Profilin 1 controls a microglial cytoskeleton checkpoint to prevent senescence and premature synaptic decline

Abstract

Profilin 1 (Pfn1) expression decreases significantly in aged human microglia, suggesting that loss of cytoskeletal integrity may trigger microglial senescence and increased synaptic vulnerability. To test this hypothesis, we used an inducible, microglia-specific Pfn1 knockout in adult mice, a strategy designed to isolate the direct effects of acute Pfn1 loss at the cellular and circuit levels, avoiding confounding factors from development or chronic aging. Using a multi-omics approach combined with intravital two-photon imaging, we found that Pfn1 ablation disrupts actin–microtubule coupling, leading to a collapse of microglial morphodynamics and a complete failure to respond to focal brain injury. This cytoskeletal disruption triggers a cell-autonomous, senescence-associated secretory phenotype (SASP), driven by the ERK/NF-κB signaling axis. SASP factors secreted by Pfn1-deficient microglia reprogram the synaptic environment, resulting in significant deficits in mitochondrial energy production and a selective reduction in the frequency of GABAergic inhibitory postsynaptic currents in the prefrontal cortex. These circuit-level disturbances lead to behaviors characterized by altered anxiety and risk assessment. Our findings identify Pfn1 as a critical checkpoint against microglial senescence and show that its loss is sufficient to drive circuit-specific synaptic decline, highlighting the Pfn1-cytoskeleton axis as a potential therapeutic target to enhance brain resilience.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12974-025-03588-z.

Introduction

Microglia are essential immune sentinels of the central nervous system (CNS) that continuously survey their surroundings, support synaptic plasticity, and coordinate inflammatory responses to injury and disease [1]. However, with advancing age, microglia undergo profound morphofunctional changes—shifting from a highly ramified, dynamic state to a senescence-like phenotype characterized by retracted processes, decreased motility and phagocytic capacity, and a heightened pro-inflammatory secretory profile [2–4]. These age-associated changes exacerbate chronic neuroinflammation and increase neuronal vulnerability, contributing to cognitive and behavioral deficits as well as the pathogenesis of age-related neurodegenerative disorders such as Alzheimer’s and Parkinson’s disease [2–4]. Notably, the pro-inflammatory milieu produced by senescent microglia resembles the SASP observed in other aging cells [5, 6], suggesting that common cellular aging mechanisms may be at play in the brain’s immune compartment.

Transcriptomic analyses of human microglia have revealed that the expression of several cytoskeleton-regulating genes, most prominently Pfn1, declines significantly with age [7]. This finding raises the possibility that progressive cytoskeletal breakdown is a key driver of the microglial aging phenotype, rather than a mere bystander effect of aging. In other words, a loss of cytoskeletal regulators, such as Pfn1, might actively instigate microglial senescence and dysfunction.

The cytoskeleton, composed of actin filaments and microtubules, is fundamental to cellular architecture, motility, and intracellular transport [8]. Pfn1, a small (~ 15 kDa) evolutionarily conserved protein, plays a pivotal role in regulating these cytoskeletal elements [9, 10]. As a gatekeeper for actin dynamics, Pfn1 controls the availability of ATP-bound G-actin monomers, accelerates filament elongation, and modulates filament nucleation via interactions with formins and Ena/VASP proteins [11]. Through these actions, Pfn1 orchestrates the formation of distinct actin structures, enabling the cell to adapt its morphology and behavior in response to changing environmental cues. In addition, Pfn1 contributes to microtubule stability and coordinates actin–microtubule crosstalk, which is essential for proper intracellular trafficking and overall cellular integrity [11].

While Pfn1’s roles in neurons and astrocytes have been relatively well-characterized, influencing cell migration, neurite outgrowth, and synaptic remodeling [12, 13], its function in microglia remains largely unexplored. The observed decline of Pfn1 in aged human microglia [7] suggests that disrupted cytoskeletal dynamics may contribute directly to the functional deterioration of microglia. Since microglia critically regulate synapse remodeling and plasticity, perturbations to their actin cytoskeleton can propagate through neural circuits, weakening connectivity and contributing to network deficits. On this basis, we hypothesized that loss of Pfn1 in adult microglia is sufficient to initiate a senescence-like program and synaptic decline.

Although Pfn1 expression declines with age in human microglia [7], this trajectory is not consistently observed in laboratory mice. Due to this species divergence, our goal here was not to phenocopy mouse aging, but to test causal sufficiency. We therefore used an adult-onset, tamoxifen-inducible, microglia-specific Pfn1 conditional knockout (cKO) to isolate the acute consequences of microglial cytoskeletal disruption after brain development is complete—independent of developmental and chronic-aging confounds. Under this design, measuring Pfn1 in aged mouse microglia is not probative but orthogonal to our hypothesis. Correlative evidence from aged human microglia [7] and research on cytoskeletal alterations and senescence [14–18] contextualizes the approach, justifying the evaluation of synaptic and behavioral outcomes as downstream consequences of microglial Pfn1 loss.

Here, integrating RNA-seq, deep proteomics and phosphoproteomics, intravital two-photon imaging, ex vivo patch-clamp electrophysiology, and behavioral assessments, we comprehensively determined how acute Pfn1 deficiency reshapes microglial function, perturbs the synaptic landscape, and impacts circuit-level behavior. Our findings provide novel evidence linking microglial cytoskeletal integrity to the maintenance of synaptic homeostasis.

Results

Pfn1 is essential for microglia morphodynamic plasticity

To investigate Pfn1’s role in microglia, we developed a conditional knockout model by crossing Cx3cr1^CreERT2 mice with mice carrying floxed Pfn1 alleles (Fig. 1A). The CreERT2 transgene is active in Cx3cr1 + brain myeloid cells and microglia [19, 20], enabling tamoxifen (TAM)-induced deletion of floxed alleles in adult microglia in the brain parenchyma [21, 22].

Fig. 1.

Loss of Pfn1 in microglia impairs cytoskeletal dynamics and surveillance capacity. A A schematic of the conditional knockout (cKO) model for deleting Pfn1 in microglia. The model was generated by crossing Cx3cr1^CreERT2 mice with Pfn1 floxed mice (Cx3cr1^CreERT2+:Pfn1.^fl/fl). TAM was administered to 4–5-week-old mice, and analyses were conducted 6–8 weeks after induction. B A Western blot showing Pfn1 protein levels in brain microglia isolated via MACS from cKO and control (CT) mice. The blot displays results from 4 independent biological replicates (4 mice per genotype). C Representative Iba1 + microglia images from cortical brain sections, acquired via deepSIM microscopy. The images display the morphology of microglia from both genotypes (n = 3-4 cells/animal, 5 animals/group). D 3D reconstructions of CD68 + vesicles in cortical microglia. A graph presents the quantification of vesicular volume per cell (n = 3-4 cells/animal, 5 animals per group). E and F Images from intravital two-photon microscopy of EYFP + cortical microglia at baseline and following a laser-induced injury (LI). A graph shows the trajectory analysis of protrusion velocity and directional motility (n = 5–11 protrusions/mouse, 3 mice/genotype). G A graph presenting the directional polarization analysis after LI. The plot displays the global and instantaneous velocities of microglial processes (n = 12–20 cells or 189–496 protrusions pooled from 3 mice per genotype). Data are presented as mean ± SD. Statistical significance was determined using Student’s t-test for panels B and D, two-way ANOVA for panel C, and Linear mixed models for panels F and G: *p < 0.05, **p < 0.01, ****p < 0.0001. Scale bars: 50 µm (C-low), 10 µm (C-high), 10 µm (D), 20 µm (E)

TAM was administered to 4–5-week-old control (Pfn1^fl/fl, CT) and conditional knockout (Cx3cr1^CreERT2+:Pfn1^fl/fl, cKO) mice to induce Pfn1 deletion in adult microglia. All subsequent analyses were performed 6–8 weeks post-induction. Western blotting of microglia isolated by MACS confirmed a significant and efficient reduction of Pfn1 protein in cKO mice compared to controls (Fig. 1B), validating the model.

In a healthy brain, microglia maintain a ramified shape to monitor the parenchyma [23]. Aging is linked to a shift toward a less complex, more reactive morphology [24, 25], a process closely connected to cytoskeletal regulation [26]. To determine if Pfn1 loss mimics these features, we used confocal and structured-illumination super-resolution microscopy (deepSIM) on Iba1 + microglia. Morphological analysis showed significant structural changes in Pfn1-deficient microglia (Fig. 1C). Compared to the highly branched CT cells, cKO microglia had hypertrophic primary processes, less overall branching, shorter total filament length, fewer branch points, less filaments, fewer terminal points, and increased soma volume (Fig. 1C and Suppl. Figure 1 A).

Aging microglia show deficits in vesicular trafficking and phagocytic activity, often linked to the accumulation of CD68 + lysosomal vesicles [27]. Consistent with this, IMARIS-based 3D rendering revealed a significant increase in the total volume of CD68 + vesicles within each Pfn1 cKO microglia, an effect that was independent of overall cell volume (Fig. 1D). This finding suggests that Pfn1 loss disrupts intracellular vesicular dynamics in a way that parallels features of microglial aging [25, 28].

Building on these observations, we next sought to evaluate the functional consequences of Pfn1-related changes on microglial dynamics in vivo in the brain parenchyma. To this end, we performed intravital two-photon (2P) imaging in CT (Cx3cr1^CreER+) and Pfn1 cKO mice. Imaging was conducted on EYFP + cortical microglia under baseline conditions and after laser-induced injury (LI) to the cortical parenchyma, a paradigm that mimics chemotactic responses to focal brain damage [29].

To quantify the trajectories and dynamics of microglial protrusions, we utilized a computational approach using cartographic methods measuring Euclidean distances (Suppl. Figure 1B). This approach allowed for precise tracking of microglial cell bodies and protrusions over time, enabling automated analysis of key parameters such as protrusion size, velocity, directionality, and polarization dynamics, particularly in response to LI.

Pfn1 cKO microglia exhibited significantly reduced surveillance capacity compared to controls (Fig. 1E, Suppl. Figure 1). Quantification of motility parameters revealed substantial impairments in both global average velocity (VGlobal), reflecting overall surveillance efficiency (Suppl. Figure 1 C), and instantaneous velocity (VInst), capturing real-time responsiveness (Suppl. Figure 1D).

The most prominent finding was the six-fold reduction in protrusion motility in Pfn1 cKO microglia compared to controls, as reflected by global extension and retraction velocities (VGlobalExt-cKO = 0.03 ± 0.018 µm/min, VGlobalRet-cKO = 0.07 ± 0.055 µm/min vs. VGlobalExt-CT = 0.24 ± 0.23 µm/min, VGlobalRet-CT = 1.12 ± 0.61 µm/min; Suppl. Figure 1 C). Instantaneous velocity analysis corroborated this result, showing a three-fold reduction in real-time extension and retraction responsiveness in cKO microglia (VInstExt-cKO = 0.69 ± 0.69 µm/min, VInstRet-cKO = 0.69 ± 0.67 µm/min vs. VInstExt-CT = 2.08 ± 2.02 µm/min, VInstRet-CT = 2.46 ± 2.16 µm/min; Suppl. Figure 1D). These findings suggest that Pfn1 deficiency disrupts the structural rearrangements required for efficient protrusion dynamics, leaving microglia less capable of surveilling their microenvironment.

Following LI, control microglia, as expected, rapidly extended protrusions toward the injury site. This robust response involved a marked increase in protrusion extension amplitude and directional polarization toward the injury (Fig. 1E-G). In contrast, Pfn1 cKO microglia exhibited significantly blunted responses, with protrusion amplitude reduced by more than 50% relative to controls (Fig. 1F). Furthermore, their directional efficiency, as measured by polarization velocities (global and instantaneous), was significantly reduced in cKO microglia (Fig. 1G). These reductions suggest that Pfn1-deficient microglia extend protrusions more slowly and are less capable of responding to injury cues. Such deficits in motility and directional precision highlight the critical role of Pfn1 in enabling microglia to react effectively to localized brain damage. These results demonstrate how Pfn1-dependent cytoskeletal dynamics support microglial morphofunctional plasticity, enabling them to efficiently monitor their environment and respond to tissue damage.

