Abstract
Experimental evidence suggests that activated microglia induce astrocyte reactivity in neurodegenerative disorders, such as Alzheimer’s disease (AD). In this study, we investigated the association between microglial activation and amyloid-β (Aβ) with reactive astrogliosis in individuals across the AD spectrum. We examined 101 individuals using positron emission tomography radiotracers to assess Aβ deposition ([18F]AZD4694), tau aggregation ([18F]MK-6240) and microglial activation ([11C]PBR28), along with plasma biomarkers for astrocyte reactivity (GFAP) and tau phosphorylation (p-tau217). We further evaluated 251 individuals with cerebrospinal fluid levels of the microglial marker sTREM2. We found that Aβ pathology was associated with astrocyte reactivity across cortical brain regions only in the presence of microglial activation. The microglia-dependent effects of Aβ on astrocyte reactivity were further related to cognitive impairment through tau phosphorylation and aggregation. Our results suggest that microglial activation plays a key role in Aβ-related astrocyte reactivity, which, in turn, contributes to downstream pathological features of AD.
Similar content being viewed by others
Main
AD is biologically defined by the brain accumulation of Aβ and tau proteins1. However, a growing body of work indicates that these proteinopathies alone do not fully explain the clinical progression of AD2,3. This suggests that additional pathological mechanisms are involved in the AD pathological cascade. In fact, it has been proposed that neuroinflammation plays a pivotal role in determining patients’ susceptibility to dementia, with microglia and astrocytes orchestrating the repertoire of immune responses4.
In pathological conditions, astrocytes undergo molecular, functional and morphological changes, a process referred to as reactive astrogliosis5. Despite increasing evidence of astrocyte involvement in AD progression3, the biological underpinnings of reactive astrogliosis in this neurodegenerative condition remain elusive. Although Aβ pathology is an important trigger of astrocyte reactivity6,7,8, other pathological components of AD have also been implicated in promoting reactive astrogliosis, notably microglial activation/reactivity3. Instead of solely being a response to Aβ deposition, compelling experimental data demonstrate that activated microglia can induce astrocyte reactivity through the secretion of inflammatory cytokines9, highlighting a close relationship between Aβ and glial cells. This raises the possibility that microglial activation may account for the fact that Aβ drives astrocyte reactivity in some individuals but not in others10. However, to our knowledge, no previous clinical study has addressed whether microglial activation can modulate astrocyte reactivity in response to Aβ pathology in the living human brain. Better understanding the microglia−astrocyte crosstalk in AD may provide valuable insights for developing therapies targeting glial cells.
Using imaging and fluid biomarkers for the quantification of glial reactivity and AD hallmark proteinopathies, we tested the hypothesis that microglia impact the effects of Aβ pathology on reactive astrogliosis in individuals across the aging and AD clinical spectrum. In addition, we assessed whether gene expression patterns resemble the in vivo relationship between Aβ pathology and glial reactivity. Lastly, we investigated whether glial reactivity could aid in explaining the associations among Aβ, tau and cognitive impairment.
Results
We studied 101 individuals aged 50 years or older from the Translational Biomarkers in Aging and Dementia (TRIAD) cohort who had complete positron emission tomography (PET) data for Aβ ([18F]AZD4694) and translocator protein (TSPO) ([11C]PBR28) as well as clinical assessments and plasma glial fibrillary acidic protein (GFAP) measurements. Participants were also assessed for plasma phosphorylated tau at threonine 217 (p-tau217) and tau PET imaging ([18F]MK-6240). Selection of study participants for the primary sample is displayed in Extended Data Fig. 1. Individuals were genotyped for the rs6971 single-nucleotide polymorphism (SNP) in the TSPO gene, which determines the binding affinity of the [11C]PBR28 radiotracer for the TSPO11. Because the rs6971 SNP is a methodological caveat that does not influence TSPO levels, glial reactivity or AD pathophysiology12, only high-affinity binders were included in the present study to mitigate noise related to artificial uptake variations. Among the primary sample, a total of 62 were cognitively unimpaired, and 39 were cognitively impaired (26 with mild cognitive impairment (MCI) and 13 with AD dementia; Table 1 and Supplementary Table 1). Analyses with cerebrospinal fluid (CSF) soluble triggering receptor expressed on myeloid cells 2 (sTREM2) were conducted in the secondary sample, which was composed of 251 participants (214 cognitively unimpaired and 37 cognitively impaired) from the Wisconsin Registry for Alzheimer’s Prevention (WRAP) and TRIAD cohorts (Supplementary Table 2). These individuals had available data for Aβ PET ([18F]AZD4694 or [11C]PiB), CSF sTREM2 and plasma GFAP. Here, TSPO PET and CSF sTREM2 were used as markers of microglial activation, and plasma GFAP was used as a marker of astrocyte reactivity. Time differences among imaging, fluid and cognitive data are reported in Supplementary Tables 3 and 4.
Aβ pathology associates with astrocyte reactivity only in the presence of microglial activation
We first investigated the association of TSPO PET and Aβ PET with plasma GFAP in the primary sample. Regression analyses revealed a significant positive association between global Aβ PET standardized uptake value ratio (SUVR) and plasma GFAP levels in TSPO microglial activation-positive (MA+TSPO) individuals (β = 0.618, 95% confidence interval (CI): 0.310−0.926, P < 0.001) but not in TSPO microglial activation-negative (MA−TSPO) individuals (β = 0.185, 95% CI: −0.133 to 0.502, P = 0.247) as well as a significant interaction of Aβ PET SUVR and MATSPO status with plasma GFAP levels (β = 0.528, 95% CI: 0.172−0.883, P = 0.004; Fig. 1a and Supplementary Table 5). Additionally, a significant interaction between continuous values of Aβ PET and TSPO PET on plasma GFAP levels (β = 0.302, 95% CI: 0.116−0.489, P = 0.002; Fig. 1b and Supplementary Table 5) further supported that microglial activation affects the association of Aβ pathology with reactive astrogliosis, independently of the threshold used to define MA+TSPO. Analysis of variance corroborated the adequacy of the interaction model compared to the reduced models (including only Aβ PET, only TSPO PET or their additive effects; all P < 0.05). Similar findings were observed in sensitivity analyses using other summary measures of TSPO PET (Extended Data Fig. 2) as well as including outliers (Supplementary Table 6). Region-wise linear regressions showed that Aβ PET burden was positively associated with plasma GFAP levels only in the presence of MA+TSPO across cortical areas, including the frontal, parietal, temporal and cingulate cortices (Fig. 1c and Extended Data Fig. 3). In a subsequent analysis assessing continuous values for the topographical distribution of microglial activation, we found that the relationship between Aβ PET burden and plasma GFAP levels was influenced by TSPO PET uptake in the cingulate and frontal brain regions (Fig. 1d and Extended Data Fig. 4). Using microarray-based postmortem data from the Allen Human Brain Atlas (AHBA), we found that the greatest microglia-related effects of Aβ on astrocyte reactivity observed in our study population occurred in cortical regions with elevated TSPO mRNA expression, with the variogram estimation null model supporting a spatial correlation between brain maps (r = 0.253, P < 0.001; Fig. 1e). In the secondary sample, we examined the association of CSF sTREM2 and Aβ PET with plasma GFAP. Regression analyses demonstrated that higher Aβ PET burden was related to higher plasma GFAP levels, with a smaller magnitude in sTREM2 microglial activation-negative (MA−sTREM2) individuals (β = 0.009, 95% CI: 0.003−0.015, P = 0.004) and a greater magnitude in sTREM2 microglial activation-positive (MA+sTREM2) individuals (β = 0.024, 95% CI: 0.017−0.031, P < 0.001). The stronger effects according to microglial activation levels were supported by a significant interaction of Aβ PET and MAsTREM2 status on plasma GFAP levels (β = 0.013, 95% CI: 0.005−0.021, P = 0.001; Fig. 1f) as well as by the model assessing the association between continuous values of Aβ PET, CSF sTREM2 and plasma GFAP (β = 0.005, 95% CI: 0.001−0.008, P = 0.012; Fig. 1g). Analysis of variance confirmed that the investigated relationships were best described by the model with the interaction term in comparison to reduced models (including only Aβ PET, only CSF sTREM2 or their additive effects; all P < 0.05). Similar results were observed when stratifying participants by study cohort (Extended Data Fig. 5). Exploratory analyses demonstrated that sex affected the association between CSF sTREM2 (but not TSPO PET) and Aβ PET with plasma GFAP (Extended Data Fig. 6).
a, Scatter plot displaying the association between global Aβ PET and plasma GFAP according to MATSPO status (MA+TSPO: exact P = 2.01 × 10−4). Error bands represent the 95% CI of regression lines. b, Continuous association among global Aβ PET, posterior cingulate TSPO PET and plasma GFAP. c, Region-wise linear regression T-map showing the association between Aβ PET and plasma GFAP across cortical areas according to MATSPO status. d, Region-wise linear regression T-map showing the interaction of global Aβ PET and TSPO PET on plasma GFAP across cortical areas. e, Brain map showing the result of voxel-wise linear regression testing the interaction of Aβ PET and TSPO PET on plasma GFAP across the cortex. The color scale represents β estimates of the interaction term (top left). Brain map depicting TSPO mRNA expression across the cortex in six individuals from the AHBA. The color scale represents log2 mRNA expression intensity (top right). Spatial similarity between maps was assessed using the variogram-matching method with two-sided testing. Histogram depicting the null distribution, with the point indicating the empirical Pearson correlation coefficient (bottom; exact P = 5.00 × 10−4). f, Scatter plot displaying the association between global Aβ PET and plasma GFAP according to MAsTREM2 status (MA+sTREM2: exact P = 1.15 × 10−10). Error bands represent the 95% CI of regression lines. g, Continuous association among global Aβ PET, CSF sTREM2 and plasma GFAP. β estimates and corresponding two-sided P values were computed from linear regression models (a,b,f,g). In all regressions, biomarker levels were standardized, and models were adjusted for age, sex and cognitive status. The interaction models also accounted for the main effects of the interaction terms. For secondary analysis involving CSF sTREM2, Aβ PET was assessed in centiloids, and cohort was included as a variable for adjustment in regression models. The tilde symbol (~) indicates that the variable on the left (response variable) is regressed on the variable(s) on the right (predictor variable(s)).
