Lewy body dementia promotion by air pollutants

Abstract

Evidence links air pollution to dementia, yet its role in Lewy body dementia (LBD) remains unclear. Here we showed in a cohort of 56.5 million individuals across the U.S. that PM_2.5 exposure raises LBD risk. Mechanistically, we found PM_2.5 exposure led to brain atrophy in wild-type mice, an effect not seen in α-synuclein (αSyn)-deficient mice. PM_2.5 exposure generated a highly pathogenic αSyn strain, PM-PFF, with enhanced proteinase K-resistance and neurotoxicity, resembling αSyn LBD strains. PM_2.5 samples from China, the U.S., and Europe consistently induced proteinase-resistant αSyn strains and in vivo pathology. Transcriptomic analyses revealed shared responses between PM_2.5-exposed mice and LBD patients, underscoring PM_2.5’s role in LBD and the stresses the need for interventions to reduce air pollution and its associated neurological disease burden.

Emerging evidence suggests that urban and roadside air pollution are contributing factors to dementia (1). Higher annual mean concentrations of fine particulate matter (PM_2.5) in the United States (US) are associated with an increased risk of first hospital admissions for Alzheimer’s disease (AD) and related dementia incidence (2–4). Lewy body dementia (LBD), the second most common form of dementia, is characterized by the aggregation of α-synuclein (αSyn) and includes Dementia with Lewy Bodies (DLB) and Parkinson’s disease (PD) with dementia (PDD) (5–8). The link between air pollution and LBD is unknown.

In LBD and related α-synucleinopathies, one of the most pivotal discoveries is the prion-like behavior of αSyn, which reflects the presence of distinct αSyn strains across different α-synucleinopathies (9, 10). These strains drive the spread of αSyn pathology and contribute to subsequent behavioral impairments (11–18). Emerging data also suggest that environmental factors can modify αSyn strains, potentially worsening or mitigating disease progression (18–21).

αSyn strains appear to progress from PD without dementia to PDD in both cross-sectional and longitudinal cohorts (22, 23). We propose that air pollutants, particularly PM_2.5, contribute to the development of LBD by promoting the formation of LBD-like αSyn strains. To test this hypothesis, we combined epidemiological evidence with the multi-modal analyses of PM_2.5-induced αSyn strains derived from mouse models, comparing these with αSyn strain amplified from the cerebrospinal fluid (CSF) of LBD patients. Additionally, we examined transcriptomic responses in PM_2.5-exposed and PM_2.5-induced αSyn strain-inoculated mouse brains, comparing these results with previously published data from LBD brain tissues (24). This comprehensive approach aims to clarify the mechanisms by which PM_2.5 exposure may promote LBD and support interventions that reduce PM_2.5 exposure.

Study Design, Population, and Data Collection in Epidemiological study

To examine the relationship between PM_2.5 exposure and the occurrence of α-synucleinopathies resulting in hospital admissions, we conducted a longitudinal cohort study. Our study population is a longitudinal open cohort that includes all Medicare-fee-for-service beneficiaries who were aged 65 years or older in the U.S. from Jan 1, 2000, to Dec 31, 2014, using the Medicare part A data. People are eligible to enter Medicare after they turn 65 years of age, and for each beneficiary, follow-up started on Jan 1, 2000, or Jan 1 of the year following entry into the Medicare program, until first admission with diagnosis codes for each outcome separately, death, or the end of the study period, whichever came first. In this study, PD includes both PD without dementia and PDD. Here, PD without dementia specifically refers to patients who have not progressed to PDD. Four α-synucleinopathy related outcomes considered were first hospital admissions for PD, PD without dementia, PDD, and DLB. We obtained the Medicare inpatient hospital claims from the Medicare Provider and Analysis Review files, which include one record per hospital admission. We extracted the primary or secondary diagnoses in hospital admissions for each beneficiary based on International Classification of Diseases (ICD) codes (detailed in the Materials and Methods). We also extracted age, sex, race, postal code (such as ZIP code) of residence, and Medicaid eligibility for each beneficiary in each follow-up year. Medicaid eligibility is a proxy for individual-level socioeconomic status, where a Medicare beneficiary eligible for Medicaid is likely to have a lower socioeconomic status. Table S1 provides descriptive information on the study cohort. This epidemiological study was conducted under a protocol approved by the Columbia University Mailman School of Public Health Human Subjects Committee.

The cohort included more than 56.5 million individuals residing in 34,824 ZIP codes, with a mean entry age of 70.1 years (SD 7.2). The total follow-up time ranged from 406.7 to 409.3 million person-years for four α-synucleinopathy related outcomes. The number of first hospital admissions was 0.31 million for DLB, 0.39 million for PDD, and 0.73 million for PD without dementia, with 1.14 million for PD by the end of follow-up. The interquartile range (IQR) for an annual average PM_2.5 during the study period was 4.14 μg/m³.

Geographical Distribution of PM_2.5 and α-Synucleinopathies

As shown on the US map in Fig. 1A, the mean annual PM_2.5 concentration during the study period was 9.7 μg/m³ (IQR 4.14 μg/m³). PM_2.5 concentrations were generally higher in the eastern US than in the western US, although certain areas in California exhibited the country’s highest concentrations. Fig. 1B illustrates the occurrence of first hospital admissions for the four α-synucleinopathies diagnoses per 100,000 Medicare beneficiaries across the contiguous US, indicating geographical similarities to the distribution of PM_2.5.

Fig. 1. Correlation between PM_2.5 exposure with LBD/PD outcomes.

(A) Nationwide PM_2.5 concentrations across the contiguous United States from 2000 to 2014. (B) Occurrences of first hospital admissions for four α-synucleinopathies across the contiguous United States from 2000 to 2014, per 100,000 Medicare fee-for-service beneficiaries. (C) Hazard Ratios (HRs) and 95% Confidence Intervals (CIs) for long-term PM_2.5 exposure and first hospital admissions for four α-synucleinopathies. HRs were calculated using Cox proportional hazard models, adjusted for socioeconomic factors, behavioral risk factors, meteorological variables, geographic region, and year. (D) HRs and 95% CIs for long-term PM_2.5 exposure and first hospital admissions for four α-synucleinopathies across different exposure windows. 10-Year and 5-Year Moving Average refer to time-varying averages over the 10- and 5-year periods preceding the hospital admission. 10-Year and 5-Year Lagged refer to annual PM_2.5 concentrations from 10 and 5 years before the hospital admission.

Associations between PM_2.5 Exposure and Risk of α-Synucleinopathies

Motivated by the geographical similarities between PM_2.5 and α-synucleinopathies diagnoses, we performed an epidemiological study adjusting for spatiotemporal, socioeconomic, behavioral, and meteorological covariates (See Methods). As shown in Fig. 1C, long-term exposure to PM_2.5 was associated with increased risk for first hospital admissions with all four α-synucleinopathies diagnoses. In the full cohort, we observed a hazard ratio (HR) of 1.10 (95% CI 1.09–1.10) for PD admissions, consistent with our previously published results (3), and an HR of 1.07 (95% CI 1.07–1.08) for PD without dementia per IQR increase in annual PM_2.5 concentrations. For LBD outcomes, we found a higher risk with an HR of 1.17 (95% CI 1.16–1.18) for PDD and an HR of 1.12 (95% CI 1.11–1.13) for DLB admissions. Overall, the risk increases were more pronounced for LBD (PDD and DLB) compared to PD and PD without dementia, highlighting a greater adverse effect of PM_2.5 exposure on LBD admissions.

Analysis of Different PM_2.5 Exposure Windows

Fig. 1D illustrates the relationships between α-synucleinopathy outcomes and PM_2.5 concentrations across different exposure windows. PM_2.5 exposure was assigned to patients based on time-varying averages over the 10- and 5-year periods preceding their first hospital admission with an α-synucleinopathy diagnosis (10-Year and 5-Year Moving Average). Alternatively, PM_2.5 exposure was assigned using annual PM_2.5 concentrations from 10 and 5 years before the first hospital admission (10-Year and 5-Year Lagged). Overall, we observed positive associations between PM_2.5 exposure and the risk of first hospital admission for all four α-synucleinopathies across various exposure windows. The use of multi-year moving average exposures yields consistent results compared to those based on more recent annual PM_2.5 exposures shown in Fig. 1C. In contrast, relying on lagged exposure from many years ago resulted in less pronounced, yet still significant (p < 0.05), positive associations.

Together, these findings underscore that PM_2.5 exposure is one of the environmental risk factors associated with increased hospital admissions for the LBD, including PDD and DLB. Hospital admissions serve as a proxy for neurodegeneration reaching a severe disease stage and may indicate accelerated disease aggravation. We also conducted an extensive set of sensitivity analyses to show the robustness of the observed associations (see Supplementary Materials and Table S2).

Genetic Depletion of αSyn Prevents PM_2.5-induced Brain Atrophy and Neurodegenerative Pathology

Building on these epidemiological findings that demonstrate a strong association between PM_2.5 exposure and LBD, we sought to investigate the underlying mechanisms through controlled animal studies. To explore the causal mechanisms of the association between PM_2.5 and LBD and model the delayed onset of neurodegeneration, a long-term PM_2.5 exposure paradigm was conducted using wildtype (WT) mice followed by nest building tests at 4, 6, and 8 months post-exposure. No deficits in fine motor function or emotional behavior were observed (fig. S1A)(25–27). However, by 10 months post-exposure, clear impairment emerged, suggesting that prolonged exposure is required to induce behavioral deficits (fig. S1A). Similarly, pS129-positive αSyn pathology was minimal at 2 months post-exposure but showed substantial accumulation by 10 months, reinforcing a time-dependent association between chronic PM_2.5 exposure and progressive LBD-like pathology (fig. S1B, C).

