Nanoplastics trigger glial–neuronal collagen signaling miscommunication to exacerbate cognitive impairment in Alzheimer's disease

Abstract

INTRODUCTION

Alzheimer's disease (AD) is a progressive neurodegenerative disorder with limited treatments and poorly defined environmental risks. Micro‐ and nanoplastics (MNPs) are widespread pollutants linked to neurotoxicity, but their role in AD remains unclear.

METHODS

We investigated the effects of 90‐day intragastric exposure to polystyrene nanoplastics (PS‐NPs) in amyloid precursor protein/presenilin 1 (APP/PS1) mice using behavioral tests, brain imaging, histopathology, and cell‐type‐resolved proteomics.

RESULTS

PS‐NPs exacerbated cognitive deficits and hippocampal damage in APP/PS1 mice. Proteomic and CellChat analyses revealed PS‐NPs enhanced neuroglial communication through the collagen–integrin axis. In vitro triculture demonstrated that PS‐NPs strengthened collagen‐mediated astrocyte–microglia–neuron signaling, whereas in vivo blockade with TC‐I 15 suppressed collagen activation and improved cognition in PS‐NP‐exposed APP/PS1 mice. Single‐nucleus RNA sequencing of human AD brains validated conserved activation of collagen signaling.

DISCUSSION

Our findings highlight that PS‐NPs exacerbate cognitive impairment in AD by driving collagen‐dependent neuroglial dysfunction, establishing MNPs as modifiable environmental risk factors.

Highlights

• MNPs act as environmental risk factors that worsen cognitive impairment in AD.

• PS‐NPs trigger glial–neuronal communication via the collagen–integrin axis in AD.

• PS‐NP‐induced astrocyte‐ and microglia‐derived collagen, driving neurotoxicity in AD.

• TC‐I 15 blocked collagen signaling and rescued cognition in PS‐NP‐exposed AD mice.

• Collagen signaling was upregulated in human AD brains, confirming disease relevance.

1. BACKGROUND

Dementia, characterized by progressive cognitive decline that profoundly disrupts daily life, has emerged as a major global public health challenge. ¹ Alzheimer's disease (AD) is the primary cause of dementia and is one of the most burdensome diseases of the 21st century. ² Currently, approximately 50 million people worldwide are living with dementia. ³ By 2050, this number is expected to exceed 150 million, driven by the aging global population. ⁴ Dementia is now the fifth leading cause of death globally, ⁵ with AD contributing significantly to disability‐adjusted life years lost, particularly among individuals aged 75 and older. ⁶ The accelerating prevalence of AD underscores the urgent need to identify modifiable risk factors and therapeutic target.

The 2024 update from the Lancet Commission on dementia prevention, intervention, and care highlighted environmental pollution, including fine particulate matter, as a modifiable risk factor for late‐life dementia. ⁷ Epidemiological studies have linked fine particulate matter to accelerated cognitive decline. ⁸ , ⁹ Among the most concerning components of fine particulate matter are micro‐ and nanoplastics (MNPs), particles less than 5 mm in diameter, which are pervasive in both ecosystems and the human body. ¹⁰ , ¹¹ , ¹² A recent study detected MNPs in the human brain and reported significantly higher levels in individuals with dementia, establishing for the first time a population‐level association between MNP exposure and AD. ¹³ Our previous work demonstrated that orally administered polystyrene nanoplastics (PS‐NPs) can cross the intestinal barrier, enter the bloodstream, and disseminate hematogenously to multiple organs, including the brain. ¹⁴ Notably, PS‐NPs induced AD‐related transcriptomic changes in the hippocampus of wild‐type C57BL/6J mice. ¹⁵ However, the mechanisms by which these changes contribute to cognitive impairment remain unclear.

Experimental evidence suggests that MNPs may exacerbate AD pathology by promoting amyloid beta (Aβ) aggregation and triggering neuroinflammation. ¹⁶ The hippocampus is essential for memory and is among the first brain region affected in AD. ¹⁷ , ¹⁸ Yet, direct cytotoxicity alone may not fully explain the observed cognitive deficits in exposed animals. Instead, recent studies propose that MNPs may influence intercellular communication within the brain parenchyma, a process fundamental to maintaining neural integrity and function. ¹⁹ Perturbations in glial–neuronal signaling have been implicated in the amplification of neurodegenerative cascades and cognitive decline, ²⁰ suggesting that MNP‐induced communication deficits mechanistically link between environmental exposure and AD‐related cognitive impairment.

To explore this hypothesis, we designed a stepwise investigation focusing on the hippocampus. Using a 90‐day intragastric PS‐NP exposure model in APPswe/PSEN1dE9 (amyloid precursor protein [APP]/presenilin 1 [PS1]) mice, we evaluated memory and learning performance, structural brain alterations, and molecular changes. Hippocampal neurons, astrocytes, microglia, and oligodendrocytes were subsequently isolated for cell‐type‐resolved proteomic analysis. Intercellular communication analysis using CellChat revealed distinct signaling signatures across cell types, with pronounced enhancement of the collagen signaling axis following PS‐NP exposure. An in vitro triculture system of neurons, astrocytes, and microglia was used to reconstruct intercellular interactions under PS‐NP exposure. Protein interactions were analyzed by co‐immunoprecipitation (CO‐IP). Pharmacological inhibition of the collagen pathway was applied during PS‐NP exposure to assess functional relevance in vivo. Finally, we examined single‐nucleus RNA sequencing (snRNA‐seq) data from human AD brain samples to evaluate glial–neuronal communication through the collagen pathway. Our study revealed a mechanistic link between environmental MNP exposure and exacerbation of cognitive impairment in AD, offering new therapeutic targets and strategies to mitigate the risks posed by environmental pollutants in aging‐related dementias.

2. METHODS

2.1. Animal study

We followed the guidelines set by the Scientific Research Committee at Southern Medical University on Ethics in the Care and Use of Laboratory Animals (SMUL202404002) for all animal procedures. APP/PS1 transgenic mice on a C57BL/6J background were obtained from the Guangdong Medical Laboratory Animal Center (Foshan, China). We maintained and bred these mice in a specific pathogen‐free facility at the Laboratory Animal Center, Southern Medical University (Guangzhou, China). To identify the transgenic mice, we used polymerase chain reaction (PCR) and confirmed the genotypes via agarose gel electrophoresis. We housed the mice in climate‐controlled conditions (22°C ± 2°C, 30% to 70% humidity) with a 12‐h light/dark cycle (lights on at 8:00 a.m. and off at 8:00 p.m.). The mice had ad libitum access to food and water. To ensure unbiased results, we randomly grouped the mice by body weight (BW) for each experiment.

We used both fluorescent and non‐fluorescent PS‐NPs (Magsphere Inc., CA, USA) in this study. Their physicochemical properties were characterized in our previous work. ¹⁴ , ¹⁵ In earlier work, we detected trace sodium azide in the stock solution. ²¹ We measured its concentration in the working solution and confirmed it remained below neurotoxicity thresholds reported by the National Toxicology Program. ²¹ , ²² We also verified that fluorescence leakage from labeled PS‐NPs was virtually absent under our experimental conditions. ¹⁴ , ¹⁵

To investigate the impact of PS‐NP exposure on cognitive deficits in AD, we used 60 APP/PS1 male mice, aged 2 months, for a 90‐day exposure experiment. A single oral gavage was administered to each mouse, with PS‐NPs (Magsphere Inc., Catalog No. PS050NM) suspended in pure water to prepare the gavage solution. The control group received pure water, while the experimental groups were treated with PS‐NPs at doses of 0.25 or 25 mg/kg BW, with a gavage volume of 20 mL/kg BW.

RESEARCH IN CONTEXT

1. Systematic review: We searched PubMed for studies examining environmental exposures and AD. Prior work has linked fine particulate matter to accelerated cognitive decline. Recent reports identified MNPs in human brains, with higher levels in individuals with dementia. However, the mechanistic link between MNPs and AD remains poorly understood.

2. Interpretation: Our findings demonstrate that oral exposure to PS‐NPs aggravates cognitive impairment in AD by enhancing collagen‐dependent signaling among astrocytes, microglia, and neurons. This aberrant communication drives hippocampal injury, whereas pharmacological inhibition of collagen signaling reversed PS‐NP–induced neurotoxicity and cognitive deficits. SnRNA‐seq further confirmed upregulated collagen signaling in human AD brains.

3. Future directions: Our study highlights plastic pollution as a modifiable environmental contributor to dementia. Interventions that reduce MNP exposure or target collagen‐mediated neuroglial dysfunction may help mitigate pollution‐aggravated cognitive decline.

To quantify brain accumulation of PS‐NPs, additional APP/PS1 mice received daily oral gavage of fluorescent PS‐NPs (Magsphere Inc., Catalog No. PSYF050NM; 0.25 or 25 mg/kg BW) or double‐distilled water for 90 days. Following anesthesia and transcardial perfusion with saline, hippocampal tissues were collected, weighed, and digested in buffer containing proteinase K (Vetec, Darmstadt, Germany, Catalog No. V900887), sodium dodecyl sulfate (SDS; Biosharp, Hefei, China, Catalog No. BS088), Na₂HPO₄ (Guangzhou Chemical Reagent Factory, Guangzhou, China, Catalog No. BG06), and NaH₂PO₄ (Guangzhou Chemical Reagent Factory, Catalog No. BG05). Fluorescence intensity was measured by spectrophotometry using tissue‐matched standard curves, as described previously. ¹⁴ In parallel, frozen hippocampal sections were subjected to fluorescence scanning to assess PS‐NP distribution.

To evaluate the effect of blocking the collagen pathway on PS‐NP‐exacerbated cognitive deficits in AD, we dissolved 10 µg of TC‐I 15 (MedChemExpress, NJ, USA, Catalog No. 916734‐43‐5) in 100 µL phosphate‐buffered saline (PBS; Servicebio, Wuhan, China, Catalog No. G0002) and administered it intraperitoneally to APP/PS1 mice every other day throughout the 90‐day PS‐NP exposure period (TC‐I 15 10 µg/mouse). ²³ , ²⁴

At the experimental endpoint, following behavioral testing, we euthanized the mice humanely. We administered a 3% pentobarbital injection to induce deep anesthesia, confirmed by the absence of response to tail or toe pinch stimuli. Using tissue scissors, we carefully incised the ribcage to expose the heart and lungs. Transcardial perfusion with ice‐cold PBS was performed until the effluent from the right atrium ran clear. Brain tissues were collected for subsequent analyses. Samples designated for biochemical tests were immediately frozen at −80°C, while those for proteomics studies underwent hippocampal cell isolation. For pathological examination, we perfused the mice with PBS followed by 4% paraformaldehyde (PFA; Servicebio, Catalog No. G1101) to ensure proper tissue fixation. These brains were then fixed in 4% PFA for further analysis. All major reagents and instruments used were listed in the key resources table (Table S1).

2.2. Cell culture and treatment

Human astrocyte line SVG p12 was purchased from the American Type Culture Collection (ATCC, Manassas, VA, USA, Catalog No. CRL‐8621), while human microglia line HMC3 (Catalog No. CL_0620) and neuron line SH‐SY5Y were obtained from Procell (Wuhan, China, Catalog No. CL_0208). SH‐SY5Y cells were selected as the neuronal model due to their capacity to express Aβ following retinoic‐acid‐induced differentiation into neuron‐like cells (RA‐SH‐SY5Y), making them a widely accepted in vitro system for AD research. ²⁵ All cell lines were cultured in a 5% CO₂ humidified atmosphere at 37°C. We maintained the cells in Dulbecco's modified Eagle's medium (DMEM; Sigma‐Aldrich, St. Louis, MO, USA, Catalog No. D6429) supplemented with 10% fetal bovine serum (FBS; Gibco, Carlsbad, CA, USA, Catalog No. 10099141C) and 1% penicillin‐streptomycin (Meilunbio, Dalian, China, Catalog No. MA0110). Mycoplasma contamination was ruled out using a One‐step Quickcolor Mycoplasma Detection Kit (Yise Medical Technology, Shanghai, China, Catalog No. MD001) every 3 months. To ensure consistency, all experiments were conducted with cell passages ranging from 10 to 20. None of the cell lines used in this study appear in the database of commonly misidentified cell lines maintained by the International Cell Line Authentication Committee (ICLAC, http://iclac.org/databases/cross‐contaminations).