Microglial and astrocyte reactivity jointly relate to tau phosphorylation and aggregation
Next, we assessed the association of TSPO PET and plasma GFAP with tau biomarkers. Although fluid p-tau and tau PET detect tau abnormalities in AD, these biomarkers reflect distinct aspects of tau pathology: fluid p-tau captures soluble hyperphosphorylated tau forms, whereas tau PET targets insoluble tau aggregates13. Regression analysis showed that plasma GFAP levels were positively associated with plasma p-tau217 levels in MA+TSPO individuals (β = 0.658, 95% CI: 0.323−0.993, P < 0.001) but not in MA−TSPO individuals (β = 0.110, 95% CI: −0.091 to 0.312, P = 0.276) and that the interaction between plasma GFAP and MATSPO status was significantly associated with higher plasma p-tau217 levels (β = 0.471, 95% CI: 0.125−0.816, P = 0.008; Fig. 2a and Supplementary Table 7). Moreover, a significant interaction was also observed between continuous levels of plasma GFAP and TSPO PET uptake on plasma p-tau217 levels (β = 0.201, 95% CI: 0.024−0.377, P = 0.026; Fig. 2b and Supplementary Table 7), reinforcing that the relationship between reactive astrogliosis and tau phosphorylation depends on microglial activation levels, irrespective of the threshold used to determine MA+TSPO. The adequacy of the interaction model compared to the reduced models (including only plasma GFAP, only TSPO PET or their additive effects) was supported by analysis of variance (all P < 0.05). Similar results were observed in sensitivity analyses including outliers (Supplementary Table 8). In region-wise linear regressions, we observed that higher plasma GFAP levels were associated with tau PET accumulation only in the presence of MA+TSPO mainly in neocortical brain regions, comprising temporal structures as well as association and sensorimotor cortices (Fig. 2c and Extended Data Fig. 7).
a, Scatter plot displaying the association between plasma GFAP and plasma p-tau217 according to MATSPO status (MA+TSPO: exact P = 2.96 × 10−4). β estimates and corresponding two-sided P values were computed from linear regression models. Error bands represent the 95% CI of regression lines. b, Continuous association among plasma GFAP, posterior cingulate TSPO PET and plasma p-tau217. β estimate and corresponding two-sided P value were computed from a linear regression model. c, Region-wise linear regression T-map showing the association between plasma GFAP and tau PET across cortical areas according to MATSPO status. In all regressions, biomarker levels were standardized, and models were adjusted for age, sex and cognitive status. The interaction models also accounted for the main effects of the interaction terms. d,e, Structural equation modeling testing the associations among Aβ (global Aβ PET), reactive astrogliosis (plasma GFAP), tau phosphorylation (plasma p-tau217), tau tangles (temporal meta-ROI tau PET) and cognition (MMSE) in MA−TSPO (d) and MA+TSPO (e) individuals. Associations were adjusted for age, sex and (if involving cognition) years of education. Solid lines with standardized β estimates and 95% CI represent significant associations, and dashed lines represent non-significant effects. Analyses involving plasma p-tau217 were performed in a subset of 93 individuals. From the total study population of 101 individuals, five participants did not have available plasma p-tau217 measurements, and three were excluded as they were considered outliers (plasma p-tau217 levels 3 s.d. above the mean of the population). MMSE, Mini-Mental State Examination.
Glial reactivity contributes to AD-related cognitive impairment
Lastly, we applied structural equation modeling to test the associations among Aβ, astrocyte reactivity, tau phosphorylation, tau tangles and cognition according to MATSPO status. In MA−TSPO individuals, we observed direct effects of Aβ PET load only on higher plasma p-tau217 levels and tau PET burden, with no significant associations with cognitive impairment (Fig. 2d). This model fit the data poorly (root mean squared error of approximation (RMSEA) = 0.257, standardized root mean square residual (SRMR) = 0.069, comparative fit index (CFI) = 0.888). Conversely, in MA+TSPO individuals, we found that plasma GFAP levels partially mediated the effects of Aβ PET load on higher plasma p-tau217, which was further associated with cognitive impairment through increased tau PET accumulation. The model also demonstrated pathways by which Aβ PET was related to cognitive deterioration via direct associations with tau pathology biomarkers (plasma p-tau217 and tau PET; Fig. 2e). This construct explained 76% of the variance in cognitive impairment and fit the data well (RMSEA = 0.000, SRMR = 0.018, CFI = 1.000). Supplementary Tables 9 and 10 report detailed coefficients and associated statistics for structural equation models. Exploratory analyses assessing tau PET across Braak-like regions and evaluating the associations of biomarkers with memory function are reported in Extended Data Figs. 8 and 9, respectively.
Discussion
In the present study, we observed that microglia modulate Aβ effects on astrocyte reactivity in the living human brain. We also found that the physiological distribution of TSPO gene expression in the postmortem brain resembles this in vivo relationship between Aβ pathology and glial reactivity. Lastly, we showed that the microglia-dependent impact of Aβ on astrocyte reactivity was further associated with cognitive impairment through tau phosphorylation and aggregation.
We found that Aβ pathology was associated with GFAP levels across cortical brain regions only in the presence of increased TSPO PET signal. Notably, we observed similar results when using CSF sTREM2, a different marker of microglial activation. Experimental studies revealed that Aβ is closely related to astrocyte reactivity6,7,14,15. Accordingly, neuropathological investigations showed that GFAP+ reactive astrocytes surround Aβ plaques and mirror the topographical distribution of Aβ deposition in the AD brain8,16. This is further supported by observations that CSF and, more prominently, plasma levels of GFAP are associated with Aβ biomarkers17,18. Beyond Aβ accumulation, an animal study demonstrated that microglial activation also triggers astrocyte reactivity by secreting interleukin 1α (IL-1α), tumor necrosis factor (TNF) and complement component 1, subcomponent q (C1q) cytokines9. Microglial activation is an important pathological feature of AD and has an intimate communication with reactive astrocytes in coordinating the innate immune response in the brain4,19. Here, we build on previous experimental data by showing clinical evidence that microglia modulate the Aβ-induced astrocyte reactivity in the living AD human brain. Analyzing postmortem data from the AHBA, we identified a spatial similarity between TSPO mRNA expression patterns and the topography of Aβ−glia interactions in the AD brain. Nonetheless, although strong correspondence was observed in some brain regions, the modest strength of the spatial correlation also revealed regional discrepancies between brain maps. We interpret this finding as biologically meaningful, suggesting that the physiological distribution of TSPO gene expression may help to partially explain microglia-driven effects of Aβ on astrocyte reactivity across the cortex. It is likely that regional heterogeneity of glial cells in aging and disease20,21 contributes to explaining the mismatch across some cortical brain areas. Interestingly, exploratory analyses revealed sex differences in the association between Aβ and glial reactivity when assessing microglial activation with CSF sTREM2 but not with TSPO PET. In line with previous studies suggesting that sex modulates microglial effects in AD and also contributes to microglial heterogeneity22,23,24, our results might reflect distinct sex effects across the microglial phenotypes captured by each biomarker. However, these findings should be interpreted with caution, as this study was not designed to test sex effects and included a higher proportion of women. Whether the role of activated microglia in modulating Aβ-related astrocyte reactivity differs by sex warrants further investigation.
