This study employed all female mice. The THY-Tau22 transgenic mouse model was generated to model AD-like tau pathology that is associated with learning and memory deficits (Schindowski et al., 2006). Tau22 mice overexpress the 4R tau RNA mutated at G272V and P301S and develop, mainly in the hippocampus, aggregation of hyperphosphorylated tau in addition to progressive tau-related neuropathology. In Tau22 mice, hyperphosphorylated and aggregated forms of tau first appear in the CA1 subfield and are present throughout the hippocampus at the peak of pathology at 12 months (Schindowski et al., 2006). Tau22 mice display learning and memory deficits and, although long-term potentiation (LTP) is intact, they show changes in NMDA-dependent long-term depression (LTD), and hippocampal synaptic plasticity (Schindowski et al., 2006; Van der Jeugd et al., 2011).

TauKO mice are meant to mimic the loss of tau protein from its normal physiological compartments (Tucker et al., 2001). The behavioral findings in the TauKO model include loss of contextual and cued fear conditioning, but normal motor, exploratory, and anxiety behaviors (Ahmed et al., 2014). Our TauKO mice display better spatial learning than controls in the water maze test. Additionally, in electrophysiological tests TauKO mice showed no change in basal synaptic transmission and paired-pulse facilitation from the CA1 hippocampal region. However, LTP, but not LTD, showed severe deficits.

All animals were kept in standard animal cages (12 h/12 h light/dark cycle, at 22 °C), with ad libitum access to food and water. The animals were maintained in compliance with institutional protocols (Comité d’éthique en expérimentation animale du Nord Pas-de-Calais, no. 0508003). All the animal experiments were performed in compliance with and following the approval of the local Animal Ethical Committee (agreement #12787–2, 015,101,320,441,671 v9 from CEEA75, Lille, France), standards for the care and use of laboratory animals, and the French and European Community rules.

Animals were maintained in compliance with institutional protocols (Comité d’éthique en expérimentation animale du Nord Pas-de-Calais, no. 0508003). All the animal experiments were performed in compliance with and following the approval of the local Animal Ethical Committee (agreement #12787–2, 015, 101, 320, 441, 671 v9nfrom CEEA75, Lille, France), standards for the care and use of laboratory animals, and the French and European Community rules.

Adult mice were sacrificed, and the brains were removed from the cranial cavity, embedded in OCT, and snap-frozen in isopentane pre-cooled with dry ice and liquid nitrogen. The left hemispheres were sectioned on the cryostat at 10 μm thickness. Sections were placed on the spatially barcoded arrays with one section per well.

Sections were fixed in 3.6–3.8% formaldehyde (Sigma) in PBS, washed in PBS, then treated for 1 min with isopropanol and air-dried. To stain the tissue, sections were incubated in Mayer’s Hematoxylin (Dako) for 7 min, then Bluing buffer for 2 min and Eosin (Sigma) for 20 s. After drying, the slides were mounted with 85% glycerol and images of sections were taken using Metafer Slide Scanning Platform (Metasystems). Raw images were stitched together using VSlide software (Metasystems).

To pre-permeabilize the tissue, sections were incubated for 20 min at 37 °C with 0.5 U/ul collagenase (Thermofisher) in HBSS buffer mixed with 0.2 ug/ul BSA (NEB). Following washing in 0.1x SSC buffer (Sigma), sections were permeabilized with 0.1% pepsin/HCl (Sigma) at 37 °C for 10 and 6 min, respectively. Then, the sections were washed with 0.1x SSC buffer.