Brain atrophy is a hallmark of dementia, including LBD, characterized by gray matter reduction in the cortical and hippocampal regions due to neurodegeneration (25, 28–35). To investigate the effects of PM_2.5 on brain structure, wild-type (WT) mice were exposed nasally to PM_2.5 every other day for 10 months (details in Methods) (Fig. 2, A and B). Compared to PBS-treated controls, the WT mice exposed to PM_2.5 exhibited brain atrophy, evidenced by an enlargement of the posterior lateral ventricle (LV) and a reduction in the medial temporal lobe (MTL) and hippocampus (Hip) areas (Fig. 2, C to F). αSyn knockout (αSyn^−/−) mice were protected from this PM_2.5-induced brain atrophy, showing no changes in these brain regions (Fig. 2, C to F).

Fig. 2. αSyn Depletion Prevents PM_2.5-Induced Brain Atrophy and Cell Death.

(A) PM_2.5 extraction and preparation method. (B) In vivo experiment timeline: nasal PM_2.5 administration to WT and αSyn^−/− mice. (C) Nissl staining of 12-mon-old mouse brains after 10-mon PM_2.5 exposure. (D-F) Volume measurements of (D) posterior lateral ventricle (LV), (E) medial temporal lobe (MTL), and (F) hippocampus (Hip) (n = 6). (G, H) Representative images showing TUNEL (red) and NeuN (green) staining in the cortex of WT and αSyn^−/− mice after exposure to either PBS or PM (n = 4). Normalized Statistics: Data are presented as violin plots showing all individual data points. Statistical significance was determined by using one-way ANOVA with Tukey’s correction; **p < 0.01, ***p < 0.001, ****p < 0.0001; ns, not significant.

Additionally, increased tau hyper-phosphorylation (p-tau217, p-tau181, p-tau202/205) (fig. S2), microglial activation (fig. S3) (36–38), astroglial activation (fig. S4) (39–41), elevated oxidative stress (fig. S5) (42–44), a reduced percentage of cortex myelination (fig. S6) (45, 46), increased DNA double-strand breaks (fig. S7) (47–49) and DNA fragmentation (Fig. 2, G and H) (50, 51) were observed in PM_2.5-exposed WT mice. These pathological changes were mitigated in PM_2.5-exposed αSyn^−/− mice (fig. S2 to S7, and Fig. 2, G and H). No amyloid-β (Aβ) accumulation was detected in our study (fig. S8). These results highlight the critical role of αSyn in mediating PM_2.5-induced pathology and neurodegeneration.

αSyn^−/− Mice Exhibit Resilience to PM_2.5-Induced Cognitive Decline

To assess the behavioral impact of PM_2.5 exposure, open field testing was performed. No differences were observed in activity and the time spent in the center area (Cen) versus the center plus peripheral (Cen+Per) area across the four groups: WT-PBS, αSyn^−/−-PBS, WT-PM_2.5, and αSyn^−/−-PM_2.5 (fig. S9A), indicating that 10-month PM_2.5 exposure did not cause a change in activities among the different groups of animals (11, 52). Next, spatial recognition memory using the Y-maze was evaluated. WT mice exposed to PM_2.5 spent less time in the novel arms, indicating impaired spatial recognition memory compared to the WT-PBS group (53, 54). In contrast, αSyn^−/− mice exposed to PM_2.5 did not show cognitive deficits relative to the αSyn^−/−-PBS group (fig. S9B). To further evaluate memory and cognitive function, the Novel Object Recognition (NOR) test was performed (55, 56). Although no differences were observed during the training phase among the four groups, WT mice exposed to PM_2.5 exhibited a reduction in the recognition index (%) compared to the WT-PBS group (fig. S9C). Similarly, αSyn^−/− mice showed no differences in recognition index, regardless of PM_2.5 exposure (fig. S9C). Collectively, these findings suggest that PM_2.5 exposure induces cognitive decline in WT mice, whereas αSyn^−/− mice are protected from these PM_2.5-induced deficits.

Mapping of αSyn Pathology in the Brain and Peripheral Organs of Non-Transgenic Mice following PM_2.5 Exposure

To investigate the effects of PM_2.5 exposure on αSyn pathology, phosphorylated serine 129 (pS129) αSyn immunoreactivity in both the brain and peripheral organs of WT mice was examined. In the brain of WT mice, we observed pS129-positive αSyn pathology in key regions such as the Cornu Ammonis 1 (CA1) region, dentate gyrus (DG), and parietal cortex (PaC) compared to PBS-treated controls (fig. S10). In addition to the brain, we examined αSyn pathology in the gastrointestinal (GI) tract and the lung. In the GI tract, PM_2.5 exposure induced pS129-positive αSyn pathology in the stomach, duodenum, jejunum, ileum, and colon (fig. S11, A to F). In the lung, PM_2.5 exposure also led to pS129-positive αSyn pathology (fig. S11, A and G). In contrast, αSyn^−/− mice exposed to PM_2.5 exhibited minimal pS129 αSyn signal in both the brain, GI tract and the lung (fig. S10 and S11).

PM_2.5 Exposure Induces a Proteinase-Resistant αSyn Strain In Vivo, Contributing to Cognitive Decline in hA53T Mice

Environmental factors can influence the formation of distinct αSyn strains, potentially leading to diverse disease trajectories (18–21). Building on these findings and the role of αSyn in PM_2.5-induced deficits, we aimed to determine whether exposure of αSyn to PM_2.5 could induce a distinct αSyn strain.

Since WT mice do not naturally form αSyn aggregates during the aging process, we used human A53T mutant (hA53T) transgenic mice, which are prone to developing αSyn aggregates and associated pathology as they age. This model allowed us to compare the αSyn strain formed due to natural aging with those induced by PM_2.5 exposure.

PM_2.5 was nasally administered every other day to hA53T mice (57–59) over a period of 5 months (fig. S12A). Substantial αSyn pathology was observed in multiple brain regions of PM_2.5-exposed hA53T (PM_2.5- hA53T) mice compared to the PBS-hA53T control group (fig. S12B). Brain regions exhibiting elevated pS129-positive αSyn pathology included the olfactory bulb (OB), prefrontal cortex (PFC), PaC, piriform cortex (PiC), Hip, hypothalamus (HyP), and brainstem (BS), but not the midbrain (MB) (fig. S12, C and D).

To determine whether ambient PM_2.5 from different geographic regions could yield similar results, samples of PM_2.5 were collected from three distinct geographical locations: Shaanxi, China (PM_2.5-1), Georgia, US (PM_2.5-2) and Prague, Czech Republic (PM_2.5-3). Table S3 shows a comparison of PM_2.5 samples collected from different locations in this study. Nasal administration of the PM_2.5 samples from the 3 different geographical locales to hA53T transgenic mice for 2 months induced pS129-positive αSyn pathology that was similar between the three locales (fig. S13).

To investigate whether PM_2.5 can cause the formation of a different αSyn strain, we performed a seeding amplification assay (SAA) using the TX-100 insoluble fractions of brain lysates from PM_2.5-hA53T mice (Fig. 3A), followed by proteinase K (PK) digestion (18, 19) and dot blot analysis. We used aged (12-mon-old) symptomatic hA53T mice as the control (Fig. 3A). After centrifugation (100k rpm, 1 hour), we collected the amplified αSyn aggregates samples from the pellet and resuspended in PBS and sonicated to generate preformed fibrils (PFF), using a prior protocol (12). The amplified αSyn strain from PM_2.5-hA53T mice showed higher Thioflavin T (ThT) fluorescence compared to that of aged hA53T mice (Fig. 3B), dot blot results indicated that the αSyn strain from PM_2.5-hA53T mice were more resistant to PK digestion than the strain from the aged hA53T mice (Fig. 3, C and D). After 5-mon PM_2.5 exposure the hA53T mice exhibited cognitive and psychiatric impairments, as demonstrated by poor performance in the NOR test (fig. S14A), the Y-maze test (fig. S14B), and the Nest building behavior (fig. S14C). Motor dysfunction was also observed in the PM_2.5-hA53T mice (fig. S14D).

Fig. 3. PM_2.5 Exposure Induces a Proteinase-Resistant, Highly Pathogenic αSyn Strain with Enhanced Neuropathology, Neurotoxicity, and Dementia-Inducing Potential.

(A) Insoluble fraction extraction schematic: 5-mon PM_2.5-exposed hA53T (PM_2.5-hA53T) mice vs. naturally aging symptomatic hA53T (Aged hA53T) mice. (B) ThT-mfi (max fluorescence intensity) of seeded αSyn: PM_2.5-hA53T vs. Aged hA53T insoluble fractions in seeding amplification assay (SAA). (C) Dot blot: remaining αSyn after proteinase K (PK) digestion (0, 5, 15, 30 min) of PM_2.5-hA53T and Aged hA53T insoluble fractions. (D) Quantification: remaining αSyn at 15 minutes PK digestion vs. 0 minute, PM_2.5-hA53T vs. Aged hA53T. (E) PM-PFF and PFF generation schematic. (F) ThT-mfi: PM-PFF vs. PFF. (G) Dot blot: remaining αSyn after PK digestion (0, 5, 15, 30 minutes) of PFF and PM-PFF insoluble fractions. (H) Quantification: remaining αSyn at 15 minutes PK digestion vs. 0 minute, PFF vs. PM-PFF. (I) pS129-positive neuropathology in primary cortical neurons; n = 8. (J) Neurotoxicity (NeuN staining) in primary cortical neurons; n = 10. (K) In vivo protocol: striatal PM-PFF/PFF injection in 2-month-old hWT mice; behavior at 6 months, then euthanasia/staining. (L) Novel object recognition, (M) Nest building scores: PM-PFF/PFF injected mice at 6 months; n = 7–11. (N) Fluorescent images, (O) Quantification: cortical pS129 signals at 6 months post-injection (PBS, PFF, PM-PFF). (D, H, I, J, L, M, O) Data are presented as violin plots showing all individual data points. Statistics: (B, D, F, H) Means ± SEM, Student’s t-test; (I, J, L, M, O) Statistical significance was determined by using one-way ANOVA with Tukey’s correction; *p < 0.05, **p < 0.01, ****p < 0.0001; ns, not significant.

