Astrocytes contribute to remote memory formation by modulating hippocampal-cortical communication during learning

Abstract

Remote memories depend on coordinated activity in the hippocampus and frontal cortices, but the timeline of these interactions is debated. Astrocytes sense and modify neuronal activity, but their role in remote memory is scarcely explored. We expressed the G_i-coupled designer receptor hM4Di in CA1 astrocytes and discovered that astrocytic manipulation during learning specifically impaired remote, but not recent, memory recall and decreased activity in the anterior cingulate cortex (ACC) during retrieval. We revealed massive recruitment of ACC-projecting CA1 neurons during memory acquisition, which was accompanied by the activation of ACC neurons. Astrocytic G_i activation disrupted CA3 to CA1 communication in vivo and reduced the downstream response in the ACC. In behaving mice, it induced a projection-specific inhibition of CA1-to-ACC neurons during learning, which consequently prevented ACC recruitment. Finally, direct inhibition of CA1-to-ACC-projecting neurons spared recent and impaired remote memory. Our findings suggest that remote memory acquisition involves projection-specific functions of astrocytes in regulating CA1-to-ACC neuronal communication.

Extended Data Fig. 1. Prolonged Gi pathway activation in CA1 astrocytes reduces their calcium activity (Related to Figure 1).

Following an injection of AAV8-GFAP∷hM4Di-mCherry, hM4Di was expressed in 87% (491/552 cells from 4 mice) of CA1 astrocytes (A), with >96% specificity (491/507 cells, from 4 mice)(B). (C-D) Minimal co-localization with the neuronal nuclear marker NeuN was detected (scale bar 50μm; 0.9% expression in neurons, 7/766 cells). (E,G) Representative astrocytes expressing GCaMP6f (white) and hM4Di-mCherry (not visible, but see Figure 1D) were exposed either to ACSF (E,F) or CNO (G,H). Representative ROIs, and their activity in corresponding colors are presented. Scale bars = 100μm.CNO application triggered a decrease in baseline intracellular Ca²⁺ levels, and reduced the total size of Ca²⁺ events in these cells (see Figure 1F,G). Data presented as mean ± standard error of the mean (SEM).

Extended Data Fig. 2. Astrocytic Gi activation in CA1 during learning had no effect on auditory-cued remote memory (Related to Figure 1).

GFAP∷hM4Di mice were injected with Saline (n=7) or CNO (n=6) 30 min before Fear Conditioning (FC) acquisition. CNO application before training had no effect on exploration of the conditioning cage (A), or on auditory-cued memory recall either 24 hr after acquisition (B) or 20 days after that (C) in a novel context, with both groups showing increased freezing during tone presentation (p<0.001, p<0.01, respectively). (D) Bilateral double injection of AAV5-CaMKIIα∷hM4Di-mCherry resulted in hM4Di-mCherry expression in CA1 Neurons only (top). Scale bar - 100 μm. The groups did not differ in the percent of hM4Di-expressing cells level of expression (Saline - 20.5%, CNO - 20.6%; bottom). CaMKIIα∷hM4Di mice were injected with either Saline (n=9) or CNO (n=10) 30min before FC acquisition. CNO application before training had no effect on exploration of the conditioning cage (E), or on auditory-cued memory recall either 24 hr after acquisition (F) or 20 days after that (G) in a novel context, with both groups showing increased freezing during tone presentation (p<0.000001, p<0.00001, respectively). (H) In a new group of GFAP∷hM4Di mice, CNO administration (n=12) only during the recall tests had no effect on either recent or remote memory, compared to Saline-injected controls (n=12). In these mice, CNO administration during recall also had no effect on auditory cued memory either 24 hr after acquisition (I) or 20 days after that (J), compared to Saline-injected controls. When CNO was not administered during acquisition of the non-associative place recognition task, the GFAP∷hM4Di mice (n=6) from Figure 1L showed equivalent performance to controls (n=7; p<0.01)(K). Example exploration traces and average Δ are shown (right). Data presented as mean ± SEM.

Extended Data Fig. 3. CNO application itself during learning had no effect on remote memory (Related to Figure 1).

