Early β-amyloid accumulation in the brain is associated with peripheral T cell alterations
Christoph Gericke and Tunahan Kirabali contributed equally to this work.
Valerie Treyer, Maria Teresa Ferretti and Anton Gietl jointly directed this work.
Abstract
INTRODUCTION
Fast and minimally invasive approaches for early diagnosis of Alzheimer's disease (AD) are highly anticipated. Evidence of adaptive immune cells responding to cerebral β-amyloidosis has raised the question of whether immune markers could be used as proxies for β-amyloid accumulation in the brain.
METHODS
Here, we apply multidimensional mass-cytometry combined with unbiased machine-learning techniques to immunophenotype peripheral blood mononuclear cells from a total of 251 participants in cross-sectional and longitudinal studies.
RESULTS
We show that increases in antigen-experienced adaptive immune cells in the blood, particularly CD45RA-reactivated T effector memory (TEMRA) cells, are associated with early accumulation of brain β-amyloid and with changes in plasma AD biomarkers in still cognitively healthy subjects.
DISCUSSION
Our results suggest that preclinical AD pathology is linked to systemic alterations of the adaptive immune system. These immunophenotype changes may help identify and develop novel diagnostic tools for early AD assessment and better understand clinical outcomes.
1 BACKGROUND
Alzheimer's disease (AD) is a highly prevalent neurodegenerative disorder that leads to progressive memory loss and cognitive as well as functional decline, eventually resulting in dementia symptoms. Accumulation and aggregation of misfolded β-amyloid peptide (Aβ) in the brain represents a primary pathological event of AD1 that occurs over decades before symptom onset in a silent, preclinical phase of the disease.2 Cognitively healthy control subjects (HCS) with elevated brain β-amyloid are individuals at such an early, preclinical stage of the AD continuum, with higher risk for disease progression toward mild cognitive impairment (MCI) and established AD dementia stage.3, 4 Therefore, preclinical AD is currently considered an ideal window of opportunity for disease-modifying therapeutic strategies.5 However, due to a lack of knowledge about prognostic AD biomarkers, it is difficult to identify AD risk populations and populations in which therapeutic interventions might prevent impending neuropathological events. Early disease diagnosis is the subject of intense investigation and in particular, blood-based biomarkers such as isoforms of phosphorylated tau (p-tau181, p-tau217, and p-tau231), as well as Aβ42/Aβ40 ratio in blood plasma have recently been linked to early β-amyloid accumulation in the brain.6, 7 Better characterization of additional systemic changes in the AD continuum would be crucial to improve diagnostic accuracy in the preclinical stage.
The immune system is designed to respond to any homeostatic perturbation, and both innate and adaptive immune cells have been shown to react to AD pathology.8-12 Moreover, emerging genetic and clinical data indicate an involvement of inflammation-related pathways in AD pathogenesis and disease progression.13-16 Thus, whether immune markers could be used as proxy of β-amyloid accumulation in the brain is a clinically highly relevant question. In this study, we therefore applied state-of-the-art single-cell immunophenotyping, focusing on peripheral T lymphocytes, the most abundant cellular subtype of the adaptive immune system. After antigen exposure, T cells undergo clonal expansion and differentiation into antigen-specific effector and long-lived memory cells that can spark a faster and more vigorous immune response upon second encounter with their cognate antigen.17 Priming of the adaptive immune system by brain-derived antigens is still a matter of debate. The existence of protective antibodies against β-amyloid aggregates in healthy aged donors18, 19 strongly indicates that the peripheral immune system can ‘sense’ the brain, most likely via the drainage of brain-derived solutes through dural lymphatic vessels.20-23 Human T cells have been shown to be reactive against specific epitopes derived from the linear Aβ1-42 peptide sequence in vitro.9, 24 However, while research has focused on the link between T cell reactivity and cognitive outcomes in MCI and AD,252627 the association of peripheral T cell alterations with key pathological biomarkers such as cerebral β-amyloid deposition and AD plasma markers in a preclinical stage of AD has not yet been addressed in detail.
Multivariate mass cytometry (CyTOF) combined with data-driven, unbiased analysis tools allows for comprehensive characterization of those complex, disease-associated cellular patterns. The technology has been recently refined28 and established in cancer biology (e.g., leukemia),29, 30 neurological conditions such as narcolepsy31 or multiple sclerosis,32 and infectious diseases (e.g., immune factors in M. tuberculosis infections).33 Here, we investigate whether systemic adaptive immune cell alterations reflect early changes in AD biomarkers. We employed CyTOF to analyze peripheral blood mononuclear cells (PBMCs) from well-characterized subjects in cross-sectional and longitudinal studies. We show that increased levels of brain β-amyloid and changes in plasma AD biomarkers as well as cerebral β-amyloid accumulation over time are indeed associated with a systemic increase in effector subpopulations of adaptive immune cells, particularly T cell lineages, even in the absence of cognitive symptoms. Importantly, we demonstrate that such systemic alterations are independent of typically latent and widespread Herpesviridae infections.
2 METHODS
2.1 Resource availability
2.1.1 Contact for reagent and resource sharing
Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Dr Christoph Gericke ([email protected]).
2.1.2 Data and code availability
The data sets and code generated during this study are publicly available at Zenodo (CERN, European Organization for Nuclear Research) under https://doi.org/10.5281/zenodo.7157682.