Molecular Evidence of PM_2.5-Induced αSyn Strain Formation (PM-PFF) and Distinct Cellular Response In Vitro, compared to PFF

Building on the observation that hA53T mice exposed to PM_2.5 induces a distinct αSyn strain that leads to cognitive deficits, we determined whether PM_2.5 could directly induce recombinant human αSyn misfolding into a αSyn strain that was comparable to that generated in hA53T mice exposed to PM_2.5. Accordingly, recombinant human WT αSyn monomers were exposed to 5% PM_2.5-1, PM_2.5-2, PM_2.5-3 (w/w) to generate PM_2.5-αSyn preformed fibrils, PM1-PFF, PM2-PFF, PM3-PFF, respectively and control PFF using a prior protocol (12) (Fig. 3E). ThT fluorescence assays revealed that PM_2.5 accelerated αSyn aggregation after 7 days of agitation at 37 °C (Fig. 3F, fig. S15A). Dot blot results indicated that the PM-PFF were more resistant to PK digestion than the PFF (Fig. 3G, H, fig. S15, B and C). Given the similar properties of PM_2.5 from all three regions, we used PM_2.5-1 for subsequent experiments (unless otherwise specified, PM_2.5 refers to PM_2.5-1). Circular dichroism (CD) spectra showed a peak at 218 nm, indicating increased β-sheet structure formation with higher PM_2.5 concentrations (fig. S16, A and B).

Since different αSyn strains can drive different cellular responses and strain-specific α-synucleinopathies (9, 60, 61), the cellular responses to the PM-PFF strain were compared to control PFF. Primary mouse cortical neurons were exposed to PM-PFF and control PFF (5 μg/mL) and pS129 immunoreactivity was evaluated 10 days after treatment. Immunostaining showed that the PM-PFF strain caused a strong elevation of pS129 immunoreactivity that was ~2-fold greater than neurons treated with the control PFF strain (Fig. 3I, fig. S17A). Neurons treated with PM_2.5 alone showed minimal pS129 signal (Fig. 3I, fig. S17A), indicating that the increased pS129 signal was specifically induced by the PM-PFF strain and not by PM_2.5. Neurotoxicity was evaluated by quantifying NeuN immunostaining, as previously described (12, 17). Exposure to control PFF for 21 days resulted in approximately 50% neuronal cell death, whereas the PM-PFF strain induced higher neurotoxicity (~70%) (Fig. 3J, fig. S17B). PM_2.5 alone did not cause neurotoxicity in primary neuronal cultures compared to the PBS control (Fig. 3J, fig. S17B).

These results taken together indicate that PM_2.5 promotes αSyn aggregation that leads to the formation of a propagative and neurotoxic strain. Moreover, PM_2.5 alone does not induce pS129-positive αSyn pathology or neurotoxicity in primary neuronal cultures, nor does it cause neurodegeneration in αSyn knockout mice.

Differential Effects of PM-PFF and PFF Inoculation on Cognitive and Motor Functions in Humanized WT αSyn (hWT) Mice

We next investigated whether PM-PFF can directly induce cognitive deficits in mice. Given that cross-seeded aggregation of human and mouse αSyn is bidirectionally restricted (62), we utilized the humanized WT αSyn (hWT) mice, PAC-Tg (SNCAWT) (strain: 010710) (15, 63, 64), in which endogenous mouse αSyn expression is replaced by human αSyn. As illustrated in Fig. 3K, stereotaxic injections of PM-PFF and PFF were administered into the dorsal striatum of 2–3-mon-old hWT mice. In the αSyn spreading model, PBS was chosen as the negative control based on extensive previous studies, where PBS served as the vehicle control for PFF and demonstrated high reproducibility (12, 17, 18, 65–67). Afterwards, a series of behavioral tests and immunostaining were performed to evaluate cognitive and motor functions, as well as the distribution of pS129-positive αSyn pathology.

Behavioral tests revealed that PM-PFF-inoculated mice exhibited a reduction in the recognition index during the NOR test compared to PBS groups, whereas PFF-inoculated mice exhibited no difference from PBS controls (Fig. 3L). Additionally, PM-PFF-inoculated mice left more unused nestlet material than both PFF and PBS controls, resulting in lower Nest building scores (Fig. 3M). There’s no difference between PFF and PBS groups (Fig. 3M) (25–27). There are no differences observed in activity and the time spent in the center area (fig. S18A). These results suggest that PM-PFF induces more pronounced cognitive deficits than PFF.

Motor function was assessed using the pole test, which is a sensitive indicator of dopaminergic function (68). In line with previous studies (16–18), PFF inoculation caused substantial motor deficits, including increased time to turn and total time to reach the base, compared to PBS controls (fig. S18B). In contrast, PFF inoculation resulted in more pronounced motor dysfunction compared to PM-PFF (fig. S18B).

To investigate the link between neuropathology and behavioral deficits, we assessed neurodegeneration of dopaminergic neurons in the substantia nigra (SN) using anti-tyrosine hydroxylase (TH) and Nissl staining. Both PM-PFF and PFF induced dopaminergic neuronal loss (fig. S19, A to C). However, compared with PM-PFF group, PFF group exhibited motor impairment in the pole test (fig. S18B), potentially due to the greater dopamine cell loss induced by PFF compared to PM-PFF (fig. S18B). Moreover, PM-PFF inoculation resulted in a higher accumulation of pS129-positive αSyn pathology in the cortex compared to PFF (Fig. 3N and O). In the SN, PFF induced substantial pS129-positive αSyn pathology, whereas PM-PFF caused less pS129 signal than PFF (fig. S19, D and E).

PM_2.5-Induced αSyn Strains Exhibit Passage Stability and Mimic Key Features of LBD αSyn Strains

To determine whether the PM-PFF strain exhibits passage stability and shares characteristics with LBD αSyn strains, we compared the biophysical, biochemical, pS129 αSyn immunoreactivity and neurotoxic properties of αSyn strains derived from LBD (5 PDD and 2 DLB), PD without dementia (n = 5) and healthy control (HC) (n = 5) individuals versus PM_2.5 induced αSyn second passage strains (PM-PFF^2nd) (Fig. 4A). LBD αSyn strains displayed higher ThT fluorescence compared to PD without dementia strains (Fig. 4B), and PM-PFF^2nd similarly showed higher ThT fluorescence than the second-passage PFF strains (PFF^2nd) (Fig. 4C), suggesting a stronger fibrillar structure in PM-PFF^2nd that parallels the properties of LBD strains.

Fig. 4. PM_2.5-Induced αSyn Strain Shares Key Pathological Features with LBD Patient-Derived Strain and Maintains These Features Across Passages.

(A) Schematic of SAA comparing features of LBD patient cerebrospinal fluid (CSF) samples (n = 7 LBD [5 PDD, 2 DLB], n = 5 PD without dementia, n = 5 HC) and PM_2.5-related mouse brain lysate samples (hWT mice inoculated with PM-PFF, PFF, or PBS, n = 5) using biophysical, biochemical, and cellular assays. 2nd passage samples (PM-PFF^2nd, PFF^2nd, PBS^2nd) were derived from 1st passage inoculations. (B-G) Biochemical comparisons show highest ThT signal and strongest proteinase resistance in LBD αSyn strain vs. PD without dementia and HC, and in PM-PFF^2nd vs. PFF^2nd and PBS^2nd. (H, I) pS129-positive pathology in primary cortical neurons treated with αSyn strains derived from HC, PD without dementia, and LBD, with corresponding quantification. (J, K) pS129-positive pathology in primary cortical neurons treated with PBS^2nd, PFF^2nd, PM-PFF^2nd, with corresponding quantification. (L, M) Neurotoxicity in primary cortical neurons treated with αSyn strains from HC, PD without dementia, and LBD, with corresponding quantification. (N, O) Neurotoxicity in primary cortical neurons treated with PBS^2nd, PFF^2nd, PM-PFF^2nd, with corresponding quantification. Statistics: (B, C) Means ± SEM, statistical significance was determined by using one-way ANOVA with Tukey’s correction with data at 7d; (E, G, I, K, M, O) Data are presented as violin plots showing all individual data points. Statistical significance was determined by using one-way ANOVA with Tukey’s correction; *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Next, LBD αSyn strains demonstrated greater PK resistance compared to PD without dementia strains (Fig. 4, D and E), and PM-PFF^2nd mirrored this with higher PK resistance than PFF^2nd (Fig. 4, F and G), indicating that PM-induced strains adopt a more protease-resistant conformation similar to that LBD-derived strains. Additionally, LBD αSyn strains induced more pronounced pS129 αSyn immunoreactivity compared to PD without dementia strains (Fig. 4, H and I), and PM-PFF^2nd caused more severe pS129-positive αSyn pathology than PFF^2nd (Fig. 4, J and K).

