Background

The number of persons affected by Alzheimer’s disease (AD) across its pathological continuum was recently estimated to be as high as 416 million [1]. From this global estimate, 3/4 of individuals were classified as preclinical AD, characterized by the presence of beta-amyloid (Aβ) plaques but absence of clinical symptoms [1]. In addition to the hallmark pathologies, i.e., Aβ plaques and neurofibrillary tangles, inflammation in the central nervous system (CNS) is recognized to have an important, partly independent, role in Alzheimer’s continuum [2]. In the brain, inflammation is mainly mediated by microglia and astrocytes [3], and in AD, compiling evidence suggests that increased microglial and astrocytic reactivity could be present already during early, possibly protective processes [4,5,6,7].

Apolipoprotein E (APOE) ε4 allele is the strongest genetic risk factor of sporadic AD; it increases the risk of disease and decreases the age of onset when compared with the most common APOE ε3 or the protective APOE ε2 alleles [8]. APOE ε4 gene dose-related increase in brain Aβ load is present already in cognitively normal individuals [9,10,11], and it has been suggested to be caused by impaired degradation and clearance of Aβ, a task which is performed by glial cells and affected by APOE isoforms [12, 13]. In neuropathological studies, APOE ε4 has been seen to associate with increased microglial number in the brains of individuals with AD [14] and higher microglial cell reactivity around Aβ plaques in a mouse model of Aβ deposition and human APOE alleles [15]. Still, clinical investigations concerning the relationship between early ongoing neuroinflammatory processes and Aβ in cognitively unimpaired APOE ε4 carriers are scarce.

Investigation of regional neuroinflammation in AD in vivo has been enabled by PET imaging and specific ligands such as 11C-PK11195 that target 18-kDa translocator protein (TSPO) as a proxy for microglial reactivity. TSPO is present in the outer mitochondrial membranes of microglia and elevated in the brain in relation to injuries or pathology [16]. In humans, increased TSPO ligand-binding has recently been suggested to represent changes in cell density rather than protein overexpression [17], and to be mostly covered by microglia, and to a lesser extent astrocytes and endothelial cells [18, 19]. Previous studies using TSPO PET imaging have shown increased regional ligand-binding in patients with AD [20,21,22,23], mild cognitive impairment [4, 24, 25], and some also in Aβ-positive compared with Aβ-negative controls [7, 26]. However, results are partly inconclusive since also minor or no differences between diagnostic groups have been reported [27,28,29].

In addition to imaging, more easily accessible biomarkers for AD pathology measured in blood have become available recently thanks to the development of more sensitive methods [30]. Unfortunately, since proteins expressed by microglia in the CNS are also present in peripheral macrophages, the development of blood-based assays targeting microgliosis is demanding, and interpretation of measurements from blood is complicated [31]. However, one interesting fluid biomarker for glial reactivity, glial fibrillary acidic protein (GFAP, a marker of reactive astrocytosis), is measurable from blood using the single molecule array (Simoa) technology and has been recently shown to be associated with Aβ deposition and increased already in early stages of AD [32,33,34].

Based on previous literature suggesting an early involvement of neuroinflammation during the AD continuum, we hypothesized that in vivo TSPO binding and plasma GFAP concentrations would be elevated in cognitively normal APOE ε4 homozygotes or APOE ε4 heterozygotes, representing different genetically increased risk for Aβ accumulation and sporadic AD, compared to age-matched non-carriers. To test this hypothesis, we evaluated differences in (i) regional Aβ and TSPO PET and (ii) plasma Aβ42/40 and GFAP concentrations and their associations primarily among cognitively normal APOE ε4 homozygotes, heterozygotes, and non-carriers. In addition, we performed secondary analyses between Aβ-positive (representing Alzheimer’s pathological change or preclinical AD [35]) and Aβ-negative individuals and investigated the association between imaging and fluid biomarkers of neuroinflammation and Aβ deposition and markers of disease progression (cognitive performance and volumetric brain changes) in our cohort comprised by cognitively unimpaired participants enriched with APOE ε4 carriers.

Methods

Study design and participants

The study design is illustrated in Fig. 1 and detailed study protocol including power calculations has been previously published [36]. Briefly, participants in this cross-sectional, observational study were recruited in collaboration with the local Auria biobank (Turku, Finland). Set inclusion criteria were 60–75 years of age and CERAD total score > 62 points at screening. Main exclusion criteria were dementia or cognitive impairment; other severe neurological or psychiatric disease; diabetes; chronic inflammatory condition; and contraindication for MRI or PET. In total, sixty-three cognitively normal individuals were recruited. However, one individual from the APOE ε4 heterozygotes and of the non-carriers discontinued the study (two experienced claustrophobia inside the scanner; one had an unexpected and excluding finding in MRI); thus, sixty participants from three groups were included in the analysis (APOE ε4 homozygotes: n = 19, median MMSE = 28 (interquartile range (IQR) 27-29); APOE ε4 heterozygotes: n = 21, median MMSE = 29 (IQR 28-30), non-carriers: n = 20,  median MMSE = 29 (IQR 27-30)).

