Longitudinal analysis reveals myeloid cell contributions to murine neuroPASC pathogenesis Authors: Lu Tan, Abhishek K. Verma, Shea Lowery, Andrew Thurman, Alan Sariol, Cori Fain, Zhaoyuan Liu, Florent Ginhoux, John Harty, David K. Meyerholz, Stanley Perlman Published: 2026 ABSTRACT Neurological and neuropsychiatric symptoms, collectively termed neuroPASC, are among the most prevalent Post-Acute Sequelae of COVID-19 (PASC). Neuroinflammation – particularly microglia reactivity – has been implicated in neuroPASC. We previously established a PASC model in which SARS-CoV-2-infected mice developed persistent behavior alterations and prolonged neuroinflammation for up to 120 days post-infection (dpi). Here, we extend these results to a longitudinal single-cell RNA sequencing analysis of brain immune cells collected at 0, 6, 30, and 100 dpi. We identify a coordinated contribution of infiltrating and resident myeloid cells to the initiation and persistence of neuroinflammation. In specific, microglia display sustained expansion of subclusters characterized by inflammatory, stress response, and metabolic signatures. Border-associated macrophages upregulate monocyte attractants during acute infection. Concurrently, peripherally derived monocytes and neutrophils mount transient inflammatory responses, potentially triggering long-term microglial reactivity. Together, these findings provide a high-resolution atlas of brain myeloid immune dynamics during neuroPASC and highlight a central role for microglia in sustaining chronic neuroinflammation. Introduction Many COVID-19 patients develop long-term effects encompassing a group of multisystem disorders following the resolution of acute SARS-CoV-2 infection, collectively termed post-acute sequelae of COVID-19 (PASC). Owing to its complexity and prevalence, PASC poses substantial impacts on individual lives along with broad economic and healthcare consequences1. In particular, neurological and psychiatric manifestations, referred to as neuroPASC, are among the most common sequelae, affecting both previously hospitalized and non-hospitalized patients and persisting for at least three years in some cases2. Typical symptoms/signs include, but are not limited to, brain fog, fatigue, dizziness, olfactory dysfunction, memory problems, and abnormal movements2. Accumulating evidence implicates multifaceted pathological changes in neuroPASC. Reductions in gray matter volume3,4, elevations of brain injury markers in serum4,5, blood-brain barrier (BBB) disruption6, and gliosis7 have been documented in different patient cohorts. However, the cellular and molecular alterations occurring within the brain remain understudied. Current knowledge is derived largely from studies of post-mortem brain samples from COVID-19 patients. Histological, proteomic, and transcriptomic analyses of these samples have revealed broad cellular perturbations in the central nervous system (CNS), involving neurons, microglia, oligodendrocytes, astrocytes, and T cells, among others8–12. Importantly, neuroinflammation and microglia reactivity have been consistently reported in SARS-CoV-2-infected patients and experimental animals with neurological abnormalities9,12–16. Microglia, yolk sac-derived, CNS parenchymal resident myeloid cells, are key mediators of neuroinflammation across diverse contexts such as neurodegenerative diseases and aging17–19. In the context of COVID-19, microglial reactivity was evidenced by morphological changes, upregulation of activation markers, and increased cell numbers. The underlying transcriptional reprogramming has also been profiled using single-cell and single-nucleus RNA sequencing (scRNA-seq and snRNA-seq)8,10–12,20,21. However, most findings were from patients who succumbed to acute COVID-19 and had heterogeneous disease courses and comorbidities8,10–12,20 and were often restricted to specific brain regions8,10–12,20,21. Furthermore, the causal consequences of microglial reactivity remain largely unknown. For example, while metabolic failure and mitochondrial damage are linked to microglia dysfunction in the brains of SARS-CoV-2-infected patients at sites of synapse loss12, how microglia gene expression contributes to these processes is not known. IL-1β and IL-6 have been implicated in microglia proinflammatory, pathogenic immune responses12, but these analyses of human specimens are by necessity cross-sectional. In addition, it is not known whether microglial gene expression returns to baseline levels in patients who recover from acute COVID-19 and if normalization occurs, how rapidly. Border-associated macrophages (BAMs) constitute another population of brain-resident sentinel cells localized to meninges, choroid plexus (CP), and perivascular spaces22. Alongside hematogenous monocyte-derived macrophages entering the CNS under pathological conditions22, they exhibit both redundant and non-redundant functions relative to microglia. However, contributions of BAMs and monocyte-derived macrophages in the brain during SARS-CoV-2 infection remain undefined. Of note, outside the CNS, monocytes and macrophages are potent drivers of systemic inflammation and COVID-19 pathogenesis23–25. Similarly, neutrophils have been implicated in acute COVID-19 severity and post-COVID-19 pulmonary sequelae26–28 and in the pathogenesis of human neurodegenerative diseases such as Alzheimer’s disease (AD)29, but their role in neurological PASC has not been explored. Altogether, a comprehensive and longitudinal characterization of whole-brain myeloid cell dynamics following peripheral SARS-CoV-2 infection is still lacking. Given the scarcity of post-mortem samples, especially from neuroPASC patients, relevant animal models are essential. Because ancestral strains of SARS-CoV-2 cannot infect mice, we and others generated mouse-adapted viruses. We introduced a N501Y mutation into the spike protein using reverse genetics, allowing the virus to use mouse ACE2 for cellular entry30,31. We then passaged the virus 30 times through mouse lungs to isolate a virulent, mouse-adapted SARS-CoV-2 (SARS2-N501YMA30)31. Using this virus, we showed that mice develop anosmia during acute infection and persistent neuroinflammation with behavioral changes32,33. In this study, we use this experimental infection model to longitudinally assess the CNS inflammatory milieu of neuroPASC, up to 100 days post-infection (dpi). We find that inflammatory myeloid cell infiltration into the brain and antiviral responses are maximal at early times after infection and slowly decrease over the next few weeks. In contrast, microglial reactivity is detected at all time points post-infection and occurs independent of viral neuroinvasion. Together, our data provide a high-resolution atlas of brain immune responses during neuroPASC progression. These data underscore the predominant role of microglia in sustaining long-term neuroinflammation with implications for understanding microglial function in neurodegeneration in general. Results Mice with neuroPASC A sublethal dose (1000 PFU per mouse) of SARS2-N501YMA30 was intranasally administered to 4–10-month-old C57BL/6 mice to establish a neuroPASC model (Fig. 1a), based on our previous results32. This age range was chosen because young C57BL/6 mice (6–10 weeks) are only modestly susceptible to SARS2-N501YMA30 infection31, while mice greater than 4 months uniformly develop dosage-dependent clinical disease. Here, we showed that infected mice exhibited an approximate 20% body weight loss by 5-7 dpi, followed by gradual recovery, although an approximate 10% weight loss persisted at 14 dpi (Fig. 1b). Meanwhile, more than one-quarter of animals succumbed to infection between 5 and 10 dpi (Fig. 1c). These data suggested that this model recapitulated more severe disease and a prolonged clinical course compared with other reported long COVID-19 mouse models14,16,21. As expected for a respiratory tract infection, high levels of viral RNA (vRNA) were detected in lungs at 2 dpi and remained detectable even at 60 dpi (Fig. 1d), although infectious virus was no longer recoverable by 10 dpi. In contrast, vRNA was only transiently detected in the brain, declining to the mock level thereafter (Fig. 1d). This is consistent with our previous report demonstrating that a small number of infectious virus particles can be found in some mouse brains at 2 dpi31, likely as a consequence of viremia, since no virus-infected cells were observed in the brain34. Therefore, we consider productive CNS infection in our model to be unlikely, consistent with previous reports describing the presence of viral RNA or protein in the infected human CNS in only a minority of studies9,12,35–38. Meanwhile, no overt lesions or evidence of hypoxic injury were identified in the brains of infected mice when examined at 6 dpi (Supplementary Fig. 1).Fig. 1Long-term behavioral changes in a neuroPASC mouse model.a | Schematic of study design. dpi, days post-infection. b | Weight change of C57BL/6 mice (4–6 months old) following intranasal infection with SARS2-N501YMA30 (1000 PFU, n = 9 mice from two independent experiments). c | Survival of infected mice (n = 18 mice from four independent experiments). d | Viral RNA loads in lungs (mock: n = 2 mice; 2 dpi: n = 4; 20 dpi: n = 4; 40 dpi: n = 4; 60 dpi: n = 4 from one experiment) and brains (mock: n = 5; 2 dpi: n = 5; 20 dpi: n = 4; 40 dpi: n = 4; 60 dpi: n = 5 from one experiment) at indicated times post-infection. e | Schematic of the three-chamber social preference