Mouse brain transcriptome responses to inhaled nanoparticulate matter differed by sex and APOE in Nrf2-Nfkb interactions

The neurotoxicity of air pollution is undefined for sex and APOE alleles. These major risk factors of Alzheimer’s disease (AD) were examined in mice given chronic exposure to nPM, a nano-sized subfraction of urban air pollution. In the cerebral cortex, female mice had two-fold more genes responding to nPM than males. Transcriptomic responses to nPM had sex-APOE interactions in AD-relevant pathways. Only APOE3 mice responded to nPM in genes related to Abeta deposition and clearance (Vav2, Vav3, S1009a). Other responding genes included axonal guidance, inflammation (AMPK, NFKB, APK/JNK signaling), and antioxidant signaling (NRF2, HIF1A). Genes downstream of NFKB and NRF2 responded in opposite directions to nPM. Nrf2 knockdown in microglia augmented NFKB responses to nPM, suggesting a critical role of NRF2 in air pollution neurotoxicity. These findings give a rationale for epidemiologic studies of air pollution to consider sex interactions with APOE alleles and other AD-risk genes.


Introduction
Air pollution is a major global environmental risk factor of morbidity and mortality across the human lifespan (Landrigan et al., 2018;Shaffer et al., 2019;Finch, 2018). Air pollution exposure is also associated with neurodegeneration, accelerated cognitive decline of aging and increased risk of Alzheimer's disease (AD) (Kulick et al., 2020;Calderó n-Garcidueñas et al., 2020). However, little is known of interaction of air pollution neurotoxicity for sex and APOE alleles and other AD risk factors (Finch and Kulminski, 2019).
Epidemiological studies of air pollution neurotoxicity have not identified interactions of gender by APOE alleles. Findings are typically 'adjusted or controlled' for gender differences (Clifford et al., 2016;Chen and Schwartz, 2009;Ailshire and Clarke, 2015;Gatto et al., 2014). In the WHIMS cohort of elderly women, APOE4 homozygotes had a greater risk of dementia and accelerated cognitive decline (Cacciottolo et al., 2017). The APOE4 vulnerability for accelerated cognitive aging was recently extended to ozone, as well as PM10 and PM2.5 in a large sample of both sexes from New York City (Kulick et al., 2020). A recent study from China suggested greater male vulnerability to air pollution for verbal test deficits . Sex-APOE interactions for air pollution neurotoxicity remain undefined. In a small sample from polluted Mexico City, APOE4 heterozygous females with high BMI had higher risk of severe cognitive deficit than other groups (Calderó n-Garcidueñas and de la Monte, 2017). Developmental air pollution exposure has received greater attention for gender because of the consistent male excess vulnerability for behavioral and cognitive deficits in the pre-adolescent and young adult (Chiu et al., 2013;Sunyer et al., 2015).
Mouse models have not addressed sex and APOE in responses to air pollution. Our initial study examined female EFAD (APOE-TR/5xFAD-Tg +/-) mice carrying transgenes for familial AD genes (5xFAD) together with human APOE alleles by targeted replacement (APOE-TR), which had APOE-e3 +/+ (E3; APOE3) or APOE-e4 +/+ (E4; APOE4) genotype. Consistent with WHIMS findings, E4FAD female mice accumulated more brain amyloid in response to nPM than the E3FAD (Cacciottolo et al., 2017). However, for ozone exposure, male APOE-TR showed the converse of greater behavioral impairments in APOE3 than APOE4 (Jiang et al., 2019). For further study of both sexes, we examined the cerebral cortex transcriptomic responses of APOE-TR and wildtype mice (C57BL/6J) by RNAseq for the main regulators of air pollution toxicity in AD pathways.
Sex and APOE alleles can also alter NRF2 and NFKB activities, as shown for the larger response of female mice for hepatic NRF2 activation by phenobarbital and oxazepam and other xenotoxins (Rooney et al., 2018a). NRF2 downstream genes including Gsta2, Ho1, and Nqo1 showed lower hepatic expression in APOE4-TR than APOE3-TR mice (Graeser et al., 2011). We therefore examined both sex and APOE allele for interactions with NRF2/NFKB responses of air pollution neurotoxicity.