After permeabilization, reverse transcription mix containing Superscript III reverse transcriptase (Thermofisher) was added to each section and incubated overnight at 42 °C as described previously (Stahl et al., 2016). Next, to remove tissue from the slide, sections were incubated for 1 h at 56 °C with Proteinase K in PKD buffer (both from Qiagen). Surface probes with bound mRNA/cDNA were then cleaved from the slide by USER enzyme (NEB) (Stahl et al., 2016). Released probes were collected from each well and transferred to separate tubes. Next, 2nd strand synthesis, cDNA purification, in vitro transcription, aRNA purification, adapter ligation, post-ligation purification, a second 2nd strand synthesis, and purification were carried out using an automated MBS 8000 system, as described previously (Jemt et al., 2016). cDNA was amplified by PCR using Illumina Indexing primers (Stahl et al., 2016) and purified using carboxylic acid beads on an automated MBS robot system (Lundin et al., 2010). An Agilent Bioanalyzer High Sensitivity DNA Kit (Agilent) was used to analyze the size distribution of the final libraries. The concentration of the libraries was measured with Qubit dsDNA HS (Thermofisher). The libraries were sequenced on the Illumina Nextseq platform using paired-end sequencing. Thirty bases were sequenced on read one to determine the spatial barcode and UMI, and 55 bases were sequenced on read two to cover the genetic region. Probes were collected from each well and transferred to separate tubes. Next, 2nd strand synthesis, cDNA purification, in vitro transcription, aRNA purification, adapter ligation, post-ligation purification, a second 2nd strand synthesis and purification were carried out using an automated MBS 8000 system as described previously (Jemt et al., 2016). cDNA was amplified by PCR using Illumina Indexing primers (Stahl et al., 2016) and purified using carboxylic acid beads on an automated MBS robot system (Lundin et al., 2010). An Agilent Bioanalyzer High Sensitivity DNA Kit (Agilent) was used to analyze the size distribution of the final libraries. The concentration of the libraries was measured with Qubit dsDNA HS (Thermofisher).

After the probes were released from the slide surface, the features with remaining non-cleaved DNA probes were detected by incubation with hybridization mixture containing Cyanine-3 labeled oligonucleotides, as described previously (Stahl et al., 2016). Fluorescent images were acquired using the same microscope as for the bright field images.

The libraries were sequenced on the Illumina Nextseq platform using paired-end sequencing. Thirty bases were sequenced on read one to determine the spatial barcode and UMI, and 55 bases were sequenced on read two to cover the genetic region.

Bright field-stained images (H&E) and fluorescent images (Cy3) were aligned using the ST Spot Detector (Wong et al., 2018) software. The pixel and respective array coordinates of the detected spot centroids (inside tissue) were exported to a file that was used for down-stream analysis and visualization.

Sequenced raw data was processed using the open-source ST Pipeline v1.45 (Navarro et al., 2017) with the genome reference Ensembl GRCm38 v86 and reference Mouse GenCode vM11 (Comprehensive gene annotation). The ST Pipeline was executed with the following settings: Enable homopolymers filter (A, G, T, C, N) with a length of 10, enable two-pass mode for the alignment step, remove non-coding RNA (using the v86 non coding RNA database from Ensembl), discard reads whose UMI has more than 6 low quality bases, and discard trimmed reads shorter than 20.

The matrices of counts (spots by genes) generated by the ST Pipeline were filtered to replace Ensembl IDs by gene names and to keep only protein-coding, long-non-coding-intergenic, and antisense RNAs. The matrices of counts underwent another filtering step where only spots inside the tissue were kept using the file generated in the previous step (image alignment).

The Tau22 dataset is composed of 2 sections per animal (mice) and 3 animals per genotype (Tau22 and littermate WT). Likewise, the TauKO dataset is composed of 2 sections per animal (mice) and 3 animals per genotype (TauKO and littermate WT). The two WT strains are not the same since they were continuously bred to their respective genetically modified tau strains. We did not do a direct Tau22 to TauKO analysis.

The filtered and aligned matrices of counts were analyzed jointly with the Scanpy (Wolf et al., 2018) package v1.8.2. Briefly, the spots with a total count (UMIs) less than 2,000 or bigger than 40,000 were discarded. The spots with less than 1,000 RNAs detected (count > 0) or with a percentage of mitochondrial RNAs above 15 were also discarded, mitochondrial RNAs were consequently removed from the filtered data and the remaining RNAs that were detected in less than 10 spots were also discarded. This resulted in 7,403 spots and 12,748 RNAs after filtering. The filtered data was normalized using the normalize_total function in Scanpy, the normalized data was log-transformed using a pseudo count of 1. The normalized and log-transformed data was adjusted to remove the unwanted batch effect of the animal (mice) using the regress_out function in Scanpy. Using the batch-corrected data, we selected the top 2000 variable RNAs using the “Seurat” flavor implemented in Scanpy. We used the 7,403 spots and the 2000 RNAs to perform unsupervised clustering which consisted in: (1) scale data to unit variance, (2) dimensionality reduction with PCA (Pearson, 1901), (3) compute the k-nearest neighbors (k = 15), (4) build a 2D manifold with UMAP (McInnes et al., 2018), and (5) compute clusters with the leiden algorithm (Traag et al., 2019) using a resolution of 0.75. The Allen Brain Atlas (Lein et al., 2007) and the tissue sections (H&E) were used to annotate the clusters. The clustering was validated by looking for technical effects by tissue morphology, count, animal, and genotype.