Fig. 1
figure 1

Study flowchart. Altogether 60 individuals were recruited based on their APOE ε4 gene dose (APOE ε4/ε4, n = 19, APOE ε4/ε3, n = 21, APOE ε3/ε3 n = 21). All underwent positron emission tomography (PET) imaging targeting Aβ using 11C-PiB, 18-kDa translocator protein (TSPO) as a proxy for glial reactivity using.11C-PK11195, magnetic resonance imaging (MRI), and cognitive testing. A blood sample was drawn for laboratory measurements, including plasma markers of Aβ pathology (Aβ1-40 and Aβ1-42) and reactive astrocytosis (glial fibrillary acidic protein, GFAP)

Full size image

Brain imaging measurements

Structural T1-weighted brain MRI scan was performed on either a Philips Ingenuity 3.0 T TF PET/MRI (n = 38; Philips Healthcare, Amsterdam, the Netherlands) or a Philips Ingenia 3.0 T (n = 22; Philips Healthcare, Amsterdam, the Netherlands). PET scans were acquired on an ECAT high-resolution research tomograph (HRRT, Siemens Medical Solutions, Knoxville, TN). For amyloid imaging, 11C-PiB scans (n = 60) were acquired 40 to 90 min post injection (mean injected dose 497 (standard deviation (SD) 30) MBq), and for TSPO imaging, dynamic 11C-PK11195 scans (n = 57) were acquired for 60 min post injection (mean injected dose 494 (SD 21) MBq). All images were reconstructed with 3D ordinary Poisson ordered subset expectation maximization algorithm (OP-OSEM3D), and list mode data was histogrammed into 8 (6 × 5 + 2 × 10 min, 11C-PiB) and 17 (2 × 15; 3 × 30; 3 × 60; 7 × 300; 2 × 600 s, 11C-PK11195) time frames.

Brain image analysis

PET and MR image preprocessing and analysis was performed using an automated pipeline at Turku PET Centre [37] which executed the PET data frame by frame realignment, PET-MRI co-registration, FreeSurfer ROI parcellation, and PET data kinetic modeling. Regional and voxel level 11C-PiB-binding was quantified as standardized uptake value ratios (SUVR) calculated for 60 to 90 min post injection, using the cerebellar cortex as reference region. Regional 11C-PK11195-binding was quantified as distribution volume ratios (DVR) within 20–60 min post injection using a reference tissue input Logan’s method with pseudo-reference region extracted using supervised clustering algorithm [38, 39]. Voxel-level kinetic modeling for 11C-PK11195 was carried out using basis function implementation of simplified reference tissue model with respect to the aforementioned clustered pseudo-reference region and with 300 basis functions calculated within the Θ3 parameter limits 0.06 ≤ Θ3 ≤ 0.6 [40]. Partial volume effect (PVE)-corrected data was used for all 11C-PK11195 analysis in order to minimize the effect TSPO uptake in sinuses to cortical regions. PVE correction was carried out using PETPVE12 toolbox [41] in both region-of-interest (ROI, geometric transfer matrix method) and voxel-level (Muller-Gartner method) data. ROI-level analysis for both 11C-PiB and 11C-PK11195 data was performed in a priori defined regions known for early Aβ deposition (prefrontal cortex, parietal cortex, anterior cingulum, posterior cingulum, precuneus, lateral temporal cortex, and a volume weighted composite containing all the regions).41 For 11C-PK11195, additional volume-weighted ROIs for transentorhinal (Braak I–II) and limbic composite (Braak III–IV) regions [42] were analyzed to investigate TSPO-binding in regions associated with early tau deposition. Details of the combined FreeSurfer regions are previously published [36]. Spatially normalized parametric SUVR and BPND images in MNI152 space were smoothed using Gaussian 8 mm FWHM filter and used for all voxel-wise statistical analysis. For all figures, BPND were transformed to DVRs for clarity, using the formula: DVR = BPND + 1. Aβ positivity was defined as cortical composite 11C-PiB SUVR > 1.5 [43, 44].

Total hippocampal volume and total entorhinal area volume (left + right, ml) normalized for intracranial volume, age, and sex were obtained from the T1-weighted MR images using an automatic cNeuro image analysis tool (Combinostics Oy, Tampere, Finland) [45, 46]. Since two different instruments were used for acquiring MRI images the used scanner was added as a covariate in all analyses including hippocampal or entorhinal volumes.

