Reciprocal molecular and cellular cholinergic working memory impairments in Alzheimer's disease model mice Authors: Chloe I Rybicki‐Kler, Isla A W Brooks, Izabela Jedrasiak‐Cape, Tao Deng, Cassandra Avila, Elizabeth Brouns, Alex Lekander, Omar J Ahmed Published: 2026 ABSTRACT AbstractINTRODUCTIONThe degeneration of basal forebrain cholinergic neurons is a key pathophysiological feature of Alzheimer's disease (AD). These cholinergic neurons target cortical neurons including those in the granular retrosplenial cortex (RSG) to facilitate essential cognitive functions. At the cellular level, acetylcholine supports working memory by inducing persistent firing that outlasts the stimulus, but how such cholinergic‐induced persistent firing is altered in AD remains unknown.METHODSUsing multiscale molecular and cellular analyses, we investigated neuron type–specific transcriptomic and physiological impairments of cholinergic persistent signaling in the RSG of 5xFAD mice.RESULTSCholinergic inputs to RSG were reduced in 5xFAD mice. Expression of the Chrm1 gene encoding M1 muscarinic receptors was also reduced in RSG neurons, coupled with impaired cellular persistent firing.DISCUSSIONThe loss of M1 receptors suggests that allosteric M1 modulators being considered for treatment of AD symptoms may not be as effective on key cell types, necessitating further multiscale investigation.Keywords: acetylcholine, after depolarization, carbachol, cognition, g protein, granular retrosplenial cortex, Calcium‐activated non‐selective cation current (ICAN), M1 receptors, memory, muscarinic receptors, persistent activity, regulator of g protein signaling, retrosplenial cortex, slow afterdepolarization, spatial navigation, spatial orientation BACKGROUND The cholinergic system is critical for successful attention, learning, memory, perception, and spatial navigation. 1 , 2 , 3 , 4 , 5 , 6 Cholinergic neurons in the basal forebrain project to, and alter the firing of, cortical neurons to support these cognitive functions; 7 , 8 , 9 , 10 , 11 for a review see Sarter et al. 12 and Sarter and Lustig 13 ). Dysfunction of these basal forebrain cholinergic neurons is one of several key pathophysiological features of Alzheimer's disease (AD), often occurring before the appearance of amyloid beta deposits. 14 , 15 , 16 This early degeneration of basal forebrain neurons and the downstream molecular and cellular consequences for postsynaptic cortical neurons have formed the basis of the longstanding cholinergic hypothesis of AD and led to the development of therapies targeting components of the cholinergic system (for a review, see Contestabile 17 and Fisher and Levey 18 ). Most current treatments aim to compensate for the loss of cholinergic inputs by boosting levels of acetylcholine (ACh) available at synapses, without factoring in changes in postsynaptic receptors and downstream signaling cascades, potentially explaining the limited success that current cholinergic medications have had thus far. 17 , 18 Rational design of improved therapies focused on the cholinergic system may benefit from a multimodal molecular and cellular understanding of changes in the downstream components of the cholinergic signaling pathway. ACh induces the persistent firing of cortical pyramidal cells, allowing neurons to stay depolarized and fire additional action potentials even after the stimulus of interest has ended. 19 , 20 , 21 , 22 , 23 This persistent firing is one of the cellular correlates supporting working memory functions and is dependent on postsynaptic M1 muscarinic receptors (encoded by the Chrm1 gene) as well as downstream calcium channels. 23 , 24 , 25 , 26 , 27 M1 receptors are g protein coupled receptors (GPCRs), specifically using the Gq subtype. Gq proteins themselves are regulated by a large family of proteins called regulators of g protein signaling (RGS), 28 highlighting the rich molecular machinery working together to support cholinergic‐dependent persistent firing. Despite its relevance as a key cellular mechanism supporting working memory, cholinergic persistence and its underlying molecular machinery has, to the best of our knowledge, yet to be examined in the context of AD. Here, we study the cellular and molecular changes in the cholinergic persistence signaling pathway in a brain region that is critical for learning, memory, navigation, and also for retaining information over long timescales: the granular retrosplenial cortex (RSG 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 ). The RSG is part of the default mode network and an important integrator of subcortical and higher order cognitive inputs. 37 , 38 , 39 , 40 , 41 , 42 The RSG is also one of the first regions to show hypoactivity in people with AD. 43 , 44 , 45 While recent work in animal models has started to focus on understanding RSG dysfunction in AD, 46 , 47 , 48 there remains a stark gap in knowledge regarding the molecular and physiological basis for retrosplenial hypometabolism in AD and the possible role played by impaired cholinergic signaling in this region. We have previously shown that pyramidal cells in layer 5 (L5) of RSG display robust cholinergic persistence and that this response is dependent on M1 receptor activation. 