Entorhinal Cortex Layer III Input to the Hippocampus Is Crucial for Temporal Association Memory

See allHide authors and affiliations

Science  09 Dec 2011:
Vol. 334, Issue 6061, pp. 1415-1420
DOI: 10.1126/science.1210125


Associating temporally discontinuous elements is crucial for the formation of episodic and working memories that depend on the hippocampal-entorhinal network. However, the neural circuits subserving these associations have remained unknown. The layer III inputs of the entorhinal cortex to the hippocampus may contribute to this process. To test this hypothesis, we generated a transgenic mouse in which these inputs are specifically inhibited. The mutant mice displayed significant impairments in spatial working-memory tasks and in the encoding phase of trace fear-conditioning. These results indicate a critical role of the entorhinal cortex layer III inputs to the hippocampus in temporal association memory.

A critical feature of episodic memory shared by some forms of working memory is the ability to associate temporally discontinuous elements, called temporal association memory (13). However, the neural circuits within the entorhinal cortex (EC)–hippocampus (HP) network subserving this type of association have remained unknown. The EC provides inputs to the HP via two major projections (Fig. 1A): the trisynaptic pathway (TSP) originating from EC layer II and the monosynaptic pathway (MSP) originating from EC layer III (ECIII). Studies on genetically engineered mice (47) and lesioned rats (811) have demonstrated crucial roles of the TSP in several features of episodic-memory processing, such as pattern completion (58) and separation (4, 8). In contrast, the MSP contributions to episodic-memory processing remain poorly known. We tested the hypothesis that the MSP is necessary for temporal association memory.

Fig. 1

Spatial specificity of Cre-loxP recombination in pOxr1-Cre transgenic mice. (A) Schematic of inputs from EC to HP, layer III (red, MSP) to CA1 and subiculum (Sub), and layer II (blue) to CA3 and DG. PrS, presubiculum; PaS, parasubiculum; V, layer V. (B and C) Parasagittal (B) and horizontal (C) brain sections of 12-week-old pOxr1-Cre/Rosa26 mice stained with X-gal and nuclear fast red. Adjacent sections stained with thionin (C). Arrowheads in (C) indicate MEC and LEC boundaries. Cb, cerebellum; Ctx, cortex; Str, striatum; r, rostral; c, caudal; l, lateral; m, medial. (D and E) Double immunofluorescence staining of horizontal sections of pOxr1-Cre/Rosa26 mouse with anti–β-gal [(D) and (E), red] and anti-NeuN [(D), green] or anti-parvalbumin [(E), green]. Colabeled cells in (D) are shown in yellow. (F to I) Injection sites of retrograde tracer, CTB (green), in the SLM of dorsal CA1 (F), intermediate CA1 (G), ventral CA1 (H), and the hippocampal fissure of dorsal CA1 (I). Arrowheads in (G) and (H) indicate the injection sites. (J to M) Parasagittal sections visualized with CTB-labeled cell body (green) and immunostained by anti–β-gal (red) in the EC. The dotted line denotes the lamina dissecans (J). (N) Magnified image from (M). Scale bars, 100 μm.

We created a Cre transgenic mouse, pOxr1-Cre (12). X-gal staining and thionin staining on brain sections from the progeny of pOxr1-Cre and Rosa26 (lacZ reporter line) crosses revealed that, by 12 weeks of age, Cre-loxP recombination was restricted to superficial layers of the dorsal portion of the medial EC (MEC) and occurred sparsely in the lateral EC (LEC) (Fig. 1, B and C, and figs. S1 and S2). Immunofluorescence studies with anti–β-galactosidase (β-gal, recombination marker) (Fig. 1, D and E, and fig. S3), anti-NeuN (Fig. 1D and figs. S3 and S4), anti-PGP9.5 (fig. S3), and anti-parvalbumin (Fig. 1E and fig. S3) or anti–GAD-67 (fig. S3) indicated that recombination was mostly confined to excitatory neurons in the MEC superficial layers.

