Impaired TGFβ Signaling in Plaque-Associated Microglia

TL;DR

The study finds that microglia in Alzheimer's disease mouse models have abnormal shapes and disrupted TGFβ signaling, which may contribute to the disease's progression.

Contribution

This work reveals impaired TGFβ signaling in plaque-associated microglia in an Alzheimer's disease mouse model.

Findings

APP/PS1 mice show disturbed glial morphology compared to wild-type mice.

Altered pSMAD2 distribution suggests impaired canonical TGF-β signaling in microglia.

Dysfunctional TGF-β signaling may play a role in Alzheimer’s disease pathogenesis.

Abstract

Aging and Alzheimer’s disease (AD) are associated with profound changes in glial cell morphology and signaling. This study investigates the three-dimensional morphology of microglia and the intracellular localization of phosphorylated SMAD proteins as downstream effectors of transforming growth factor β (TGF-β) signaling in the amyloid precursor protein and presenilin-1 (APP/PS1) transgenic mouse model of Alzheimer’s disease. Using confocal microscopy and Simple Neurite Tracer software, we reconstructed and quantitatively analyzed glial cell morphology in aged wild-type and APP/PS1 mice. Immunofluorescence staining revealed altered pSMAD2 distribution in microglia, suggesting impaired canonical TGF-β signaling. Our findings indicate a disturbed glial morphology and dysfunctional TGF-β signaling cascade in the APP/PS1 model, underlining their potential role in Alzheimer’s disease pathogenesis.

Figures

Full text

1. Introduction

The progressive neuroinflammation observed in aging and neurodegenerative diseases, such as Alzheimer’s disease (AD), is increasingly linked to both functional and morphological changes in glial cells. Microglia and astrocytes are central to maintaining central nervous system (CNS) homeostasis and respond dynamically to pathological stimuli, including β-amyloid (Aβ) accumulation [1]. AD poses a growing global health challenge. Since 1990, its incidence and prevalence have risen markedly, and projections indicate a continued increase through 2040 [2]. The disease typically begins decades before clinical symptoms: biomarkers such as Aβ and neurofilament light chain may rise as early as 20 years prior to diagnosis [3,4]. Recent work has expanded our understanding of AD pathogenesis to include not only protein aggregation but also synaptic dysfunction and altered synaptic plasticity [5].

Aβ peptides—especially the aggregation-prone Aβ_42_—arise from sequential cleavage of the amyloid precursor protein (APP) by β-secretase and γ-secretase. In AD, aberrant APP processing favors Aβ overproduction, leading to self-assembly into soluble oligomers, protofibrils, and eventually extracellular plaques. These species impair synaptic plasticity, disrupt neuronal calcium homeostasis, and trigger inflammatory cascades [6]. In addition to direct amyloid toxicity, increasing evidence implicates neuroinflammation as a driver of both the onset and progression of AD [7]. Neuroinflammation involves the activation of the brain’s innate immune system in response to stimuli, such as misfolded or aggregated proteins [8]. A key consequence is disruption of the blood–brain barrier, enabling the infiltration of peripheral immune cells and, in some cases, pathogens into the CNS. This phenomenon occurs not only in AD but also during normal aging, where low-grade inflammation may create a permissive environment for pathology. In amyloid contexts, inflammation both responds to and exacerbates disease processes [9].

Microglia are the brain’s resident immune cells, responsible for continuous surveillance of the CNS microenvironment. In their homeostatic state, they support neuronal function through synaptic pruning, debris clearance, and the release of trophic factors. Upon encountering pathological stimuli, such as aggregated Aβ, microglia undergo profound morphological and transcriptional changes, shifting from a ramified, surveillant state to an activated phenotype. This activation is accompanied by altered motility, increased phagocytic capacity, and secretion of pro- and anti-inflammatory mediators. While acute microglial activation can be protective—facilitating removal of harmful aggregates—chronic activation is associated with sustained inflammation, oxidative stress, and synaptic dysfunction. In AD, microglial activation clusters spatially around Aβ plaques, where cells adopt specialized phenotypes to engage with the aggregated material. Emerging evidence suggests that microglial responses are highly heterogeneous, encompassing both beneficial and detrimental effects on disease progression depending on activation state, local environment, and signaling cues, such as transforming growth factor β (TGF-β) [10].

TGF-β, in particular TGF-β1, is a key regulator of microglial development, homeostasis, and reactivity in the central nervous system. TGF-β1 signaling is required for proper microglial maturation and for maintaining a homeostatic, non-inflammatory phenotype in the adult brain, and disruption of TGF-β signaling in microglia leads to abnormalities in microglial number, morphology, and gene expression, as well as neurodegeneration [11,12]. In AD, however, TGF-β signaling becomes dysregulated: ligand and receptor levels are altered, and downstream SMAD signaling is modified in ways that can both limit and promote neuroinflammatory responses and amyloid pathology [13]. SMAD proteins are intracellular signal transducers of the TGF-β superfamily, comprising receptor-regulated SMADs (R-SMADs), such as SMAD2/3 (TGF-β branch) and SMAD1/5/8 (BMP branch); the common mediator SMAD4 (Co-SMAD); and inhibitory SMADs (I-SMADs), such as SMAD6/7, which provide negative feedback on the pathway [14,15]. Both branches of the pathway are shown in Figure 1. In the canonical pathway, TGF-β ligands bind to Type II receptors, which recruit and phosphorylate Type I receptors to form an active complex [16]. The adaptor protein SARA then recruits SMAD2/3 to the receptor for phosphorylation [17]. Phosphorylated SMAD2/3 forms complexes with SMAD4 that accumulate in the nucleus to regulate transcription [15,18,19]. Canonical TGF-β signaling thus depends on the dynamic import and export of SMAD proteins across the nuclear envelope [14]. How canonical TGF-β/SMAD2/3 and bone morphogenic protein (BMP)/SMAD1/5/8 branches are differentially engaged or impaired in microglia during Aβ pathology remains poorly understood and is a key focus of the present study.