Pfn1 regulates a cytoskeletal program akin to aged human microglia

To map the molecular underpinnings of the observed morphological and functional deficits, we performed RNA-seq on sorted microglia from Pfn1 cKO and CT brains 6–8 weeks after TAM induction. Differential expression analysis identified 2,015 significantly altered transcripts in cKO microglia, with 1,362 upregulated and 653 downregulated genes (Fig. 2A; Supplementary Table 1). Notably, several of these genes are associated with aging-related processes, including Cxcr4, Ms4a4a, Runx3, Cd163, and Irf4.

Fig. 2.

RNA sequencing reveals transcriptomic alterations in Pfn1-deficient microglia, mimicking aged phenotypes. A A volcano plot from RNA sequencing of sorted microglia from Pfn1 cKO and CT brains 6–8 weeks after TAM induction (n = 3 mice per genotype). B A bubble-plot-like chart showing Gene Ontology (GO) analysis of cytoskeleton-related genes, highlighting alterations in structures such as membrane ruffles, lamellipodia, and stress fibers. C A Venn diagram from a comparative analysis showing the overlap in gene expression changes between Pfn1 cKO microglia and published transcriptomes of aged human microglia. D A heatmap highlighting key commonly altered cytoskeletal regulatory pathways, including cortical actin organization, membrane ruffling, and stress fiber formation. E A series of Venn diagrams from contingency analysis comparing transcriptomic profiles. The diagrams show comparisons between Pfn1 cKO microglia and aged mouse microglia (upper), aged mouse microglia and aged human microglia (middle), and Pfn1 cKO microglia and aged human microglia (bottom). F A heatmap of commonly affected transcripts across the compared profiles in E

Given the critical role of the cytoskeleton in microglial function, we focused on genes related to cytoskeletal organization. Gene Ontology (GO) analysis revealed significant alterations in components of both the actin and microtubule cytoskeletons, affecting structures such as membrane ruffles, lamellipodia, actomyosin complexes, and stress fibers (Fig. 2B). These changes suggest that Pfn1 loss disrupts cytoskeletal integrity at multiple levels.

To determine whether these molecular changes relate to those occurring in aged human microglia, we compared our dataset with published transcriptomes of microglia isolated from aged human subjects [7]. Comparative analysis revealed a significant overlap in gene expression changes between Pfn1 cKO microglia and aged human microglia, particularly in genes related to cytoskeletal regulation (Fig. 2C; Supplementary Table 2). Chi-square analysis confirmed a strong association between the datasets (p ≈ 2.42 × 10^ − 59), indicating that the cytoskeletal dysregulation observed in Pfn1 cKO microglia mirrors, to a large extent, that of aged human microglia. Key cytoskeletal regulators commonly altered in both datasets include genes involved in cortical actin organization, membrane ruffling, and stress fiber formation (Fig. 2D).

To further investigate the aging-like phenotype induced by Pfn1 ablation, we compared the transcriptomic profiles of Pfn1 cKO microglia with those of microglia from aged mice (Hickman et al., 2013). Contingency analysis revealed a robust positive correlation between Pfn1 cKO microglia and aged mouse microglia transcriptomes (Fig. 2E; Supplementary Table 2). Interestingly, the transcriptomic signature of aged human microglia showed a stronger correlation with Pfn1 cKO microglia than with aged mouse microglia. Among these profiles, commonly affected transcripts include Cxcr4, Ms4a4a, Itgal, Runx3, Emilin1, Rgs7bp, Tmc7, Tmem204, Dlc1, and Adgra3 (Fig. 2F).

Pfn1 suppresses the emergence of a senescence-like transcriptomic program

The cytoskeletal alterations observed in Pfn1 cKO microglia suggest that the loss of Pfn1 may lead to broader transcriptomic changes beyond structural modifications. We hypothesized that Pfn1 deficiency not only mirrors aging-associated transcriptomic alterations but also induces a senescence-like state in microglia. To test this hypothesis, we performed comprehensive bioinformatic analyses of our RNA-seq data to identify transcriptomic signatures characteristic of cellular senescence [30].

Initially, we conducted contingency analyses to compare the transcriptomic profile of Pfn1 cKO microglia with established microglial gene expression signatures commonly influenced during normal aging, such as inflammation [31], oxidative stress [32], and engulfment [33]. The chi-square analyses revealed a strong and significant association between these traditional aging-related signatures and the transcriptome of Pfn1 cKO microglia (Fig. 3A; Supplementary Table 2).

Fig. 3.

Loss of Pfn1 induces a senescence-like transcriptional phenotype in microglia. A Contingency analysis comparing the transcriptomic profile of Pfn1 cKO brain microglia 6–8 weeks after TAM induction (n = 3 mice/genotype) with established gene signatures for inflammation, oxidative stress, and engulfment that are commonly influenced during normal aging. The panel includes a Venn diagram and associated heat maps for selected transcripts to illustrate the association between the datasets. B A Cytoscape-based visualization of pathway clustering derived from enrichment analysis of differentially expressed genes. C A heatmap displaying the expression levels of genes comprising a Senescence-Associated Secretory Phenotype (SASP)-like signature in Pfn1 cKO microglia. D A heatmap showing the expression of genes within the MAPK-ERK signaling pathway. The visualization illustrates critical components that contribute to ERK activation and subsequent NF-κB-mediated transcription of SASP factors

To better understand the affected biological processes, we performed statistical over-representation analysis (ORA) coupled with GO and pathway analyses on the differentially expressed genes. This approach revealed significant alterations in 116 pathways and 467 biological processes in the microglia following Pfn1 ablation (6–8 weeks after TAM induction). After consolidating redundant terms, we identified nine major Pfn1-regulated signaling pathways in microglia (Fig. 3B). These pathways include immune function, growth factor signaling, energy metabolism, protein processing and degradation, RNA metabolism, DNA repair, intracellular trafficking, cell cycle regulation, and mitochondrial function. Notably, these Pfn1-regulated transcriptomic modules are strongly associated with the transition to a senescence-like state in microglia.

A hallmark of cellular senescence is the SASP, characterized by the elevated secretion of pro-inflammatory cytokines, chemokines, growth factors, and proteases [5, 6]. SASP factors contribute to a chronic inflammatory environment and extracellular matrix (ECM) degradation, exacerbating tissue dysfunction and promoting aging-related pathologies. Moreover, SASP is linked to alterations in intracellular signaling pathways, including deficits in DNA damage response (DDR) mechanisms and metabolic reprogramming toward glycolysis [5, 6].

Consistent with these characteristics, Pfn1 ablation in microglia led to changes in multiple transcripts aligning with a SASP-like signature (Fig. 3C). Genes involved in DNA damage and repair, such as Cdc14b and Ticrr, were upregulated, suggesting compromised genomic integrity and activation of DDR pathways. This suggests that Pfn1 cKO microglia may experience increased DNA damage or reduced repair capacity, which are common features of senescent cells.

Genes linked to glycolytic metabolism, including Igf1, Nupr1, and Pde2a, were also upregulated, indicating a metabolic shift toward glycolysis. Senescent cells often rely on glycolysis for energy production, a phenomenon known as the Warburg effect [34]. This metabolic reprogramming supports the energy demands of SASP factor production and secretion. Additionally, genes encoding ECM-degrading enzymes, such as Adamts20 and Mmp9, were significantly elevated, highlighting increased matrix remodeling activity. The upregulation of these proteases can lead to ECM degradation, facilitating the infiltration of immune cells and further promoting inflammation. Notably, classical pro-inflammatory mediators C3, Il1β, and Tnf were upregulated, underscoring a heightened inflammatory state in Pfn1 cKO microglia. Elevated levels of these cytokines can have deleterious effects on neuronal function and survival, contributing to neuroinflammation and neurodegeneration.

The ERK signaling pathway directly mediates cellular senescence and inflammatory responses, especially in microglial aging [35]. In line with this, we found that Pfn1 cKO microglia displayed substantial alterations in the MAPK-ERK signaling pathway (Fig. 3D). Mechanistically, Pfn1 cKO microglia exhibited upregulation of critical components in mitogen signaling that directly contribute to ERK activation (Fig. 3D). This modulation of the MAPK-ERK pathway facilitates the transcription of SASP factors, including pro-inflammatory cytokines, chemokines, and matrix-degrading enzymes, through the activation of NF-κB (Fig. 3D).

Activation of NF-κB is a central event in the induction of SASP, leading to increased expression of pro-inflammatory cytokines, chemokines, and matrix-degrading enzymes [36]. The upregulation of Il1β, Tnf, and Mmp9 in Pfn1 cKO microglia supports the involvement of NF-κB signaling in promoting a senescent phenotype.

The alterations observed in MAPK/ERK signaling align with the characteristics of aged microglia [35] and establish a mechanistic basis for acquiring a SASP-like signature by Pfn1 cKO microglia. These findings suggest that Pfn1 loss activates pro-senescent pathways, driving microglia toward a state characterized by chronic inflammation and altered functionality.

Loss of Pfn1 in microglia triggers age-related transcriptomic changes in the young brain

To assess whether this senescence-like state of Pfn1 cKO microglia broadly impacts the brain environment, we performed RNA-seq on cortical tissue from Pfn1 cKO and CT mice 6–8 weeks after TAM induction (Fig. 4A). Gene set enrichment analysis (GSEA) revealed substantial changes in the brain transcriptome, affecting functional modules such as cellular metabolism, proteostasis, synaptic transmission, and cell surface receptor signaling (Fig. 4B).

Fig. 4.

Pfn1 deficiency in microglia reprograms the brain transcriptome. A A diagram of the experimental workflow for bulk RNA-seq performed on cortical tissue from Pfn1 cKO and CT brains 6–8 weeks after TAM induction (n = 3 per genotype). B Signaling modules and clustered heatmaps generated from the RNA-seq data, linked to cellular metabolism, proteostasis, synaptic transmission, and RTK-GPCR signaling

Mitochondrial function and energy production pathways were dysregulated, reflecting impairments in cellular metabolism commonly observed during aging [37]. Proteostasis networks, including components of the ubiquitin–proteasome system and chaperones, were also disrupted (Fig. 4B), consistent with age-related declines in protein quality control [38].

Notably, we also identified alterations in pathways related to synaptic transmission, particularly those involved in GABAergic signaling (Fig. 4B). Synaptic dysfunction is a hallmark of aging and neurodegenerative diseases, contributing to cognitive decline [39]. Changes in receptor tyrosine kinase (RTK) and G protein-coupled receptor (GPCR) signaling pathways, including EGF and Ras signaling, were also observed (Fig. 4B). These pathways are critical for neuronal survival, synaptic plasticity, and response to environmental stimuli and are often dysregulated during aging.

Deep proteomic and phosphoproteomic profiling reveals alterations consistent with synaptic aging in Pfn1 cKO brains

The brain RNA-seq data instructed us to conduct a more in-depth investigation of the impact of microglial Pfn1 ablation on the synaptic landscape. We performed deep, high-throughput proteomics and phosphoproteomics on synaptosomal preparations from Pfn1 cKO and CT mice (6–8 weeks after TAM induction) using liquid chromatography-tandem mass spectrometry (LC–MS/MS) coupled to unbiased label-free quantifications (Fig. 5A).

Fig. 5.