To further define the distribution of recombination within the superficial layers, we conducted neuronal tracing experiments using the retrograde tracer AlexaFluor488–cholera toxin subunit B (CTB) injected into hippocampal subregions of the progeny of pOxr1-Cre and Rosa26 crosses. CTB injection into the stratum lacunosum moleculare (SLM) of dorsal CA1 (Fig. 1F) led to colabeling of only dorsal MEC layer III (MECIII) (Fig. 1J) by CTB and β-gal, whereas the injection of CTB into intermediate CA1 SLM (Fig. 1G) resulted in colabeling restricted to intermediate MECIII (Fig. 1K). In contrast, injection of the tracer into ventral CA1 SLM (Fig. 1H) led to labeling of ventral MECIII, which was poorly colabeled with β-gal (Fig. 1L). When CTB was injected into the hippocampal fissure, which covers both CA1 SLM (receives ECIII projections) and dentate gyrus (DG) stratum moleculare (receives ECII projections) (Fig. 1I) (13), we observed colabeling in the dorsal MECIII (Fig. 1, M and N). More importantly, there was a clear segregation of the layer II and layer III cells, confirming that recombination was primarily restricted to the MECIII (Fig. 1N and fig. S4).

To selectively inhibit synaptic transmission at the MECIII to CA1/subiculum synapses by use of the tetanus toxin light chain (TeTX), we crossed the pOxr1-Cre mouse (Tg1) with the pαCaMKII-loxP-STOP-loxP-tTA mouse (Tg2) and the tetO-TeTX mouse (Tg3) (Fig. 2A) (5). To examine the input-output relationship of transmission at the ECIII-CA1 synapses, we employed the fluorescent voltage-sensitive dye (VSD) imaging method (Fig. 2B and fig. S5) (14) on hippocampal slices prepared from the triple-transgenic mutant, MECIII-TeTX, and control mice (Tg1 x Tg2), which were raised on a doxycycline (Dox)–containing diet followed by 4 to 8 weeks of a Dox-free diet (Fig. 2, C and F). The MECIII-TeTX mice showed a significant reduction (78.4 to 95.9%) in postsynaptic fluorescence with SLM stimulation compared with controls (Fig. 2, B and D), whereas no difference of fluorescence was observed between the genotypes with stratum radiatum (SR) stimulation (Fig. 2, B and E). When hippocampal slices were prepared from MECIII-TeTX mice off Dox for 4 weeks, the postsynaptic fluorescence was reduced and matched that of the slices from MECIII-TeTX mice off Dox for 8 weeks (Fig. 2F and fig. S6). With 4 weeks of Dox withdrawal followed by 4 weeks of Dox readministration, the postsynaptic fluorescence at ECIII-CA1 synapses was restored in MECIII-TeTX mice to levels comparable to controls (Fig. 2F and fig. S6). These results demonstrate that transmission in MECIII-TeTX mice can be inhibited specifically at the MECIII-CA1 synapses in an inducible and reversible manner. Combined with the CTB tracing data from the pOxr1-Cre/Rosa26 mouse (Fig. 1, F to N), we conclude that inhibition of neuronal transmission in the MECIII-TeTX mice was restricted to synapses of the projections from the dorsal and intermediate MECIII neurons to the HP. The MECIII-TeTX mice raised on a Dox diet followed by 4-weeks of Dox withdrawal will be referred to as mutants hereafter, unless indicated otherwise.

Fig. 2

Triple-transgenic MECIII-TeTX mouse (mutant) and inducible and reversible inhibition of MECIII input to CA1. (A) Breeding strategies for mutant and control mice. (B) VSD imaging of transverse hippocampal sections. A stimulating electrode (red bar) was placed in SLM or SR, and fluorescent signal was measured in SR (green square). Dotted black lines indicate knife cuts to separate CA3 and DG from CA1. Fluorescent signal changes are displayed in pseudocolor after SLM or SR stimulation in control and mutant slices. Dashed white lines indicate the boundary between SR and SLM. (C) Dox withdrawal schedule in VSD experiments for (D) and (E). (D and E) Input-output relationships after SLM or SR stimulation in mutant and control slices. dF, fractional changes in fluorescence. (F) Reversibility of the synaptic inhibition. VSD imaging was performed 4 weeks after Dox withdrawal (solid circles) and after 4 weeks of Dox withdrawal followed by 4 weeks of Dox readministration (empty circles) in mutant slices after SLM stimulation. Asterisks indicate the significance at the 0.001 level; error bars indicate SEMs.