23 Here, using a mouse model of AD together with transcriptomic, physiological, and immunohistochemical analyses, we demonstrate that there is a simultaneous loss of cholinergic inputs to the RSG as well as significant reduction in the expression of the M1 receptor by postsynaptic L5 pyramidal cells, providing new, molecular insight into why simply boosting ACh levels may not always lead to more activation of postsynaptic neurons. We also investigate the functional cellular consequences of these changes and find that cholinergic persistent firing in L5 of RSG is decreased, despite several homeostatic changes in proteins regulating g protein signaling. Together, this work provides additional mechanistic insight into why RSG may become hypoactive in AD and highlights possible limitations of drugs designed to target M1 receptors. RESEARCH IN CONTEXT Systematic review: PubMed was used to review literature related to (a) retrosplenial dysfunction in preclinical and clinical Alzheimer's disease (AD), (b) basic science research on the subtypes of neurons in the retrosplenial cortex (RSC), and (c) the key components of the cholinergic persistent signaling pathways. This literature is cited where relevant. Systematic review: PubMed was used to review literature related to (a) retrosplenial dysfunction in preclinical and clinical Alzheimer's disease (AD), (b) basic science research on the subtypes of neurons in the retrosplenial cortex (RSC), and (c) the key components of the cholinergic persistent signaling pathways. This literature is cited where relevant. Interpretation: We discovered both molecular and cellular changes in how acetylcholine impacts retrosplenial neurons in a mouse model of AD. Importantly, we found that muscarinic M1 receptors—critical for cholinergic‐dependent working memory—are decreased in key retrosplenial cell types. This means that therapeutic drugs designed to target and activate M1 receptors may not be very effective in treating working memory as these receptors are decreased in number. Interpretation: We discovered both molecular and cellular changes in how acetylcholine impacts retrosplenial neurons in a mouse model of AD. Importantly, we found that muscarinic M1 receptors—critical for cholinergic‐dependent working memory—are decreased in key retrosplenial cell types. This means that therapeutic drugs designed to target and activate M1 receptors may not be very effective in treating working memory as these receptors are decreased in number. Future directions: This work provides preclinical and translational research into the mechanisms underlying retrosplenial circuit deficits in AD and motivates additional multiscale molecular and cellular analyses to identify alternative cholinergic signaling pathways to target with rational therapeutics. Future directions: This work provides preclinical and translational research into the mechanisms underlying retrosplenial circuit deficits in AD and motivates additional multiscale molecular and cellular analyses to identify alternative cholinergic signaling pathways to target with rational therapeutics. METHODS Animals All animal use and housing were approved by the University of Michigan Institutional Animal Care and Use Committee. The 5xFAD (034848‐JAX) transgenic mouse model of AD was used for all experiments, with 5xFAD(+/tg) mice (henceforth referred to simply as 5xFAD mice) compared to wildtype (WT) littermates. All mice were maintained on C57BL6 backgrounds. Male and female mice > 9 months of age were used for all experiments, except an additional control test of cholinergic cellular persistent firing that was also run in a group of 4‐month‐old 5xFAD and WT littermates and revealed no change in cholinergic persistent firing at the younger time point. Vesicular ACh transporter immunohistochemistry Animals were deeply anesthetized (in the dark phase of the animals’ circadian cycle) with isoflurane and transcardially perfused with 4% paraformaldehyde (PFA). Brains were dissected and post‐fixed overnight in 4% PFA at 4°C and then transferred into 30% sucrose solution for at least 72 hours or until sectioning (no more than 1 week). Brains were sectioned at 80 µm using Leica SM2000 R sliding microtome and stored in 0.1 M phosphate‐buffered saline solution until further processing (up to 2 weeks) For vesicular Ach transporter (VAChT) immunohistochemistry (IHC), slices were first blocked for 1 hour at room temperature in tris‐buffered saline (TBS) containing 5% normal donkey serum (NDS) and 0.4% TritonX. Primary antibody was then added (1:500 dilution, Synaptic Systems #139 008) and slices incubated overnight on a shaker at 4°C. Afterward, slices were washed for 3 × 5 minutes with TBS and incubated for 4 hours at 4°C in fresh blocking buffer containing secondary antibody (1:250 dilution, Invitrogen #A32795TR) and NeuroTrace (NT, 1:200 dilution, Invitrogen #N21479). NT was added to visualize cytoarchitecture and lamination of cortical tissue. Finally, slices were washed for 3 × 10 minutes in TBS, mounted using Fluoromout‐G (Southern Biotech #0100‐01), and left overnight to set before imaging. VAChT imaging Cortical slices stained for VAChT expression (AP range of −2.8 to −3.1 from bregma following Paxinos Mouse Brain Atlas coordinates) were imaged using a Zeiss Imager M2 equipped with an LSM 700 confocal system. For each slice (n = 14, one per animal) a RSG cortical column with clear lamination was selected for imaging. Care was taken to select similar areas midway through the dorsoventral axis of RSG. Once the area was chosen, an 8 µm z stack of the cortical column was acquired (z step 0.25 µm, frame size 1024 × 1024 px) in two channels: NT for cytoarchitecture, and AlexaFluor 647 for VAChT expression. VAChT quantification