Cognitive testing

All participants completed CERAD cognitive test battery at screening, as well as more extensive neuropsychological testing during one of the study visits as previously described [36]. CERAD total score, mini-mental state examination (MMSE) score, and Alzheimer’s Prevention Initiative Preclinical Composite Cognitive test (APCC) score were used to investigate the association between both imaging and blood biomarkers and cognitive performance.

Blood biomarker measurements

All plasma biomarker measurements were performed in the Clinical Neurochemistry Laboratory, Mölndal, Sweden. Plasma Aβ1-40 and Aβ1-42 concentrations were measured using an in-house immunoprecipitation mass spectrometry method (IP-MS) described in detail elsewhere [47, 48]. Briefly, Aβ peptides were immunoprecipitated from 250 μl of sample using 4G8 and 6E10 anti-Aβ antibodies (BioLegend) coupled to Dynabeads™ M-280 Sheep Anti-Mouse IgG magnetic beads and a KingFisher Flex instrument (Thermo Fisher Scientific) and further analyzed by liquid chromatography-tandem mass spectrometry (LC–MS/MS). Recombinant Aβ1-40 and Aβ1-42 peptides were used as calibrators, and heavy labeled peptides were added to both samples and calibrators for internal standards.

Plasma GFAP concentration was measured using the Single molecule array (Simoa) platform, a HD-X analyzer (Quanterix, Billerica, MA), and a commercial GFAP discovery kit (Quanterix, #102,336) following the instructions provided by the manufacturer. Two internal quality control (QC) samples with mean concentrations of 100 pg/ml and 608 pg/ml were measured in the beginning and after samples in both plates. Calibrators and QC samples were measured as duplicates and samples as singlicates. The intra-assay precision (variation within run, CVr (%)) and inter-assay precision (variation between runs, CVrw (%)) were < 5% and < 15%, respectively.

Statistical analysis

All data following normal distribution are presented as mean (SD), otherwise as median (IQR). Normality of the data was established visually and from the residuals. Missing data points for each variable are presented in eTable 1 in Additional file 1. For continuous variables, differences in group demographics and in regional 11C-PiB and 11C-PK11195-binding between the three APOE ε4 gene doses were tested using one-way ANOVA with Tukey’s honest significance test or Kruskal–Wallis test with Dunn's method for multiple comparisons depending on the distribution of data. χ2 test was used for testing categorical variables. Associations between regional PET data and fluid biomarker concentrations were evaluated using Spearman’s rank correlation. Differences in 11C-PK11195-binding between Aβ-positive and Aβ-negative individuals were first tested with Student's t-test. We also wanted to see if regional 11C-PK11195-binding differed between Aβ-positive and Aβ-negative individuals accounting for APOE ε4 status, so we additionally tested the effect of APOE ε4 gene dose, amyloid positivity, and their interaction (ε4 gene dose ˟ amyloid positivity) on 11C-PK11195 in a priori defined ROIs with linear regression models. If an interaction term with P < 0.1 was found, a post hoc comparison of all groups was performed to explore the nature of the interaction.

Voxel-level differences in 11C-PIB and 11C-PK11195-binding between APOE ε4 gene doses were evaluated using one-way ANOVA, followed by post hoc pairwise comparisons in Statistical Parametric Mapping (SPM12 v12; Wellcome Trust Centre for Neuroimaging, London, UK) running on MATLAB, whereas voxel-level 11C-PIB SUVRs and 11C-PK11195 BPND Spearman’s rank correlation coefficients were calculated using built-in MATLAB functions. False discovery rate-corrected cluster level threshold was set at P < 0.05. Differences in blood biomarker concentrations between APOE ε4 gene doses were analyzed using Kruskal–Wallis test with Dunn's method for multiple comparisons.

Finally, we used multivariable linear regression models adjusted for age, sex, and education (and MRI scanner for models explaining hippocampal or cortical volumes) to test how well PET and fluid biomarkers of Aβ and glial reactivity could explain different cognitive and structural variables that could be interpreted as markers of disease progression. For comparison, standardized βs were calculated and presented in figures.

All statistical analyses were performed using SAS JMP Pro v.15.1.0 (SAS institute, Cary, NC) and visualizations using GraphPad Prism version 9.0.1 (GraphPad, San Diego, California, USA). A P-value < 0.05 (2-tailed) was considered statistically significant in all analysis, except for interaction effects, where stratified analysis was run already if P (interaction) < 0.1.