2.2 Experimental model and subject details
2.2.1 Ethics statement
The participants included in the analyses participated in in-house research studies conducted by the Center for Prevention and Dementia Therapy at the Institute for Regenerative Medicine at the University of Zurich, Switzerland. All study participants gave written informed consent. All studies and further use for the current analyses were approved by the local ethics committee (Kantonale Ethikkommission Zürich) and conducted in accordance with their guidelines and the Declaration of Helsinki.34
2.2.2 Study participants – exploratory cross-sectional study I
In an exploratory study named Cross-sectional I, we included cryopreserved PBMCs, blood plasma and serum from cognitively healthy control subjects (HCS) (n = 23), subjects diagnosed with MCI (n = 18) according to consensus criteria,35 and patients with probable AD (n = 9) according to NINCDS-ADRDA 198036 and ICD-10 criteria.37 All subjects previously participated in cohort studies with comparable research protocols. Cerebral β-amyloid load was assessed by positron emission tomography (PET) imaging with [11C]-Pittsburgh compound B (PiB) tracer except for AD subjects who were assessed in a protocol without amyloid-PET. We used the standardized uptake value ratios (SUVR) and the Centiloid method for better comparability among different tracers.38 Subjects were selected to compare low PiB SUVR (β-amyloid negative) and high PiB SUVR clearly above the cutoff (β-amyloid positive). The lowest PiB+ SUVR was 1.514 while the highest PiB- SUVR was 1.250, the pre-defined SUVR cutoff level was 1.265.39 Study participants received extensive blood examinations including apolipoprotein E genotyping (APOE ε2, ε3, ε4), viral antibody titer analysis (against Herpesviruses) and assessment of acute inflammatory markers (C-reactive protein, CRP). The median time difference between PBMC sampling and PET imaging was 19 days. MCI β-amyloid– subjects were included as internal control, since they are considered to be cognitively impaired due to reasons other than AD. MCI β-amyloid+ patients showed both early signs of cognitive impairment and β-amyloid deposition, and were categorized as MCI due to AD. Detailed group sizes, demographics and characterization of Cross-sectional I are shown in Figure 1, Figure S1, and Table S1. Due to the exploratory nature of Cross-sectional I, no a priori power analysis was conducted to estimate sample sizes.
RESEARCH IN CONTEXT
-
Systematic review: We screened PubMed for literature on multiparameter immune cell analysis in early Alzheimer's disease (AD). Other research projects focused on the link between immunological changes and cognitive outcomes in mild cognitive impairment (MCI) and AD, but the association of peripheral immune cell alterations with important biomarkers such as cerebral β-amyloid deposition and AD plasma markers in a preclinical stage of AD has not yet been studied in detail.
-
Interpretations: Here we investigated whether blood immune alterations reflect preclinical changes in AD biomarkers. We found that increased levels of brain β-amyloid and changes in plasma AD biomarkers, as well as cerebral β-amyloid accumulation over time, were indeed associated with a systemic increase in adaptive immune cells, particularly effector T cell lineages, even in the absence of cognitive symptoms.
-
Future directions: Future long-term longitudinal studies will help determine the conversion rate of cognitively healthy individuals with altered immune profiles into MCI or AD. Possible epitope targets of the identified T cells will be investigated by targeted epitope mapping.
2.2.3 Study participants – validation cross-sectional study II
A second cross-sectional study, termed Cross-sectional II, included 95 HCS and 47 MCI cases from the baseline of an ongoing longitudinal cohort study whose participants were recruited via newspaper advertisements. Neuropsychological examination according to consensus criteria and further blood examinations were performed in a comparable way as in Cross-sectional I. All subjects received PET imaging with [18F]-Flutemetamol (FMM) tracer (Vizamyl, GE Healthcare). Finding optimal cutoffs was important for the cohort design; therefore, we included FMM Centiloids representing AD biomarker-positive cases (12 < Centiloid < 30) and established AD pathology cases (Centiloid > 30),40 with an overall cutoff of FMM Centiloid > 12 for β-amyloid+ subjects. All β-amyloid– cases had Centiloid levels below 9. This resulted in a more homogenous and therefore more natural β-amyloid PET signal distribution compared to Cross-sectional I, including β-amyloid+ subjects with FMM SUVR and Centiloid close to the cutoff (Figure S1). Further selection was based on health status (excluding current/recent tumor patients or patients with Hashimoto's disease) and age range. Considering these criteria, Cross-sectional II contained 50 HCS β-amyloid–, 45 HCS β-amyloid+, 26 MCI HCS β-amyloid–, and 21 MCI HCS β-amyloid+ subjects. According to a priori power analysis considering the effect size in CMV titer-adjusted CD8+ TEMRA/Effector cell analysis, the minimum sample size for a validation HCS β-amyloid+ group should be n = 26 (Power = 0.8; Type I error = 0.1).41 Detailed group sizes, demographics, and characterization of Cross-sectional II are shown in Figure 1, Figure S1, and Table S2.
2.2.4 Study participants – longitudinal study
For the longitudinal study, participants were followed over a total time period of 3 years. In order to be eligible, subjects had to be between 55 and 80 years old and cognitively healthy at baseline which was ascertained by an MMSE (Mini-Mental State Examination) minimum score of 2742, 43 as well as comprehensive clinical and neuropsychological examination. Every eligible subject for the current analysis (n = 59) received two separate [11C]-PiB-PET scans at Baseline and after 3 years at the end of the study. PBMCs were isolated from participants’ blood donations at Baseline, after 18 months (= Follow-up1), and after 3 years (= Follow-up2). While all individuals received β-amyloid PET imaging at Baseline and Follow-up2, PBMC samples were not available for all visits of each subject. Availability of PBMCs from different visits is summarized in Table S3. Due to the exploratory nature of this longitudinal study, no a priori power analysis was conducted to estimate sample sizes.