Z stacks of VAChT expression were acquired as described above and Z projected (maximum intensity) in ImageJ for further analysis. Cortical column widths were then adjusted to crop out tissue imperfections or noise, resulting in cortical column widths averaging 364 µm (11 columns width of 375 µm, one 467 µm, one 300 µm, and one 200 µm, n = 14 total). To further reduce the noise, the NT channel was subtracted from the VAChT channel using ImageJ Image Calculator and only the resulting subtracted VAChT expression was used. Each image was manually curated to ensure that minimal signal was lost during pre‐processing. Manually curated cortical column VAChT expression was then turned into a binary (per pixel) image with VAChT signal being white and background black for each pixel and quantified per layer using previously described RSG layer demarcations. 41 Slice preparation Retrosplenial slice preparation was similar to our previously described methods. 23 , 41 , 49 Briefly, mice were deeply anesthetized (in the dark phase of the animals’ circadian cycle) using isoflurane and decapitated. The brain was rapidly removed and immersed in an ice‐cold, carbogen‐saturated, high‐sucrose solution. Three hundred µm thick slices were cut using a Leica 1200VT vibratome and transferred into high‐magnesium artificial cerebrospinal fluid (aCSF, prepared as below but with an additional 150 µL MgCl per 150 mL added) held at 32°C. After 20 minutes, the recovery chamber was allowed to cool to room temperature and slices were held there for the remainder of the experiment. Experimental recordings began at least 45 minutes after this transition. Whole‐cell recording Slices were transferred from the recovery chamber into a submerged recording chamber and continuously perfused at 3 mL/minute with 32°C, carbogen‐saturated aCSF containing: 126 mM NaCl, 1.25 mM NaH2PO4, 26 mM NaHCO3, 3 mM KCl, 10 mM dextrose, 1.20 mM CaCl2, and 1 mM MgSO4. Cells were visualized using an Olympus BX51WI microscope with Olympus 60x water immersion lens and Andor Neo sCMOS camera (Oxford Instruments). Patch pipettes were pulled using a horizontal puller and had resistances between 4 and 7 MΩ. Recordings were done using a potassium gluconate intracellular solution containing: 130 mM K‐gluconate, 2 mM NaCl, 4 mM KCl, 10 mM HEPES, 0.2 mM EGTA, 0.3 mM GTP‐Tris, 14 mM phosphocreatine‐Tris, 4 mM ATP‐Mg (pH 7.25, osmolarity 290 mOsm), and 0.5% biocytin to fill cells for reconstruction. Patch clamp recordings were obtained with Multiclamp 700B and Digidata 1440A (Molecular Devices). Neurons from layer 5a of RSG were chosen based on their shape and position in the superficial half of L5, and all recordings were performed in current clamp mode. Only cells with a resting membrane potential of ≤ −55 mV were used for the analysis. 41 Cells were adjusted for series resistances and the voltage adjusted (in current clamp) to −65 mV at the start of each recording protocol (the voltage was thus allowed to change in current clamp mode in response to current injection or drug exposure). Recordings were not liquid junction potential corrected. Each recorded cell was first characterized for intrinsic and firing properties and subsequently tested for persistence (see Brennan et al. 41 , 49 and Jedrasiak‐Cape et al. 23 for detailed testing protocols). Pharmacology After cell type was confirmed by physiology, the persistence protocol was run. The protocol consisted of repetitive sweeps of a baseline period followed by three 250 ms current injections, each at an intensity at least twice that of the rheobase and spaced 13 seconds apart to initiate persistent firing, ending with a short −5 to −15pA current step to monitor input resistance changes. All cells were biased toward –65 mV before persistence recording began, and for the duration of the experiment the resting membrane potential was not readjusted. Cells were recorded for ≈ 2 minutes in aCSF before the non‐selective muscarinic receptor agonist, carbachol (40 µM; CCh), was washed into the bath, and responses were recorded for at least 12 sweeps. Recordings of intrinsic properties in the presence of carbachol were performed after persistence recordings, at least 10 minutes after carbachol was initially added to the bath and with continuous perfusion of carbogenated aCSF containing 40 µM CCh. Analysis of intrinsic physiological properties All intrinsic properties were analyzed in Python as previously described. 50 Briefly, spike amplitude, width, threshold, and adaptation ratio were calculated from the first spike generated during a 600 to 1000 ms depolarizing current step (or, during the first step containing a spike train of ≥ 10 Hz to calculate adaptation ratio). Rheobase and latency were calculated from recordings of small, 1000 ms depolarizing current steps increasing by 5 mV per sweep. These trials began at a current that elicited no action potentials and continued until three consecutive sweeps containing action potentials were recorded. Input resistance, capacitance, and decay time constant tau were derived from recordings wherein each cell was delivered a small negative current step resulting in ≈ 4 mV deflection in membrane potential. Sag was calculated from larger hyperpolarizing current injections, in which cells reached a Vm < −80 mV. Physiology index (PI) is a metric that helps to distinguish distinct cell populations in L5 of cortex based on their sag and input resistance, and was calculated here following the procedures developed by Allan Gulledge. 