The filtered and aligned TauKO matrices of counts were analyzed in the same way as the Tau22 dataset with the exception that 6,437 spots and 11,073 RNAs were obtained after filtering, the unsupervised clustering produced 16 well-defined clusters, and no batch effect could be observed. Similarly, to the Tau22 dataset, the Allen Brain Atlas (Lein et al., 2007) and tissue section images (H&E) were used to annotate the clusters.

We used sepal (Andersson and Lundeberg, 2021) to infer RNAs with spatially distinct patterns. Sepal can only be run on individual sections, and it ranks the RNAs by a score. We averaged the scores for all the Tau22 and TauKO sections separately and then selected the top 25 RNAs, respectively. We validated the results by plotting the normalized expression of the RNAs onto the tissue sections.

We used the diffpy (Fischer and Hölzlwimmer, 2021) package to leverage on the power of its GLM zero inflated negative binomial models to infer RNAs that were differentially expressed for each region of interest. Each strain was compared to its respective littermate WT controls, thus Tau22 vs. WT and TauKO vs. WT were compared for the regions of interest (hippocampus and ventricle). An RNA was considered differentially expressed if the adj p-value ≤ 0.05 and a fold change of |0.5| for hippocampus and ventricle regions.

The sets of differentially expressed RNAs were queried for GO biological processes (Ashburner et al., 2000) enriched pathways using the gprofiler (Kolberg et al., 2023) package. Only the terms labelled as “significant” (p-value below 0.05) were reported with the corresponding significantly changed RNAs.

Immunofluorescence (IF) on mouse brain sections was performed as described previously (Zheng et al., 2020). Briefly, sagittal (5 μM) brain slices were deparaffinized and unmasked using citrate buffer (12 mM citric acid, 38 mM sodium phosphate dibasic, pH 6) for 8 min in a pressure tank. The slices were submerged for 1 h in 1% goat serum (Vector Laboratories #S-1000), and the primary antibodies were incubated overnight at 4 °C in the presence of PBS-0.2% Triton. Primary antibodies were revealed via secondary antibodies coupled to Alexa 488 or 568 (Life Technologies; 1/1000). The sections were counterstained with 4′,6-Diamidino-2-phenylindole (DAPI) and mounted with fluorescence mounting medium (Agilent Dako #S3023). The following primary antibodies were used: TTR (ThermoFisher Scientific PA5-88094; 1/100), AT8 (PSer202/Thr205tau; ThermoFisher Scientific MN1020; 1/400), Aβ (MOAB2; reactive to Aβ aa 1–4, Novus Biologicals NBP2-13075; 1/500) (4G8; reactive to Aβ aa 17–24; BioLegend SIG-39200; 1/5000). Fluorescence was quantified using the FIDJI macro application of ImageJ (confocal microscopy platform, PBSL, UAR2014/US41, Lille). Quantification corresponds to the z stack of serial confocal sections covering the entire thickness of the brain section. For CP epithelial cells, in each section the cell set was manually delimited and fluorescence was quantified in all cells. For Aβ deposits, in each section, labeled clusters were manually delimited and fluorescence was quantified. The quantification shows the mean of fluorescence values per individual.

Fluorescence from mouse brain sections was acquired using an LSM 710 confocal laser-scanning microscope (clsm) (Carl Zeiss). The confocal microscope was equipped with a 488-nm Argon laser, 561-nm diode-pumped solid-state laser, and a 405-nm ultraviolet laser. The images were acquired using an oil 63X Plan-APOCHROMAT objective (1.4 NA). All recordings were performed using the appropriate sampling frequency (16 bits, 1,024–1,024 images, and a line average of 4).