Deep proteomic and phosphoproteomic profiling reveals synaptic aging-like phenotypes in Pfn1 cKO brains. A A diagram of the experimental workflow for deep proteomics. Cortical synaptosomal preparations from Pfn1 cKO and CT brains (n = 5 per genotype) were analyzed 6–8 weeks after TAM induction using liquid chromatography-tandem mass spectrometry (LC–MS/MS) with label-free quantification. B A summary of the proteomic analysis in Pfn1 cKO and CT cortical synaptosomes. The panel highlights results from Gene Ontology (GO) and pathway enrichment analyses. C A diagram of the experimental workflow for deep phosphoproteomics. The analysis was performed on independent cortical synaptosomal preparations (n = 5 per genotype) using a workflow similar to that in (A), but with an additional titanium dioxide enrichment step for phosphopeptides. D Representative examples of proteins with altered phosphorylation states, including those involved in mitochondrial turnover, proteostasis, energy metabolism, calcium signaling, and mitochondrial transport

Differential expression analysis identified significant reconfigurations of the synaptic proteome, with 85 proteins upregulated and 52 downregulated in Pfn1 cKO synaptosomes compared to CTs (Fig. 5A and Suppl. Table 3). Gene Ontology (GO) and pathway enrichment analyses highlighted alterations in modules related to synaptic mitochondria, energy metabolism, proteostasis, and local protein synthesis (Fig. 5B). The observed changes in mitochondrial proteins and components of the respiratory chain suggest impairments in synaptic energy production, which can affect neurotransmitter release and synaptic plasticity [40]. Disruptions in proteostasis mechanisms, including chaperone complexes and ubiquitin-mediated degradation, indicate deficits in protein quality control at the synapse [41]. Additionally, altered expression of ribosomal and RNA-binding proteins involved in local translation suggests impairments in synaptic protein synthesis, critical for long-term synaptic modifications [42].

Phosphorylation is a critical post-translational modification that regulates protein activity, interactions, and signaling pathways, thereby providing additional layers of regulation beyond protein abundance. To complement the global proteomic data and gain further mechanistic insight into the impact of microglia Pfn1 ablation on the synapses, we conducted phosphoproteomic analyses using titanium dioxide enrichment and independent downstream deep LC–MS/MS runs on the same synaptosomal preparations (Fig. 5C and Suppl. Table 4). The phosphoproteomic data revealed extensive changes in the phosphorylation states of proteins involved in mitochondrial function, energy metabolism, and proteostasis (Fig. 5C and D), further broadening the functional impact of microglial Pfn1 ablation on synapses.

The altered phosphorylation of proteins such as Bnip3l and Bnip3, closely linked to mitochondrial-mediated apoptosis and mitophagy, suggests significant disruptions in mitochondrial turnover and integrity at the synapse [43]. These proteins interact with Huwe1, an E3 ubiquitin ligase involved in mitochondrial protein degradation and apoptotic regulation, thereby creating a pathway critical for mitochondrial quality control and apoptosis [43]. Additionally, Prkce interacts with Camk2a, a calcium/calmodulin-dependent kinase, impacting calcium regulation and mitochondrial energy production through intricate signaling cascades [44].

Phosphorylation changes in chaperones like Hsp90aa1 and Hspa4 potentially highlight compromised protein folding and stress response mechanisms essential for proteostasis at the synapse [45]. Huwe1 also showed altered phosphorylation states, indicating potential deficits in ubiquitin-mediated protein degradation pathways vital for maintaining protein quality control. These alterations suggest that the proteostasis network is impacted in the synapses of Pfn1 cKO mice.

Signaling pathways directly linked to energy metabolism were also notably affected, with kinases such as Gsk3b, Prkaca, and Prkce displaying altered phosphorylation states. Gsk3b interacts with Prkaca and Prkce, forming a regulatory network crucial for energy signaling and glycogen metabolism, thereby maintaining synaptic energy balance [46]. Bckdk, another metabolic regulator, works alongside Gsk3b to modulate amino acid and glucose metabolism, which is essential for sustaining synaptic energy supply [47]. These phosphorylation changes suggest significant disruptions in the signaling related to energy metabolism critical for synaptic function.

Calcium signaling was also potentially affected, as evidenced by altered phosphorylation of Camk2a and Calm2, a key calcium-binding protein. Camk2a plays a pivotal role in calcium-dependent synaptic plasticity, and its dysregulated phosphorylation can impair long-term potentiation and memory formation. Calm2, interacting with Ppp3cb (calcineurin), modulates calcium-dependent signaling pathways vital for synaptic plasticity and energy consumption. These disruptions in calcium homeostasis can adversely affect mitochondrial energy production and overall synaptic homeostasis.

Furthermore, cytoskeletal components linked to mitochondrial transport were impacted, with proteins such as Kif1b, Vps35, and Tppp showing altered phosphorylation states. Kif1b is a motor protein responsible for mitochondrial transport along axons, and Vps35 regulates mitochondrial trafficking and recycling [48]. Tppp promotes tubulin polymerization, supporting cytoskeletal stability and ensuring proper mitochondrial positioning within neurons [48]. These phosphorylation changes suggest impaired mitochondrial transport and cytoskeletal organization.

Collectively, the proteomic and phosphoproteomic changes observed in the Pfn1 cKO synapses not only corroborate the brain RNA-seq data but also pinpoint molecular alterations indicative of premature synaptic aging, including mitochondrial dysfunction, impaired energy metabolism, disrupted calcium signaling, and compromised proteostasis [49].

Metabolic analysis reveals premature mitochondrial energy deficits in Pfn1 cKO synapses

To validate and extend the molecular insights gained from our proteomic and phosphoproteomic analyses, we conducted a mitochondrial stress test using the Seahorse XF Analyzer on synaptosomal preparations from Pfn1 cKO and CT mice 6–8 weeks after TAM induction (Fig. 6A). This assay quantitatively measures key parameters of mitochondrial respiration, providing functional corroboration of the observed proteomic and phosphoproteomic alterations.

Fig. 6.

Metabolic profiling reveals mitochondrial dysfunction in synaptosomes from Pfn1 cKO brains. A A diagram of the experimental workflow using the Seahorse XF Analyzer to perform mitochondrial stress tests on cortical synaptosomes of Pfn1 cKO and CT mice 6–8 weeks after TAM induction (n = 3 mice/genotype). B A graph showing the level of basal respiration in cKO synaptosomes compared to controls. C A graph showing the level of maximal respiration, assessed after FCCP uncoupling, in cKO synaptosomes compared to controls. D A graph showing the level of proton leak in cKO synaptosomes compared to controls. E A graph showing the spare respiratory capacity, calculated as the difference between maximal and basal respiration, in cKO synaptosomes compared to controls. F A graph showing the level of ATP production-linked respiration in cKO synaptosomes compared to controls. G A graph showing the coupling efficiency in cKO synaptosomes compared to controls. Data are presented as mean ± SEM. Statistical significance was determined using paired t-test: *p < 0.05, **p < 0.01

The Seahorse analysis revealed a significant reduction in basal respiration in Pfn1 cKO synaptosomes compared to CTs (Fig. 6B). This decrease in basal oxygen consumption rate (OCR) aligns with the downregulation of mitochondrial proteins involved in energy metabolism, indicating a lower baseline mitochondrial activity in the absence of microglial Pfn1. Maximal respiration, assessed after the addition of the uncoupler agent carbonyl cyanide-p-trifluoromethoxyphenylhydrazone (FCCP), was significantly lower in Pfn1 cKO synaptosomes (Fig. 6C). This reduction indicates a diminished capacity for oxidative phosphorylation, corroborating the proteomic findings of impaired energy metabolism pathways and suggesting that Pfn1 cKO mitochondria have a reduced ability to meet increased energy demands. Proton leak was also significantly diminished in Pfn1 cKO synaptosomes (Fig. 6D), further reflecting an overall decline in mitochondrial function. Spare respiratory capacity, calculated as the difference between maximal and basal respiration, was significantly decreased in Pfn1 cKO synaptosomes (Fig. 6E). This reduction implies a limited ability of Pfn1 cKO mitochondria to respond to additional energetic stress, rendering synapses more vulnerable under conditions of increased metabolic demand. ATP production coupled respiration was also significantly impaired in Pfn1 cKO synaptosomes (Fig. 6F). This aligns with the observed reductions in both basal and maximal respiration, reinforcing the conclusion that oxidative phosphorylation is compromised in the absence of microglial Pfn1, thereby diminishing the ATP generation necessary for proper synaptic function. Finally, coupling efficiency was significantly decreased in Pfn1 cKO synaptosomes compared to controls (Fig. 6G).

Coupling efficiency represents the proportion of mitochondrial oxygen consumption used for ATP synthesis relative to total respiration. A decrease in coupling efficiency indicates that a smaller fraction of the consumed oxygen is effectively utilized for ATP production in the Pfn1 cKO synaptosomes. This reinforces the overall inefficiency of the oxidative phosphorylation process. The reduction in coupling efficiency and diminished ATP production and proton leak imply that mitochondrial dysfunction in Pfn1 cKO synaptosomes involves impaired electron transport chain activity and compromised ATP synthesis efficiency. Collectively, our data suggest that microglial Pfn1 deficiency induces synaptic alterations by promoting the release of SASP factors and creating a pro-inflammatory environment that adversely reconfigures the synaptic energetic landscape, which may underlie synaptic activity impairments and behavioral deficits associated with aging.

Loss of Pfn1 in microglia leads to deficits in GABAergic transmission

Thus far, our analyses revealed significant alterations in mitochondrial function and energy metabolism—all crucial for synapse functioning. GABAergic synapses are particularly sensitive to disruptions in these pathways [50]. For instance, efficient GABA release relies on robust mitochondrial function [51, 52], calcium signaling [53], and the cytoskeleton [54], all of which are impacted by the molecular changes observed in our results.

Furthermore, existing literature underscores the vulnerability of GABAergic synapses to metabolic and proteostatic stress. Studies have demonstrated that impaired mitochondrial function and disrupted calcium homeostasis can disproportionately affect inhibitory synapses, reducing GABAergic tone and subsequent neural circuit dysfunction [55]. Thus, we sought to elucidate whether microglial Pfn1 deficiency compromises inhibitory neurotransmission, thereby contributing to the broader synaptic aging phenotype observed in Pfn1 cKO mice.

We specifically examined alterations within GABAergic synapses to further elucidate the impact of microglial Pfn1 deficiency on synaptic function. Our phosphoproteomics analysis identified significant phosphorylation changes in proteins composing two major GABAergic clusters: one related to GABAergic synapse composition and the other to Inhibitory Postsynaptic Currents (IPSC) (Fig. 7A and B).

Fig. 7.

