versus CA2 neurons elicited by distal EC stimulation is likely to reflect intrinsic differences in the extent to which the distal dendrites of the two neurons integrate and attenuate the local EPSP as it propagates to the soma, rather than to differences in the efficacy of local synaptic transmission. Thus, we found that the distal synaptic current density (measured by the fEPSP) in response to distal stimulation in SLM is identical in CA1 and CA2. Moreover, whereas CA1 dendrites attenuate distal EPSPs by a factor of 50–100 at the soma (Golding et al., 2005; Spruston, 2008), the CA2 dendritic cable can produce at most a 7-fold attenuation of the distal EPSP, given that the CA2 somatic EPSP in response to LIII stimulation is up to 10 mV in amplitude and the distal EPSP can be at most 70 mV (the difference between the EPSP reversal potential and the resting potential). In principle, the enhanced propagation of the distal EPSP to the CA2 neuron soma may also reflect active properties of the CA2 dendrites to amplify the distal EPSPs. However, the presence of sublinear summation between layer II and layer III EPSPs argues against this possibility.
Additional studies will be needed to elucidate the dendritic mechanisms underlying the differences in the propagation of EPSPs in CA2 and CA1 neurons. However, certain qualitative differences are already apparent in the structure and functional properties of the dendrites of these two neurons that might contribute to their differential attenuation of the distal EPSPs. First, whereas CA1 neurons have a large number of oblique secondary dendrites in SR that shunt the synaptic current as it propagates down the dendrite, CA2 neurons have very few oblique dendrites in SR. Second, distal EPSPs in CA1 neurons are locally attenuated by the active properties of the dendrites, including the presence of a very high density of HCN1 channels in the distal dendrites, which act as a depolarizing shunt conductance (George et al., 2009; Magee, 1999). In contrast, immunocytochemical light micrographs show that HCN1 channels fail to accumulate in the distal CA2 dendrites (Notomi and Shigemoto, 2004; Santoro et al., 2004). Moreover, based on the size of the depolarizing sag in response to a hyperpolarizing current pulse, there is also a much lower density of Ih in the CA2 somato-dendritic compartment compared to CA1 neurons.
Our results in hippocampal slices can account for a number of previous findings from in vivo recordings and behavioral analyses. Thus the powerful excitatory disynaptic circuit linking EC to CA2 to CA1, combined with the marked feed-forward inhibition from CA3 to CA2, can explain how in vivo electrical stimulation of entorhinal cortex inputs leads to the initial firing of CA2 neurons, followed by firing of CA1 and then CA3 neurons (Bartesaghi and Gessi, 2004; Bartesaghi et al., 2006). Moreover, the disynaptic circuit can account for the residual hippocampal-dependent memory and high levels of CA1 place cell firing in vivo following surgical or genetic lesions of CA3 (Brun et al., 2002; Nakashiba et al., 2008). Although this residual hippocampal function was originally proposed to result from the ability of the direct connection from LIII EC neurons to drive the firing of CA1 neurons, such inputs generate weak EPSPs in the CA1 neuron soma, display little LTP, and recruit strong feed-forward inhibition that suppresses CA1 neuron output. In contrast, the EC inputs to CA2 neurons are strong and undergo robust LTP that enhances the excitatory drive of the disynaptic path. This excitation may actually be enhanced upon lesion of CA3 due to the loss of feed-forward inhibition, providing a potential explanation for the elevated rate of CA1 neuron place cell firing in the lesioned animals (Nakashiba et al., 2008).
The cortico-CA2 synapses are also likely to contribute to hippocampal function and memory formation under normal conditions when CA3 inputs are intact. Thus, mice lacking the vasopressin V1b receptor, which is highly expressed in CA2 neurons (Young et al., 2006), show a selective impairment in episodic memory of the temporal order of events (DeVito et al., 2009). In addition, lesions at the CA3/CA2 border impair hippocampal-dependent operant conditioning (Samuel et al., 1997). Moreover, the finding that CA2 neurons exhibit place cell firing in vivo similar to CA1 neurons (Martig and Mizumori, 2010) indicates that the distal EC inputs are likely to provide a strong excitatory drive to the CA2 neurons in awake animals, given the fact that the SC inputs to CA2 neurons recruit a large net feedforward inhibition. Finally, CA2 neurons are the only hippocampal target of the supramammillary nucleus (which also targets the dentate gyrus), a structure involved in controlling the frequency of the theta rhythm (Pan and McNaughton, 2002) and the spread of epileptic activity (Saji et al., 2000) in the hippocampus. This suggests that CA2 neurons may play an important role in the modulation and synchronization of activity throughout the hippocampal network. Future studies using chemical or genetic lesions of the CA2 neurons will be important to further define the role of the CA2 neurons and the robust disynaptic circuit they mediate in the physiological function of the intact hippocampus.