Results
To define brain transcriptional responses of air pollution and interactions with sex and APOE alleles, we examined responses of adult C57BL/6J (wild type, 'B6') and B6 mice carrying human APOE3 and APOE4 alleles by targeted replacement (APOE-TR) to nPM, a subfraction of ultrafine PM (PM0.2). Three independent exposures used different batches of nPM at specified durations of exposure (details on sample collection and chemical composition in Figure 1-figure supplement 1 and Zhang et al., 2019). In vitro studies on BV2 microglia examined the role of NRF2 and NFKB in responding inflammatory pathways.
RNAseq data were stratified to identify sex-and APOE-specific nPM responses by linear models and by weighted gene co-expression (WGCNA). Females had more DEG than males for both B6 and APOE-TR mice, by up to two-fold ( Figure 1D). Females of APOE3 and B6 had the most nPM responding genes (153 vs 189, respectively). Gene modules identified by WGCNA also had more female responses to nPM for B6 and APOE-TR ( Figure 1D). Modules were constrained to a modules (kME intermodule connectivity) were used for IPA analysis. Significance was calculated from the Pearson correlation of eigengene of the modules with nPM. (E) Upstream regulators and F) canonical pathways associated with nPM transcriptome responses in B6 and APOE-TR mice. Solid horizontal lines separate responses that are shared and sex-specific. Heatmaps were sorted by the sum of -log10 (p-values) in each row. p-values<10 À6 were converted to 10 À6 for better visualization; grey, not significant. RNAseq sample size was 4/genotype/sex/treatment. The online version of this article includes the following figure supplement(s) for figure 1:    maximum of 150 hub genes, based on connectivity (Materials and methods). Both analyses (DE, WGCNA) showed more nPM-responding gene responses for APOE3 than APOE4.
Upstream regulators and canonical pathways were identified by IPA for sex-specific and shared nPM responses. The top upstream candidate was Nfe2l2 (Nrf2), a regulator of Phase II detoxification ( Figure 1E, Figure 1-figure supplements 2-3), which had the strongest associations for B6 and APOE4. Sexes differed in immune-related upstream regulators of gene responses. Female-specific responses included Pparg (peroxisome proliferator activated receptor gamma), Sp1 (specificity pro-tein1 transcription factor), and Tnfsf11 (TNF superfamily 11). Male-specific responses included Traf6 (TNF associated receptor factor 6), Camk2 (regulator of synaptic plasticity and AMPA glutamate receptors), and Ccdc22 (regulator of NFKB signaling by interaction with COMMD proteins). These results paralleled the enrichment of NRF2 and immune response pathways in the combined multivariate model above.
Stratified analysis by APOE and sex for canonical pathways showed nPM responses of neuronal pathways; for example G-protein-coupled receptors, axonal guidance, ephrin receptors, synaptic long-term depression, and endocannabinoid development neuron pathway ( Figure 1F). Other nPM responding genes were related to relaxin, GM-CSF, and c-AMP-mediated signaling. Female-specific responses include genes associated with the following pathways: AMPK, dopamine-PARPP32 feedback in cAMP, gap junction signaling, and nitric oxide production. Male-specific responses in both mouse strains were enriched for 'cardiac hypertrophy' signaling; for example, Elk1 (transcription factor) and Hsp27 ( Figure 1F). APOE3 and APOE4 of both sexes had different inflammatory responses for NFKB, IL6, CREB, and IL22 pathways (

Baseline effects of APOE4 allele and the overlap with nPM responses
Baseline differences by APOE in non-exposed controls were analyzed by sex in two steps. The combined multivariate model showed 133 DEGs differed in baseline APOE allele effect (5% FDR) ( Figure 2A). These DEGs were enriched for immune-related pathways including rheumatoid arthritis, granulocyte adhesion, IL10, and NFKB signaling. APOE4 baseline differences included pathways of glutamate metabolism, and production of nitric oxide, superoxide and hydrogen peroxide, the LXL/ RXR pathway of cholesterol efflux, and atherosclerosis.
AD-related pathways differed by sex and APOE alleles for nPM responses ( Figure 3A-B). Only female APOE3 responded to nPM in five AD pathway genes for amyloid precursor protein (APP) processing and for tau: App, Bace1, Psen1 ( Figure 3A); Tau and Gsk3b ( Figure 3B). In contrast to mRNA changes in amyloidogenesis genes, the levels of amyloid peptides (Ab40 and Ab42) were not affected by nPM exposure in the cerebral cortex of APOE-TR mice ( Figure 3C). However, Ab40 peptide had a 50% lower baseline in males than females of both APOE3 and APOE4-TR mice ( Figure 3C). Ab40 peptide also had a significant negative correlation with mRNA levels of amyloidogenesis pathway including App (r, À0.65) and Psen1 (r,À0.51 to À0.67) for both APOE3 and APOE4-TR cerebral cortex ( Figure 3D).