Results

Demographics

Demographics and descriptive data for the APOE ε4 gene dose groups are presented in Table 1. No statistically significant differences in age, sex, education, body mass index (BMI), or CERAD total score were present between the APOE ε4 gene dose groups (P > 0.37 for all). APOE ε4 heterozygotes had significantly higher MMSE than homozygotes (P = 0.036). Age had positive correlation with 11C-PiB cortical composite SUVRs in APOE ε4 homozygotes (Rho = 0.63, P = 0.0039), but not in heterozygotes, non-carriers, or the whole cohort (P > 0.19 for all). There was no correlation between age and composite cortical 11C-PK11195 DVRs (P > 0.42 for all), plasma GFAP (P > 0.17 for all), or plasma Aβ1-42/1–40 (P = 0.22 for all). Using a cut-off value of cortical composite 11C-PiB SUVR > 1.5, 84% (n = 16) of the APOE ε4 homozygotes, 48% (n = 10) of the heterozygotes, and 40.0% (n = 8) of non-carriers in our cohort were classified as amyloid positive.

Table 1 Demographics and descriptive data for cognitively unimpaired APOE ε4 homozygotes, heterozygotes, and non-carriers included in the study
Full size table

For secondary analyses, we also stratified the cohort based on Aβ-positivity (composite 11C-PiB SUVR > 1.5). Demographics are presented in eTable 2 in Additional file 1. Significant differences between Aβ-positive and Aβ-negative individuals were found in education level (P = 0.046), CERAD total score (P = 0.0034), and MMSE score (P = 0.0074).

Fibrillar Aβ deposition distinguishes cognitively unimpaired APOE ε4 homozygotes from heterozygotes and non-carriers

APOE ε4 gene dose-related differences in fibrillar amyloid load were visually detectable from mean 11C-PiB distribution maps in regions typical for early amyloid deposition (Fig. 2A). ROI-level analysis verified the findings, revealing significant differences in 11C-PiB-binding between gene doses in all evaluated regions (P < 0.016 for all regions, Kruskal–Wallis test). After post hoc comparison of all groups, 11C-PiB-binding was significantly higher in APOE ε4 homozygotes compared with heterozygotes in the prefrontal cortex (P = 0.023) and in all evaluated regions when compared with non-carriers (P < 0.017 for all regions) (Fig. 2B, Table 2).

Fig. 2
figure 2

Mean 11C-PiB and 11C-PK11195 distribution maps and regional ligand-binding in cognitively unimpaired volunteers stratified by APOE ε4 gene dose. A Mean 11C-PiB standardized uptake value ratio (SUVR) distribution maps and B region-of-interest analysis showed significantly higher uptake in APOE ε4 homozygotes compared with non-carriers in all evaluated regions and compared with heterozygotes in anterior cingulate and prefrontal cortex (Kruskal–Wallis test with Dunn's method for multiple comparisons). C Mean11C-PK11195 standardized distribution volume ratio (DVR) maps showed regional differences in tracer-binding, D but no significant differences between the APOE ε4 gene dose groups (one-way ANOVA with Tukey’s honest significance test for multiple comparisons). ACIN, anterior cingulate; HIP, hippocampus; LTC, lateral temporal cortex; PFC, prefrontal cortex, PARC, parietal cortex; PCIN, posterior cingulate cortex; PREC, precuneus. * P < 0.05; ** P < 0.01; *** P < 0.001

Full size image
Table 2 Regional 11C-PiB SUVR values for APOE ε4 homozygotes, heterozygotes, and non-carriers
Full size table
Table 3 Regional 11C-PK11195 DVR values for APOE ε4 homozygotes, heterozygotes, and non-carriers
Full size table

Voxel-level comparisons verified the findings showing significantly higher 11C-PiB-binding in the prefrontal cortex, precuneus, and lateral temporal cortex of the APOE ε4 homozygotes compared with non-carriers (eFigure 1A in Additional file 1). Weaker effects with similar spatial distribution were seen in APOE ε4 homozygotes compared with heterozygotes (eFigure 1B in Additional file 1). No significant clusters were found when comparing heterozygotes and non-carriers.

Regional 11C-PK11195 binding does not differ between APOE ε4 gene doses

Mean 11C-PK11195 DVR distribution maps for each APOE ε4 gene dose are shown in Fig. 2C. Even though amyloid imaging results showed presence of Aβ pathology in the APOE ε4 homozygote group, we did not observe hypothesized increase in TSPO-binding in the same brain regions as a response to presence of Aβ (P > 0.08 for all, one-way ANOVA, Fig. 2D, Table 3) measured by 11C-PK11195 PET. In agreement with the ROI-level analyses, no significant clusters were detected in voxel-level comparisons between the APOE ε4 gene dose groups.

