Introduction

The retrosplenial cortex (RSC) is a neocortical structure that acts as an integration center between sensory, limbic, and higher-order cortical brain regions [1]. In humans, the RSC has been shown to be integral for topographical and episodic memories [1, 2]. The RSC has also been identified as a key structure for spatial [3,4,5,6,7,8] and contextual fear memories [9,10,11] in rodents. The RSC integrates spatial navigational information using place-like and head-direction cells [12], and damage to the RSC leads to deficits in spatial cognition [2]. Furthermore, the RSC has been linked to numerous neurological disorders, including post-traumatic stress disorder [13], schizophrenia [14, 15], and Alzheimer’s Disease and related dementias (ADRD) [16, 17]. RSC dysfunction is observed during the early stages of ADRD, and lesions in RSC lead to memory impairments [18,19,20,21,22]. Additionally, neuronal tracing studies have identified several major RSC connections to other brain regions, including the hippocampus and entorhinal cortex [23,24,25]. These regions are critical within the spatial memory circuit, and the connectivity between the RSC and these regions indicates its essential role in spatial navigation and memory [26]. Spatial gene expression changes in the dorsal hippocampus during an early time window after learning are essential for long-term memory consolidation [27, 28]. However, the spatial transcriptional signature of the RSC during this early time point of memory consolidation has not been clearly described.

Recent single-cell transcriptomic studies have identified multiple spatially restricted excitatory neuronal sub-types within the RSC, many of which are unique to this region in comparison to other cortical areas [29, 30]. The RSC is anatomically diverse, comprising of anterior and posterior as well as granular and agranular subregions. These subregions receive distinct inputs, project to a wide variety of brain regions [31, 32], and display differing roles in behavior. Glutamatergic neurotransmission in the RSC has been implicated in contextual memory consolidation [9, 10, 33,34,35]. Furthermore, these excitatory neurons within the RSC laminar layers exhibit functional and behavioral differences; for example, ablation of neurons in layers 4 and 5 of granular RSC caused amnesia in rats [36]. But, given this diversity, it has been challenging to delineate specific roles for each excitatory neuronal subtype within the RSC laminae. Therefore, it is critical to investigate the molecular signature of layer-specific patterns of excitatory neurons in the RSC to determine which neuronal populations are critical for long-term memory.

The molecular mechanism underlying spatial memory consolidation involves precisely timed transcriptional events. Memory-encoding neuronal ensembles comprising cells exhibiting the induced expression of immediate early genes (IEGs), also known as engram cells, have been extensively studied in the forebrain [37, 38]. Studies have shown the upregulation of activity-induced genes such as Arc and Fos in the RSC after learning in a contextual fear conditioning task [39, 40]. Using in vivo IEG imaging, a recent study showed engram activation in the RSC during a spatial memory task [41]. The study also demonstrated that the stability of the RSC engrams recruited during a spatial navigation task is associated with memory performance, providing evidence for a relationship between spatial memory and RSC engram ensembles [41]. Bulk transcriptomic profiling of the RSC and hippocampus identified a common induction pattern of an engram-specific gene expression signature after spatial learning in adult mice [42]. A study on the expression of Fos, an engram marker, suggests a unique role the RSC plays in tracking an animal’s position in the environment [43]. However, such learning-induced gene expression in the RSC at single-cell resolution across excitatory neuronal layers has never been performed. Therefore, understanding the precise transcriptomic signature of IEG that are markers of engrams in the RSC will provide mechanistic insights into the role of the RSC in spatial memory consolidation.

Single nuclei RNA sequencing and spatial transcriptomics have been used to study RSC-specific cell type identity [29] and gene expression following nerve injury [44]. However, the literature lacks single-cell and spatial transcriptomic studies investigating the anatomically distinct RSC neurons following a learning experience. In this study, we used cutting-edge molecular approaches to delineate the transcriptional signatures within the RSC during spatial memory consolidation. First, we investigated the unbiased transcriptional signature in the RSC from our previously published spatial transcriptomics dataset after spatial learning. Next, we utilized a newly developed spatial transcriptomics platform, Xenium, to investigate learning-induced gene expression in the RSC at single-cell resolution following spatial learning in adult mice. We utilized a computational tool to predict the neuronal activity pattern following learning in the RSC of adult mice. We next examined how tau pathology impacts the predicted neuronal activity after learning in a mouse model of ADRD. Lastly, using a chemogenetic approach, we demonstrate the importance of learning-induced activation of excitatory neurons in the RSC for long-term spatial memory consolidation. Together, our findings provide insights into the transcriptomic signature of the RSC at single-cell resolution during long-term memory consolidation.