51 Analysis of cholinergic persistent activity Files were loaded into Easy Electrophysiology (v2.5.2) and action potentials were identified using the “AP Counting” function. Only action potentials generated outside of depolarizing current steps were counted toward the total number of spikes observed during cholinergic persistent firing. The frequency of cholinergic persistent firing was calculated as the number of action potentials generated between current steps divided by the duration of that period in seconds. Cells that were classified as responding to carbachol included those that displayed cholinergic persistent firing (see definition below) as well as those that developed slow afterdepolarizations (sADPs) without persistent firing. 52 We defined sADPs as a depolarization event of at least 1 mV, not resulting in action potentials, occurring no less than 0.5 seconds and no more than 1.5 seconds after cessation of evoked action potentials. Cells exhibiting cholinergic persistent activity were identified as follows: in at least one sweep during which carbachol was present in the bath, three or more action potentials must be generated outside the depolarizing current steps. Antidromic action potentials were not counted toward cholinergic persistent activity. Statistical analysis of electrophysiological properties Statistical analyses were performed in GraphPad Prism (v10.6.1). All data were tested for normal or non‐normal distributions; comparisons of WT versus 5xFAD intrinsic properties were performed using either a Welch t test (when all data were normally distributed) or Mann–Whitney test (if one or more groups were non‐normally distributed). Comparisons of intrinsic properties in plain aCSF versus during 40 uM CCh exposure were performed using either a paired t test (for normally distributed groups) or a paired Wilcoxon test (for non‐normally distributed groups). Tissue dissection and processing for 10x RNA sequencing Mice were deeply anesthetized using isoflurane and decapitated. The brain was rapidly removed into ice cold aCSF (same composition as for patch clamp recordings) and then placed in a brain matrix. One mm thick coronal sections were cut with a razor blade at distances −1.7 to −2.9 mm from bregma. Each slice was transferred to a glass culture dish filled with fresh ice cold aCSF. Slices were then cut in half width‐wise just below the hippocampus and the lower half of the slice was discarded. To help limit the dissection to only the RSG region, hippocampus was removed from the top half of the slice, leaving only a strip of cortex, which included the area of interest. Then, a cut was made along the boundary of the RSG/dysgranular retrosplenial cortex border in each hemisphere, and the dissected RSG from both hemispheres was combined and immediately flash‐frozen in a polymerase chain reaction tube. Remaining cortical regions from the same strips were saved as controls for cell sorting. All tubes were kept at −80°C until further processing. For the subsequent nuclei isolation, microdissected RSG regions from three 5xFAD mice and three WT littermates were processed using standard protocols. 53 Briefly, the tissue samples were homogenized in a lysis buffer using a glass dounce, filtered through a 30 µm MAC strainer, and centrifuged at 500 g for 5 minutes and the pellet resuspended in a wash buffer. Nuclei were filtered and strained again and resuspended in a wash buffer with propidium iodide and went through fluorescence‐activated cell sorting on a MoFlo Astrios Cell Sorter. Afterward, the sample was centrifuged at 100 g for 6 minutes and the pellet resuspended in a wash buffer for a concentration of 750 to 1200 nuclei/µL. The sample was then submitted for single‐nucleus RNA sequencing (snRNA‐seq) processing. 10x RNA‐seq analysis Raw RNA‐seq data was generated by the University of Michigan Advanced Genomics Core, delivered as FASTQ files containing raw reads. We ran Cell Ranger 7.1.0 with default parameters, using the GRCm38 RefSeq assembly (https://www.ncbi.nlm.nih.gov/datasets/genome/GCF_000001635.20/) as a reference genome, to generate counts matrices. The initial cell counts (defined as barcodes that Cell Ranger included in its filtered matrix of putative cells) were 8979 in the 5xFAD sample, and 7184 in the WT sample. All analyses related to clustering and quality control were performed using Scrublet in Python 3.11.2. First, in both datasets, we used Scrublet 54 to remove putative doublets. We then removed all cells with < 1500 total counts or 750 total detected genes, as well as cells with > 50,000 counts. We additionally excluded cells whose total mitochondrial gene content was > 0.5%, as extremely little mitochondrial RNA is expected in single‐nucleus sequencing. Cholinergic innervation of the RSG is impaired in the 5xFAD mouse model of Alzheimer's disease. A, Pattern of expression of VAChT in the RSG of WT mice is characterized by a dominant band of cholinergic axons in L1, increasing in intensity toward pia. B, Comparison of VAChT expression in RSG cortical columns from WT and 5xFAD animals, rotated 90 degrees from (A). Note the large drop in L1 expression which has the highest density of cholinergic terminals. C, Mean + standard error of the mean VAChT expression of RSG cortical columns from (B; n = 7 mice for each group). L1 is the dominant band for both groups, but the amount of VAChT in L1 is noticeably smaller in 5xFAD animals compared to WT animals. D, Expression in L1 relative to