For secondary analysis, we stratified the cohort based on Aβ-positivity (11C-PiB SUVR > 1.5). Again, we found no regional differences in TSPO-binding between Aβ-positive and -negative individuals (P > 0.21 for all regions, Student’s t test, eTable 3 in Additional file 1). To further evaluate the possible effects of amyloid status on TSPO-binding in different APOE ε4 gene doses, we analyzed also the interaction of Aβ-positivity × APOE ε4 gene dose for predicting regional TSPO-binding. Whereas amyloid status (accounted for APOE ε4 gene dose) did not have a significant effect on TSPO-binding (P > 0.28 for all regions), the interaction term approached statistical significance in the cortical composite (P = 0.089), lateral temporal cortex (P = 0.063), transentorhinal (Braak I–II, P = 0.052), and limbic (Braak III–IV, P = 0.019) ROIs (Table 4). Exploratory analysis in those regions showed that median TSPO-binding was higher in Aβ-positive APOE ε4 carriers than in non-carriers and, interestingly, also in Aβ-negative non-carriers compared with Aβ-positive non-carriers. However, these differences did not reach statistical significance after post hoc comparison between all six groups (eFigure 2 in Additional file 1). Since we had two highly 11C-PiB-positive non-carriers (with cortical composite SUVRs of 3.4 and 2.2) included in our cohort, we also verified that the found interactions were still present for LTC (P = 0.079), Braak I–II (P = 0.063), and Braak III–IV, (P = 0.032) when these two individuals were excluded.

Table 4 Test effects from multivariate linear regression models explaining regional 11C-PK11195 binding
Full size table

Regional 11C-PK11195-binding correlates with Aβ load measured by 11C-PiB only in cognitively unimpaired APOE ε4 carriers

No significant correlation between 11C-PiB and PVE-corrected 11C-PK11195-binding was present in any of the a priori chosen ROIs in the total study population (Rho =  − 0.11–0.13, P > 0.35 for all) or Aβ-positive individuals (Rho =  − 0.13–0.21, P > 0.23 for all). However, when stratified by APOE ε4 gene dose, higher 11C-PiB-binding in the cortical composite ROI associated with higher TSPO-binding in the cortical (Rho = 0.47, P = 0.043) and limbic (Rho = 0.49, P = 0.032) composite ROIs in APOE ε4 homozygotes (Fig. 3A), but not in APOE ε4 heterozygotes. In contrast, a negative correlation between 11C-PiB-binding in the cortical composite ROI was observed for non-carriers in the transentorhinal (Rho =  − 0.63, P = 0.0065) and limbic ROIs (Rho = -0.68, P = 0.0025).

Fig. 3
figure 3

Regional association between amyloid PET and TSPO PET in cognitively unimpaired volunteers stratified by APOE ε4 gene dose. A Scatterplots from ROI level data showed positive correlation (Spearman’s rank correlation) for APOE ε4 carriers in cortical and Braak III–IV composite regions, whereas negative associations were present for non-carriers. B Most significant voxel-wise positive correlations between 11C-PiB and 11C-PK11195-binding were present in the APOE ε4/ε3 heterozygotes (yellow scale) and in APOE ε4 homozygotes (red scale), whereas only sparse significant voxels were seen in non-carriers (green scale). Partial volume corrected 11C-PiB SUVR and.11C-PK11195 BPND images smoothed using Gaussian 8 mm FWHM filter were used for all voxel-wise analysis. False Discovery Rate corrected cluster level threshold was set at P < 0.05

Full size image

Voxel-wise analysis (not limited to specific predefined regions) did reveal clusters with significant positive correlation between 11C-PiB- and 11C-PK11195-binding in both APOE ε4 homozygotes (Fig. 3B, red scale) and heterozygotes (Fig. 3B, yellow scale), whereas only small spare clusters were found in non-carriers (Fig. 3B, green scale). However, many of the clusters were located outside our primary regions of interest (chosen based on presence of early amyloid or tau pathology), such as in the white matter and the paracentral lobule.

Differences in plasma GFAP concentrations between APOE ε4 gene doses

Absolute plasma GFAP concentrations were higher in APOE ε4 homozygotes (186 pg/ml, 124–269) compared with APOE ε4 heterozygotes (150 pg/ml, 104–170) and non-carriers (128 pg/ml, 105–147), (P = 0.077, Kruskal–Wallis test, Fig. 4A). A trend towards positive association between plasma GFAP and cortical 11C-PiB-binding was present in the whole cohort (Rho = 0.23, P = 0.085), and a significant positive correlation was observed in Aβ-positive individuals (Rho = 0.35, P = 0.040). No association between plasma GFAP and cortical TSPO-binding was present in the whole cohort (Rho = 0.069, P = 0.61) or in Aβ-positive individuals (Rho = 0.13, P = 0.47, Fig. 4A).