In the resting state, R-SMADs continuously cycle between the cytoplasm and nucleus, with the equilibrium biased toward the cytoplasm, whereas SMAD4 is more evenly distributed. The ligand-induced phosphorylation of R-SMADs promotes the formation of SMAD2/3–SMAD4 or SMAD1/5/8–SMAD4 complexes, which accumulate in the nucleus and, together with additional transcription factors, drive target gene transcription. Nuclear export requires the complex dissociation and dephosphorylation of R-SMADs, a comparatively slow step that favors the transient nuclear accumulation of pSMADs under physiological conditions. Consequently, alterations in the balance between cytoplasmic and nuclear pSMAD provide a sensitive readout of disturbed TGF-β/SMAD signaling in aging and neurodegeneration [20,21]. However, it remains unclear how this pathway—especially the phosphorylation and distribution of SMAD proteins—changes in glia during aging and AD.

To interrogate these questions in AD microglia, we used the APP/PS1 transgenic mouse, which co-expresses human mutant APP (Swedish mutation, K594N/M595L) and mutant presenilin-1 (PS1; e.g., L166P). These alleles increase β-secretase processing, elevate Aβ_42_ production, and accelerate amyloid deposition [22,23]. APP/PS1 mice develop cortical and hippocampal plaques in ~ 4–6 months, accompanied by microglial activation, astrocytosis, and a pro-inflammatory milieu, as well as synaptic and cognitive deficits—including impaired long-term potentiation and spatial learning—prior to overt neuronal loss [24].

Single-cell transcriptomics has defined a disease-associated microglia (DAM) state in neurodegeneration [25]. DAM are characterized by the downregulation of homeostatic genes (e.g., P2ry12, Tmem119) and the upregulation of genes linked to phagocytosis, lipid metabolism, and inflammatory signaling (e.g., Apoe, Trem2, and Tyrobp) [26]. Their activation proceeds in two steps: an initial TREM2-independent phase followed by a TREM2-dependent phase that amplifies microglial responses to aggregated proteins, such as Aβ and hyperphosphorylated tau. While DAM may aid aggregate clearance, they can also sustain chronic inflammation and contribute to synaptic dysfunction [27]. Because DAM-like phenotypes arise alongside plaque formation, this model is well-suited to dissect amyloid-driven and immunological components of AD [28].

By combining the morphological analysis of microglia with the intracellular mapping of phosphorylated SMAD proteins in aged and APP/PS1 brains, this study aims to clarify whether age- and disease-related inflammatory environments alter TGF-β signaling in ways that could contribute to AD progression.

2. Materials and Methods

2.1. Materials

2.1.1. Anesthetics and Chemicals

Mice were euthanized using carbon dioxide, and rigor mortis was assessed. Paraformaldehyde (PFA), Phosphate-buffered saline (PBS) citrate solution, and Triton X-100 were from Sigma, St. Louis, MO, USA. Copper (II) sulfate was from Merck, Søborg, Denmark. Mounting medium was from Thermo Fisher Scientific, Waltham, MA, USA. DAPI was from Roth, Karlsruhe, Germany. Chemicals otherwise used were of the purest grade available from regular commercial sources.

2.1.2. Antibodies

Primary and secondary antibodies used in this study are summarized in Table 1 and Table 2, respectively.

2.2. Animals

The present study was carried out in accordance with the European Communities Council Directive of 22 September 2010 (2010/63/EEC) for the care of laboratory animals. The breeding and euthanasia were approved by the local Animal Research Committee (Landesamt für Landwirtschaft, Lebensmittelsicherheit und Fischerei (LALLF)) of the state of Mecklenburg-Western Pomerania (LALLF M-V/TSD/7221.3-2-034/17; approval date: 24 July 2017).

Three young wild-type (female, 6 months), four aged wild-type (female, 24 months old) and five aged APP/PS1 mice (female, 24 months old) (stock# 034829, Jackson Laboratory, Bar Harbor, ME, USA) were sacrificed by deep carbon dioxide inhalation. The dead mice were decapitated in order to remove brain tissue. APP/PS1 mice express both the chimeric amyloid precursor protein (human APP695swe) with Swedish double mutations (K594N and M595L) and the human presenilin protein 1 carrying the exon-9-deleted variant (PS1-dE9, L166P) [23]. The APPswe/PS1delta9 mice were hemizygotes on B6xC3H and C57BL6 mouse back-grounds. Additional information is available at: https://www.jax.org/strain/004462 (accessed on 26 January 2026). Mice had free access to food and water and were housed in a 12 h light/dark cycle. The health status of the mice was checked daily.