Immunohistological analyses revealed no obvious indications of molecular and cytoarchitectural abnormalities in the EC (figs. S7, S8, and S10) and HP (figs. S7 to S9). The mutants displayed no detectable abnormalities in anxiety, motor coordination, or pain sensitivity (fig. S11). The mutants were also normal in the acquisition, recall, and consolidation of spatial reference memory (figs. S12 and S13) and in the basic properties of place fields of CA1 pyramidal neurons and interneurons (fig. S14) (15).

The mutants exhibited a deficit in a water-maze version of the delayed matching-to-place (DMP) task (Fig. 3A) designed to test spatial working memory, a form of temporal association memory. During early training (block 0 to block 2), the mutants showed normal escape latencies compared to controls (Fig. 3B), presumably because they were still learning the task requirements and rules (16, 17). As training advanced and the platform size was reduced (blocks 3 and 4), however, the mutants’ escape latencies became greater compared with the controls in runs 2 to 4. Consequently, the savings (escape latency difference between runs 1 and 2) were less for the mutants in block 3 (t = 2.564, P < 0.05) and block 4 (t = 4.132, P < 0.001) (Fig. 3C), indicating an impairment in spatial working memory.

Fig. 3

Impairment of MECIII-TeTX mice in the DMP and DNMP tasks. (A) Protocol and Dox-schedule for a water-maze version of the DMP task. (B) Averaged escape latencies for run 1 (R1), R2, R3, and R4 for each 4-day block. (C) Averaged savings. Asterisks in (B) and (C) represent run-specific significance. (D) T-maze version of the DNMP task. Black dots represent rewards. (E) Success rate on 10 trials each day for 12 days. The gray dashed line denotes the success rate expected by a random arm selection on a choice run (50%). (F) Success rates averaged over 12 days. Asterisks indicate the significance at the level of 0.05 or less; error bars indicate SEMs.

We next subjected both genotypes to a delayed non–matching-to-place (DNMP) version of the T-maze task (Fig. 3D). The controls showed a significant improvement in performance between day 1 and day 12 (fig. S15). In contrast, the mutants were impaired in this task over the 12-day period (10 trials per day) [two-way analysis of variance: genotype, F1,360 = 96.75, P < 0.0001] (Fig. 3, E and F) and did not display an improvement (fig. S15), confirming the mutants’ impairment in spatial working memory.

To investigate whether the MECIII input to the HP plays a role in nonspatial temporal association, we subjected the mutant and control mice to trace fear-conditioning (TFC). In this task, a tone [conditioned stimulus (CS)] must be associated with a footshock [unconditioned stimulus (US)] delivered subsequently with a 20-s time gap. The mutants froze less than controls during the training composed of three CS-US pairs (Fig. 4A). The next day, in a distinct novel chamber, mutants froze less than controls in response to tone presentation (Fig. 4A). Conversely, when the temporal gaps between the CS and US were eliminated [delay fear-conditioning (DFC)], there were no freezing differences between the two genotypes during either the acquisition or the recall phase (Fig. 4B). To investigate whether the MSP is required during the acquisition and/or recall phase in TFC, we targeted the inhibition of synaptic transmission to each of these two phases. When inhibition was targeted to the acquisition phase, mutants froze less than controls in response to tone (Fig. 4C), whereas when inhibition was targeted to the recall phase, we observed no freezing difference between the genotypes (Fig. 4D).

Fig. 4

Fear conditionings with two mutants, MECIII-TeTX and CA3-TeTX, and pharmacologically treated mice. (A and B) Time course of freezing observed in MECIII-TeTX mice in the TFC (A) or DFC task (B) during training on day 1 (left) and testing on day 2 (middle). Gray and green bars represent tone and shock, respectively. Freezing levels during the testing were averaged over three epochs of the 60-s tone period and of the entire 240-s tone plus post-tone periods (right). (C and D) Time course of freezing (left) and averaged freezing levels (right) of MECIII-TeTX mice during testing of the TFC task in which the synaptic inhibition was targeted to the training (C) or testing (D). (E) Time course of freezing of CA3-TeTX mice during the training (left) and testing (middle) of the TFC task. (F) Time course of freezing (left) and averaged freezing levels (right) during testing for controls and mutants in the TFC task. Mice were bilaterally injected with either an antagonist mixture or vehicle into dorsal MEC before the training. Asterisks indicate the significance at the level of 0.05 or less; error bars indicate SEMs. n.s., not significant.