Fig. 4
figure 4

Plasma GFAP and plasma Aβ1-42/1–40 concentrations in cognitively unimpaired volunteers stratified by APOE ε4 gene dose. Differences in biomarker concentrations between APOE ε4 gene doses, correlations with cortical composite amyloid PET standardized uptake value ratios (SUVRs), and TSPO PET distribution volume ratios (DVRs) for A plasma glial fibrillary acidic protein (GFAP) and B plasma Aβ1-42/1–40. Differences between groups were tested with Kruskal–Wallis with Dunn's method for multiple comparisons, and correlations with Spearman’s rank correlation

Full size image

Differences in soluble Aβ concentrations estimated by plasma Aβ1-42/1–40 between APOE ε4 gene doses

Despite the clear differences in regional Aβ PET, plasma Aβ1-42/1–40 was not significantly different between APOE ε4 homozygotes (0.077, 0.059–0.098), APOE ε4 heterozygotes (0.087, 0.068–0.11)), and non-carriers (0.086, 0.076–0.10) (P = 0.50, Kruskal–Wallis test, Fig. 4B). In our cohort, plasma Aβ1-42/1–40 did not correlate with either cortical composite amyloid load measured by 11C-PiB PET (Rho = -0.18, P = 0.18), or with cortical composite TSPO-binding measured by 11C-PK11195-binding (Rho = 0.066, P = 0.64; Fig. 4B).

Amyloid and neuroinflammatory biomarker associations with markers for disease progression

Finally, we wanted to compare how the different biomarkers associate with cognitive (MMSE, CERAD total score, APCC score) and structural variables (total hippocampal and entorhinal volume) that could be seen as proxies for future disease progression (Fig. 5). In the whole cognitively unimpaired cohort, higher cortical composite 11C-PiB-binding (βstd =  − 0.29 (95% CI − 0.52 to − 0.067), P = 0.012), but not higher 11C-PK11195-binding (βstd =  − 0.041 (− 0.26 to 0.18), P = 0.70), was associated with lower APCC scores. However, higher cortical 11C-PK11195-binding was associated both with lower hippocampal volume (βstd =  − 0.35 (− 0.61 to −0.10), P = 0.0066) and entorhinal volume (βstd =  − 0.47 (− 0.72 to − 0.22), P = 0.0004). Higher plasma GFAP concentration was associated with both lower hippocampal volume (βstd =  − 0.34 (− 0.60 to − 0.084), P = 0.011), MMSE (βstd =  − 0.35 (− 0.59 to − 1.10), P = 0.0060), and APCC scores (βstd =  − 0.29 (− 0.51 to − 0.070), P = 0.011). Plasma Aβ1-42/1–40 was not associated with any of the cognitive or volumetric variables (P > 0.18 for all analysis). All models were adjusted for age, sex, and education and used MRI scanner for structural variables.

Fig. 5
figure 5

Comparison of PET and blood biomarkers of Aβ deposition and glial reactivity and their association with cognitive performance and brain structure. Higher cortical composite 11C-PiB-binding (A) but not plasma Aβ1-42/1–40 (B) was associated with lower entorhinal volumes and lower scores in the Alzheimer’s Prevention Initiatives preclinical cognitive composite (APCC) battery. C Cortical composite 11C-PK11195 PET was associated only with lower hippocampal and entorhinal volume, whereas D elevated plasma GFAP levels were associated with lower APCC and Mini Mental State Examination (MMSE) scores. The results are shown as standardized estimates (betas) derived from liner models adjusted for age, sex, and education (and used MRI scanner for structural variables). CERADTS, Consortium to Establish a Registry for Alzheimer’s Disease total score; VT, total volume

Full size image

Discussion

Microglial and recently also astrocytic reactivity has been suggested to be early events during the long AD continuum [3, 4, 6]. In this study, we used TSPO PET imaging and plasma GFAP to investigate early neuroinflammatory differences in cognitively unimpaired individuals stratified by their APOE ε4 gene dose and thus risk for AD. Despite the verified presence of Aβ pathology in APOE ε4 carriers, we found no significant regional differences in TSPO-binding between APOE ε4 gene doses or cognitively normal Aβ-positive subjects (presenting Alzheimer’s pathological change or preclinical AD) compared with Aβ-negative individuals. However, in the whole cognitively unimpaired sample, higher levels of both neuroinflammatory markers were associated with lower hippocampal volume, used as a proxy for future disease progression.