In the longitudinal study, subjects were considered β-amyloid+ at PiB SUVR > 1.223 (75th percentile of baseline SUVR). We allocated subjects into groups of cerebral β-amyloid Non-accumulators and β-amyloid Accumulators based on their change in cortical PiB SUVR over time (termed ΔPiB). Previous studies have used basic discriminations considering every subject with an annual ΔPiB SUVR > 0 as Accumulator2, 44 or have established cutoffs based on inflexion points between bimodal distributions of ΔPiB SUVR/year.45 Because negative ΔPiB SUVR in Baseline β-amyloid– subjects presumably reflects measurement noise,44 we allocated individuals with ΔPiB SUVR < 0 automatically to a group of Non-accumulators. We aimed to identify only Accumulators with clearly increasing PiB retention over time compared to Non-accumulators. For that reason, all individuals with ΔPiB SUVR ≥ 0 were split by median, and as a result subjects with ΔPiB SUVR ≥ 0.04 were classified as Accumulators, whereas subjects with ΔPiB SUVR < 0.04 were categorized as Non-accumulators. In an alternative group allocation, we divided subjects according to the change of plasma p-tau181 levels over time (Δp-tau181). Via median split, we defined a cutoff at Δp-tau181 = 0.13pg/mL, to distinguish between p-tau181 Accumulators and Non-accumulators. Detailed group sizes, demographics, and characterization of the cohort are shown in Table S4.
2.3 Method details
2.3.1 APOE genotyping
APOE genotyping was either performed by restriction isotyping as previously described46 or by commercially available Sanger Sequencing (Microsynth AG).
2.3.2 PET imaging and image analysis
Dynamic [11C]-PiB-PET or [18F]-FMM-PET images were acquired over 70 min on whole-body PET/CT cameras in 3D mode (Discovery RX and Discovery STE, GE Healthcare). Subjects received approximately 350 MBq of tracer. Image reconstruction included standard corrections and filtered back projection algorithms with an image resolution of 2.34*2.34*3.27 mm. For magnetic resonance imaging (MRI), 3D TFE T1-weighted imaging was performed on a Philips 3T Achieva with an image resolution of 0.9375*0.9375*1 mm.
Images were processed with PMOD 3.7 NeuroTool (PMOD Technologies, Zurich, Switzerland) and brain parcellation method. MR images were segmented (3 probability maps methods), and cortical regions of interest were defined according to ‘Adult brain maximum probability map’ by Hammers et al. as implemented in PMOD NeuroTool. For non-cortical region definitions, deep nuclei parcellation method as implemented in NeuroTool was performed (each hemisphere is processed individually, 16 reference sets were taken into account to improve accuracy). Gray matter–white matter segmentation cutoff was defined at a standard 50% probability. This segmentation was applied to region of interest definitions to exclude white matter voxels from analysis. For the follow-up scans in the longitudinal study, the same MR baseline image and the same regions of interest definitions were taken to reduce variability due to automatic segmentation and outlining process between the two time points. PET images were co-registered to the baseline MR images based on the early frames signal (6 min) of the dynamic PET scan, which resembles a perfusion map. Manual correction was applied where needed to ensure optimal co-registration. On the fused individual images in MR space, correct outlining of the regions of interest was checked and corrected if necessary. Late phase uptake values were calculated as average late frames activity (50 to 70 min post injection) of cortical regions (averaging cortical regions as defined by the ‘Centiloid’ approach and divided by cerebellar cortex activity).38 Due to the standard procedure and standard calibration of equipment, the SUVR approach was considered sufficient and most robust to evaluate changes over 3 years in the longitudinal study cohort.
2.3.3 Neuropsychological testing
The test selection for the descriptive analysis was based on availability of identical neuropsychology tests over all cohorts and should provide information on the cognitive status or changes of the different subgroups. In all studies, the participants received first a short informal interview to assess subjective cognitive complaints and factors with possible influence on the test results (e.g., alcohol/drug abuse, medication, recent travelling over time zones) as well as characteristics of free language expression. The here-reported neuropsychological tests (Table S5) were comparable in all studies but were not necessarily performed in the same order. Especially in the AD group, not all available tests were performed. Z-scores (not reported, available upon request) were calculated based on norm values of the individual tests as defined in the respective test manuals.
2.3.4 Plasma biomarker analysis
Blood plasma samples were obtained on the same day as PBMC extraction was performed. Blood of study participants was collected in ethylenediaminetetraacetic acid (EDTA) tubes (Vacutainer EDTA Tubes, BD), inverted 10 times, and centrifuged at 1620 × g for 12 min at 6°C. Plasma supernatant was aliquoted and immediately stored at −80°C. For ultrasensitive AD biomarker quantification, EDTA-plasma of Cross-sectional I and the longitudinal study was shipped to Quanterix corporation (Quanterix, Billerica, MA, USA). Analysis was performed on a Simoa HD-X instrument using Simoa Human Neurology 3-plex A (N3PA) and P-tau181 V2 immunoassays for measuring plasma Aβ1-42, Aβ1-40, total tau, and phospho-tau181 concentrations.
2.3.5 CyTOF antibody panels
Monoclonal, heavy metal isotope-conjugated anti-human antibodies were purchased from Fluidigm or self-made from purified antibodies and heavy metal isotopes using the Maxpar X8 Multi-Metal Labeling Kit (Fluidigm) according to manufacturer's instructions. Antibody panels were designed for barcoding purposes as well as for characterization of T cells for both cross-sectional and longitudinal studies. All panels and further panel information are listed in the Supplemental Materials.
2.3.6 CyTOF barcoding strategies
In order to reduce technical noise due to staining variabilities among single samples, we applied different barcoding strategies to allow for sample pooling and bulk PBMC analysis. For all studies, we applied custom-made live-cell barcoding based on combinations of anti-human CD45 antibodies labelled with different metal tags. For the T cell study in Cross-sectional I, we applied an 8-choose-3 approach (56 possible combinations), whereas for Cross-sectional II and the longitudinal T cell study we applied a 10-choose-4 approach (210 possible combinations). Detailed live-cell barcoding panels are listed in Table S6.