To further investigate the role of the EC-HP circuits in temporal association memory, we examined whether the indirect ECII input to CA1 via the TSP also played a role in TFC by using a second mutant, the CA3-TeTX mouse (5). These mice, in which CA3 input to CA1 is inhibited under Dox withdrawal (5), performed similar to controls in the TFC task (Fig. 4E).

We hypothesized that the essential function of the MECIII input to the HP in TFC may be associated with the persistent activity observed in vitro in their cells (18, 19). As this activity depends on activation of the metabotropic glutamate receptor 1 (19) and/or cholinergic muscarinic receptors (20), we injected a mixture of their respective antagonists (LY367385 and scopolamine) or the vehicle bilaterally into the dorsal EC and subjected these mice to TFC. In response to tone, the freezing level of antagonist-injected control mice was less than vehicle-injected control mice, but was not different from antagonist-injected mutants (Fig. 4F). With the same pharmacological treatment, we observed no freezing difference to tone between the genotypes in the DFC task (fig. S16)

In this study, we created a transgenic mouse line in which the synaptic output of the excitatory MEC layer III cells is specifically and reversibly inhibited. Our genetic manipulation and behavior data provided insights into the specific role of MEC layer III projection in the processing of hippocampal-dependent memory.

First, the MECIII input to the HP, an essential component of the MSP, is crucial for temporal associations in spatial working memory. The DMP task required animals to update the association between the platform location and spatial cues daily on the first run and maintain that association for subsequent runs that same day. The DNMP task required animals to associate the sample arm and the alternative reward arm across a delay period. Both of these tasks require a temporal association memory with a temporal gap of 15 to 30 s. Although it is possible that the deficits observed in the mutants are due to their inability to encode spatial/contextual information, this is unlikely because these animals were normal in other spatial/contextual memory tasks (figs. S12 to 14 and S17). Second, a blockade of the MECIII input into the HP resulted in an impairment in TFC, specifically during day 1 when the animals needed to associate the CS and US delivered with an intervening 20-s gap. Again, it is unlikely that the deficits are due to an inability to encode the CS or US because the same mutants were normal in DFC and even in a TFC task with a 40-s trace that may not be hippocampal-dependent (fig. S18) (21). Third, our demonstration that the alternative and indirect input to CA1 mediated by the TSP is dispensable for TFC highlights the importance of the MSP in temporal association memory.

In addition to the major projections to CA1/subiculum, MECIII neurons send axons to other EC cells (22). We expect that these connections are also blocked in the mutants and therefore cannot exclude the possibility that an impairment of intrinsic ECIII circuits contributed to the observed behavioral deficits. However, such an impairment alone can not explain the observed deficits in the hippocampal-dependent memory tasks unless it is translated into an MSP impairment. Thus, a more attractive possibility would be that the lack of transmission of persistent activity from the MECIII (19) to CA1/subiculum via the MSP resulted in the behavioral deficits in these hippocampal-dependent tasks, although such activity has yet to be demonstrated in vivo in the mouse. We hypothesize that tone induces sustained activity in the MECIII after its cessation. This persistent activity, perhaps with the help of theta and/or gamma frequency coupling between the MECIII and CA1 (and/or subiculum) (23, 24), may allow the delivery of the CS to the amygdala through CA1 (and/or subiculum) in a manner coincidental with the US signal. The differential effects of the mGluR1 and cholinergic muscarinic receptor antagonists injected into MEC on the TFC and DFC tasks (Fig. 4F and fig. S16) support this hypothesis.

Apart from specific mechanisms, our overall results demonstrate a crucial role of the EC layer III input to the HP in hippocampal-dependent temporal association memory.

Supporting Online Material

Materials and Methods

SOM Text

Figs. S1 to S18


References and Notes

  1. Materials and methods are available as supporting material on Science Online.
  2. Relevant discussion regarding data in figs. S12 and S14 can be found in the supporting online material.
  3. Acknowledgments: We thank C. Ragion, C. Twiss, M. Pfau, M. Ragion, C. Carr, and S. Perry for technical help; N. Arzoumanian for help with manuscript preparation; and D. Buh, J. Young, and other members of the Tonegawa lab for discussion. This work was supported by NIH grants R01-MH078821 and P50-MH58880 (to S.T.) and the RIKEN Brain Science Institute.

Stay Connected to Science

Navigate This Article