2.3. Tissue Preparation, Immunohistochemistry, and Image Acquisition

After decapitation, brains were fixed in 4% PFA for 24 h, then washed in PBS for 10 min, and, lastly, cryoprotected in 30% sucrose for 72 h. Finally, the brains were stored at −80 °C until cryostat sectioning. Coronal brain slices (50 μm) were cut using a cryostat (CM 3050 S, Leica Microsystems, Wetzlar, Germany) and stored in PBS at 4 °C. Immunofluorescence was performed using two double-staining panels:

• (1)Aβ with Iba-1 (rabbit anti–Iba-1, FUJIFILM Wako, Osaka, Japan);
• (2)TGF-β1, pSMAD2, or pSMAD1/5/8 with Iba-1 (guinea-pig anti–Iba-1, Synaptic Systems, Göttingen, Germany).

All sections were counterstained with DAPI to visualize nuclei.

Brain slices from the frontal lobe were processed over three days. Primary and secondary antibodies were applied sequentially, with all steps conducted under reduced light exposure. To reduce lipofuscin autofluorescence, sections were finally treated with a solution of 35 mM CuSO_4_ in 50 mM ammonium acetate buffer (pH 5), both were purchased from Merck EMSURE, Darmstadt, Germany for 1 h after antibody staining and washed afterwards with PBS to remove any residues of CuSO_4_. Sections from young and old animals were treated equally, as CuSO_4_ solution can also reduce the intensities of immunofluorescent labeling [29]. Stained slices were mounted on glass slides and stored at 4 °C. Fluorescence microscopy was performed using a C1 confocal system, Nikon, Tokyo, Japan with a PCM-2000 confocal microscope scanning system, Nikon, Tokyo, Japan coupled to an ECLIPSE E400 microscope, Nikon, Tokyo, Japan. Images (Z-Stacks) were captured in 1024p and processed in EZ-C1 Software (Ver. 3.80), Nikon, Tokyo, Japan. Exposure times were optimized per image to prevent overexposure, and images were taken at 60× magnification across multiple regions of interest (ROIs) in the frontal cortex and white matter. Images were acquired in pre-defined cortical ROIs that include peri-plaque tissue. Because plaque distance was not prospectively encoded for all stacks, microglia were quantified per cell and summarized per animal without explicit plaque-distance stratification. Representative insets illustrated peri-plaque signal enrichment, while statistical analyses reflected group-level effects within these ROIs. Negative control images with secondary antibodies alone can be found in Figure S3.

2.4. Data Analysis

2.4.1. Quantification and Cell Morphology

Histological quantification was conducted using Fiji (Ver. 2.1.0). Cells and Aβ-Plaques were counted in Maximal Intensity Projections (MIP) of the Z-Stacks, and their number was then related to the volume of the data set. Microglia morphology was examined using Sholl analysis. For this purpose, each individual cell was first reconstructed in a semi-automated procedure using a Fiji plugin called Simple Neurit Trace (SNT, Ver. 3.2.4), a tool previously validated for highly branched cellular architectures, including neurons, Purkinje cells, and astrocytes [30,31,32].

For a reconstruction that was as complete as possible, a preselection of the cells was performed as per the following criteria:

• (1)A soma with a DAPI signal should be recognizable;
• (2)Processes should not be cut off in the x- and y-axis by image borders;
• (3)The cell should be as central as possible in relation to the z-axis in order to reconstruct extensions in this plane as well as possible;
• (4)It should be possible to differentiate between neighboring cells in order to be able to assign extensions as clearly as possible to a cell.

To improve process demarcation prior to SNT tracing, we applied background subtraction and adjusted brightness/contrast under fixed settings across all images. After reconstruction, finally, a Traces File gave information about every single reconstructed process, so the sum could be calculated. From the reconstructed cell, a skeletal rendering could be created from which information about the number of branches and endpoints could be obtained. Sholl analysis was also performed. The number of primary processes and the maximum number of processes crossing a certain radius could be taken from the Sholl results file, and the Schoenen ramification index (SRI) could be determined. The surface area was measured manually as a convex hull. For the measurement of the soma, Z-stacks of the native images were used as a maximum projection. The area covered by processes was measured using the SNT reconstructions.

The ramification index (RI) was calculated according to [33] as

$$[eqn]$$

at 1 µm radius for each cell. Primary processes were defined as first-order branches emerging from the soma. This definition captured proximal arbor elaboration normalized by the initial process number.

2.4.2. Quantification of Fluorescence Intensities

The fluorescence intensities were also quantified using Fiji. First, the integrated density of the ROI fluorescence and the area of the ROI were measured. The integrated density is the product of the measured area and the average gray value. The average gray value is the sum of the gray values of all the pixels in the selection divided by the number of pixels. In addition to the ROI, five further intensities were measured above the background of the image, and an average value was calculated.

A corrected total cell fluorescence (CTCF) according to [34] could then be calculated:

$$[eqn]$$

For measurements of nuclear intensities, the DAPI signal was used as an ROI. For measurements of the cytoplasm of microglia, the ROI was defined by the Iba-1 signal and the DAPI signal. Microglia processes were not part of the ROI.