test. f | Time spent by mice in social (TS) versus non-social (TNS) chambers (mock: n = 16; 20 dpi: n = 10; 30–40 dpi: n = 16; 100 dpi: n = 7 from three independent experiments). g | Social preference index (ISP) of individual time points (mock: n = 16; 20 dpi: n = 10; 30–40 dpi: n = 16; 100 dpi: n = 7 from three independent experiments). ISP was calculated as: ISP = (TS − TNS)/(TS + TNS). h | Schematic of the marble burying test. i | Number of marbles buried (mock: n = 10 mice; 20 dpi: n = 7; 30–40 dpi: n = 18 from two independent experiments). A marble was scored as buried if ≥ 60% of its surface was covered with bedding. All data are presented as mean ± standard error of the mean (SEM). Statistical significance in (d, f) was determined using two-sided ordinary two-way ANOVA with Sidak’s multiple comparisons test, and significance in (g, i) was assessed by two-sided ordinary one-way ANOVA with Dunnett’s multiple comparisons test. Non-significant p values in (d) are not shown. Graphical illustrations in (a, e, and h) were created using BioRender (BioRender.com). Long-term behavioral changes in a neuroPASC mouse model. a | Schematic of study design. dpi, days post-infection. b | Weight change of C57BL/6 mice (4–6 months old) following intranasal infection with SARS2-N501YMA30 (1000 PFU, n = 9 mice from two independent experiments). c | Survival of infected mice (n = 18 mice from four independent experiments). d | Viral RNA loads in lungs (mock: n = 2 mice; 2 dpi: n = 4; 20 dpi: n = 4; 40 dpi: n = 4; 60 dpi: n = 4 from one experiment) and brains (mock: n = 5; 2 dpi: n = 5; 20 dpi: n = 4; 40 dpi: n = 4; 60 dpi: n = 5 from one experiment) at indicated times post-infection. e | Schematic of the three-chamber social preference test. f | Time spent by mice in social (TS) versus non-social (TNS) chambers (mock: n = 16; 20 dpi: n = 10; 30–40 dpi: n = 16; 100 dpi: n = 7 from three independent experiments). g | Social preference index (ISP) of individual time points (mock: n = 16; 20 dpi: n = 10; 30–40 dpi: n = 16; 100 dpi: n = 7 from three independent experiments). ISP was calculated as: ISP = (TS − TNS)/(TS + TNS). h | Schematic of the marble burying test. i | Number of marbles buried (mock: n = 10 mice; 20 dpi: n = 7; 30–40 dpi: n = 18 from two independent experiments). A marble was scored as buried if ≥ 60% of its surface was covered with bedding. All data are presented as mean ± standard error of the mean (SEM). Statistical significance in (d, f) was determined using two-sided ordinary two-way ANOVA with Sidak’s multiple comparisons test, and significance in (g, i) was assessed by two-sided ordinary one-way ANOVA with Dunnett’s multiple comparisons test. Non-significant p values in (d) are not shown. Graphical illustrations in (a, e, and h) were created using BioRender (BioRender.com). Given the wide range of COVID-19-related CNS symptoms and signs2, we performed several behavior tests to evaluate the clinical manifestations of neuroPASC at later time points when mice no longer exhibited overt clinical signs. We previously reported motor dysfunction in SARS2-N501YMA30-infected mice, as demonstrated by deficits in rotarod and open field tests32. Next, we assessed potential social impairment by conducting three-chamber social testing (Fig. 1e)39. Healthy (mock-infected) mice are considered social animals, and as expected, they spent more time exploring another mouse (“social stimulus”) than an empty chamber (“non-social stimulus”) (Fig. 1f). In contrast, SARS2-N501YMA30-infected mice failed to show a significant preference for the social stimulus at 20, 30–40, and 100 dpi (Fig. 1f). Social preference indices39 were calculated accordingly and found to be significantly decreased at 20 dpi and trend to significant reduction at 30–40 dpi compared with mock-infected controls (Fig. 1g), indicating persisting social interaction deficits. We also performed marble burying tests to evaluate changes in anxiety-like behaviors (Fig. 1h). Consistent with a previous study in hamsters40, infected mice showed significant reductions in burying activity at 30–40 dpi relative to controls (Fig. 1i), suggestive of an abnormal behavioral pattern41,42. Collectively, these findings showed that peripheral sublethal infection with SARS2-N501YMA30 induced various long-term behavioral changes in mice, in the apparent absence of virus neuroinvasion. Altered brain immune cell composition following peripheral SARS2-N501YMA30 infection While the mechanisms underlying neuroPASC are almost certainly multifactorial38, neuroinflammation is considered to be one of the major contributory factors in its development. Therefore, we next investigated the indirect effects of peripheral SARS2-N501YMA30 infection of respiratory tract, sustentacular31,33 or possibly other cells on immune responses in the brain. To this end, we isolated CD45+ immune cells from whole brains of SARS2-N501YMA30-infected mice following perfusion at 6, 30, and 100 dpi, corresponding to acute disease, post-acute early, and late recovery phases, respectively, and performed scRNA-seq (Fig. 2a). Mock-infected mice were processed in parallel as controls. This approach enabled us to track dynamic changes in brain immune cell composition and transcriptional profiles across disease stages. As a result, a total of 20,153 cells were recovered, revealing 11 immune cell types, including microglia, two discrete subsets of BAMs, monocytes, neutrophils, T and B lymphocytes, and minor populations of other immune cells (Fig. 2b; Supplementary Fig. 2; Supplementary Data 1). The most prominent compositional shift was seen at 6 dpi, characterized by marked increases in monocyte and neutrophil frequencies compared with mock-infected controls (Fig. 2c).Fig. 2Peripheral myeloid cell infiltration into the brain following intranasal SARS2-N501YMA30 infection.a | Schematic of the experimental workflow. Whole brains were dissociated, followed by either magnetic bead enrichment of CD45+ cells and single-cell RNA sequencing (scRNA-seq) or multiplex flow cytometry. b | Uniform Manifold Approximation and Projection (UMAP) visualization of brain immune cells identified by scRNA-seq, including samples from mock, 6, 30, and 100 dpi. BAMs, border-associated macrophages; DCs, dendritic cells; NK cells, natural killer cells; ILCs, innate lymphoid cells. c | Relative composition of brain immune cells across post-infection time points in the scRNA-seq dataset. d–j | Representative contour plots, numbers, and frequencies of microglia, CD11b+CD45hi cells, neutrophils, and Ly6C+ monocytes in brains at 6 and 30 dpi, measured by multiplex flow cytometry. Data in (e, f, h, j) are represented as mean with SEM and are pooled from two independent experiments per time point (mock: n = 12 mice; 6 dpi: n = 9; 30 dpi: n = 9). Outliers were removed using the ROUT method (Q = 1%). Normality was assessed by the Shapiro–Wilk test; normally distributed data were compared by two-sided ordinary one-way ANOVA with Dunnett’s multiple comparisons test, while non-normally distributed data were analyzed using two-sided Kruskal–Wallis test with Dunn’s multiple comparisons test. Graphical illustrations in (a) were created with BioRender (BioRender.com). Peripheral myeloid cell infiltration into the brain following intranasal SARS2-N501YMA30 infection. a | Schematic of the experimental workflow. Whole brains were dissociated, followed by either magnetic bead enrichment of CD45+ cells and single-cell RNA sequencing (scRNA-seq) or multiplex flow cytometry. b | Uniform Manifold Approximation and Projection (UMAP) visualization of brain immune cells identified by scRNA-seq, including samples from mock, 6, 30, and 100 dpi. BAMs, border-associated macrophages; DCs, dendritic cells; NK cells, natural killer cells; ILCs, innate lymphoid cells. c | Relative composition of brain immune cells across post-infection time points in the scRNA-seq dataset. d–j | Representative contour plots, numbers, and frequencies of microglia, CD11b+CD45hi cells, neutrophils, and Ly6C+ monocytes in brains at 6 and 30 dpi, measured by multiplex flow cytometry. Data in (e, f, h, j) are represented as mean with SEM and are pooled from two independent experiments per time point (mock: n = 12 mice; 6 dpi: n = 9; 30 dpi: n = 9). Outliers were removed using the ROUT method (Q = 1%). Normality was assessed by the Shapiro–Wilk test; normally distributed data were compared by two-sided ordinary one-way ANOVA with Dunnett’s multiple comparisons test, while non-normally distributed data were analyzed using two-sided Kruskal–Wallis test with Dunn’s multiple comparisons test. Graphical illustrations in (a) were created with BioRender (BioRender.com). We subsequently validated these changes in cellular composition using multiplex flow cytometry. Myeloid cell populations were distinguished based on CD45 and CD11b expression (Supplementary Fig. 3). The number and frequency of microglia (CD45midCD11b+ cells) did not significantly change between mock-infected and post-infection samples (Fig. 2d, e). In contrast, the frequency of CD45hiCD11b+ cells was elevated at both 6 and 30 dpi (Fig. 2d, f). Further analysis of CD45hiCD11b+ cells revealed increased infiltration of neutrophils (Ly6G+ cells) (Fig. 2g, h) and classical monocytes (Ly6C+Ly6G- cells) (Fig. 2i, j). These results are generally consistent with the scRNA-seq results (Fig. 2c), although the magnitude of infiltration differed modestly when results from the two approaches were compared. On the other