Fixing a key protein complex restores nerve cell communication and internal cell transport problems in an Alzheimer's mouse model

The thiophene thioureas, R55 and R33, were obtained as hydrochloride salts from CaymanChem (Ann Arbor, MI) and MedKoo (Morrisville, NC), respectively. To prevent degradation due to hydrolysis and oxidation, the compounds were dissolved in high-purity anhydrous DMSO at a concentration of 300 mM, aliquoted, and stored at − 70 °C. Immediately before intracranial microinjection, one aliquot was thawed and diluted to 10 mM under sterile conditions using freshly prepared 20% sulfobutylether beta-cyclodextrin (Thermo Fisher) in saline. A vehicle control solution containing the same proportions of DMSO and beta-cyclodextrin (final concentration of 3% DMSO) was prepared alongside the drug solutions to ensure consistency across treatments.

Although pharmacokinetic data for R55 and R33 is not available, the estimated final cortical concentration following a 500 nL injection was approximately 25 µM, assuming homogeneous distribution. This concentration exceeds the estimated dissociation constant (Kd) for VPS35 stabilization (~ 5 µM) and the IC₅₀ for Aβ reduction (~ 12 µM) in primary neurons, as previously described (Mecozzi et al., 2014).

Experiments were conducted using 16-week-old female 5xFAD/C57BL6 mice (B6), specifically Cg-Tg(APPSwFlLon, PSEN1*M146L*L286V)6799Vas/Mmjax, obtained from The Jackson Laboratory (JAX) and locally bred with C57BL/6 J mice from the same source. Compounds were administered via stereotaxic microinjection into the lateral parietal associative cortex using the following coordinates: caudal − 2.06 mm, lateral −1.5 mm, and ventral − 0.8 mm. Infusions were performed at a rate of 100 nL/min, with a total volume of 500 nL.

Twenty-four hours post-injection, mice were euthanized, and brains were rapidly dissected. For the transcriptomics study, six mice per treatment group (n = 6 per group) were used. A cortical tissue block measuring approximately 1 mm3 was isolated from each hemisphere, with the injection site serving as the focal point in the ipsilateral hemisphere. The contralateral hemisphere was used as an internal control. Care was taken during microdissection to ensure that no parts of the corpus callosum were included, thereby preventing contamination from interhemispheric structures. To preserve RNA integrity, tissue blocks were submerged in RNAlater (Thermo Fisher) and stored at − 70 °C until processing. For histological analysis, an independent cohort of mice (n = 3 per treatment) underwent intracardiac perfusion with PBS followed by 4% PFA in PBS before whole-brain dissection. Tissue assignments for drug or vehicle treatment were randomized, and researchers were blinded to treatment identity until the completion of data analysis. All procedures were approved by the Animal Ethics Committee of Linköping University and conducted in accordance with ethical guidelines for animal research.

Total RNA was extracted using the RNeasy + Mini Kit (Qiagen) and processed with the QIAcube automated system (Qiagen). RNA integrity was assessed using Bioanalyzer RNA 6000 Nano Chips (Agilent), and all samples had an RNA integrity number (RIN) exceeding 8.5. RNA quantification was performed using Quant-iT (Invitrogen), and indexed sequencing libraries were constructed with the Illumina Stranded mRNA Prep Kit. Deep RNA sequencing was conducted using a NextSeq 2000 (Illumina, P50 flow cell) following the standard Illumina pipeline.

Paired-end reads were aligned to the mouse GRCm39 genome (June 2020) using HISAT2, with sequencing depths ranging between 20 and 40 million mapped reads per sample. Differential expression analysis was performed using DESeq2 within the R (R Core Team, 2023, www.r-project.org↗) and Bioconductor [14] framework, comparing gene expression in the injected hemisphere to that in the contralateral hemisphere. Additional comparisons were made between treatment groups, including R55 vs. vehicle, R33 vs. vehicle, and R55 vs. R33. Pathway and functional enrichment analyses were conducted using multiple databases, including Gene Ontology (GO), InterPro, KEGG, Monarch, Pfam, SMART, STRINGdb, UniProtKB, and WikiPathways.