SH‐SY5Y neurons were pretreated with 10 µM retinoic acid (RA) for 7 days. Medium was changed daily. RA‐induced cells were used in both mono‐ and triculture systems for AD modeling. ²⁶ , ²⁷ To evaluate the effects of PS‐NP exposure, we separately cultured SVG p12, HMC3, and RA‐SH‐SY5Y cell lines in six‐well plates at 50% to 70% confluence. After overnight incubation, we treated the cells with PS‐NPs at concentrations of 0, 2, 10, or 50 µg/mL for 48 h, following established protocols. ²⁶ To investigate the effects of PS‐NPs on intercellular communication among these cell types, we established an in vitro triculture model using transwell polycarbonate membrane inserts in six‐well plates following the published article (BIOFIL, Guangzhou, China, Catalog No. TCP001006). ²⁸ , ²⁹ , ³⁰ Initially, we seeded HMC3 microglia (1.5 × 10⁶ cells) into the upper chamber with serum‐free DMEM, while the lower chamber contained DMEM supplemented with 10% FBS. After 2 days, the HMC3 microglia migrated and adhered to the underside of the transwell membrane, and residual cells in the upper chamber were removed using a sterilized cotton swab. The transwell insert with HMC3 microglia was transferred to a six‐well plate preseeded with RA‐SH‐SY5Y neurons (3.0 × 10⁶ cells). We subsequently seeded SVG p12 astrocytes (1.5 × 10⁶ cells) into the upper chamber, and the medium was replaced with fresh DMEM containing 10% FBS. Following 24 h of triculture, we treated the cells with PS‐NPs (0, 2, 10, or 50 µg/mL) for another 48 h.

We investigated the involvement of the collagen pathway in PS‐NP exposure under triculture conditions by employing two approaches: small interfering RNA (siRNA)‐mediated gene silencing and ligand–receptor inhibition. To suppress the COL1A1 ligand, we used two independent siRNAs targeting the COL1A1 gene (Table S2), procured from Tsingke Biological Technology (Beijing, China). SVG p12 astrocytes were seeded in six‐well plates at 50% to 70% confluence 24 h prior to transfection. For transfection, we introduced 30 nM siRNA into the cells using RNAFit (HanBio, Shanghai, China, Catalog No. HB‐RF‐1000) according to the manufacturer's protocol. To inhibit ligand–receptor binding, we used the collagen‐binding integrin inhibitor TC‐I 15. During the migration and adhesion of HMC3 microglia to the underside of the transwell membrane, we supplemented the culture medium with 2 µM TC‐I 15 to block ITGA1/ITGB1 interaction with ligand. ²³ , ²⁴ Following these treatments, the triculture experiment was carried out as previously described.

2.3. Single‐nucleus transcriptomic analysis of PS‐NP‐exposed mice and AD patients

To assess the relevance of collagen‐mediated signaling across species and disease states, we analyzed snRNA‐seq data from PS‐NP‐exposed wild‐type C57BL/6J mice brain from our previous study ¹⁵ and from human AD brains (GSE188545). ³¹ Mouse brain nuclei were isolated from control (n = 3) and 250 mg/kg PS‐NP‐treated (n = 3) groups. The human dataset comprised post mortem prefrontal cortex samples from AD patients (n = 6) and age‐matched healthy controls (n = 6) (Table S3). Raw reads were processed using the Seurat R package in RStudio. Cell clustering was performed with a graph‐based approach applied to principal component analysis‐reduced data, then a shared nearest neighbor graph was constructed to identify distinct cell populations. We focused our analysis on nuclei representing neurons, astrocytes, microglia, and oligodendrocytes, which are key players in this study. For cell–cell interaction mapping, we employed the CellChat R package to identify and visualize communication networks among these cell types. We focused specifically on interactions mediated by the collagen signaling pathway, emphasizing its critical role in facilitating crosstalk among neurons, astrocytes, microglia, and oligodendrocyte.

2.4. Y‐maze test

We evaluated the impact of PS‐NP exposure on long‐term memory deficits in APP/PS1 mice using the Y‐maze test. The Y‐maze apparatus consisted of three identical arms, each 35 cm long, 5 cm wide, and 20 cm high, made from white polyvinyl chloride. The arms were labeled “Target,” “Reference,” and “Start,” positioned at 120° angles from each other and extended from a central triangular platform. All Y‐maze experiments were conducted by the same investigator in a quiet room, and the tests were performed in a blinded manner. During the training phase, we closed the “Reference” arm for all groups. Mice were placed in the “Start” arm and allowed to explore the “Start” and “Target” arms for 5 min. We placed food in the “Target” arm to encourage the mice to select the food‐reinforced arm and complete feeding. After exploration, the mice were returned to their home cages and given a 1‐h intertrial interval. In the test phase, we opened all arms and allowed the mice to explore the maze for 5 min. Mice were trained for five consecutive days. The Y‐maze was cleaned with 75% ethanol between trials. The number of entries into and the first choice of entry were registered from video recordings by an observer blinded to the experimental groups. Escape latency and arm entry counts were analyzed using a maze video tracking system (Chucai Electronic Technology Co., Shanghai, China).

2.5. Open field test (OFT)

We conducted OFTs on the APP/PS1 mice to assess the impact of PS‐NP exposure on activity during AD, as previously described in our earlier studies. ¹⁵ , ²⁶ The same investigator performed all OFT experiments in a quiet room and ensured that all groups were tested in a blinded manner. To maintain objectivity, the analysts reviewing the video data were also blinded to the group assignments. All mice were naive to the OFTs, with no prior exposure to the test. We placed each mouse individually in the central area of an opaque box (60 × 60 × 30 cm, Flyde, Guangzhou, China) and allowed it to explore freely for 5 min. We captured the mice's movements with an overhead camera, recording the total trajectory during the 5‐min session. Before each trial, we cleaned the box with alcohol, followed by water, to remove any residual odors that could influence the behavior of the mice. We analyzed the frequency and duration of time spent in the central area using a maze video tracking system (Chucai Electronic Technology Co., Shanghai, China).

2.6. Novel object recognition (NOR)

We conducted NOR tests to assess the effect of PS‐NP exposure on short‐term memory impairment in APP/PS1 mice. The tests took place in the same open field arena as the OFT (60 cm × 60 cm × 30 cm, Flyde). All mice completed the OFT before starting the NOR test. This setup provided a 5‐min habituation period to help mice adjust to the exploratory environment. In the learning phase, the arena contained two identical objects, matched for size, texture, and color. Mice explored these objects for 5 min. After this phase, the mice returned to their home cages for a 1‐h intertrial interval. In the testing phase, one of the identical objects was replaced with a novel object with a different appearance. Mice explored the arena again for 5 min. ³² We recorded the movement of each mouse using a digital charge‐coupled device camera placed overhead. The ezTrack Location Tracking Module, an open‐source software, analyzed the recordings. ³³ Exploration included walking within 2 to 3 cm of an object while sniffing or sweeping vibrissae. We calculated the NOR‐frequency ratio (%)  = (frequency of explored novel object: C _novel)/(C _novel + C _familiar) × 100%. The discrimination ratio was calculated using the following formula: (spent time for each object: T _novel − T _familiar)/(T _novel + T _familiar) × 100%. To avoid confounding factors, the animals used in the Y‐maze test were not included in the NOR test, ensuring that frequent environmental changes or prolonged training‐induced fatigue did not influence the experimental results. We cleaned the arena thoroughly with 75% ethanol and water between trials to remove any residual odors. Light intensity and environmental conditions remained consistent throughout the experiments.

2.7. Magnetic resonance imaging (MRI)

We conducted MRI to assess hippocampal damage linked to PS‐NP‐exacerbated cognitive deficits in APP/PS1 mice, following established protocols. ³⁴ , ³⁵ At the end of the exposure period, we anesthetized the mice with 3.5% isoflurane for induction and positioned them on a MRI‐compatible cradle. Isoflurane was reduced to 2% and maintained throughout the procedure while we secured the mice's heads with ear bars and stabilized their teeth using a bite bar. Rectal temperature was maintained at 37°C ± 0.1°C using a magnetic resonance (MR)‐compatible heating system (Small Animal Instruments, USA). We continuously monitored physiological parameters, including rectal temperature, breathing rate, heart rate, and oxygen saturation, using a MR‐compatible monitoring system (Small Animal Instruments, USA). The setup for each mouse required approximately 15 min, followed by an additional 15 min for localization and anatomical MRI scans. During this period, the physiological parameters stabilized to normal ranges (rectal temperature: 37°C ± 0.1°C, breathing rate: 80 breaths per min, heart rate: 350 to 420 beats per min, oxygen saturation: > 95%) before initiating functional MRI (fMRI) acquisition.

We performed all fMRI scans on a Bruker 7T MRI scanner (Bruker BioSpin, Germany) equipped with a mouse head cryocoil (MRI CryoProbe, Bruker, Germany). A T₂‐weighted pilot scan was conducted using the following parameters: RARE sequence (repetition time = 3,000 ms, echo time = 60 ms, RARE factor = 18, slice thickness = 1 mm, field of view = 40 × 55 mm², in‐plane spatial resolution = 0.5 × 0.5 mm², matrix size = 80 × 110, and one signal average). MRI data were analyzed using custom ParaVision 360 (Bruker, Karlsruhe, Germany). We calculated T₂* maps for each dynamic frame to facilitate quantitative analysis.

2.8. Nissl staining

We performed Nissl staining to assess neuronal injury associated with PS‐NP‐exacerbated cognitive deficits in APP/PS1 mice. After post‐fixing the brains in 4% PFA at 4°C overnight, we embedded the tissues in paraffin. Brain tissues were sectioned into 5‐µm slices and stained with Nissl staining solution (Servicebio, Catalog No. GP1043) for 2 to 5 min, followed by washing with running water until clear. Differentiation was achieved using 0.1% glacial acetic acid until the Nissl bodies appeared dark blue, and the background was light blue or colorless. Slides were dried and mounted using Permount TM Mounting Medium (Servicebio, Catalog No. WG10004160). We examined and imaged the sections under a microscope. For each mouse, we captured three representative images from five evenly spaced brain sections to ensure comprehensive coverage of the hippocampus. To minimize bias, all analyses were conducted in a blinded manner. Quantification of Nissl‐positive neurons in the hippocampus was performed using ImageJ version 1.52 (National Institutes of Health [NIH], Bethesda, MD, USA).

2.9. Reactive oxygen species (ROS) detection fluorescence assay on brain slides

We detected ROS in brain tissues using dihydroethidium (DHE) (Sigma‐Aldrich, Catalog No. D7008‐10). Frozen brain tissue sections (5 µm thick) were thawed at room temperature for 20 min. Nuclei were stained by incubating the sections with 4′,6‐diamidino‐2‐phenylindole (DAPI; Servicebio, Catalog No. G1407‐25ML, 1:500 working solution) for 3 min in the dark. Following nuclear staining, the sections were washed three times with Tris‐buffered saline with Tween (TBST; Servicebio, Catalog No. G0004), with each wash lasting 5 min. For ROS detection, we prepared a DHE working solution by diluting the stock solution 1:400 with TBST and applied 50 to 100 µL of the solution to each section, adjusted according to tissue size. The sections were incubated in a humidified chamber at room temperature for 1 h, protected from light. After incubation, the sections were washed three additional times with TBST for 5 min per wash. We mounted the slides with an antifade mounting medium to preserve fluorescence and stored them at 4°C in the dark. Fluorescence microscopy was used to capture and analyze images, ensuring consistent methodology across all samples.