hand, Ly6C-Ly6G- populations, corresponding to non-classical monocytes and BAMs, were stably populated across samples (Supplementary Fig. 4a). Notably, frequencies of neutrophils and Ly6C+ monocytes largely returned to mock levels by 100 dpi (Supplementary Fig. 5), indicating that their infiltration was transient. Using Ms4a3Cre-RosaTdT reporter mice43, which efficiently label bone marrow-derived monocytes and granulocytes but not yolk sac-derived myeloid cells or lymphocytes, we further observed that these infiltrating cells localized to both parenchyma and choroid plexus (Supplementary Fig. 6). In addition to myeloid cells, we also detected increased numbers and frequencies of B cells at 30 dpi and Ly6C+CD8+ T cells44 at 6 and 30 dpi but not total CD4 or CD8 T cells at either time point (Supplementary Fig. 4b-e). Altogether, the presence of these infiltrating immune cells was consistent with a disturbance of brain immune cell homeostasis following pulmonary SARS2-N501YMA30 infection, motivating subsequent transcriptional analyses of individual brain immune cell types. Microglial state heterogeneity and compositional shift after SARS2-N501YMA30 infection Microglia are considered to be key drivers of neuroinflammation. To delineate their contributions in SARS2-N501YMA30-infected mice, we extracted all microglial cells (n = 15,086) from our scRNA-seq dataset and performed high-resolution subclustering (resolution = 2) to capture rare populations. Similar subclusters were merged based on their proximity in the UMAP space (Supplementary Fig. 7a), shared marker gene expression, and inter-subcluster correlation (Supplementary Fig. 7b), resulting in a final set of 9 subclusters (Fig. 3a, b; Supplementary Fig. 7c; Supplementary Data 2). A clear segregation between naïve and post-infection microglia was revealed by UMAP projection (Fig. 3c) and was accompanied by markedly different subcluster compositions (Fig. 3d), indicating broad microglial transcriptional alterations following SARS2-N501YMA30 infection.Fig. 3Microglia heterogeneity and long-term compositional shifts in response to intranasal SARS2-N501YMA30 infection.a | UMAP plot showing subclustering of microglia from mock, 6, 30, and 100 dpi samples. IEG, immediate early gene; DAM, disease-associated microglia; TIM, terminally inflammatory microglia; NIC, non-inflammatory cluster. b | Dot plot showing representative marker genes for each microglial subcluster. c | UMAP plots illustrating sample-specific distributions of microglia. Cells belonging to the indicated sample are highlighted in green, whereas cells from all other samples are shown in gray. d | Relative proportions of microglial subclusters across time points. e | Gene ontology biological process (GO:BP) enrichment of TIM-like microglial signature genes. f | Comparison of pseudotime distributions among microglial subclusters. Pseudotime trajectory analysis was performed using monocle3, with homeostatic microglia from mock-infected mice positioned at the root. g | Comparison of pseudotime distributions between samples. Box plots in (f, g) show median (center line), interquartile range (box), and minimum to maximum values (whiskers). Median values for each group are indicated. Microglia heterogeneity and long-term compositional shifts in response to intranasal SARS2-N501YMA30 infection. a | UMAP plot showing subclustering of microglia from mock, 6, 30, and 100 dpi samples. IEG, immediate early gene; DAM, disease-associated microglia; TIM, terminally inflammatory microglia; NIC, non-inflammatory cluster. b | Dot plot showing representative marker genes for each microglial subcluster. c | UMAP plots illustrating sample-specific distributions of microglia. Cells belonging to the indicated sample are highlighted in green, whereas cells from all other samples are shown in gray. d | Relative proportions of microglial subclusters across time points. e | Gene ontology biological process (GO:BP) enrichment of TIM-like microglial signature genes. f | Comparison of pseudotime distributions among microglial subclusters. Pseudotime trajectory analysis was performed using monocle3, with homeostatic microglia from mock-infected mice positioned at the root. g | Comparison of pseudotime distributions between samples. Box plots in (f, g) show median (center line), interquartile range (box), and minimum to maximum values (whiskers). Median values for each group are indicated. Among identified clusters, one was present across all time points at comparable frequencies (Fig. 3d) and was primarily characterized by high expression of immediate-early genes (IEGs), including Egr1 and Ier2 (Fig. 3b). This pattern likely reflects dissociation-induced activation during tissue preparation45–47. We will resolve this in the future by adding the transcription inhibitor actinomycin D at the time of cell harvest to inhibit de novo RNA transcription48. Similarly, a smaller cluster was enriched for Hspa1b and Hspa1a, which have also been associated with dissociation-induced activation45–47. Beyond these two clusters, we identified a homeostatic subset abundantly present in all samples (Fig. 3a, d), defined by the high expression of canonical microglial homeostatic genes such as P2ry12 and Tmem119, and P2ry1349 (Fig. 3b; Supplementary Fig. 8a). However, homeostatic cells in mock-infected and post-infection samples showed minimal overlap in the UMAP space (Fig. 3c). Comparative analysis revealed upregulation of genes involved in the regulation of metabolic process and macromolecule modification in post-infection homeostatic microglia (Supplementary Data 3). In addition, minor but disease-relevant populations were also detected. A small cluster (0.4–0.8% per sample) recapitulated disease-associated microglia (DAM)-like signatures, including upregulation of Lpl and Itgax50 (Fig. 3b). These cells are predicted to possess enhanced phagocytic activity compared to other microglia51. Unexpectedly, the frequency of DAM-like microglia was similar across samples. Additionally, interferon-responsive microglia were also detected at a low frequency, defined by elevated expression of interferon-stimulated genes such as Ifit3 and Isg1552 (Fig. 3b). Notably, their frequency increased two-fold at 6 dpi before declining at 30 and 100 dpi, indicative of a transient antiviral response. These clusters comprised over 90% of mock microglia. Several other clusters, on the other hand, were detected at low frequencies under mock-infected conditions but markedly expanded after infection. One such cluster was distinguished by high expression of inflammatory chemokines Ccl3 and Ccl4 (Fig. 3b), whose upregulation has been described in animal models of aging53,54, brain injury54, and neurodegenerative diseases53,55. These chemokines may function by regulating the trafficking and effector functions of peripheral immune cells56 or by directly influencing neuronal activity and survival55. This cluster also showed enrichment of Ccrl2 and Csf1 (Supplementary Data 2), genes implicated in modulating microglial activation57 and in supporting microglial development and maintenance58, respectively. Another inflammatory cluster was annotated as terminally inflammatory microglia (TIM)-like, defined by concomitant expression of inflammatory genes (such as Tnf and Nlrp3) and stress markers (including Egr1, Ier2, and multiple AP-1 family transcription factors)59 (Fig. 3b). Gene Ontology (GO) enrichment analysis of TIM-like markers confirmed heightened activity of these cells in inflammatory and stress responses (Fig. 3e). Originally hypothesized to represent an exhausted-like state of inflammatory microglia in an AD model59, TIM-like cells in our dataset may represent a similar phenotype. We also identified two closely related non-inflammatory clusters (NICs) (Fig. 3b; Supplementary Fig. 7c), both exhibiting high module scores for genes involved in RNA processing and protein turnover (Supplementary Fig. 8b). One of these clusters lacked other clearly distinguishable features (Supplementary Data 2, 3) and was designated NIC-1. The other cluster (NIC-2) showed elevated expression of mitochondrial DNA (mtDNA)-encoded transcripts (Fig. 3b; Supplementary Fig. 8b) and was therefore termed high-mito microglia. Compared with homeostatic microglia, these four infection-enriched clusters showed reduced expression of homeostatic genes (Supplementary Fig. 8a). Moreover, transcriptional profiles of TIM-like, NIC-1, and NIC-2 microglia correlated poorly with those of homeostatic microglia (Supplementary Fig. 7c), highlighting their transcriptional divergence. Pseudotime trajectory analysis using monocle360 further reconstructed a continuum of microglial states, with homeostatic microglia from mock-infected mice positioned at the root (Supplementary Fig. 8c). TIM-like microglia occupied the most advanced pseudotime states (Fig. 3f), consistent with their designation as exhausted-like inflammatory cells. Overall, microglial pseudotime positions changed markedly over time compared with the mock-infected control (Fig. 3g), underscoring infection-driven transcriptional reprogramming. Persistent pro-inflammatory signatures and mitochondrial dysfunction in post-infection microglia Microglial responses to SARS2-N501YMA30 infection were further assessed by examining gene expression dynamics over time. All microglial subclusters exhibited dysregulated gene expression following infection, with these alterations persisting through 100 dpi (Supplementary Fig. 9a). Overall, expression