Loss of Pfn1 disrupts inhibitory synaptic transmission in the prefrontal cortex. A A histogram and chart showing the phosphorylation status of proteins involved in GABAergic synapse composition. B A histogram and chart showing the phosphorylation status of proteins associated with inhibitory postsynaptic currents. C Representative traces of miniature inhibitory postsynaptic currents (mIPSCs) recorded from PFC pyramidal neurons in Pfn1 cKO and CT mice (n = 6 animals/genotype, 1–6 cells/animal). D Representative traces of miniature excitatory postsynaptic currents (mEPSCs) recorded under the same conditions. E and F Graphs of the mIPSC frequency and amplitude from PFC pyramidal neurons in Pfn1 cKO and CT mice. G and H Graphs of the mEPSC frequency and amplitude from PFC pyramidal neurons in Pfn1 cKO and CT mice. Data are presented as mean ± SD. Statistical comparisons were performed using Linear mixed models: **p < 0.01

Several key proteins exhibited altered phosphorylation states critical for GABAergic signaling within the GABAergic Synapse cluster (Fig. 7A). Git1 (GIT ArfGAP 1), essential for regulating synaptic vesicle recycling, showed two different phosphorylation alterations. These modifications potentially impair the cytoskeleton-dependent recycling of synaptic vesicles necessary for consistent GABA release, leading to diminished inhibitory tone and reduced synaptic strength (Fig. 7A). Bsn (Bassoon), a structural scaffolding protein localized at the presynaptic active zone, exhibited multiple phosphorylation alterations, suggesting compromised synaptic architecture and disrupted neurotransmitter release mechanisms (Fig. 7A). Similarly, Pclo (Piccolo), which works alongside Bsn to control synaptic vesicle cycling, showed significant multiple phosphorylation changes, further indicating deficits in vesicle availability and neurotransmission efficiency (Fig. 7A). Pak1 (p21-activated kinase 1), involved in cytoskeletal dynamics, displayed reduced phosphorylation, which may lead to impaired actin remodeling and synaptic stability (Fig. 7A). This cytoskeletal disruption can affect the structural maintenance of GABAergic synapses, compromising their ability to sustain inhibitory neurotransmission. Gphn (Gephyrin), a scaffolding protein critical for clustering GABA_A receptors at the postsynaptic membrane, was found to have substantially increased phosphorylation, potentially leading to a disarray in receptor availability and impaired inhibitory synaptic transmission (Fig. 7A). Additionally, Slc6a17, a sodium- and chloride-dependent transporter involved in neurotransmitter regulation, exhibited decreased phosphorylations, which may disrupt the balance of neurotransmitter precursors necessary for GABA synthesis and release (Fig. 7A).

In the Inhibitory Postsynaptic Currents (IPSC) cluster (Fig. 7B), key proteins involved in GABA synthesis and synaptic vesicle dynamics were significantly affected by phosphorylation changes. Gad1 (Glutamate Decarboxylase 1), the key enzyme responsible for converting glutamate to GABA, showed decreased phosphorylation, indicating a potential reduction in GABA production and subsequent inhibitory neurotransmission (Fig. 7B). Dagla (Diacylglycerol Lipase Alpha), involved in the production of the retrograde signaling molecule 2-arachidonoylglycerol (2-AG), exhibited a considerable phosphorylation increase, which may impair feedback inhibition mechanisms regulating GABA release (Fig. 7B). Syn1 (Synapsin 1), which regulates synaptic vesicle docking and availability, had multiple phosphorylation alterations, suggesting reduced vesicle tethering and increased synaptic fatigue (Fig. 7B). Syt1 (Synaptotagmin 1), a calcium sensor crucial for synaptic vesicle fusion, showed decreased phosphorylation, potentially leading to impaired timing and efficiency of GABA release in response to calcium influx (Fig. 7B). Apba1 (Amyloid Beta A4 Precursor Protein-Binding Family A Member 1), known for its role in modulating amyloid precursor protein processing, was increased in its phosphorylated state, which may affect synaptic vesicle dynamics and neurotransmitter regulation in GABAergic synapses (Fig. 7B). Lastly, Prrt2 (Proline-Rich Transmembrane Protein 2), associated with synaptic transmission and membrane dynamics, exhibited decreased phosphorylation, indicating potential disruptions in synaptic vesicle cycling and membrane fusion processes essential for effective inhibitory neurotransmission (Fig. 7B).

These phosphorylation alterations collectively suggest a profound impairment in inhibitory synaptic function in Pfn1 cKO mice. Disruptions in synaptic vesicle recycling, neurotransmitter synthesis, receptor clustering, and calcium-dependent vesicle fusion may contribute to deficits in GABAergic tone and synaptic efficacy. To test the hypothesis that loss of Pfn1 in microglia disrupts inhibitory synaptic activity, we performed whole-cell patch clamp in pyramidal neurons from the prefrontal cortex (PFC) of Pfn1 cKO and CT mice 6–8 weeks after TAM induction to assess GABA-mediated miniature inhibitory postsynaptic currents (mIPSC). We observed a decrease in the frequency of mIPSC (Fig. 7C, E) but no change in mIPSC amplitude (Fig. 7C, F) in Pfn1 cKO. To confirm that Pfn1 microglial ablation did not change excitatory synapses, we also performed AMPA-mediated miniature excitatory postsynaptic currents (mEPSC). Interestingly, no differences in either frequency or amplitude were observed in these recordings (Fig. 7D, G, and H). In line with the phosphoproteomics data, these results suggest that inhibitory synaptic transmission is decreased in putative excitatory neurons of the PFC and highlight the critical role of microglial Pfn1 in maintaining inhibitory synaptic homeostasis and overall neural circuit integrity.

Inhibitory neurotransmission impairments underlie altered anxiety-like and risk-assessment behaviors in Pfn1 cKO mice

Considering the significant disruptions in PFC inhibitory neurotransmission, we hypothesized that Pfn1 deficiency in microglia would selectively impair behaviors regulated by this circuitry, such as anxiety and risk assessment, while sparing cognitive domains primarily dependent on other regions, like the hippocampus. To test this, we subjected Pfn1 cKO and CT mice (6–8 weeks post-TAM induction) to a battery of behavioral assays.

First, to assess hippocampus-dependent cognitive function, we performed the Morris Water Maze (MWM) test, which evaluates spatial learning and memory. Pfn1 cKO mice exhibited learning curves and probe trial performance comparable to those of CT mice, indicating that spatial memory is intact (Fig. 8A, B). To further probe cognition, we conducted a Novel Object Recognition (NOR) test. Consistent with the MWM results, Pfn1 cKO mice showed a similar preference for the novel object as controls, demonstrating that recognition memory is also preserved (Fig. 8C).

Fig. 8.

Behavioral characterization of Pfn1 cKO mice. A and B Morris water maze (MWM) test (n = 6–9 mice/genotype), displaying escape latencies during the training phase and performance during the probe trial. C Novel object recognition (NOR) test showing the discrimination index (n = 6–7 per group). D Light–dark box (LDB) test (n = 6–7 mice/genotype), displaying the number of transitions between the light and dark compartments. E Open-field (OF) test (n = 7 mice/genotype), displaying the time spent in the center and periphery of the arena and the total distance each mouse traveled in the arena. F Elevated plus maze (EPM) test (n = 8–10 mice/genotype), displaying the time spent in the open arms, the total distance traveled in the maze, the number of head-dips over the open arms, and the number of stretch-attend postures (SAP). *Data are presented as mean ± SEM. Statistical analyses were performed using two-way ANOVA in A and Student’s t-tests in B-F: *p < 0.05. The behavioral test schematics were created with BioRender and are for representative purposes only

In stark contrast, behaviors related to anxiety and risk assessment were significantly altered. In the light–dark box (LDB) and open field (OF) tests, Pfn1 cKO mice spent significantly more time in the more anxiogenic zones (light compartment and center of the arena, respectively) and showed increased transitions in the LDB, suggesting a lower aversion to brightly lit, open spaces (Fig. 8D, E). Similarly, in the elevated plus maze (EPM), cKO mice spent more time in the open arms, performed more exploratory head-dips, and showed fewer stretch-attend postures (Fig. 8F). Importantly, total distance traveled was comparable between genotypes across all relevant tests, ruling out hyperactivity or gross motor impairments as confounding factors (Fig. 8E and F).

Altogether, these data suggest that the loss of microglial Pfn1 leads to changes in behavior consistent with altered anxiety and risk assessment, aligning with the selective disruption of inhibitory circuits we identified in the PFC.

Discussion

Aging drives a gradual decline in cellular function and increases susceptibility to neurodegenerative disease [56]. In the CNS, microglia undergo profound aging-related changes in morphology and function [57], which fuel chronic neuroinflammation and neuronal impairment. Our findings position Pfn1 as a central mechanistic regulator of microglial aging, as its loss alone is sufficient to trigger microglial senescence, disrupt inhibitory–excitatory synaptic balance, and cause behavioral deficits. Importantly, microglial Pfn1 deletion led to selective dysfunction of inhibitory synapses in the PFC, while excitatory synapses were largely spared.

Galatro et al. (2017) reported that Pfn1 transcript levels are significantly reduced in aged human microglia [7], suggesting that Pfn1 loss might underlie microglial dysfunction during aging. We hypothesized that Pfn1 is crucial for maintaining microglial homeostasis and that its decline actively drives the aging process. Indeed, our data confirm this premise and establish a causal cascade where decreased Pfn1 expression leads to cytoskeletal collapse, the emergence of a SASP signature, and ultimately, inhibitory synapse dysfunction (Fig. 9).

Fig. 9.

Microglial Profilin-1 loss collapses the cytoskeleton, elicits a SASP, and depresses inhibitory synapse activity. The left-to-right schematic shows: steady-state microglia with an intact cytoskeleton → loss of Pfn1 → disruption of the cytoskeleton → emergence of a SASP with pro-inflammatory cytokine release and extracellular matrix remodeling → reduced inhibitory synapse activity. The inhibitory terminal is marked with Bassoon (Bsn) and Piccolo (Pclo) to indicate presynaptic changes. Representative mIPSC traces display a reduction in event frequency, with amplitude mostly unchanged. Arrows indicate causal flow. Color and shape coding differentiate microglia (cell silhouettes, left), SASP/ECM processes (starburst and fibrillar icons, center), and the inhibitory synapse module (bouton–postsynaptic contact, right)

The cytoskeleton, composed of actin filaments and microtubules, underpins key microglial functions such as motility, environmental scanning, and phagocytosis [57]. Continuous actin remodeling allows microglia to extend and retract processes for effective surveillance and injury responses [29, 58]. Pfn1, a key regulator of actin polymerization, ensures the availability of actin monomers for filament elongation and proper cytoskeletal organization [9, 59]. By removing Pfn1, we found that microglia exhibit markedly reduced process motility and blunted injury responses, reflecting impaired actin dynamics. These deficits mirror hallmarks of aged microglia, such as shorter, less branched protrusions and diminished surveillance capacity [2–4], indicating that Pfn1 is required to prevent these aging-like morphodynamic changes. Our findings align with reports of age-related microglial cytoskeletal decline. Aged microglia in vivo show significantly slower process motility and responses to injury, correlating with reduced levels of actin-regulatory proteins [24] and exhibiting simplified, less-branched morphologies [28]. The ability of Pfn1 loss to recapitulate these age-related cytoskeletal deficits suggests that Pfn1 is essential for preserving microglial morphodynamic plasticity during aging.

In addition to actin maintenance, Pfn1 is essential for coordinating actin–microtubule crosstalk, critical for intracellular transport and cellular organization [9, 59]. Microtubules provide tracks for moving vesicles and organelles to distant microglial processes during surveillance [60, 61]. Pfn1 loss likely destabilizes microtubules and impairs this transport, compounding microglial dysfunction. Additionally, without Pfn1’s interactions with formins and Ena/VASP proteins to nucleate and elongate actin filaments [9, 59], microglia may fail to generate the specialized cytoskeletal structures needed to adapt to a dynamic environment. In this way, Pfn1 serves as a linchpin maintaining the integrative architecture of the cytoskeleton, and its absence broadly deregulates cytoskeletal dynamics. Cytoskeletal remodeling in microglia is also governed by Rho family GTPases (Rac1 and RhoA), which drive lamellipodia and filopodia formation [57]. Pfn1 likely operates in concert with these signaling pathways to ensure proper protrusive activity. Perturbations in Rac1 or RhoA alone can impair microglial motility and morphology [21, 22], and the blunted protrusive activity in Pfn1 cKO microglia suggests that Pfn1 may act as a key node integrating Rho GTPase signals with the actin machinery.