2.3.7 Sample preparation and surface staining
Cryopreserved PBMCs were thawed in a 37°C water bath and transferred into tubes with pre-warmed cell medium (RPMI 1640, Sigma Aldrich with 10% heat-inactivated FBS (Gibco) and 1% Glutamax (Gibco)). After two washes in cell medium (350 × g, 7 min), 1 million PBMCs from each subject were transferred into a 96-well plate, washed twice in PBS (350 × g, 5 min), and incubated with Cell-ID Cisplatin-198Pt (Fluidigm) 1:10,000 in PBS for dead-cell exclusion staining for 10 min at room temperature (RT). Cells were incubated with Fc receptor blocking antibodies (Human TruStain FcX, Biolegend) 1:20 in Cell Staining Buffer (CSB, Fluidigm) for 5 min at RT before applying the respective barcoding strategies. After barcoding, cells were washed twice in CSB, pooled, and incubated as bulk sample with their antibody cocktail (see Supplemental Materials) for 20 min at RT. After washing in CSB, cells were fixed with 1.6% formaldehyde in CSB and stored overnight at 4°C. The next day, stained samples were aliquoted in max. 20 Mio. cell batches and frozen in 10% dimethyl sulfoxide (DMSO, Thermo Fisher) in FBS for long-term storage at −80°C. On the day of analysis, frozen samples were thawed in pre-warmed phosphate buffered saline (PBS) and acquired on the mass cytometer (CyTOF) according to acquisition protocol (Figure 1).
2.3.8 CyTOF acquisition
Prior acquisition, cells were incubated in Cell-ID Intercalator-Ir (Fluidigm) 1:5000 in Maxpar Fix and Perm Buffer (Fluidigm) for 1 h at RT. After one wash step in deionized metal-free water and two wash steps in Cell Acquisition Solution (CAS, Fluidigm), samples were diluted to 1.2 million cells/mL in CAS containing 1:10 EQ Four Element Calibration Beads (Fluidigm). Samples were analyzed with a Helios CyTOF2 mass cytometer (DVS Sciences, Fluidigm).
2.3.9 Serology for virus antibody titers
Viral antibody titers in serum samples from both cross-sectional studies were analyzed via enzyme-linked immunosorbent assay (ELISA) or indirect immunofluorescence at the Institute of Medical Virology, University of Zurich. Antibody titers against cytomegalovirus (CMV) (immunoglobulin [Ig]G and IgM), Epstein-Barr virus (EBV) (IgG and IgM against viral capsid antigen (VCA), and IgG against Epstein-Barr nuclear antigen (EBNA) were tested), Herpes simplex virus (HSV) (IgM for HSV-1+2, HSV-1 IgG, HSV-2 IgG) and Varicella-Zoster-Virus (VZV) (IgG and IgM) were investigated. Quantitative results were obtained for the following antibody titers: VZV IgG (in mIU/mL, threshold for positive test > 100), CMV IgG (in AE/mL, threshold ≥6), HSV-1 IgG (in “signal divided by cutoff” [S/CO], threshold > 1), and HSV-2 IgG (in S/CO, threshold > 1). All other titers were analyzed qualitatively (positive/negative).
2.3.10 C-reactive protein (CRP) assessment
CRP levels for serum samples from both cross-sectional and longitudinal study participants were analyzed as part of routine diagnostics externally in certified laboratories to exclude acute background infections and inflammation.
2.4 Quantification and statistical analysis
2.4.1 CyTOF data preprocessing
Bead normalization over time, file concatenation, debarcoding, and compensation for overspilling high-intensity barcoding channels were performed using Premessa (R package, version 0.3.4) and Catalyst (R package, v1.22.0) in an R environment (R, v4.2.2) implemented in RStudio (v2022.12.0-353). All R packages used are listed in the Supplemental Materials.
2.4.2 Multivariate data analysis
Custom-made R scripts (see data/code repository) were used for data analysis and visualization. Mass signals from calibration beads, cisplatin-positive (198Pt) dead cells, iridium-low (191Ir, 193Ir) debris, and iridium-high (191Ir, 193Ir) cell doublets were cleaned out. For further analysis, CD45+ PBMCs, CD8+ T cells (defined as CD3+ CD4– CD8+), and CD4+ T cells (defined as CD3+ CD4+ CD8–) were selected and exported as individual cell subsets. A maximum of 10,000 events (CD45+ leukocyte analysis and T cell analysis) per FCS file were randomly selected and processed via flowCore (R package, v2.10.0). In case of files that included less than 10,000 events, all events were used for the analysis. All events were transformed using an inverse hyperbolic sine (arcsinh) function with a cofactor of 5.47 Next, the multivariate n-dimensional data were processed by tSNE (t-distributed stochastic neighbor embedding), a nonlinear dimensionality reduction algorithm (Rtsne package, v0.16). Spatial separation of events in the resulting tSNE map is controlled by the ‘perplexity’ parameter.
2.4.3 FlowSOM clustering
Automated and unsupervised clustering was performed in R using the FlowSOM algorithm48 implemented in the FlowSOM R package (v2.6.0). For meta-clustering, consensus hierarchical clustering was used, and respective k-values defining the number of resulting clusters were determined based on estimations of expectable subclusters.31
2.4.4 Identification of immune cell subsets
Via data visualization package ggplot2 (R package, v3.4.0) automatically identified FlowSOM clusters were overlaid with their corresponding cell events in the tSNE map, and heatmaps displaying median signal intensities per cluster were created. Based on signal intensities, immune cell subpopulations were identified and similar clusters were merged to reflect biologically meaningful immune cell subsets. Eventually, relative cluster abundances were determined as quantification readout.