To contextualize transcriptional alterations observed in 5xFAD mice, external reference datasets were used, including publicly available RNA-seq data from MODEL-AD datasets (GSE168137, [11]), which document transcriptomic changes in 5xFAD mice across disease progression, from Bliim, Leshchyns'ka, Keable, Chen, Curry-Hyde, Gray, Sytnyk and Janitz [5] (GSE110908), which describes gene expression changes following chemically induced long-term potentiation (cLTP) and [19] describing the disease-associated microglia (DAM) phenotypes in 5xFAD.

Following transcardial perfusion, brains were cryoprotected in sucrose in 4% PFA until they lost buoyancy, then snap frozen. Sections were obtained using a cryostat and immunostained with the following primary antibodies: anti-VPS35 (clone 7E4, NBP2-78823, Novus Biologicals, CO), anti-RAB7 (clone E9O7, Cell Signaling, MA), anti-VPS13b (PA5-34406, Invitrogen), anti-SORL1 (NB100-785, Novus) and anti-IBA1 (ab5076, Abcam).

Fluorescence microscopy was performed using a Zeiss LSM700 confocal microscope, and image deconvolution was applied for signal enhancement. Cortical layer V neurons adjacent to the injection site, but sufficiently distant from the needle track to maintain normal parenchymal morphology, were segmented using QuPath [2]. Fluorescence intensity was quantified in approximately 700 segmented neurons per protein. The quantification was done in the cortex layer V, aiming to analyse as close to the injection track as possible but without artefacts from the procedure. For the Sorl1 study, the morphometry, location and distance to nuclear envelope of about 20,000 Sorl1 + particles was analysed using ImageJ DiAna plugin [12].

All data were analyzed using R, with results presented as mean ± standard error (SE). For multiple comparisons, FDR corrected p-values are presented. The statistical significance threshold was set at α = 0.05. Two-tailed statistical tests were applied for comparisons.

Network analysis of the R55-regulated genes identified 247 interactions with an average node degree of 2.38 (Fig. 2B), significantly exceeding the expected 101 interactions from a random dataset (p = 1.10e-16), demonstrating that R55 selectively modulates interconnected pathways. Hierarchical clustering revealed distinct gene expression signatures across hemispheres and treatment groups (Fig. 2C). Interestingly, gene expression changes were observed in both ipsilateral and contralateral hemispheres, suggesting widespread parenchymal distribution of the compounds.

Enrichment analysis using multiple databases, revealed that the majority of the enriched GO terms were neuronal and synaptic-related, as well as some related to calcium handling and axonal organization (Fig. 2D). The AD-related pathway WP2075 was significantly altered, along with synaptic organization pathways and other CNS-related phenotypes. Structural protein domain enrichment analysis highlighted common domains in the differential expression genes, specifically the ones related to PSD-95, Egf-like, and Vps13 families (Fig. 2E). Notably, a volcano plot of WP2075 genes (Fig. 2F) confirmed that both R55 and R33 modulated gene expression in the same direction, with R55 exerting stronger effects.

Finally, to obtain a statistical quantification, we build a series of generalized linear models of increasing complexity that express the changes we observed in R55/veh (log2FC.R55) as a function of the data in plot 8B separately (logFC.m4, log2FC.DAMfromHOM, log2FC.DAM2fromDAM1) or in combination. We calculated the Akaike information criterion (AIC) for each model and found that the model that included all factors had the best AIC (8C) and was statistically different from the others on the likelihood ratio test. We then estimated the statistical contribution of each parameter to the model and the combinations between them, and found that logFC.m4, log2FC.DAMfromHOM, log2FC.DAM2fromDAM1 contribute significant separately, and that log2FC.DAMfromHOM actually has a greater effect than log2FC.m4. The combination of microglial activation phenotypes (log2FC.DAMfromHOM * log2FC.DAM2fromDAM1) almost reaches statistical significance. Taking this all together, shows that the DAM from HOM and DAMII from DAM I changes explain a significant part of the changes we observe in microglial genes after R55 treatment (Figs. 9C, D).

