Report

Interregional synaptic maps among engram cells underlie memory formation

See allHide authors and affiliations

Science  27 Apr 2018:
Vol. 360, Issue 6387, pp. 430-435
DOI: 10.1126/science.aas9204

Memories are stored in synapses

Memory formation is thought to change the strength of synaptic connections between neurons. However, direct measurements between neurons that participate in a learning process are difficult to obtain. Choi et al. developed the “dual-eGRASP” technique to identify synaptic connections between hippocampal CA3 and CA1 pyramidal cells. This method could label two different sets of synapses so that their convergence on the same dendrites would be quantified. After contextual fear conditioning in mice, the number and size of spines were increased on CA1 engram cells receiving input from CA3 engram cells.

Science, this issue p. 430

Abstract

Memory resides in engram cells distributed across the brain. However, the site-specific substrate within these engram cells remains theoretical, even though it is generally accepted that synaptic plasticity encodes memories. We developed the dual-eGRASP (green fluorescent protein reconstitution across synaptic partners) technique to examine synapses between engram cells to identify the specific neuronal site for memory storage. We found an increased number and size of spines on CA1 engram cells receiving input from CA3 engram cells. In contextual fear conditioning, this enhanced connectivity between engram cells encoded memory strength. CA3 engram to CA1 engram projections strongly occluded long-term potentiation. These results indicate that enhanced structural and functional connectivity between engram cells across two directly connected brain regions forms the synaptic correlate for memory formation.

Memory storage and retrieval require specific populations of neurons that show increased neuronal activity during memory formation. Several studies identified these engram cells throughout various brain regions and demonstrated that activated engram cells can induce artificial retrieval of stored memories (16). To explain how memory is encoded in the engram, Hebb proposed a hypothetical mechanism, often paraphrased as “fire together, wire together” (7). This hypothesis suggests that synaptic strengthening between coactivated neurons forms the neural substrate of memory. However, it has not been possible to delineate whether memory formation enhances synapses between engram cells in connected brain regions because we could not distinguish presynaptic regions originating from engram cells and nonengram cells.

To compare two different presynaptic populations that project to a single postsynaptic neuron, we modified the green fluorescent protein (GFP) reconstitution across synaptic partners (GRASP) technique (8, 9). GRASP uses two complementary mutant GFP fragments (10), which are expressed separately on presynaptic and postsynaptic membranes and reconstitute in the synaptic cleft to form functional GFP. This GFP signal indicates a formed synapse between the neuron expressing the presynaptic component and the neuron expressing the postsynaptic component. We developed an enhanced GRASP (eGRASP) technique, which exhibits increased GRASP signal intensity by introducing a weakly interacting domain that facilitates GFP reconstitution and a single mutation commonly found on most advanced GFP variants (fig. S1) (11). We further evolved eGRASP to reconstitute cyan or yellow fluorescent protein (Fig. 1, A and B, and fig. S2) (1214). Placing the color-determining domain in the presynaptic neuron (cyan/yellow pre-eGRASP) and the common domain to the postsynaptic neuron (post-eGRASP) enabled visualization of the two synaptic populations that originated from two different presynaptic neuron populations and projected to a single postsynaptic neuron. We named this technique dual-eGRASP (Fig. 1A). We demonstrated that two colors reveal the contact interface in human embryonic kidney (HEK) 293T cells expressing the common domain with cells expressing either of the color-determining domains (Fig. 1C). We successfully applied this technique to synapses on dentate gyrus (DG) granule cells originating from either the lateral entorhinal cortex (LEC) or the medial entorhinal cortex (MEC) that projected to the outer and middle molecular layers of the DG, respectively (Fig. 1D) (15). This technique can also separately label intermixed synapses that do not have a unique spatial distribution on CA1 pyramidal neurons that originate from either the contralateral CA3 or ipsilateral CA3 (Fig. 1E) (16). We confirmed that the eGRASP formation itself does not induce undesired strengthening of the synaptic transmission between the neurons expressing pre-eGRASP and post-eGRASP (fig. S3).

Fig. 1 Dual-eGRASP differentiates two population of synapses on a single neuron.

(A) Schematic illustration of cyan and yellow eGRASP. Cyan pre-eGRASP and yellow pre-eGRASP are expressed in two different presynaptic population, and common post-eGRASP is expressed in a single postsynaptic cell. (B) Coexpression of either cyan or yellow pre-eGRASP with post-eGRASP and iRFP670 in HEK293T cells. (C) Three populations of HEK293T cells were separately transfected using nucleofection. One population expressed cyan pre-eGRASP and mCherry, another population expressed yellow pre-eGRASP and mCherry, and the third population expressed post-eGRASP and iRFP670. (D) Cyan pre-eGRASP and yellow pre-eGRASP were expressed in the LEC and MEC, respectively. Post-eGRASP was expressed together with myristoylated TagRFP-T (myr_TagRFP-T) in the DG. (E) Cyan pre-eGRASP and yellow pre-eGRASP were expressed in the right CA3 and left CA3, respectively. Post-eGRASP was expressed together with myr_TagRFP-T in CA1.