Sex-and APOE-specific nPM-mediated NFKB responses
Next, we examined genes of the NFKB pathway, which regulate pro-inflammatory responses to nPM, as shown for responses of wildtype mice (B6 males) in hippocampus to nPM (Woodward et al., 2017a). In cerebral cortex, APOE-TR mice responded to nPM with a subset of genes downstream of NFKB (13%, 8/133) that differed by sex and APOE ( Figure 4A). Two clusters of nPM responses were identified by Principal Component Analysis for these 133 NFKB downstream genes: Principal component (PC) 2 (20% of variance, nPM: sex interaction) and PC4 (2.5% of variance nPM:APOE4 interaction) ( Figure 4B).

Inhibitory effects of NRF2 on NFKB response to nPM
The relationship of NRF2 and NFKB responses of nPM was further explored in an independent doseresponse experiment. The duration of inhalation exposure was only 3 weeks to assess the initial responses of nPM by male C57BL/6 mice. After 3-week exposure to 300 mg/m 3 nPM, the cerebral cortex had nuclear translocation of NRF2 protein (+50%) and increased cytosolic NFKBP65 (+25%) ( Figure 6A). Downstream of NRF2, the rate limiting enzyme of glutathione synthetase (GCLC) had dose-dependent increase correlated with Nrf2 mRNA (r, 0.6, p=0.005) ( Figure 6B). Nrf2 and Nfkb responses of B6 male mice at three nPM doses had opposing changes of increased Nrf2 mRNA, but decreased Nfkbp65, Nfkbp50 mRNA and IL2 protein, all with dose-dependence ( Figure 6C). (C) Levels of Ab40 and Ab42 peptides did not respond to nPM. However, females had higher Ab40 (pg/ml) of both APOE3 and APOE4 mice. **p<0.01, *p<0.05 in t-test. (D) Ab40 peptide was negatively correlated with amyloidogenesis gene expression including App (r = À0.65), and Psen1 (r = À0.51 to À0.67). The expression was reported as Log2 count per million (cpm). (E) Abamyloid clearance pathway responses to nPM. A small subset (10%, 5/46) of amyloid clearance genes differed by APOE or nPM (genes identified in the IPA database for phagocytosis, proteolysis, degradation, deposition). Only APOE3 responded to nPM. Mean ± SEM. ANOVA; FDR multiple test correction. * Adj. p-value, 0.05. Sample size: 4/genotype/sex/treatment.
NRF2 and NFKB are potential regulators of the nPM responses in the cerebral cortex and mixed glial culture The top canonical pathways related to nPM exposure in the cerebral cortex of adult mice comprised of calcium signaling, HIF1a signaling, circadian rhythm pathway, AMPK signaling, SAPK/JNK signaling, endocannabinoid pathways and NRF2 oxidative stress responses ( Figure 1A). These highly interconnected pathways comprise a larger network of oxidative and inflammatory responses ( Figure 7). Thus, the hub regulators of nPM responses could broadly affect these pathways. Using IPA analysis, we built two networks from the NRF2 and NFKB downstream genes in the identified canonical pathways. Exposure to nPM caused expression changes in four of these genes in the cerebral cortex of adult mice: Smarca4 (+25%), Cftr (+25%), Hdac1 (À65%), and Vegfc (À65%).
We hypothesized that these networks are among the initial responses to nPM exposure and contributed to chronic damage at later stages. The transcriptional responses of these networks were examined in mixed glial culture after 24 hr exposure to 10 mg/ml nPM (Dataset from prior nPM studies Woodward et al., 2017a). Major increases of both Nrf2 (300%) and Nfkb (400%) mRNA were induced by acute nPM exposure (Figure 7). A large portion of genes from the constructed networks was also upregulated by nPM: 17 NRF2 and 30 NFKB DEGs from these downstream networks. These genes were related to all selected signaling pathways: HIF1A, calcium, AMPK, SAPK/JNK, and the endocannabinoid system. Several of these genes are also transcriptional regulators that could initiate further downstream changes, body-wide. Some of these genes include Brca1 (+30%), Atf4 (350%), Smarca4 (50%), Hif1a (400%), and Stat3 (350%) (Figure 7).