Disruptions in synaptic transmission and plasticity are early hallmarks of AD. Endosomal trafficking, mediated by the retromer complex, plays a pivotal role in synaptic function by regulating intracellular protein transport. Our study demonstrates that stabilizing the retromer complex reverses key molecular alterations in AD, shedding light on downstream mechanisms and signaling pathways affected by retromer dysfunction. Our results supports previous reports that dysfunctional retromer linked to neurodegenerative diseases is not primarily caused by loss of expression of Vps35 but by decreased endosomal association causing loss of function that can be mitigated by retromer stabilizing compounds such as R55 and R33 [10]. Additionally, our analysis confirms that 5xFAD mice exhibit extensive gene expression changes in the retromer interactome and its cargo, resembling patterns seen in human AD cohorts — a degree of alteration not observed in other transgenic AD models.

A strong body of evidence links synaptic dysfunction to early AD progression, with Aβ oligomers interfering and inducing changes similar to those involved in developmental synaptic pruning and with plasticity-related processes such as LTP and LTD [23]. Under normal conditions, these interactions are transient, but in AD, Aβ oligomers induce persistent synaptic alterations, contributing to disease progression. For example, Aβ oligomers increase cholesterol concentration at synapses, promoting Aβ/PrP complexes that disrupt NMDAR signaling via GRIN2a/GRIN2b S-nitrosylation [3]. In line with these events, we observed compensatory increases in Grin2a, Grin2b, iNos, and VLDL/ApoE-related genes such as Lrp1, Lrp1b, and Fbn1 in 5xFAD mice (MODEL-AD datasets). Notably, retromer stabilization by R55 counteracted these transcriptional upregulations, suggesting enhanced receptor recycling and trafficking reduced the need for compensatory gene expression changes.

The retromer complex is crucial for synaptic AMPA receptor trafficking. Loss of VPS35 impairs Gria1 (AMPA) receptor insertion and disrupts LTP, while restoring VPS35 reverses these deficits [39]. In our study, R55 treatment normalized Gria1 and Grip1 expression, the latter being essential for AMPA receptor endosomal sorting and membrane targeting. This is particularly significant given recent findings implicating the semaphorin family (Plexin a4, Sema5a) and netrin receptors (Unc5c, Dcc) in AMPA receptor targeting during LTP-associated plasticity [17]. In 5xFAD mice, these genes were dysregulated, but their expression was restored by retromer stabilization, further supporting a role for retromer function in synaptic integrity.

Defective intracellular calcium homeostasis, particularly endoplasmic reticulum (ER) calcium leakage, is an early event in AD-linked synaptic dysfunction. The neuronal ryanodine receptor (Ryr2) is overactivated in AD, leading to enhanced receptor-operated calcium entry and decreased buffering capacity [18, 21]. Furthermore, RyRs interact functionally with L-type calcium channels (Cacna1c, Cacna1d, Cacna1e), particularly in CA1 pyramidal neurons, where calcium-induced calcium release contributes to synaptic plasticity [35].

In 5xFAD mice, we observed elevated Ryr2 and Cacna1c/Cacna1d/Cacna1e expression, consistent with previous findings. R55 treatment effectively reversed these alterations, suggesting restored calcium signalling homeostasis. Of Ryr2-associated regulators [38], only Jph4 showed a weak response to R55, indicating selective rather than global modulation of calcium-handling genes.

Given that the retromer complex regulates Aβ processing, we examined key AD-related cargos. R55 treatment increased γ-secretase components while reducing Sorl1 and Col25a1 expression. Col25a1 encodes a collagen-like amyloid plaque component, known for forming protease-resistant aggregates. These changes suggest that recovered retromer function may reduce excessive amyloidogenic processing caused by extended cargo residency at the Golgi by normalizing Golgi-retromer dynamics.