The Shapiro–Wilk test of normality (GraphPad Prism 7) was used to test if the data were normally distributed. Two-tailed, unpaired t-test (GraphPad Prism 7) was used for statistical analysis of immunofluorescence in murine brains. Each biological replicate corresponds to one mouse. The number of biological replicates is indicated in the legends. The experimenters were not blinded. Data are presented as mean ± SEM, *p < 0.05; **p < 0.01.

A graphic overview of our experimental approach is shown in Figure 1. Briefly, snap frozen tissue samples from three animals of each genotype (Tau22, TauKO, and their respective littermate WT controls), were cryo-sectioned and processed as described in the Methods. Only female mice were used in this study. The Tau22 dataset is composed of 2 sections per animal (mice) and 3 animals per genotype (Tau22 and WT), the total number of spots under the tissue is 7,483 with 22,174 RNAs. Box plots showing total read counts and number of detected RNAs per spot are shown in Supplementary Figure S1. The average number of reads (UMIs) per spot is 13,784 with an average of 4,797 detected RNAs per spot (Supplementary Figures S1A,B). Similarly, we obtained 6,857 spots and 21,974 unique RNAs for the TauKO dataset with an average number of UMI reads per spot of 6,714 and 3,116 detected RNAs per spot (Supplementary Figures S1C,D).

Clustering analysis was performed for each dataset (see Methods). Characterization of Tau22 mice data is shown in Supplementary Figure S2 and TauKO data is shown in Supplementary Figure S3. UMAP manifolds derived from the cluster analysis were colored by genotype, animal, and total expression to test for the presence of batch effects and to assess the robustness of the clustering (Supplementary Figures S2A, S3A). In all cases, the RNA expression clusters corresponded well to the anatomical layers in the brain hemisphere (Supplementary Figures S2B,C, S3B,C). The H&E-stained tissue sections are shown in panel Supplementary Figure S2B for Tau22 and Supplementary Figure S3B for TauKO, while the clusters for each mouse overlayed on the sections are shown in Supplementary Figure S2C for Tau22 and Supplementary Figure S3C for TauKO. The annotation of the clusters was performed using the H&E images and the Allen Brain Atlas (Lein et al., 2007). We selected, as mentioned previously, the clusters corresponding to the hippocampus and ventricle regions for further analysis. We validated the clustering results by plotting the normalized expression of the RNAs onto the tissue sections (Tau22, Supplementary Figure S2C; TauKO, Supplementary Figure S3C).

We elected to focus our attention on the hippocampus and ventricles because tau mice have hippocampus-dependent defects (Schindowski et al., 2006) and the CP dysfunction was recently reported as a newly defined subgroup of AD (Tijms et al., 2024). The clustering analysis was followed by a differential expression (DE) analysis to identify DE RNAs between Tau22-WT (Figures 2A–C) and TauKO-WT (Figures 2D–F) in the regions of interest (hippocampus and ventricles). As noted in the methods, each strain has their own WT mouse littermates. After applying a significance cut off adj. p-value of ≤ 0.05 and a fold change of |0.5| on the Tau22 datasets, we obtained 5 DE in the hippocampus (Thy1, Sez6, Ttr, Gm42418, Lars2) and 4 RNAs that were DE in the ventricles (Thy1, Sez6, Sgk1, Supt7l). Two RNAs were shared between the two regions, Thy1 and Sez6. Thy1 is a cell surface glycoprotein that functions in cell-to-cell communication (Hu et al., 2022) and Sez6 encodes a protein altered in Alzheimer’s disease patient’s cerebral spinal fluid and important in neuronal signaling (Munro et al., 2016). For TauKO, there were 10 RNAs that were DE in the hippocampus and 11 RNAs that were DE in the ventricles, and 1 RNA was found in both regions, Mapt (Figures 2C,F). The following genes were significantly changed in the hippocampus: Mapt, Meg3, Serinc1, Nme7, Rtn4, Malat1, Mt3, Crym, Lars2, Gm42418. In the ventricles of TauKO mice, Mapt, Nrgn, Mpc1, Gria2, Ppp3cb, Pla2g16, St8sia3, Gm10076, Dbi, Cadm2, and Ppp3ca were found to be significantly changes. A full list of DE genes per region is listed in Supplementary Table 1.