2.4.5 CITRUS
In parallel to FlowSOM, CITRUS (Cluster Identification, Characterization and Regression) was used as an alternative method for unsupervised clustering and identification of statistically significant differences in immune cell abundances between diagnostic groups.49 The correlative significance analysis of microarrays (SAM) association model was used with maximum 10,000 cells per sample, 1% minimum cluster size and 5% false discovery rate (FDR). Clusters that showed differential relative abundances among diagnostic groups were quantified and visualized.
2.4.6 Cross-sectional studies – statistical analysis
Statistical analysis was performed using Graphpad Prism (v9.1.2) and R (v4.2.2). Assumption of normal distribution of variables was tested with the Shapiro-Wilk normality test (α = 0.05). If the null-hypothesis of normality was not rejected, significant differences between two groups were tested with a two-tailed, unpaired Student's t test. Correlations for normally distributed data were calculated with Pearson's correlation (α = 0.05, confidence interval 95%, two-tailed). In case of a deviation from normal distribution, significant differences between two groups were tested with a non-parametric, two-tailed, unpaired Mann-Whitney test. Correlations were calculated with non-parametric Spearman's correlation (α = 0.05, confidence interval 95%, two-tailed). The difference of cluster abundances between multiple diagnostic groups was compared using the non-parametric Kruskal-Wallis test. All p values were corrected for multiple comparisons within each cluster using the Benjamini-Hochberg false discovery rate (FDR) method. Age-, sex-, or virus-related effects were estimated and controlled using linear regression models, and analysis was performed on the basis of the corresponding residuals. All p values are reported as absolute values including an estimate of FDR in each figure. Due to small cohort size, significance was defined as FDR < 10%.
2.4.7 Longitudinal study – descriptive analysis and statistical modelling
For the longitudinal study, a linear mixed model (LMM) was fitted in R. For simplicity, a two-time point analysis was chosen. If available, the relative abundance of immune cell clusters and the corresponding cortical PiB SUVR or plasma p-tau181 data at time points Baseline and Follow-up2 were used. If there were no immune cell data available at Baseline or Follow-up2, time point Follow-up1 served as replacement. Cortical PiB SUVR or plasma p-tau181 at each time point considered was used as response variable (see ‘response vector’ in Formula S1). Cortical PiB SUVR and plasma p-tau181 were measured only at Baseline and Follow-up2, hypothetical values for Follow-up1 were generated by linear interpolation (see imputed PiB SUVR values). The relative abundance of immune cell clusters at each time point considered was used as covariate. The LMM was used to further investigate the association of individual changes in cortical PiB SUVR or plasma p-tau181 over time with covariates such as relative abundance of immune cell clusters, group allocation (PiB SUVR or plasma p-tau181 Accumulators vs. Non-accumulators) and age (see “design matrix for fixed effects” in Formula S1). Additionally, we included an interaction term describing the combined effect of changes in relative abundance of immune cell clusters over time and allocation to the Accumulator group (see “design matrix for fixed effects” in Formula S1). Relative cell abundances were transformed using the arcsinh transformation. The LMM was fitted with the lme4 R package (v1.1-26).
3 RESULTS
3.1 CyTOF-based immunophenotyping of cross-sectional study populations covering the early AD spectrum
We analyzed blood-derived material from two separate study populations focusing mainly on the early spectrum of AD, namely cognitively healthy control subjects (HCS) and subjects with MCI with biomarker evidence for AD. All study participants underwent neuropsychological testing, plasma AD biomarker analysis, and cerebral β-amyloid imaging via PET scanning using PiB or FMM tracers (Figure 1A, Table S1 and S2). To immunophenotype adaptive immune cell populations using CyTOF, we stained live-cell-barcoded PBMCs with self-designed, heavy metal-conjugated antibody panels for the analysis of surface markers on PBMCs, in particular on T cells (Figure 1B). For a first exploratory study (termed Cross-sectional I), we selected PBMCs from 50 individuals (Figure 1C) to represent different cognitive stages of the disease and markedly different β-amyloid status based on PiB PET results. Cross-sectional I includes five diagnostic groups based on clinical assessment and brain β-amyloid load, with the exception of AD patients, who were derived from a protocol that did not include amyloid PET: AD (n = 9), MCI (β-amyloid+, n = 9 and β-amyloid–, n = 9), and HCS (β-amyloid+, n = 10 and β-amyloid–, n = 13). As expected, cerebral β-amyloid load (PiB SUVR and Centiloid) was significantly different between positive and negative groups (Figure S1A and S1B), with β-amyloid+ groups clearly above the cutoff. Both β-amyloid+ groups were characterized by higher frequency of APOE ε4 carriers (approx. 2x in β-amyloid+ compared to β-amyloid– groups) (Figure 1C, Table S1). Furthermore, AD subjects were older than the other groups (p < 0.01, AD vs. HCS β-amyloid–) and more likely to be women (Table S1). Since altered levels of plasma AD biomarkers such as Aβ42/Aβ40 ratio,50 p-tau181,51 and p-tau181/Aβ42 ratio52 have been well associated with preclinical AD and MCI due to AD, and are deemed suitable for early diagnosis,53 we included plasma markers for further characterization (Figure S1C-S1E). As expected, most significant alterations were observed for HCS β-amyloid+, MCI β-amyloid+ and AD compared to β-amyloid– groups. Plasma total tau, an approximate marker of neurodegeneration, was most elevated in AD patients compared to HCS β-amyloid– and MCI β-amyloid+ (Figure S1F).