Results

Learning in a spatial object recognition (SOR) task induces immediate early gene expression changes in the RSC

RSC neuronal activation has been reported during tasks that utilize spatial locations and landmarks [45,46,47,48]. Interestingly, introducing an object to the environment alters the mean firing rate of RSC neurons [48]. More specifically, activity within the anterior RSC has been shown to be required for long-term spatial memory [49, 50]. Therefore, we first investigated the unbiased transcriptomic signature of the anterior RSC following learning in a spatial object recognition (SOR) task from our previously published datasets (GSE223066 and GSE201610) [27, 51]. Using these datasets, we have previously demonstrated the spatial transcriptomic signature of the dorsal hippocampal subregions one hour after training in the SOR task [27]; however, the learning-induced gene expression signature in the RSC was not studied. To investigate the gene expression changes in the RSC after spatial learning, we performed differential expression of genes (DEG) analysis on the RSC region comparing learning with homecage controls, and this analysis revealed 64 upregulated and 13 downregulated genes (Fig. 1a, b, and Supplemental Table 1). Several of the top significant genes identified were IEGs such as Fos, Arc, Nr4a1, and Egr1 (Fig. 1c). These activity-induced genes have been well characterized for their roles in learning and memory in the hippocampus and are used as markers of neuronal activity and engram ensembles [27, 39]. We further validated the induction of a few IEGs after learning from an independent experiment using qPCR analysis from RSC tissue (Fig. 1d). Next, we performed a gene ontology (GO) enrichment analysis to identify the molecular functions most represented among the DEGs in the RSC following learning (Supplemental Table 2). The top enriched pathways in the RSC include RNA polymerase II-specific DNA binding transcription factor binding, ubiquitin-protein ligase biding and ubiquitin-like protein ligase binding, unfolded protein binding, protein chaperone, heat shock protein binding, MAP kinase activity, MAP kinase phosphatase activity, ATPase regulator activity and misfolded protein binding (Fig. 1e). Genes related to DNA binding transcription factor binding such as Nr4a1, Nr4a2, and Nr4a3 are closely linked to long-term memory consolidation and mice expressing a dominant negative form of Nr4a that blocks the transactivation function of all the Nr4a family members in the excitatory neurons in forebrain exhibits long-term memory deficits in contextual fear conditioning [52] and SOR tasks [53].

Fig. 1: Spatial transcriptomic analysis using Visium reveals learning-induced gene expression in the RSC.

a Schematics of the experimental paradigm being utilized for spatial gene expression and the region of interest being analyzed. b Number of differentially expressed genes found within the RSC 1 h after learning (n = 7 for both groups). c Heatmap displaying the top 50 significant differentially expressed genes within the RSC. d Expression of Egr1, Nr4a1, and Dusp5 within the RSC 1 h after learning. Normalized to homecage animals (homecage n = 3, and learning n = 4). Unpaired t-test: *p < 0.05 (Nr4a1: p = 0.0178, Dusp5: 0.0198), **p < 0.005 (Egr1: p = 0.0013). The data is shown as group mean ± SEM. e Cnet plot exhibiting gene ontology (molecular function) enrichment analysis of the differentially expressed genes that were identified within the RSC.

Single-cell resolution spatial transcriptomic analysis reveals distinct and overlapping learning-induced gene expression changes across major cell types in the RSC

Our unbiased gene expression analysis using the spatial transcriptomics (Visium) approach revealed a learning-induced gene expression signature in the RSC (Fig. 1). However, this approach lacks information regarding which cell types exhibit these learning-induced gene expression changes. Therefore, to determine the precise cell type-specific gene expression signature of the RSC, we utilized Xenium, an in situ- hybridization approach for single-cell analysis of each RNA molecule at a high spatial resolution within anatomically distinct tissue regions. We trained adult male mice in the SOR task and collected brains one hour after training for Xenium analysis. Homecage animals were used as baseline controls (Fig. 2a). We employed a 297-gene panel composed of a standard probe panel including cell-type markers, genes responsive to learning and memory, and genes involved in other processes (full probe list: Supplemental Table 3). We obtained 26,484 high-quality cells from coronal sections in the RSC across learning and homecage animals.