To apply dual-eGRASP on synaptic connections between engram cells from two different regions, we used a Fos promoter–driven reverse tetracycline–controlled transactivator (rtTA) delivered by adeno-associated virus (AAV) to express specific genes of interest in the engram cells at particular time points (1720). Doxycycline injection 2 hours before either seizure induction or contextual fear conditioning (CFC) successfully labeled the cells activated during these events (figs. S4 and S5). Using this Fos-rtTA system, we expressed post-eGRASP together with membrane-targeted mScarlet-I (21) unilaterally in CA1 engram cells and yellow pre-eGRASP in the contralateral CA3 engram cells to avoid possible coexpression of pre-eGRASP and post-eGRASP. This system labeled CA3 engram to CA1 engram (E-E) synapses with yellow eGRASP signals on red fluorescently labeled dendrites. To compare these synapses with other synapses [nonengram to engram (N-E), engram to nonengram (E-N), and nonengram to nonengram (N-N) synapses], we expressed post-eGRASP together with membrane-targeted iRFP670 (22) in a sparse neuronal population from the ipsilateral CA1, while expressing cyan pre-eGRASP in a random neuronal population from the contralateral CA3. We achieved strong expression in the random neuronal population using a high titer of double-floxed inverted open reading frame (DIO) AAV with a lower titer of Cre recombinase expressing AAV (Fig. 2A). We confirmed that yellow pre-eGRASP expression is doxycycline dependent, demonstrating that this system can label synapses originating from engram cells of a specific event (fig. S6). We successfully distinguished four types of synapses in the same brain slice after CFC. Based on the percentage of overlapping fluorescence, CA3 cells expressing cyan pre-eGRASP, yellow pre-eGRASP, CA1 cells expressing iRFP and mScarlet-I are estimated to be 78.38, 40.25, 11.61, and 20.93%, respectively (fig. S7). Cyan and yellow puncta on red (mScarlet-I) dendrites indicated N-E and E-E synapses, respectively, whereas cyan and yellow puncta on near-infrared (iRFP670) dendrites indicated N-N and E-N synapses (Fig. 2, B and C). We considered puncta expressing both cyan and yellow fluorescence as synapses originating from engram cells, because these synapses originate from CA3 cells expressing both cyan pre-eGRASP (randomly selected population) and yellow pre-eGRASP (engram cells). We found no significant differences between the density of N-N and N-E synapses (Fig. 2D and fig. S8, A and C); however, the density of E-E synapses was significantly higher than E-N synapses (Fig. 2D and fig. S8, B and D). This difference indicates that presynaptic terminals from CA3 engram cells predominantly synapsed on CA1 engram cells rather than CA1 nonengram cells. We also examined the size of spines in each synapse population. E-E spine head diameter and spine volume were significantly greater than N-E synaptic spines, whereas N-N and E-N did not show any significant differences (Fig. 2E).

Fig. 2 CA3 engram to CA1 engram synapses exhibited higher synaptic density and larger spine size after memory formation.

(A) (Left) Schematic illustration of injected AAVs. (Middle) Illustration of virus injection sites. Injection in each site was performed with a complete cocktail of all the virus infected in each site. (Right) Schematic of the experimental protocol. (B) (Left) Schematic diagram of the four possible synapse populations among engram and nonengram cells. (Right) Classification of the four synaptic populations indicated by four colors. Green, N-N; orange, E-N; blue, N-E; red, E-E. The color for each group applies to Figs. 2 and 3. (C) Representative image with three-dimensional modeling for analysis. (D) Normalized cyan/yellow eGRASP per dendritic length. The densities of cyan-only (left) or yellow puncta (right) on red dendrites are normalized to the corresponding cyan-only or yellow puncta on near-infrared dendrites from the same images in order to exclude the effect of different number of CA3 cells expressing each presynaptic components. Each data point represents a dendrite. n = 43 for CA1 nonengram dendrites; n = 45 for CA1 engram dendrites; 9 images from 3 mice. Mann Whitney two-tailed test. n.s., not significant; **P = 0.0017. (E) Normalized spine head diameters and spine volumes of dendrites from CA1 nonengram cells (left) and engram cells (right) with schematic illustration. Sizes of the spines with yellow puncta were normalized to those of the spines with cyan-only puncta of the same dendrite. Each data point represents a spine. N-N, n = 81; E-N, n = 107; N-E, n = 93; E-E, n = 55. Mann Whitney two-tailed test. n.s., not significant; **P = 0.0014; ****P < 0.0001. Data are represented as mean ± SEM.