We observed that TSPO PET and plasma GFAP levels were jointly associated with tau biomarkers. Rather than merely being a response to the accumulation of AD hallmark proteins, a growing body of work supports the notion that both microglia and astrocytes are involved in tau pathogenesis through the exacerbation of tau spread and hyperphosphorylation as well as tau-driven neurodegeneration25,26,27,28,29,30. Accordingly, recent biomarker-based clinical studies have suggested that glial reactivity might help to explain the link between Aβ and tau pathology. More specifically, it was demonstrated that astrocyte reactivity, indexed by plasma GFAP, impacts Aβ effects on tau pathology in individuals across the AD continuum10,31. In addition, neuroimaging data demonstrated that the interaction between Aβ and activated microglia is crucial for tau pathology progression32. In the present study, we propose a model in which the interplay between Aβ and reactive astrogliosis contributes to tau pathology in the presence of activated microglia. It is important to consider that fluid p-tau and tau PET are complementary tools to assess tau pathology, with notable differences in their timing relationships13. Fluid p-tau captures soluble hyperphosphorylated tau forms and has been shown to become abnormal in preclinical stages, preceding detectable changes in tau PET signal33,34,35. The stronger association with Aβ PET than with tau PET36, along with evidence that it mediates Aβ effects on tangle accumulation37,38, supports p-tau as a marker reflecting pathological tau phosphorylation driven by Aβ pathology. By contrast, tau PET targets insoluble tau tangles, with abnormalities in the neocortex typically detected in AD symptomatic stages39. The topography of tau PET signal has been shown to closely relate to downstream AD features, as it is a major predictor of brain atrophy40 and clinical progression41, corroborating the use of tau PET as a late measure of tau pathology that indicates tau aggregation. Our results align with these concepts by revealing glial-mediated effects of Aβ on pathological tau phosphorylation, which were further associated with subsequent tau aggregation, ultimately leading to the manifestation of cognitive symptoms. Altogether, these observations reinforce the importance of microglial activation and astrocyte reactivity in AD progression, suggesting glial cells as potential therapeutic targets for future disease-modifying clinical trials.
Methodological strengths of our study include the evaluation of a large cohort genetically enriched to increase the reliability of TSPO PET imaging. Moreover, the most advanced methodologies for the brain quantification of Aβ, tau and glial reactivity were used. Notably, microglial activation was assessed using two biomarkers (TSPO PET and CSF sTREM2) across well-characterized cohorts. This study also has methodological limitations. TRIAD and WRAP participants are volunteers who were motivated to participate in a study about dementia, which might limit the generalizability of our findings due to self-selection bias. Here, we used [11C]PBR28 TSPO PET and CSF sTREM2 as indices of microglial activation and plasma GFAP as an index of astrocyte reactivity. However, neuroinflammation involves various glial phenotypes during AD progression5,24, which could not be captured in our biomarker-based study. Although TSPO is predominantly expressed in the AD brain by microglia42,43,44,45, it has also been detected in other cell types (for example, astrocytes and endothelial cells)12,44,45,46. In addition, there is an ongoing discussion about whether the TSPO PET signal more accurately reflects microglial activation/reactivity or density/recruitment47. To elucidate the biological mechanisms underlying our findings, future research integrating biomarkers with phenotypically characterized glial cells is needed to better understand the neuroinflammatory response in AD. Lastly, it is important to note that the cross-sectional design of the present study limits causal inferences. Our results should be expanded with multimodal longitudinal data to capture the temporal dynamics and sequential relationships between glial and AD biomarkers.
In conclusion, our results support the construct that microglia activation is a key factor in Aβ-dependent reactive astrogliosis, which further contributes to cognitive impairment via the aggregation of phosphorylated tau in the living AD human brain.
Methods
The present study complies with all relevant ethical regulations, in accordance with the Declaration of Helsinki. The TRIAD study was approved by the Montreal Neurological Institute (MNI) PET Working Committee and the Douglas Mental Health University Institute Research Ethics Board (IUSMD-16-60). The WRAP study was approved by the University of Wisconsin-Madison Institutional Review Board (2023-1522). All participants provided written informed consent, and they were compensated in accordance with the number of study procedures completed.
Participants
The present work included individuals enrolled in the TRIAD (https://triad.tnl-mcgill.com) and WRAP (https://wrap.wisc.edu) cohorts. TRIAD is a biomarker-based study launched in 2017 that comprises participants from the community or outpatients at the McGill University Research Centre for Studies in Aging, Canada. Exclusion criteria encompassed inability to speak English or French, inadequate auditory and visual capacities for neuropsychologic testing, contraindications for PET or magnetic resonance imaging (MRI), recent head trauma or major surgery, inadequately treated conditions, current enrollment in other studies or active substance abuse. Cognitively unimpaired individuals had no objective cognitive impairment and a global Clinical Dementia Rating (CDR) of 0. Participants with MCI had preserved activities of daily living, subjective and/or objective cognitive impairment and a global CDR of 0.5 (ref. 48). Patients with mild-to-moderate AD dementia met the National Institute on Aging and the Alzheimer’s Association (NIA-AA) criteria for probable AD49 and had a global CDR score between 0.5 and 2. WRAP is an observational study50 conducted by the Wisconsin Alzheimer Institute at the University of Wisconsin School of Medicine and Public Health. Since 2001, the cohort has enrolled dementia-free individuals from the community, enriched for a parental history of probable AD dementia. Inclusion criteria comprised fluency in English, adequate visual and auditory acuity to complete neuropsychological assessments and good overall health, with no medical conditions expected to affect study participation. Cognitive status at each study visit was defined based on a consensus diagnosis by a multiprofessional committee of dementia specialists50, with MCI diagnosis following the NIA-AA criteria51.
In the primary sample, individuals underwent genotyping for the Ala147Thr polymorphism in the TSPO gene (rs6971, https://www.ncbi.nlm.nih.gov/snp/rs6971). Depending on this genotype, participants can present three [11C]PBR28 binding affinity phenotypes: low-affinity, mixed-affinity or high-affinity binders11. [11C]PBR28 signal is negligible in low-affinity binders and heterogeneous in mixed-affinity binders52. Thus, only high-affinity binders were included to minimize artificial inter-subject TSPO PET signal variability32. All participants included in the primary sample were from the TRIAD study and had available Aβ PET, TSPO microglial activation PET, MRI, plasma GFAP and clinical assessments. We also assessed plasma p-tau217 and tau tangle PET data. In the primary sample, two individuals who had plasma GFAP levels 3 s.d. above the mean of the population were considered outliers and were excluded from the analyses. Secondary analyses assessing microglial activation with CSF sTREM2 were performed in individuals without dementia from the WRAP and TRIAD studies with available Aβ PET, MRI, CSF sTREM2, plasma GFAP and clinical assessments. To minimize inter-cohort differences, this set of analyses was restricted to cognitively unimpaired and MCI individuals, given that no WRAP participant had a clinical diagnosis of dementia. In the secondary sample, three TRIAD participants and four WRAP participants with biomarker levels (Aβ PET, CSF sTREM2 or plasma GFAP) 3 s.d. above the mean of the population were identified as outliers, as defined a priori and excluded from the analyses. Of note, 43 TRIAD participants were shared between the primary and secondary samples.
Fluid biomarkers
Blood and CSF collection followed TRIAD36,53 and WRAP54,55 procedures. Plasma biomarkers were quantified using single-molecule array (Simoa) methods on the HD-X platform (Quanterix). In TRIAD and WRAP, plasma GFAP concentration was measured with a commercially available assay from Quanterix at the Clinical Neurochemistry Laboratory, University of Gothenburg17. Plasma p-tau217 concentration in TRIAD was measured at Johnson & Johnson56. CSF sTREM2 was measured using an in-house assay on a SECTOR Imager 6000 (MesoScale Discovery)57 in TRIAD and the NeuroToolKit on a cobas e 411 analyzer (Roche Diagnostics)58 in WRAP, both at the Clinical Neurochemistry Laboratory, University of Gothenburg. We used a residual approach to calculate batch-adjusted biomarker values when appropriate. In TRIAD, sTREM2 microglial activation positivity (MA+sTREM2) was determined as CSF sTREM2 levels surpassing the mean plus 2 s.d. from a separate population of 24 cognitively unimpaired Aβ− young adults (Supplementary Table 11). In WRAP, the positivity threshold was set to correspond to the same percentile of MA+sTREM2 observed among cognitively unimpaired TRIAD participants aged 65−75 years.
Brain gene expression
We obtained TSPO gene expression data in the entire brain from the open-source AHBA (http://www.brain-map.org)59. In brief, mRNA expression intensity values were derived from microarray data of 3,702 samples from six healthy postmortem human brains (Supplementary Table 12). Microarray-based TSPO mRNA brain expression map, generated using Gaussian process regression60, was retrieved from https://www.meduniwien.ac.at/neuroimaging/mRNA.html.