2.10. Immunofluorescence (IF) staining

We conducted multiplex IF staining to analyze protein localization in brain tissue sections. Using the same paraffin‐embedded brain specimens from the Nissl staining experiment, we performed IF staining following established protocols. ¹⁵ , ³⁶ , ³⁷ Brain samples were coronally sectioned into 5‐µm slices and mounted on slides. Sections were washed three times with PBS (pH 7.4) on a pendulum shaker (Servicebio, Catalog No. SYC‐Z100), with each wash lasting 5 min. Antigen retrieval was conducted in citric acid antigen retrieval buffer (pH 6.0, Servicebio, Catalog No. G1202), heated to boiling for 10 min, ensuring buffers remained moist throughout. Slides were cooled naturally to room temperature and washed three times with PBS. To block endogenous peroxidase activity, tissue boundaries were marked with a tissue pen (Servicebio, Catalog No. G6100), and sections were incubated in 3% hydrogen peroxide for 25 min in a lightproof wet box (Servicebio, Catalog No. SIB‐20F), followed by three PBS washes. Serum blocking was performed with 3% BSA (Servicebio, Catalog No. GC305006) for 30 min at room temperature. Primary antibodies were applied and incubated overnight at 4°C in a lightproof wet box. After washing, HRP (Servicebio, Catalog No. G3431‐200UL)‐labeled secondary antibodies were added and incubated for 50 min at room temperature. Tyramide signal amplification (TSA; Servicebio, Catalog No. G1255) reagent was applied for 10 min, followed by three washes with TBST. For subsequent primary antibody cycles, the sequence of blocking, antibody application, HRP‐labeled secondary antibody incubation, TSA reagent application, and microwave antigen retrieval in citric acid buffer was repeated, up to four cycles. Following immunostaining, nuclei were stained with DAPI (Servicebio, Catalog No. G1012) for 10 min in a lightproof wet box. Autofluorescence was quenched by treating slides with a quenching agent (Servicebio, Catalog No. G1221) for 5 min, followed by a 10‐min water rinse. Slides were mounted with an anti‐fluorescence quenching medium (Servicebio, Catalog No. G1401). Fluorescence imaging was performed using a microscope equipped with specific excitation and emission filters. Spectral imaging and panoramic scans ensured detailed and comprehensive analysis. A complete list of primary and secondary antibodies is presented in a key resources table (Table S1).

2.11. Thioflavin‐S staining

To assess PS‐NP‐induced amyloid pathology, we performed Thioflavin‐S staining on brain sections newly cut from the same paraffin‐embedded tissue blocks used for Nissl staining. Following deparaffinization and rehydration, sections were washed in PBS (3 × 5 min) and encircled with a hydrophobic barrier pen to retain staining solutions. Sections were incubated in 0.3% Thioflavin‐S (Servicebio, Catalog No. GDP1028) dissolved in 50% ethanol at 37°C for 10 min in the dark, followed by PBS washes (3 × 5 min) under light‐protected conditions. To minimize background fluorescence, sections were treated with 75% ethanol (3 × 5 min), also in the dark. After rinsing, we applied an antifade mounting medium containing DAPI and mounted coverslips. Fluorescence signals were captured using a fluorescence microscope.

2.12. Transmission electron microscopy (TEM)

To observe PS‐NPs in the hippocampus, 1‐mm³ hippocampal tissue samples were collected from anesthetized mice after 90 days of PS‐NP exposure. The tissue was immediately placed in TEM fixative (Servicebio, Catalog No. G1102‐100ML) at 4°C for 24 h, then post‐fixed with 1% osmium tetroxide for 2 h at room temperature. After dehydration in a graded ethanol series, the tissue underwent resin infiltration and embedding, followed by dehydration in acetone, and kept in a 37°C oven overnight. The embedded tissue was polymerized at 60°C for 48 h. Semi‐thin sections (1.5 µm) were cut, stained with toluidine blue, and examined under a light microscope for initial analysis. Ultra‐thin sections (60 nm) were then cut using an ultramicrotome, placed on formvar‐coated copper grids, and stained with 2% uranyl acetate for 8 min. After rinsing in 70% ethanol and ultra‐pure water, the grids were stained with 2.6% lead citrate for 8 min, followed by further rinsing and drying with filter paper. The prepared grids were placed in grid holders and stored overnight at room temperature. Finally, the grids were observed under a transmission electron microscope, and images were captured for detailed analysis.

2.13. Isolation of oligodendrocytes, neurons, microglia, and astrocytes

We isolated oligodendrocytes, neurons, microglia, and astrocytes from the hippocampus of APP/PS1 mice following a 90‐day intragastric exposure to PS‐NPs. The isolation process followed established protocols. ³⁸ Hippocampal tissues from mice exposed to PS‐NPs for 90 days underwent intracardial perfusion to remove blood. We minced the tissues into small pieces (< 1 mm) using scalpels in a fresh Petri dish containing TrypLE Select trypsin (Gibco, Catalog No. 12563029). The tissue was digested at 37°C for 15 min. Digestion was stopped by adding DMEM supplemented with 10% FBS. We gently pipetted the suspension to generate single cells and filtered it through a 70‐µm cell strainer (SORFA, Zhejiang, China, Catalog No. SCS701). Using MACS magnetic sorting (Miltenyi Biotec, Bergisch‐Gladbach, Germany), we sequentially isolated specific cell types. Filtered cells were centrifuged at 300 × g for 5 min, and the supernatant was discarded. The pellet was resuspended in 80 µL of buffer (pH 7.2 PBS with 0.5% BSA, MACS BSA Stock Solution, Catalog No. 130‐091‐376), and 20 µL of anti‐myelin microbeads (Miltenyi, Catalog No. 130‐104‐262) were added to label oligodendrocytes. The mixture was incubated at 4°C for 10 min, washed with 2 mL of buffer, and centrifuged at 300 × g for 5 min at 4°C. The resuspended cells were applied to a prewetted LS column (Miltenyi, Catalog No. 130‐042‐401) placed in a MidiMACS Separator (Miltenyi, Catalog No. 130‐042‐302). The flow‐through, containing neurons, microglia, and astrocytes, was collected after two washes with 500 µL of buffer. Oligodendrocytes were retained within the column. The column was removed from the magnetic holder and placed into a clean 1.5‐mL conical tube. Oligodendrocytes were eluted by adding 500 µL of buffer and pushing the solution through the column using a plunger. We repeated the same procedure to isolate neurons, microglia, and astrocytes. For neurons, we used the MACS Neuron Isolation Kit (Miltenyi, Catalog No. 130‐115‐389). For microglia, we used MACS CD11b⁺ Microbeads (Miltenyi, Catalog No. 130‐093‐634). For astrocytes, we used the MACS ACSA‐2 MicroBead Kit (Miltenyi, Catalog No. 130‐097‐678). After each isolation, we centrifuged the cells at 300 × g for 3 min at 4°C and discarded the supernatant. We stored the final isolated cell samples at −80°C.

2.14. Label‐free cell proteomics analysis

We performed proteomic analysis of oligodendrocytes, neurons, microglia, and astrocytes to examine cell–cell communication changes driving PS‐NP‐exacerbated cognitive deficits in APP/PS1 mice. The protocol followed methods described in our previous studies. ³⁷ , ³⁹ We began by homogenizing the isolated cells using a high‐speed, low‐temperature tissue homogenizer (Servicebio) operating at 60 Hz, in IP lysis buffer (Beyotime, Shanghai, China, Catalog No. P0013) containing phosphatase and protease inhibitor cocktail (Keygen, Nanjing, China, Catalog No. KGB5101‐100; Catalog No. KGB5101‐2) at 4°C for 3 min. The homogenates were incubated on ice for 30 min to allow complete lysis and then centrifuged at 12,000 × g for 30 min at 4°C to separate the soluble proteins. For each cell type, we pooled two samples to create one biological replicate, resulting in three replicates per group for downstream analysis. The protein suspension was filtered through 3‐kDa molecular weight cutoff devices (Millipore, CA, USA, Catalog No. UFC5003) to remove contaminants and concentrate the protein. We reduced the proteins using dithiothreitol (10 mM final concentration) for 60 min at room temperature to break disulfide bonds. This was followed by alkylation with iodoacetamide (25 mM final concentration) for 60 min in the dark at room temperature to block free sulfhydryl groups. For protein digestion, we added trypsin (Thermo Fisher Scientific, Waltham, MA, USA, Catalog No. 90057) at a 1:50 enzyme‐to‐protein mass ratio. The mixture was incubated overnight at 37°C to digest the proteins into peptides. The resulting peptides were desalted using Pierce Peptide Desalting Spin Columns (Thermo Fisher Scientific, Catalog No. 89851). We activated the columns with 300 µL of acetonitrile (ACN) and equilibrated them by washing twice with 0.1% formic acid (FA). After loading the peptide digested, we washed each column twice with 0.1% FA to remove impurities and eluted the peptides twice using 50% ACN with 0.1% FA. Finally, we dried the desalted peptide solutions at 45°C using a SpeedVac and stored them at −20°C for subsequent proteomic analysis.

We conducted quantitative proteomic analysis on pooled samples using label‐free mass spectrometry with a high‐resolution Orbitrap mass spectrometer (Orbitrap Exploris 480, Thermo Fisher Scientific, USA). First, we dissolved peptides in 0.1% FA and loaded them onto a C18 trap column (Thermo Fisher Scientific, Catalog No. 164535), followed by separation on a C18 analytical column (75 µm × 250 mm, 2 µm particle size, 100 Å pore size, Acclaim PepMap C18; Thermo Fisher Scientific, Catalog No. 164941). Each sample was independently injected three times for analysis. To account for technical variability, proteins were considered quantifiable if detected in at least two out of three technical injections per sample. We then processed the raw data using Proteome Discoverer 2.1 (Thermo Fisher Scientific), searching against the Uniprot mouse protein database. Protein quantification was performed using the label‐free quantification algorithm in Proteome Discoverer.

We processed the data further using the DEP package in RStudio, beginning by filtering out missing values, particularly proteins identified in only one of the three biological replicates. To correct for background and ensure proper normalization, we applied variance‐stabilizing transformation. Missing values were imputed using the k‐nearest neighbor method. We then performed differential protein expression analysis, using empirical Bayes statistics with a p‐value significance threshold of < 0.05 for further analysis. Gene Ontology (GO) enrichment analysis was conducted using the enrichplot package. To explore intercellular communication, we used the CellChat package following established protocols. ⁴⁰ Normalized protein expression matrices for oligodendrocytes, neurons, microglia, and astrocytes were independently loaded into the workflow. We then calculated the probability of cell‐to‐cell communication and assessed intercellular communication at the signaling pathway level by aggregating the communication probabilities of all ligand–receptor interactions associated with each pathway.

2.15. Ultra‐high‐performance liquid chromatography coupled with high‐resolution mass spectrometry (UHPLC‐HRMS) analysis

We quantified TC‐I 15 levels in mouse hippocampal tissue using UHPLC‐HRMS. For each biological replicate, three unilateral hippocampal samples were pooled, with three replicates analyzed for both the control and PS‐NPs + TC‐I 15 groups. Tissue homogenates were prepared by grinding samples in 300 µL of ultrapure water with a cryogenic grinder, followed by the addition of 700 µL methanol and ultrasonication in an ice bath for 20 min. The samples were centrifuged at 16,000 × g at 4°C for 15 min, and the supernatant was collected, vacuum‐dried, reconstituted in 100 µL of 40% methanol‐water solution, vortexed, and centrifuged again under the same conditions. The final supernatant was used for analysis.

Chromatographic separation was achieved using a Vanquish UHPLC system (Thermo Fisher Scientific) equipped with an ACQUITY UPLC HSS T3 column (2.1 mm × 100 mm, 1.8 µm; Waters, Milford, MA, USA, Catalog No. 176001608). The column temperature was maintained at 35°C, with a flow rate of 0.3 mL/min. The mobile phases were 0.1% FA in water (A) and 0.1% FA in acetonitrile (B). Mass spectrometric data were collected using a Q‐Exactive HFX mass spectrometer (Thermo Fisher Scientific) operating in Full‐MS/dd‐MS2 mode with an m/z scan range of 90 to 1,300. The spray voltage was set to 3,800 V in positive ion mode (ESI⁺) and 3,500 V in negative ion mode (ESI⁻). The sheath gas flow rate was 40 L/min, with an ion transfer tube temperature of 320°C and a vaporizer temperature of 350°C. The top 10 MS1 ions were selected for fragmentation using stepped normalized collision energies of 20, 40, and 60, providing comprehensive MS/MS data. Compound annotation was done by matching accurate mass, isotopic distributions, and MS/MS spectral to reference data from an in‐house standards TCM database (Shanghai Applied Protein Technology co., Shanghai, China) and public databases GNPS, ReSpect, and Massbank.