Finally, in terms of neurotoxicity, LBD αSyn strains caused greater neurotoxicity in primary cortical neurons compared to PD without dementia strains (Fig. 4, L and M), and PM-PFF^2nd proved more neurotoxic than PFF^2nd (Fig. 4, N and O), aligning with the toxic properties seen in LBD strains. Together, these findings strongly suggest that PM-PFF exhibits passage stability and shares critical pathogenic features with LBD αSyn strains.

Transcriptomic Responses to PM_2.5 exposure and PM-PFF inoculation in hWT Mice and LBD Postmortem Brains

To evaluate the trajectory of transcriptomic responses induced by PM_2.5 exposure and PM-PFF inoculation, and their resemblance to those observed in LBD compared to PD without dementia, we performed RNA-seq gene expression profiling in PM_2.5-exposed and PM-PFF-inoculated hWT mice. We selected the anterior cingulate cortex (ACC) to ensure consistency in comparisons between these experimental groups and the disease patterns observed in patients (24). Mice were divided into five groups, each exposed for 2 months under the following conditions: 1) PBS (control), 2) PM_2.5, 3) PFF, 4) PM-PFF and group 5, which received a PM-PFF inoculation followed by 7 weeks of PM_2.5 nasal administration, beginning one week post-injection (PM-PFF with PM_2.5 exposure, 8 weeks total post-injection). This fifth group, PM-PFF with PM_2.5 exposure, intended to model real-world conditions where patients with PM_2.5-induced αSyn seeds exposed to PM_2.5.

Principal component analysis of the gene expression data revealed a clear separation between control samples and those treated under the various conditions (Fig. 5A). Differential gene expression analysis identified similar numbers of differentially expressed genes (DEGs) for PM-PFF (n = 1,222) and PM_2.5 (n = 1,205), followed by PM-PFF with PM_2.5 exposure (n = 639), with the fewest DEGs observed in the PFF condition (n = 109) (table S5, fig. S20A). Comparison of DEGs across conditions revealed many unique and overlapping DEGs (fig. S20B). For instance, of the 1,205 DEGs identified in the PM_2.5 condition, 582 were unique to PM_2.5, whereas 323 shared with PM-PFF.

Fig. 5. PM_2.5 Exposure Induced Gene Expression Changes in Mice Resemble Those Observed in LBD Patients.

(A) Principal Component Analysis (PCA) of gene expression profiles shows clear separation between control samples and those exposed to four conditions: PM_2.5, PFF, PM-PFF, and PM-PFF with PM_2.5 exposure. PM: Nasal administration of PM_2.5 to hWT mice for 2 months; PFF: Striatal injection of PFF into hWT mice, followed by 2-month observation; PM-PFF: Striatal injection of PM-PFF into hWT mice, followed by 2-month observation; PM-PFF with PM_2.5 exposure: Striatal injection of PM-PFF into hWT mice, followed one week later by nasal administration of PM_2.5 for 7 weeks (8 weeks total post-injection). (B) Tables comparing differentially expressed genes (DEGs) signatures between human diseases [DLB, PD without dementia, PDD] and mice condition [PM_2.5, PFF, PM-PFF, PM-PFF with PM_2.5 exposure]. The upper table illustrates the correlation of DEGs effect sizes between mice and humans, whereas the lower table shows the proportion of DEGs with consistent direction in effects. Color gradients represent significance levels as -log₁₀(p-value), with white color indicating nominal significance (p = 0.05). (C) Scatter plot showing the correlation between DEGs effect sizes in PM_2.5-exposed mice and those observed in PD without dementia, PDD and DLB patients. Each point represents a gene, with the x-axis indicating DEGs effects in humans and the y-axis indicating DEG effects in mice. The shaded area around the blue regression line represents the 95% confidence interval. (D) Scatter plot showing the correlation of DEGs effect sizes between PFF or PM-PFF-injected mice and those observed in DLB and PDD patients. (E) Venn diagram showing the overlap of dysregulated gene sets among PFF- or PM-PFF-injected mice and DLB and PDD patients.

Gene set enrichment analysis further revealed the greatest number of dysregulated gene sets in the PM_2.5 condition (n = 365), followed by PM-PFF with PM_2.5 exposure (n = 271) and PM-PFF (n = 266), whereas the PFF condition displayed the fewest dysregulated gene sets (n = 13) (table S5). Consistent with the DEGs results, both unique and shared dysregulated gene sets were observed across conditions (fig. S20C), suggesting that these exposures may perturb both distinct and overlapping biological pathways. For example, of the 365 dysregulated gene sets in the PM_2.5 condition, 144 were unique to PM_2.5, including 11 downregulated pathways related to cell division whereas 116 gene sets were shared with both PM-PFF and PM-PFF with PM_2.5 exposure, all of which were upregulated and related to immune system pathways.

PM_2.5 Exposure Induced Gene Expression Changes in Mice Resemble Those Observed in LBD Patients:

Gene expression changes induced by each experimental condition in mice were compared with those observed in the anterior cingulate cortex of human individuals with DLB, PDD, and PD without dementia (Fig. 5, B to D). In Fig. 5B, Spearman correlations of log₂ (fold) changes of gene expression and the percentages of genes with concordant directional changes are presented for each comparison. For PM_2.5 exposed mice, we found the strongest correlation with PDD (R = 0.35, p = 1.7 × 10⁻⁴), followed by a weaker but still significant correlation with DLB (R = 0.17, p = 3.0 × 10⁻⁴), and no significant correlation with PD without dementia (R = 0.08, p = 0.61) (Fig. 5C). Additionally, the percentage of genes that changed in the same direction was 67% for PDD (p = 3.7 × 10⁻⁴) and 56% for DLB (p = 0.02), both of which were significantly higher than the 50% expected by chance. This pattern of gene expression correlation aligns with our epidemiological findings, suggesting that PM_2.5 exposure has the greatest impact on PDD, followed by DLB, with the smallest effect on PD without dementia.

Gene Expression Changes Induced by PM-PFF in Mice More Closely Mimic LBD Patients than Those Induced by PFF:

For gene expression changes induced by PFF injected mice, we observed weak correlations with both DLB (R = 0.2, p = 0.0028) and PDD (R = 0.24, p = 0.051) and no significant correlation with PD without dementia (R = −0.18, p = 0.34). In contrast, gene expression changes induced by PM-PFF injection showed much stronger correlations with both DLB (R = 0.37, p = 1.3 × 10⁻¹⁵) and PDD (R = 0.46, p = 1.2 × 10⁻⁶), whereas no significant correlation was found with PD without dementia (R = 0.059, p = 0.69) (Fig. 5, B and D). Furthermore, the percentage of genes with concordant directional changes under PM-PFF was 81% for PDD (p = 1.6 × 10⁻¹⁰) and 69% for DLB (p = 3.7 × 10⁻¹⁵), both higher than the 50% expected by chance (Fig. 5B). This stronger correlation with PM-PFF-induced gene expression changes, compared to PFF, suggests that PM-PFF more closely resembles the transcriptomic patterns seen in LBD patients. This aligns with our previous findings that PM-PFF exhibits similar features to LBD-derived αSyn strains in terms of biophysical, biochemical, neurotoxic, and pS129-positive αSyn pathology. The similarity in gene expression profiles induced by PM-PFF further validates that PM-PFF functions as an LBD-like αSyn strain, providing additional support for the role of environmental factors such as PM_2.5 in the development of LBD.

Additionally, the PM-PFF with PM_2.5 exposure condition exhibited the highest correlation and the largest proportion of genes with consistent directional changes for both PDD and DLB (Fig. 5B). This condition closely mimics real-world scenarios, where individuals with PM_2.5-induced αSyn seeds are continually exposed to PM_2.5, leading to progressive disease pathology. Taken together, these findings suggest a powerful link between PM_2.5 exposure, the formation of LBD-like αSyn strains, and the development of LBD, highlighting the role of environmental factors in disease progression.

Dysregulation of Immune-Related Pathways Underlies both PM-PFF-Inoculated Mice and LBD Patients:

To further investigate the biological mechanisms underlying the shared gene expression signatures between PM-PFF-inoculated mice and LBD patients (PDD and DLB), we conducted gene set enrichment analysis using a set of 14,066 homologous genes common to both species (table S6). In PFF-inoculated mice, 33 significant gene sets were identified (FDR < 0.05), with 29 of these overlapping with gene sets identified in PDD or DLB (Fig. 5E, fig. S21). In contrast, PM-PFF-inoculated mice exhibited a much broader effect on pathways, with 243 significant gene sets (FDR < 0.05), 222 of which overlapped with those in both PDD or DLB (Fig. 5E). These 222 overlapping gene sets were clustered into several major groups (fig. S22), most of which were upregulated and primarily related to immune system processes. Comparative analysis of enriched pathways between PM_2.5-exposed or PM-PFF with PM_2.5 exposure mice and LBD patients revealed similar immune-related processes underlying both mouse models and LBD patients (fig. S23 and S24). These results suggest that dysregulation of immune-related pathways plays a central role in the pathogenesis of LBD, further supporting the hypothesis that environmental factors like PM_2.5 may contribute to LBD development by potentially triggering immune system disturbances similar to those observed in individuals with LBD.

Conclusion

In this study, we explored the role of PM_2.5 air pollution in promoting LBD and identified an LBD-like PM-PFF strain. Our epidemiological analysis established an association between PM_2.5 exposure and LBD outcomes, particularly in the context of PDD and DLB. Using transgenic and humanized αSyn mouse models, we demonstrated that PM_2.5 exposure induces LBD-like pathology, including pS129-positive αSyn aggregation, neurodegeneration, brain atrophy, and cognitive decline.