(A) Bilateral double injection of AAV8-GFAP-eGFP resulted in eGFP expression in CA1 astrocytes only. Scale bar – left 300μm, right 50 μm. Mice expressing eGFP in their CA1 astrocytes were injected with either Saline (n=6) or CNO (n=7) 30min before fear conditioning acquisition. CNO administration before training to eGFP-expressing mice had no effect on baseline freezing or recent contextual memory recall one day later (B). Neither did CNO have any effect on remote memory 20 days later or 45 days after that (C). In the non-associative place recognition test, CNO application before a first visit to a new environment had no effect on remote memory 28 days later (D), reflected by a similar decrease (p<0.0001) in the exploration between Saline injected (n=6) and CNO-treated mice (n=7) Example exploration traces and the average change (Δ) following treatment are shown on the right. Data presented as mean ± SEM.

Extended Data Fig. 4. CNO administration during acquisition reduces CA1 and ACC activity at the time of remote recall only in GFAP∷hM4Di mice, and does not affect neuronal proliferation, differentiation, or survival (Related to Figure 2).

(A) Active neurons expressing cFos were quantified in the CA1, ACC, dentate gyrus (DG), retrosplenial cortex (RSC), and basolateral amygdala (BLA). GFAP∷hM4Di mice from figure 2A,B that were injected with CNO (n=6) before fear conditioning and showed impaired remote recall compared to Saline controls (n=6), also demonstrated reduced number of cFos expressing neurons in CA1 and ACC (p<0.05 for both)(B). No changes in cFos expression in the DG or RSC were observed in these mice, but the reduced fear was accompanied by a significant reduction in cFos expression in the BLA (p<0.05)(C). GFAP∷eGFP control mice were injected with CNO (n=5) or Saline (n=5) before fear conditioning, and then tested on the next day. No changes were observed in recent memory (D) or in the number of neurons active during recent recall in the CA1 or ACC (E). Other GFAP∷eGFP mice were injected with CNO (n=5) or Saline (n=6) before fear conditioning, and then tested on the next day and again 21 days later. No changes were observed in recent or remote memory (F), or in the number of neurons active during remote recall in the CA1 or ACC (G). Representative images of GFAP∷eGFP (green) and cFos (red in H,J green in I,K) following recent (H-I) or remote (J-K) recall in the CA1 (H,J) and ACC (I,K) are presented. (L) GFAP∷eGFP mice were injected with CNO or Saline together with BrdU before fear conditioning, and then tested on the next day. No changes were observed in stem cell proliferation (Brdu in white)(M) or in the number of young, Doublecortine (DCx)-positive neurons (white)(N). (O) GFAP∷eGFP mice were injected with CNO or Saline and BrdU before fear conditioning, and then tested 21 days later. No changes were observed in stem cell proliferation and differentiation (P) or in the number of young, DCx-positive neurons (Q). Scale bars = 100μm for CA1, ACC and whole DG, 10 μm for zoomed-in cells. Data presented as mean ± SEM.

Extended Data Fig. 5. Gi pathway activation in CA1 astrocytes during memory acquisition does not affect the recruitment of the RSC and DG (Related to Figure 3).

(A) GFAP∷hM4Di mice that were injected with CNO (n=9) or Saline (n=9) 30 minutes before fear conditioning showed similar immediate freezing following shock administration to Saline-injected controls. (B) Active neurons expressing cFos were quantified in the in the CA1, basolateral amygdala (BLA), ACC, retrosplenial cortex (RSC) and dentate gyrus (DG) of GFAP∷hM4Di mice that were injected with CNO (n=9) or Saline (n=9) 30 minutes before fear conditioning, or in home-caged mice (CNO n=4, Saline n=4). (C) Representative images of hM4Di (red) and cFos (green) in the CA1 (C) and ACC (D) of home caged GFAP∷hM4Di mice showing no effect of CNO administration on cFos levels. cFos-expressing astrocytes are observed below and above the CA1 pyramidal layer. Scale bars=100μm. (E) Fear-conditioned GFAP∷hM4Di mice showed increased cFos levels in the BLA compared to home-caged mice (p<0.01), but CNO administration had no effect on either group. Fear-conditioning and CNO administration had no effect on cFos levels in the RSC and DG. Representative images of hM4Di (red) and cFos (green) in the BLA (F), RSC (G) and DG (H) are presented. (I) Double staining for cFos and GFAP showed a negligible (0.34%) percent of ACC astrocytes that express cFos. (J) An electrode dipped in DiI was placed in the ACC to record the response to CA1 activation. (K) The location of the electrode in the ACC is shown in crimson, and no ChR2-eYFP positive axons (green) are observed in this region. All scale bars = 100μm. Data presented as mean ± SEM.