2.16. CO‐IP and Western blot

We homogenized cell and hippocampal tissue samples in IP lysis buffer supplemented with phosphatase and protease inhibitors. After incubating the homogenates on ice, we centrifuged them at 12,000 × g to extract soluble proteins. Protein concentrations were measured using an Enhanced BCA Protein Assay Kit (Beyotime, Catalog No. P0009). To investigate ligand–receptor interactions involved in altered cell–cell communication during PS‐NP exposure, we performed CO‐IP assays. We targeted the interactions between COL1A1 and ITGA1/ITGB1 in HMC3 microglia and between COL1A2 and ITGA1/ITGB1 in RA‐SH‐SY5Y neurons in the triculture cell model. Before starting, we prewashed Protein A/G magnetic beads (Solarbio, Beijing, China, Catalog No. M2400) with IP binding buffer (Solarbio) and conjugated them with anti‐COL1A1, anti‐COL1A2, or mouse IgG antibodies for 30 min at room temperature. We reserved 15% of the total protein lysate as input, mixed it with 5× loading buffer, and boiled it at 100°C for 5 min. The remaining protein extracts were incubated with antibody‐conjugated beads at room temperature for 2 h with constant shaking. Isotype IgG served as a negative control. After incubation, we washed the beads four times with IP washing buffer (Solarbio) to remove non‐specific binding. We eluted the bound proteins by heating them with 5× loading buffer for 5 min at 100°C and then analyzed via Western blot.

We separated the proteins on 8% or 12% SDS‐polyacrylamide gel electrophoresis and transferred them onto polyvinylidene fluoride membranes (Bio‐Rad, CA, USA, Catalog No. 1620177). To block non‐specific binding, we incubated the membranes with 5% skim milk in TBST for 1 h at room temperature. Subsequently, we incubated the membranes overnight at 4°C with primary antibodies. We visualized the protein bands using an enhanced chemiluminescence kit (Millipore, Catalog No. WBKLS0500) and captured images with a Tanon‐5200 chemiluminescence imaging system (Tanon, Shanghai, China). For quantification, we analyzed target protein bands using ImageJ software (version 1.52, NIH, Bethesda, MD, USA). All protein expression levels were normalized to β‐ACTIN as the housekeeping protein. For CO‐IP assays, we normalized protein expression levels to the immunoprecipitated protein levels in each sample. Details of all antibodies used are provided in the key resources table (Table S1).

2.17. Enzyme‐linked immunosorbent assay (ELISA)

We measured Aβ₄₀ and Aβ₄₂ levels in RA‐SH‐SY5Y cells and in hippocampal tissues from mice. Both samples were lysed in PBS containing 0.5% Triton X‐100 (Beyotime, Catalog No. P0096). Human‐specific ELISA kits (ELK Biotechnology; Aβ₄₀: Catalog No. ELK8921, Aβ₄₂: Catalog No. ELK8920) were used according to the manufacturer's instructions. Absorbance was recorded at 450 nm using a microplate reader (Tecan Spark, Austria). Total protein concentrations were determined with an Enhanced BCA Assay Kit (Beyotime, Catalog No. P0009), and Aβ levels were normalized to total protein and expressed as picogram per microgram (pg/µg) protein.

2.18. Quantitative reverse transcription‐PCR

We extracted total RNA from cells using Trizol reagent (Accurate Biology, Shanghai, China, Catalog No. AG21101) according to the manufacturer's instructions. RNA concentrations were measured using a Nanodrop 2000 spectrophotometer (Thermo Fisher Scientific). Reverse transcription was performed using 1 µg total RNA with HiScript II Q RT SuperMix for qPCR (Vazyme, Catalog No. R323‐01) to synthesize complementary DNA (cDNA). Quantitative reverse transcription‐PCR (qRT‐PCR) was carried out on an Applied Biosystems ViiA 7 Real‐Time PCR system (Thermo Fisher Scientific) using SYBR Green Realtime PCR Master Mix (Accurate Biology, Changsha, China, Catalog No. AG11718). Primers specific to the target genes (Table S4) were used. The PCR conditions included an initial denaturation step at 95°C for 30 s, followed by 40 cycles of denaturation at 95°C for 5 s and annealing/extension at 60°C for 30 s. Relative mRNA expression levels were normalized to β‐actin as the reference gene. We calculated relative expression levels of mRNAs using the 2^{^–ΔΔCt} method.

2.19. Flow cytometry

We utilized flow cytometry to evaluate microglial polarization in HMC3 microglia and neuronal death in RA‐SH‐SY5Y neurons under triculture conditions after PS‐NP exposure. To evaluate microglial polarization, treated HMC3 microglia were collected and stained with CD206‐APC (Proteintech, Rosemont, IL, USA, Catalog No. APC‐65155) and CD86‐PE antibodies (Biogems, Westlake Village, CA, USA, Catalog No. 08911‐60). Cells were incubated in the dark at 37°C for 20 min, followed by two washes with 1 mL staining buffer through centrifugation at 1,500 rpm for 5 min. The supernatant was discarded, and the cell pellet was resuspended in 300 µL staining buffer. The samples were analyzed using a flow cytometer (Accuri C6, BD Biosciences). Data were processed and visualized using FlowJo software (version 10, Treestar). Gating was based on Side Scatter‐Height/Forward Scatter‐Height (SSC‐H/FSC‐H) patterns to identify HMC3 microglia. Dead cells were excluded by examining the dispersion along the diagonal of FL3‐H/FL3‐A, while live cells clustered in the center were retained. Microglia were further categorized as M1 (CD86⁺CD206⁻) or M2 (CD86⁻CD206⁺) subtypes (Figure S1A).

For neuronal death analysis, RA‐SH‐SY5Y neurons were stained using the FITC Annexin V Apoptosis Detection Kit with Propidium Iodide (PI) (Elabscience, Wuhan, China, Catalog No. E‐CK‐A151). After two washes with ice‐cold PBS, both adherent and suspended neurons were resuspended in staining buffer provided with the kit. Cells were incubated with 2.5 µL Annexin‐V‐FITC and 2.5 µL propidium iodide in the dark at 37°C for 15 min. After staining, 400 µL of diluted 1× binding buffer was added to resuspend the cells, and the mixture was gently agitated. Samples were analyzed using a flow cytometer (Accuri C6, BD Biosciences), and data were processed using FlowJo software. Gating was based on SSC‐H/FSC‐H patterns to include RA‐SH‐SY5Y neurons. Dead cells were gated as PI⁺ (Figure S1B).

2.20. Quantification and statistical analysis

We first tested all datasets for normality using the Shapiro–Wilk test, as appropriate. We presented normally distributed data as mean ± standard deviation (SD) and compared multiple groups using one‐way ANOVA, followed by Tukey's post hoc test for pairwise comparisons. For non‐normally distributed data, we reported medians and analyzed them using the Kruskal–Wallis H Test with Bonferroni correction for multiple comparisons. We also calculated effect sizes where relevant (Table S5). We performed all statistical analyses with SPSS 24.0 (IBM, Armonk, NY, USA) and generated figures using GraphPad Prism 8.0 (San Diego, CA, USA). We considered differences statistically significant at two‐sided p < 0.05. Detailed statistical parameters, including exact p values and effect sizes, are provided in Table S5.

2.21. Illustrations

Illustrations were generated with BioRender (www.biorender.com) under an academic license.

3. RESULTS

3.1. PS‐NP exposure aggravated AD‐related cognitive deficits in APP/PS1 mice

We assessed the effects of PS‐NPs on AD‐related cognitive deficits in APP/PS1 transgenic mice following 90 days of intragastric exposure (Figure 1A). Fluorescently labeled PS‐NPs accumulated in the hippocampus in a dose‐dependent manner, reaching 0.03 mg/g and 1.8 mg/g in the 0.25 mg/kg and 25 mg/kg groups, respectively (Figure S2A). Imaging revealed that the fluorescent particles were strongly retained in the hippocampus (Figure S2B). TEM confirmed the presence of spherical PS‐NPs (∼50 nm) in the hippocampus of the high‐dose group (Figure S2C), implicating particle deposition in AD‐related pathology.

FIGURE 1.

PS‐NP exposure exacerbated AD‐related cognitive deficits in APP/PS1 mice. (A) Experimental design. (B–D) Behavioral analysis in Y‐maze test. Representative activity trajectory of APP/PS1 mice following 90‐day PS‐NP exposure (B), target arm errors (C), and escape latency (D) across five training days (n = 5 male APP/PS1 mice/group). (E) Behavioral analysis in OFT for APP/PS1 mice after 90‐day PS‐NP exposure (n = 10 male mice/group). (F–H) Behavioral analysis in NOR test. Experimental design for NOR test, including representative activity trajectories and heatmap during testing phases of APP/PS1 mice after 90‐day PS‐NP exposure (F), discrimination index for familiar and novel objects (G), and NOR object exploration frequency (H) during testing phase (n = 7 male APP/PS1 mice/group). (I and J) MRI of mouse hippocampus. Representative MRI images of APP/PS1 mouse brains (I) and hippocampal T₂* relaxation times (J) after 90‐day PS‐NP exposure (n = 3 male APP/PS1 mice/group). (K and L) Nissl staining. Representative Nissl staining images from hippocampus of APP/PS1 mice (K) and quantification of neurons per field of view (l) after 90‐day PS‐NP exposure (n = 3 male mice/group). (M and N) ROS generation assessed by DHE staining. Representative images of ROS generation in hippocampus of APP/PS1 mice (M) and relative DHE fluorescence intensity in control group after 90‐day PS‐NP exposure (N) (n = 3 male APP/PS1 mice/group). (O–S) Aβ protein levels in APP/PS1 mice following 90‐day PS‐NP exposure. Representative IF images of Aβ protein (O), Thioflavin‐S staining of amyloid plaques (P), immunoblot of APP and Aβ proteins (Q), and ELISA quantification of Aβ₄₀ (R) and Aβ₄₂ (S) concentrations in hippocampus of APP/PS1 mice after 90‐day PS‐NP exposure (n = 8 male mice/group). We used one‐way ANOVA followed by Tukey's post hoc test for panels C, E, G, H, J, L, N, and R. Data were presented as mean ± SD. For panels D and S, we used the Kruskal–Wallis H test, with Bonferroni correction applied for multiple comparisons and data presented as median with range. We considered differences statistically significant at a two‐sided p value < 0.05. *p < 0.05 (Table S5). ANOVA, one‐way analysis of variance; DHE, dihydroethidium; IF, immunofluorescence; NOR, novel object recognition; OFT, open field test; PS‐NP, polystyrene nanoplastic; SD, standard deviation.

We evaluated cognitive function in APP/PS1 mice after PS‐NP exposure using behavioral tests. In the Y‐maze test, mice exposed to 0.25 and 25 mg/kg PS‐NPs made more errors on day 5 of training (Figure 1B,C). Compared to controls, only the 25 mg/kg PS‐NP group exhibited prolonged escape latencies on days 4 and 5 of training (Figure 1B,D). These results indicated long‐term memory impairments following PS‐NP exposure. To exclude the potential influence of activity, we conducted OFTs, which showed no differences in locomotor activity across groups (Figure 1E; Figure S2D–I). In the NOR test, both the 0.25 and 25 mg/kg PS‐NP groups had significantly reduced discrimination indices (Figure 1F,G), pointing to short‐term memory deficits, though NOR frequency remained unchanged (Figure 1H).

We then examined histological and biochemical changes in the hippocampus to understand the underlying mechanisms. MRI revealed decreases in hippocampal T₂* relaxation times, particularly in the 25 mg/kg group (Figure 1I,J). Nissl staining confirmed significant neuronal loss in the hippocampus after PS‐NP exposure (Figure 1K,L). DHE staining showed elevated levels of ROS in the hippocampus, with fluorescence intensity increasing dose‐dependently in PS‐NP‐treated groups (Figure 1 M,N). These data linked PS‐NP exposure to heightened hippocampal injury, further implicating these processes in the observed cognitive deficits.