RNA enrichment analysis using Gene Ontology biological process revealed a single term changed in Tau22 mice, in the hippocampus the term negative regulation of neuron projection development was identified, and the DE RNAs found in the term were Sez6 and Thy1. In contrast, several terms associated with axon extension and projection, and gliogenesis were identified in the TauKO hippocampus (Supplementary Figure S4A). Across the hippocampus terms, three significantly changed RNAs were identified: tau (Mapt), metallothionein 3 (Mt3), and reticulon-4 (Rtn4) (Supplementary Figure S4B). Metallothionein 3 binds heavy metals like zinc and copper, has antioxidant properties, and is downregulated in AD brains (Vasak and Meloni, 2017). Reticulon-4 is a potent neurite outgrowth inhibitor and is thought to promote neuroinflammation and neurodegeneration in AD (Kulczynska-Przybik et al., 2021). The term learning or memory and several terms related to synaptic signaling terms were enriched in the TauKO ventricles dataset (Supplementary Figure S4C). There were five RNAs from TauKO ventricle in the terms, Gria2, Ppp3ca, Ppp3cb, Nrgn, and Mapt and most of the terms involved synaptic signaling.

In a separate analysis which included all regions (Supplementary Figure S5), we aimed to detect RNAs that had region-wise distinct spatial patterns, in other words RNAs that were spatially and differentially expressed with regards to the other regions (S-DE). To begin, we identified the top DE RNAs from each genotype comparison (Supplementary Figures S5A,C). Next, we identified the top DE RNAs that also displayed a spatial pattern. The top DE and S-DE RNAs are shown in the heatmaps per genotype and region (Supplementary Figures S5A,B, Tau22 versus its WT, Supplementary Figures S5C,D, TauKO versus its WT). Notably, Ttr was both a top DE RNA and top S-DE RNA in the Tau22 dataset. It was not in the TauKO significantly changed RNA list. Additionally, we found minimal overlap, two RNAs Lars2 and Gm42418, among the RNAs in the hippocampus datasets in the Tau22 and TauKO comparisons by region and no shared genes across the ventricle’s datasets.

The ST results revealed a number of RNAs whose transcript expression were dysregulated in the presence of tau pathology and/or tau (Mapt) deletion in the hippocampus and the lateral ventricles (Figure 3). Among the RNA lists, we noted the RNA Ttr. We elected to focus on Ttr because of its known neuroprotective properties, it is important for memory, and plays a role in Aβ sequestration and clearance (Brouillette and Quirion, 2008; Buxbaum, 2023; Buxbaum and Johansson, 2017; Buxbaum et al., 2008; Cascella et al., 2013; Ciccone et al., 2020; Costa et al., 2008; Gomes et al., 2016; Gonzalez-Marrero et al., 2015; Iqbal, 2018; Li et al., 2013; Liz et al., 2020; Nilsson et al., 2018; Santos et al., 2010; Schwarzman et al., 1994; Ueda, 2022; West et al., 2021). Further, Ttr was a top RNA in both the S-DE and DE analysis in the Tau22 data set (Supplementary Figures S5A,B). Figures 3A–C displays the expression pattern and transcript levels of Ttr from Tau22 mice from the hippocampus and ventricles, whereas Figures 3D–F show similar results from the TauKO mice. The plots show the mean and median values to facilitate easier comparison. The RNA plots of Ttr expression are overlaid on the tissue slices in Figure 3G for Tau22 mice and Figure 3H for TauKO mice. Ttr is an RNA coding for an extracellular chaperone deregulated in AD but there is little known about a regulatory relationship between tau and TTR protein.

TTR protein localization and levels were analyzed by IF in sagittal sections from 12 months-old THY-Tau22 and WT littermate mouse brains using TTR antibody. In both genotypes, TTR protein was only strongly detected in CP epithelial cells, where it is secreted (Stauder et al., 1986), with an increasing intensity towards the basal surface (Figures 4A,B). IF analysis revealed a significant increase of TTR protein level in CP epithelial cells from Tau22 mice versus WT littermates (mean intensity WT: 106.6%; Tau22: 130.6%) (Figures 4B,C). We did not observe accumulation of phosphorylated tau in CP epithelial cells (Figure 4D), indicating the absence of tau pathology there.