2.4.3. Statistical Analyses

The statistical analysis was carried out using Excel and GraphPad PRISM 10.4.0. Outliers of the individual values of each cell were not taken into account in the further calculation, as they could be caused by possible measurement errors. Outliers were defined as values that were more than two standard deviations away from the mean value. Mean values for each animal were calculated from the remaining values collected from individual cells. Three groups were compared: 6 months WT, 24 months WT, and 24 months AD. The samples were first examined for normal distribution using the Shapiro–Wilk test. For normally distributed data with comparable variances, group differences were assessed by one-way ANOVA followed by Tukey’s multiple comparisons test. For non-normally distributed data, the Kruskal–Wallis test followed by Dunn’s multiple comparisons test was used. p values < 0.05 were considered statistically significant (* p < 0.05, ** p < 0.01, *** p < 0.001; ns, not significant). Data are presented as mean ± SEM, with each dot representing the mean value of one animal.

2.4.4. Bulk RNA-seq Data Acquisition and Processing

Publicly available bulk RNA-sequencing (RNA-seq) data of microglia from amyloid-β mouse models were re-analyzed to complement our histological findings. Raw count matrices and metadata were obtained from [35], under accession GSE205048. Gene-level expression values (pseudocounts) for all expressed genes were obtained from Table S3 of the supplementary material, which contains bulk transcriptome profiles of three microglial populations: control microglia (CM), plaque-distant microglia (PCM), and plaque-associated microglia (PAM), sampled at 8 and 12 months of age.

Microglia were identified based on canonical marker expression (P2ry12, Tmem119) and subclustered into homeostatic and disease-associated states.

For descriptive comparison across groups, we calculated gene-wise means for CM, PCM, and PAM at each time point. Where indicated, values were log_2_-transformed after adding a small offset to avoid taking the logarithm of zero. We focused on curated gene sets comprising components of the TGF-β–SMAD pathway, including ligands (Tgfb1–3), receptors (Tgfbr1, Tgfbr2), R-/co-/I-SMADs (Smad1, Smad2, Smad3, Smad4, Smad5, Smad7, and Smad8), and representative downstream target genes. No additional normalization beyond the pseudocount transformation provided by the original authors was applied, and all analyses were restricted to the microglial samples defined in the original study.

3. Results

3.1. Amyloid Burden and Validation of the Mouse Model

To verify that the 24-month-old APP/PS1 mice display the amyloid pathology, the number of amyloid plaques and their sizes were quantified by immunostaining (Figure S1). Immunostaining results confirmed extensive cortical Aβ plaque formation in the 24-month-old APP/PS1 mice. Quantification of plaque density (per mm^3^) and mean plaque diameter confirmed robust cortical amyloid pathology in APP/PS1 mice, thereby validating the model for subsequent downstream analyses.

3.2. Glial Cell Density in the Brain of Aged WT and APP Mice

Next, we investigated the glial response to amyloid deposition. Immunostaining for the microglial marker Iba-1 demonstrated a marked increase in microglial density in 24-month-old APP/PS1 mice compared to both 6-month and 24-month-old WT groups, whereas no significant difference was observed between young and aged WT mice (Figure S2a,b). This indicates that microgliosis in the APP/PS1 cortex is primarily driven by amyloid pathology rather than aging alone.

To further assess microglial alterations, we performed 3D reconstructions and Sholl analyses of the Iba-1^+^ cells. Microglia morphology has been investigated extensively, but most prior work relied on two-dimensional projections or cell-culture models. Here, we adopted a three-dimensional analysis pipeline to obtain accurate ex vivo reconstructions of these glial cell types and to compare age-related remodeling with changes observed in advanced Alzheimer’s-like pathology. We combined semi-automated segmentation with continuous manual quality control to maximize accuracy and reproducibility.

A specific methodological challenge in the Alzheimer group was the high local density of astrocytes and microglia, frequently arranged concentrically around β-amyloid plaques. In such fields, an unambiguous assignment of individual processes was not always possible a priori. To mitigate this, trajectories were iteratively verified and, when needed, corrected during the semi-automated reconstruction, yielding largely faithful process attribution even in plaque-rich regions.

Morphological evaluation revealed highly ramified structures in the 6-month WT mice, moderate simplification in the aged WT mice, and a hypertrophic, amoeboid phenotype with retracted processes in the APP/PS1 mice (Figure 2a). Sholl profiles confirmed an age-dependent reduction in arbor complexity, with APP/PS1 microglia displaying the smallest maximum radius (42 µm vs. 62 µm in WT, Figure 2b). Interestingly, microglia from APP/PS1 mice exhibited an increased number of primary processes at the soma, resulting in an early peak of intersections at 10–14 µm.

Microglia from 6-month-old WT mice exhibited the highest number of intersections, indicating greater arbor complexity. Both aged WT and APP/PS1 microglia showed markedly reduced branching, particularly in distal regions, reflecting age- and disease-associated morphological simplification.