expression across the whole cortical column reveals a significant deficit in the number of cholinergic terminals in L1 in 5xFAD mice (n = 7 mice, rank sum = 31, p = 0.0041). HIPP, hippocampus; L1, layer 1; RSD, dysgranular retrosplenial cortex; RSG, granular retrosplenial cortex; VAChT, vesicular acetylcholine transporter; WM, white matter; WT, wild type. We next performed a round of initial clustering to broadly identify non‐neuronal populations and exclude them from downstream analyses. We performed principal component analysis (PCA) on the top 2000 highly variable genes (identified with scanpy's highly_variable_genes function, with flavor = “seurat” supplied), and used the Harmonypy package 55 for batch integration. We used scanpy's neighbors function, with n_neighbors = 15, and leiden, with resolution = 1.0, to compute a nearest neighbor graph and generate the initial clustering. We then aligned these clusters to the Allen Institute's Whole Mouse Brain 10xv3 (WMB‐10xv3) dataset 56 to assign putative neuronal and non‐neuronal identities. We did this by computing mutual nearest neighbors (MNNs), which represent pairs of cells of the same type between two biological samples, 57 between our dataset and the WMB‐10xv3 dataset. We assigned identities at the “class” level, the broadest classification in the WMB dataset, based on the class from which the majority of a cluster's MNNs were derived. Any of our clusters that aligned with non‐neuronal WMB‐10xv3 classes were excluded. The RSG contains the same set of glutamatergic neuronal clusters in WT and 5xFAD mice. A, UMAP projection of all RSG glutamatergic neurons, color coded by genotype. UMAP projections of WT (B) and 5xFAD (C) glutamatergic neurons, color coded by cell type. Note the existence of an identical set of clusters, including the highlighted L5a RSG cluster in each genotype. GLU, glutamatergic; L2/3 IT, layer 2/3 intratelencephalic cells; L2/3 LR, layer 2/3 low rheobase cells; L5a RSG, layer 5a cells restricted to retrosplenial granular cortex; L5 IT, layer 5 intratelencephalic cells; L5 ET, layer 5 extratelencephalic cells; L5 NP_1, layer 5 near‐projecting cells cluster 1; L5 NP_2, layer 5 near projecting cells cluster 2; L6 IT, layer 6 intratelencephalic cells; L6 CT, layer 6 corticothalamic cells restricted to retrosplenial granular cortex; L6 CT mCTX, layer 6 corticothalamic medial cortical cells; L6b, layer 6b cells; RSG, granular retrosplenial cortex; UMAP, Uniform Manifold Approximation and Projection; WT, wild type. We then performed a second, final clustering on only the putatively neuronal cells in our dataset. We computed highly variable genes, PCA, and Harmony batch integration as above. We additionally applied the leiden clustering algorithm, but this time performing clustering in a series of hierarchical steps to be more sensitive to the fine‐grain scale of neuronal clustering, rather than attempting to broadly assign neuronal and non‐neuronal identities as before. In this methodology, the leiden algorithm is run with a much lower “resolution” parameter, 0.08, and any clusters whose highly variable gene correlation exceeds 0.95 are recombined at each step. Clustering proceeds by hierarchically splitting the dataset until no new clusters are generated. As above, we again used the WMB‐10xv3 dataset to assign identities to our clusters, this time naming our clusters according to the “supertype” in the WMB‐10xv3 dataset with which they aligned, or with previously defined names for transcriptomic RSG cell types in the case of L2/3 low rheobase (LR) and L5a RSG cells. 58 At the end of clustering and non‐neuronal cell exclusion, our total cell counts were 3591 in the 5xFAD sample and 3434 in the WT sample. Differential expression analysis Statistical analysis of differential expression between WT and 5xFAD clusters, as well as between L5a and L5 extratelencephalic (ET) neurons within the WT condition, were carried out using the PyDESeq2 Python library. 59 P values are reported as the “padj” statistic directly output by the PyDESeq2 pipeline, and log‐fold change values are reported from the “Log2FoldChange” metric. RESULTS Layer‐specific loss of cholinergic inputs to RSG in a mouse model of AD Loss of cholinergic neurons in the basal forebrain and the consequent cholinergic denervation of the cortex are one of the hallmarks of AD. Decline of cholinergic terminals has also been described in the RSG after electrolytic lesions of the basal forebrain. 60 , Cholinergic persistence signaling pathway transcriptomic changes across retrosplenial cell types in 5xFAD mice. A, Schematic representation of cholinergic persistence signaling pathway. B–E, left, Dot plots of gene expression counts in WT animals, (right) corresponding heatmaps of log fold change in expression of those genes between WT and 5xFAD, by cluster, of muscarinic receptors (B), relevant G‐proteins (C), downstream ion channels (D), and RGS genes (E). For dot plots, dot size corresponds to the percentage of cells expressing the gene, and dot color intensity reflects the mean number of transcripts of that gene. For heatmaps, blue indicates lower expression in 5xFAD relative to WT, and red indicates higher expression. GLU, glutamatergic; L2/3 IT, layer 2/3 intratelencephalic cells; L2/3 LR, layer 2/3 low rheobase cells; L5a RSG, layer 5a cells restricted to retrosplenial granular cortex; L5 IT, layer 5 intratelencephalic cells; L5 ET, layer 5 extratelencephalic cells; L5 NP_1, layer 5 near‐projecting cells