A key finding in this study is the identification of the PM-PFF strain, which shares biophysical, biochemical, and neuropathological features with αSyn strains derived from LBD patients. Transcriptomic analyses further revealed that gene expression changes induced by PM-PFF in mice closely resemble those observed in individuals with LBD, more than those induced by PFF alone. This strong alignment between the PM-PFF strain and LBD-associated gene expression profiles strengthens the conclusion that the PM-PFF strain represents an LBD-like αSyn strain, potentially linking PM_2.5 exposure to LBD pathogenesis.

Moreover, depletion of αSyn mitigated the pathological and cognitive deficits induced by PM_2.5 exposure, underscoring the central role of αSyn in PM-induced neurodegeneration. Although our study confirms αSyn’s key role in this process, the absence of a pronounced cognitive difference between WT and αSyn^−/− mice under PM_2.5 exposure suggests longer observation periods may be necessary for PM_2.5 exposure-induced deficits. Additionally, other prion-like proteinopathies, such as tau, amyloid-β, may contribute to PM_2.5 exposure-induced deficits.

Recent studies have demonstrated a strong association between PM_2.5 exposure and neurodegenerative diseases, highlighting mechanisms including αSyn aggregation, neuroinflammation, and gut-brain axis disruption. However, these studies primarily focused on the links between PM_2.5 and either AD or PD (69–72). These findings reinforce the role of PM_2.5 in AD/PD pathogenesis; our study further expands this understanding by emphasizing the impact of PM_2.5 exposure on α-synucleinopathy-associated dementia, including PDD and DLB.

Limitations of the study

Despite these findings, the following limitations should be noted in the context of an epidemiological study. First, our analysis used first hospital admissions as outcomes, which can only be interpreted as increased need for inpatient care and accelerated disease aggravation associated with higher PM_2.5 exposure. Hospital admission data did not allow us to examine the relationship between PM_2.5 and true disease onset or incidence. Furthermore, hospital admission records may miss some cases of disease aggravation and misclassify certain non-α-synucleinopathy outcomes. However, they capture a substantial number of α-synucleinopathy cases progressing to a severe disease stage as indicated by clinical studies on LBD and PD hospitalizations (73, 74). Second, using ICD codes in Medicare records to identify disease outcomes is subject to outcome misclassifications, which could either over- or under-estimate the number of outcomes in the study population (75–77). The large scale of this analysis prevents us from reviewing and applying consensus diagnostic criteria to individuals’ clinical or medical records. However, we presented a single outcome combining all first hospital admission with αSyn-related disorder (including PD without dementia and LBD) and used multiple criteria defined by different sets of ICD codes to identify PD without dementia and LBD outcomes. The consistency across results from multiple sensitivity analyses showed the robust relationship between PM_2.5 exposure and the aggravation of α-synucleinopathies. Third, determining the exposure window during which air pollution increases the risk of hospital admissions for α-synucleinopathies is challenging. However, through analyses using different exposure windows, we found consistent associations between PM_2.5 and the risk of hospital admissions for all four α-synucleinopathies, regardless of the chosen window. Fourth, exposure measurement error could potentially affect effect estimates. In this study, PM_2.5 exposure was estimated from a well-validated ensemble prediction model (78). Although the model demonstrated excellent overall accuracy, its performance varied across different U.S. regions, partly due to the limited number of monitoring sites. To this end, in a region-specific analysis of four geographic regions—Northeast, South, Midwest, and West—the results were generally consistent across regions. Yet, in the Northeast, we observed a strong association with PD outcomes but a substantially weaker association with DLB outcomes, warranting further investigation. Furthermore, ZIP code-level PM_2.5 exposure may not fully represent accurate residential address’ or personal exposures of each beneficiary. Several studies have used personal exposure measurements to correct exposure measurement error in studies of comparable populations (79, 80). Empirical evidence indicates that assigning larger-scale ambient air pollution exposure to individuals typically biases results toward the null, consistent with theoretical analyses based on non-differential measurement error (81). However, real-world scenarios may involve both non-differential and differential measurement error. The use of ZIP code- and personal-level PM_2.5 exposure involves trade-offs. When exposure estimates are not at the individual-level, they are less prone to confounding biases from personal factors that are difficult to control, such as individual-level socioeconomic and behavioral risk factors (82). Fifth, nearly one-fifth of Medicare beneficiaries changed their residential ZIP codes at least once between 2000 and 2014. Since we update exposure data for each beneficiary every year, their movement will not affect our exposure assessment (83). To this end, a sensitivity analysis of Medicare beneficiaries who did not move throughout the follow-up period (referred to as non-movers) showed results largely consistent with the main analysis, with only a slight attenuation in effect sizes, suggesting that residential mobility during the study did not substantially alter the observed associations between PM_2.5 exposure and α-synucleinopathies. Acknowledging the limitations of the epidemiological study, we do not interpret its results as confirmatory or causal; rather, they serve as the key foundation for the subsequent laboratory study.

In addition to the epidemiological study limitations, there are also inherent limitations in our laboratory study. We cannot conclude whether PM_2.5 exposure preferentially influences the trajectory of DLB or PDD, as this may involve complex interactions between specific PM_2.5 species and genetic factors. Key distinguishing features of DLB, such as autonomic dysfunction and (rapid eye movement) REM sleep behavior disorder, may also manifest in PM_2.5-exposed mice, which warrants further evaluation in future studies. Besides, we applied nasal administration rather than whole-body exposure for PM_2.5. We acknowledge that nasal administration differs from inhalation, as it delivers a high dose over a short period rather than continuous low-dose exposure, and the exposure route and subsequent systemic distribution vary slightly. However, due to the large sample requirements and high material loss associated with whole-body exposure, we opted for nasal administration in this study. To ensure rigor, the PM_2.5 dose used in our study was precisely calculated (see Methods) and is comparable to inhalation exposure in polluted urban environments. Furthermore, previous research (84) has shown that, at equivalent doses, initial lung burdens of nanoparticles were similar between instillation and inhalation, though instillation led to greater short-term retention. In the future, our research aims to explore chronic whole body exposure models for a more accurate representation of real-world exposure.

Overall, these findings establish a mechanistic link between PM_2.5 exposure and the development of LBD, with the PM-PFF strain emerging as a potential environmental intervention target. Further investigation into specific PM_2.5 chemical components and their distinct effects on neurodegenerative diseases is warranted, given the spatial and temporal variation in PM_2.5 chemical composition across the U.S. (85). A more comprehensive integration of medical records, gene-environment interactions, and human epidemiological and neurobiological studies would enhance our understanding of how environmental pollutants contribute not only to the progression but also to the onset of LBD and related disorders.

Materials and Methods

Data Procedures for Epidemiological Studies

We used International Classification of Diseases (ICD) codes to identify patients who had a first admission with primary or secondary diagnoses corresponding to DLB (ICD-9: 331.82; ICD-10: G31.83) and PD (ICD-9: 332.0, 332.1; ICD-10: G20, G21.11, G21.19, G21.3, G21.4, G21.8, G21.9) in hospital admissions among U.S. Medicare beneficiaries during the study period of 2000–2014. Given that the U.S. officially transitioned to ICD-10 on October 1, 2015, most diagnoses recorded during the study period were coded using the ICD-9 categorization scheme. For the four α-synucleinopathy related outcomes considered, PD and DLB diagnoses were given to patients who had a first hospital admission with a diagnosis of PD and DLB, respectively. PDD diagnosis was given to patients with a prior diagnosis of PD and who had a first hospital admission with a dementia diagnosis (ICD-9: 294.10, 294.11; ICD-10: F02.80, F02.81) at least one year after the PD diagnosis in the follow-up years. PD-without dementia patients are defined as patients who had a first hospital admission with a PD diagnosis during the study period yet without identifying any dementia diagnosis during their entire follow-up period. Various sensitivity analyses were performed to assess the robustness of the findings to different uses of ICD codes, including: (1) excluding secondary parkinsonism diagnoses from PD diagnosis; (2) combining PD without dementia and LBD diagnoses into a single category for αSyn-related disorder diagnosis; and (3) restricting the analysis to dementia patients who had to have a primary diagnosis of PD or DLB and a secondary diagnosis of dementia in one admission.

We obtained daily PM_2.5 concentrations at a high spatiotemporal resolution using a 1-km² grid network across the contiguous United States and a well-validated ensemble-based prediction model (58). This model used ensemble-learning approaches to combining three machine learning models: a random forest regression, a gradient boosting machine, and an artificial neural network. These machine learning algorithms used more than 100 predictor variables from satellite data, land-use information, weather variables, and output from chemical transport model simulations. The ensemble-based model was trained and calibrated using daily PM_2.5 concentrations from 2,156 monitoring sites in the U.S. EPA’s Air Quality System database and the IMPROVE monitoring network. It achieved an average cross-validated R² of 0.86 for daily PM_2.5 predictions and 0.89 for annual predictions, with performance ranging from 0.77 in mountainous regions of the USA to 0.92 in the eastern Midwest, demonstrating excellent prediction performance.