Furthermore, the Sholl analysis indicates that microglia in the AD group possessed a higher number of primary processes, as evidenced by an increased number of intersections at a radius of 1 μm. These primary processes subsequently branched out, resulting in a peak number of intersections at a radius of 10–14 μm in all groups. Regarding the number of primary processes, the 6-month WT group showed the greatest increase in intersections, whereas the AD group exhibited only a modest rise (Figure 2).

Sholl analysis at small radii captured proximal process crowding, whereas total endpoints and branch points were dominated by distal arborization. Accordingly, the increased intersections near the soma (small radii) could co-occur with unchanged or reduced distal endpoints/branch points, indicating proximal hypertrophy with distal simplification, a hallmark of reactive microglia.

Quantitative analysis of morphological parameters further supported these observations (Figure 3). The number of branch points, endpoints, and the ramification index were highest in the 6-month WT microglia and significantly reduced in both aged WT and APP/PS1 mice, with no additional decrease in the AD group compared to age-matched controls (Figure 3c).

The number of branch points per cell was significantly higher in 6-month-old WT mice compared to both 24-month-old WT and APP/PS1 mice (Figure 3a; p < 0.05 and p < 0.01, respectively), indicating a decline in process arborization with aging and disease. Notably, no significant difference was observed between aged WT and APP/PS1 groups (ns).

Similarly, the number of endpoints (Figure 3b), representing the terminal extensions of microglial processes, was significantly reduced in both 24-month-old groups compared to the young WT (p < 0.01), with no further reduction in APP/PS1 mice.

The ramification index, a size-normalized measure of process complexity, was highest in 6-month WT and significantly lower in 24-month APP/PS1 (Figure 3c; p < 0.05), with no significant difference between 6- and 24-month WT, nor between 24-month WT and 24-month APP/PS1. Accordingly, the RI indicated reduced complexity relative to the young baseline in AD, but did not show a standalone aging effect in our cohort.

Collectively, these findings highlight an age-dependent reduction in microglial complexity, with APP/PS1 mice showing similar levels of morphological simplification as their age-matched wild-type counterparts.

The morphological profile observed in aged and APP/PS1 cortices—proximal hypertrophy with distal simplification—provides a structural substrate for altered signaling. In subsequent analyses, we show that, in PAM, this architecture coincides with reduced nuclear accumulation of pSMAD2 despite the elevated TGF-β1/TGFBR2 and robust cytoplasmic pSMAD2, consistent with a decoupling between strong upstream input and incomplete canonical nuclear output.

3.3. TGF-β Signaling

We next examined TGF-β pathway activity in microglia. Immunostaining confirmed TGF-β1 expression in glial populations (Figure 4). Quantitative analysis of TGF-β1 immunoreactivity in Iba-1^+^ microglia revealed a disease-related increase in signal intensity. Microglia from 24-month WT mice showed a modest elevation of TGF-β1 corrected total cell fluorescence (CTCF) compared with 6-month WT animals, but this difference did not reach statistical significance (ns). In contrast, microglia from 24-month APP/PS1 mice displayed a robust increase in TGF-β1 CTCF, with values clearly exceeding both 6-month WT and age-matched 24-month WT levels. Despite some inter-animal variability, all APP/PS1 mice showed a consistently higher TGF-β1 signal in microglia, indicating a pronounced upregulation of microglial TGF-β1 expression in the aged, AD-like brain. Group-wise comparisons (one-way ANOVA with post hoc testing) are reported in Figure 4d.

Furthermore, we assessed receptor-regulated SMAD phosphorylation as a readout of pathway engagement. In cortical microglia, pSMAD2 immunoreactivity was predominantly cytoplasmic with comparatively weaker nuclear accumulation in aged groups, indicative of impaired nuclear translocation (Figure 5a–c). The quantification of compartmental signals (DAPI-segmented nuclei and cytoplasm defined as Iba-1 mask minus the nucleus) showed a shift in the nuclear/cytoplasmic (N/C) pSMAD2 ratio toward cytoplasmic predominance in both 24-month WT and 24-month APP/PS1 mice relative to 6-month WT controls (Figure 5d–f).

To test whether this pattern could be generalized across the canonical branch, we analyzed pSMAD1/5/8 distributions (Figure 6). The quantification of pSMAD1/5/8 immunoreactivity in cortical microglia revealed a marked age-dependent increase in both cytoplasmic and nuclear signal intensity. Cytoplasmic pSMAD1/5/8 CTCF was significantly higher in the 24-month WT and APP/PS1 mice compared with the 6-month WT mice, with no additional increase in the AD group beyond age-matched controls (Figure 6d). A similar pattern was observed for nuclear pSMAD1/5/8, which was elevated in both aged groups relative to the young WT (Figure 6e). In contrast, the nuclear-to-cytoplasmic pSMAD1/5/8 ratio remained unchanged across all three groups, indicating that aging and APP/PS1 pathology increased overall pSMAD1/5/8 levels but did not appreciably alter its subcellular partitioning (Figure 6f).

Together, the compartmental data indicated a convergent shift toward cytoplasmic retention and/or diminished nuclear import of R-SMADs in aging and APP/PS1 cortices. Thus, while ligand and phospho-signals are detectable—and even elevated near plaques—the nuclear delivery step appeared curtailed, consistent with a decoupling between receptor–proximal activation and transcriptional execution in plaque-associated, reactive microglia.