of homeostatic markers P2ry12, Tmem119, and P2ry13 was decreased in post-infection samples (Supplementary Fig. 9b), indicating overall elevated microglial reactivity. Functional enrichment analysis of genes upregulated relative to the mock-infected control (avg_log2FC > 1, pct.1 > 0.2, and p_val_adj <0.01) (Supplementary Data 3) revealed significant alterations in the regulation of immune responses at 6 dpi (Fig. 4a). Notably, several immune-related changes persisted into the subacute and long-term phases. For example, Tap1 and H2-D1, genes associated with MHC class I antigen presentation61,62, along with additional genes in the KEGG (Kyoto Encyclopedia of Genes and Genomes) antigen processing and presentation pathway, remained consistently elevated at 6, 30, and 100 dpi relative to mock-infected samples (Fig. 4b). A group of inflammatory genes also exhibited sustained upregulation throughout these time points (Fig. 4c). In particular, although not uniformly produced in all microglia, elevated expression of proinflammatory cytokines–including Tnf, Il1a, Il1b, Il18, Ccl3, and Ccl4—was seen in all infected samples (Fig. 4c, d). In line with this, genes encoding key inflammasome components required for IL-1β and IL-18 maturation and release, including Nlrp3, Nlrp1b, Casp1, and Gsdmd63–65, were concordantly upregulated (Fig. 4c, d). Regulatory factors such as the nuclear receptor Nr4a166 and NF-κB inhibitors Nfkbia and Nfkbid67 were also transcribed at higher levels in post-infection samples (Fig. 4c), suggesting dynamic regulation of microglial inflammation. TIM-like and Ccl3/Ccl4 inflammatory microglia appeared to be the major contributors to these inflammatory signatures (Supplementary Fig. 9c). This inflammatory microglial phenotype in infected mice was further validated by intracellular cytokine staining (ICS) analysis (Supplementary Fig. 10). Following 6 h of LPS stimulation, microglia from 6 dpi mice exhibited a significantly increased proportion of TNF⁺ cells compared with mock controls (Fig. 4e, f), whereas microglia from 30 dpi mice showed elevated IL-1β production (Fig. 4g, h). Although the cytokine responses differed between the two time points, microglia from post-infection mice consistently displayed a sensitized response to stimulation. We also verified upregulation of Ccl3 and Ccl4 by RT-qPCR analysis of brain CD11b+ cells (Supplementary Fig. 11), the great majority of which are microglia.Fig. 4Persistent inflammatory response and mitochondrial dysfunction in post-infection microglia.a | Functional enrichment of genes upregulated at 6 dpi relative to mock. Selected terms are shown. KEGG, Kyoto Encyclopedia of Genes and Genomes. b | Regulation of genes across samples under the KEGG pathway antigen processing and presentation. pct.1 indicates the proportion of microglia in each sample expressing indicated genes; avg_log2FC indicates the log2 fold change relative to mock. c | Regulation of genes across samples under the GO:BP term inflammatory responses. d | Expression levels of inflammasome-dependent cytokines and inflammasome components across samples. e | Representative contour plots of TNF⁺ microglia measured by intracellular cytokine staining (ICS). Mouse brain single-cell suspensions were stimulated with LPS for 6 h prior to analysis. f | Comparison of the proportion of TNF ⁺ microglia between mock-infected and post-infection samples as measured by ICS. g | Representative contour plots of IL-1β⁺ microglia measured by ICS following LPS stimulation. h | Comparison of the proportion of IL-1β⁺ microglia between mock-infected and post-infection samples as measured by ICS. Data in (f, h) are represented as mean with SEM and are pooled from two independent experiments per time point (mock: n = 12; 6 dpi: n = 7; 30 dpi = 7). Statistical analysis was performed using two-sided ordinary one-way ANOVA with Dunnett’s multiple comparisons test. i | Module scores of nuclear-encoded mitochondrial genes (NEMGs) across microglial subclusters. j | Proportion of dysregulated NEMGs among total NEMGs in TIM-like, NIC-1, and NIC-2 subclusters compared with others. A cutoff of adjusted p < 0.01, calculated using two-sided Wilcoxon rank-sum tests with Bonferroni correction implemented in Seurat’s FindMarkers function, was applied to identify dysregulated genes. k | Functional enrichment of downregulated genes in TIM-like, NIC-1, and NIC-2 subclusters compared with others. GO:MF, Gene Ontology Molecular Function. Persistent inflammatory response and mitochondrial dysfunction in post-infection microglia. a | Functional enrichment of genes upregulated at 6 dpi relative to mock. Selected terms are shown. KEGG, Kyoto Encyclopedia of Genes and Genomes. b | Regulation of genes across samples under the KEGG pathway antigen processing and presentation. pct.1 indicates the proportion of microglia in each sample expressing indicated genes; avg_log2FC indicates the log2 fold change relative to mock. c | Regulation of genes across samples under the GO:BP term inflammatory responses. d | Expression levels of inflammasome-dependent cytokines and inflammasome components across samples. e | Representative contour plots of TNF⁺ microglia measured by intracellular cytokine staining (ICS). Mouse brain single-cell suspensions were stimulated with LPS for 6 h prior to analysis. f | Comparison of the proportion of TNF ⁺ microglia between mock-infected and post-infection samples as measured by ICS. g | Representative contour plots of IL-1β⁺ microglia measured by ICS following LPS stimulation. h | Comparison of the proportion of IL-1β⁺ microglia between mock-infected and post-infection samples as measured by ICS. Data in (f, h) are represented as mean with SEM and are pooled from two independent experiments per time point (mock: n = 12; 6 dpi: n = 7; 30 dpi = 7). Statistical analysis was performed using two-sided ordinary one-way ANOVA with Dunnett’s multiple comparisons test. i | Module scores of nuclear-encoded mitochondrial genes (NEMGs) across microglial subclusters. j | Proportion of dysregulated NEMGs among total NEMGs in TIM-like, NIC-1, and NIC-2 subclusters compared with others. A cutoff of adjusted p < 0.01, calculated using two-sided Wilcoxon rank-sum tests with Bonferroni correction implemented in Seurat’s FindMarkers function, was applied to identify dysregulated genes. k | Functional enrichment of downregulated genes in TIM-like, NIC-1, and NIC-2 subclusters compared with others. GO:MF, Gene Ontology Molecular Function. Microglial metabolic failure and mitochondrial injury have been reported in post-mortem brain samples from COVID-19 patients and are associated with synapse and myelin loss12. Consistent with this observation, our functional enrichment analysis of differentially expressed genes (DEGs) suggested broad metabolic changes in microglia across post-infection time points (Supplementary Data 3). Intriguingly, three infection-enriched subclusters (TIM-like, NIC-1, and NIC-2) displayed pronounced mitochondrial gene dysregulation, evidenced by markedly reduced module scores of nuclear-encoded mitochondrial genes (NEMGs)68 (Fig. 4i). Of 944 detectable NEMGs, 14.8% were significantly downregulated (log2FC ≤ –1) in these clusters (Fig. 4j), including 44 of 89 genes encoding oxidative phosphorylation (OXPHOS) subunits. This prominent downregulation of NEMGs was predicted to impair crucial cellular processes such as mitochondrial organization, OXPHOS, aerobic respiration, ER stress response, unfolded protein response, and antioxidation, and contribute to pathways implicated in neurodegenerative diseases, including prion disease, Huntington’s Disease, Parkinson’s Disease, and AD (Fig. 4k). Conversely, these clusters showed increased expression of mtDNA-encoded transcripts, particularly in NIC-2 microglia (Supplementary Fig. 8b). Previous studies have demonstrated asynchronous kinetics between NEMGs and mtDNA-encoded genes69, and also shown that OXPHOS transcripts encoded by the two genomes increase in parallel even if not concordantly70. However, the divergent expression we observed was unexpected and underscored mitochondrial dysfunction as a possible key feature of microglial responses to SARS2-N501YMA30 infection. To assess the potential effects of microglia activation on neuronal integrity, we examined neuronal gene expression in the olfactory bulb (OB) by bulk RNA sequencing (RNA-seq). We focused on the OB because microglial activation is particularly prominent at this site32,40. RNA-seq data revealed downregulation of a group of neuron-specific and neuron-enriched genes at both 6 and 30 dpi compared with controls (Supplementary Fig. 12). Among these, Gabra2, which encodes the GABA A receptor α2 subunit, is highly expressed across multiple brain regions71. Together, persistent microglial activation and concomitant changes in neuronal gene expression suggest that microglial alterations in infected mice contribute to impaired neuronal integrity and function. Post-infection monocyte recruitment and chemokine activity revealed by BAM transcriptomic changes BAMs are another group of brain-resident myeloid cells located in the perivascular compartment, meninges, and CP. In our scRNA-seq dataset, we identified two clusters of cells expressing universal BAM signature genes, including Mrc1, Pf4, and Ms4a746,72 (Fig. 2b; Supplementary Fig. 13a; Supplementary Data 1). They corresponded to two discrete populations: MHCIIlo and MHCIIhi BAMs (Fig. 2b). Unlike the dynamically shifting microglia composition