Although the number of engram cells may remain constant across different memory strengths (23), we predicted that connectivity between pre- and post-engram cells could encode memory strength. We investigated whether memory strength correlates with connectivity between engram cells using the same combination of AAVs and injection sites (Fig. 3A) as described in Fig. 2. To induce different strengths of memory, we divided mice into three groups. Mice were exposed to either weak (one shock of 0.35 mA) or strong (three shocks of 0.75 mA) electric foot shocks during CFC, while mice in the context-only group were exposed to the context without any foot shocks (Fig. 3B). Increasing electric foot shock intensity during memory formation produced higher freezing levels (Fig. 3C). When we quantified the number of CA3 and CA1 engram cells, we found no significant differences among the three groups (fig. S9) (23). There were no significant differences between the density of N-N and N-E synapses in all groups. However, we found a significantly higher density of E-E synapses in the strong shock group compared with the context only and weak shock group (Fig. 3, D and E). We further investigated whether the size of spines was positively correlated with memory strength. E-E spine head diameter and spine volume were significantly greater in the strong shock group than in the other groups, whereas N-N and E-N did not show any significant differences in all groups (Fig. 3F and fig. S10).

Fig. 3 Synaptic connectivity between pre- and post-engram cells is correlated to memory strength.

(A) Schematic illustration of injected AAVs, illustration of virus injection sites, and experimental protocol. (B) Schematic illustration of the conditioning and retrieval process. (C) Freezing levels for each group: context, n = 6; weak, n = 5; strong, n = 5, Tukey’s multiple comparison test after one-way analysis of variance (ANOVA); F(2,13) = 15.85; *P < 0.05; ***P < 0.001. (D) Schematic illustrations of hypothesized results showing higher density of E-E synapses with increasing memory strength. (E) Synaptic density of each connections. n = 74, context N-N; n = 67, context N-E; n =79, weak N-N; n = 80, weak N-E; n = 92, strong N-N;n = 91, strong N-E; n = 74, context E-N; n = 67, context E-E; n = 79, weak E-N; n = 80, weak E-E; n = 92, strong E-N; n = 91, strong E-E. Fifteen images from six mice for context group. Sixteen images from five mice for weak group. Nineteen images from five mice for strong group. Mann-Whitney two-tailed test, n.s.: not significant, *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. (F) Spine head diameter of each connection. n = 107, context N-N; n = 64, context E-N; n = 72, weak N-N; n = 34, weak E-N; n = 112, strong N-N; n = 46, strong E-N; n = 103, context N-E; n = 77, context E-E; n = 85, weak N-E; n = 84, weak E-E; n = 57, strong N-E; n = 110, strong E-E, six mice for context group, five mice for weak shock group, five mice for strong shock group. Mann Whitney two-tailed test. n.s., not significant; *P < 0.05, **P < 0.01; ***P < 0.001; ****P < 0.0001. Data are represented as mean ± SEM.

Because we found increased structural connectivity between CA3 and CA1 engram cells after memory formation, we investigated the synaptic strength of these synapses. We selectively stimulated two different inputs from CA3 neurons using two opsins, Chronos and ChrimsonR, that can be independently activated using blue and yellow wavelength lasers, respectively (24). We expressed ChrimsonR in CA3 engram neurons using Fos-rtTA, while we expressed Chronos primarily in CA3 excitatory neurons under the calcium/calmodulin-dependent protein kinase type II alpha (CaMKIIα) promoter (Fig. 4A) (25). We labeled CA1 engram neurons with nucleus-targeted mEmerald (mEmerald-Nuc) using Fos-rtTA and then performed whole-cell recordings from either CA1 engram or nonengram neurons. We investigated the following four combinations of synaptic responses in a single hippocampal slice after CFC: total excitatory to nonengram (T-N), total excitatory to engram (T-E), engram to nonengram (E-N), and engram to engram (E-E) (Fig. 4B). First, we investigated presynaptic transmission using paired-pulse ratios (PPR) (Fig. 4, C and D). PPR from CA3 engram inputs were significantly decreased at 25-, 50-, and 75-ms interstimulus intervals, which suggests increased release probability from CA3 engram inputs to CA1. The decrease was most prominent in E-E synaptic responses (Fig. 4E). We then examined postsynaptic α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptor levels in individual synapses from the four combinations of synaptic responses by replacing Ca2+ with Sr2+ in the external recording solution (26, 27). Sr2+ desynchronized evoked release and induced prolonged asynchronous release, which enabled measurement of quantal synaptic response (Fig. 4F). We measured the amplitude of evoked miniature excitatory postsynaptic currents (mEPSCs) 60 to 400 ms after light stimulation. Synapses from CA1 engram cells exhibited significantly increased levels of postsynaptic AMPA receptors compared with CA1 nonengram cell levels (Fig. 4G). These results indicate that the synapses of CA1 engram cells were potentiated after memory formation but not the synapses of CA1 nonengram cells. Alterations in both presynaptic release probability and postsynaptic potentiation are important for long-term potentiation (LTP) (28). To measure the existence of LTP during memory formation, we examined the extent of LTP occlusion by inducing pairing LTP separately in the four synaptic types (Fig. 4H) (29). After 5 min of baseline recording, we delivered pairing LTP stimuli. We found robustly potentiated T-N synaptic responses (~150%). T-E and E-N synaptic responses were potentiated to a lower extent than T-N synaptic responses (~120%), but these differences were not significant. Interestingly, we found that pairing LTP in E-E synaptic responses was completely blocked and potentiation was significantly lower than T-N synaptic responses (Fig. 4I).