cluster 1; L5 NP_2, layer 5 near projecting cells cluster 2; L6 IT, layer 6 intratelencephalic cells; L6 CT RSG, layer 6 corticothalamic cells restricted to retrosplenial granular cortex; L6 CT mCTX, layer 6 corticothalamic medial cortical cells; L6b, layer 6b cells; RGS, regulators of g protein signaling; RSG, granular retrosplenial cortex; WT, wild type. We first investigated cholinergic innervation in the RSG and whether it was altered in 5xFAD mice relative to their WT littermates. IHC using VAChT, a known marker of cholinergic neurons and synaptic terminals, 61 revealed a consistent pattern of cholinergic synapses in RSG in WT animals (Figure 1A). Cholinergic axons were present throughout the cortical column but formed dense bands in upper layer 1 (L1) and, to a lesser degree, in the bottom of layer 3 (Figure 1A, B left panel). Quantification of VAChT fluorescent labeling confirmed this pattern and further revealed increasing L1 density of expression from the bottom to top of this layer (Figure 1C, black WT trace). This is consistent with work in somatosensory cortex showing that L1 ACh release regulates synaptic integration by key L5 pyramidal neurons. 62 Having established a consistent pattern of WT RSG cholinergic innervation, we next compared it to the RSG of the 5xFAD mouse model of AD. VAChT labeling still showed a similar overall pattern with L1 dominating cholinergic expression (Figure 1B right, 1C red trace). However, a comparison of the fraction of VAChT concentrated in L1 relative to expression across L1 to L5 revealed a sharp decline in 5xFAD versus WT (Figure 1D, n = 7 animals for each genotype, ranksum = 31, p = 0.0041). This suggests not only a deficit in cholinergic innervation in general, but a targeted, localized denervation of specifically L1, the main source of cholinergic inputs into RSG. Widespread cell type–specific transcriptomic alterations in cholinergic persistent signaling genes, including decreased expression of M1 receptor gene We next aimed to characterize cell type–specific changes in key genes related to cholinergic signaling in the retrosplenial cortex in the 5xFAD model of AD. We have previously identified 10 to 11 glutamatergic cell types in the RSG of mice and rats and shown that these cell types are remarkably well conserved across millions of years of evolution. 58 We conducted similar transcriptomic analyses here, generating a 10x snRNA‐seq dataset of both WT and 5xFAD RSG. We observed a nearly identical set of glutamatergic cell types in RSG in this dataset as in our previous WT dataset. Importantly, all clusters in this new dataset contained cells from both the WT and 5xFAD conditions, emphasizing that the same cell types are reliably found in both genotypes (Figure 2). We next analyzed changes in key cholinergic persistence signaling genes across all these glutamatergic principal neuronal types (Figure 3 and Table S1 and S2 in supporting information). We focused on muscarinic receptor genes, coupled g proteins, RGS, and downstream ion channels that help to generate the calcium currents that support cholinergic‐induced persistent firing. 23 Several changes were observed across cell types, including a decrease in Chrm1, the gene encoding the M1 receptor, in most clusters. Perhaps the most notably consistent change, however, was seen in the gene Rgs4, which showed severe downregulation in the 5xFAD condition across nearly all clusters. To better understand these changes, we carefully analyzed a cell type that is also known to show physiological persistent activity in the presence of cholinergic agonists: the L5a RSG cell type. 23 , 58 This RSG‐specific cell type is evolutionarily conserved between mice and rats and is easily identifiable within mouse RSG by its selective expression of the gene Scnn1a. 58 We confirmed this expression pattern in the present dataset for both WT and 5xFAD mice (Figure S1 in supporting information). Examining this cell type's differentially expressed cholinergic signaling genes between the WT and 5xFAD conditions, we observed significantly reduced Chrm1 gene expression in 5xFAD (Figure 4A–C). Considered alone, this reduction in expression could lead to impairment of M1‐dependent cholinergic signaling and persistent activity; however, strong and possibly compensatory reductions in expression were also seen in the genes Rgs4 and Rgs17 (Figure 4D). These genes encode regulators of g protein signaling, molecules which negatively modulate the activity of Gq proteins, and hence their reduction in number can be viewed as a compensatory change in the opposite direction of decreased Chrm1 expression. Cholinergic persistence signaling pathway transcriptomic changes specific to L5a RSG neurons in 5xFAD mice. A, UMAP projection of all glutamatergic neurons, highlighting the L5a RSG cluster. B, Schematic representation of the cholinergic persistence signaling pathway, highlighting the observed decrease in Chrm1 (M1) and increase in Trpm4 expression. C, D, Bar graphs showing, in L5a RSG neurons, log fold change of key cholinergic persistence genes (C) and RGS genes (D) between WT and 5xFAD. Positive values indicate higher expression in 5xFAD relative to WT. Asterisks represent p values (* p < 0.01; ** p < 0.001; *** p < 0.0001). L5a RSG, layer 5a cells restricted to retrosplenial granular cortex; RGS, regulators of g protein signaling; RSG, granular retrosplenial cortex; UMAP, Uniform Manifold Approximation and Projection; WT, wild type. Together, these results