Fig. 2: Single-cell resolution spatial transcriptomic approach reveals learning-induced gene expression across major cell types in RSC.

a Schematics depicting the spatial learning task, the region of interest collected, and the processing for Xenium pipeline. b Spatial position of cells of a single coronal tissue section colored by major cell types in RSC. Ex neurons = Excitatory neurons, In neurons = Inhibitory neurons, Oligo = Oligodendrocytes, Endo = Endothelial cells, VLMC = Vascular leptomeningeal cells, OPC = Oligodendrocyte progenitor cells c UMAP plot displaying unique cell types identified within the RSC. d Bar plot comparing the proportion of cell types found among the two groups (homecage, n = 4 and learning, n = 4). e Logistic regression results for predicting experimental groups using NEUROeSTIMator predicted activity. The colored dots represent the ß coefficients for predicted activity in the learning vs homecage experiment across major cell types. f Volcano plots demonstrating the top significant differentially expressed genes for the excitatory, inhibitory neuronal cell types, astrocytes, and oligodendrocytes. The genes with FDR < 0.05, absolute log2 fold change > 0.2 were considered significant and colored by cell type identity. g Upset plot illustrating the unique and overlapping upregulated genes across the major cell types identified within the RSC. h Sankey plot shows the common genes in both Xenium and Visium experiment that were significantly differentially expressed (red: upregulated, blue: downregulated) in the major cell types identified in Xenium analysis.

Within the RSC, we used the pre-defined marker gene expression in the Xenium gene panel to identify the different cell types. We identified eight major cell types within the RSC, including excitatory neurons, inhibitory neurons, astrocytes, oligodendrocytes, endothelial cells, vascular leptomeningeal cells (VLMCs), microglia, and oligodendrocyte progenitor cells (OPCs) (Fig. 2b, c and Supplementary Fig. 1). The proportion of cells in each of the clusters between homecage and learning samples was comparable between the two groups (Fig. 2d). Given the role of neurons (excitatory and inhibitory) and glial cells (astrocytes [54], oligodendrocytes [55], and microglia [56]) in memory consolidation, we primarily focused our analysis on these five major cell types in the RSC. We applied a deep learning computational model, NEUROeSTIMator [51], to identify activation patterns of RSC cells following learning in the SOR task. The NEUROeSTIMator analysis predicted higher activity in all cell types in the learning group compared to the homecage control group. Highest increases in predicted activity were observed in excitatory and inhibitory neurons following SOR learning (Fig. 2e). Next, we investigated the differentially expressed genes following learning in these five major RSC cell types. Among the 297 genes analyzed from the Xenium panel, a comparison between learning and homecage groups revealed 96 DEGs in excitatory neurons (32 up and 64 down), 23 DEGs in inhibitory neurons (20 up and 3 down), 43 DEGs in astrocytes (21 up and 22 down), 44 DEGs in oligodendrocytes (17 up and 27 down), and 14 DEGs in microglia (13 up and 1 down) (Fig. 2f, Supplementary Fig. 2, and Supplemental Table 4). Further, Gene Ontology (GO: molecular function) enrichment analysis revealed RNA pol II-specific DNA-binding transcription factors to be the most represented function across the five major cell types (Supplementary Fig. 3 and Supplementary Table 5). Using an UpSet plot, we examined the distinctly upregulated and downregulated genes after learning within these cell types in the RSC. Among the upregulated genes, Arc, Egr1, Nr4a1, Hspa1b, Nr4a3, Per1, and Dnajb1 were induced after learning in all the five major cell types (excitatory and inhibitory neurons, astrocytes, oligodendrocytes, and microglia), while Bdnf, Egr4, Bhlhe40, Trib2, Cbln4, Sik2, Dusp5, Dnajb5, and Tnfrsf25 were induced exclusively in excitatory neurons. We also found that neuronal activity response genes Fosb and Homer1 were exclusively upregulated in neurons, while Sgk1 and Xbp1 were exclusively upregulated in oligodendrocytes and astrocytes (Fig. 2g). Among the genes downregulated after learning, Cirbp, Dner, Dpy19l1, Lypd6, Slc17a7 and Garnl3 were downregulated in excitatory neurons, astrocytes, and oligodendrocytes. Id2, Igfbp5, Parm1, Cdh20, Nrep, and Hpcal1 were downregulated in excitatory neurons, and oligodendrocytes. Rbm3 was downregulated in excitatory neurons, inhibitory neurons, and oligodendrocytes. Zc3h6 was downregulated in excitatory and inhibitory neurons, astrocytes, and oligodendrocytes, while Plekha2 was downregulated in oligodendrocytes and microglia (Supplementary Fig. 4). Zinc finger CCCH containing 6 (Zc3h6) is predicted to have RNA binding and metal ion binding activity [57]. However, very little is known about Zc3h6 and its function in the brain. Comparing the learning-responsive DEGs from Visium and Xenium suggests that neurons in the RSC showed the highest overlap of learning-induced genes, particularly the upregulated genes in excitatory neurons (Fig. 2h).