Strikingly, loss of Pfn1 triggered a SASP in microglia, as revealed by our transcriptomic analyses. Pfn1-deficient microglia upregulated numerous SASP-related genes encoding pro-inflammatory cytokines (IL-1β, TNF-α), chemokines, growth factors, and matrix metalloproteinases like MMP9 [5]. The induction of this pro-inflammatory secretome promotes chronic neuroinflammation and tissue damage, compounding the detrimental effects of microglial dysfunction on the aging brain [62]. Consistent with this, Pfn1-deficient microglia exhibited substantial MAPK/ERK pathway alteration, a signaling cascade implicated in both senescence and inflammation [35]. ERK1/2 activation can enhance SASP factor transcription through downstream effectors like NF-κB and AP-1 [5, 36]. Thus, we propose that Pfn1 loss, by destabilizing cytoskeletal–integrin connections, abnormally amplifies ERK signaling [63], reinforcing the pro-inflammatory, senescent microglia phenotype. This feedback loop directly links the initial cytoskeletal disruption to a SASP-related inflammatory signaling.

It is essential to contextualize our in vivo motility impairment within the framework of the Pfn1 phenotype. While the intravital LI paradigm is a powerful tool to assess a reactive response, it is a simplified model. Therefore, we interpret our in vivo motility data not as a single event, but as two facets of a concatenated phenotype: a failure in homeostatic surveillance under baseline conditions, and a failure in reactive chemotaxis following an acute stimulus. We propose that this global state of morphofunctional impairment represents a major cellular stressor. This dysfunctional state, rather than altered synaptic pruning or microglia-synapse contact, is a more plausible trigger for the cell-autonomous stress response and subsequent SASP activation that we identified through our transcriptomic analysis. This framework suggests a direct mechanistic link, where the primary cytoskeletal collapse leads to a pathogenic secretory state, and the soluble factors of the SASP (e.g., IL-1β, TNF-α, MMP9) then act non-cell-autonomously on the surrounding neural environment to cause the mitochondrial and synaptic deficits we observed.

The selective vulnerability of inhibitory circuits in the PFC warrants further consideration, and our data suggest a multi-layered mechanism targeting fast-spiking, parvalbumin-positive (PV +) interneurons. These cells are characterized by a high metabolic burden required to sustain high-frequency firing [64], likely rendering them intrinsically susceptible to the mitochondrial energy deficits we identified. We propose that the microglial SASP then launches a direct, 'two-hit assault' on these already-stressed neurons. First, the upregulation of pro-inflammatory cytokines, such as IL-1β and TNF-α, core components of SASP, can further exacerbate the metabolic impairments by directly affecting neuronal mitochondrial function. In parallel, the significant increase in MMP9 expression executes a direct structural attack by degrading the perineuronal nets (PNNs), the specialized extracellular matrix essential for the stability and synaptic integrity of PV + neurons [65].

Compounding this targeted assault on the neuron is the broader environmental degradation, also mediated by MMP9. By degrading key tight junction proteins, MMP9 is a potent driver of blood–brain barrier (BBB) dysfunction [66]. This breach in the neurovascular unit allows for the infiltration of neurotoxic, serum-derived proteins, such as albumin and fibrinogen, into the brain parenchyma, which can amplify the initial inflammatory insult and trigger further synaptotoxicity. Thus, the selective failure of inhibitory neurotransmission appears to be the result of a concerted multi-layered process in which a direct SASP-mediated 'two-hit assault exploits an intrinsic neuronal vulnerability', which is then amplified by a broader, MMP9-driven degradation of the neurovascular environment.

Our data showed a clear presynaptic deficit (reduced mIPSC frequency) co-occurring with phosphorylation changes in postsynaptic scaffolding proteins (e.g., Gephyrin). We argue this reflects a sequential, cause-and-effect relationship. The primary insult is presynaptic. The reduced mIPSC frequency strongly indicates fewer spontaneous vesicle fusion events. This aligns with our phosphoproteomic data, which show altered phosphorylation of key presynaptic proteins involved in vesicle dynamics (Syn1, Syt1, Bsn, Pclo) and GABA synthesis (Gad1), as well as the mitochondrial energy deficits that would impair vesicle recycling. In response to this presynaptic reduction in GABAergic input, the postsynaptic terminal likely initiates homeostatic compensatory mechanisms aimed at preserving synaptic strength. The increased phosphorylation of Gephyrin could be one such mechanism, attempting to stabilize or increase the number of GABAA receptors at the remaining functional synapses to maximize the response to the limited GABA released. This compensatory effort appears partially successful, as mIPSC amplitude is maintained. However, this compensation is ultimately insufficient to overcome the drastic reduction in presynaptic drive, resulting in a net decrease in inhibitory tone at the circuit level, which manifests as the observed behavioral phenotype.

The selective disruption of inhibitory synapses in the PFC has significant implications for behavior. GABAergic signaling in this region is essential for inhibitory control and decision-making [67–69], and the reduced inhibitory tone we observed likely underlies the altered anxiety and risk-assessment behaviors in Pfn1 cKO mice. This circuit-specific vulnerability is particularly striking given that hippocampus-dependent spatial and recognition memory remained intact.

Aging is often described as anxiogenic, but empirical data supporting this claim are inconsistent. In rodents, studies report decreased [70], unchanged [71], or task- and time-dependent changes in anxiety-like behavior [72–74]. In humans, accelerated biological aging can increase anxiety risk ([75], yet trait anxiety often decreases with age and exhibits distinct neural patterns [76], while psychosocial and health factors substantially influence prevalence [77–79]. Our data are therefore more in line with, rather than contradictory to, this diverse body of research on aging-related anxiety.

Furthermore, microglial Pfn1 loss leads to a specific prefrontal inhibitory deficit (decreased mIPSC frequency with preserved excitatory measures), aligning with models where inhibitory populations are more vulnerable with aging [80]. Behaviorally, increased open/center/light exploration at matched locomotion levels, accompanied by sustained hippocampus-dependent memory, might suggest a PFC-weighted disinhibition rather than a strict overall reduction in anxiety (Figs. 7, 8 and 9). Thus, our findings allow us to speculate that an early phase of PFC dominance (weeks after induction) leads to an approach bias in conflict tasks, followed by a potential later engagement of limbic circuits that may shift behavior toward avoidance. The circadian/time-window sensitivity observed in aged/APP mice may support this temporal aspect [72].

It is essential to interpret these findings within the framework of our experimental model. We deliberately used an inducible microglia-specific cKO system to establish direct causality—an aim that traditional aging models cannot satisfy. This approach demonstrates that the loss of Pfn1, a molecular event observed in aged human microglia [7], is sufficient to elicit a coordinated phenotype encompassing senescence, synaptic dysfunction, and behavioral change; therefore, we interpret these data as a test of causal sufficiency rather than an observational model of aging. Additionally, our cKO model does not replicate the slow, multifactorial progression of natural aging. Moreover, the robust decline in Pfn1 expression in aged human microglia is not consistently observed in aged laboratory mice, likely reflecting species-specific differences in lifespan and immune-senescence dynamics. Consequently, our cKO model should be viewed as a mechanism-discovery tool that isolates the consequences of Pfn1 insufficiency, not as a comprehensive replica of the aged brain.

Because Pfn1 does not consistently decrease with age in mouse microglia, a microglia-targeted “rescue” (i.e., restoring Pfn1 levels) in aged mice is not logically justified: there is no species-specific deficit to restore. Therefore, we do not frame a mouse “rescue” as an informative test of our central claims. Overexpressing Pfn1 in aged mouse microglia would instead constitute a gain-of-function, pro-resilience experiment. While interesting scientifically and valuable for therapeutic development, such a study would require dose-controlled microglial enhancement of Pfn1 in aged cohorts with multiple endpoints and extensive tool validation; accordingly, this approach lies outside the scope of the present work.

A key finding of our study is the apparent circuit-specific nature of the dysfunction. While we observed robust synaptic and behavioral deficits linked to the PFC, hippocampus-dependent cognitive functions remained intact. This raises questions about the mechanisms behind this apparent hippocampal resilience. One possibility is a difference in regional sensitivity to the microglial SASP, where inhibitory interneurons in the hippocampus might be less vulnerable to the specific inflammatory mediators released by Pfn1-deficient microglia compared to their PFC counterparts. Alternatively, the hippocampus may have more effective homeostatic or compensatory mechanisms at the network level that buffer the impact of decreased inhibitory tone, thereby maintaining memory function. Understanding why some circuits remain stable while others falter is crucial to comprehending how the brain resists age-related damage. Future studies using longitudinal designs in naturally aging rodents or accelerated aging models (such as the SAMP8 line) will be essential to understand the effects of a gradual Pfn1 decline and how it interacts with other age-related stressors. Similarly, electrophysiological recordings from other brain regions, such as the hippocampus, will be needed to map the full range of regional vulnerabilities and/or compensation in the Pfn1 cKO brain.

In conclusion, our data establish Pfn1 as a critical molecular lever maintaining microglial cytoskeletal integrity, whose loss precipitates microglial senescence and features of premature synaptic aging. The preferential vulnerability of PFC inhibitory synapses in Pfn1-deficient mice underscores how microglial cytoskeletal dysfunction can erode the excitatory–inhibitory balance in cortical circuits, highlighting the importance of sustaining microglial cytoskeletal dynamics for cognitive resilience. In this context, targeting Pfn1-dependent cytoskeletal pathways in microglia could be a viable strategy for mitigating age-related brain changes and fostering resilience against neurodegenerative processes.

Material and methods

Key resources

Supplementary Table 5 provides a list of key reagents, antibodies, mouse strains, equipment, and software, including their sources, identifiers, and Research Resource Identifiers (RRIDs).

Experimental models

Animals

Mice used in this work were bred at the i3S animal facility. Mice were housed under specific pathogen-free conditions in standard laboratory conditions with a light/dark cycle of 12 h, 20ºC, 45–55% humidity, and access to water and food ad libitum. All procedures were conducted according to the European Union guidelines for animal welfare (European Union Council Directive 2010/63/EU) and Portuguese law (DL 113/2013) under Direção-Geral de Alimentação e Veterinária (license 2022–02–18 003669), and all procedures meet ARRIVE 2.0 guidelines. All experimental cohorts were randomly assigned to treatment or genotype groups using block randomization balanced by litter and cage origin to minimize potential genetic or environmental bias. Experimenters were blinded to genotype during all data acquisition and analysis. Mice were monitored daily for health and welfare. Humane endpoints followed institutional veterinary guidelines: any animal showing >15% weight loss, persistent piloerection, abnormal posture, or impaired ambulation. No animals reached these endpoints in the present study. Cohort sizes were specified a priori from historical variance in equivalent assays in our laboratory and practical throughput limits per modality. For multi-level designs, we used linear mixed effects models with mouse as a random effect to avoid inflating n by technical/within-mouse replicates.

Conditional Pfn1-deficient mice

Cx3cr1^CreER−EYFP mice were purchased from Jackson Laboratories and used as before [22]. In such mice, the Cx3cr1 promoter drives high expression of the CreER cassette in microglia [19]. Mice homozygous for the Pfn1 floxed allele [81] were backcrossed for at least ten generations and were kept at the I3S animal facility. PCR determined all genotypes on genomic DNA. Pfn1 floxed mice were crossed with Cx3cr1^CreER−EYFP mice. Progeny of interest were: Controls (Pfn1^fl/fl or Cx3cr1^CreER+) and mutants (Pfn1^fl/fl:Cx3cr1^CreER+). All mice (4–5 weeks old) were given tamoxifen as before [21, 22] and analyzed 6–8 weeks after tamoxifen administration.

Experimental procedures

Intravital two-photon laser-scanning microscopy (2P-LSM) and analysis of microglia protrusion dynamics

Control and mutant mice were subjected to a three mm-diameter cortical craniotomy (lateral = 1.5 mm and longitudinal = 2 mm from bregma) as described previously [82, 83]. A 3 mm coverslip was placed on the brain and fixed with dental cement (RelyX®, 3M-ESPE, Neuss, Germany). Immediately, mice were placed under the 2P-LSM microscope, and an area with 256 µm × 256 µm was recorded for two hours with a 950 nm laser beam. Images were recorded at a thickness of 20–30 µm with 2 µm intervals, using two images per layer, every 2 min for a period of 2 h. After baseline recording, a laser injury was performed in the center of the region of interest (ROI) by parking the laser beam (800 nm) at a single point for 250 ms, followed by another two hours of imaging in the same ROI with the same settings.