Neuroimaging
For TRIAD, all participants had a T1-weighted MRI that was used for co-registration18. Structural MRI data were acquired at the MNI on a 3T Siemens Magnetom scanner using a standard head coil. High-resolution structural images of the whole brain were obtained with the magnetization-prepared rapid acquisition gradient echo MRI sequence. All PET scans were acquired at the MNI on the same brain-dedicated Siemens High-Resolution Research Tomograph. Aβ PET with [18F]AZD4694 (acquired 40−70 minutes after injection), tau tangle PET with [18F]MK-6240 (acquired 90−110 minutes after injection) and TSPO microglial activation PET with [11C]PBR28 (acquired 60−90 minutes after injection) were reconstructed with the ordered subset expectation maximization (OSEM) algorithm on a four-dimensional volume with three frames (3 × 600 seconds)61, four frames (4 × 300 seconds)61 and six frames (6 × 300 seconds)32, respectively. After acquisition, PET images were corrected for attenuation, motion, dead time, decay and random and scattered coincidences. T1-weighted MRI images were corrected for non-uniformity and field distortions in accordance with an in-house pipeline. Then, imaging co-registration and spatial normalization to the Alzheimer’s Disease Neuroimaging Initiative (ADNI) template space were performed. The ADNI template space is a stereotactic template aligned to MNI coordinates, which is derived from MRI scans of cognitively unimpaired elderly individuals, individuals with MCI and individuals with AD dementia and, thus, represents an anatomical template optimized for neuroimaging studies in older adults across the aging and AD spectrum62. PET images were automatically registered to the native T1-weighted MRI with linear transformations, and T1-weighted MRI images were linearly and nonlinearly registered to the ADNI template space. Subsequently, PET images were registered to the ADNI template space by applying the linear and nonlinear transformation parameters from PET to native MRI and native MRI to the ADNI template space. Aβ PET SUVR and TSPO PET SUVR were calculated using the whole cerebellar gray matter as reference region32,63. Tau PET SUVR was calculated using the inferior cerebellar gray matter as reference region39. PET images were spatially smoothed to achieve a final resolution of 8-mm full width at half-maximum. The Desikan−Killiany−Tourville atlas was used to determine the anatomical regions of interest (ROIs)64. The global Aβ PET SUVR composite was estimated from the following brain regions: precuneus, prefrontal, orbitofrontal, parietal, temporal and cingulate65. Aβ positivity (Aβ+) was established as global [18F]AZD4694 SUVR ≥ 1.55 (ref. 66). We used the temporal meta-ROI as a summary measure for tau PET, as it has been proposed to be an appropriate composite for detecting early tau aggregation in AD by including both medial temporal and neocortical areas, thereby encompassing regions affected across the aging and AD continuum65,67. The temporal meta-ROI tau PET SUVR composite was estimated from the following brain regions: entorhinal, hippocampus, fusiform, parahippocampal, inferior temporal and middle temporal65. Tau PET SUVR was also estimated across Braak-like regions, as described elsewhere39. We assessed TSPO PET in the posterior cingulate as we identified higher [11C]PBR28 uptake in this brain region as a cortical microglial activation/reactivity signature of AD (Extended Data Fig. 10)32. We also conducted sensitivity analyses using other summary measures of TSPO PET, defined as follows: (1) a composite brain region encompassing all areas with increased TSPO PET SUVR in AD Aβ+ compared to cognitively unimpaired Aβ− and (2) a composite brain region comprising areas within the top decile (that is, above the 90th percentile) for TSPO PET SUVR differences between AD Aβ+ and cognitively unimpaired Aβ− individuals. Microglial activation positivity (MA+TSPO) was defined as SUVR values 2 s.d. above the mean from a separate population of 17 cognitively unimpaired Aβ− young adults (Supplementary Table 11).
For WRAP, all participants included in this study had acquired a T1-weighted MRI for co-registration purposes on a 3T GE Signa 750 scanner using an eight-channel head coil at the University of Wisconsin-Madison Waisman Center Brain Imaging Lab68. PET imaging with [11C]Pittsburgh Compound-B ([11C]PiB) was acquired on a Siemens HR+ scanner at the University of Wisconsin-Madison Waisman Center Brain Imaging Lab68. Dynamic [11C]PiB PET images were acquired 0−70 minutes after injection of the radiotracer. Images were reconstructed with a filtered back-projection algorithm, with five 2-minute frames followed by 12 5-minute frames (total of 17 frames). Corrections were applied for random events, attenuation of annihilation radiation, dead time, scanner normalization, scatter radiation and motion. [11C]PiB PET data were co-registered to each participant’s T1-weighted MRI and transformed into voxel-wise parametric distribution volume ratio (DVR) maps using Logan graphical analysis (35−70-minute interval and mean efflux constant of 0.149 per minute), with the cerebellum gray matter as the reference region. DVRs were spatially normalized to the MNI space and smoothed with an 8-mm full width at half maximum Gaussian kernel. Global Aβ PET burden was calculated by averaging the [11C]PiB DVR from eight bilateral ROIs (angular gyrus, anterior cingulate gyrus, posterior cingulate gyrus, frontal medial orbital gyrus, precuneus, supramarginal gyrus, middle temporal gyrus and superior temporal gyrus)69, which were defined using the Automated Anatomical Labeling atlas70 and linearly translated to the centiloid scale71.
Statistical analysis
Statistical analyses were carried out using R software (version 4.0.2, http://www.r-project.org/), unless stated otherwise. The associations between biomarkers were tested using regression models accounting for age, sex and cognitive status. Regression models including interaction terms also accounted for the main effects of the variables involved in the interaction. Biomarker levels were standardized prior to model entry to facilitate comparison across estimates. For secondary analysis involving CSF sTREM2, fluid biomarkers were standardized based on the mean and s.d. of each cohort’s cognitively unimpaired Aβ− subset, and cohort was included as a variable for adjustment in regression models combining WRAP and TRIAD data. Normality of residuals, homoscedasticity and linearity were visually assessed, but this was not formally tested. Additionally, exploratory analyses were conducted to investigate sex effects on the association of microglial activation markers (TSPO PET and CSF sTREM2) and Aβ PET with plasma GFAP. Multiple comparisons correction with the false discovery rate method at P < 0.05 and sensitivity analysis were performed for primary analyses. Analysis of variance was used to test the adequacy of models with the interaction term compared to reduced models. The Python neuromaps toolbox72 was used to test whether TSPO mRNA expression intensity resembles Aβ−glia relationships across cortical brain regions in our population, applying the spatial autocorrelation-preserving variogram-matching model73 at the voxel level (2-mm resolution), with null distributions derived from 1,000 null maps. For this, voxel-wise analysis assessing Aβ−glia relationships was carried out using the MATLAB package VoxelStats74. Structural equation modeling was applied using the R package lavaan75 to test the associations among Aβ, tau, glial reactivity and cognition. Associations were adjusted for age, sex and (if involving cognition) years of education. Notably, structural equation models were constructed to assess specific hypotheses represented in the figure’s meta-models, and the following thresholds were used for considering a good fit: CFI > 0.97; RMSEA < 0.05; SRMR < 0.05 (refs. 76,77). Sample sizes were based on data availability, and, thus, no statistical methods were used to predetermine sample sizes. The collection and processing of biomarkers were performed in a blinded manner for clinical diagnosis. For all analyses, we considered a two-sided P < 0.05 as statistically significant.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
The data from the TRIAD study used in the present work are not publicly available, as the information could compromise participant privacy. Thus, anonymized data will be made available from the corresponding author upon reasonable request from a qualified academic researcher for the purpose of replicating the procedures and results presented in this article. Arrangements for data sharing are subject to standard data-sharing agreements, and further information can be found on the study’s website (https://triad.tnl-mcgill.com/). The data from the WRAP study can be requested through an online submission process (https://wrap.wisc.edu). AHBA microarray data are openly available at http://www.brain-map.org, and mRNA brain expression maps can be downloaded from https://www.meduniwien.ac.at/neuroimaging/mRNA.html.
Code availability
The codes used for data analysis in the present work are publicly available via GitHub at https://github.com/Tyrev/2025_JPFS_NatureNeuroscience. The codes were implemented in R (version 4.0.2) and Python (version 3.12.3) using open-source packages.
References
Knopman, D. S. et al. Alzheimer disease. Nat. Rev. Dis. Primers 7, 33 (2021).
Herrup, K. The case for rejecting the amyloid cascade hypothesis. Nat. Neurosci. 18, 794–799 (2015).
Carter, S. F. et al. Astrocyte biomarkers in Alzheimer’s disease. Trends Mol. Med. 25, 77–95 (2019).
Heneka, M. T. et al. Neuroinflammation in Alzheimer’s disease. Lancet Neurol. 14, 388–405 (2015).
Escartin, C. et al. Reactive astrocyte nomenclature, definitions, and future directions. Nat. Neurosci. 24, 312–325 (2021).
Hu, J., Akama, K. T., Krafft, G. A., Chromy, B. A. & Van Eldik, L. J. Amyloid-β peptide activates cultured astrocytes: morphological alterations, cytokine induction and nitric oxide release. Brain Res. 785, 195–206 (1998).
Johnstone, M., Gearing, A. J. & Miller, K. M. A central role for astrocytes in the inflammatory response to β-amyloid; chemokines, cytokines and reactive oxygen species are produced. J. Neuroimmunol. 93, 182–193 (1999).
Osborn, L. M., Kamphuis, W., Wadman, W. J. & Hol, E. M. Astrogliosis: an integral player in the pathogenesis of Alzheimer’s disease. Prog. Neurobiol. 144, 121–141 (2016).
Liddelow, S. A. et al. Neurotoxic reactive astrocytes are induced by activated microglia. Nature 541, 481–487 (2017).
Bellaver, B. et al. Astrocyte reactivity influences amyloid-β effects on tau pathology in preclinical Alzheimer’s disease. Nat. Med. 29, 1775–1781 (2023).