described thus far (Fig. 3d), these BAM clusters showed stable relative abundance across different time points (Fig. 2c). Phenotypically, MHCIIlo BAMs exhibited high expression of M2 macrophage markers (Mrc1 and Cd163)73 and multiple anti-inflammatory genes (Gas6, Stab1, and Maf), while MHCIIhi cells were enriched for antigen processing and presentation genes (H2-Ab1, H2-Aa, and Cd74) as well as proinflammatory genes (Il1b, Icam1, and Fcgr4) (Fig. 5a). These two populations also expressed distinct sets of chemokine and chemokine receptors (Fig. 5a), indicating complementary roles in chemotaxis. Based on reported tissue-specific signature genes46,74, we expected that MHCIIlo BAMs originated primarily from the meningeal and perivascular niches, whereas the MHCIIhi population included meningeal, perivascular, and CP BAMs (Fig. 5b).Fig. 5Monocyte recruitment and transient enhancement of chemokine activity in BAMs after SARS2-N501YMA30 infection.a | Signature genes distinguishing MHCIIlo and MHCIIhi BAMs. b | Expression of tissue-specific signature genes in MHCIIlo and MHCIIhi BAMs. CP, choroid plexus. c | Proportion of Ccr2+ cells within MHCIIhi BAMs across samples. d | Signature genes distinguishing Ccr2- and Ccr2+ MHCIIhi BAMs. e | Expression of the monocyte-attracting chemokines Ccl2, Ccl7, and Ccl12 across samples. f | Expression of Tnf across samples. Monocyte recruitment and transient enhancement of chemokine activity in BAMs after SARS2-N501YMA30 infection. a | Signature genes distinguishing MHCIIlo and MHCIIhi BAMs. b | Expression of tissue-specific signature genes in MHCIIlo and MHCIIhi BAMs. CP, choroid plexus. c | Proportion of Ccr2+ cells within MHCIIhi BAMs across samples. d | Signature genes distinguishing Ccr2- and Ccr2+ MHCIIhi BAMs. e | Expression of the monocyte-attracting chemokines Ccl2, Ccl7, and Ccl12 across samples. f | Expression of Tnf across samples. Notably, Ccr2, a marker of infiltrating monocytes, was exclusively detected in MHCIIhi macrophages (Fig. 4a, b), consistent with a monocytic origin for a fraction of these cells46,75. This gradual renewal of specific BAM subsets by peripheral cells happens under homeostatic conditions46, but disruption of CNS homeostasis significantly accelerates this replenishment75. To investigate the impact of SARS2-N501YMA30 infection, we compared the frequency of Ccr2+ macrophages at different post-infection time points. We found that Ccr2+ BAMs were most abundant at 30 dpi, increasing from 40.07% of total MHCIIhi BAMs in the mock control to 61.65% (Fig. 5c). This elevation suggested ongoing monocyte recruitment and transition to macrophages, possibly reflecting peripheral infection–driven changes to brain homeostasis. By 100 dpi, the frequency of Ccr2+ macrophages declined, probably due to the gradual resolution of inflammation and phenotypic adaptation of infiltrating cells to the brain microenvironment niche. Indeed, the Ccr2- population showed higher expression of certain BAM-associated genes, such as Ms4a7, Stab1, Vcam1, and Cp46,76, while Ccr2+ cells maintained several monocyte-associated genes, including Lsp1, Crip1, Tmsb10, Capg, and Adgre5 (Fig. 5d). Nevertheless, compared with Ccr2+ monocytes, Ccr2+ macrophages exhibited robust BAM signatures and high MHC II gene expression (Supplementary Fig. 13b). Consistent with the elevated frequency of Ccr2+ macrophages in post-infection samples, upregulation of monocyte-attracting chemokines, including Ccl2, Ccl7, and Ccl1277, was observed in both MHCIIlo and MHCIIhi BAMs at 6 dpi (Fig. 5e). Furthermore, the expression of Tnf, a prototypic proinflammatory cytokine, increased across all post-infection time points (Fig. 5f), indicating a persistently inflammatory milieu, potentially driven by infiltrating monocytes78. Supporting this, within the MHCIIhi BAM population, expression of Il1b, Il18, and their associated inflammasome components Nlrp3, Casp1, and Gsdmd was also upregulated (Supplementary Fig. 13c). Inflammatory and antiviral programs in infiltrating monocytes and neutrophils As shown by scRNA-seq and flow cytometry, monocyte and neutrophil frequencies in the brain markedly increased during acute infection (Fig. 2). In mice, two major monocyte subsets have been described: classical Ly6C+ (CX3CR1intCCR2+CD62L+ CD43lowLy6Chi) and nonclassical Ly6C- (CX3CR1hiCCR2-CD62L-CD43hiLy6Clo)79,80. Because Ly6c2 transcripts were undetectable in our dataset for technical reasons81, we used Ccr2 expression to distinguish these two populations. Ccr2+ cells constituted the majority of infiltrating monocytes, although Ccr2- cells were also increased relative to the mock-infected control (Fig. 6a). Transcriptionally, Ccr2+ monocytes expressed higher levels of Sell (encoding CD62L) and Lyz1, consistent with a classical phenotype (Fig. 6b)82. Conversely, Spn (encoding CD43), Itgax, Cd36, Pparg, and Fcgr4 were enriched in Ccr2- cells, indicating a nonclassical signature82,83. These findings align with the established role of classical monocytes as the primary subset recruited to inflamed tissues79,80. DEG analysis between mock and infected samples (Supplementary Data 4) underscored a strong antiviral response in monocytes at 6 dpi. Upregulated genes were functionally enriched in defense response to virus, cellular response to interferon-beta and type II interferon, activation of innate immune response, and negative regulation of viral process (Fig. 6c). Correspondingly, ISGs such as Isg15, Oas2, Ifit3, and Irf7 were robustly induced at this time point (Fig. 6d). By 30 and 100 dpi, these genes and other antiviral signatures had returned to baseline levels (Fig. 6d). Moreover, no sustained induction of inflammatory genes was detected (Supplementary Fig. 14a).Fig. 6Acute infiltration of inflammatory monocytes and neutrophils into the brain following peripheral infection and their antiviral responses.a | Relative abundance of Ccr2+ and Ccr2- monocytes across samples. b | Expression of classical and non-classical monocyte signature genes in Ccr2+ and Ccr2- monocytes. c | GO:BP enrichment of genes upregulated in monocytes at 6 dpi relative to mock-infected samples. Selected terms are shown. d | Regulation of genes in monocytes across samples under the GO:BP term response to virus. e | Relative abundance of neutrophil subclusters across samples. f | Module scores of featured gene sets across neutrophil subclusters. g | Expression of neutrophil marker genes and typical inflammatory genes across neutrophil subclusters. h | Regulation of genes in neutrophils across samples under the GO:BP termresponse to virus. Acute infiltration of inflammatory monocytes and neutrophils into the brain following peripheral infection and their antiviral responses. a | Relative abundance of Ccr2+ and Ccr2- monocytes across samples. b | Expression of classical and non-classical monocyte signature genes in Ccr2+ and Ccr2- monocytes. c | GO:BP enrichment of genes upregulated in monocytes at 6 dpi relative to mock-infected samples. Selected terms are shown. d | Regulation of genes in monocytes across samples under the GO:BP term response to virus. e | Relative abundance of neutrophil subclusters across samples. f | Module scores of featured gene sets across neutrophil subclusters. g | Expression of neutrophil marker genes and typical inflammatory genes across neutrophil subclusters. h | Regulation of genes in neutrophils across samples under the GO:BP termresponse to virus. Neutrophils were partitioned into three subpopulations via the initial unsupervised clustering (Supplementary Fig. 2b). Their abundance varied across different time points, with subcluster 1 becoming prominently enriched at 6 dpi (Fig. 6e). This cluster displayed the highest maturation score (Fig. 6f), reflecting elevated expression of genes related to neutrophil development and terminal differentiation84. Neutrophils in this cluster appeared to be more “aged” than in other subclusters (Fig. 6f), a feature enabling these cells to more rapidly respond to inflammatory signals85. Moreover, subcluster 1 exhibited the highest chemotaxis score, but expressed various granule-associated genes at the lowest levels84 (Fig. 6f). In contrast, subcluster 2 showed a granule-enriched phenotype consistent with strong cytotoxic activity, whereas cluster 3 represented an intermediate state between subclusters 1 and 2 (Fig. 6f). Notably, subclusters 1 and 3 uniquely expressed inflammatory mediators such as Cxcl2 and Nlrp3 (Fig. 6g). Collectively, these findings indicated that brain-infiltrating neutrophils following SARS2-N501YMA30 infection predominantly skewed toward a mature, aged, and inflammatory state (subcluster 1). Similar to monocytes, neutrophils also mounted a transient antiviral response at 6 dpi (Fig. 6h; Supplementary Fig. 14b; Supplementary Data 5), highlighting their involvement in the early response to peripheral viral infection. Modification of brain immune cell-cell communication by peripheral SARS2-N501YMA30 infection Cell-cell communication analysis was subsequently performed using CellChat86 to identify interactions among peripheral cells and microglial subclusters at different time points. Consistent with the recruitment of monocytes and neutrophils into the brain at 6 dpi, these infiltrating cells showed strengthened interactions with microglia, particularly homeostatic and IEG-enriched microglia, compared with the mock control (Fig. 7a). Specifically, 59 ligand–receptor interactions were identified between monocytes and microglia at 6 