For single-cell and protrusion quantitative dynamics analysis, data were extracted from the i-th frame, where r[i] = {x[i], y[i]} represents the Cartesian coordinates of specific points: rLI[i], the laser injury center; rCC[i], the cell center; and rPr[k][i], the k-th protrusion endpoint. Euclidean distances (D(p1, p2)) between these points were calculated for each frame. Specifically, D(rLI[i], rCC[i]) was used to measure the distance between the laser injury and the cell center, D(rLI[i], rPr[k][i]) for the distance between the laser injury and the k-th protrusion endpoint, and D(rCC[i], rPr[k][i]) for the distance between the cell center and the k-th protrusion endpoint. In the absence of a laser injury, only D(rCC[i], rPr[k][i]) was calculated. Automated analyses, including distance measurements and protrusion velocity computations, were performed in Mathematica using custom-written code that utilized the initial coordinate data.

Protrusion velocities were computed using two distinct approaches. The first method calculated the global velocity (VGlobal) by dividing the displacement of the protrusion—defined as the distance between its initial and final points—by the total trajectory duration, yielding an average displacement velocity. The second method calculated the instantaneous velocity (VInst) by determining the distance between two adjacent points along the protrusion trajectory and dividing it by the acquisition time step. The average instantaneous velocity was then obtained by averaging these values over the entire trajectory. Additionally, the trajectory amplitude, defined as the maximum distance between any two points along the trajectory, was computed.

Prior to statistical analysis of instantaneous velocity data, statistical outliers were identified and removed using the ROUT method (Q = 1%) to account for disproportionate, aberrant measurements that may represent tracking artifacts rather than typical process motility in both CT and Pfn1 cKO data. This pre-processing step resulted in the removal of 15 out of 204 CT datapoints and 62 out of 558 cKO datapoints. All subsequent analyses were performed on the cleaned dataset.

To account for the hierarchical structure of the data (multiple protrusions nested within each mouse), statistical comparisons of protrusion amplitude and velocity were performed using a linear mixed-effects model (LMM) implemented in Python with the statsmodels library. In all analyses, genotype was treated as a fixed effect, and mouse_id was included as a random effect to account for within-subject correlations.

Tissue preparation and immunofluorescence

Mice were perfused transcardially with ice-cold PBS, and brains were fixed by immersion in 4% paraformaldehyde (PFA) in PBS (pH 7.2) overnight at 4°C. Fixed brains were rinsed in PBS and cryoprotected in sucrose using a two-step gradient (15% followed by 30% w/v) until they sank. After at least 24 hours in sucrose, brains were positioned in the cryostat with the cerebellum stabilized against the mounting block, and coronal sections (30 μm thick) were cut in an anterior-to-posterior sequence—from the olfactory bulb to the cerebellum—perpendicular to the rostrocaudal axis, ensuring that each section encompassed both hemispheres. All sections used for microglia-related analyses were collected in a free-floating configuration. Immunofluorescence staining was performed on free-floating sections under gentle agitation. Matched cortical regions from control and experimental mice were processed in parallel to ensure identical conditions. Sections were permeabilized in 0.25% Triton X-100 for 15 minutes, rinsed in PBS for 10 minutes, and blocked for 1 hour in a solution containing 5% BSA, 5% FBS, and 0.1% Triton X-100. Primary antibodies were incubated for 48 hours at 4°C in blocking buffer within a humidified chamber. Then, sections were washed three times in PBS and incubated with appropriate secondary antibodies for 2 hours at room temperature in blocking buffer. After final PBS washes (3 × 10 minutes), sections were mounted onto glass slides and coverslipped using anti-fading mounting media.

Confocal imaging and morphometric analysis

As before [21], images from cortical tissue sections were acquired in 8-bit sequential mode using a Leica SP8 confocal microscope in standard TCS or resonant mode at. The pinhole was kept at one airy in the Leica TCS SP8 confocal microscope. Images were illuminated with different laser combinations and resolved using HyD detectors at 512 × 512 or 1024 × 1024 pixels format. Entire Z-series were acquired from tissue sections with a step size of 0.3-0.5 µm. For each slide, equivalent Z-series and matched regions were obtained across all tissue sections.

IMARIS Analysis

All smoothing values and rolling ball radii were empirically determined and applied to every image. All threshold values were determined specifically for each image as before [21].

Microglia Morphological Analysis: The morphological analysis of microglia was conducted using the filaments function of the IMARIS software (version 10.1.0). Initially, all images were uploaded and pre-processed using a background subtraction filter with a rolling ball radius of 10 µm. An exaggerated surface object was subsequently created from the Ionized calcium-binding adapter molecule 1 (Iba1) channel, ensuring the inclusion of every process, using the default smoothing of 0.283 µm. This surface was utilized to mask the Iba1 channel, which facilitated the creation of the filaments object. During the creation of the filaments object, the soma diameter was adjusted to each cell and reconstructed when possible. The seed point threshold was determined for every image, and the seed points were manually classified. The segment classification option was deactivated, and no filter was applied to the segments at the end. Finally, after the filament object creation, every segment was individually verified, and any missing or incorrect segments were manually corrected and redrawn using either the auto-depth or auto-path drawing mode.

CD68: For each cell, CD68 was masked using the same microglia surface as above, and vesicles were reconstructed using the Surface object with a surface smoothing of 0.120 µm. The surfaces were filtered by volume to exclude any that were under 0.01 µm^3. After reconstruction, the vesicles were checked to ensure they were all inside the microglia being reconstructed.

Behavioral tests

All behavioral procedures were conducted during the dark phase of the light/dark cycle and were performed by an experimenter blind to the genotypes of the mice. The number of animals per group is reported in the respective figure legends. MWM: The MWM was used to evaluate spatial memory. The apparatus consisted of a circular pool (110 cm diameter) of fiberglass filled with water (21 ± 1ºC). This test is divided into three consecutive steps. Firstly, the cued training was conducted for two days with four trials per animal per day.

MWM: The MWM was used to evaluate spatial memory. The apparatus consisted of a circular pool (110 cm diameter) of fiberglass filled with water (21 ± 1ºC). This test is divided into three consecutive steps. Firstly, the cued training was conducted for two days with four trials per animal per day. A non-visible escape platform (7 × 8 cm) was submerged 1 cm below the water surface in the quadrant center. Mice were trained to find the platform with a visual cue. Animals were subjected to four swimming trials with different start and goal positions. The training phase lasted seven days, with four trials per animal per day and a completion time of 1 min. The visual cue on top of the platform was removed, and other cues were placed on the room walls to allow identifying the platform position. At this stage, mice were released from different positions, while the platform’s location remained unaltered. Again, if the mice could not reach the platform in 1 min, they were guided to it. Mice were allowed 10 s on top of the platform for each trial, allowing them to learn its location. The test day occurred 24 h after the last training. At this stage, the platform was removed, challenging the mice to recall where the platform was supposed to be. All mice were released from the same position (opposite quadrant) and allowed to freely swim around the apparatus for 30 s to search for the platform. All parameters were automatically evaluated by SMART v3.0 software (Panlab, Barcelona, Spain).

NOR: The NOR test was conducted in a square open-field arena (40 × 40 × 40 cm) made of opaque gray plastic. The test was divided into three parts. On the first day, the procedure started with a 10 min habituation session, during which each mouse was allowed to freely explore the empty arena. Twenty-four hours later, during the familiarization phase, two identical objects were placed side by side, equidistant from the corners of the arena, and each mouse was allowed to explore them for 10 min. Following a 4 h inter-trial interval, one of the familiar objects was replaced with a novel object of similar size but distinct shape and texture for a 3 min test phase. The placement of the novel object was counterbalanced across animals to prevent location bias. Exploration was defined as the mouse touching/directing its nose toward an object at a distance shorter than 2 cm. The time spent exploring each object was recorded automatically using a video-tracking system (SMART v3.0, Panlab). The arena and objects were thoroughly cleaned with a neutral soap solution between each trial to eliminate olfactory cues. A discrimination index (a metric for recognition memory) was calculated as (Time exploring novel object—Time exploring familiar object) (Total exploration time).

DLB: A DLB test was conducted to evaluate anxiety-related behaviors, risk-assessment and exploratory drive in Pfn1-deficient and control mice, leveraging their innate aversion to brightly illuminated spaces and the competing drive for exploration. The apparatus consisted of a rectangular chamber measuring 45 × 25 × 25 cm, divided into two distinct compartments: a dark, enclosed area covering one-third of the total space, designed to minimize light penetration, and a light compartment occupying the remaining two-thirds, illuminated at an intensity of 1000 lx. The two chambers were connected by a small 5 × 5 cm opening at floor level, allowing the mice to transition freely between them. The light compartment was constructed with white plastic walls to maximize brightness, whereas the dark chamber was designed with black opaque walls to create an environment of relative darkness, enhancing the contrast between the two spaces.

To ensure experimental reliability, all mice underwent a habituation period of 30 min in the testing room prior to the assessment. Each mouse was carefully introduced into the dark compartment, facing the back wall, and was allowed to explore the apparatus freely for a total of 10 min. Behavioral parameters were recorded using an automated tracking system (SMART v3.0, Panlab) and included the latency to first transition, defined as the time taken for the mouse to enter the light compartment for the first time; the number of transitions, representing the total number of crossings between the two chambers; the percentage of total time spent in the illuminated compartment, which served as an index of reduced anxiety-like behavior; and the total distance traveled within the light compartment, providing insight into locomotor activity and exploratory tendencies.

OF: OF test was conducted to assess general locomotor activity, exploratory behavior, and anxiety-like responses in Pfn1-deficient and control mice, leveraging their natural tendency to explore novel environments while avoiding open, unprotected spaces. The apparatus consisted of a square arena measuring 40 × 40 × 40 cm, constructed from opaque gray polyvinyl to reduce potential reflections and unintended visual cues. The arena was uniformly illuminated to create a distinction between the peripheral and center zones, with the periphery providing a more enclosed and reassuring environment and the center representing a more exposed, potentially aversive space.

Each mouse was individually placed in the center of the arena and allowed to explore freely for a total duration of 10 min. Behavioral tracking and locomotor parameters were recorded using an automated video tracking system (SMART v3.0, Panlab), ensuring objective and high-precision movement detection. The primary parameters analyzed included the total distance traveled, serving as a measure of overall locomotor activity; the time spent in the center of the arena, considered an inverse indicator of anxiety-like behavior; and the number of entries into the center zone, reflecting exploratory drive and risk assessment. Mice exhibiting increased time in the center and a higher frequency of center entries were considered to display reduced anxiety-like behavior, whereas a preference for the periphery was interpreted as heightened anxiety and an avoidance phenotype.

EPM: The EPM test was conducted to evaluate anxiety-like behavior and risk assessment in Pfn1-deficient and control mice, leveraging their innate aversion to open, elevated spaces and their drive to explore novel environments. The test was performed during the dark phase of the light–dark cycle to maximize behavioral consistency across experimental subjects. The EPM apparatus was constructed from opaque gray polyvinyl and consisted of a cross-shaped platform with four arms, each measuring 37 × 6 cm. Two opposing arms were enclosed by surrounding walls 18 cm in height, while the remaining two arms were open, lacking any protective barriers. The entire structure was elevated 50 cm above the floor to create an environment that induces approach-avoidance conflict, a key parameter in anxiety assessment.