Residential addresses are not available for Medicare enrollees, only residential ZIP Codes. For each standard ZIP Codes, we used zonal statistics to calculate the daily average PM_2.5 concentration based on all 1-km² grid cell predictions within the ZIP Codes via aggregations. More specifically, we first overlaid the ZIP Codes boundaries to the 1-km² grid cells and then averaged the predictions at 1-km² grid cells whose centroids fall within the boundary of that ZIP Codes (59). For P.O. Box–only ZIP Codes, the average PM_2.5 concentrations were calculated by linking to the predictions from the nearest 1-km² grid cell. Based on these results, we estimated the annual ZIP Code average and assigned the ZIP code-wide annual PM_2.5 concentration means to Medicare beneficiaries according to the ZIP Code of residence and calendar year. PM_2.5 exposure was assigned to patients for each follow-up year using various exposure windows, based on the time-varying average PM_2.5 concentrations within the specified window preceding their first hospital admission with a α-synucleinopathy diagnosis. In the US, the mean population per ZIP Code is approximately 7500. Each postal code can cover a small area in cities but can be larger in rural areas. The median land area of a ZIP Code is around 92 km².

We collected neighborhood-level socioeconomic status and behavior risk factors, available at county- or ZIP Code tabulation areas (ZCTA)-level, which have both been associated with ambient air pollution and implicated in neurological health. These variables derived from the Behavioral Risk Factor Surveillance System from 2000 to 2014, the 2000 and 2010 US Census, and the American Community Survey for each year from 2005 to 2014. Specifically, we included (i) two county-level behavior risk variables: average body mass index and smoking rate; (i) eight ZIP Code-level socioeconomic status and demographic variables: proportion of Hispanic residents, proportion of Black residents, median household income, median home value, proportion of residents in poverty, proportion of residents with a high school diploma, population density, and proportion of residents that own their house; and (iii) four ZIP Code-level meteorological variables: the summer (June to September) and winter (December to February) averages of maximum daily temperatures and relative humidity. We obtained ZIP Code-level meteorological variables using area-weighted aggregations based on daily temperature and humidity data on 4-km² gridded rasters from Gridmet via Google Earth Engine. We also considered two indicator variables indicating (i) the four geographic regions of the United States (Northeast, South, Midwest, and West) and (ii) calendar years to adjust for residual spatial and temporal confounding, respectively.

Statistical analysis

We applied Cox-equivalent re-parameterized Poisson models for each of the four outcomes with parallel computing. The Poisson re-parameterization, although mathematically equivalent to conventional Cox proportional hazards models (86), addresses the computational challenges faced by the conventional time-varying Cox proportional hazards model in such large data (3). Specifically, we fit a stratified quasi-Poisson model to estimate associations between the stratified rates of first hospital admissions with diagnosis codes of neurodegenerative disorder and time-varying annual mean PM_2.5 concentrations. The dependent variable was the count of outcome-related hospital admissions in each follow-up year, calendar year, and postal code location within strata specified by individual characteristics, using the corresponding total person-time of Medicare-fee-for-service beneficiaries as the offset. By stratifying on individual characteristics—sex, race, Medicaid eligibility, and age at study entry in 5-year categories—we allowed for flexible strata-specific baseline rates. To account for correlated observations from patients within the same ZIP Code across years, we applied an m-out-n bootstrap method using ZIP Code units to calculate CIs.

To adjust for potential confounding, we included area-level social economic, behavior risk factors, and meteorological variables in our analyses. To account for potential residual confounding by spatial and temporal trends, we included indicator variables for region and calendar years. The Anderson-Gill model was used to model time-varying covariates and address left truncation (87). This approach ensures that individuals contribute risk time only after entering and before exiting the open cohort, effectively handling lefts truncation due to individuals must survive until age 65 to become eligible for Medicare.

PM_2.5 sampling

Fine particulate matter (PM_2.5) samples for this study were collected from three locations across Asia, North America, and Europe, representing typical PM_2.5 exposure in urban environments. A comparison of these samples is presented in Table S3.

For atmospheric PM_2.5 collected from China (PM_2.5-1), samples were obtained from March 2011 to Feb. 2013 in Baoji, Shaanxi, China. The site is to the east of Baoji city and next to a commercial center where there is heavy traffic, but no industries are nearby. Other possible sources include open and residential biomass burning, and coal combustion processes for residential heating and energy generation. The sampler was set up 10 m above the ground on the rooftop of the Baoji Municipal Environmental Protection Bureau. PM_2.5 samples were collected with mini-volume samplers (Airmetrics, Oregon, USA) with a flow rate of 5 L min⁻¹ for 24 hours. The sampling was conducted every 6 days and a total number of 64 samples were collected. The PM_2.5 samples were collected on 47-mm quartz fiber filters (Whatman QM/A, Maidstone, UK), that were pre-combusted at 900°C for 3 hours. To minimize the evaporation of volatile components, each loaded sample filter was placed in a clean polystyrene petri dish and stored in a refrigerator at < 4°C prior to analysis.

Additional atmospheric PM_2.5 samples (PM_2.5-2) were collected in the USA, from the South Dekalb (SDK) site located in Atlanta, GA (33.6878°N, 84.2905°W). The SDK site is operated by the Georgia Environmental Protection Division (EPD) (88), and serves as a Southeaster US urban background site in the Atmospheric Science and Chemistry mEasurement NeTwork (ASCENT). This site is heavily forested with surrounding residential properties. However, it can be influenced by local traffic (Interstate I-285) and episodically biomass burning events. A major airport is located ~10 km to the southwest of the site. A total of two weeks of sampling was conducted from February 9, 2024, to February 22, 2024. PM_2.5 filter samples were collected over a one-week period on 47-mm PTFE filter (Teflo, Pall, USA) using a mini-volume sampler (URG-2000–30EH, URG, USA). The flow rate of the mini-volume sampler was 16.7 lpm.

The standard atmospheric PM_2.5 were purchased from Sigma-Aldrich (NIST2786, St. Louis, MO, USA; PM_2.5-3), which was prepared from atmospheric PM_2.5 collected from the air intake filtration system of a large exhibition center in Prague, Czech Republic in 2005. The collected total suspended particulates were resuspended, and a cyclone was used to remove coarse particles. This atmospheric PM_2.5 was certificated as Standard Reference Material by National Institute of Standard & Technology (NIST) in July 2021. The NIST2786 standard material was previously characterized for mass fractions of selected organic and inorganic constituents (89).

The filter membranes were cut into small pieces and sonicated in methanol: dichloromethane (2:1) for 20 minutes at 25% amplitude with 0.5-second pulse on/off cycle (Branson Digital Sonifier SFX 250, Danbury, CT, USA) to obtain an extract of total PM_2.5. Subsequently, the liquid containing PM_2.5 was placed in a Savant SpeedVac^™ Vacuum Concentrator (Thermo Fisher Scientific, Grand Island, NY, USA) to evaporate the solvent. Maintain the sample under vacuum at 60°C for 2 hours to obtain the dried total PM_2.5. The mass of the PM_2.5 is determined by measuring the difference in mass between the empty centrifuge tube and the tube containing the dried sample. The resulting sample was resuspended in Endotoxin-Free Dulbecco’s PBS (TMS-012-A, Sigma-Aldrich, St. Louis, MO, USA) at a final concentration of 5 mg/mL and stored at −20°C. Suspensions were sonicated for 30 seconds before use.

The PM_2.5 used in Fig. S13 and S15 are PM_2.5-1, PM_2.5-2, and PM_2.5-3, whereas for the other figures, unless otherwise specified, PM_2.5-1 was used.

Animal procedures

The following mouse strains (Mus musculus) were used: The C57BL/6 wildtype mice (strain 000664), humanized WT (hWT) αSyn mice (strain 010710), hA53T transgenic mice (strain 006823), and αSyn^−/− mice (strain 003692) were obtained from The Jackson Laboratory. App^NL-GF/NL-GF (Strain Name: App^tm3.1Tcs/App^tm3.1Tcs) mice were provided by Dr. Takaomi Saido (90). Animals were housed in ventilated cages with unlimited access to food and water. All the procedures were approved by the Johns Hopkins Animal Care and Use Committee and carried out following the guidelines from NIH for Care and Use of Experimental Animals.

Intranasal treatment with PM_2.5

The mouse was gently restrained by the scruff, positioning it in a supine state with the head slightly tilted backward. A pipette was then used to slowly administer 2 μL of PM_2.5 suspension (5mg/mL in PBS) into the nostrils once every two days, whereas the mouse was kept in the supine position until the PM_2.5 was absorbed. According to the high-resolution PM_2.5 exposure dataset used in epidemiological studies, the average concentration of PM_2.5 in the contiguous U.S. is 9.7 μg/m³. Adult mice, with a characteristic minute ventilation rate of 11–36 ml min⁻¹, inhale 0.15–0.5 μg of PM_2.5 per day at this exposure amount (91). The dosage of PM_2.5 administered intranasally to the mice corresponds to an inhalation exposure of 100–300 μg m⁻³ PM_2.5, a concentration one order of magnitude higher than the US average, but still atmospherically relevant during pollution events, such as urban haze episodes (92) and wildfires (92, 93). We acknowledge that nasal instillation delivers a high dose over a short period, differing from continuous inhalation exposure at a lower dose rate. A previous study (84) suggested that, at equivalent doses, initial lung burdens of nanoparticles were similar between instillation and inhalation, though instillation led to greater short-term retention. Nevertheless, the nasal instillation method was adopted in this study as it requires a lower amount of material, enabling us to examine the effects of authentic fine particulate matter rather than engineered nanoparticles.