We also evaluated the phosphorylated form of tau (ptau) and TTR expression in the hippocampus of WT and Tau22 mice, Supplementary Figures S6A–E, respectively. Although increased levels of TTR protein have been previously reported in hippocampal cells from AD-like transgenic mouse models (Li et al., 2011; Stein and Johnson, 2002), no difference in the level of TTR was observed in CA1 hippocampal cells between 12 months-old WT and Tau22 mouse brains (Supplementary Figures S6D,E). Additionally, no TTR expression was detected in neurons displaying hyperphosphorylated tau (Supplementary Figure S6C).

To better understand whether the elevation of TTR observed in Tau22 mice associates with a gain or a loss of tau function, we further explored the influence of tau deletion on TTR expression by IF in sagittal sections from 12 months old TauKO and WT littermate mouse brains using TTR antibody. Again, in WT mice, high TTR expression was restricted to CP epithelial cells from ventricles (Figure 4E). Quantification of TTR IF revealed a statistically significant decrease of TTR in TauKO compared to WT CP epithelial cells (mean intensity WT: 106.4%; TauKO: 64.44%) (Figure 4F). We also evaluated TTR levels in the hippocampus of TauKO mice, and similar levels of TTR were observed in TauKO and WT littermates (Supplementary Figures S6F,G). Our results indicate that tau may positively modulate the level of TTR protein in the CP epithelial cells.

TTR synthesized in CP epithelial cells is secreted into the CSF that bathes the whole brain. In transgenic Tg2576 mice, an AD-like mouse model that expresses mutant APP(sw), Aβ peptide is overexpressed and TTR expression increases in hippocampal neurons likely as a compensatory mechanism to prevent Aβ aggregation in the brain (Li and Buxbaum, 2011; Stein and Johnson, 2002). Since TTR may prevent Aβ peptide aggregation, we hypothesized that tau deletion in aged mice might induce the formation of Aβ amyloid deposits. To test this proposition, we compared the effect of tau deletion-induced TTR downregulation on the presence of Aβ deposits in sagittal sections from 12 and 24 months-old TauKO and WT littermate mouse brains using the Aβ antibody MOAB2. Although only few Aβ deposits were detected both in 12 months-old TauKO and WT littermate mouse brains (Supplementary Figure S7), a strong increase of the area and IF intensity of Aβ aggregates was observed in the hippocampus and/or near the lateral ventricles of 24 months-old TauKO compared to WT mouse brains (mean intensity WT: 100%; TauKO: 596.3%) (Figures 5A,B). Similar results were obtained using a second anti-Aβ antibody (4G8, mean intensity WT: 100%; TauKO: 387.9%) (Figures 5C,D). It is noticed that the aspect of the plaques detected by MOAB2 and 4G8 is different and potentially linked to the distinct epitopes recognized by the two antibodies. Nevertheless, it should be stressed that the use of 4G8 to detect murine Aβ is controversial. No Aβ accumulation was observed in CP epithelial cells of 24 months-old TauKO compared to WT mouse brains (data not shown).

For comparison, we also evaluated TTR protein level, and it was downregulated in CP epithelial cells from 24 months-old TauKO mouse brains (mean intensity WT: 102.7%; TauKO: 69.42%) (Figures 5E,F) in a similar range as previously observed in 12 months-old TauKO mice (Figures 4E,F), showing that the regulatory role of tau on TTR protein expression is conserved during aging. Altogether these results show that tau deletion potentiates Aβ deposit formation of endogenous Aβ peptide in aged mice, and that this correlates with a decrease in TTR expression.

To explore the interconnectedness between tau (Mapt), APP (presumably Aβ), and TTR, we submitted those genes plus our DE gene lists, by genotype, to GeneMANIA (Warde-Farley et al., 2010). The output graphics are shown in Figure 6. There were no known direct linkages between tau and TTR, neither physical nor predicted. No transcription factors were revealed by this analysis either. However, there was an indirect link between tau, APP (Aβ), and TTR. Stringdb analysis produced similar results (data not shown). It appears that the shortest connection between Ttr and tau is through Aβ.