Each mouse was placed on the central platform of the maze, facing an open arm, and was allowed to explore freely for a total duration of 5 min. Behavioral parameters were recorded using an automated video tracking system (SMART v3.0, Panlab ) equipped with an infrared-sensitive camera to ensure accurate movement detection. The primary parameters analyzed included the time spent in the open arms, the number of entries into the open arms, and the total distance traveled within the maze. The percentage of time spent in the open arms was considered an inverse measure of anxiety-like behavior, with a greater proportion of time reflecting reduced anxiety levels and a diminished perception of risk. The total number of entries into the open arms was analyzed as an additional measure of risk-taking behavior, while overall locomotor activity was assessed to ensure that observed differences were not attributable to motor impairments.

Flow cytometry and cell sorting

Microglia were collected from the brains and other tissues of control and mutant mice using density gradient separation as before [21, 84]. Single-cell suspensions (5 × 10⁵ cells) were incubated with antibodies for 30 min at 4 °C in the dark. Compensation settings were determined using the spleen from both the control and mutants. Cell sorting was performed on a FACS ARIA cell sorter [22].

RNA from brain microglia or brain cortical tissue was isolated using the RNeasy Plus Micro Kit. RNA integrity was analyzed using the Bioanalyzer 2100 RNA Pico chips (Agilent Technologies, CA, USA).

Library preparation, RNA sequencing, and bioinformatics

Ion Torrent sequencing libraries were prepared according to the AmpliSeq Library prep kit protocol as we did before [21, 85]. Briefly, one ng of highly intact total RNA was reverse transcribed. The resulting cDNA was amplified for 16 cycles by adding PCR Master Mix and the AmpliSeq mouse transcriptome gene expression primer pool. Amplicons were digested with the proprietary FuPa enzyme, and then barcoded adapters were ligated onto the target amplicons. The library amplicons were bound to magnetic beads, and residual reaction components were washed off. Libraries were amplified, re-purified, and individually quantified using Agilent TapeStation High Sensitivity tape. Individual libraries were diluted to a 50 pM concentration and pooled equally. Emulsion PCR, templating, and 550 chip loading were performed with an Ion Chef Instrument (Thermo Scientific MA, USA). Sequencing was performed on an Ion S5XL™ sequencer (Thermo Scientific MA, USA) as we did before [85].

Data from the S5 XL run was processed using the Ion Torrent platform-specific pipeline software Torrent Suite v5.12 to generate sequence reads, trim adapter sequences, filter and remove poor signal reads, and split the reads according to the barcode. FASTQ and BAM files were generated using the Torrent Suite plugin FileExporter v5.12. Automated data analysis was done with Torrent Suite™ Software using the Ion AmpliSeq™ RNA plugin v.5.12 and target region AmpliSeq_Mouse_Transcriptome_V1_Designed as we did before [21, 85]. Raw data was loaded into Transcriptome Analysis Console (4.0 Thermo Fisher Scientific, MA, EUA) and first filtered based on ANOVA eBayes using the Limma package and displayed as fold change. For the microglia dataset, transcripts were filtered to identify significant changes (fold change < −1.5 or > 1.5, p-value < 0.05, and FDR < 0.1). For the brain cortex dataset, the complete transcript list was subjected to Gene Set Enrichment Analysis (GSEA), with fold change values used as ranking metrics and no pre-filtering threshold applied. Functional enrichment analyses were further performed using STRING [86]. Pathway enrichment was conducted using the REACTOME database with default settings. Enrichment scores for gene sets were calculated using an FDR cutoff of 0.05. Enriched pathways were manually recategorized to core transcriptomic modules and displayed as a network (constructed using Cytoscape) as before [21].

MACS isolation of microglia

Mice were perfused with ice-cold PBS, and their brains were removed. The right hemisphere was mechanically dissociated in ice-cold Dounce buffer (15mM HEPES; 0,5% Glucose; and DNAse) by six strokes in a tissue potter. Homogenate was pelleted by centrifugation, resuspended in MACS buffer (0.5% BSA; 2 mM EDTA in PBS), followed by incubation with 80 μL myelin removal microbeads. Homogenate was negatively selected using LS columns, pelleted, washed twice, resuspended in MACS buffer, and incubated 10 μL CD11b microbeads Kit. CD11b⁺ fraction was selected using LS columns according to the manufacturer’s instructions. Enriched fractions were centrifuged (9300 g; 1 min; 4ºC) and reserved for preparing protein extracts and Western blotting [21, 22].

Synaptosomal preparations

Synaptosomes were acutely prepared from the brain cortex using SynPer as before [84, 87]. One hundred micrograms of synaptosomal proteins from each sample were processed for proteomic analysis following the solid-phase-enhanced sample-preparation protocol as before [88]. Enzymatic digestion was performed overnight at 37 °C with trypsin/LysC (2 μg) at 1000 rpm. The resulting peptide concentration was measured by fluorescence. Enrichment for phosphorylated peptides was performed using Titanium dioxide beads (TiO2; ThermoFisher Scientific) as described in the manufacturer's protocol.

High-throughput proteomics, data acquisition, and quantification

Protein identification and quantitation were performed by nanoLC-MS/MS using an Ultimate 3000 liquid chromatography system coupled to a Q-Exactive Hybrid Quadrupole-Orbitrap mass spectrometer (Thermo Scientific, Bremen, Germany). Peptides of each sample were loaded onto a trapping cartridge (Acclaim PepMap C18 100 Å, 5 mm × 300 µm i.d., 160,454, Thermo Scientific, Bremen, Germany) in a mobile phase of 2% ACN, 0.1% FA at 10 µL/min. After 3 min loading, the trap column was switched in-line to a 50 cm × 75 μm inner diameter EASY-Spray column (ES803, PepMap RSLC, C18, 2 μm, Thermo Scientific, Bremen, Germany) at 300 nL/min. Separation was achieved by mixing A: 0.1% FA and B: 80% ACN, 0.1% FA with the following gradient: 5 min (2.5% B to 10% B), 120 min (10% B to 30% B), 20 min (30% B to 50% B), 5 min (50% B to 99% B), and 10 min (hold 99% B). Subsequently, the column was equilibrated with 2.5% B for 17 min. Data acquisition was controlled by Xcalibur 4.0 and Tune 2.9 software (Thermo Scientific, Bremen, Germany).

The mass spectrometer was operated in the data-dependent (dd) positive acquisition mode alternating between a full scan (m/z 380–1580) and subsequent HCD MS/MS of the 10 most intense peaks from a full scan (normalized collision energy of 27%). The ESI spray voltage was 1.9 kV. The global settings were as follows: use lock masses best (m/z 445.12003), lock mass injection Full MS and chrom. peak width (FWHM) of 15 s. The full scan settings were as follows: 70 k resolution (m/z 200), AGC target 3 × 10⁶, maximum injection time 120 ms; dd settings: minimum AGC target 8 × 10³, intensity threshold 7.3 × 10⁴, charge exclusion: unassigned, 1, 8, > 8, peptide match preferred, exclude isotopes on, and dynamic exclusion 45 s. The MS2 settings were as follows: microscans 1, resolution 35 k (m/z 200), AGC target 2 × 10⁵, maximum injection time 110 ms, isolation window 2.0 m/z, isolation offset 0.0 m/z, dynamic first mass, and spectrum data type profile.

The raw data were processed using the Proteome Discoverer 2.5.0.4 software (Thermo Scientific, Bremen, Germany). Protein identification analysis was performed with the data available in the UniProt protein sequence database for the Mus Musculus Proteome (2020_02 version, 55,398 entries) and a common contaminant database from MaxQuant (version 1.6.2.6, Max Planck Institute of Biochemistry, Munich, Germany). Sequest HT tandem mass spectrometry peptide database search program was used as the protein search algorithm. The search node considered an ion mass tolerance of 10 ppm for precursor ions and 0.02 Da for fragment ions. The maximum allowed missing cleavage sites was set as 2. For the phosphoproteomics, the IMP-ptmRS node, with the PhosphoRS mode (set to false), was used to localize phosphorylation sites. The Inferys rescoring node was considered, and the processing node Percolator was enabled with the following settings: maximum delta Cn 0.05; decoy database search target False Discovery Rate—FDR 1%; validation based on q-value. Protein-label-free quantitation was performed with the Minora feature detector node at the processing step. Precursor ion quantification used the processing step with the following parameters: Peptides: unique plus razor; precursor abundance was based on intensity; normalization mode was based on the total peptide amount; the pairwise protein ratio calculation and hypothesis test were based on a t-test (background based). The Feature Mapper node from the Proteome Discoverer software was used to create features from unique peptide-specific peaks within a short retention time and mass range. This was achieved by applying a chromatographic retention time alignment with a maximum shift of 10 min and 10 ppm of mass tolerance allowed for mapping features from different sample files. For feature linking and mapping, the minimum signal to noise (S/N) threshold was set at 5.

Further downstream analyses were performed using Proteome Discoverer 3.0. Peptide groups were filtered by modifications, excluding those that had no phosphorylation and contained oxidation or deamidation. Peptides with carbamidomethylation were maintained. The table file with the final phosphoprotein list was extracted to Microsoft Excel and peptides were further filtered for each comparison using a custom-made Python script using the following parameters: (1) only master proteins detected with high/medium confidence FDR; (2) a protein/phosphoprotein must be detected in more than 50% of samples in each experimental group (except for proteins that were depleted entirely in one of the experimental groups); (3) the P-value adjusted using Benjamini–Hochberg correction for the FDR was set to < 0.05; (4) at least 50% of samples with protein-related peptides sequenced by MS/MS; (5) the peptide spectrum matches (PSMs) was set to ≥ 3.

Proteomics network analyses

DE proteins retrieved from the LFQ experience were uploaded to STRING [86] within Cytoscape to construct granular networks. Enrichment analyses (FDR cutoff < 0.05 with the Benjamini–Hochberg multiple test adjustments) were conducted with REACTOME and GO as functional databases. Network construction and topography were carried out following manual pathway annotation and clustering in Cytoscape.

Phosphoproteome bioinformatic analyses

DE phosphoproteins of the datasets were inspected and cross-checked using the comprehensive web portal PhosphositePlus (https://www.phosphosite.org/homeAction.action [89]). Putative protein kinases controlling the phosphorylation of the filtered phosphopeptides were identified using NetworKIN 3.0 (http://networkin.info/index.shtml [90]). DE phosphoproteins were then screened for synaptic interactors/partners using Stitch and Mechismo [91]. Pathway enrichment analyses (FDR cutoff < 0.05 with the Benjamini–Hochberg multiple test adjustments) were carried out in STRING [86].

Metabolic analysis

Metabolic analyses were conducted using the Seahorse XF24 Analyzer to assess mitochondrial function in synaptosomal fractions acutely isolated from the cortex of Pfn1-deficient and control mice (as above). The assay was performed using the Seahorse XF Cell Mito Stress Test Kit (Agilent, 103,015–100), allowing real-time measurement of the oxygen consumption rate (OCR) and extracellular acidification rate (ECAR) as indicators of mitochondrial respiration and glycolytic activity, respectively.

To ensure optimal sensor calibration, XF24 cartridges were hydrated overnight at 37 °C in a non-CO₂ incubator with 1 ml of Seahorse XF Calibrant (Agilent, 100,840–000) per well. Prior to metabolic measurements, 50 μl of oligomycin (30 μM), carbonyl cyanide-4-(trifluoromethoxy)phenylhydrazone (FCCP, 40 μM), and rotenone/antimycin A (20 μM) were loaded into the corresponding ports of the injection cartridge, ensuring final working concentrations of 3 μM, 4 μM, and 2 μM, respectively, upon injection during the assay. The cartridge was then loaded onto the Seahorse XF24 Analyzer for calibration.