To our knowledge, this study is the first one to compare TSPO PET findings in cognitively normal at-risk individuals grouped by their APOE ε4 gene dose. In previous human in vivo studies, most robust increases in TSPO-binding have been found in Alzheimer’s dementia in comparison to controls [20,21,22, 49], but also in Aβ-positive MCI [4, 24, 25] and Aβ-positive controls [7, 26]. Thus, we had hypothesized that early Aβ related neuroinflammatory changes should be present also in either cognitively normal APOE ε4 homozygotes or APOE ε4 heterozygotes, both representing a genetically increased risk for Aβ accumulation and sporadic AD. Here, despite clearly increased fibrillar Aβ load, we did not find associated increased TSPO binding in APOE ε4 homozygotes. In addition, our secondary analyses were not able to replicate the reported increased TSPO-binding in Aβ-positive “at-risk” individuals using 11C-PK11195 PET, even with a larger sample size compared with previous reports. Our study included approximately 20 participants in each APOE ε4 gene dose group and 34 cognitively unimpaired Aβ-positive individuals, whereas the previous studies included only six [7] or seven [26] Aβ-positive controls. However, it should be remembered that our cohort, and especially its Aβ-positive participants, was highly enriched for APOE ε4 carriers. Neuritic Aβ plaques are known to be surrounded by reactive microglia in AD; however, APOE is also directly liked to immune responses [50, 51], and APOE ε4 genotype has been suggested to attenuate the response of “amyloid related microglia” towards AD pathology [52]. Here, using TSPO imaging, we are unable to differentiate microglial phenotypes and cannot exclude a direct down modulation of microglial response in APOE ε4 carriers that could explain the lack of hypothesized increased TSPO-binding in regions with abundant Aβ pathology.

In addition to TSPO PET, we also performed 11C-PiB PET for all participants to verify the presence of Aβ pathology in our cognitively unimpaired sample. As expected, Aβ deposition in the brain increased in an APOE ε4 gene dose dependent fashion; significantly elevated cortical 11C-PiB retention was present in APOE ε4 homozygotes compared with both heterozygotes and non-carriers in all evaluated regions. These findings are in line with previous PET studies [9,10,11, 53, 54] as well as with a recent study by the Amyloid Biomarker Study Group summarizing APOE ε4 gene dose-related effects on temporal course of Aβ accumulation [55]. In our study, all APOE ε4/ε4 participants over the age of 63 were already Aβ-positive, whereas approximately 50% of the heterozygotes were still Aβ-negative, and 40% of non-carriers were classified as Aβ-positive. Activated microglia are known to be located in the proximity of Aβ plaques in AD, and using PET imaging in vivo, Aβ pathology has been shown to correlate with TSPO-binding in some, although not all studies [7, 25, 27, 56,57,58]. Our voxel level analysis showed significant positive correlations between 11C-PiB and 11C-PK11195 both in cognitively normal APOE ε4 homozygotes and heterozygotes. However, significant clusters were found also in regions outside our a priori chosen regions of interest, such as the white matter, suggesting that these effects might not all be due to higher TSPO binding associated with Aβ plaques.

Here, we also found that Aβ positivity modulated the effect of APOE ε4 gene dose on 11C-PK11195-binding in regions known for early tau deposition, where Aβ-positive APOE ε4 homozygotes seemed to have increased TSPO binding compared with Aβ-positive non-carriers. In addition to Aβ, APOE ε4 is known to accelerate tau pathology that again has been suggested to be closely associated with microglial reactivity [59], and increased tau PET signal in the entorhinal cortex has been reported for cognitively unimpaired Aβ-positive APOE ε4 homozygotes and heterozygotes compared with Aβ-positive non-carriers [53]. Since Aβ build up starts earlier in APOE ε4 carriers, we could hypothesize that increased tau deposition and related inflammatory processes in APOE ε4 carriers would be driving this interaction. Unfortunately, lack of tau PET or CSF tau measurements in our cohort prevented us from investigating the interaction with TSPO-binding and tau further in our cohort.