To assess Aβ pathology following PS‐NP exposure, we performed IF staining in APP/PS1 mice. Sparse, compact Aβ deposits were observed in the 0 and 0.25 mg/kg PS‐NP groups, whereas the 25 mg/kg PS‐NP group showed widespread, diffuse aggregation with scattered plaques throughout the hippocampus (Figure 1O). Thioflavin‐S staining further distinguished plaque morphology. Mice in the 0 and 0.25 mg/kg PS‐NP groups displayed reticular, coreless plaques (Figure 1P). In the 25 mg/kg PS‐NP group, dense‐core fibrillar plaques were observed (Figure 1P), resembling the “burned‐out” morphology described in human AD brains. ⁴¹ Western blot analysis revealed no significant change in APP levels (Figures 1Q; S2J) but a notable increase in Aβ protein levels in the 25 mg/kg PS‐NP group (Figures 1Q; S2K). ELISA further indicated that PS‐NPs primarily increased Aβ₄₂ species (Figure 1R,S). Collectively, our results demonstrated that PS‐NPs aggravated AD neuropathology in APP/PS1 mice.

3.2. PS‐NPs exacerbated cognitive deficits via glial–neuronal signaling in APP/PS1 mice

To evaluate PS‐NP‐induced changes in intercellular communication, we isolated hippocampal oligodendrocytes, neurons, microglia, and astrocytes from APP/PS1 mice after 90 days of exposure and performed cell‐type‐resolved proteomic analysis. We used specific biomarkers to isolate each cell type, confirmed by proteomic data showing high expression of the selected markers (Figure 2A). Principal component analysis revealed notable variability across cell types (Figure S3A), while Pearson correlation analysis showed distinct clustering within each cell types (Figure S3B). Proteomic profiles showed dose‐dependent changes in protein expression within each cell type after PS‐NP exposure (Figure S3C). Correlation analysis further confirmed these changes, showing strong correlations within each cell type and between different exposure groups (Figure S3D).

FIGURE 2.

PS‐NPs aggravated cognitive deficits through collagen‐mediated intercellular communication in APP/PS1 mice. (A) Cell‐type annotation based on canonical marker protein expression in oligodendrocytes, neurons, microglia, and astrocytes from APP/PS1 mice after 90 days of PS‐NP exposure (n = 3 pool samples/group). (B) Mapping of cell‐to‐cell communication between oligodendrocytes, neurons, microglia, and astrocytes in APP/PS1 mice following PS‐NP exposure for 90 days (n = 3 pool samples/group). (C) Overview of intercellular communication pathways in APP/PS1 mice after 90 days of PS‐NP exposure (n = 3 pool samples/group). (D) Identification of ligand–receptor interactions associated with intercellular communication during PS‐NP exposure (n = 3 pool samples/group). (E) Visualization of collagen pathway between astrocytes and microglia and between microglia and neurons (n = 3 pool samples/group). (F and G) Colocalization of COL1A1 in astrocytes (GFAP⁺) in hippocampus of APP/PS1 mice. Representative images (F) and quantification (G) (n = 3 male mice/group). (H and I) Colocalization of COL1A1 (ligand) with ITGA1–ITGB1 (receptor) in microglia (TMEM119⁺) in hippocampus of APP/PS1 mice. Representative images (H) and quantification (i) (n = 3 male mice/group). (J and K) Colocalization of COL1A2 in microglia (TMEM119⁺) in hippocampus of APP/PS1 mice. Representative images (J) and quantification (K) (n = 3 male mice/group) (n = 3 male mice/group). (L and M) Colocalization of COL1A2 (ligand) with ITGA1–ITGB1 (receptor) in neuron (NEUN⁺) in hippocampus of APP/PS1 mice. Representative images (L) and quantification (M) (n = 3 male mice/group). Data were expressed as mean ± SD. We performed one‐way ANOVA followed by Tukey's post hoc multiple comparisons test. We considered differences statistically significant at a two‐sided p value < 0.05 (Table S5). ANOVA, analysis of variance; Ast, astrocytes; IF, immunofluorescence; Mic, microglia; Neu, neurons; Oli, Oligodendrocytes; PS‐NP, polystyrene nanoplastic; SD, standard deviation.

We conducted GO analysis to identify the biological processes linked to the proteins altered by PS‐NP exposure. The results matched the phenotypic changes observed in APP/PS1 mice following PS‐NP exposure. Astrocytes showed oxidative stress emerged (Figure S4A), aligning with the elevated ROS levels detected earlier. In microglia, proteins related to learning were identified (Figure S4B), consistent with the cognitive deficits seen in behavioral tests. Neurons showed proteins tied to cell death induced by oxidative stress (Figure S4C), echoing the neuronal loss observed in histology. Oligodendrocytes displayed proteins involved in cognition and oxidative stress response (Figure S4D), further linking PS‐NP exposure to exacerbated cognitive deficits in AD. These findings confirmed that PS‐NPs disrupt key cellular processes.

We used CellChat to investigate how PS‐NP exposure affects cell‐to‐cell communication. The results revealed a dose‐dependent increases in interaction strength between astrocytes and microglia, as well as between microglia and neurons, following PS‐NP exposure (Figure 2B). Conversely, oligodendrocytes exhibited a dose‐dependent decline in communication with microglia and astrocytes following PS‐NP exposure (Figure 2B). Further signaling analysis revealed distinct changes across cell type. Oligodendrocytes exhibited decreased incoming signaling, while astrocytes and neurons showed increased incoming signaling following PS‐NP exposure (Figure S4E). Microglia acted as key mediators, with significantly elevated outgoing and incoming signaling in response to PS‐NP exposure (Figure S4E). With neuronal loss observed in the hippocampus of APP/PS1 mice after PS‐NP exposure, we identified neurons as the final signaling endpoint. Only signals from microglia increased in a dose‐dependent manner among those received by neurons (Figure 2B). Microglia also received signals from astrocytes in a dose‐dependent manner (Figure 2B). Based on this communication network, we proposed that PS‐NPs enhanced cell‐to‐cell communication, propagating signals from astrocytes to microglia and ultimately to neurons.

To explore global signaling changes due to PS‐NP exposure, we compared information flow between control and exposed groups. We observed dose‐dependent increases in communication across key pathways, including laminin, collagen, agrin (AGRN), contactin (CNTN), occludin (OCLN), junctional adhesion molecule (JAM), and neurexin (NRXN), among astrocytes, microglia, and neurons (Figure 2C). Notably, only the laminin and collagen pathways formed complete cascades, with upregulated ligand–receptor pairs across all three cell types (Figure 2D). These pathways mediate extracellular matrix (ECM)‐receptor signaling, a critical mechanism for neural cell communication. ⁴² , ⁴³ While the laminin pathway showed decreased communication at 0.25 mg/kg PS‐NP exposure (Figure S4F), the collagen pathway exhibited a dose‐dependent increase, particularly between astrocytes and microglia, and microglia and neurons, matching the dose‐dependent communication changes between these cell types (Figure 2B,C,E). This suggested that PS‐NPs amplified signaling through the collagen pathway, where astrocytes served as signal senders, neurons as receivers, and microglia as both senders and receivers (Figure 2E). Dose‐dependent ligand–receptor interactions within the collagen pathway further supported this model. COL6A1‐ and COL1A1‐(ITGA1/ITGB1) interactions from astrocytes to microglia, and COL1A2‐(ITGA1/ITGB1) interactions from microglia to neurons, increased with exposure (Figure 2D). Notably, COL1A1 levels were upregulated in astrocytes at 25 mg/kg PS‐NP exposure (Figure S4G), while COL6A1 levels remained unchanged (Figure S4G). These results suggested that PS‐NPs enhanced communication through the collagen pathway, where COL1A1 from astrocytes interacts with ITGA1/ITGB1 receptors on microglia and COL1A2 from microglia binds to the same receptors on neurons.

To validate these ligand–receptor interactions, we conducted co‐expression IF analysis in the hippocampus of APP/PS1 mice. Receptors mediate their function by binding to ligands, thereby enhancing cell communication. Labeling astrocytes (GFAP⁺), microglia (TMEM119⁺), and neurons (NEUN⁺), we observed a marked increase in COL1A1 expression in astrocytes following 25 mg/kg PS‐NP exposure (Figure 2F,G). This suggested that PS‐NP exposure enhanced COL1A1 production in astrocytes. Similarly, we observed a significant increase in the number of COL1A1⁺ microglia (TMEM119⁺) expressing the ITGA1/ITGB1 receptor (Figure 2H,I), indicating elevated interactions between COL1A1 from astrocytes and the ITGA1/ITGB1 receptor in microglia. Microglia also exhibited a significant increase in COL1A2 expression (Figure 2J,K), with colocalization of COL1A2 and ITGA1/ITGB1 in neurons (NEUN⁺) being significantly higher in PS‐NP‐treated animals (Figure 2L,M).

To determine whether these alterations were specific to AD‐related cognitive impairment, we re‐analyzed single‐cell transcriptomic from wild‐type C57BL/6J mice following 28 days of 250 mg/kg PS‐NP exposure. ¹⁵ In the absence of genetic susceptibility, PS‐NP exposure did not significantly alter the expression of genes associated with the collagen pathway (Figure S5), indicating that activation of this signaling axis was restricted to the AD context. These findings supported a model in which PS‐NPs exacerbated cognitive deficits in AD by potentiating glial–neuronal communication through the collagen pathway.

3.3. PS‐NPs potentiated collagen signaling to drive microglial activation and neuronal death in in vitro AD models

To investigate the in vivo mechanisms suggested by hippocampal proteomics and cell–cell communication analyses, we established mono‐ and triculture systems simulating brain cell interactions under PS‐NP exposure. Among these, RA‐SH‐SY5Y were used as a neuron‐like cell model known to express key AD‐related proteins, including Aβ, thereby serving as an in vitro AD neuronal model. ²⁶ , ²⁷ , ⁴⁴ , ⁴⁵ In monocultures of SVG p12 astrocytes, HMC3 microglia, and RA‐SH‐SY5Y neuron‐like cells, PS‐NPs minimally affected ligand and receptor expression (Figure S6A). Ligand mRNA levels COL1A1 in SVG p12 astrocytes were significantly reduced (Figure S6B), while COL1A2 levels in microglia remained unchanged after PS‐NP treatment (Figure S6C). For receptors, PS‐NPs increased ITGA1 expression in HMC3 microglia but decreased ITGA1 levels in RA‐SH‐SY5Y neurons, while ITGB1 expression remained stable across all cell types (Figure S6D,E). At the protein level, PS‐NPs did not alter COL1A1 levels in SVG p12 astrocytes (Figure S6F,G). CO‐IP analysis confirmed that PS‐NPs did not affect ligand–receptor interactions between COL1A1‐ITGA1/ITGB1 in HMC3 microglia (Figure S6H‐J) and COL1A2‐ITGA1/ITGB1 in RA‐SH‐SY5Y neurons (Figure S6K‐M). Together, these results suggested that PS‐NPs exerted limited direct effects on ECM signaling components in isolated cell types. This pointed to a context‐dependent mechanism, likely requiring multicellular interactions to trigger pathological collagen signaling and downstream neurotoxicity.

The collagen pathway facilitates intercellular communication through ligand–receptor interactions between ECM proteins and integrins, functioning similarly to cytokines. ⁴⁶ , ⁴⁷ To model the ECM‐mediated communication axis identified in our in vivo cell–cell interaction analyses following PS‐NP exposure, we established a triculture transwell system incorporating astrocytes, microglia, and neurons (Figure 3A). Unlike in monoculture, PS‐NP (50 µg/mL) notably enhanced COL1A1 expression in SVG p12 astrocytes, increasing the ligand source (Figure 3B,C). This indicated that PS‐NP exposure triggered multicellular crosstalk that specifically drove COL1A1 induction in astrocytes, rather than acting directly on individual cells. CO‐IP analysis showed a dose‐dependent enhancement of COL1A1 binding to ITGB1 on HMC3 microglia, achieving statistical significance, while binding to ITGA1 trended upward without reaching significance (Figure 3B,D,E). In HMC3 microglia input samples, PS‐NPs did not alter COL1A1, ITGA1, or ITGB1 expression levels (Figures 3B,F; S7A,B). IgG controls confirmed the specificity of these interactions (Figure 3B). These results demonstrated that PS‐NPs enhanced astrocyte–microglia interactions by promoting astrocyte COL1A1 binding to integrins on microglia. We next evaluated how these interactions affected microglial behavior. Collagen–integrin binding activates microglial signaling, leading to phenotypic changes. ⁴⁸ , ⁴⁹ Consistently, PS‐NP treatment in the triculture system polarized HMC3 microglia toward a pro‐inflammatory M1 phenotype, as shown by an increased CD86/CD206 ratio (Figure 3H,I). PS‐NPs also upregulated M1 markers, such as NF‐κB1 and NF‐κB2, in APP/PS1 mice (Figure S8A). These findings highlighted that PS‐NP drove astrocyte–microglia interactions and induced microglial activation through the collagen pathway.