For the experimental setup, each well of a Seahorse XF24 plate was precoated with poly-D-lysine (50 μg/ml) to enhance synaptosomal adherence. A total of 10 μg of synaptosomal protein was seeded per well in a final volume of 100 μl of ionic medium containing 20 mM HEPES, 10 mM D-glucose, 1.2 mM Na₂HPO₄, 1 mM MgCl₂, 5 mM NaHCO₃, 5 mM KCl, and 140 mM NaCl, adjusted to pH 7.4 at 4°C. The plate was centrifuged at 2,000 g for 45 min at 4 °C using a swinging-bucket rotor in a Thermo Fisher Scientific 3000R centrifuge to promote synaptosomal attachment. Following centrifugation, the ionic medium was replaced with 180 μl of incubation medium consisting of 3.5 mM KCl, 120 mM NaCl, 1.3 mM CaCl₂, 0.4 mM KH₂PO₄, 1.2 mM Na₂SO₄, 2 mM MgSO₄, 4 mg/ml BSA, 15 mM D-glucose, 5 mM pyruvate, and 2.5 mM malate, pH 7.4 at 37 °C, ensuring physiological mitochondrial function.

OCR measurements began with the determination of basal respiration, followed by sequential injections of mitochondrial inhibitors to deconstruct specific mitochondrial parameters. First, oligomycin was injected to inhibit F0-F1 ATP synthase, allowing the quantification of ATP-linked OCR by measuring the respiration decrease upon ATP synthesis blockade. The residual OCR following oligomycin treatment represented the proton leak, reflecting mitochondrial uncoupling and inner membrane integrity. FCCP was subsequently injected to induce maximal mitochondrial respiration by uncoupling the electron transport chain (ETC), allowing the calculation of spare respiratory capacity, which is defined as the difference between maximal and basal OCR. Lastly, the complex I inhibitor rotenone and the complex III inhibitor antimycin A were added to completely suppress mitochondrial respiration, enabling the measurement of non-mitochondrial OCR, which was subtracted from all other parameters to isolate mitochondrial contributions.

Coupling efficiency, representing the proportion of oxygen consumption allocated to ATP synthesis relative to total basal OCR, was calculated as the fraction of ATP-linked OCR over basal OCR. Spare respiratory capacity was determined as the difference between maximal and basal respiration, providing insight into the ability of mitochondria to respond to increased energy demands.

Data acquisition and analysis were performed using Wave Software (Agilent Technologies), with normalization to total protein content measured by the BCA assay.

Electrophysiology

Solutions used for electrophysiology experiments were composed of (in mM): Sucrose-enriched artificial cerebrospinal fluid (Sucrose-aCSF)—Sucrose 75, NaCl 86.93, KCl 2.55, NaHCO₃ 25, NaH₂PO₄.H₂O 2.09, Glucose 25.03, MgSO₄ 7, CaCl₂ 0.5; Artificial cerebrospinal fluid (aCSF)—NaCl 130.90, KCl 2.55, NaHCO₃ 24.04, NaH₂PO₄.H₂O 1.23, Glucose 12.49, MgSO₄ 2, CaCl₂ 2; Cesium-based internal solution (Cs-Int) for recording mEPSC—CsMeSO₃ 115, CsCl 20, HEPES 10, EGTA 0.6, Na-phosphocreatine 10, MgCl₂ 2.5, ATP-Na⁺ salt 4, GTP-Na²⁺ salt 0.4; Chloride-enriched cesium-based internal solution (CsCl-Int) for recording mIPSC – CsMeSO₃ 110, CsCl 60, HEPES 2, Lidocaine 5, MgCl₂ 2, ATP-Na⁺ salt 2, GTP-Na²⁺ salt 0.4.

The osmolarity of solutions was adjusted to 300–310 mOsm for aCSF, 330–340 mOsm for sucrose-aCSF, and 295–298 mOsm for both internal solutions. The pH of all solutions was adjusted to 7.36, using HCl for external solutions and CsOH for internal solutions. All external solutions were oxygenated (95%:5% O₂:CO₂ mix) before and during usage.

Acute brain slices were prepared from Pfn1 cKO and CT mice. Mice were deeply anesthetized with isoflurane (Abbot) and immediately perfused with oxygenated, ice-cold, sucrose-aCSF. The brain was quickly removed and glued to a vibratome support filled with this same solution. Coronal slices (300 µm) containing the prefrontal cortex were obtained using a vibratome (Leica VT1200s, Leica Microsystems, USA) and the following settings: 0.12 mm/s slicing speed and 0.8 mm amplitude. Slices were immediately recovered in a holding chamber containing oxygenated aCSF at 32 °C for 30 min after which they were removed from the bath and left to recover for at least 1 h at room temperature. Slices were finally moved to the recording chamber and perfused with aCSF (2 to 3 mL/min) at 25°C.

The prefrontal cortex was visualized using an Axio Examiner.D1 microscope (Carl Zeiss) equipped with a Q-capture Pro7 camera (Teledyne). Pyramidal neurons were identified under infrared differential interference contrast microscopy with a 40 × water immersion objective. Cells were patched with borosilicate glass recording electrodes (3–5 MΩ; Science Products, Germany) filled either with Cs-Int solution for mEPSC recordings or CsCl-Int solution for mIPSC recordings. Neurons were voltage-clamped at −70 mV. To isolate mEPSCs and mIPSCs, recordings were performed in the presence of 1 μM tetrodotoxin (TTX; Tocris), 5 μM D-APV (Tocris), and either 25 μM bicuculline (Enzo) or 10 μM CNQX (Hello Bio), respectively.

Criteria for acceptance of cells was determined as a stable access resistance (≤ 20% change) which initial value was under 25 MΩ. Recordings were filtered at 2 kHz and digitized at 20 kHz. All data was acquired with a Multiclamp 700B amplifier and Digidata 1550A (Molecular Devices, USA) and analyzed using Clampfit software (v11, Molecular Devices).

To account for the hierarchical structure of the data (multiple recordings nested within each mouse), statistical comparisons of frequencies or amplitudes were performed using LMM implemented in Python with the statsmodels library. In all analyses, genotype was treated as a fixed effect, and mouse_id was included as a random effect to account for within-subject correlations.

Sex as a biological variable

Sex as a biological variable was a key consideration in designing this study. Our approach used a cross-sex strategy to ensure the generality of our findings while following the 3Rs principle of reducing animal numbers. This experimental design was reviewed and approved by the Portuguese National Authority for Animal Health (Direção-Geral de Alimentação e Veterinária, license 2022–02–18 003669), and all procedures meet EU Directive 2010/63/EU and ARRIVE 2.0 guidelines.

Specifically, intravital two-photon imaging and all molecular profiling (RNA-seq, proteomics) were performed on female mice. Females were selected for imaging due to their thinner skulls, which enhance optical clarity, and for omics to establish a more stable molecular baseline by avoiding the transcriptional variability caused by male pubertal androgens. Behavioral tests and electrophysiology, however, were conducted on male mice. This approach was chosen to eliminate the significant variability caused by the estrous cycle in behavioral and synaptic plasticity measurements, thus improving the ability to detect deficits specifically caused by Pfn1 loss. The core static morphological changes (as assessed by confocal morphometry) were mainly identified in the male group. They were also observed in a small cohort of female mice, confirming the direction of the effect. To validate Pfn1 depletion, Western blotting was performed on microglia isolated from both male and female mice. This confirmed that TAM-induced protein depletion was equally efficient across sexes, providing a solid basis for all subsequent experiments. After confirming that there was no significant difference in Pfn1 levels between sexes in the control group, the densitometry data were aggregated for final analysi

Statistics

Experimenters were blinded to genotype during data collection and analysis. Data normality and variance homogeneity were assessed with the D’Agostino–Pearson and Levene tests, respectively. Statistical significance was set at p < 0.05 (two-tailed). Data are shown as mean ± SD, except for Seahorse and behavioral results (mean ± SEM). Group comparisons used unpaired two-tailed Student’s t-tests when assumptions were met; Welch’s correction was applied when variances differed (e.g., Figs. 1B, 1D; Suppl. Fig. 1A). Datasets with multiple factors or repeated measures were analyzed by two-way ANOVA with Tukey’s or Šidák’s post-hoc tests (e.g., Figs. 1C, 8A). Hierarchical data, including microglial protrusion dynamics (Figs. 1F–G; Suppl. Fig. 1C and D) and miniature postsynaptic current recordings (Figs. 7E–H), were modeled with linear mixed effects models (LMMs) in statsmodels v0.14.1 (Python 3.11), using genotype as a fixed effect and mouse identity as a random intercept. Seahorse XF mitochondrial stress tests (Fig. 6) were analyzed with paired t-tests comparing matched synaptosomal preparations within each experimental batch. For omics datasets, false-discovery-rate thresholds were dataset-specific:– microglia RNA-seq: p < 0.05, FDR < 0.10 (Benjamini–Hochberg).– Proteomics and phosphoproteomics: p < 0.05, FDR < 0.05 (Proteome Discoverer workflow). The figure legends accompanying the results provide details of the statistical tests, including the exact value of n (number of animals, cells, or protrusions). All statistical analyses and the creation of graphical representations of the data were conducted using GraphPad Prism (version 9.0.2 for macOS) or Python 3.11.

The assembly of figure panels for publication was performed using Adobe Illustrator 2020 (version 24.3).

Schematic diagrams were created using BioRender.

Authors’ contributions

Conceptualization—CCP, RS, JBR; formal analyses—CCP, TOA, JTM, JG, TC, CS, AM, XB, IM, JC, JG, IMP, RS; software— BR, IMP; funding acquisition—CCP, TS, IMP, JP, RS, JBR; investigation— CCP, TOA, JTM, JG, TC, CS, AM, XB, IM, JC, JG, IMP, RS; resources—BR, FK, TS, JP, JBR.; supervision—R.S. and J.B.R.; validation—CCP, TOA, RS, JBR; visualization—CCP, RS, JBR; writing original draft—CCP, IMP, XB, FK, JG, JP, RS, JBR.

Funding

This work was supported by the project NORTE2030-FEDER-01711200 – ERA Chair NCbio_2030, supported by Norte Portugal Regional Operational Programme (NORTE 2030), under the PORTUGAL 2030 Partnership Agreement, via the European Regional Development Fund (FEDER). Portuguese funds financed this work through FCT—Fundação para a Ciência e Tecnologia/Ministério da Ciência, Tecnologia e Ensino Superior in the framework of the project PTDC/MED-NEU/1677/2021 (to João Relvas) and COMPETE2030-FEDER-00657300–15536 (to Renato Socodato). Inês Mendes Pinto acknowledges funding from the European Union’s Seventh Framework Program for research, technological development and demonstration (Marie Curie Actions) under grant agreement no 600375. This work was funded (in part) by Programa Operacional Regional do Norte and co-funded by the European Regional Development Fund under the project "The Porto Comprehensive Cancer Center" with the reference NORTE-01–0145-FEDER-072678—Consórcio PORTO.CCC – Porto. Comprehensive Cancer Center. Renato Socodato and Teresa Summavielle held employment contracts funded by FCT (2023.06380.CEECIND and 2022.03699.CEECIND, respectively). João Galvão, Tiago O. Almeida, and Joana Tedim-Moreira held Ph.D. fellowships financed by FCT (2024.03386.BD, SFRH/BD/147277/2019, SFRH/BD/147981/2019, and UI/BD/151552/2021, respectively).

Data availability

The data that support the findings of this study are available from the authors upon reasonable request; however, restrictions apply, and the data are not publicly available.

Declarations

Ethics approval and consent to participate

All animal experiments were conducted in accordance with the European Union Directive 2010/63/EU and were approved by the Ethics Committee for Animal Experimentation of the University of Porto and the Portuguese National Authority for Animal Health (Direção-Geral de Alimentação e Veterinária), under the license reference number 2022–02-18 003669.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.