dpi, involving pathways such as CCL (CCL3, CCL6, and CCL9), ICAM, GALECTIN, SEMA4, APP, and SIRP. Meanwhile, 41 interactions were identified between neutrophils and microglia at 6 dpi, including pathways associated with CCL (CCL3, CCL6, and CCL9), ICAM, APP, among others (Fig. 7d). At 30 and 100 dpi, these interactions nearly returned to mock-infected levels (Fig. 7b, c), suggesting potential roles for monocytes and neutrophils in the initiation rather than maintenance of microglial reactivity. Meanwhile, T cells also showed enhanced interactions with microglia, which persisted through 100 dpi (Supplementary Fig. 15). These interactions involved pathways associated with immune cell recruitment and inflammatory signaling, including CCL (CCL6 and CCL12), CXCL (CXCL16), IL1, and IFN-II; antigen presentation and T cell co-stimulation, including MHC-I, CD80, CD86, and ICOS; as well as signaling through additional pathways.Fig. 7Altered brain immune cell-cell communication following SARS2-N501YMA30 infection.a–c | Heatmap showing changes in interaction weights between cell populations in post-infection samples compared with mock controls, revealed by the CellChat analysis. Results from analyses using microglial subclusters are shown. d | Ligand-receptor interactions between total microglia and monocytes/neutrophils at 6 dpi. e | Ligand-receptor interactions between TIM-like microglia and homeostatic microglia across different time points. Statistical significance of inferred interactions in (d, e) was determined using permutation testing implemented in CellChat. Altered brain immune cell-cell communication following SARS2-N501YMA30 infection. a–c | Heatmap showing changes in interaction weights between cell populations in post-infection samples compared with mock controls, revealed by the CellChat analysis. Results from analyses using microglial subclusters are shown. d | Ligand-receptor interactions between total microglia and monocytes/neutrophils at 6 dpi. e | Ligand-receptor interactions between TIM-like microglia and homeostatic microglia across different time points. Statistical significance of inferred interactions in (d, e) was determined using permutation testing implemented in CellChat. Notably, interactions between different microglial subclusters were strengthened following infection (Fig. 7a–c, e). The interaction between homeostatic microglia and TIM-like microglia was the strongest and persisted through 100 dpi (Fig. 7a–c, e). Their interactions involved pathways related to immune activation (TNF, CCL, BAFF, OSM, TGFβ, and VISFATIN) as well as cell adhesion, migration, and tissue remodeling (ICAM, JAM, CADM, PCDH, PTPRM, COLLAGEN, and SEMA4). This sustained communication, as well as T cell-microglia interaction, may contribute to the chronic inflammatory transcriptional signatures observed in microglia. Discussion Neuroinflammation is a prominent feature observed in COVID-19 patients with acute and persistent neurological abnormalities8–12. Here, using an established mouse model of neuroPASC, we described a comprehensive analysis of the longitudinal contributions of different brain myeloid cells to this process. We showed that microglia underwent profound and lasting transcriptional shifts after peripheral SARS2-N501YMA30 infection (Fig. 3c, d, g). Most notably, expression of multiple inflammatory mediators was consistently upregulated across acute, subacute, and long-term time points (Fig. 4c, d), indicative of their major role in chronic neuroinflammation. Persistent elevation of pro-inflammatory cytokines was also observed in another brain-resident macrophage population, BAMs (Fig. 5f; Supplementary Fig. 13c). Despite being a smaller population than microglia, these cells also potentially contribute to the chronic inflammatory milieu. Meanwhile, BAMs transiently enhanced the production of monocyte attractants during acute infection (Fig. 5e), concordant with increased Ccr2+ cell frequency at later time points (Fig. 5c), suggesting disturbed brain homeostasis. In contrast to sustained alterations in resident myeloid cells, inflammatory monocytes and neutrophils were transiently recruited into the brain and exhibited strong antiviral signatures at 6 dpi (Figs. 2, 6; Supplementary Fig. 5). Our findings recapitulate several hallmarks of microglial reactivity previously reported in COVID-19 post-mortem brain samples. Similar to those studies, we observed the emergence of COVID-19-associated microglial clusters10 (Fig. 2c, d), downregulation of homeostatic markers such as P2RY1210–12 (P2ry12 in mice) (Supplementary Fig. 9b), and mitochondrial injury reflected by metabolic pathway dysregulation and significant suppression of mitochondrial gene expression12 (Fig. 4i-k). Importantly, our data revealed that these alterations were initiated during acute infection and persisted for at least 100 dpi, whereas existing human datasets largely represent acute or subacute disease stages. Moreover, in line with their established role in neuroinflammation, microglia in infected mice exhibited sustained upregulation of neurotoxic factors such as Tnf and Il1b (Fig. 4c, d). These changes in gene expression were further validated by increased microglial cytokine expression in response to LPS stimulation (Fig. 4e-h). Meanwhile, some microglial features we identified are like those reported in neurodegenerative conditions. The TIM-like subcluster, which was almost exclusively enriched in post-infection samples (Fig. 3d), resembles an exhausted-like microglial population in brains of aged AD mice bearing the human APOE4 allele, the single greatest monoallelic risk factor for late-onset AD59. In experimental autoimmune encephalomyelitis, a model of human MS, a subpopulation of DAMs in the spinal cord upregulates both nuclear and mtDNA-encoded genes for mitochondrial complex I (CI) subunits, sustaining microglial activation and neuroinflammation through CI-driven reverse electron transport and consequent reactive oxygen species production87. In parallel, the NIC-2 subcluster enriched in our post-infection samples showed enhanced expression of all mtDNA-encoded CI subunit genes. Although augmented expression of nuclear-encoded CI components (e.g., Ndufs4) and oxidative stress markers (e.g., GP91-PHOX, Sod1) was not evident, the transcriptional profile of NIC-2 microglia may reflect mitochondrial dysfunction coupled to neuroinflammation. However, the core DAM signature, originally described in AD50 and later generalized to aging and diverse CNS diseases51, was only modestly enriched in a small fraction of microglia in our study (Fig. 3a–c; Supplementary Fig. 8b). Since DAMs feature increased expression of genes involved in the processing and clearance of phagocytosis substrates51, peripheral SARS2-N501YMA30 infection likely does not substantially enhance microglial phagocytic activity. Instead, the downregulation of homeostatic genes together with inflammatory activation in microglia isolated from post-infection mice more closely resembles transcriptional changes induced by systemic LPS challenge88. Notably, the transcriptional alterations induced by LPS challenge were detected at 24 h after stimulation88, whereas the microglial changes observed in our model persisted through at least 100 dpi. In addition to parenchymal microglia, our scRNA-seq data identified MHCIIhi and MHCIIlo BAMs and Ccr2+ and Ccr2- monocytes in mouse brains. Their compositional and transcriptional shifts collectively underscored the recruitment of Ccr2+ monocytes and their transition into MHCIIhi macrophages in response to SARS2-N501YMA30 infection (Figs. 5c, 6a). While circulating monocytes minimally enter the healthy CNS, they infiltrate the brain in different CNS disease contexts and contribute to an inflammatory immune response89,90, similar to our observations here. Notably, although inflammatory response genes were enriched in monocytes only at 6 dpi (Supplementary Fig. 14a), persistent inflammation markers were found in BAMs (Fig. 5f), especially the MHCIIhi subpopulation (Supplementary Fig. 13c). Moreover, the upregulation of monocyte attractants in BAMs (Ccl2, Ccl7, Ccl12) and microglia (Ccl3 and Ccl4) at 6 dpi suggests that they may cooperatively promote monocyte recruitment into the CNS. Although less well studied, a growing body of evidence has shown the contribution of neutrophils to CNS disease pathogenesis. Their extravasation into the brain and subsequent release of neutrophil extracellular traps (NETs) and IL-17 have been reported in AD neuropathology91, where IL-17 interacts with microglia to suppress their response to neurodegeneration92. Other neurodegenerative diseases can also be exacerbated by neutrophils, producing an inflammatory state and neuronal death93. In SARS2-N501YMA30-infected mice, inflammatory neutrophils, instead of those enriched in cytotoxic responses, migrated into the brain during acute infection (Figs. 2c, f, 6e-g), likely functioning through cell-cell communication with other immune cells. Indeed, SARS2-N501YMA30 infection appears to reshape the brain immune cell-cell communication landscape, as inferred by CellChat (Fig. 7; Supplementary Fig. 15). Consistent with their transient recruitment into the brain at 6 dpi, monocytes and neutrophils exhibited enhanced interaction weights with microglia at 6 dpi, followed by a decline at 30 dpi and return to mock levels at 100 dpi. Although we did not directly verify the spatial association between these infiltrating cells and microglia in the