To further identify the gene expression signature of active neurons after learning, we separated RSC excitatory and inhibitory neurons into active and non-active states based on Fos expression levels in the learning group. DEG analysis comparing Fos+ and Fos- excitatory neurons in the RSC revealed 42 DEGs, and only 7 DEGs in inhibitory neurons in the RSC (Supplementary Fig. 5, Supplementary Table 6). We found that the classical immediate-early genes Fos, Fosb, Nr4a1, Nr4a2, Bdnf, Egr1, Egr4, and Dusp6 were upregulated in Fos+ excitatory neurons compared to Fos- neurons in the learning group. Fos, Fosb, Fosl2, and Egr1 were upregulated in Fos+ inhibitory neurons (Supplementary Fig. 5). Thus, our spatial transcriptomic analysis using Xenium revealed a cell-type-specific signature of learning-induced genes in the RSC.

Learning induces IEG expression changes across layer-specific neurons of the RSC

Our learning-induced gene expression analysis in the RSC using the Visium platform identified a higher number of upregulated genes in the RSC than downregulated genes. Importantly, we found a robust learning-induced upregulation of genes commonly used as engram markers in the RSC. RSC excitatory neuronal sub-types exhibit spatially recognizable laminar structural features [1, 29]. Additionally, these layer-specific glutamatergic neurons show unique electrophysiological characteristics [58, 59] and receive distinct inputs from the hippocampus [60, 61]. We performed unsupervised clustering from both the groups (homecage and learning) to identify the neuronal sub-types localized across different layers in the RSC. Investigating the expression of marker genes within each cluster of excitatory neurons further allowed us to classify the major layer-specific excitatory neuronal sub-types (Supplementary Fig. 6 and 7). We identified seven major excitatory neuronal sub-types: layer 2/3 (L2/3), retrosplenial specific layer 2/3 (L2/3 RSP), layer 4 (L4), retrosplenial specific layer 4 (L4 RSP), layer 5 (L5), layer 6 (L6), and near projecting-subiculum (NP SUB) (Fig. 3a, b). Similarly, we also identified five major inhibitory neuronal populations (Pvalb, Sst, Vip, Lam5, and Sncg) along with glial cells (astrocytes, oligodendrocytes, microglia and OPC) in the RSC (Fig. 3a, c and d). The proportion of cells in each cluster, comparing homecage and learning samples, was comparable between the two groups (Fig. 3e). Further, NEUROeSTIMator analysis revealed higher predicted activity in all the RSC neuronal subtypes comparing learning and homecage (Fig. 3f), suggesting a global increase in neuronal activity following learning.

Fig. 3: Single-cell resolution spatial transcriptomic analysis reveals the induction of IEGs after learning within different classes of RSC neurons.

a UMAP exhibiting the layer-specific neuronal subtypes identified within the RSC. b–d Spatial map of b) excitatory neuronal sub-types c) inhibitory neuronal sub-types and d) glial sub-types within RSC. e Bar plot comparing the proportion of all the cell types found among the two groups (homecage and learning) in RSC. f Logistic regression coefficients for NEUROeSTIMator-predicted activity across excitatory and inhibitory neuronal sub-types, comparing the learning group to homecage. g, h Upset plot showing the unique and overlapping expression patterns of upregulated genes across different g) excitatory neuronal sub-type and h) inhibitory neuron sub-types within the RSC.