suggest that while the loss of Chrm1 gene expression is likely to impair persistent activity, the possibility remains of compensatory effects due to Rgs4 and Rgs17 downregulation. To more concretely define the physiological consequences of these transcriptomic changes, we next conducted direct electrophysiological study of persistent firing in L5a RSG neurons in 5xFAD and WT mice. Cholinergic‐dependent persistent firing is impaired in 5xFAD RSG neurons We next focused on physiological recordings from L5a RSG cells to ascertain whether, and how, transcriptomic changes seen in these cells reflect their ability to fire persistently in the presence of cholinergic agonists. These L5a RSG neurons express a surprising combination of canonically intratelencephalic (IT) and ET marker genes and repeatedly separate into a distinct transcriptomic cluster from the L5 ET RSC cells, as confirmed in both WT and 5xFAD in our present work (Figure 2; Table S3, 23 , 63 , 64 , 65 in supporting information). Based on spatial transcriptomics, L5a RSG cells and IT cells are found in L5a of the RSG, while ET cells are largely restricted to L5b. 23 Therefore, to target neurons corresponding to the L5a RSG transcriptomic cluster, we patched L5a cells and identified a way to physiologically separate these neurons from the other pyramidal cell population in L5a IT cells (Figure S2 23 , 41 , 58 in supporting information). Several features distinguished the two cell types: L5a RSG cells displayed a prominent sag potential when hyperpolarized, reflecting high expression of hyperpolarization‐activated cyclic nucleotide‐gated (HCN) channels (Figure S2B, top). Their action potential amplitudes decreased and then increased (Figure S2C–D, top) when spike trains were generated using large current steps (≥ 300 pA). Conversely, L5 IT cells showed very little sag during hyperpolarization (Figure S2B, bottom), consistent with prior characterizations of IT cells in other cortical regions. 51 , 66 The action potential amplitudes of IT cells did not change distinctly during spike trains generated at high current injections (Figure S2C–D, bottom), in contrast to what was seen in L5a RSG neurons. A metric called the PI, developed by Gulledge et al., uses sag and input resistance to distinguish IT and ET cells in other cortical regions. 51 We found that the PI also reliably divided the L5a cells in RSG into two categories, with IT cells having very low PI values and putative L5a RSG cells having significantly higher PI values (Figure S2E). Once L5a RSG cells were identified by their intrinsic physiology (Table S4 in supporting information provides a comparison of L5a RSG cells’ intrinsic physiology properties in WT vs. 5xFAD mice), each cell was tested for cholinergic persistent firing (persistence) during bath application of 40uM carbachol in 5xFAD mice and WT littermates (Figure 5). A schematic of the persistence protocol is shown in Figure S3 in supporting information. Consistent with prior work, we confirmed that cholinergic‐induced persistent firing was dependent on muscarinic M1 receptors (Figure S4 in supporting information). The majority (77%) of WT L5a RSG neurons showed persistent responses and/or their subthreshold correlate, sADP; Figure 5B, D). Significantly fewer 5xFAD L5a RSG neurons showed such sADP and persistent responses (chi‐squared = 4.526, p = 0.0334; Figure 5C, D). Importantly, these experiments were done using the same bath‐applied concentrations of the cholinergic agonists and thus are likely to reflect the decreased M1 receptor expression revealed in the transcriptomic analyses. This impairment in persistent firing was not seen in a younger 4‐month‐old cohort of WT and 5xFAD animals (Figure S5 and S6 in supporting information). Cholinergic‐dependent persistent firing of retrosplenial neurons is impaired in aged 5xFAD mice. A, Cholinergic‐dependent persistence is assessed via intracellular slice recordings of L5a RSG neurons before and during wash‐in of the cholinergic agonist CCh (40 uM). B, Bath application of CCh generates sADPs and persistent firing in a representative WT L5a RSG neuron. C, Bath application of CCh fails to elicit either sADPs or persistent firing in a representative 5xFAD L5a RSG cell. D, Proportion of L5a RSG cells responding with either sADP or persistence in WT (N = 26, 77% responding) and 5xFAD (N = 14, 43% responding) mice. Chi‐squared = 4.526, p = 0.0334. ACSF, artificial cerebrospinal fluid; CCh, carbachol; L5a RSG, layer 5a cells restricted to retrosplenial granular cortex; sADP, slow afterdepolarization potential; WT wild type. DISCUSSION The cholinergic hypothesis relates to pronounced dysfunction and degeneration of basal forebrain cholinergic neurons as an early pathophysiological feature of AD that, together with other pathology, precipitates many of the cognitive symptoms associated with AD. 17 , 18 Here, we have explored the mechanisms of cholinergic impairments as they relate to cholinergic inputs to, and processing in, neurons of the RSG in a model of AD. We find impairments at multiple levels: First, reduced VAChT labeling in RSG, indicative of reduced cholinergic inputs to the RSG, is consistent with the cholinergic hypothesis. A second impairment is decreased expression of M1 muscarinic receptors that are essential for cholinergic‐dependent persistent firing of postsynaptic neurons and myriad additional functions. 