Owen, D. R. et al. An 18-kDa translocator protein (TSPO) polymorphism explains differences in binding affinity of the PET radioligand PBR28. J. Cereb. Blood Flow Metab. 32, 1–5 (2012).
Gui, Y., Marks, J. D., Das, S., Hyman, B. T. & Serrano-Pozo, A. Characterization of the 18 kDa translocator protein (TSPO) expression in post-mortem normal and Alzheimer’s disease brains. Brain Pathol. 30, 151–164 (2020).
Jack, C. R. Jr et al. Revised criteria for diagnosis and staging of Alzheimer’s disease: Alzheimer’s Association Workgroup. Alzheimers Dement. 20, 5143–5169 (2024).
Diniz, L. P. et al. Astrocyte transforming growth factor beta 1 protects synapses against Aβ oligomers in Alzheimer’s disease model. J. Neurosci. 37, 6797–6809 (2017).
Preman, P. et al. Human iPSC-derived astrocytes transplanted into the mouse brain undergo morphological changes in response to amyloid-β plaques. Mol. Neurodegener. 16, 68 (2021).
Serrano-Pozo, A. et al. Reactive glia not only associates with plaques but also parallels tangles in Alzheimer’s disease. Am. J. Pathol. 179, 1373–1384 (2011).
Benedet, A. L. et al. Differences between plasma and cerebrospinal fluid glial fibrillary acidic protein levels across the Alzheimer disease continuum. JAMA Neurol. 78, 1471–1483 (2021).
Ferrari-Souza, J. P. et al. Astrocyte biomarker signatures of amyloid-β and tau pathologies in Alzheimer’s disease. Mol. Psychiatry 27, 4781−4789 (2022).
Matejuk, A. & Ransohoff, R. M. Crosstalk between astrocytes and microglia: an overview. Front. Immunol. 11, 1416 (2020).
Serrano-Pozo, A. et al. Astrocyte transcriptomic changes along the spatiotemporal progression of Alzheimer’s disease. Nat. Neurosci. 27, 2384–2400 (2024).
Bottcher, C. et al. Human microglia regional heterogeneity and phenotypes determined by multiplexed single-cell mass cytometry. Nat. Neurosci. 22, 78–90 (2019).
Kodama, L. et al. Microglial microRNAs mediate sex-specific responses to tau pathology. Nat. Neurosci. 23, 167–171 (2020).
Casaletto, K. B. et al. Sex-specific effects of microglial activation on Alzheimer’s disease proteinopathy in older adults. Brain 145, 3536–3545 (2022).
Fumagalli, L. et al. Microglia heterogeneity, modeling and cell-state annotation in development and neurodegeneration. Nat. Neurosci. 28, 1381−1392 (2025).
Garwood, C. J., Pooler, A. M., Atherton, J., Hanger, D. P. & Noble, W. Astrocytes are important mediators of Aβ-induced neurotoxicity and tau phosphorylation in primary culture. Cell Death Dis. 2, e167 (2011).
Richetin, K. et al. Tau accumulation in astrocytes of the dentate gyrus induces neuronal dysfunction and memory deficits in Alzheimer’s disease. Nat. Neurosci. 23, 1567–1579 (2020).
Mann, C. N. et al. Astrocytic α2-Na+/K+ ATPase inhibition suppresses astrocyte reactivity and reduces neurodegeneration in a tauopathy mouse model. Sci. Transl. Med. 14, eabm4107 (2022).
Hopp, S. C. et al. The role of microglia in processing and spreading of bioactive tau seeds in Alzheimer’s disease. J. Neuroinflammation 15, 269 (2018).
Ising, C. et al. NLRP3 inflammasome activation drives tau pathology. Nature 575, 669–673 (2019).
Mancuso, R. et al. CSF1R inhibitor JNJ-40346527 attenuates microglial proliferation and neurodegeneration in P301S mice. Brain 142, 3243–3264 (2019).
Cogswell, P. M. et al. Modeling the temporal evolution of plasma p-tau in relation to amyloid beta and tau PET. Alzheimers Dement. 20, 1225–1238 (2024).
Pascoal, T. A. et al. Microglial activation and tau propagate jointly across Braak stages. Nat. Med. 27, 1592–1599 (2021).
Mila-Aloma, M. et al. Plasma p-tau231 and p-tau217 as state markers of amyloid-β pathology in preclinical Alzheimer’s disease. Nat. Med. 28, 1797–1801 (2022).
Mattsson-Carlgren, N. et al. Aβ deposition is associated with increases in soluble and phosphorylated tau that precede a positive Tau PET in Alzheimer’s disease. Sci. Adv. 6, eaaz2387 (2020).
Janelidze, S. et al. Associations of plasma phospho-tau217 levels with tau positron emission tomography in early Alzheimer disease. JAMA Neurol. 78, 149–156 (2021).
Therriault, J. et al. Association of phosphorylated tau biomarkers with amyloid positron emission tomography vs tau positron emission tomography. JAMA Neurol. 80, 188−199 (2022).
Mattsson-Carlgren, N. et al. Soluble P-tau217 reflects amyloid and tau pathology and mediates the association of amyloid with tau. EMBO Mol. Med. 13, e14022 (2021).
Pichet Binette, A. et al. Amyloid-associated increases in soluble tau relate to tau aggregation rates and cognitive decline in early Alzheimer’s disease. Nat. Commun. 13, 6635 (2022).
Pascoal, T. A. et al. 18F-MK-6240 PET for early and late detection of neurofibrillary tangles. Brain 143, 2818–2830 (2020).
La Joie, R. et al. Prospective longitudinal atrophy in Alzheimer’s disease correlates with the intensity and topography of baseline tau-PET. Sci. Transl. Med. 12, eaau5732 (2020).
Ossenkoppele, R. et al. Amyloid and tau PET-positive cognitively unimpaired individuals are at high risk for future cognitive decline. Nat. Med. 28, 2381–2387 (2022).
Cosenza-Nashat, M. et al. Expression of the translocator protein of 18 kDa by microglia, macrophages and astrocytes based on immunohistochemical localization in abnormal human brain. Neuropathol. Appl. Neurobiol. 35, 306–328 (2009).
Venneti, S., Wang, G., Nguyen, J. & Wiley, C. A. The positron emission tomography ligand DAA1106 binds with high affinity to activated microglia in human neurological disorders. J. Neuropathol. Exp. Neurol. 67, 1001–1010 (2008).
Ji, B. et al. Detection of Alzheimer’s disease-related neuroinflammation by a PET ligand selective for glial versus vascular translocator protein. J. Cereb. Blood Flow Metab. 41, 2076–2089 (2021).
Tournier, B. B. et al. TSPO and amyloid deposits in sub-regions of the hippocampus in the 3xTgAD mouse model of Alzheimer’s disease. Neurobiol. Dis. 121, 95–105 (2019).
Tournier, B. B. et al. Fluorescence-activated cell sorting to reveal the cell origin of radioligand binding. J. Cereb. Blood Flow Metab. 40, 1242–1255 (2020).
Kreisl, W. C. et al. PET imaging of neuroinflammation in neurological disorders. Lancet Neurol. 19, 940–950 (2020).
Petersen, R. C. Mild cognitive impairment as a diagnostic entity. J. Intern. Med. 256, 183–194 (2004).
McKhann, G. M. et al. The diagnosis of dementia due to Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement. 7, 263–269 (2011).
Johnson, S. C. et al. The Wisconsin Registry for Alzheimer’s Prevention: a review of findings and current directions. Alzheimers Dement. 10, 130–142 (2018).
Albert, M. S. et al. The diagnosis of mild cognitive impairment due to Alzheimer’s disease: recommendations from the National Institute on Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer’s disease. Alzheimers Dement. 7, 270–279 (2011).
Kreisl, W. C. et al. A genetic polymorphism for translocator protein 18 kDa affects both in vitro and in vivo radioligand binding in human brain to this putative biomarker of neuroinflammation. J. Cereb. Blood Flow Metab. 33, 53–58 (2013).
Karikari, T. K. et al. Blood phosphorylated tau 181 as a biomarker for Alzheimer’s disease: a diagnostic performance and prediction modelling study using data from four prospective cohorts. Lancet Neurol. 19, 422–433 (2020).
Jonaitis, E. M. et al. Plasma phosphorylated tau 217 in preclinical Alzheimer’s disease. Brain Commun. 5, fcad057 (2023).
Johnson, S. C. et al. Identifying clinically useful biomarkers in neurodegenerative disease through a collaborative approach: the NeuroToolKit. Alzheimers Res. Ther. 15, 25 (2023).
Triana-Baltzer, G. et al. Development and validation of a high-sensitivity assay for measuring p217+tau in plasma. Alzheimers Dement. 13, e12204 (2021).
Jensen, C. S. et al. Exercise as a potential modulator of inflammation in patients with Alzheimer’s disease measured in cerebrospinal fluid and plasma. Exp. Gerontol. 121, 91–98 (2019).