Discussion
These finding address the gap in how sex and APOE allele interactions may alter neurodegenerative responses to air pollution. Cerebral cortex genomic responses to nPM, were examined in C57BL/6 mice (B6, wildtype) and B6 mice carrying human transgenes for APOE3 and APOE4 alleles. Female mice had two-fold more genes that responded to nPM, further enhanced for APOE3. Responding genes included neuronal pathways (e.g. axonal guidance; glutamate synapse genes), inflammation (e.g. AMPK, and APK/JNK signaling), and antioxidant and hypoxic signaling (e.g. NRF2, HIF1A signaling). Genes in pathways downstream of NFKB and NRF2 responded oppositely to nPM. These interactions of NRF2 and NFKB may modulate sex and APOE risk for AD and accelerated cognitive aging during air pollution exposure.
The nPM responding genes may help to identify GxE in neurodegenerative risks from air pollution and cigarette exposure. For example, a combination of air pollution and a specific IL1B variant increased the risk of Parkinson disease (Lee et al., 2016). Our findings extend microarray analysis of frontal cortex of children and young adults from Mexican cities differing in air pollution levels: the 134 responding genes include inflammation (e.g. NFKB) and antioxidant responses (e.g. GPX2, GPX3) (Calderó n- Garcidueñas et al., 2012). Microarray analysis of rat brain chronically exposed to PM0.2 also overlapped with our results: S100a9, calcium channels (e.g. Cacna1i), and glutamate receptors (Ljubimova et al., 2013).
These glutamatergic gene responses extend findings that hippocampal neurites are selectively sensitive to nPM (Morgan et al., 2011;Woodward et al., 2017b;Fonken et al., 2011). The increased levels of the ionotropic receptor NMDA type subunit 1 (Grin1) is notable: a GRIN polymorphism is associated with the risks of Parkinson (Wu et al., 2010), schizophrenia (Demontis et al., 2011), and also interacts with APOE alleles for earlier AD onset (de Oliveira et al., 2016), GRIN variants might also alter air pollution neurodegenerative responses, for example mutations in GRIN2A and GRIN2B increased the risk of cognitive impairments for lead poisoning of children (Rooney et al., 2018b). The nPM responses of Grin2a and Grin2b mRNA, while modest (p=0.05-0.06), merit further study among xenobiotics.
AD-associated genes responded differently to nPM by sex and APOE allele. APOE3 females had the lowest baseline and the highest nPM response in genes associated with amyloidogenesis and TAU. The smaller responses of APOE4 mice to nPM may be a ceiling effect, because about 7% of the nPM responsive genes also had APOE4 baseline differences in both sexes. For non-exposed controls (baseline), APOE4 vs APOE3 differed in 300 genes related to known APOE pathways including LXR/RXR (Courtney and Landreth, 2016), Atherosclerosis (Mahley et al., 2009), and Rheumatoid arthritis (Toms et al., 2012). A similar APOE allele specificity was found in responses to ozone (O 3 ), which impaired memory in APOE3, but not APOE4-TR mice (Jiang et al., 2019). This pattern may be compared with ceiling effects of aging on responses to air pollution which we observed in middle-aged B6 mice. Both sexes at age 18 months have elevated baseline levels of antioxidant and inflammatory gene mRNA and protein, which did not respond to nPM or O3, unlike young mice (Jiang et al., 2019;Woodward et al., 2017b;Zhang et al., 2017a).