18 This decrease is also accompanied by changes that appear to be compensatory, with prominent decreases in transcript levels for regulators of g‐protein signaling (Rgs4 in particular), which would non‐specifically increase Gq signaling downstream of M1 and other receptors. The third impairment is cholinergic‐induced persistent firing in a key retrosplenial cell type, consistent with the transcriptomic decreases in M1 receptor expression. In the rest of this discussion, we explore the importance of retrosplenial failure in AD, and the therapeutic implications of these mechanistic observations, together with the limitations of this work and necessary next steps on the path toward translational relevance. Cholinergic signaling is important for retrosplenial function. ACh is released in the RSC during active spatial navigation. 6 Indeed, compared to baseline, navigation leads to a larger increase in ACh levels in the RSC than in the hippocampus. 6 Consistent with this involvement of retrosplenial ACh in spatial navigation and memory, infusion of cholinergic antagonists into RSC can impair spatial performance, 67 suggesting that a similar loss of cholinergic inputs, together with the postsynaptic molecular and cellular changes reported in the present work, may contribute to impaired RSC function in AD. The RSC receives input from basal forebrain cholinergic sources including the horizontal limb of the diagonal band of Broca (H‐DBB), the medial septum (MS), and the nucleus basalis of Meynert (NBM). 68 , 69 In WT animals, this leads to a unique terminal lamination, with high density labeling in L1 and at the L3/5 boundary in the RSG, as confirmed by our VAChT observations (Figure 1). Our results show that this pattern is altered in AD model mice, with specific decreases in cholinergic inputs to L1 of RSG. L5a RSG neurons send their apical dendrites to L1, 23 , 41 and would thus be directly impacted by the loss of cholinergic inputs in AD. Most people with AD suffer from pronounced spatial disorientation, an inability to link external landmark cues with internal navigational goals. 70 , 71 , 72 This results in an increased likelihood of getting lost, even when navigating familiar environments. 71 , 72 Strikingly, the same disorientation symptoms are seen in non‐AD patients with specific lesions, due to a stroke or hemorrhage, of the RSG. 73 , 74 , 75 The present results suggest the hypothesis that the combination of decreased cholinergic inputs, decreased M1 receptor expression in principal neurons, and impaired persistent firing may contribute to RSG dysfunction in AD, and by extension contribute to the associated spatial disorientation symptoms. This study has several limitations that need to be addressed before taking any translational steps based on the present findings. First, our results are based on one model of AD, and it will be important to use the same multimodal methods to study the RSG and cholinergic signaling in other models of AD. Second, while the present work used both male and female animals, it was not powered to study sex differences, and future work will need to investigate sex differences in RSG cholinergic signaling. Third, similar methods need to be applied to other neocortical and hippocampal circuits to confirm if the transcriptomic and cellular observations seen in RSG are seen elsewhere or if they are specific to RSG processing. Finally, the transcriptomic work can be readily extended to human brain bank tissue to investigate if there are similar cholinergic signaling changes in the RSG and other regions at different stages of human AD disease progression. In conclusion, we have identified multiple failure points in the cholinergic signaling pathway in the retrosplenial cortex in AD model mice. Chief among these is the loss of muscarinic M1 receptors that are the target of several allosteric modulator drugs that are currently being considered for treatment of AD symptoms. 18 If M1 receptors are being reduced as part of the pathophysiology of AD disease progression, then there would be reduced target receptors for such drugs to bind to, limiting utility. Alternative downstream components of M1‐dependent cholinergic signaling, such as antagonists of RGS4 proteins, may act as useful alternative therapeutic targets in the search to restore cholinergic balance in AD. CONFLICT OF INTEREST STATEMENT The authors declare no conflicts of interest. Author disclosures are available in the supporting information. CONSENT STATEMENT Human subjects were not involved in this preclinical study. Supporting information Supporting Information: alz71490‐sup‐0001‐tableS1.xlsx Supporting Information: alz71490‐sup‐0002‐tableS2.xlsx Supporting Information: alz71490‐sup‐0003‐tableS3.xlsx Supporting Information: alz71490‐sup‐0004‐SuppMat.docx Supporting Information: alz71490‐sup‐0005‐SuppMat.pdf ACKNOWLEDGMENTS This work was supported by National Institute of Mental Health (R01MH129282); National Institute of Neurological Disorders and Stroke (R34NS127101, P50NS123067, T32NS007222, T32NS076401); National Institute on Aging (P30AG072931, F31AG087629); Alzheimer's Association (>AARG‐NTF‐21‐846572); National Institute on Deafness and Other Communication Disorders (T32DC000011); Berger Endowment (G102865/321033) REFERENCES Associated Data Supplementary Materials Supporting Information: alz71490‐sup‐0001‐tableS1.xlsx Supporting Information: alz71490‐sup‐0002‐tableS2.xlsx Supporting Information: alz71490‐sup‐0003‐tableS3.xlsx Supporting Information: alz71490‐sup‐0004‐SuppMat.docx Supporting Information: alz71490‐sup‐0005‐SuppMat.pdf