Unexpectedly, R55 treatment significantly reduced Vps13 family gene expression, which was otherwise elevated in 5xFAD mice. The Vps13 complex regulates endosomal recycling towards the Golgi, mitochondria, and autophagosomes by promoting organelle bilayer bridging, and has been implicated in mitochondrial dysfunction in neurodegenerative diseases, including atypical PARK23 [40]. Genetic variations in VPS13 genes are emerging risk factors for sporadic, late-onset AD, with VPS13 variants exceeding even ApoE in risk contribution [27]. Immunohistochemistry confirmed that Vps13 protein levels decreased following R55 treatment, while RAB7 expression increased, this suggests restored endosomal maturation. Previous studies have shown a stagnation of large endosomes both in AD and in 5xFAD [9, 24] and impaired recycling pathway, in 5xFAD in mice of almost the same age [24]. We used SORL1 expression as a marker of endosomal trafficking. SORL1 expression has been shown to have a reciprocal role prototypical cargos such as transferrin to in cell surface recycling [30]. Sorl1 distribution that has been reported to be altered both in APP overexpressing AD models such as 5xFAD and in Sorl1 mutations that have been linked to AD pathology [29]. The changes to Vps13 and RAB7, taken together with the increased proximity of both VPS35 and SORL1 to the plasma membrane, compensation in synaptic-related pathways, and the absence of changes in lysosomal degradation-associated genes, our data support improved sorting and membrane trafficking that may favor recycling over degradation.

In the neuroinflammatory domain, R55 treatment increased both pro-inflammatory mediators (Il1, Nfkb2) and pro-trophic glial markers (Il4r, Il13ra1), as well as downstream anti-inflammatory transcription factors (Stat6, Stat3). This coordinated shift in crosstalk between microglial polarization and astrocytic responses mirrors findings reported in other AD studies [45]. Similarly, Vps35 KO impaired microglial activation in response to Aβ [33].

IBA1 staining revealed that R55-treated 5xFAD microglia exhibited reduced ramification, indicating an altered activation state. Further gene expression comparisons to MODEL-AD and DAM datasets showed significant shifts in microglial phenotypes [8, 15]. Given that phagocytic receptors and cytokine receptors utilize retromer-mediated recycling in glia, these findings provide a mechanistic link between retromer function and neuroinflammation although further studies are needed to assess if the identified changes have long term positive or negative effects.

This study focuses on dysregulation of the transcriptome in early AD and the acute molecular effects of retromer stabilizing drugs. Although our results suggests that retromer dysfunction can contribute to many AD related alterations and that the retromer stabilizing drugs have the potential to counteract these effects, further studies are needed to confirm these findings. Comparing the changes we identify to downstream effects on the proteome would be interesting, but large-scale proteomics are still challenging and currently there are no database for changes to the proteome similar to the MODEL-AD. As the focus of this study was on acute changes, we can only speculate on the long-term effects on AD-related phenotypes such as memory, cognition, and synaptic plasticity. Further studies assessing pharmacokinetics and pharmacodynamics of this type of drugs are also needed.

While our study provides acute mechanistic insights, extrapolating these findings to chronic AD progression remains challenging. However, our results strongly support retromer dysfunction as a key contributor to synaptic pathology in AD, raising the possibility that retromer stabilization could mitigate disease progression.

One intriguing question is why VPS35 mutations (e.g., D620N) cause Parkinson's disease (PARK17) rather than AD. The D620N mutation disrupts VPS35 binding to WASH complex proteins, affecting actin nucleation, while AD-linked VPS35 mutations (e.g., L625P) impair VPS35-VPS26-VPS29 assembly [25, 34]. Notably, R55 enhances VPS35 assembly without affecting WASH binding [4]. However, R55-treated 5xFAD mice exhibited changes in WASH complex components, sorting nexins (Snx1, Snx2, Snx3), RAB7, and Vps13, suggesting a broader role for retromer stabilization in intracellular trafficking beyond cargo selection.

Our findings suggest that retromer stabilization influences not only endosomal cargo selection but also endosomal anchoring to other organelles. The observed redistribution of VPS35-positive vesicles away from the nucleus suggests that endosomes are repositioned towards the plasma membrane, likely improving membrane protein recycling, the decrease in Vps13 and increase in RAB7 further support this hypothesis.