While representative insets demonstrated the peri-plaque enrichment of pSMAD2, our quantitative analyses were designed a priori to capture group-level nuclear/cytoplasmic differences in Iba-1^+^ microglia within cortical ROIs, and therefore, were not stratified by plaque proximity.

3.4. Microglia Transcriptional Context

To place our morphological and signaling data into a broader molecular context, we re-analyzed the microglial RNA-sequencing dataset from Hemonnot-Girard et al., which reported pseudocount expression values for control microglia (CM), plaque-distant microglia (PCM), and PAM at 4, 8, and 12 months of age [35]. Within this framework, CM represented microglia from non-amyloid controls, PCM reflected cells located away from plaques in Aβ mice, and PAM corresponded to microglia directly associated with plaques.

As expected, classical homeostatic genes were most strongly expressed in CM and PCM, whereas PAM adopted a more disease-associated profile. P2ry12 and Tmem119 showed high average expression in CM and PCM and were reduced in PAM, particularly at 12 months of age. To characterize the DAM transcriptional profile, we examined the expression of key DAM markers (Apoe, Trem2, Tyrobp, and Cst7) across CM, PCM, and PAM at different stages of amyloid-β pathology (Figure 7). While Apoe was already low-expressed in CM and PCM, it was further elevated in PAM at 8 and 12 months of age, consistent with a prominent activation of the APOE-driven DAM program in plaque-associated microglia. Trem2 and Tyrobp followed a similar pattern, with relatively low expression in CM, intermediate levels in PCM, and the highest levels in PAM, reflecting progressive engagement of the TREM2–TYROBP axis in the plaque niche.

The cytochrome b-245 (Cybb) and the lysosomal cysteine protease inhibitor Cst7 showed the most striking plaque-associated upregulation. Both genes were weakly expressed in CM and only moderately increased in PCM, but strongly induced in PAM, where they clearly segregated PAM from PCM. Together, this marker profile confirmed that PAM adopted a robust DAM-like transcriptional state, characterized by the activation of the TREM2–APOE–TYROBP module and pronounced upregulation of Cybb and Cst7, distinguishing them functionally and molecularly from both control and PCM. Regarding the neuroinflammatory profile, markers involved in interferon signaling, such as Irf7, Usp18, and Pyhin1, were predominantly upregulated in 12-month PAM. In addition to interferon-related genes, PAM showed increased expression of Mmp12, a macrophage metalloelastase implicated in matrix degradation and chronic neuroinflammation, further supporting a shift toward a tissue-remodeling, pro-inflammatory microglial phenotype in the amyloid niche.

Apoc1 encodes apolipoprotein C-I, a component of lipoprotein particles that modulates lipid and cholesterol metabolism, and its upregulation in PAM was consistent with a lipid-remodeling, ApoE-linked DAM phenotype in the amyloid plaque niche.

We next examined components of the TGF-β pathway. Tgfb1 increased in CM compared to PCM and was further elevated in PAM—most prominently at intermediate disease stages—consistent with enhanced ligand availability in plaque-associated microglia. Tgfb2 showed a similar but more modest rise across CM→PCM→PAM, whereas Tgfb3 remained relatively stable or slightly reduced in PAM. Among the receptors, Tgfbr1 exhibited a gradual increase in CM compared to PCM and PAM, while Tgfbr2 displayed a more pronounced plaque-related gradient, with clearly higher levels in PCM and a peak in PAM at later stages. Thus, PAM upregulated not only DAM markers but also key TGF-β inputs, particularly Tgfb1 and Tgfbr2.

Core SMAD transcripts were largely maintained, with gene-specific shifts. Smad1, Smad2, and Smad5 were broadly similar across CM, PCM, and PAM, with only modest increases in PAM. In contrast, Smad3 tended to be lower in PAM than in CM/PCM, indicating a relative depletion of this canonical TGF-β R-SMAD in plaque-associated cells. Smad4 was slightly higher in PCM than in CM or PAM, and Smad7 remained comparable across groups. Together, these data point to a transcriptional configuration in PAM characterized by increased ligands/receptors but relatively lower Smad3, suggesting a subtle imbalance between upstream receptor availability and downstream canonical SMAD factors (Figure 7). Among commonly TGF-β-responsive genes, Fn1 and Tgfb1 were higher in PAM relative to CM/PCM, Smad7 showed no consistent reduction, and the homeostatic marker Tmem119 was lower in PAM. Overall, the transcriptome indicates strong upstream TGF-β input with selective, context-dependent target modulation, rather than a uniform decrease in TGF-β targets in PAM.

4. Discussion

This study demonstrated that aging and amyloid-β pathology jointly reshaped microglial structure and signaling, with convergent effects on TGF-β/SMAD pathways and disease-associated gene expression. At the morphological level, we observed a progressive simplification of microglial arborization in aged and APP/PS1 mice, characterized by reduced branch points, endpoints, and ramification index. Young microglia displayed a highly ramified, surveillant morphology, whereas microglia in the 24-month WT and APP/PS1 mice adopted a more compact, hypertrophic shape with shorter and fewer processes. Notably, the extent of morphological simplification in APP/PS1 microglia did not markedly exceed that of age-matched WT animals, indicating that aging alone drove the substantial structural remodeling of the microglial network. These findings aligned with the concept of “inflammaging”, in which age-related low-grade inflammation and chronic stress gradually erode the homeostatic microglial phenotype and lower the threshold for pathological activation [36].