Stereotaxic injection of PFF and PM-PFF

2–3 mon-old humanized WT αSyn mice (FVB; 129S6-Snca^tm1Nbm Tg (SNCA)1Nbm/J, Strain: 010710), obtained from The Jackson Laboratory (Bar Harbor, ME, USA), were used to identify the distinct propagation patterns of PFF and PM-PFF. Mice were deeply anesthetized with a ketamine mixture (100 mg/kg) and then secured on a stereotaxic apparatus. Unilateral injections of either PFF (10 μg) or PM-PFF (10 μg) were administered into the striatum at the following coordinates: anterior-posterior (AP), +0.2 mm; medial-lateral (ML), +2.0 mm; dorsal-ventral (DV), +2.8 mm (relative to the bregma). Mice injected with PBS served as the control group. The injections were performed using a 2 μL syringe (Hamilton, Reno, NV, USA) at a rate of 0.2 μL/min, and the needle was left in place for 10 minutes post-injection to ensure complete absorption of the solution. The needle was then withdrawn at a rate of 0.2 mm/min. Postoperative monitoring and care were provided to the mice. Behavioral tests were conducted 6 months post-surgery. All mice were euthanized, and tissues were collected for biochemical and histological analyses. Notably, for biochemical analyses, tissues were frozen and stored at −80°C. For histological analyses, mice were perfused with PBS and 4% PFA and transferred to 30% sucrose for cryoprotection (14, 17, 94).

Immunohistochemistry, immunofluorescence, mapping of αSyn pathology and volumetric analysis

Immunohistochemistry (IHC) and immunofluorescence (IF) staining were performed on 30 μm thick serial brain sections. The primary antibodies and their working dilutions were detailed in Table S8. For IHC studies, the blocked sections were incubated with primary antibodies against phosphorylated p-αSyn or tyrosine hydroxylase, followed by incubation with biotin conjugated anti-mouse/anti-rabbit antibodies (Vector Laboratories, Burlingame, CA, USA), Avidin-Biotin Complex (ABC)-HRP Detection Kit (Vector Laboratories, Burlingame, CA, USA), and ImmPACT^® DAB Substrate Kit (SK-4105, Vector Laboratories, Burlingame, CA, USA). The sections were then counterstained with Nissl (0.09% Thionin), dehydrated in graded alcohols and xylene, then mounted with DPX (06522, Sigma-Aldrich, St. Louis, MO, USA). IHC images were captured using an AxioCam Mrc camera connected to an Observer. Z1 microscope (Zeiss, Oberkochen, Germany). For IF staining, the sections were blocked in 10% goat serum for two hours before being transferred to primary antibodies and incubated overnight at 4°C. After washing with PBST, the sections were incubated at room temperature (RT) for four hours with appropriate secondary antibodies conjugated to Alexa-Fluor 488, 568, or 647, along with Hoechst. Following washes with PBST, the sections were mounted. IF images were acquired using a Keyence Microscope (BZ-X710, Itasca, IL, USA) and a confocal scanning microscope (LSM880, Zeiss, Dublin, CA, USA). All images were analyzed and quantified using the Fiji distribution package of ImageJ. The mapping of pS129 was performed by capturing images at 4x magnification using a Keyence microscope and subsequently stitching these images together to obtain whole brain images. The distribution of pS129 was then plotted based on the average distribution across 4 samples for each group. These sections corresponded at approximate +4.28, +2.68, +0.86, −0.58, −1.34, −2.06, −2.54, −3.28, −5.52, and −7.64 relative to Bregma. For volumetric analysis, every 10 coronal brain section (300 μm between sections) starting rostrally at bregma +0.14 mm to bregma −2.92 mm was mounted for each mouse. All mounted sections were stained with 0.09% Thionin at RT for 20 minutes, then washed in ultrapure water, formic acid, 80%, 90%, 100%, 100% ethanol, xylene, xylene for 10 minutes. The sections were finally mounted with DPX. The stained slices were imaged with the Keyence. The volume was calculated using the following formula: volume = (sum of area) × 0.3 mm. For hippocampus, quantification started from bregma −1.22 mm and ended at bregma −2.92 mm. For posterior lateral ventricle, quantification started from bregma 0.14 mm and ended at bregma −2.92 mm. For entorhinal cortex, quantification started at bregma −2.3 mm and ended at bregma −2.92 mm. Sholl analysis was manually performed for analyzing the morphology of microglia (36–38).

αSyn purification, PFF and PM-PFF preparation

The purification process for recombinant human αSyn monomers was carried out as described in previous studies. In brief, pRK172-αSyn plasmid was transduced in BL21 (DE3) cells and cultured overnight at 37°C. This culture was then scaled up in LB broth containing ampicillin and centrifuged to collect E. coli pellets. For lysis, the pellets were resuspended in osmotic shock buffer and homogenized, followed by osmotic shock and centrifugation to remove cell debris. The clear supernatant was then dialyzed against buffer to remove impurities. αSyn monomers were purified using Fast Protein Liquid Chromatography (FPLC) on an anion exchange column, collecting fractions that contain the target protein. Subsequently, bacterial endotoxins were removed from the purified αSyn monomers using the ToxinEraser Endotoxin Removal Kit (L00338, GenScript Biotech Corp., Piscataway NJ, USA). The αSyn monomer solution was centrifuged at 12,000 rpm for 20 minutes at 4°C to remove aggregates, and the supernatant was transferred to a new Eppendorf tube. The αSyn monomers were dissolved in PBS at a concentration of 2 mg/mL and incubated at 37°C while stirring at 1,000 rpm for 7 days to generate PFF. The fibrils were sonicated for 30 seconds (0.5-second pulse on/off cycle) at 20% amplitude before use. PM-PFF was prepared as described above by adding 5% PM_2.5 in αSyn fibrillization reaction. All the αSyn monomer, PFF and PM-PFF were kept at −80°C. The purity was assessed using Coomassie brilliant blue staining and immunoblot (14, 17, 94).

Behavioral analysis

Behavioral analyses were performed 10 days prior to the euthanasia of mice to assess behavioral deficits induced by PFF, PM-PFF, and PM_2.5 exposure. Tests conducted included the pole test, open field, nest building, grip strength, Y-maze, and novel object recognition. Sample sizes were determined based on previously published studies using similar experimental paradigms. To minimize bias, animals were administered by experimenters who were blinded to the treatments and randomly assigned to groups. Mice that developed infections due to treatment and required euthanasia prior to the experimental endpoint were excluded from analysis.

Pole test

In the pole test, a metal rod, 75 cm in length and 9 mm in diameter, wrapped in adhesive gauze, was utilized. The rod was cleaned with 70% ethanol before and after each test. Mice were placed 7.5 cm from the top of the pole with heads upward. On the first day, mice were induced to turn and descend to the bottom of the pole 3 times; on the second day, unguided training was conducted 3 times; on the third day, the time required for turns and the total time to reach the bottom were recorded for 3 times. The maximum cutoff time for ending the experiment was set at 2 minutes. Both turnaround time and total time were reported (13, 14, 16, 17, 94).

Open field

For the open field test, an apparatus consisting of a cubic plastic box (40 cm × 40 cm × 40 cm), divided into 36 equal areas (6.6 cm × 6.6 cm), was used. The field was subdivided into central and peripheral areas, with the central area comprising four central squares (2 × 2). Prior to testing, mice were placed in the apparatus room to acclimatize for about 30 minutes. Then, the mice were allowed to freely explore the box for 30 minutes. The field was cleaned with 70% ethanol between each test. A tracking system (PAS software) monitored the mice’s trajectories, recording the beam breaks as an assessment of locomotor activity. Counts occurring in the center served as an anti-anxiety index (13, 52, 53, 59).

Nest building

The nesting test was conducted using a standard method. Mice were required to use orofacial and forelimb movements, such as pulling nesting material apart with forelimbs and teeth and incorporating shredded material into their bedding. Briefly, mice were individually placed in a cage containing a new 2.5 g nestlet (5 cm × 5 cm, Johns Hopkins Medical Research Animal Resources Center, Baltimore, MD, USA). Nest building scores were evaluated after 16 hours (13, 25–27).

Grip strength

Grip strength testing, an indicator of neurodegenerative disease progression, was performed by placing a grid-equipped dynamometer on a table. Mice were gently held by the tail with their forelimbs or all limbs on the grip strength grid. The mice were gently pulled downward, and the maximum reading displayed by the dynamometer was recorded (13, 14, 16, 17, 94).

Y-Maze

The Y-maze was used to assess the spatial memory of mice. The Y-shaped maze consisted of three symmetric arms, each 40 cm long, 10 cm wide, and 15 cm high, with adjacent arms forming a 120-degree angle. During the training phase, one arm (novel arm) was blocked with a removable barrier, allowing mice to freely explore from the starting arm to familiar arm for 8 minutes. After training, mice were returned to their home cages. Two hours later, the test was conducted by removing the barrier, and mice were again placed in the starting arm to freely explore all three arms. The Y-maze was cleaned with 70% ethanol between successive tests. The Any-Maze software recorded and analyzed the time spent in the novel arm, total distance moved, and alternation count (13, 53, 54).

Novel object recognition

The novel object recognition test was conducted following previously reported methods. The apparatus, an opaque polyethylene plastic box (45 cm wide × 45 cm deep × 50 cm high) was used. Initially, all mice were placed in the box without any objects for five minutes. Two identical objects (Object A) were then placed in the box, and mice were allowed to explore for five minutes. The time spent by the mice on each object was recorded during the training session. One hour after training, one Object A was replaced with a new Object B, identical in material and size but different in shape. Mice were then placed in the box to explore for five minutes, and the time spent on Object A or B was recorded (defined as the short-term test phase). The box and objects were cleaned with 70% ethanol after each trial. The video tracking system (ANY-Maze software) recorded the mice’s behaviors, considering sniffing, biting, or facing the objects as exploratory activities. Data were defined as the percentage of new object performance = time spent on new Object B / (time on Object A + time on new Object B) × 100 (13, 55, 56).