Van Hulle, C. et al. An examination of a novel multipanel of CSF biomarkers in the Alzheimer’s disease clinical and pathological continuum. Alzheimers Dement. 17, 431–445 (2021).
Hawrylycz, M. J. et al. An anatomically comprehensive atlas of the adult human brain transcriptome. Nature 489, 391–399 (2012).
Gryglewski, G. et al. Spatial analysis and high resolution mapping of the human whole-brain transcriptome for integrative analysis in neuroimaging. Neuroimage 176, 259–267 (2018).
Pascoal, T. A. et al. In vivo quantification of neurofibrillary tangles with [18F]MK-6240. Alzheimers Res. Ther. 10, 74 (2018).
Fonov, V., Coupe, P., Eskildsen, S. & Collins, D. P4-053: Atrophy specific MRI brain template for Alzheimer’s disease and mild cognitive impairment. Alzheimers Dement. 7, S717 (2011).
Cselenyi, Z. et al. Clinical validation of 18F-AZD4694, an amyloid-β−specific PET radioligand. J. Nucl. Med. 53, 415–424 (2012).
Klein, A. & Tourville, J. 101 labeled brain images and a consistent human cortical labeling protocol. Front. Neurosci. 6, 171 (2012).
Jack, C. R. Jr et al. Defining imaging biomarker cut points for brain aging and Alzheimer’s disease. Alzheimers Dement. 13, 205–216 (2017).
Therriault, J. et al. Determining amyloid-β positivity using 18F-AZD4694 PET Imaging. J. Nucl. Med. 62, 247–252 (2021).
Villemagne, V. L. et al. What is T+? A Gordian knot of tracers, thresholds, and topographies. J. Nucl. Med. 62, 614–619 (2021).
Johnson, S. C. et al. Amyloid burden and neural function in people at risk for Alzheimer’s disease. Neurobiol. Aging 35, 576–584 (2014).
Sprecher, K. E. et al. Amyloid burden is associated with self-reported sleep in nondemented late middle-aged adults. Neurobiol. Aging 36, 2568–2576 (2015).
Tzourio-Mazoyer, N. et al. Automated anatomical labeling of activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single-subject brain. Neuroimage 15, 273–289 (2002).
Betthauser, T. J. et al. Multi-method investigation of factors influencing amyloid onset and impairment in three cohorts. Brain 145, 4065–4079 (2022).
Markello, R. D. et al. neuromaps: structural and functional interpretation of brain maps. Nat. Methods 19, 1472–1479 (2022).
Burt, J. B., Helmer, M., Shinn, M., Anticevic, A. & Murray, J. D. Generative modeling of brain maps with spatial autocorrelation. Neuroimage 220, 117038 (2020).
Mathotaarachchi, S. et al. VoxelStats: a MATLAB package for multi-modal voxel-wise brain image analysis. Front. Neuroinform.10, 20 (2016).
Rosseel, Y. lavaan: an R package for structural equation modeling. J. Stat. Softw. 48, 1–36 (2012).
Mueller, R. O. & Hancock, G. R. in Best Practices in Quantitative Methods (ed. Osborne, J.) 488−508 (SAGE, 2008).
Schermelleh-Engel, K., Moosbrugger, H. & Müller, H. Evaluating the fit of structural equation models: tests of significance and descriptive goodness-of-fit measures. Methods Psychol. Res. 8, 23–74 (2003).
Acknowledgements
We would like to express our sincere gratitude to all study participants and relatives who made this research possible. We thank the TRIAD and WRAP team members for their time and dedication. We acknowledge the assistance of R. Hopewell and G. Massarweh with radiochemistry production at the McConnell Brain Imaging Centre, McGill University. Acknowledgement is also made to the brain donors as well as to collaborators and contributors for the AHBA. We thank Johnson & Johnson for the quantification of plasma p-tau217. This study was supported by the following funding agencies: Alzheimer’s Association (grants 24AARFD-1243899 to G.P.; AARFD-22-974627 to B.B.; AARFD-22-923814 to P.C.L.F.; AARFD-23-1148735 to M.A.D.B.; 24AACSF-1200375 to C.S.A.; AACSF-D 22-928689 to W.V.B.; ADSF-21-831376-C, ADSF-21-831381-C, ADSF-21-831377-C and ADSF-24-1284328-C to H.Z.; ZEN-21-848495 and SG-23-1038904 QC to K.B.; NIRG-12-92090 and NIRP-12-259245 to P.R.-N.; AARGD-21-850670, AACSFD-22-928689, AARFD-23-1148735 and BFECAA2024 to E.R.Z.); Alzheimer’s Drug Discovery Foundation (ADDF; grants 201809-2016862 to H.Z. and RDAPB-201809-2016615 to K.B.); Bluefield Project (grant to H.Z.); Brain Canada Foundation (Canadian Foundation for Innovation Project 34874 and 33397 to P.R.-N.); Canadian Consortium on Neurodegeneration in Aging (grant MOP-11-51-31-team 1 to P.R.-N.); Canadian Institutes of Health Research (grants MOP-11-51-31 and RFN 152985, 159815 and 162303 to P.R.-N.); Colin J. Adair Charitable Foundation (grant to P.R.-N.); Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq; grants 200691/2021-0 to J.P.F.-S. and 12410/2018-2, 435642/2018-9, 312306/2021-0, 409066/2022-2, 447074/2023-7, 409595/2023-3 and 444880/2024-0 to E.R.Z.); Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES; grants 88887.627297/2021-00 to J.P.F.-S.; 88887.951210/2024-00 to C.S.A.; 88887.687008/2022-00 to G.C.-C.; 88881.996985/2024-01 to E.R.Z.); Cure Alzheimer’s Fund (grant to H.Z.); European Union Joint Program – Neurodegenerative Disease Research (grants JPND2021-00694 to H.Z. and JPND2019-466-236 to K.B.); European Partnership on Metrology, co-financed by the European Union’s Horizon Europe research and innovation program and by the participating states (NEuroBioStand; grant 22HLT07 to H.Z.); the European Union’s Horizon Europe research and innovation program (grant 101053962 to H.Z.); the European Union’s Horizon 2020 research and innovation program under the Marie Skłodowska-Curie agreement (grant 860197 (MIRIADE) to H.Z.); Erling-Persson Family Foundation (grant to H.Z.); Familjen Rönströms Stiftelse (grant to H.Z.); Fonds de Recherche du Québec – Santé (Chercheur Boursier, grants 2020-VICO-279314 and 2024-VICO-356138; linked projects https://doi.org/10.69777/324345, https://doi.org/10.69777/356138 and https://doi.org/10.69777/312994 to P.R.-N.); Fundação de Amparo à Pesquisa do Estado do Rio Grande do Sul (FAPERGS; grants 21/2551-0000673-0 and 85053.824.30451.24062024 to E.R.Z.); Global Brain Health Institute, Alzheimer’s Association and Alzheimer’s Society (grant GBHI ALZ UK-23-971089 to C.S.A.); Hjärnfonden (grants FO2022-0270 to H.Z. and FO2017-0243 and ALZ2022-0006 to K.B.); Instituto Nacional Saúde Cerebral (grant 406020/2022-1 to E.R.Z.); Instituto Nacional Saúde em Excitotoxicidade Neuroproteção (grant 465671/2014-4 to E.R.Z.); Instituto Serrapilheira (grants Serra-1912-31365 and R-2401-47242 to E.R.Z.); Michael J. Fox Foundation (grant MJFF-023158 to E.R.Z.); Ministério da Saúde (grant 00030420240118-003490 to E.R.Z.); National Academy of Neuropsychology and Alzheimer’s Association (grant 22-92838 to E.R.Z.); National Institutes of Health (NIH; grant R01AG068398 to K.B.; AG027161, AG021155 and AG062715 to S.C.J.; R01AG075336 and R01AG073267 to T.A.P.); National Institute for Health and Care Research University College London Hospitals Biomedical Research Centre (grant to H.Z.); Olav Thon Foundation (grant to H.Z.); Stiftelsen för Gamla Tjänarinnor (grant to H.Z.); Swedish Alzheimer’s Foundation (grants AF-930351, AF-939721 and AF-968270 to K.B.); Swedish Research Council (grants 2023-00356, 2022-01018 and 2019-02397 to H.Z. and 2017-00915 and 2022-00732 to K.B.); Swedish State Support for Clinical Research (grant ALFGBG-71320 to H.Z.); the Swedish state under the agreement between the Swedish government and the county councils, the ALF-agreement (grants ALFGBG-715986 and ALFGBG-965240 to K.B.); UK Dementia Research Institute at UCL (grant UKDRI-1003 to H.Z.); and the Weston Brain Institute (grant to P.R.-N.).