Transverse hippocampal slices were prepared from 4–6 week old C57BL6 male mice. In brief, animals were anesthetized and killed by decapitation in accordance with institutional regulations. Hippocampi were dissected out and transverse slices (400 μM thickness) were cut on a vibratome (Leica VT1200S, Germany) in ice-cold extracellular solution containing (in mM): 10 NaCl, 195 sucrose, 2.5 KCl, 15 glucose, 26 NaHCO3, 1.25 NaH2PO4, 1 CaCl2 and 2 MgCl2. The slices were then transferred to 30°C ACSF (in mM: 125 NaCl, 2.5 KCl, 10 glucose, 26 NaHCO3, 1.25 NaH2PO4, 2 Na Pyruvate, 2 CaCl2 and 1 MgCl2) for 30 minutes and then kept at room temperature for at least 1.5 h before transfer to the recording chamber. Cutting and recording solutions were both saturated with 95% O2 and 5% CO2, pH 7.4. All experiments were performed at 33°C.
Whole-cell recordings were obtained from pyramidal neurons in current clamp mode at the resting membrane potential or held at −73 mV with a patch pipette (3–5 MΩ) containing (in mM): 135 KMeS03, 5 KCl, 0.1 EGTA-Na, 10 Hepes, 2 NaCl, 5 ATP, 0.4 GTP, 10 phosphocreatine (pH 7.2; 280–290 mOsm). Series resistance (typically 15–25 MΩ) was monitored throughout each experiment; cells with more than 15% change in series resistance were excluded from analysis. Extracellular field potentials were recorded with a patch pipette containing ACSF. Synaptic potentials were evoked by monopolar stimulation with a patch pipette filled with ACSF and located in the middle of stratum radiatum, the middle of stratum lacunosum moleculare or in the middle of the molecular layer of the dentate gyrus.
Recordings were obtained from CA2 pyramidal neurons located in the pyramidal cell layer in the area following the end of the mossy fiber track connecting the dentate gyrus to CA3. Neurons were identified according to specific electrophysiological properties summarized in table 1. The amplitude of the sag during a 1 second hyperpolarizing step was quantified for hyperpolarization to −100 mV (from a initial potential of −70 mV). Stimulation of layer III inputs was performed in the inner third of stratum lacunosum moleculare between the recording sites for CA2 and CA1 neurons (~200 μm from CA2 area). Some CA1 neurons were also recorded at the border with CA2 (see Figure S2). Layer II inputs were activated by stimulation in the middle molecular layer of the dentate gyrus. Polysynaptic responses resulting from successive activation of granule cells and CA3 neurons were blocked with 1 μm DCG-IV. CA3 neurons were recorded in the middle of CA3 (CA3b). Schaffer collateral inputs were activated by stimulation in the middle third of stratum radiatum at ~200 μm from CA2 area.
Neurons were hold at −70 mV for input-output curves, or at resting potential to study the impact of layer III and Schaffer collaterals on firing.
EPSPs were recorded in the presence of GABAA and GABAB blockers (100 μM Picrotoxin and 2 μM CGP 55845A). Inhibitory transmission was quantified by subtracting the synaptic potential in the presence of GABA blockers from the response in the absence of blockers. Field recordings of EPSPs were performed in stratum lacunosum moleculare or stratum radiatum (Figure 1, ,2,2, ,5A)5A) in different hippocampal areas, or in the somatic layer (Figure 5C). The fiber volley was quantified after blocking excitatory and inhibitory synaptic transmission with 50 μM D-APV, 10 μM NBQX, 100 μM Picrotoxin and 2 μM CGP 55845A.
LTP was induced by tetanic stimulation (100 pulses at 100Hz, repeated twice at 20 seconds interval) after 10 or 20 minutes of stable baseline recording for whole cell or field recordings, respectively. The magnitude of LTP was estimated by comparing averaged responses 40–50 minutes after the induction protocol with baseline-averaged responses 0–10 minutes before the induction protocol. Statistical comparisons were performed using Student’s t-test. Results are reported as mean ± SEM. All drugs were bath-applied following dilution into the external solution from concentrated stock solutions.
Whole cell recordings from CA2 pyramidal neurons were obtained with the fluorescent dye Alexa-fluor 594 cadaverin (50μM) in the patch pipette solution. The location of the recorded neurons in CA2 was subsequently confirmed after fixation of the slice for 1 hour in 4% paraformaldehyde in phosphate buffer (PBS). We further confirmed the location in CA2 by staining the slice for α-actinin 2, a protein enriched in the CA2 area (Wyszynski et al., 1998) After fixation, slices were washed in PBS and permeabilized with 0.1% triton in PBS. Slices were then incubated overnight in 3% goat serum in PBS with 0.1% triton followed by overnight incubation with the mouse monoclonal anti-α-actinin antibody (Sigma, St Louis, MO; 1/50) at 4ºC. Slices were then washed and incubated for 3 hours with a goat anti-mouse-Alexa 488 secondary antibody (Jackson ImmunoResearch Lab; 1/250). Slices were washed overnight in PBS and then mounted in Fluoro-Gel and studied by confocal fluorescence microscopy.
We thank Rebecca Piskorowski for her help on the morphometric analysis and Bina Santoro and Jayeeta Basu for helpful comments on the manuscript. This work was supported by the Howard Hughes Medical Institute.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.