Despite broadly similar morphology, prior transcriptomic work has shown that microglia in APP/PS1 models diverge functionally from aged WT counterparts, acquiring a distinct DAM profile characterized by the upregulation of Trem2, Apoe, Itgax, and Cst7 and the loss of homeostatic markers, such as P2ry12 and Tmem119 [26]. Our data were consistent with this view. The increased density of Iba-1^+^ microglia in the 24-month APP/PS1 cortex, together with their plaque-associated, hypertrophic morphology, suggests that aging provided the structural substrate onto which amyloid pathology imprinted a specialized, plaque-engaged state. The TREM2–TYROBP axis has been identified as a central driver of this transition, integrating lipid and damage-associated cues and promoting phagocytic, DAM-like programs [37,38]. It is therefore plausible that, even in the presence of similar arbor simplification, the APP/PS1 microglia were functionally distinct from aged WT cells due to additional TREM2-dependent reprogramming in the amyloid niche.

At the level of TGF-β signaling, we found clear evidence for disease-related alterations. Microglial TGF-β1 protein levels were highest in the 24-month APP/PS1 mice, indicating that plaque-engaged microglia were exposed to enhanced TGF-β ligand availability [38]. Our analysis of TGF-β pathway activity revealed more complex changes downstream. In cortical microglia, pSMAD2 immunoreactivity displayed a predominantly cytoplasmic distribution in the aged, AD group, with comparatively weaker nuclear accumulation.

Quantitative analysis of nuclear versus cytoplasmic pSMAD2 CTCF supported a model of impaired nuclear translocation: despite a robust cytoplasmic signal and local enrichment around plaques, nuclear pSMAD2 remained relatively low in aged APP/PS1 microglia. Together with the increased TGF-β1 and TGFBR2 expression observed in plaque-associated microglia, this pattern suggests a decoupling between strong upstream pathway activation and incomplete canonical nuclear SMAD2 output.

Our re-analysis of the Hemonnot-Girard bulk RNA-seq dataset provided independent support for this interpretation [35]. Focusing on microglial subclusters classified as CM, PCM, and PAM at different disease stages, we found that PAM adopted a robust DAM-like transcriptional signature characterized by the loss of homeostatic genes (P2ry12, Tmem119), strong induction of Apoe, Trem2, Tyrobp, and Cst7, and increased expression of Tgfb1 and Tgfbr2 relative to CM and PCM. Thus, PAM appeared to experience high TGF-β ligand and receptor availability in parallel with a full DAM program [38,39].

At the same time, canonical SMAD composition was subtly altered: while several SMAD family members (Smad1, Smad2, and Smad5) remained relatively stable or showed only modest increases, Smad3 tended to be reduced in PAM compared with CM and PCM. Given the critical role of SMAD3 as an R-SMAD in the canonical TGF-β pathway, its relative depletion provided a plausible transcriptional correlate of the reduced nuclear pSMAD2 accumulation that was observed histologically.

Taken together, these data support a model in which plaque-associated microglia receive strong TGF-β input (increased Tgfb1 and Tgfbr2, robust cytoplasmic pSMAD2, and peri-plaque enrichment), but transmit this signal only partially into canonical nuclear SMAD2/3-dependent transcription, in part due to altered SMAD3 availability [40]. In this scenario, TGF-β signaling may be sufficient to sustain aspects of the DAM phenotype and limit overt cytotoxicity, but insufficient to fully engage the homeostatic, anti-inflammatory, and synapse-supporting programs typically associated with microglial TGF-β signaling in the healthy brain.

The selective vulnerability of the SMAD2/3 branch contrasted with our observations on pSMAD1/5/8. Both cytoplasmic and nuclear pSMAD1/5/8 levels were significantly elevated in microglia from 24-month WT and APP/PS1 mice compared with young WT mice, yet the nuclear-to-cytoplasmic ratio remained unchanged across groups. This indicates that aging and amyloid pathology globally increased SMAD1/5/8 signaling in microglia, while preserving the balance between cytoplasmic and nuclear localization. Given that SMAD1/5/8 are classically associated with BMP rather than TGF-β ligands, this pattern suggests that the BMP–SMAD axis remained functionally competent and may even be upregulated in the aged and amyloid-bearing cortices [41,42]. Enhanced SMAD1/5/8 activity could support tissue remodeling or survival responses, but it might also contribute to maladaptive matrix degradation and gliosis, depending on the downstream gene programs engaged [43,44]. Importantly, the preserved nuclear access of pSMAD1/5/8 argues against a global failure of SMAD trafficking and instead points to a selective impairment of the TGF-β–SMAD2/3 arm, consistent with studies showing that linker-region phosphorylation and MAPK-dependent modification of SMAD2/3 can specifically inhibit their nuclear translocation and canonical TGF-β signaling [45].