Brain amyloid must also be considered in the complex interactions of APOE genotype and sex with air pollution. The current study of APOE-TR mice response to nPM and a study of response to ozone (Jiang et al., 2019) showed that neither air pollutant altered brain Abeta40 and Abeta42 peptide levels for mice of ages 3 to 18 months (young to middle-age). The APOE-TR mice have the normal murine wildtype amyloid peptide Abeta42, which differs from the human in three amino acids and aggregates less avidly (Otvos et al., 1993). Thus, our study in APOE-TR mice do not include additional stress due to amyloid aggregates present in the FAD transgenic models. Unlike APOE-TR, nPM exposure increased the amyloid plaque loads in EFAD (APOE-TR plus five mutation in amyloidogenic genes) and J20 AD mouse models (Cacciottolo et al., 2017;Cacciottolo et al., 2020). In EFAD mice, nPM caused greater amyloid plaque load in E4FAD than E3FAD. Even in that study, nPM increased levels of Abeta oligomers, but not plaque load in E3FAD more than for E4FAD mice. These differences suggest the hypothesis that APOE alleles cause differences in amyloid clearance, phagocytosis, and proteolysis. As expected, some nPM-responding genes that mediate amyloid clearance differed by the APOE allele. The complement factor C3, which mediates amyloid clearance (Maier et al., 2008), had a higher baseline in APOE3, but lacked nPM response, whereas Vav and S100 calcium-binding protein A9 (S100a9) responses to nPM were restricted to APOE3. Vav and S100a9 regulate phagocytosis of fibrillary Abeta by microglia (Wilkinson et al., 2006;Kummer et al., 2012). The APOE3 and 4 proteins, and Abeta compete differentially for uptake by astrocytes via the lipoprotein receptor-related protein 1 (LRP1)-dependent pathway  Figure 1A). The network was made by IPA software. The networks were overlaid with the significant responses to nPM in mixed glial culture (Woodward et al., 2017a). The numbers indicate log2 fold-changes of gene expression in Affymetrix microarray. Dataset from prior studies (Woodward et al., 2017a). In vitro sample size, 4/group. (Verghese et al., 2013). Moreover, APOE4 astrocytes have less efficient Abeta clearance, attributed to acidification of endosomes (Verghese et al., 2013;Prasad and Rao, 2018). Possibly, an efficient clearing system could compensate even for a greater amyloidogenic effect of nPM in APOE3 than APOE4 mice. This hypothesis will be tested further in vitro and in vivo.
Similar to the amyloid pathway, APOE4 had higher baseline level of inflammation, while their response to nPM was less than the APOE3. This allele specificity was also shown for responses to ozone (Jiang et al., 2019). In contrast, responses to LPS endotoxin by intra-peritoneal injection were greater in APOE4 male mice for microglial activation and systemic inflammatory cytokines . In vitro, APOE4 macrophages also had higher induction of NFKB, TNFA, and heme oxygenase one in response to LPS (Jofre-Monseny et al., 2007). While air pollution can include endotoxins, the comparisons with injected LPS for sex and APOE are limited because the systemic responses are downstream of the lung, which receives most inhalants.
Female mice had greater response to nPM than males for immune and antioxidant pathways, for wildtype and APOE-TR mice. Sex hormones and early life gonadal programming during brain development could underly these differences. Ex-vivo microglial cultures from male and female brains had divergent inflammatory response to estradiol (E2) and LPS (Loram et al., 2012). A gene expression microglial developmental index showed a sex difference in maturation and immune reactivity, which correlated with the risk of AD and autism spectrum disorders (Hanamsagar et al., 2017). In SH-SY5Y neuroblastoma cells, E2 increased cell survival and NRF2 antioxidant defense against homocysteine (Chen et al., 2013). Similarly, E2 treatment of postnatal rats ameliorated acute ethanolinduced oxidative stress, neuroinflammation, and neuronal cell death through increase of sirt1, P53 acetylation inhibition, and reduction in phospho-NFKB nuclear localization (Khan et al., 2018). Higher adaptive genomic response by females might favor faster detoxification or recovery from air pollution. Metformin mediated NRF2 activation could ameliorate tight junction proteins, blood-brain barrier (BBB) integrity, reduce inflammation and oxidative stress, and also normalize the levels of BBB glucose transporter GLUT1 protein after cigarette smoke exposure in mice (Prasad et al., 2017).
As opposed to potential protective effects of estrogen hormones against air pollution, female mice had higher baseline levels of Abeta40 than males. This results parallels with female excess amyloid plaque load of aged AD mouse models (Callahan et al., 2001;Hirata-Fukae et al., 2008). Neonatally demasculinized or defeminized 3xTg-AD or other AD transgenic mice shows the major role of sex steroids in determining adult sex differences in Abeta accumulation (Carroll et al., 2010;Pike et al., 2009). Thus, sex steroids can act as a double-edged sword for amyloidogenic responses to air pollution and other environmental neurotoxins. Resolving the role of steroid hormones in this intricate relationship of background biology and air pollution requires studies on the recovery after nPM exposure in gonadectomized and older mice. We plan further studies of the complex interface of sex-APOE allele and nPM in mixed glial cultures derived from male and female APOE3 and APOE4-TR mice.