A second, independent cross-sectional study (termed Cross-sectional II) was analyzed to extend and reproduce key findings of Cross-sectional I. Here, the study population consisted of 142 participants including 4 diagnostic groups: HCS (β-amyloid+, n = 45 and β-amyloid–, n = 50) and MCI (β-amyloid+, n = 21 and β-amyloid–, n = 26) (Figure 1D). In this study, subjects with FMM SUVR and Centiloid more closely around the cutoff were included, resulting in a more homogeneous distribution of cerebral β-amyloid load among these diagnostic groups compared to Cross-sectional I. Cerebral β-amyloid load (FMM SUVR and Centiloid) was significantly different between β-amyloid+ and β-amyloid– groups of Cross-sectional II (Figure S1G and S1H). However, when comparing the β-amyloid+ MCI groups of both studies, Cross-sectional II had a significantly lower median FMM Centiloid = 38.92 (interquartile range [IQR] = 58.49) compared to PiB Centiloid = 95.54 (IQR = 27.86) in Cross-sectional I (Figure S1I). The β-amyloid+ groups of Cross-sectional II had a higher frequency of APOE ε4 carriers (Figure 1D, Table S2).
After CyTOF acquisition, obtained datasets were depicted via automated dimensionality reduction tools (here: tSNE) and analyzed with data-driven, unbiased clustering algorithms (here: FlowSOM) (Figure 1E and 1F). The resulting clusters were compared to heatmaps, which contain the median signal intensities for every marker of interest (Figure 1G and 1H) and allow for assignment of clusters to biologically meaningful cell types. Automated clustering successfully identified major CD45+ PBMC immune cell populations such as CD4+ and CD8+ T cells, double-negative (DN) / γδ T cells, CD19+ B cells, CD14+/low CD16+/– monocytes, CD16+/low CD56+ natural killer (NK) cells and blood-derived dendritic cells (DCs), which split into BDCA1+ myeloid DCs (mDCs) and BDCA2+ plasmacytoid DCs (pDCs) (Figure 1E-1H). Adaptive immune cell populations were largely comparable among the different diagnostic groups (Figure S1J and S1K). Of note, the relative abundance of total CD8+ T cells were slightly increased in HCS β-amyloid+ subjects compared to AD samples in Cross-sectional I (Figure S1J). In the following, we took further insight into adaptive immune cell populations by subclustering T cell subpopulations of interest for in-depth analysis.
3.2 Cross-sectional I – increase of peripheral CD8+ TEMRA/effector cells in AD-biomarker-positive cognitively healthy individuals
CD3+ CD4– CD8+ T cells were preselected and analyzed via FlowSOM automated clustering (Figure 2A). Six major CD8+ T cell subpopulations were identified via median signal intensities of lineage and activation markers (Figure 2B). One of the most prominent CD8+ T cell subclusters was a mixed cluster of CD197– CD45RA-reactivated T effector memory cells (TEMRA cells) and T effector cells (together termed TEMRA/effector cells) (Median = 23.4% of CD8+ T cells; IQR = 21.2%).54 Moreover, we found CD197+ CD45RA+ naïve T cells, CD56+ NK T cells, CD197+ CD45RA– central memory (CM) T cells, CD197– CD45RA– effector memory (EM) T cells, as well as a potentially exhausted variant of EM T cells (HLA-DR+ PD-1+).
Relative abundances of CD8+ T cell subpopulations were compared between an HCS β-amyloid+ group of interest (= preclinical AD, positive for β-amyloid PiB PET) and HCS β-amyloid–, MCI β-amyloid+, as well as AD patients (Figure 2C). Statistical analysis indicated that relative abundance of naïve CD8+ T cells was significantly reduced in HCS β-amyloid+ study participants compared to HCS β-amyloid–, MCI β-amyloid+ and AD subjects. In contrast, CD8+ TEMRA/Effector cells showed only a tendency to increase in HCS β-amyloid+ study participants compared to HCS β-amyloid– and MCI β-amyloid+ subjects (p = 0.1). The CD8+ TEMRA/Effector cell population is a heterogeneous group; therefore, in order to characterize its increase in HCS β-amyloid+ individuals in greater detail, we subclustered the TEMRA/Effector cell cluster and compared cell abundances across groups. We found that a terminally differentiated, reactivated, and highly antigen-specific TEMRA subpopulation, CD57+ TEMRA cells,55-57 was increased in HCS β-amyloid+ subjects compared to HCS β-amyloid– and MCI β-amyloid+ (Figure 2D). The sex of the study participants did not affect immune cell clusters of interest (Pearson's chi-squared test: p(CD8+ naïve T cells) = 1.00; p(CD8+ TEMRA/Effector cells) = 1.00; p(CD8+ CD57+ TEMRA cells) = 1.00), despite the imbalanced male/female distribution in Cross-sectional I (Figure 1C, Table S1). Age differences among diagnostic groups had only minor effect on CD8+ T cells’ relative cluster abundances (Figure S2A, Spearman correlation: −0.5 < rho < 0.5). Additional multiple linear regression analysis for age and sex was consistent with these results, with the only difference being that naïve CD8+ T cells were significantly affected by age, but only with low estimate close to 0 (Figure S2B).
We confirmed our findings using an alternative clustering and analysis via CITRUS and its association models, which enables a fully automated discovery of statistically significant biological signatures within single-cell datasets.49 The most significantly different clusters resulting from CITRUS analysis in combination with the correlative SAM association model were highlighted in hierarchical clustering maps (Figure 2E). Marker expression analysis revealed that these cluster subsets belonged to naïve CD8+ T cells as well as a subpopulation of CD57+ TEMRA/Effector cells. Quantification of these clusters revealed that results on naïve and TEMRA/Effector CD8+ T cells could be replicated using correlative CITRUS analysis (Figure 2F).