To examine the learning-induced gene expression within the different populations of neurons, we performed DEG analysis on the 297 genes from the Xenium panel on all the neuronal sub-types (Supplemental Table 7). We used an Upset plot to investigate the learning-induced genes upregulated in each of the seven major excitatory neuronal populations identified in the RSC (Fig. 3g). Genes upregulated in only one sub-type of excitatory neurons include Cbln4 upregulated in L2/3, Hspa1a in L2/3 RSP, Ankrd33b in L4, and Gadd45a in L6. Memory-related IEGs Arc, Egr1, Fos, Fosb, Nr4a1, Nr4a3, and Per1 were commonly upregulated across all the seven major excitatory neuronal cell types, while TNF-receptor superfamily-related gene Tnfrsf25 was upregulated in all the excitatory neuronal cell types except NP SUB. Nr4a2, Per2, Sik2, and Jdp2 were upregulated in L2/3, L2/3 RSP, L4 and L6. Unfolded protein binding factors Dio2 and Dnajb1 were upregulated in L2/3, L2/3 RSP, L4, L6, and NP SUB, while chaperones Sdf2l1 and Pdia6 were upregulated in L2/3 RSP and L6. MAPK pathway-related genes Egr3, Dusp6 were upregulated in L2/3, L2/3 RSP, L4, L5 and L6. Brain-derived neurotrophic factor and memory-related gene Bdnf was upregulated in L2/3, L2/3 RSP, and L4, and immediate early response gene Ier5 was upregulated in L2/3, L4, and L4 RSP (Fig. 3g, Supplemental Table 7). Among the inhibitory neurons, Arc and Nr4a3 were commonly upregulated across all four inhibitory neuronal populations, while Egr1 and Nr4a1 were upregulated in Pvalb, Sst, and Vip inhibitory neurons. Fos was upregulated in Sst and Vip inhibitory neurons, and Hspa5, Per1, and Dnajb1 were upregulated in Pvalb and Sst neurons (Fig. 3h, Supplemental Table 7). Therefore, our spatial transcriptomic analysis after learning suggests an overall induction of IEG expression in the major neuronal populations, particularly across excitatory neuronal layers.

Neuronal activation after learning is reduced in the RSC of a mouse model of ADRD

Long-term memory impairment and hyperphosphorylation of microtubule-associated protein tau (MAPT) are common in patients with ADRD [62]. We have previously shown that a tauopathy model of ADRD (rTg4510 mice: Tau-P301L) that expresses mutant human tau in excitatory neurons shows spatial memory impairment in the SOR task [53]. An immunofluorescence analysis showed hyperphosphorylation of tau in the RSC of Tau-P301L (rTg4510) mice (Fig. 4a–c). Next, we performed spatial transcriptomic analysis using Xenium from the RSC of Tau-P301L and control mice after learning in the SOR task (Fig. 4d). Using the same gene panel as we used for the wild-type mice (Figs. 2, 3), we identified the major cell types in the RSC of Tau-P301L and control mice based on marker gene expression (Fig. 4e, f, Supplementary Figs. 8, 9). The proportion of different cell types was similar between the two experimental groups (Tau-P301L and control mice) (Fig. 4g). We then utilized the NEUROeSTIMator tool to predict activity in different RSC cell types. We found that excitatory and inhibitory neurons of Tau-P301L mice exhibited reduced activity after learning compared to controls (Fig. 4h). Additionally, Fos expression was downregulated only in excitatory neurons of Tau-P301L mice compared to controls (Fig. 4i, j). Taken together, our results suggest reduced activation of RSC excitatory neurons after learning in this tauopathy model of ADRD.