Amplification of pathogenic αSyn with seeding amplification assay (SAA)

The amplification of αSyn strains from mouse brain homogenates were performed using SAA method by referring to previous work (23). The SAA equipment included a microplate horn (431MPX), a sound enclosure (432MP), and a thermoelectric cooler (4900) purchased from Qsonica (Qsonica, Newtown, CT, USA). Briefly, recombinant αSyn monomers were centrifuged at 100,000 g for 30 minutes at 4°C to remove any preformed aggregates before use. The αSyn was then diluted in SAA buffer (1% Triton X-100 in PBS) and 100 μL was transferred to PCR tubes containing an appropriate amount of silicon beads (d = 1.0 mm, 11079110z, BioSpec Products, Bartlesville, OK, USA), with 10 μL of insoluble brain homogenate or CSF samples from LBD patients (5 PDD and 2 DLB), PD without dementia patients (n = 5) or health control (HC, n = 5) added as a seed. The final concentration of αSyn was 0.5 mg/mL. After mixing, the samples were subjected to sonication (amplitude: 5; 40 seconds sonication followed by incubation at 37°C for 29 minutes and 20 seconds). The amplification monitored by measuring the ThT fluorescence.

ThT fluorescence assay

5 μL aliquot of the sample was added into 55 μL ThT solution (20 μM). Subsequently, the mixture was dispensed in triplicate into a 384-well black/clear bottom plate (P6491, Sigma-Aldrich, St. Louis, MO, USA). The fluorescence was measured using a microplate reader (Varioskan LUX microplate reader, Thermo Fisher Scientific, Grand Island, NY, USA) with excitation/emission at 450/485 nm (23).

Circular dichroism (CD) spectroscopy measurements

The far-UV CD spectra of PFF and PM-PFF were recorded using a spectrophotometer (Aviv 420, Aviv Biomedical, Lakewood, NJ, USA) in a 1 mm path length cuvette. The scanning range was set from 190 to 250 nm at 25°C, with a scan speed of 50 nm/min, data interval of 1 nm, and bandwidth of 2 nm. The average of three scans was taken for all spectra. Noise in the signal was reduced using Origin software to smooth the CD data (95).

Proteinase K digestion, dot blot and silver staining

The SAA amplified samples (8 μg) or PFF (8 μg) were mixed with 1μL proteinase K (PK, 0.1 mg/mL) and PBS to achieve a final volume of 20 μL. The mixture was incubated in a water bath at 37°C for various time points (0, 5, 15, and 30 minutes) and the digestion was terminated by the addition of 2 μL of PMSF (40 mM). For dot blot assay, 2 μL of the PK-digested samples were loaded onto the nitrocellulose membrane (1620112, Bio-Rad, Hercules, CA, USA) and blocked with 5% milk (1706404XTU, Bio-Rad, Hercules, CA, USA) in TBST for one hour at RT. The membrane was then incubated with mouse anti-αSyn monoclonal antibody (mAb) (1:2000, 610787, BD Biosciences, Franklin Lakes, NJ, USA) at 4°C overnight. After washing with TBST (four times, 10 minutes for each) the membrane was incubated with anti-mouse IgG-HRP (1:5000, 31430, Thermo Fisher Scientific, Grand Island, NY, USA) at RT for one hour. Following additional TBST washes, the chemiluminescent signal was developed using SuperSignal West Pico Plus chemiluminescent substrate (34096, Thermo Fisher Scientific, Grand Island, NY, USA). All images were processed using an Amersham Image 600 (GE Healthcare Life Sciences, Chicago, IL, USA) (23).

Primary neuronal culture and PFF/PM-PFF transduction

CD1 mice were obtained from the Jackson Laboratory (Bar Harbor, ME, USA). Primary cortical neurons were extracted from the cortex of E15.5 embryos, with careful removal of the meninges and hippocampus. The dissected tissues were then digested in 0.25% trypsin-EDTA at 37°C for 15 minutes in a water bath. Gently pipetting into a single-cell suspension, then passed through a 40 μm nylon mesh. The single-cell suspension was seeded on tissue culture plates pre-coated with poly-L-lysine and cultured in Neurobasal medium supplemented with B27, 0.5 mM L-glutamine, penicillin and streptomycin. The inhibitor 5-Fluoro-2’-deoxyuridine (1 mM) was added 3 days after seeding. After culturing in vitro for 7 days, primary neurons were treated with sonicated PFF/PM-PFF at a final concentration of 5 μg/mL, incubating for another 10 or 21 days before pathological or toxicity assay (14, 17, 94).

Human RNA-Seq data processing

We obtained raw FASTQ files from a previous study (24) that performed bulk RNA sequencing of anterior cingulate cortex samples from 6 PDD, 6 DLB, 7 PD patients, and 6 non-neurological control samples. We processed the FASTQ files using the SPEAQeasy (96) RNA-seq pipeline, which integrates several bioinformatics tools to convert raw FASTQ files into ready-to-use R objects for differential gene expression analysis. In brief, reads were initially quality checked with FastQC (24, 97) v0.11.8, followed by trimming with Trimmomatic (98) v0.39 when necessary. The quality-checked reads were then aligned to the human reference genome (hg38/GRCh38) using HISAT2 (99) v2.2.1. Finally, gene-level expression was quantified using featureCounts (100) based on GENCODE release 25 annotations.

Mouse RNA sequencing

RNA was extracted from the anterior cingulate cortex of mice treated under each condition, with four replicates per condition. 2-month-old humanized WT αSyn mice (FVB; 129S6-Snca^tm1Nbm Tg (SNCA)1Nbm/J, Strain: 010710), obtained from The Jackson Laboratory (Bar Harbor, ME, USA), were used for RNA sequencing experiments. The mice were divided into five groups: the PBS group (intranasal administration of PBS), the PM_2.5 group (intranasal administration of PM_2.5), the PFF group (PFF injection), the PM-PFF group (PM-PFF injection), and the PM-PFF with PM_2.5 exposure group (PM-PFF injection followed by intranasal PM_2.5 administration one week later). All treatments were administered for a duration of 2 months. Sequencing libraries were prepared using the RNeasy Lipid Tissue Mini Kit, following the manufacturer’s protocol. The RNA-seq data were processed using the same SPEAQeasy RNA-seq pipeline, with reads aligned to the mouse genome (mm10) and gene-level expression quantified based on GENCODE release M25. For principal component analysis (PCA) of mouse RNA-seq profiles, lowly expressed genes were first filtered out using the filterByExpr function from the edgeR (101) package. The filtered data were then normalized using the trimmed mean of M-values (TMM) method, implemented in the calcNormFactors function. PCA was performed on the mouse RNA-seq data using the top 500 most variable genes, determined by variation of expression concentrations measured as log₂ normalized counts per million (CPM) across samples.

Differential gene expression analysis

Differential gene expression analysis for both human and mouse RNA-seq datasets was conducted using the DESeq2 (102) package. We retained genes with at least 10 reads in at least 6 samples for the human dataset and in at least 4 samples for the mouse dataset. For the human RNA-seq data, we adjusted for potential confounders in the regression model, including sex, age at death, post-mortem interval (PMI), and RNA integrity number (RIN). We only included condition status for the mouse data in the regression model to detect group difference of gene expression. To enhance visualization and ranking of genes, we applied the apeglm method (103) for effect size (Log₂FC) shrinkage implemented in DESeq2.

Gene set enrichment analysis

Gene Set Enrichment Analysis (GSEA) was conducted using the clusterProfiler (104) package, focusing on gene sets from Gene Ontology (GO) related to biological processes. We set the minimum gene set size (minGSSize) to 20 and the maximum gene set size (maxGSSize) to 500. The input for GSEA was a ranked gene list derived from the shrunken effect sizes obtained with DESeq2. The R package ‘aPEAR’ (105) was used to identify clusters of similar gene sets detected by GSEA, utilizing the default parameters of the “enrichmentNetwork” function for gene sets clustering and network visualization.

Comparison of transcriptional responses in mice and human patients

We compared the transcriptional responses in mice under each condition to those in human patients with PDD, PD without dementia, and DLB. Following a previously published approach (106), comparison was limited to genes that showed differential expression in humans, defined by a p-value < 0.05 and an absolute effect size (log₂FC) greater than 1.2. To ensure a sufficient number of genes for cross-species comparison, we included mouse genes with only nominal evidence of differential expression (p-value < 0.05) without restricting the magnitude of effect sizes. The strength of correlation was assessed using Spearman correlation. We applied a binomial test to assess the statistical significance of the percentage of genes with consistent directional effects between mice and humans, assuming an expected 50% consistency by chance.

Quantification and Statistical Analysis.

All data (except epidemiological and transcriptomic data) were analyzed using GraphPad Prism 8. Statistics Data are presented as the mean ± SEM (Standard Error of the Mean) or presented as violin plots showing all individual data points. With at least 3 independent experiments. Representative morphological images were obtained from at least 3 experiments with similar results. Statistical significance was assessed via a t-test or one way ANOVA test followed by indicated post-hoc multiple comparison analysis. Assessments with p < 0.05 were considered significant.

Supplementary Material

Figures S1–S24

Tables S1–S8

Data and materials availability:

Raw Medicare data used in the epidemiological study were obtained from the Centers for Medicare & Medicaid Services under a data sharing agreement and we are not permitted to directly share the third-party raw data used in the analyses. The raw and processed RNA-seq data generated in this study have been deposited in the Gene Expression Omnibus (GEO) under accession number GSE297423. The code used for epidemiological data analysis and RNA-seq is publicly available on Zenodo at https://doi.org/10.5281/zenodo.15374633. Other data are available in the main text or the supplementary materials and from Dryad.