In an alternative diagnostic grouping, HCS and MCI subjects were subdivided according to median split of plasma p-tau181 levels (1.89 pg/mL) instead of β-amyloid PiB PET. For readability, we refer to subjects who are above the median as “p-tau181+”, being aware that this is not a clinical cutoff for tau positivity. Relative abundance of CD8+ naïve T cells was reduced in HCS p-tau181+ subjects compared to HCS p-tau181–, MCI p-tau181+, and AD, whereas TEMRA/Effector and CD57+ TEMRA cells were increased in HCS p-tau181+ study participants in comparison with HCS p-tau181– and MCI p-tau181+ (Figure 2G-2J). Thus, using two independent clustering methods as well as two different AD biomarkers, we show that a preclinical AD stage with early cerebral β-amyloidosis in cognitively healthy subjects (HCS β-amyloid+/p-tau181+) is characterized by an increase in CD8+ TEMRA/Effector cells with a concomitant decrease in CD8+ naïve T cells. In addition, specifically when we analyzed plasma p-tau181, we found a decrease in PD-1+ (CD279+), potentially exhausted EM cells in HCS p-tau181+ compared to HCS p-tau181– (Figure 2K).
To control for association with APOE ε4 genotype, the most prominent risk factor for AD, we compared relative abundances of our immune populations of interest among APOE ε4 carriers and non-carriers. We found that CD8+ TEMRA/Effector and CD57+ TEMRA cells were elevated in APOE ε4 carriers compared to non-carriers (Figure 2L and 2 M). However, APOE ε4 genotype might be a confounding variable in this case because we observed, as expected, significantly higher median β-amyloid load in APOE ε4 carriers (Figure 2N).
3.3 β-Amyloid-positive, cognitively healthy individuals are further characterized by reduced naïve CD4+ T cell numbers
CD3+ CD4+ CD8– cells were preselected, clustered, and identified as shown before (Figure S3A and S3B). Similar to what we observed for CD8+ T cells, relative abundance of CD4+ naïve T cells was lower in HCS β-amyloid+ individuals compared to MCI β-amyloid+ and AD patients (Figure S3C). However, the majority of other CD4+ T cell subclusters did not exhibit statistically significant differences in relative abundances. Only CD4+ effector memory T cells were increased in HCS β-amyloid+ subjects compared to AD patients. Age differences between diagnostic groups had no effect on relative cluster abundances of CD4+ T cells (Figure S2C, Spearman correlation: −0.5 < rho < 0.5). The sex of the study participants did not affect CD4+ T cell clusters of interest (Pearson's chi-squared test: p(CD4+ naïve T cells) = 1.00; p(CD4+ EM cells) = 1.00). No differences in CD4+ T cell clusters were observed when diagnostic groups were formed based on plasma AD biomarkers.
3.4 Cross-sectional II – confirmation of increased CD8+ TEMRA/effector cell abundance in early stages of AD
To specifically investigate the reproducibility and robustness of the prominent CD8+ T cell results from Cross-sectional I, we analyzed a larger study population, termed Cross-sectional II. The selection of the participant was less stringent on β-amyloid status and also included values closer to the cutoff compared to Cross-sectional I. Due to updated marker panels, we were able to conduct a more detailed cluster characterization and identification via tSNE and FlowSOM clustering, identifying seven major subpopulations (Figure 3A and 3B). We compared the median relative abundance of CD8+ T cell clusters across diagnostic groups. Based on a false discovery rate of 10%, we found a significant decrease in CD8+ naïve T cells and an increase in CD8+ TEMRA/Effector cells in HCS β-amyloid+ and, interestingly, also in MCI β-amyloid+ subjects compared to HCS β-amyloid– (Figure 3C). Age differences between diagnostic groups had only minor effect on relative cluster abundances (Figure S2D, Spearman correlation: −0.5 < rho < 0.5). Although the male/female distribution in Cross-sectional II was imbalanced (Figure 1D, Table S2), the sex of the study participants had no influence on immune cell clusters of interest (Pearson's chi-squared test: p(CD8+ naïve T cells) = 0.50; p(CD8+ TEMRA/Effector cells) = 1.00). Additional multiple linear regression analysis confirmed again that CD8+ T cell cluster abundances were largely independent of age and sex; only naïve CD8+ T cells were significantly affected by age, albeit with low estimate close to 0 (Figure S2E). We found that CD8+ TEMRA/Effector cells were elevated in APOE ε4 carriers compared to non-carriers (Figure 3D). Also here, APOE ε4 genotype is a confounding variable since we observed significantly higher median β-amyloid load in APOE ε4 carriers (Figure 3E).
We confirmed our findings using correlative CITRUS clustering. Hierarchical cluster maps generated with CITRUS for CD8+ T cells from 90 randomly selected samples show cell clusters with differential abundance (Figure 3F). Both CD8+ naïve T cells and CD8+ CD57+ TEMRA/Effector cells were significantly altered in HCS β-amyloid+ and MCI β-amyloid+ subjects compared to HCS β-amyloid– (Figure 3G).
Thus, we confirmed in a second independent analysis that a preclinical AD stage with early cerebral β-amyloidosis (HCS β-amyloid+) is accompanied by alterations in the peripheral T cell compartment. Importantly, in Cross-sectional II, also in the “MCI due to AD” stage (MCI β-amyloid+) the above-mentioned immune alterations can be observed. Since median β-amyloid load in the MCI β-amyloid+ group of Cross-sectional II was lower than in the HCS β-amyloid+ group of Cross-sectional I (Figure S1I), the extent of β-amyloid pathology is less advanced in these subjects and more comparable to the HCS β-amyloid+ stage of Cross-sectional I. Therefore, in Cross-sectional II, T cell alterations are observable in both β-amyloid+ healthy and MCI subjects.