Author information
Authors and Affiliations
Contributions
J.P.F.-S. and E.R.Z. conceived and designed the study. J.P.F.-S., G.P. and M.A.D.B. performed data analyses, with assistance from D.L.T. J.P.F.-S., G.P., N.R., P.R.-N. and E.R.Z. prepared the figures/tables and drafted the initial version of the manuscript. N.R., F.Z.L., J.T., A.C.M., S.S., J.S., S.G., A.L.B., S.C.J., T.A.P. and P.R.-N. were responsible for cohort procedures and imaging acquisition. W.S.B., I.P., G.T.-B., H.C.K., A.L.B., N.J.A., H.Z., K.B. and S.C.J. contributed to the acquisition of plasma and CSF biomarker data. B.B., P.C.L.F., M.A.D.B., D.T.L., F.Z.L., C.S.A., G.C.-C., W.V.B., D.O.S., L.P.S., M.V.L. and T.A.P. provided input on analyses and interpretation of the results. P.R.-N. and E.R.Z. supervised this work. P.R.-N. and S.C.J. acquired funding for the TRIAD and WRAP studies, respectively. All authors critically reviewed the manuscript for intellectual content and approved the final version.
Corresponding authors
Ethics declarations
Competing interests
W.V.B. served as a speaker for Novo Nordisk. J.T. has received personal fees from Alzheon and Neurotorium (outside the submitted work). S.G. is acting as scientist advisor to AmyriAD, Enigma USA, Eisai Canada, Lilly Canada, Novo Nordisk Canada and TauRx and is editor-in-chief of the Journal of Prevention of Alzheimerʼs Disease. L.P.S. has participated in educational activities organized by Aché, Apsen, Biogen, Knight, Libbs, Lilly, Novo Nordisk and Roche and has served on advisory boards for Biogen, Knight, Lilly, Novo Nordisk and Roche. G.T.-B. is an employee of Johnson & Johnson. H.C.K. was an employee of Johnson & Johnson at the time this work was performed. N.J.A. has given lectures in symposia sponsored by Lilly and Quanterix. H.Z. has served on scientific advisory boards and/or as a consultant for AbbVie, Acumen, Alector, Alzinova, ALZPath, Amylyx, Annexon, Apellis, Artery Therapeutics, AZTherapies, Cognito Therapeutics, CogRx, Denali, Eisai, LabCorp, Merry Life, Nervgen, Novo Nordisk, Optoceutics, Passage Bio, Pinteon Therapeutics, Prothena, Red Abbey Labs, reMYND, Roche, Samumed, Siemens Healthineers, Triplet Therapeutics and Wave; has given lectures in symposia sponsored by Alzecure, Biogen, Cellectricon, Fujirebio, Lilly, Novo Nordisk and Roche; and is a co-founder of Brain Biomarker Solutions in Gothenburg AB, which is a part of the GU Ventures Incubator Program (outside the submitted work). K.B. has served as a consultant and on advisory boards for AbbVie, AC Immune, ALZPath, AriBio, BioArctic, Biogen, Eisai, Lilly, Moleac Pte. Ltd, Neurimmune, Novartis, Ono Pharmaceutical, Prothena, Roche Diagnostics and Siemens Healthineers; has served on data monitoring committees for Julius Clinical and Novartis; has given lectures, produced educational materials and participated in educational programs for AC Immune, Biogen, Celdara Medical, Eisai and Roche Diagnostics; and is a co-founder of Brain Biomarker Solutions in Gothenburg AB, which is a part of the GU Ventures Incubator Program (outside the submitted work). S.C.J. serves on advisory boards for ALZPath, Enigma Biomedical, Alamar, Lilly and Merck. E.R.Z. has served on scientific advisory boards, as a consultant or as a speaker for Nintx, Novo Nordisk, Biogen, Lilly and Magdalena Biosciences; he is also a co-founder and minority shareholder of masima. The other authors declare no competing interests.
Peer review
Peer review information
Nature Neuroscience thanks Andre Altmann, Joseph Masdeu and the other, anonymous, reviewer(s) for their contribution to the peer review of this work.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data
Extended Data Fig. 1 Flowchart of study participant selection for the primary sample.
Abbreviations: AD = Alzheimer’s disease; ADAD = autosomal dominant Alzheimer’s disease; Aβ = amyloid-β; GFAP = glial fibrillary acidic protein; MCI = mild cognitive impairment; PET = positron emission tomography; QC = quality control; SNP = single nucleotide polymorphism; TSPO = translocator protein.
Extended Data Fig. 2 Sensitivity analyses testing the association of Aβ PET and plasma GFAP according to other summary measures of TSPO PET.
Scatter plots displaying the association between global Aβ PET burden and plasma GFAP levels according to MATSPO status, with error bands representing the 95% CI of regression lines (left). Continuous associations between global Aβ PET, TSPO PET, and plasma GFAP (right). (a) TSPO PET SUVR estimated from a composite brain region that includes the areas showing higher TSPO PET SUVR (T-value > 2) in AD Aβ+ compared to CU Aβ- (caudal anterior cingulate, entorhinal, fusiform, inferior temporal, isthmus cingulate, lateral orbitofrontal, lingual, medial orbitofrontal, middle temporal, paracentral, parahippocampal, pars opercularis, pars orbitalis, pars triangularis, pericalcarine, posterior cingulate, precuneus, rostral middle frontal, superior frontal; Extended Data Fig. 10). (b) TSPO PET SUVR estimated from a composite brain region that includes the areas within the top decile for absolute TSPO PET SUVR differences between AD Aβ+ and CU Aβ- individuals (paracentral, pars triangularis, posterior cingulate, and precuneus; Extended Data Fig. 10). The thresholds used to define MA+TSPO were calculated using the corresponding composite brain region. β-estimates and corresponding two-sided P-values were computed from linear regression models with standardized biomarker levels, adjusted for age, sex, and cognitive status. The interaction models also accounted for the main effects of the interaction terms. For regressions in MA+TSPO, exact P-values are (a) 2.47 × 10−6 and (b) 2.16 × 10−4. Abbreviations: AD = Alzheimer’s disease; Aβ = amyloid-β; CI = confidence interval; CU = cognitively unimpaired; GFAP = glial fibrillary acidic protein; MA = microglial activation; PET = positron emission tomography; SUVR = standardized uptake value ratio; TSPO = translocator protein.
Extended Data Fig. 3 Association of Aβ PET with plasma GFAP according to MATSPO status after FDR correction.
Region-wise linear regression T-map (FDR-corrected for multiple comparisons at P < 0.05) showing the association of Aβ PET SUVR with plasma GFAP levels across cortical areas in MA-TSPO (left) and MA+TSPO (right) individuals. Regression models with standardized biomarker levels were adjusted for age, sex, and cognitive status. Abbreviations: Aβ = amyloid-β; FDR = false discovery rate; GFAP = glial fibrillary acidic protein; MA = microglial activation; PET = positron emission tomography; SUVR = standardized uptake value ratio; TSPO = translocator protein.
Extended Data Fig. 4 Topography of TSPO PET impact on the association between Aβ PET and plasma GFAP after FDR correction.
Region-wise linear regression T-map (FDR-corrected for multiple comparisons at P < 0.05) showing the interaction of global Aβ PET and TSPO PET on plasma GFAP levels across cortical areas. Regression models with standardized biomarker levels were adjusted for age, sex, and cognitive status, as well as the main effects of the interaction terms. Abbreviations: Aβ = amyloid-β; FDR = false discovery rate; GFAP = glial fibrillary acidic protein; PET = positron emission tomography; TSPO = translocator protein.
Extended Data Fig. 5 Analysis testing the association of CSF sTREM2 and Aβ PET with plasma GFAP stratifying individuals by study cohort.
Scatter plots displaying the association between global Aβ PET burden and plasma GFAP levels according to MAsTREM2 status in the (a) WRAP (MA+sTREM2: exact P = 1.45 × 10−4) and (b) TRIAD (MA+sTREM2: exact P = 7.43 × 10−7) cohorts. Error bands represent the 95% CI of regression lines. β-estimates and two-sided P-values were computed from linear regression models using standardized fluid biomarker levels and Aβ PET in centiloids, with adjustments for age, sex, and cognitive status. The interaction models also accounted for the main effects of the interaction terms. Abbreviations: Aβ = amyloid-β; CI = confidence interval; CSF = cerebrospinal fluid; GFAP = glial fibrillary acidic protein; MA = microglial activation; PET = positron emission tomography; sTREM2 = soluble triggering receptor expressed on myeloid cells 2; TRIAD = Translational Biomarkers in Aging and Dementia; WRAP = Wisconsin Registry for Alzheimer’s Prevention.
Extended Data Fig. 6 Exploratory analyses testing whether sex affects the association between Aβ and glial reactivity.