NFKB and NRF2 had opposite responses to nPM, that included downstream genes in wildtype B6 and APOE-TR. This divergence was also shown in a short term (3 weeks) exposure of B6 male mice. The NRF2 and NFKB crosstalk was validated in BV2 microglia. This is the first evidence for NRF2 and NFKB interactions in response to air pollution of both in vivo and in vitro models. These results parallel the LPS responses of monocytes, which showed redox-mediated transcriptional cross-talk between NRF2 and NFKB responses to LPS (Zhang et al., 2017b). Concurrent increase of nuclear NRF2 and cytosolic NFKBP65 in cerebral cortex after nPM exposure suggest that NRF2 activation can attenuate NFKB nuclear localization. We hypothesize involvement of KEAP1, the NRF2 repressing protein, which can mediate IKKB degradation and inhibit NFKB nuclear localization (Kim et al., 2010;Lee et al., 2009). Other mechanisms could be mediated by direct protein-protein interaction, and by secondary messengers. For example, NRF2 can inhibit NFKB through reduction of reactive species and suppress RAC1-mediated NFKB activation (Sanlioglu et al., 2001;Cuadrado et al., 2014). In contrast, NFKB can also inhibit NRF2 activity through enhancing the recruitment of histone deacetylase (HDAC3) to ARE region (Wakabayashi et al., 2010), or competing with NRF2 for binding to CH1-KIX domain of CBP protein inside the nucleus (Liu et al., 2008). These interactions are also shown for thein nematode C. elegans: the NRF2 homologue skn-1 and the antibacterial factor 2 (abf-2) responded rapidly to nPM, with persisting developmental effects .
Since immortalized BV2 microglial cells have limited comparability with in vivo microglial cells, we further corroborated the nPM responses of adult brain in primary mixed glial culture. In adult brain, nPM affected a network of genes from a set of highly interconnected canonical pathways including NRF2 oxidative stress response, HIF1A, AMPK, circadian rhythm, and endocannabinoid related signaling pathways. Interestingly, numerous genes from this network including Nrf2 and Nfkb mRNA were upregulated in mixed glial culture after 24 hr acute exposure to nPM. A large portion of this gene network were considered as downstream of NRF2 and NFKB transcriptional factors. Thus, these two molecules are potentially the hub upstream regulators of long-term nPM neurotoxic effects. This hypothesis remained to be tested in further in vitro and in vivo studies.
The statistical power to identify all responding genes is intrinsically limited by the high dimensionality of RNAseq data with >20,000 genes. Thus, p=0.05 threshold needs to be adjusted for multiple testing to identify the real changes. Unfortunately, the small effect size of nPM exposure together with the necessarily small sample size of animals led to lack of enogh power to detect the changes at 5% FDR rate. To minimize the rate of false positives, we reported the changes at a nominal significance of p<0.005. Nonetheless, there could be small changes in some genes that we could not detect but still critical for air pollution toxicity. For example, while RNAseq analysis did not detect responses of Nrf2, Nfkb, and Gclc mRNA at p<0.005, the nPM dose-response experiment showed a 50% dose-dependent change in Nrf2, Nfkb mRNA, and GCLC protein, which confirmed findings on the cerebellum (Zhang et al., 2012). Thus, RNAseq and other high dimensional data are inherent with potential false-positive and false-negative results.
Another limitation of air pollution studies is the heterogeneity in the chemical composition and toxic activity PM. Recently, we identified physical and chemical characteristics of nPM that altered in vitro and in vivo toxicity of nPM (Zhang et al., 2019;Haghani et al., 2020). Regardless, NRF2 and NFKB responses were consistent in all the nPM batches used in these different experiments.
In summary, air pollution neurotoxicity was shown to have sex-and APOE allele-specificity, which are main risk factors for AD. These findings give a rationale for including APOE-gender interactions in epidemiological studies of cognitive aging and dementias.  (Xu et al., 1996) mice were aged 2 months at exposure onset (Cacciottolo et al., 2016;Youmans et al., 2012). For long-term nPM exposures (8-15 weeks), four mice for each sex, genotype (C57BL/6J, APOE3-TR, APOE4-TR), and exposure (48 mice total) were randomly assigned to nPM exposure or control. The dose-response experiment was done with 10 male C57BL/6J mice per group. After exposure, mice were euthanized by isoflurane anesthesia and perfused transcardially with phosphate-buffered saline. Brains were hemisected at midline; total cerebral cortex was frozen on dry ice and stored at À80˚C. Investigators were blinded to exposure groups during data measurement and analyses.