Fig. 4: RSC of a mouse model of ADRD (Tau P301L) exhibit reduced neuronal activity following spatial learning.

a Schematics depicting the control and Tau-P301L mice. b Immunofluorescence of the RSC region of Tau-P301L mice using an antibody that detects hyperphosphorylation of tau at Ser202, Thr205 (AH36). c Mean fluorescence intensity (MFI) of phosphorylated tau expression within the RSC. Unpaired T test: t (4) = 7.920, p = 0.0014. n = 3 for both control and Tau-P301L groups. The data is shown as group mean ± SEM. d Schematics depicting the spatial learning task, the region of interest collected, and the processing for Xenium pipeline. Both Tau-P301L (n = 3) and control (n = 4) mice were trained in the SOR task. e Spatial map of major cell types identified in the RSC. f UMAP plot shows clustering of cells that are colored according to the major cell type identity. g Bar plot comparing the proportion of major cell types in RSC found in both the control-learning and Tau-P301L-learning groups. h Logistic regression coefficient estimate for NEUROeSTIMator-predicted activity in Tau-P301L learning mice compared to control learning. i Violin plot showing expression of Fos across all major cell types between Tau-P301L and control mice after learning. Excitatory neurons show reduced Fos expression (Seurat Wilcox test, adjusted P value 2.05E-05, log2FoldChange: −0.23) comparing P301L and control mice. j Spatial map of Fos mRNA expression in the excitatory neurons of both control-learning and Tau P301L-learning groups.

Chemogenetic inhibition of the RSC excitatory neurons after learning impairs long-term spatial memory

Our spatial transcriptomics data identified robust learning-induced gene expression changes in the excitatory neurons of the RSC in wild-type mice. Further, the tauopathy model of ADRD showed reduced transcriptional signature of neuronal activation within the excitatory neurons after learning in RSC. To corroborate these findings, we used a chemogenetic approach to selectively manipulate the excitatory neurons immediately after learning, a critical time window for memory consolidation. We manipulated the activity of the excitatory neuronal population using a viral-based inhibitory designer receptor exclusively activated by designer drugs (DREADD) construct, which reduces the excitability of the neurons [63]. The expression of the construct was restricted within the RSC, but the CA2 region also displayed mCherry labeling, possibly due to viral affinity [64] or potential connectivity between RSC and CA2 (Fig. 5a). Following training in the SOR task, animals were injected with clozapine-n-oxide (CNO, 2 mg/kg) (Fig. 5b). The virally expressed receptors respond specifically to CNO, allowing us to selectively block learning-induced activation of these neurons during the early time point immediately after learning. CNO treatment reduced the percentage of neurons expressing Fos, a marker of engram ensembles, within mCherry positive neurons (excitatory neurons infected by the DREADD virus) one hour after training in SOR, suggesting reduced neuronal activation in the CNO treated group (Fig. 5c, d). Importantly, in homecage animals no differences were observed in the number of Fos positive neurons between saline and CNO conditions, indicating that the CNO treatment did not affect the activity level of the excitatory neurons at baseline (Supplementary Fig. 10). Given the importance of engram ensembles in memory allocation during this period after learning [65], we aimed to test the long-term spatial memory when excitatory neurons within the RSC are selectively manipulated during this early time window. We found that mice injected with CNO immediately after training showed no preference towards the displaced object in SOR during the 24-h long-term memory test (Fig. 5e). In contrast, mice injected with vehicle (saline) showed increased preference towards the displaced object during the test session compared to CNO injected mice (Fig. 5e). This finding suggests that reduced RSC excitatory neuronal activity during memory consolidation impairs long-term spatial memory (Fig. 5e). Overall, our results show the transcriptional signature of spatial memory in the excitatory neurons of the RSC and further demonstrate the importance of activation of excitatory neuronal ensembles in the consolidation of long-term spatial memory.

Fig. 5: Activation of excitatory neurons after learning is essential for long-term spatial memory consolidation.

a Graphic demonstration of the viral injection site, the viral construct that was used, and a representative immunofluorescence image showing the resulting viral expression within the RSC 4 weeks after infusion. b Schematic for the behavioral paradigm utilized. c Representative immunofluorescence images demonstrating c-Fos protein expression within DREADD positive neurons for the saline and CNO treated groups. d Dot plot showing the percent of DREADD positive neurons which are also positive for c-Fos protein expression 1 h after learning in an SOR task. Unpaired t test: t (6) = 2.894, p = 0.0276. n = 4 for both vehicle and CNO conditions. The dat