Extended Data Fig. 6. Gi pathway activation in CA1 astrocytes has no effect on cFos expression in home-caged mice (Related to Figure 4).

(A-B) Representative images of hM4Di in astrocytes (red), GFP in ACC-projecting CA1 neurons (green) and cFos (pink) in the CA1 of Saline- (A) or CNO- (B) injected home-caged mice are presented. No effect of CNO on cFos levels was observed. (C, E) Representative images of hM4Di in astrocytes (red), GFP in NAc-projecting CA1 neurons (green) and cFos (pink) in the CA1 of Saline- (C) or CNO- (E) injected fear-conditioned mice are presented, showing no effect of the astrocytic manipulation on CA1→NAc neurons activity. The GFP-positive axons of these CA1 neurons are clearly observed in the NAc (D,F), with no apparent effect on cFos expression in this region. All scale bars=50μm.

Extended Data Fig. 7. Specific inhibition of CA1-to-ACC projection during learning had no effect on auditory-cued memory (Related to Figure 5).

CA1→ACC-hM4Di mice were injected with Saline (n=9) or CNO (n=9) 30 min before FC acquisition. CNO application before training had no effect on exploration of the conditioning cage (A), or on auditory-cued memory recall either 24 hr after acquisition (B) or 20 days after that (C) in a novel context, with both groups showing increased freezing during tone presentation (p<0.00001, p<0.0001, respectively). Data presented as mean ± SEM.

Figure 1. Astrocytic Gi pathway activation in CA1 during learning specifically impaired remote contextual memory.

(A) Bilateral double injection of AAV8-GFAP∷hM4Di-mCherry resulted in hM4Di expression selectively in CA1 (scale bar 200μm). (B) hM4Di (red) was expressed in the astrocytic membrane around the soma, as well as in the distal processes (scale bar 50μm). (C) CNO administration in-vivo to mice expressing hM4Di (red) in CA1 astrocytes resulted in a significant increase in cFos expression (green) in these astrocytes, compared to Saline injected controls (p<0.00005, n = 2-4 mice, 6-15 slices per groups; scale bar 50μm). (D) hM4Di-mCherry and GCaMP6f were co-expressed in CA1 astrocytes. (E) Astrocytes were imaged for 3x3min before and after application of ACSF (109 ROIs from 5 mice) or CNO (10μM; 299 ROIs from 8 mice). CNO application triggered a decrease in baseline intracellular Ca²⁺ levels, reflected by the mode of fluorescence levels (p<0.01)(F) and reduced the total size of Ca²⁺ events in these cells (p<0.005)(G), compared to astrocytes treated with ACSF. All ROIs are presented as dots in a scatter plot, and the average change (Δ) following treatment is plotted in the insert. (H) Mice expressing hM4Di in their CA1 astrocytes were injected with either Saline (n=7) or CNO (n=6) 30min before fear conditioning (FC) acquisition. CNO application before training had no effect on baseline freezing before shock administration or on recent contextual freezing on the next day compared to Saline treated controls. (I) CNO application before training resulted in a >50% impairment (p<0.05) in contextual freezing in CNO-treated mice tested 20 days later, compared to Saline treated controls (left). An even bigger impairment of >68% (p<0.005) was observed 45 days later (right). (J) Mice expressing hM4Di in their CA1 neurons were injected with either Saline (n=9) or CNO (n=10) 30min before FC acquisition. CNO application before training had no effect on baseline freezing before shock administration, bur resulted in decreased recent contextual freezing on the next day (p<0.005), and decreased remote recall 20 days after that (p<0.05) compared to Saline treated controls. (K) In the non-associative place recognition test, astrocytic Gi pathway activation by CNO application before a first visit to a new environment had no effect on recent memory, reflected by a similar decrease (p<0.0001) in the exploration between Saline injected (n=6) and CNO-treated mice (n=8). Example exploration traces andthe average change (Δ) in exploration following treatment are shown on the right. (L) Astrocytic modulation impaired remote recognition of the environment on the second visit, reflected by a decrease in the exploration only in the Saline injected (n=7)(p<0.01), but not CNO-treated (n=6) mice. Example exploration traces and average decrease Δ are shown on the right. Data presented as mean ± standard error of the mean (SEM).