present study, previous studies have demonstrated interactions between infiltrating T cells and microglia using imaging mass cytometry and immunohistochemistry approaches9,94. Interactions among microglial subclusters were also detected. Particularly, homeostatic microglia strongly interacted with TIM-like microglia through 100 dpi, which likely contributes to the maintenance of the TIM-like status. Notably, consistent with previous studies9,12, our analyses indicated elevated interaction strength between T cells and microglia. We also identified significant enrichment of upregulated genes related to T cell activation and differentiation across post-infection time points (Supplementary Data 6). Meanwhile, enhanced B cell-microglia crosstalk occurred at 30 dpi, suggestive of a potentially delayed contribution of B cells to neuroPASC. A limitation of the present study is that it primarily relies on scRNA-seq analysis. Additional validation of our gene expression results is required, but was limited by availability of useful reagents and technical restraints. Also, despite its high resolution and breadth, there remains a significant gap in understanding how these transcriptional shifts relate to functional alterations51. For instance, while we identified marked downregulation of NEMGs in TIM-like, NIC-1, and NIC-2 microglia, the functional consequences of these metabolic disturbances remain to be further defined95. It will also be interesting to further probe the role of microglial activation in the neuronal integrity and function in different brain regions. Another limitation is that we did not demonstrate causality between infiltrating myeloid cells and microglial activation or elucidate spatial relationships between infiltrating inflammatory cells and various microglia subsets. In addition, the Fixed RNA Profiling workflow used here restricts detection to a predefined probe set targeting the mouse transcriptome. As a result, key genes such as the classical monocyte marker gene Ly6c2, microglia marker Cx3cr1, and cytokines Cxcl8 and Il10 were undetectable or captured at extremely low levels, potentially obscuring certain immune traits relevant to neuroPASC pathogenesis. Another limitation is that human PASC occurs in patients with varying ages, genetic backgrounds, and co-morbidities. Our mice with PASC do not replicate the complexity of human PASC. Moreover, given the reported sex-dependent structural and functional differences in rodent microglia96 and higher incidence of PASC in woman97,98, it will be important to determine whether male and female microglia respond similarly to peripheral SARS-CoV-2 infection. Altogether, our study delineates the dynamic immune landscape underlying neuroPASC development. Monocytes and neutrophils exhibited strong antiviral signatures during acute infection, likely reflecting direct responses to peripheral viral challenge. Their subsequent infiltration into the CNS may initiate the persistent inflammation observed in microglia and BAMs, thereby contributing to a variety of neuroPASC symptoms. While we focused on only immune components here, marked leukocyte infiltration also supports the hypothesis that BBB disruption and endothelial inflammation following SARS2-N501YMA30 infection contribute to the inflammatory milieu. Potential consequences include accelerated accumulation of senescent cells leading to premature brain aging and neurodegeneration99, loss of hippocampal neurogenesis associated with cognitive deficits15,16, and neurotransmitter dysregulation leading to corresponding neurological symptoms11,32. Together, these findings provide a framework for understanding the immunopathological basis of neuroPASC and highlight potential avenues for therapeutic interventions such as limiting peripheral myeloid cell infiltration during acute infection and mitigating chronic microglia reactivity. Methods Mice and viruses 4-10-month-old male and female C57BL/6NCrl mice (Charles River Laboratories, strain code 027) were used in all experiments, except for the experiment aimed at localizing infiltrating cells in the brain (Supplementary Fig. 6), which were performed using Ms4a3Cre-RosaTdT reporter mice43. These mice were generated by breeding C57BL/6J-Ms4a3em2(cre)Fgnx/J (Strain #:036382) with B6.Cg-Gt(ROSA)26Sortm14(CAG-tdTomato)Hze/J (Strain #:007914). Experimental and control mice were co-housed in the specific-pathogen free Animal Care Unit at the University of Iowa under standard conditions of dark/light cycle, ambient temperature and humidity. For each experiment, mouse ages were carefully matched, and animals were randomly assigned to experimental groups, with group sizes sufficient to achieve statistical significance. We did not observe sex-specific differences, so data from male and female mice were combined. SARS2-N501YMA30 was generated and propagated in the Perlman laboratory as previously described31. All virus stocks were sequenced following propagation and confirmed to match the input strain. Mouse infection Mice were anaesthetized by intraperitoneal injection of a ketamine–xylazine cocktail (87.5 mg/kg ketamine and 12.5 mg/kg xylazine) and infected intranasally with 1000 PFU of virus in a total volume of 50 μL DMEM. Animal weight and health were monitored daily. In accordance with University of Iowa Institutional Animal Care and Use Committee (IACUC) rules, all mice that lost 30% of their initial weight were humanely euthanized by carbon dioxide inhalation followed by cervical dislocation. All experiments involving SARS2-N501YMA30 were conducted in either biosafety level 3 (BSL-3) or BSL-2 facilities at the University of Iowa. All animal studies were approved by the University of Iowa IACUC and complied with the Guide for the Care and Use of Laboratory Animals. Mouse behavior testing Different cohorts of mice were analyzed at each time point. For the sociality preference test, a custom-built three-chamber plastic apparatus (60 cm × 40 cm × 20 cm) with open doors connecting each chamber was used. Prior to testing, two clean empty cups were placed in side chambers, and the test mouse was allowed to habituate to the apparatus and empty cups for 5 min. Subsequently, an age-, sex-, and strain-matched unfamiliar mouse was placed under one cup to serve as the social stimulus, while a toy was placed under the other cup to serve as the non-social stimulus. The test mouse was then placed in the center chamber and allowed to explore freely for 5 min. Time spent interacting with the social stimulus (TS) and the non-social stimulus (TNS) was recorded. Social preference indexes (ISP) were calculated as: ISP = (TS − TNS)/(TS + TNS). For the marble burying test, each test mouse was placed in a corner of a cage containing clean bedding with 20 equally spaced glass marbles arranged on the surface. Mice were allowed to explore freely for 15 min, after which numbers of buried and unburied marbles were recorded. A marble was considered buried if ≥60% of its surface was visually judged to be covered by bedding. All behavior test data were analysed manually. Histology Mice were deeply anaesthetized by intraperitoneal injection of ketamine–xylazine and perfused transcardially with 10 mL PBS. Tissues were fixed in zinc formalin and embedded in paraffin. Tissue sections ( ≈ 5 μm each) were stained with H&E. Tissues were evaluated for the presence of hypoxic injury. The examination of brain tissues focused on cellular and morphological changes (cell injury, cell death, neuronal dropout, gliosis nodules, etc.) in the hippocampus, with particular emphasis on the CA1 subregion, which is especially sensitive to hypoxic conditions. Immunofluorescence assay Ms4a3Cre-RosaTdT reporter mice were infected intranasally with 1000 PFU of virus. Brain tissue sections were prepared as described above. Sections were deparaffinized and subjected to antigen retrieval by boiling in antigen retrieval solution (eBioscience, catalog number 00-4955-58) for 20 min. After three washes in PBS (5 min each), sections were permeabilized with 0.1% Triton X-100 in PBS for 20 min. Sections were then incubated with CAS Block (Invitrogen, catalog number 008120) for 60 min, followed by incubation with primary antibody against RFP (Rockland, catalog number 600-401-379,1:100 dilution) at 4 °C overnight. Sections were subsequently washed five times with 0.1% Triton X-100 in PBS and incubated with Alexa Fluor 546-conjugated goat anti-rabbit secondary antibody (Thermo Fisher Scientific, catalog number A-11035, 1:1000 dilution) at room temperature for 1 h. After five additional washes with PBS, sections were mounted using DAPI-containing mounting medium (Vector Labs, catalog number H-1200-10). Images were acquired using the Leica THUNDER Imager. Brain cell preparation and antibodies for flow cytometric analysis Animals were deeply anaesthetized by intraperitoneal injection of ketamine–xylazine and perfused transcardially with 10 mL PBS. Intravenous labeling was not performed due to technical difficulties in the BSL-3 lab. In preliminary experiments, we found that the vast majority of brain-associated immune cells were in the brain and not in the vasculature. Brains were dissected and digested in HBSS buffer containing 2% fetal calf serum (FCS), 25 mM HEPES, 1 mg mL−1 collagenase D (Roche, catalog number 11088882001) and 0.1 mg ml−1 DNase I (Roche, catalog number 10104159001) at 37 °C for 30 min using a Miltenyi gentleMACS dissociator. Single-cell suspensions were prepared by passing through a 70-μm