Fig. 4 Enhanced synaptic transmission between CA3 engram and CA1 engram cells through pre- and postsynaptic mechanisms.

(A) (Left) Schematic illustration of injected AAVs. (Right) Illustration of virus injection sites and experimental protocol. (B) (Left) Diagram of whole-cell recording experiments. (Right) Classification of the four synaptic populations indicated by four colors. Green, T-N; orange, E-N; blue, T-E; red, E-E. The color for each group applies to all the panels below. (C) Traces from PPR recordings. (D) Results from PPR recordings. T-N, n = 11; T-E, n = 10; E-N, n = 11; E-E, n = 12. (E) Average PPR at the indicated interstimulus intervals. *P < 0.05; **P < 0.01; ***P < 0.001; Tukey’s multiple comparison test after one-way ANOVA; (25 ms) F(3,40) = 8.259, *P = 0.0276; (50 ms) F(3,40) = 7.989, ***P = 0.0003; (75 ms) F(3,40) = 7.517, ***P = 0.0004. (F) Traces of Sr2+ light-evoked mEPSCs. Arrowheads indicate quantal release events. (G) Average amplitude of the Sr2+ light-evoked mEPSCs. T-N, n = 15; T-E, n = 18; E-N, n = 12; E-E, n = 13; **P < 0.01, Tukey’s multiple comparison test after one-way ANOVA, F(3,54) = 8.540, ***P < 0.0001. (H) Pairing LTP with stimulus given after 5 min of baseline recording. T-N, n = 14; T-E, n = 10; E-N, n = 11; E-E, n = 9. (I) Average EPSC amplitude of the last 5 min of recording. *P < 0.05, Tukey’s multiple comparison test after one-way ANOVA, F(3,40) = 3.683, *P = 0.0197. Data are represented as mean ± SEM.

Our finding that synaptic populations that fired together during memory formation showed the strongest connections demonstrates that classical Hebbian plasticity indeed occurs during the learning and memory process at CA3 to CA1 synapses (7, 30). It is possible that cells with higher connectivity are allocated together into a memory circuit, in contrast to enhanced connectivity after learning. However, the allocated cell number remains constant regardless of the memory strength, whereas the connectivity is significantly enhanced with a stronger memory. This finding indicates a significant contribution of post-learning enhancement over the predetermined connectivity. The relationship between memory strength and synaptic connectivity suggests that these specific connections between engram cells across two directly connected brain regions form the synaptic substrate for memory.

Supplementary Materials

www.sciencemag.org/content/360/6387/430/suppl/DC1

Materials and Methods

Figs. S1 to S10

References (31, 32)

References and Notes

Acknowledgments: Funding: This work was supported by the National Honor Scientist Program (NRF-2012R1A3A1050385) of Korea. C.S.L. was supported by the Basic Science Research Program (NRF-2016R1D1A1B03931525) through the National Research Foundation (NRF) of Korea. Author contributions: J.-H.C., S.-E.S., J.-i.K., and D.I.C. contributed equally to this work. J.-H.C. and B.K.K. designed the experiment, developed the dual-eGRASP and Fos-rtTA systems, and contributed to the analysis. S.-E.S. designed and performed all electrophysiology experiments and processed and analyzed the data. J.-i.K. produced and purified AAVs. J.-i.K. and D.I.C. performed viral injections, contextual fear conditioning, brain preparation, imaging, and analysis. J.O., J.L., C.S.L., S.Y., T.K., and H.-G.K assisted with viral injections, contextual fear conditioning, electrophysiology experiments, and analyzing data. B.K.K. supervised the project. J.-H.C., S.-E.S., J.-i.K., D.I.C., and B.K.K. wrote the manuscript. Competing interests: The authors declare no competing financial interests. Data and materials availability: All data to understand and assess the conclusions of this study are available in the main text or supplementary materials.
View Abstract

Stay Connected to Science

Navigate This Article