(a) Association of Aβ PET with plasma GFAP in men (open dots) and women (solid dots) according to MATSPO status (primary sample). n = 101; 50 MA-TSPO (11 males, mean [SD] age of 71.4 [5.7] years) and 51 MA+TSPO (22 males, mean [SD] age of 72.4 [7.2] years). (b) Association of Aβ PET with plasma GFAP in men (open dots) and women (solid dots) according MAsTREM2 status (secondary sample). n = 251; 116 MA-sTREM2 (42 males, mean [SD] age of 65.9 [7.3] years) and 135 MA+sTREM2 (53 males, mean [SD] age of 68.3 [7.2] years). Dots and error bars represent β-estimates and 95% CI, respectively. Additional analysis using TSPO PET as an index of microglial activation demonstrated no significant interaction between sex and Aβ PET on plasma GFAP in either MA-TSPO (β = 0.250, 95% CI -0.340 to 0.841, P = 0.398) or MA+TSPO (β = 0.029, 95% CI -0.479 to 0.537, P = 0.909) individuals. When using CSF sTREM2 as an index of microglial activation, we observed a significant interaction between sex and Aβ PET on plasma GFAP in MA+sTREM2 (β = -0.019, 95% CI -0.030 to −0.007, P = 0.002) but not in MA-sTREM2 individuals (β = -0.003, 95% CI -0.014 to 0.009, P = 0.651). β-estimates and two-sided P-values were derived from linear regression models using standardized biomarker levels. All models were adjusted for age and cognitive status. The interaction models also accounted for the main effects of the interaction terms. For analysis involving CSF sTREM2, Aβ PET was assessed in centiloids, and cohort was also included as a variable for adjustment in regression models. Females were used as the reference group for the abovementioned statistics. Abbreviations: Aβ = amyloid-β; CI = confidence interval; CSF = cerebrospinal fluid; GFAP = glial fibrillary acidic protein; MA = microglial activation; PET = positron emission tomography; sTREM2 = soluble triggering receptor expressed on myeloid cells 2; TSPO = translocator protein.
Extended Data Fig. 7 Association of plasma GFAP with tau PET according to MATSPO status after FDR correction.
Region-wise linear regression T-map (FDR-corrected for multiple comparisons at P < 0.05) showing the association of plasma GFAP levels with tau PET SUVR across cortical areas in MA-TSPO (left) and MA+TSPO (right) individuals. Regression models with standardized biomarker levels were adjusted for age, sex, and cognitive status. Abbreviations: FDR = false discovery rate; GFAP = glial fibrillary acidic protein; MA = microglial activation; PET = positron emission tomography; SUVR = standardized uptake value ratio; TSPO = translocator protein.
Extended Data Fig. 8 Exploratory analyses assessing tau PET across Braak-like regions to test the associations between Aβ, tau, glial reactivity, and cognition.
(a) Representation of Braak I-II ROIs (transentorhinal, entorhinal, and hippocampus) overlaid on a structural MRI template. (b) Structural equation modeling testing the associations between global Aβ PET, plasma GFAP, plasma p-tau217, Braak I-II tau PET, and MMSE score. The model in MA-TSPO individuals (left) fit the data poorly (RMSEA = 0.272, SRMR = 0.071, CFI = 0.898), while the model in MA+TSPO individuals (right) fit the data well (RMSEA = 0.000, SRMR = 0.012, CFI = 1.000). (c) Representation of Braak III-IV ROIs (amygdala, parahippocampal gyrus, fusiform gyrus, lingual gyrus, insula, inferior temporal, lateral temporal, posterior cingulate, and inferior parietal) overlaid on a structural MRI template. (d) Structural equation modeling testing the associations between global Aβ PET, plasma GFAP, plasma p-tau217, Braak III-IV tau PET, and MMSE score. The model in MA-TSPO individuals (left) fit the data poorly (RMSEA = 0.254, SRMR = 0.068, CFI = 0.885), while the model in MA+TSPO individuals (right) fit the data well (RMSEA = 0.000, SRMR = 0.014, CFI = 1.000). (e) Representation of Braak V-VI ROIs (orbitofrontal, superior temporal, inferior frontal, cuneus, anterior cingulate, supramarginal gyrus, lateral occipital, precuneus, superior parietal, superior frontal, rostro medial frontal, paracentral, postcentral, precentral, and pericalcarine) overlaid on a structural MRI template. (f) Structural equation modeling testing the associations between global Aβ PET, plasma GFAP, plasma p-tau217, Braak V-VI tau PET, and MMSE score. The model in MA-TSPO individuals (left) fit the data poorly (RMSEA = 0.249, SRMR = 0.068, CFI = 0.883), while the model in MA+TSPO individuals (right) fit the data well (RMSEA = 0.000, SRMR = 0.013, CFI = 1.000). Associations were adjusted for age, sex, and (if involving cognition) years of education. Solid lines with standardized β-estimates and 95% CI represent significant associations, while dashed lines represent non-significant effects. Abbreviations: Aβ = amyloid-β; CFI = comparative fit index; CI = confidence interval; GFAP = glial fibrillary acidic protein; MA = microglial activation; MMSE = Mini-Mental State Examination; MRI = magnetic resonance imaging; PET = positron emission tomography; p-tau217 = phosphorylated tau at threonine 217; RMSEA = root mean squared error of approximation; ROI = region of interest; SRMR = standardized root mean square residual; TSPO = translocator protein.
Extended Data Fig. 9 Exploratory analyses evaluating the associations between Aβ, tau, glial reactivity, and the memory cognitive domain.
Structural equation modeling testing the associations between global Aβ PET, plasma GFAP, plasma p-tau217, temporal meta-ROI tau PET, and the memory composite score. (a) The model in MA-TSPO individuals fit the data poorly (RMSEA = 0.260, SRMR = 0.070, CFI = 0.912). (b) The model in MA+TSPO individuals fit the data well (RMSEA = 0.000, SRMR = 0.015, CFI = 1.000). Associations were adjusted for age, sex, and (if involving memory function) years of education. Solid lines with standardized β-estimates and 95% CI represent significant associations, while dashed lines represent non-significant effects. The memory composite score was calculated by averaging the z-scores from the Rey Auditory Verbal Learning Test immediate recall, Rey Auditory Verbal Learning Test delayed recall, Logical Memory immediate recall, and Logical Memory delayed recall. Z-scores used in the memory composite were calculated based on the mean and SD of raw test scores from CU individuals. Analyses were performed in a subset of 90 individuals (47 MA-TSPO and 43 MA+TSPO). Among the participants assessed in the main structural equation modeling analyses (Fig. 2d,e), three participants were not included because they did not have complete neuropsychological testing for calculating the memory composite score. Abbreviations: Aβ = amyloid-β; CFI = comparative fit index; CI = confidence interval; CU = cognitively unimpaired; GFAP = glial fibrillary acidic protein; MA = microglial activation; PET = positron emission tomography; p-tau217 = phosphorylated tau at threonine 217; RMSEA = root mean squared error of approximation; ROI = region of interest; SD = standard deviation; SRMR = standardized root mean square residual; TSPO = translocator protein.
Extended Data Fig. 10 [11C]PBR28 TSPO PET signature in AD.
(a) Region-wise TSPO PET SUVR mean maps in CU Aβ- individuals (top) and AD Aβ+ dementia patients (middle), along with the region-wise differences of TSPO PET SUVR mean maps (bottom), computed as mean SUVR values in AD Aβ+ minus CU Aβ-. Of note, the bottom panel displays only regions with significantly higher TSPO PET signal in AD Aβ+ compared to CU Aβ- (T-value > 2). (b) Bars display the mean and SD of TSPO PET SUVR across groups in the posterior cingulate, which is the region showing the greatest difference between AD Aβ+ and CU Aβ-. (c) Bars display the mean and SD of TSPO PET SUVR across groups in a composite brain region comprising the areas that show higher TSPO PET SUVR values (T-value > 2) in AD Aβ+ compared to CU Aβ-. This T-derived composite brain region was estimated from the following brain regions: caudal anterior cingulate, entorhinal, fusiform, inferior temporal, isthmus cingulate, lateral orbitofrontal, lingual, medial orbitofrontal, middle temporal, paracentral, parahippocampal, pars opercularis, pars orbitalis, pars triangularis, pericalcarine, posterior cingulate, precuneus, rostral middle frontal, superior frontal. (d) Bars display the mean and SD of TSPO PET SUVR across groups in a composite brain region comprising the areas within the top decile for absolute TSPO PET SUVR differences between AD Aβ+ and CU Aβ-. This difference-derived composite brain region was estimated from the following brain regions: paracentral, pars triangularis, posterior cingulate, and precuneus. Two-sided P-values correspond to group comparison (AD Aβ+ vs. CU Aβ-) from regression analyses accounting for age and sex. Exact P-values are (b) 1.99 × 10−5, (c) 6.86 × 10−5, (d) 4.51 × 10−5. n = 62; 49 CU Aβ- (11 males, mean [SD] age of 71.5 [5.7] years) and 13 AD Aβ + (6 males, mean [SD] age of 68.9 [10.3] years). Abbreviations: AD = Alzheimer’s disease; Aβ = amyloid-β; CU = cognitively unimpaired; PET = positron emission tomography; SD = standard deviation; SUVR = standardized uptake value ratio; TSPO = translocator protein.
Supplementary information
Supplementary Information
Supplementary Tables 1−12.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
About this article
Cite this article
Ferrari-Souza, J.P., Povala, G., Rahmouni, N. et al. Microglia modulate Aβ-dependent astrocyte reactivity in Alzheimer’s disease. Nat Neurosci 29, 81–87 (2026). https://doi.org/10.1038/s41593-025-02103-0
Received:
Accepted:
Published:
Version of record:
Issue date:
DOI: https://doi.org/10.1038/s41593-025-02103-0