cell strainer, followed by debris removal with the Debris Removal Solution (Miltenyi, catalog number 130-109-398). Red blood cells were lysed using the 10X Red Blood Cell Lysis Solution (Miltenyi, catalog number 130-094-183). Remaining cells were enumerated using a Countess 3 cell counter (Invitrogen). Dead cells were then labeled with the Fixable Blue Dead Cell Stain Kit (Invitrogen, catalog number L23105). Cells were washed, pelleted, and blocked with 1 μg anti-CD16/anti-CD32 (2.4G2) at 4 °C for 5 min, followed by surface staining with the following antibodies at 4 °C for 30 min: BV450 anti-mouse CD45, eFluor450 anti-mouse CD11b, FITC anti-mouse CD11c, PE/Cy7 anti-mouse Ly6G, PerCP-Cy5.5 anti-mouse Ly6C, BV605 anti-mouse I-A/I-E, BV421 anti-mouse CD3, BUV563 anti-mouse CD4, APC anti-mouse CD8a, BV785 anti-mouse CD19, and APC/Cy7 anti-mouse P2RY12. Antibody information is provided in Supplementary Table 1. After staining, cells were washed and fixed with Cytofix (BD Biosciences). Flow cytometry data were acquired on a Cytek Aurora spectral flow cytometer (Cytek Biosciences) using SpectroFlo software and analyzed using FlowJo software (FlowJo). The gating strategy is shown in Supplementary Fig. 3. Intracellular cytokine staining Brain single-cell suspensions were prepared as described above. Cells were incubated in 1 mL RP10 medium (RPMI 1640 medium containing 10% FCS, 1 mM sodium pyruvate, 50 units/mL penicillin, and 50 μg/mL streptomycin) containing 5 μg/mL Brefeldin A (BioLegend, catalog number 420601) and 5 μg/mL LPS (eBioscience, catalog number 00-4976-93) for 6 h. Dead cells and surface antigens (CD45 and CD11b) were labeled as described above. Cells were then fixed and permeabilized using Cytofix/Cytoperm (BD Biosciences, catalog number 554714) according to the manufacturer’s instructions. TNF was stained using FITC-conjugated anti-mouse TNF antibody, and IL-1β was labeled using APC-conjugated anti-mouse IL-1β (Pro-form) antibody (Supplementary Table 1). Antibodies were diluted according to manufacturers’ instructions. Data acquisition and analysis were performed as described above. The gating strategy is shown in Supplementary Fig. 10. Tissue RNA isolation and RT-qPCR Total RNA was extracted from tissues using TRIzol (Invitrogen, catalog number 15596018) according to the manufacturer’s protocol. For first-strand cDNA synthesis, 3 μg of total RNA was used as template with M-MLV reverse transcriptase (Invitrogen, catalog number 28025021). The resulting cDNA was subjected to amplification of selected genes by real-time quantitative PCR using Power SYBR Green PCR Master Mix (Applied Biosystems, catalog number 4367659). Average values from technical duplicates were used to calculate relative transcript abundance, normalized to Hprt1, and presented as 2−ΔCT. Primer sequences are provided in Supplementary Table 2. CD11b+ brain cell RNA isolation and RT-qPCR Brain single-cell suspensions were prepared using the Adult Mouse Brain Dissociation Kit (Miltenyi, catalog number 130-107-677) according to the manufacturer’s protocol. CD11b+ cells were isolated using the CD11b MicroBeads (Miltenyi, catalog number 130-097-142) according to manufacturer’s instructions. Total RNA was extracted from isolated cells using TRIzol (Invitrogen, catalog number 15596018). Glycogen (Thermo Fisher Scientific, catalog number R0561) was added to facilitate RNA precipitation. For first-strand cDNA synthesis, 500 ng of RNA was used as template with SuperScript IV reverse transcriptase (Invitrogen, catalog number 18090010). The resulting cDNA was subjected to qPCR analysis as described above. Primer sequences for Ccl3 and Ccl4 are provided in Supplementary Table 2. Single-cell RNA sequencing (scRNA-seq) Mouse information for the scRNA-seq analysis is shown in Supplementary Table 3. Animals were deeply anesthetized with ketamine–xylazine and perfused transcardially with 10 mL PBS. Brains were collected, and brain single-cell suspensions were prepared using the Adult Mouse Brain Dissociation Kit (Miltenyi, catalog number 130-107-677) according to the manufacturer’s instructions. CD45+ cells were then enriched by staining with CD45 magnetic microbeads (Miltenyi, catalog number 130-052-301) and applying them to an LS column (Miltenyi, catalog number 130-042-401) following the manufacturer’s protocol. Isolated cells were pelleted and fixed overnight in 10x Genomics Fixation Buffer. Samples from 4–5 mice per experimental time point were pooled before submission to the University of Illinois Urbana-Champaign Genomics Core for processing with the 10x Genomics Chromium Fixed RNA Profiling (Flex) Gene Expression assay. All libraries were prepared and sequenced together on the Illumina NovaSeq X Plus System, generating 28 × 90 nt reads. Notably, probes targeting the SARS-CoV-2 ORF1ab and nucleocapsid (N) genes were custom-designed and spiked into the Chromium Mouse Transcriptome Probe Set (v1.0.1), following the manufacturer’s instructions. Probe sequences are provided in Supplementary Table 4. scRNA-seq data processing and analysis The quality of raw sequencing data was assessed using FastQC (v0.12.1), and summary reports were compiled with MultiQC (v1.12). Gene-count matrix was obtained by aligning reads to a customized reference genome consisting of the GRCm38 (mm10) assembly along with the SARS-CoV-2 sequence (EPI_ISL_1666328) using Cell Ranger (v7.2.0). Subsequent analysis was conducted in R (v4.5.0) using the Seurat package (v5.3.0). Cell-level quality control was applied using the following thresholds: 400 0.80. To focus on neuroimmune populations, non-immune cells were excluded based on the absence of Ptprc or Itgam expression using the subset function in Seurat. Genes detected in fewer than ten cells were also excluded from downstream analysis. The filtered data were normalized and scaled using SCTransform, with regression of cell cycle scores and mitochondrial gene content. Principal component analysis (PCA) was conducted using default parameters. The first 40 principal components (PCs) were used to compute a shared nearest neighbor graph and perform unsupervised clustering at a resolution of 0.6. To reduce ambient RNA artifacts, contamination in individual cells was estimated using DecontX (v1.6.0)100, with Seurat-derived clustering labels provided as priors. Clusters with high estimated contamination – confirmed by the expression of marker genes for endothelial cells, astrocytes, or neurons – were removed from downstream analysis. The DecontX-corrected count matrix was then reprocessed with SCTransform, followed by clustering at a resolution of 0.6. Clusters were manually annotated based on top upregulated genes identified by Seurat’s FindMarkers function, combined with reference to previously reported marker genes. Microglia were extracted from total cells using the subset function and further subclustered following SCTransform normalization and scaling. Pseudotime trajectory analysis in microglia was performed with monocle3 (v1.4.26)60, with Seurat-derived UMAP embeddings and subcluster assignments used as inputs for trajectory graph learning and pseudotime measurement. Correlation matrices of microglial subclusters were generated from log-normalized gene expression values, and Pearson correlation coefficients were calculated. Module scores were calculated using the AddModuleScore function in Seurat, and genes included in each module are listed in Supplementary Data 7. DEG analyses between conditions were conducted with the FindMarkers function in Seurat. Genes meeting the following thresholds were used for functional enrichment in gProfiler101: avg_log2FC >= 1, p_val_adj <= 0.01, pct.1 > 0.2 or avg_log2FC <= -1, p_val_adj <= 0.01, pct.2 > 0.2. Cell-cell communication analysis was conducted using the CellChat (v2.2.0) package86. Transcriptional distances of individual microglial subclusters at different post-infection time points relative to 0 dpi were quantified via PCA of the pseudobulk expression matrices generated for each subcluster. The Euclidean distance in PC space between each infected time point and the dpi 0 centroid was designated as the transcriptional distance, with greater distances indicating larger deviations from the mock state. Bulk RNA sequencing (RNA-seq) and data analysis Total RNA was isolated from OB as described above, followed by DNase I treatment and RNA purification. Purified RNA samples were submitted to Plasmidsaurus for RNA-seq analysis. Sequencing data were explored and analyzed using the Plasmidsaurus web interface. Statistical analysis Statistical analyses were performed using GraphPad Prism (v10.3.1) for RT–qPCR, behavior test, and multiplex flow cytometry data. Outliers were identified and removed using the ROUT method (Q = 1%). Data normality was assessed using the Shapiro–Wilk test. Two-sided unpaired t test, two-way ANOVA with Sidak’s multiple comparisons test, or one-way ANOVA with Dunnett’s multiple comparisons test was applied as appropriate. Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Supplementary Data 2 Supplementary Data 3 Supplementary Data 4 Supplementary Data 5 Supplementary Data 6 Supplementary Data 7 Reporting Summary Transparent Peer Review file Supplementary Information Description of Additional Supplementary Files Supplementary Data 1 Supplementary Data 2 Supplementary Data 3 Supplementary Data 4 Supplementary Data 5 Supplementary Data 6 Supplementary Data 7 Reporting Summary Transparent Peer Review file Source data Source Data Source Data