During recent years, significant efforts have been made to measure various biomarkers of AD pathology in plasma that would provide a less invasive and more easily accessible alternative to brain imaging and lumbar puncture [30]. Plasma GFAP has been recently reported to be an early marker of astrocytosis in response to AD pathology that strongly correlates with Aβ pathology [34, 60] but not with tau when accounting for Aβ [32]. In our cohort, plasma GFAP levels showed elevated concentrations in the most Aβ positive individuals and correlated with composite amyloid PET SUVRs. Interestingly, plasma GFAP was the only biomarker showing significant associations with both cognitive performance and hippocampal volume that could be considered as markers for progression in the Alzheimer’s continuum. On the contrary, plasma GFAP concentration did not correlate with composite TSPO-binding (Fig. 4). This is not surprising, considering that the two inflammatory markers present different targets; plasma GFAP is expected to reflect more astrocytic reactivity associated with Aβ pathology [34], whereas TSPO PET is thought to reflect microglial density [17]. Our results with GFAP support the previous findings suggesting that reactive astrocytosis is present already in cognitively normal individuals and related to Aβ pathology [32,33,34]. Despite clear differences in fibrillar Aβ levels measured by PET, we did not see significant differences between the APOE ε4 gene doses in plasma Aβ1-42/1–40 measured by previously described IP-MS method [47]. Plasma Aβ1-42/1–40 was previously reported to correlate with global cortical Aβ PET signal in another study including cognitively normal individuals using the same IP-MS method [48]. We could not replicate this finding in our cohort, comprised of slightly older and highly APOE ε4-enriched cognitively normal participants, although a trend towards a negative association could be seen in the whole cohort.

Last, we also wanted to compare all the biomarkers and their associations with cognitive and structural variables that could serve as proxies for disease progression in our “at-risk” cohort. We found a negative association between composite cortical TSPO-binding and hippocampal and entorhinal volumes, suggesting that more global elevation in TSPO-binding, and thus microglial density, could be present in individuals with subtle neurodegeneration. Interestingly, higher plasma GFAP associated with both lower cognitive performance and lower hippocampal volume in our cognitively normal cohort. Previously, Hamelin and colleagues reported a positive correlation with both hippocampal volume and MMSE score, suggesting that higher glial reactivity associated with higher hippocampal volume would likely be protective [7]. However, our study population is composed of only cognitively unimpaired individuals highly enriched for APOE ε4 carriers, and all having MMSE scores > 25, thus likely presenting more subtle structural brain changes compared with the population of the previous study. In addition, we did not find any association with TSPO-binding and MMSE, CERAD total score, or the preclinical cognitive composite, in line with other studies performed with 11C-PK11195 [25]. Based on our results, increased TSPO-binding in the preclinical phase, at least in APOE ε4 carriers, could be more related to a later preclinical phase when subtle neurodegeneration already starts to be present.

Strengths and limitations

The strength of this study is our well characterized and balanced cohort of cognitively unimpaired participants stratified by their APOE ε4 gene dose and related risk for sporadic AD and a relatively large number of rare homozygotic carriers of the APOE ε4 allele. In most studies, group comparisons are done between APOE ε4 carriers against non-carriers and especially cognitively unimpaired homozygotes are rarely studied. In addition, we have measured both imaging and more easily available fluid biomarkers for amyloid deposition and neuroinflammatory processes from the same individuals and were able to combine them with additional information concerning brain structure and cognitive data. However, this study does not go without limitations. First, it would have been optimal to compare TSPO PET findings with CSF biomarkers specific to microglial responses, but unfortunately, we did not have CSF samples available from the whole cohort. Thus, plasma GFAP analysis were included. However, we are fully aware that TSPO PET and plasma GFAP reflect different neuroinflammatory processes. Second, we were not able to include tau PET or CSF tau measurements to further investigate association of glial biomarkers and early tau pathology in our cohort. The lack of tau biomarkers prevented us also from evaluating possible early tau pathology as a confounding factor that could have contributed to the reported variation in 11C-PK11195 retention seen in this study. However, since the cohort was comprised of cognitively unimpaired individuals, we did not expect wide-spread tau pathology. Third, even though 11C-PK11195 has shown robust changes in primary inflammatory conditions such as multiple sclerosis, it has been suggested that its sensitivity is limited and outperformed by the second generation TSPO ligands, such as 11C-PBR28. However, affinity of the second generation TSPO ligands is affected by a single nucleotide polymorphism rs6971 in the TSPO gene, leading to division of people into high, mixed, and low affinity binders. Due to the difficulty of recruiting rare homozygotic APOE ε4 carriers, we wanted to avoid the unfortunate scenario of having multiple homozygotic participants excluded due to low-binding TSPO genotype.

Conclusion

Our study on cognitively unimpaired “at-risk” individuals carrying either one or two copies of the APOE ε4 gene showed clear differences in fibrillar Aβ load in the brain, but the changes were not accompanied by higher glial reactivity as measured with TSPO PET either in APOE ε4 carriers, or in Aβ-positive individuals, representing preclinical AD. However, cortical TSPO binding was associated with lower hippocampal and entorhinal volumes in the whole cognitively unimpaired sample. These findings suggest that in our cognitively unimpaired cohort enriched by APOE ε4 carriers, neuroinflammatory processes measured by TSPO PET are related to a more advanced preclinical phase of AD where Aβ accumulation is accompanied by subtle structural changes.