FIGURE 3.

PS‐NPs enhanced the collagen signaling pathway, promoting microglial activation and neuronal death. (A) Experimental design. (B–G) Assessment interactions in SVG p12 astrocytes and HMC3 microglia within triculture system under PS‐NP treatment using CO‐IP. Representative immunoblot (B). Quantification of COL1A1 in SVG p12 astrocytes (C). CO‐IP analysis of COL1A1 binding to ITGA1 (D) and ITGB1 (E) in HMC3 microglia. Quantification of COL1A1 (input) (F) and COL1A2 (input) (G) in HMC3 microglia (n = 3 independent experiments/group). (H and I) Microglial activation in triculture system following PS‐NP treatment analyzed by flow cytometry. Expression of CD206 (anti‐inflammatory marker) and CD86 (pro‐inflammatory marker) is shown in representative flow cytometry plots (H) and quantification (I) (n = 3 independent experiments/group). (J–Q) Evaluation of COL1A2 expression, ligand–receptor interactions, and downstream signaling in RA‐SH‐SY5Y neurons within the triculture system. Representative immunoblot (J). CO‐IP analysis of COL1A2 binding to ITGA (K) and ITGB (L) in RA‐SH‐SY5Y neurons, with quantification of COL1A2 (input) (M). ELISA quantification of Aβ₄₀ (N) and Aβ₄₂ (O) concentrations in RA‐SH‐SY5Y neurons. Immunoblot analysis quantifying of total p38 MAPK (P) and phosphorylated p38 MAPK (Q) (n = 3 independent experiments/group). (R and S) Analysis of PS‐NP‐induced neuronal death in RA‐SH‐SY5Y neurons within triculture system by flow cytometry. Representative flow cytometry plots (R) and quantification (S) of cell death are shown (n = 3 independent experiments/group). We used one‐way ANOVA followed by Tukey's post hoc test for panels C to G, I, K to M, O, P, and Q. Data were presented as mean ± SD. For panels N and S, we used the Kruskal–Wallis H test, with Bonferroni correction applied for multiple comparisons and data presented as median with range. We considered differences statistically significant at a two‐sided p‐value < 0.05 (Table S5). ANOVA, analysis of variance; CO‐IP, co‐immunoprecipitation; RA, retinoic acid; PS‐NP, polystyrene nanoplastic; SD, standard deviation.

We further investigated how PS‐NPs affected microglia–neuron communication in the triculture system. PS‐NPs significantly increased COL1A2 levels in HMC3 microglia, identifying them as the ligand source (Figure 3B,G). CO‐IP analysis showed dose‐dependent enhancement of COL1A2 binding to ITGB1 on RA‐SH‐SY5Y neurons, achieving statistical significance, while binding to ITGA1 trended upward without significance (Figure 3J–L). PS‐NPs did not affect ITGA1 or ITGB1 expression in RA‐SH‐SY5Y neuron input samples (Figure S7C,D). The increase in COL1A2 levels in RA‐SH‐SY5Y neurons following PS‐NP treatment suggested enhanced ligand reception from HMC3 microglia (Figure 3J,M). IgG controls confirmed the specificity of these interactions (Figure 3J). PS‐NPs also increased Aβ accumulation (Figures 3J; S7E), particularly Aβ₄₂ in RA‐SH‐SY5Y (Figure 3N,O). Integrin activation, in concert with Aβ, is known to trigger mitogen‐activated protein kinase (MAPK) signaling and contribute to neuronal death. ⁵⁰ We assessed MAPK activation in neurons and found that PS‐NPs reinforced the collagen–integrin axis, facilitating p38 MAPK phosphorylation in RA‐SH‐SY5Y neurons (Figures 3J,P,Q; S7F–I). Similar p38 activation occurred in the hippocampus of PS‐NP‐exposed APP/PS1 mice (Figure S8B). Flow cytometry confirmed that PS‐NP‐induced p38 MAPK activation in RA‐SH‐SY5Y neurons led to cell death in the triculture system (Figure 3R,S). Together, these findings demonstrated that PS‐NPs reinforced the collagen–integrin interaction, driving Aβ accumulation, p38 MAPK activation, and neuronal death, thereby exacerbating cognitive impairment in AD.

3.4. Disruption of collagen pathway mitigated PS‐NP‐induced microglial activation and neuronal death

To investigate how the collagen signaling pathway regulated astrocytes, microglia, and neurons under PS‐NP exposure, we targeted its ligand and receptor components. First, we disrupted the ligand source by transfecting SVG p12 astrocytes with si‐COL1A1 and integrating them into a triculture system (Figure 4A). This transfection reduced COL1A1 expression in SVG p12 astrocytes, even during PS‐NP treatment (Figure 4B,C). CO‐IP analysis revealed reduced binding of astrocyte‐derived COL1A1 to ITGA1/ITGB1 integrins on HMC3 microglia in the si‐COL1A1 + PS‐NP group compared to the scramble + PS‐NP group (Figure 4B,D,E). HMC3 microglia input samples showed no changes in COL1A1, ITGA1, or ITGB1 expression (Figure 4B,F; Figure S7J,K). These results demonstrated that reducing COL1A1 levels in astrocytes limited its availability as a ligand for binding to ITGA1/ITGB1 integrins on HMC3 microglia during PS‐NP treatment. This restriction effectively suppressed PS‐NP‐induced polarization of HMC3 microglia toward the pro‐inflammatory M1 phenotype, as demonstrated by a decreased CD86/CD206 ratio in HMC3 microglia within the triculture system (Figure 4O,P). We then examined downstream communication between microglia and neurons in the triculture system. Si‐COL1A1 transfection in SVG p12 astrocytes reduced COL1A2 levels in HMC3 microglia (Figure 4B,G). Consequently, less COL1A2 bound to ITGA1/ITGB1 integrins on RA‐SH‐SY5Y neurons during PS‐NP exposure (Figure 4B,H,I). This reduction in binding led to decreased COL1A2 levels in neurons, reflecting impaired ligand transfer from microglia (Figure 4B,J). ITGA1 and ITGB1 expression in the RA‐SH‐SY5Y neuron input samples remained stable (Figure S7L,M). Importantly, Aβ accumulation, particularly Aβ₄₂, persisted in neurons regardless of COL1A1 silencing (Figures 4B,K,L; S7N). However, p38 MAPK activation (Figures 4B,M,N) and neuronal death (Figure 4Q,R) were significantly attenuated, indicating that Aβ alone was insufficient to induce these effects under PS‐NP exposure. Thus, under PS‐NP exposure, collagen signaling was essential to amplify Aβ neurotoxicity.

FIGURE 4.

Si‐COL1A1 treatment in SVG p12 astrocytes mitigated PS‐NP‐induced microglial activation and neuronal death in triculture system via collagen signaling pathway. (A) Experimental design. (B–N) Investigation of collagen signaling components and downstream effects in astrocytes, microglia, and neurons under PS‐NP exposure in triculture system following si‐COL1A1 treatment of SVG p12 astrocytes. Representative immunoblot (B). Quantification of COL1A1 in SVG p12 astrocytes (c). CO‐IP analysis of COL1A1 binding to ITGA (D) and ITGB (E) in HMC3 microglia. Quantification of COL1A1 (input) (F) and COL1A2 (input) (G) in HMC3 microglia. CO‐IP analysis of COL1A2 binding to ITGA (H) and ITGB (I) in RA‐SH‐SY5Y neurons, with quantification of COL1A2 (input) (J). ELISA quantification of Aβ₄₀ (K) and Aβ₄₂ (L) concentrations in RA‐SH‐SY5Y neurons. Immunoblot analysis quantifying total p38 MAPK (M) and phosphorylated p38 MAPK (N) in RA‐SH‐SY5Y neurons (n = 3 independent experiments/group). (O and P) Microglial activation assessed by flow cytometry in triculture system following si‐COL1A1‐treated SVG p12 astrocytes and PS‐NP exposure. Representative flow cytometry plots showing expression of CD206 (anti‐inflammatory marker) and CD86 (pro‐inflammatory marker) (O) and their quantification (P) (n = 3 independent experiments/group). (Q and R) Evaluation of neuronal death in RA‐SH‐SY5Y neurons within triculture system under PS‐NP exposure after si‐COL1A1 treatment in SVG p12 astrocytes. Representative flow cytometry plots (Q) and quantification of cell death (R) are shown (n = 3 independent experiments/group). We used one‐way ANOVA followed by Tukey's post hoc test for panels C to E, G to N, and P. Data are presented as mean ± SD. For panels F and R, we used the Kruskal–Wallis H test, with Bonferroni correction applied for multiple comparisons and data presented as median with range. We considered differences statistically significant at a two‐sided p value < 0.05 (Table S5). ANOVA, analysis of variance; CO‐IP, co‐immunoprecipitation; RA, retinoic acid; PS‐NP, polystyrene nanoplastic; SD, standard deviation.

Next, we targeted the receptor by using TC‐I 15, an integrin inhibitor, to block ITGA1/ITGB1 signaling (Figure 5A). In the triculture system, PS‐NP exposure still increased COL1A1 in SVG p12 astrocytes (Figure 5B,C), indicating that feedback within the multicellular network continued to influence astrocytes even when ITGA1/ITGB1 signaling was blocked. TC‐I 15 effectively blocked COL1A1 binding to ITGA1/ITGB1 receptors on HMC3 microglia (Figure 5B,D–F), thereby reducing ligand transfer from astrocytes to microglia. This inhibition suppressed PS‐NP‐induced microglial polarization toward the M1 phenotype, as shown by a lower CD86/CD206 ratio in HMC3 microglia (Figure 5O,P). HMC3 microglia input samples showed no changes in COL1A1, ITGA1, or ITGB1 expression (Figure 5B; Figure S7O, P). TC‐I 15 also lowered COL1A2 levels in HMC3 microglia (Figure 5B,G) and reduced its binding to ITGA1/ITGB1 integrins on RA‐SH‐SY5Y neurons, relative to the scramble + PS‐NP group (Figure 5B,H,I). This led to lower COL1A2 levels in RA‐SH‐SY5Y neurons (Figure 5B,J). Like si‐COL1A1, TC‐I 15 did not significantly alter ITGA1 or ITGB1 expression in HMC3 or RA‐SH‐SY5Y input samples (Figure S7Q,R). Although Aβ accumulation, particularly Aβ₄₂, in neurons persisted (Figure 5B,K,L; Figure S7S), TC‐I 15 significantly reduced p38 MAPK activation, as reflected by decreased total and phosphorylated p38 levels (Figure 5B,M,N). These molecular changes correlated with improved neuronal viability under PS‐NP exposure (Figure 5Q,R).

FIGURE 5.

Blockade of ITGA1/ITGB1‐mediated collagen signaling in microglia using TC‐I 15 mitigated PS‐NP‐induced microglial activation and neuronal death in the triculture system. (A) Experimental design. (B–N) Analysis of collagen signaling components and downstream effects in astrocytes, microglia, and neurons under PS‐NP exposure following TC‐I 15 treatment in HMC3 microglia. Representative immunoblot (B). Quantification of COL1A1 in SVG p12 astrocytes (C). CO‐IP analysis of COL1A1 binding to ITGA (D) and ITGB (E) in HMC3 microglia. Quantification of COL1A1 (input) (F) and COL1A2 (input) (G) in HMC3 microglia. CO‐IP analysis of COL1A2 binding to ITGA (H) and ITGB (I) in RA‐SH‐SY5Y neurons, with quantification of COL1A2 (input) (j). ELISA quantification of Aβ₄₀ (K) and Aβ₄₂ (L) concentrations in RA‐SH‐SY5Y neurons. Immunoblot analysis quantifying total p38 MAPK (M) and phosphorylated p38 MAPK (N) in RA‐SH‐SY5Y neurons (n = 3 independent experiments/group). (O and P) Microglial activation assessed by flow cytometry in triculture system following TC‐I 15 treatment of HMC3 microglia and PS‐NP exposure. Expression of CD206 (anti‐inflammatory marker) and CD86 (pro‐inflammatory marker) was measured in representative flow cytometry plots (O) and quantified (P) (n = 3 independent experiments/group). (Q and R) Evaluation of PS‐NP‐induced neuronal death in RA‐SH‐SY5Y neurons within triculture system after TC‐I 15 treatment in HMC3 microglia. Representative flow cytometry plots (Q) and quantification of cell death (R) are shown (n = 2‐3 independent experiments/group). Data are expressed as mean ± SD. We performed one‐way ANOVA followed by Tukey's post hoc multiple comparisons test. We considered differences statistically significant at a two‐sided p value < 0.05 (Table S5). ANOVA, analysis of variance; CO‐IP, co‐immunoprecipitation; RA, retinoic acid; PS‐NP, polystyrene nanoplastic; SD, standard deviation.