Several mechanisms could underlie the dissociation between upstream TGF-β activation and downstream SMAD2/3 function. A plausible mechanistic explanation is the physical sequestration or mislocalization of SMAD proteins by pathological tau aggregates [46]. In addition, cross-talk from other inflammatory cascades can oppose canonical SMAD2/3 signaling:

• (1)ERK/MAPK can phosphorylate the SMAD2/3 linker region, reducing nuclear import and transcriptional output [47,48];
• (2)JNK/p38 MAPK signaling can further shift SMADs toward cytoplasmic retention and modulate target selection [49,50];
• (3)type-I interferon–JAK/STAT (STAT1/2) programs can dominate the transcriptional landscape and dampen SMAD-dependent genes [51];
• (4)NF-κB can compete for limiting co-activators (CBP/p300), thereby constraining SMAD-mediated transcription [52].

In PAM, where we observed strong DAM activation, upregulation of interferon-related genes (Irf7, Usp18, and Pyhin1), increased expression of proteolytic and matrix-remodeling factors, such as Mmp12, and lipid-associated genes, including Apoc1, it is conceivable that MAPK and interferon pathways competed with or overrode canonical TGF-β–SMAD2/3 signaling [53,54]. Reduced SMAD activity would then not solely reflect intrinsic defects in the TGF-β pathway, but also the influence of these converging pro-inflammatory networks.

Our findings, therefore, place TGF-β/SMAD signaling into a broader context of microglial state transitions in aging and AD. Aging alone induces morphological simplification and increases overall SMAD1/5/8 activity, while preserving a more homeostatic transcriptomic profile. The addition of amyloid pathology drives microglia into a PAM/DAM state with pronounced TREM2–TYROBP activation, interferon signaling, lipid remodeling, and upregulated TGF-β ligand and receptor expression. Within this state, the canonical TGF-β–SMAD2/3 axis appears to be partially uncoupled, with high upstream input but incomplete nuclear output (Figure 8).

This imbalance may stabilize a chronically reactive, plaque-engaged phenotype that is only partially restrained by endogenous anti-inflammatory and pro-homeostatic signals.

Limitations

Several limitations of our study should be acknowledged. First, our morphological and signaling analyses were based on a relatively small number of animals and a cross-sectional design, which precluded firm conclusions about temporal causality between amyloid deposition, TGF-β signaling changes, and microglial remodeling.

Second, our quantification of pSMAD localization relied on immunofluorescence and corrected total cell fluorescence; while informative, this approach could not fully capture dynamic signaling kinetics or resolve microglial subpopulations beyond Iba-1 positivity. Although we quantified pSMAD nuclear translocation from z-projected confocal stacks to ensure uniform throughput and statistical power, we did not perform full 3D volumetric segmentation across the entire dataset. Future studies will need to incorporate voxel-wise 3D readouts to complement the 2D N/C metric, particularly in densely plaque-rich microenvironments.

Third, our transcriptomic conclusions were derived from the re-analysis of a published bulk-RNA dataset; although this dataset was well annotated and highly informative, it represented one specific APP/PS1 model and might not encompass the full spectrum of microglial states across brain regions and disease stages. Finally, we did not directly manipulate TGF-β/SMAD or MAPK pathways in vivo, so the causal contribution of these signaling changes to microglial function and neurodegeneration remains to be established.

Despite these limitations, our combined structural, signaling, and transcriptomic analyses highlight morphological degeneration and altered TGF-β/SMAD signaling in microglia as key features of the aged and Alzheimer’s brain. Plaque-associated microglia emerged as structurally simplified yet transcriptionally hyperactive cells that integrated strong TREM2–TYROBP, interferon, and TGF-β inputs, but relayed only an incomplete canonical SMAD2/3 response. Modulating this imbalanced signaling landscape—either by restoring effective nuclear SMAD2/3 translocation or by dampening competing pro-inflammatory cascades, such as ERK/MAPK and interferon pathways—may help re-establish glial neuroprotective functions and potentially slow disease progression. Figure 9 provides a schematic comparison of pathway organization and subcellular routing in control/plaque-distant and plaque-associated microglia according to our results.

5. Conclusions

By integrating high-resolution morphometrics, confocal readouts of subcellular SMAD localization, and a targeted re-analysis of bulk RNA-seq, we identify a branch-specific imbalance in microglial TGF-β signaling at amyloid plaques. Plaque-associated microglia show strong upstream TGF-β input—increased TGF-β1/TGFBR2 with robust cytoplasmic pSMAD2—yet attenuated nuclear SMAD2/3 output, while BMP–SMAD1/5/8 nuclear access remains preserved. This selective decoupling mechanistically links chronic plaque engagement to a stable, DAM-like state and nominates restoration of SMAD2/3 nuclear signaling as a therapeutic avenue. Re-analysis of bulk RNA-seq across plaque-proximity strata supports coordinated shifts in the TGF-β axis (ligands, receptors, and SMADs) and aligns the histological phenotype with transcriptional changes (higher Tgfb1/Tgfbr2 with comparatively lower Smad3).

Future work combining high-resolution morphometrics with spatial transcriptomics and targeted pathway manipulation will be essential to dissect how these intertwined signaling networks shape microglial behavior in the aging and amyloid-affected brain and to identify points of therapeutic leverage within the TGF-β/SMAD pathway.