Figure 2. Astrocytic Gi activation during memory acquisition reduced CA1 and ACC activity at the time of remote recall, but did not affect neurogenesis

(A) Active neurons expressing cFos were quantified in the CA1 and ACC regions. GFAP∷hM4Di mice were injected with CNO (n=5) or Saline (n=5) before fear conditioning, and then tested on the next day. No changes were observed in recent memory (B) or in the number of neurons active during recall in the CA1 or ACC (C). Representative images of hM4Di (red) and cFos (green) in the CA1 (D) and ACC (E) are presented. Other GFAP∷hM4Di mice were injected with CNO (n=5) or Saline (n=6) before fear conditioning, and then tested on the next day and again 21 days later. No changes were observed in recent memory (F left). However, CNO application before training resulted in >50% reduction (p<0.05) in contextual freezing 21 days later, compared to Saline treated controls (F right). Impaired remote recall was accompanied by reduced number of cFos-expressing neurons in CA1 and ACC (p<0.05 and p<0.01, respectively)(G). Representative images of the CA1 (H) and ACC (I) are presented. (J) GFAP∷hM4Di mice were injected with CNO (n=5) or Saline (n=5) together with BrdU before fear conditioning, and then tested on the next day. No changes were observed in stem cell proliferation (Brdu in red)(K) or in the number of young, Doublecortine (DCx)-positive neurons (white)(L). (M) GFAP∷hM4Di mice were injected with CNO (n=5) or Saline (n=6) and BrdU before fear conditioning, and then tested 21 days later. No changes were observed in stem cell proliferation and differentiation (N) or in the number of young, DCx-positive neurons (O). All scale bars = 100μm, except zoomed-in image in panel N where scale bar = 10μm. Data presented as mean ± SEM.

Figure 3. Astrocytic Gi activation in the CA1 prevents the recruitment of the ACC during memory acquisition, and inhibits CA1 to ACC communication.

(A) GFAP∷hM4Di mice were injected with CNO (n=9) or Saline (n=9) 30 minutes before fear conditioning, and brains were removed 90 minutes later for cFos quantification. (B) Fear-conditioned GFAP∷hM4Di mice showed increased cFos levels in the CA1 compared to home-caged mice (p<0.01), but CNO administration had no effect on either group. cFos levels in the ACC were increased in GFAP∷hM4Di that underwent conditioning after being injected with Saline (p<0.05), but not in CNO-injected mice. Data presented as mean ± SEM. Representative images of hM4Di (red) and cFos (green) in the CA1 (C) and ACC (D) of fear-conditioned mice are presented. cFos-expressing astrocytes are observed below and above the CA1 pyramidal layer in CNO-treated mice. (E) AAV5-CaMKII∷Channelrhodopsin-2(ChR2)-eYFP was injected into the CA3 and AAV8-GFAP∷hM4Di-mCherry into CA1. (F) ChR2-eYFP was expressed in the soma of CA3 pyramidal cells. (G) The ChR2-expressing axons (green) are observed in the CA1 stratum radiatum, and hM4Di-expressing astrocytes (red) are observed in CA1. (H) Experimental setup: Light was applied to CA1 in anesthetized mice. The response to Schaffer collaterals optogenetic stimulation was simultaneously recorded in the CA1 and ACC, after Saline administration, followed by CNO administration. (I-J) The response in the CA1 to Schaffer collaterals optogenetic stimulation had a smaller amplitude under Gi-pathway activation by CNO in CA1 astrocytes (n= 4 mice; p<0.05). The average responses (I) from one mouse under Saline and then under CNO are presented (average in a bold line, SEM in shadow, blue light illumination in semi-transparent blue). (K-L) A downstream response of CA1 activation by Schaffer collaterals optogenetic stimulation was detected in the ACC. The mean absolute value of the complex ACC response was found to have significantly smaller amplitude under Gi-pathway activation by CNO in CA1 astrocytes (n= 5 mice; p<0.01). The average responses (K) from one mouse under Saline and then under CNO are presented (average in a bold line, SEM in shadow). All scale bars=50μm.