Materials and methods
Air pollution nPM collection and exposure Mice were exposed to nPM, a nano-sized subfraction of airP particulate matter of 2.5 microns diameter (PM2.5) collected from a local urban freeway corridor (Woodward et al., 2017b;Haghani et al., 2020). Briefly, PM0.2 samples were collected by a High-Volume Ultrafine Particle (HVUP) Sampler (Misra et al., 2002) at 400 L/min flow rate on an 8 Â 10 inch-Zeflour PTFE filter (Pall Life Sciences, Ann Arbor, MI). The Particle Instrumentation Unit of University of Southern California is located within 150 m downwind of a major freeway (I-110). Chemical composition and size distribution of re-aerosolized nPM is characterized by high-resolution mass spectrometry (SF-ICPMS) and Sievers 900 Total Organic Carbon Analyzer as described before (Morgan et al., 2011;Haghani et al., 2020). Chemical characterization of the nPM batches in this study is in Supplementary data (Figure 1-figure supplement 4). Re-aerosolized nPM or filtered air (control) was delivered to the sealed exposure chambers at approximately 300 mg/m3 concentration to model chronic exposure: 5 h/day, 3 days/week, for 8 (C57BL/6J) or 15 weeks (APOE-TR). For dose-response experiment, 8 weeks male C57BL/6J mice were exposed to approximately 100, 200, and 300 mg/m 3 for 3 weeks. The duration and nPM dosages were based on brain responses in prior studies (Haghani et al., 2020;Cheng et al., 2016). The 3 weeks of intermittent exposure (5 h/day, 3 days/ week) to 300 mg/m 3 nPM yields an average hourly exposure of 27 mg/m 3 , as experienced in many cities.
RNA sequence (RNAseq) analysis of mouse cerebral cortex RNA was extracted with TRIzol (Invitrogen) and RNAeasy Mini Kit (Qiagen) with DNase digestion. Libraries were made with the TRUseq Stranded mRNA Kit (Illumina) with 1 mg of RNA. For Illumina NextSeq500sequencing, a single end-sequencing length of >50 nt was used. Reads were aligned and quantified to the mouse reference genome RefSeq mm10 with Tophat2 (v2.0.8b), restricted to uniquely mapping reads with three possible mismatches using the Partek flow software platform (Trapnell et al., 2012). Differential gene expression was calculated by linear modeling (Limma package in R). Significance was calculated at 5% FDR rate or p-value, 0.005. Gene responses were analyzed by Qiagen Ingenuity Pathway Analysis (IPA) software. In combining the datasets generated from B6 and APOE-TR experiments, the models were adjusted by the COMBAT method to control for unknown variance (Johnson et al., 2007). Additional downstream analysis and plotting were done in Rstudio and GraphPad Prism. Cell type deconvolution analysis was done using BRETIGEA (BRain cEll Type spe-cIfic Gene Expression Analysis) R package (McKenzie et al., 2018), which uses single-cell RNAseq data to identify cell-type-specific gene-signatures to predict the proportion of each cell type in bulk RNAseq data.

Weighted gene co-expression network analysis (WGCNA)
The co-expression network, based on WGCNA, was constructed from RNAseq data. WGCNA is an unsupervised clustering approach, which assigns groups of genes with shared expression patterns into modules (Langfelder and Horvath, 2008). Briefly, the adjacency matrix (correlations between genes) was converted to a scale-free network using soft threshold power (tuned in each group) of the signed matrix. The result was converted to a topological overlap matrix (TOM). Hierarchical clustering used 1-TOM distance measure (dissimilarity). A dynamic tree-cut algorithm was used to assign modules containing at least 30 genes. Module eigengenes (MEs) were calculated as the maximum amount of the variance of the model for each module, based on the Singular Value Decomposition method. Linear regression models estimated association of nPM or APOE with the MEs. The top 150 hub genes of the modules were selected for IPA analysis by the highest eigengene connectivity (kME). In total, 32 nPM associated gene modules were identified based on different analyses. Thus, the modules were renamed to M1-M32 to distinguish each analysis.