Together, these results demonstrate that PS‐NPs act as a central pathological factor in AD‐like pathology, potentiating both Aβ accumulation and integrin‐mediated signaling. Neuronal death occurred only when these pathways converged. Blocking the collagen–integrin axis uncouples Aβ from downstream toxic signaling, mitigating PS‐NP‐induced neuronal damage in this in vitro AD mode.

3.5. Blocking the collagen pathway with TC‐I 15 alleviated PS‐NP‐exacerbated cognitive impairment in APP/PS1 mice

To validate the role of the collagen pathway in PS‐NP‐exacerbated cognitive deficits, APP/PS1 mice were exposed to PS‐NPs via intragastric administration for 90 days and treated intraperitoneally with TC‐I 15 every other day (Figure 6A). At the conclusion of the experiment, we used UHPLC‐HRMS to detect TC‐I 15 in the hippocampus. We compared the base peak chromatogram spectra of the control, PS‐NPs + TC‐I 15, and TC‐I 15 standard groups, revealing clear differences between the control and PS‐NPs + TC‐I 15 groups (Figure S9A,B). Notably, the peak position in the PS‐NPs + TC‐I 15 group closely matched that of the standard TC‐I 15 sample (Figure S9A,B), confirming the presence of TC‐I 15 in the hippocampus. To further validate these findings, we analyzed the ion chromatogram at a retention time of 10.23 min. The extracted ion chromatograms (XIC) and secondary spectra for the PS‐NPs + TC‐I 15 group were consistent with the standard sample, reinforcing the presence of TC‐I 15 in the hippocampus (Figure S9C,D). Additionally, XIC analysis demonstrated a remarkable 90‐fold increase in TC‐I 15 levels in the PS‐NPs + TC‐I 15 group compared to the control group (Figure S9E), confirming that TC‐I 15 successfully reached the hippocampus following intraperitoneal administration.

FIGURE 6.

TC‐I 15 mitigated PS‐NP‐exacerbated AD‐related cognitive deficits in APP/PS1 mice. (A) Experimental design. (B–D) Behavioral analysis in Y‐maze test. Representative activity trajectories of APP/PS1 mice exposed to PS‐NPs for 90 days with TC‐I 15 treatment (B). Target arm errors (C) and escape latency (D) during the five training days (n = 5 male APP/PS1 mice/group). (E–G) Behavioral analysis in NOR test. Experimental design for NOR test, including representative activity trajectories and heatmap during testing phases of APP/PS1 mice after 90‐day PS‐NP exposure with TC‐I 15 treatment (E). Discrimination index for familiar and novel objects (F) and NOR object exploration frequency (G) during testing phase (n = 7 male APP/PS1 mice/group). (H and I) MRI of mouse hippocampus. Representative MRI images of APP/PS1 mouse brains (H) and hippocampal T₂* relaxation times (I) after 90‐day PS‐NP exposure with TC‐I 15 treatment (n = 3 male APP/PS1 mice/group). (J and K) Nissl staining. Representative Nissl staining images from hippocampus of APP/PS1 mice (J) and quantification of neurons per field of view (K) after 90‐day PS‐NP exposure with TC‐I 15 treatment (n = 3 male mice/group). (L and M) ROS generation assessed by DHE staining. Representative images of ROS generation in hippocampus of APP/PS1 mice (L) and relative DHE fluorescence intensity in control group after 90‐day PS‐NP exposure with TC‐I 15 treatment (M) (n = 3 male APP/PS1 mice/group). (N–T) Aβ protein levels in APP/PS1 mice following 90‐day PS‐NP exposure with TC‐I 15 treatment. Representative immunofluorescence images of Aβ protein (N), Thioflavin‐S staining of amyloid plaques (O), quantification of APP (P) and Aβ (Q) protein levels, immunoblot of APP and Aβ proteins (R), and ELISA quantification of Aβ₄₀ (S) and Aβ₄₂ (T) concentrations in hippocampus of APP/PS1 mice after 90‐day PS‐NP exposure during TC‐I 15 treatment (n = 5 male mice/group). We used one‐way ANOVA followed by Tukey's post hoc test for panels C, F to G, I, K, M, Q, S, and T. Data are presented as mean ± SD. For panel D and P, we used the Kruskal–Wallis H test, with Bonferroni correction applied for multiple comparisons and data presented as median with range. We considered differences statistically significant at a two‐sided p value < 0.05. *p < 0.05, **p < 0.01 (Table S5). ANOVA, one‐way analysis of variance; MRI, magnetic resonance imaging; NOR, novel object recognition; OFT, open field test; PS‐NP, polystyrene nanoplastic; SD, standard deviation.

Behavioral analyses showed that TC‐I 15 treatment improved cognitive function in APP/PS1 mice exposed to PS‐NPs. In the Y‐maze test, TC‐I 15‐treated mice made fewer errors and had shorter escape latencies after 5 days of training, indicating preserved long‐term memory compared to PS‐NP‐treated mice (Figure 6B–D). Similarly, in the NOR test, TC‐I 15‐treated mice exposed to PS‐NPs showed recovery in both NOR frequency (%) and discrimination index, while the PS‐NP group experienced a marked decline in these measures (Figure 6E–G).

MRI assessments revealed that TC‐I 15‐treated mice exposed to PS‐NPs exhibited rescued hippocampal T₂* relaxation times, in contrast to the PS‐NP group (Figure 6H,I). Histological examination showed preserved Nissl bodies and reduced neuronal loss in the hippocampus following TC‐I 15 treatment during PS‐NP exposure (Figure 6J,K). DHE staining demonstrated a slight reduction in ROS levels in the hippocampus of TC‐I 15‐treated mice exposed to PS‐NPs, compared to the PS‐NP group (Figure 6L,M). In contrast, TC‐I 15 had no effect on Aβ pathology. Both Aβ immunostaining and Thioflavin‐S staining revealed persistent dense‐core fibrillar plaques in the hippocampus of PS‐NP‐exposed mice, regardless of TC‐I 15 treatment (Figure 6N,O). Immunoblot and ELISA analyses confirmed that PS‐NP exposure elevated Aβ levels, particularly Aβ₄₂, while APP expression remained unchanged across all groups (Figure 6P–T).

To validate the mechanism, we assessed ligand–receptor interactions in the hippocampus of APP/PS1 mice using IF. In astrocytes (GFAP⁺), COL1A1 expression remained elevated in the PS‐NPs + TC‐I 15 group (Figure 7A,B), indicating that PS‐NPs maintained their effect on collagen availability within the multicellular network. However, as an integrin inhibitor, TC‐I 15 effectively blocked the interaction between COL1A1 ligands and the ITGA1/ITGB1 receptor. This was evidenced by a marked reduction in the number of COL1A1⁺ and ITGA1⁺‐ITGB1⁺ microglia (TMEM119⁺) in the PS‐NPs + TC‐I 15 group, compared to the PS‐NP group (Figure 7C,D). Moreover, the downstream effect of COL1A2 binding to microglia was attenuated after TC‐I 15 treatment during PS‐NP exposure (Figure 7E,F). Similarly, PS‐NP‐induced COL1A2 binding to ITGA1/ITGB1 in neurons (NEUN⁺) was significantly reversed in the PS‐NPs + TC‐I 15 group (Figure 7G,H).

FIGURE 7.

TC‐I 15 reversed the PS‐NP‐mediated amplification of the collagen–integrin pathway in APP/PS1 mice. (A and B) Colocalization of COL1A1 in astrocytes (GFAP⁺) in the hippocampus of APP/PS1 mice following TC‐I 15 treatment during PS‐NP exposure. Representative images (A) and quantification (B) (n = 3 male mice/group). (C and D) Colocalization of COL1A1 (ligand) with ITGA1/ITGB1 (receptor) in microglia (TMEM119⁺) in hippocampus of APP/PS1 mice following TC‐I 15 treatment during PS‐NP exposure. Representative images (C) and quantification (D) (n = 3 male mice/group). (E and F) Colocalization of COL1A2 in microglia (TMEM119⁺) in hippocampus of APP/PS1 mice following TC‐I 15 treatment during PS‐NP exposure. Representative images (E) and quantification (F) (n = 3 male mice/group). (G and H) Colocalization of COL1A2 (ligand) with ITGA1/ITGB1 (receptor) in neuron (NEUN⁺) in hippocampus of APP/PS1 mice following TC‐I 15 treatment during PS‐NP exposure. Representative images (G) and quantification (H) (n = 3 male mice/group). Data are expressed as mean ± SD. We performed one‐way ANOVA followed by Tukey's post hoc multiple comparisons test. We considered differences statistically significant at a two‐sided p value < 0.05 (Table S5). ANOVA, analysis of variance; IF, immunofluorescence; PS‐NP, polystyrene nanoplastic; SD, standard deviation.

Together, these findings demonstrated that integrin inhibition by TC‐I 15 mitigated the cognitive deficits exacerbated by PS‐NP exposure. This protection occurred independently of Aβ clearance and underscored the pivotal role of the collagen pathway in mediating PS‐NP‐exacerbated cognitive impairment in AD.

3.6. Enhanced astrocyte, microglia, and neuron communication via collagen signaling pathway in AD patients

To investigate the involvement of the collagen signaling pathway in AD, we analyzed snRNA‐seq data from AD patients and healthy controls (HCs) in the GSE188545 dataset from Gene Expression Omnibus (GEO). ³¹ For consistency with our previously discussed research, we classified the cells into four key types: astrocytes (AQP4, GFAP), microglia (CSF1R, CD74), neurons (NRGN, RBFOX3), and oligodendrocytes, based on gene marker expression (Figure 8A,B). These cell types were consistently identified in both AD and HC samples, although with notable differences in their relative proportions (Figure 8C). To delve deeper, we next examined intercellular communication by comparing the interactions between AD and HC samples. Strikingly, we observed significant enhancements in the communication networks of AD patients, particularly between neurons and microglia and between astrocytes and microglia (Figure 8D). Specifically, we found increased signaling from microglia to neurons and from astrocytes to microglia in the AD cohort, suggesting significant alterations in cellular crosstalk in the AD brain (Figure 8D). Furthermore, we observed a pronounced upregulation of the collagen signaling pathway in AD patients compared to HCs (Figure 8E), corroborating our findings from both in vivo and in vitro studies. Within the AD cohort, this pathway was critical for facilitating enhanced communication, especially between astrocytes and microglia and between microglia and neurons (Figure 8F). In stark contrast, these interactions were either absent or significantly weaker in HC samples (Figure 8F). In this network, astrocytes predominantly functioned as signal senders, neurons as receivers, and microglia as both senders and receivers (Figure 8F). These results underscored the substantial enhancement of collagen signaling pathway‐mediated intercellular communication in AD patients. Moreover, they reinforced our in vivo and in vitro findings, suggesting that the amplification of this pathway played a crucial role in driving AD progression, particularly under the influence of PS‐NP exposure.

FIGURE 8.