Figure 4. Gi pathway activation in CA1 astrocytes during memory acquisition specifically prevents the recruitment of CA1 neurons projecting to ACC.

(A) AAV-retro-CaMKII∷Cre was injected into the ACC, and AAV5-ef1α∷DIO-GFP together with AAV8-GFAP∷hM4Di-mCherry were injected into CA1. (B) Together, these three vectors induced the expression of GFP (green) in CA1 neurons projecting to the ACC, and hM4Di (red) in CA1 astrocytes. (C) GFP-positive axons of CA1 projection neurons are clearly visible in the ACC. (D) Mice expressing GFP in ACC-projecting CA1 neurons and hM4Di in their CA1 astrocytes that were injected with CNO (n=8) or Saline (n=7) 30 minutes before FC showed similar immediate freezing following shock administration. (E) Fear-conditioned mice showed increased cFos levels in the CA1 compared to home-caged mice (p<0.05), with no effect for CNO administration. cFos levels in the ACC were increased in mice that underwent conditioning after being injected with Saline (p<0.05), but not in CNO-injected mice. (F) Fear-conditioned mice injected with Saline showed an >130% increase in the percent of CA1 cells projecting into the ACC that express cFos, compared to home-caged mice (p<0.05). CNO administration completely abolished the recruitment of these cells during learning. Representative images of hM4Di in astrocytes (red), GFP in ACC-projecting CA1 neurons (green) and cFos (pink) in the CA1 of Saline- (G) or CNO- (H) injected mice are presented. (I) AAV-retro-CaMKII∷Cre was injected into the NAc, and AAV5-ef1α∷DIO-GFP together with AAV8-GFAP∷hM4Di-mCherry were injected into CA1. (J) Together, these three vectors induced the expression of GFP (green) in CA1 neurons projecting to the NAc, and hM4Di (red) in CA1 astrocytes. (K) GFP-positive axons of CA1 projection neurons are clearly visible in the NAc. (L) Mice expressing GFP in NAc-projecting CA1 neurons and hM4Di in their CA1 astrocytes that were injected with CNO (n=10) or Saline (n=8) 30 minutes before FC showed similar immediate freezing following shock administration. (M) Fear-conditioned mice showed increased cFos levels in the NAc compared to home-caged mice (p<0.05), with no effect for CNO administration. (N) Fear-conditioned mice injected with either Saline or CNO showed an >60% increase in the percent of CA1 cells projecting into the NAc that express cFos, compared to home-caged mice (p<0.05). CNO administration had no effect on the recruitment of these cells during learning. All scale bars=50μm. Data presented as mean ± SEM.

Figure 5. Specific inhibition of CA1-to-ACC projection during learning impairs the acquisition of remote, but not recent, memory.

(A) AAV-retro-CaMKII∷Cre was injected into the ACC, and AAV5-ef1α∷DIO-hM4Di-mCherry was injected into CA1. (B) Together, these vectors induced the expression of hM4Di-mCherry (red) in CA1 neurons projecting to the ACC. (C) hM4Di-mCherry-positive axons of CA1 projection neurons are clearly visible in the ACC. (D) Mice expressing hM4Di in their ACC-projecting CA1 neurons were injected with either Saline (n=9) or CNO (n=9) 30min before FC acquisition. CNO application before training had no effect on baseline freezing (left) before shock administration or on recent contextual freezing (middle) on the next day, but induced a significant decrease (p<0.05) 20 days later, compared to Saline treated controls (right). (E) Active neurons expressing cFos were quantified in the CA1 and ACC regions. Impaired remote recall was accompanied by reduced number of cFos-expressing neurons in CA1 and ACC (p<0.05 for both). (F) CNO administration reduced the recruitment of CA1→ACC cells during remote recall. Representative images of hM4Di (red) and cFos (green) in the CA1 (G) and ACC (H) are presented. All scale bars = 100μm. Data presented as mean ± SEM.