Enhanced astrocyte, microglia, and neuron communication via collagen signaling pathway in AD patients. (A) UMAP plot showing clusters of annotated cells. (B) Cell‐type annotation based on canonical marker gene expression. (C) HC or AD patients’ origin (n = 6 samples/group). (D) Cell‐to‐cell communication analysis between oligodendrocytes, neurons, microglia, and astrocytes in HC and AD patient samples (n = 6 samples/group). (E) Overview of intercellular communication pathways in HC and AD patients (n = 6 samples/group). (F) Visualization of collagen pathway between astrocytes and microglia and between microglia and neurons in HC human beings and AD patients (n = 6 samples/group). (G) MNPs enhanced neural cell crosstalk via collagen signaling pathway in AD. Astrocyte‐derived COL1A1 bound to ITGA1/ITGB1 integrins on microglia, promoting microglial activation. In turn, microglia‐derived COL1A2 interacted with the same integrins on neurons, facilitating Aβ‐induced MAPK activation. This cascade contributed to neuronal dysfunction and death. By driving astrocyte–microglia and microglia–neuron interactions, MNPs exacerbate neurodegeneration and aggravate cognitive impairment in AD. AD, Alzheimer's disease; Ast, astrocytes; ECM, extracellular matrix; HC, healthy control; MAPK, mitogen‐activated protein kinase; Mic, microglia; MNPs, micro‐ and nanoplastics; Neu, neurons; Oli, Oligodendrocytes; UMAP, uniform manifold approximation and projection.

4. DISCUSSION

This study revealed a mechanistic linked between MNP exposure and AD‐related cognitive impairment. In APP/PS1 mice, PS‐NP exposure caused cognitive decline, hippocampal injury, and altered Aβ pathology. Proteomic profiling of hippocampal cell types, coupled with intercellular signaling analysis, identified aberrant neuroglial crosstalk mediated by upregulated collagen signaling as a central pathological driver (Figure 8G). Cross‐species comparison with human AD snRNA‐seq data confirmed conserved activation of this pathway, supporting its clinical relevance. Together, these findings provide a causal framework in which MNP exposure amplifies pathological intercellular signaling and worsens cognitive deficits in AD.

Previous studies linked MNP accumulation to AD progression, ⁵¹ , ⁵² and MNPs are detected in human brains with higher burdens in individuals with cognitive impairment. ¹³ However, these associations lack causal resolution. To place our dosing strategy in context, we examined human exposure estimates. Humans ingest an estimated 0.1 to 5 g of MPs per week, equivalent to roughly 0.2 to 10 mg/kg/day (assuming an adult BW of 70 kg). ⁵³ , ⁵⁴ Indoor dust provides an additional source and can contribute 4,000 to 150,000 ng/kg/day. ⁵⁵ This value reflects polyethylene terephthalate alone, not the full range of environmental MPs, and thus likely underestimates total exposure. NP intake may further increase total daily exposure. ⁵¹ Plastic pollution continues to escalate and is expected to increase future human intake of plastics. ⁵¹ These data show that human exposure spans a wide range and that high‐end scenarios are plausible. Based on this, we used 0.25 mg/kg to model low‐level exposure and 25 mg/kg as a mechanistic dose. This approach followed standard toxicological practice for emerging contaminants. The higher dose helped reveal pathways when long‐term kinetics are uncertain and define hazard boundaries. The dose range also reflected possible real‐world conditions when cumulative intake, chronic retention, and nanoscale particles were considered. Notably, hippocampal PS‐NP accumulation in our study (0.03 to 1.8 mg/g) remained one order of magnitude below levels reported in human AD brains (26 mg/g). ¹³ Even low exposure triggered pathological changes. PS‐NPs increased Aβ plaque burden, primarily via the neurotoxic Aβ₄₂ isoform, ⁵⁶ while APP levels remained stable. This implied that PS‐NPs influenced Aβ processing rather than precursor expression. To reduce variability and detect PS‐NP effects more reliably, we used only male APP/PS1 mice. Female mice showed higher baseline Aβ deposition and stronger glial activation, ⁵⁷ , ⁵⁸ , ⁵⁹ which could mask subtle environmental influences. ⁶⁰ This choice improved consistency but limited generalizability. Future studies that include both sexes are needed to determine whether sex modulated susceptibility to PS‐NP exposure and to support a more complete risk assessment.

AD is increasingly viewed as a disorder of disrupted intercellular communication. ⁶¹ , ⁶² Beyond Aβ and tau, dysregulated glia–neuron signaling drives synaptic dysfunction, impaired Aβ clearance, and inflammation. ⁶³ Aβ does not directly induce neuronal death ⁶⁴ but amplifies maladaptive crosstalk that accelerates degeneration. ⁶⁵ We found that PS‐NP exposure intensified this signaling dose‐dependently along astrocyte–microglia–neuron pathways, causing marked neuronal loss and worsened cognitive deficits, consistent with reduced hippocampal neuronal density in APP/PS1 mice. Oligodendrocyte interactions declined under PS‐NP exposure. Although they did not directly signal to neurons and were not the focus here, this reduction might contribute to later demyelination or cognitive decline, ⁶⁶ warranting further study. While peripheral mechanisms, including immune activation and gut–brain interactions, ¹⁴ , ²¹ might contribute to neurotoxicity. More importantly, PS‐NPs were detected directly within the hippocampus, colocalizing with aberrant glial–neuronal signaling and neuronal loss. This provided strong anatomical evidence for a direct causal link. Given this spatial specificity and the hippocampus's dominant pathological response, we focused our mechanistic analysis on this region to capture the most immediate and interpretable effects of MNP exposure on AD‐related cognitive impairment.

Among the signaling changes induced by PS‐NPs, the collagen pathway emerged as a central mediator of intercellular dysfunction linked to AD‐related cognitive decline. This pathway, known to regulate critical AD‐related processes, ⁶⁷ was markedly upregulated in our animal study, consistent with its established role in AD. A similar increase appeared in human AD brains, supporting the translational relevance of our findings. Under normal conditions, collagen signaling remained largely silent. In wild‐type C57BL/6J mice lacking AD‐predisposing mutations, PS‐NPs did not activate the astrocyte–microglia–neuron collagen axis, indicating that a pathological background was required. These results suggested that MNPs acted as amplifiers of disease processes once vulnerability was present. Although human datasets did not assess MNP exposure, the parallel activation pattern supported the possibility that environmental MNPs worsened AD‐related cognitive impairment in susceptible individuals. To investigate how PS‐NPs influenced intercellular signaling, we compared astrocyte mono‐ and triculture. PS‐NPs did not alter COL1A1 expression in isolated astrocytes, indicating that the pathway was not directly triggered in single cells. In tricultures, PS‐NPs induced robust COL1A1 upregulation, even when downstream signaling was blocked, suggesting they acted as environmental amplifiers within multicellular communication loops. The amplification depended on the collagen‐based ECM network that links astrocytes, microglia, and neurons. ⁶⁷ , ⁶⁸ While our analyses focused on astrocyte–microglia–neuron signaling, microglia reinforced feedback to astrocytes, which intensified with PS‐NP exposure. Through this loop, microglial cues propagated back to astrocytes, driving further COL1A1 upregulation and amplifying downstream neurotoxic cascades. ⁶⁹ , ⁷⁰ Thus, PS‐NPs interfered with dynamic intercellular communication without directly activating collagen signaling. Multicellular crosstalk likely initiated this pathway, but the precise upstream trigger remained undefined, as the current system could not isolate the earliest signals. Future studies will manipulate candidate mediators to clarify these initiating events. Overall, this feedback shift demonstrated how PS‐NPs could destabilize intercellular networks, exacerbate AD pathology, and accelerate cognitive decline.

The collagen pathway regulates ECM remodeling, adhesion, and migration. ⁶⁷ , ⁷¹ In AD, dysregulated ECM signaling strengthens collagen–integrin interactions and promotes pathological cellular responses. ⁷² , ⁷³ , ⁷⁴ The ECM also shapes intercellular communication by sequestering cytokines and engaging integrin receptors in a juxtacrine‐like manner. ⁴⁶ , ⁴⁷ To examine these mechanisms, we used a transwell triculture system separating astrocytes, microglia, and RA‐differentiated SH‐SY5Y neurons, which stably express Aβ. This setup isolated ECM–integrin signaling in the presence of Aβ. PS‐NPs enhanced astrocyte‐derived COL1A1 signaling to microglial ITGA1/ITGB1, activating microglia and promoting the release of inflammatory and neurotoxic factors. ⁷⁵ Activated microglia then signaled to neurons through COL1A2–ITGA1/ITGB1, a receptor pair required for Aβ‐induced p38‐MAPK activation and neurotoxicity. ⁵⁰ Without this integrin complex, Aβ could not trigger downstream damage. Thus, PS‐NPs primed glial–neuronal communication through the collagen pathway and unmasked Aβ‐dependent toxic cascades.

We recognized several limitations in our in vitro models. RA‐differentiated SH‐SY5Y cells approximate mature neurons but lack key features of primary neurons, including complex network formation, robust synaptogenesis, and mature electrophysiology. ⁷⁶ , ⁷⁷ , ⁷⁸ Their molecular and stress‐response profiles also differ, which may limit translation to in vivo neuronal behavior. The transwell co‐culture system provides controlled access to ECM–integrin signaling by physically separating cell types, ²⁸ yet it cannot reproduce the three‐dimensional architecture of the native brain ECM. ⁷⁹ In vivo, ECM structure shapes mechanical cues, synapse organization, signaling niches, and collagen–integrin dynamics. ⁶⁷ , ⁸⁰ These spatial features cannot be fully reproduced in a planar transwell format, and thus the magnitude and distribution of signaling events likely differ from in vivo conditions. Although we did not isolate ECM components directly, CO‐IP allowed us to capture proteins involved in ligand–receptor interactions within the collagen pathway, most of which were ECM‐derived. Despite these constraints, the system offered a useful framework for dissecting how PS‐NPs perturb ECM–integrin signaling and amplify glial–neuronal dysfunction in an AD‐relevant context.

Another key finding was the protective effect of TC‐I 15 against PS‐NP‐accelerated AD pathology. TC‐I 15 binds the β1 domain of integrins and blocks collagen–integrin interactions that support ECM‐dependent signaling. ²³ , ⁸¹ , ⁸² In the triculture system, TC‐I 15 markedly reduced PS‐NP‐induced neuronal death, and similar protection was observed in vivo with reduced neuronal loss and improved cognition. TC‐I 15 did not alter Aβ levels, indicating that it acted on the downstream cascade triggered by PS‐NP exposure rather than on Aβ production. Mechanistically, Aβ alone fails to activate the p38 MAPK without upstream collagen–integrin engagement. ⁵⁰ PS‐NP exposure amplified ECM‐derived collagen signaling, enabling Aβ to trigger p38 MAPK activation and neuronal death. Blocking the collagen pathway with TC‐I 15 interrupted this sequence, protecting neurons despite persistent Aβ expression. TC‐I 15 may also affect other β1–integrin pathways, including α2β1 and α11β1, which could contribute only modestly to the overall inhibition of collagen‐dependent signaling, as TC‐I 15 exhibits much lower inhibitory potency for these integrins compared with α1β1. ²⁴ However, the phenotypic improvement showed a strong and consistent match with inhibition of the collagen–α1β1 axis. We monitored key indicators of this pathway, including microglial activation, p38 MAPK phosphorylation, and neuronal survival. TC‐I 15 improved each of these readouts in a pattern that mirrored the expected consequence of collagen blockade. This close alignment indicated that collagen–α1β1 inhibition was the main driver of the protective effect. Overall, these results highlighted the central role of collagen signaling in Aβ‐induced neurotoxicity and supported the idea that environmental MNPs exacerbated cognitive decline by priming this axis.

In conclusion, our study revealed that PS‐NPs exacerbated cognitive deficits in AD by dysregulating collagen‐mediated intercellular communication. Targeting this pathway might mitigate PS‐NP‐induced brain damage. Our findings highlighted both a mechanistic link between environmental pollutants and AD and the broader public health importance of addressing plastic pollution, consistent with the Lancet Commission's emphasis on modifiable environmental risk factors for dementia.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

All animal procedures were approved by the Scientific Research Committee on Ethics in the Care and Use of Laboratory Animals at Southern Medical University (Approval No. SMUL202404002) and were conducted in accordance with institutional and national guidelines for the ethical treatment of laboratory animals.

Supporting information

Supporting Information

Supporting Information

Supporting Information

Zhong Y, Fan B, Yang X, et al. Nanoplastics trigger glial–neuronal collagen signaling miscommunication to exacerbate cognitive impairment in Alzheimer's disease. Alzheimer's Dement. 2026;22:e71096. 10.1002/alz.71096