3.5 β-Amyloid-associated effects on peripheral adaptive immune cells are independent of latent herpesvirus infections
The results of both cross-sectional studies indicate that early β-amyloid accumulation in the brain is characterized by an altered immune profile in the blood, suggesting a link between brain β-amyloidosis and the peripheral immune system. However, certain diagnostic groups could have been biased by subclinical levels of viral infections. In particular, herpesvirus infections have been suggested to be associated with AD pathogenesis,58 and might be a driver of the observed peripheral immune alterations. To study this, we tested for possible associations between our predefined diagnostic groups and serum antibody titers against Herpesviridae such as VZV, HSV-1, HSV-2, EBV, and CMV. We examined IgM as well as IgG titers, to determine both acute infection and long-term immunity, respectively (Figure S4A and S4B). All study participants proved to be above the diagnostic threshold for various Herpesviridae IgG. Of note, 46% of subjects in Cross-sectional I and 40% of subjects in Cross-sectional II were tested positive for anti-CMV IgG, a virus known to affect the proportions of peripheral CD8+ TEMRA cells.59 In contrast, only very few cases with positive IgM titers were observed in both cohorts, indicating low levels of acute infections. Statistical testing revealed no significant differences in antiviral IgG serum titers between diagnostic groups, indicating that Herpesviridae infections did not preferentially occur in a specific cognitive or β-amyloid state (Figure 4A and 4F).
We next checked for specific influences of Herpesviridae on CD8+ T cell subpopulations by correlating relative cluster abundances of FlowSOM-derived subclusters with serum antibody titers of the four quantitatively assessed Herpesviridae (VZV, CMV, HSV-1, HSV-2). In both cross-sectional studies, none of the CD8+ T cell subclusters were substantially affected by increasing IgG titers against VZV, HSV-1, and HSV-2 (Figure S4C and S4D, Spearman correlation: −0.5 < rho < 0.5). However, CD8+ T cell subclusters were affected by altered anti-CMV IgG titers mainly in Cross-sectional I, less so in Cross-sectional II (Figure 4B and 4G). Specifically, CD8+ EM T cells were negatively correlated with increasing anti-CMV IgG levels (rho ≤ −0.5), whereas CD8+ TEMRA/Effector and NK T cells were positively correlated (rho ≥ 0.5) (Figure 4B). No differences in anti-CMV IgG titers have been found in cerebral β-amyloid groups and APOE genotypes (Figure S4E and S4F).
As expected, we observed that individuals who had been tested positive for anti-CMV IgG (CMV+) exhibited a significantly higher relative abundance of CD8+ TEMRA/Effector cells compared to CMV– subjects (Figure 4C and H). As this was a cell population, we identified as characteristic signature cluster associated with preclinical β-amyloidosis, we adjusted the relative abundances of our CD8+ T cell clusters of interest for anti-CMV IgG titer levels using linear regression to control for the effect of CMV infection. Statistical testing for the adjusted residuals revealed that anti-CMV IgG titer-adjusted CD8+ TEMRA/Effector cells were still significantly elevated in HCS β-amyloid+ individuals compared to HCS β-amyloid– subjects in both Cross-sectional I and II (Figure 4D and I). Additionally, in Cross-sectional II, CD8+ TEMRA/Effector cells were still increased in MCI β-amyloid+ patients compared to HCS β-amyloid– (Figure 4I). Significant decreases in naïve CD8+ T cells also remained largely unchanged after CMV titer adjustment (Figure 4E and J).
Thus, we confirmed that the observed immune cell signature clusters of CD8+ T cells, which characterize a preclinical stage of early β-amyloidosis in cross-sectional setups, are independent of latent, reactivated or acute Herpesviridae background infections. No significant association was found between herpesvirus IgG titers and CD4+ T cell clusters from Cross-sectional I.
3.6 Peripheral immune profiling for AD biomarker-accumulating subjects: A longitudinal study design
After identifying an immune pattern associated with early β-amyloid deposition in the brain of cognitively healthy subjects, we asked whether these changes were also related to AD biomarker accumulation over time. For that reason, we investigated a cohort of 59 elderly, cognitively unimpaired subjects who were followed over a time period of 3 years. For each individual we assessed the cerebral β-amyloid load via two [11C]-PiB-PET scans at baseline and after 3 years at the end of the study and collected blood for PBMC isolation at Baseline, after 1.5 years (termed Follow-up1) and after 3 years (termed Follow-up2) (Figure 5A). Plasma AD biomarkers such as p-tau181 were assessed at Baseline and Follow-up2.
We analyzed PBMCs of all available time points (Table S3) from all 59 study participants with improved versions of CyTOF antibody panels for the characterization of T cells. After CyTOF acquisition, datasets were clustered and visualized in tSNE/FlowSOM maps (Figure 5B). Single marker signal intensities per cluster were used to identify the major clusters of human PBMCs (Figure 5C). Since we were interested in understanding the association of AD pathology, particularly cerebral β-amyloid accumulation and plasma p-tau181 increase, with peripheral immune cell signatures, we divided subjects into groups of individuals whose cerebral β-amyloid or p-tau181 levels remained largely stable over time (termed Non-accumulators) and those whose levels increased over time (termed Accumulators). For β-amyloid analysis, based on the change in cortical PiB SUVR over time (ΔPiB SUVR = PiB SUVRFollow-up2 – PiB SUVRBaseline), we split all ΔPiB SUVR values ≥ 0 by median to categorize individuals with ΔPiB SUVR ≥ 0.04 as β-amyloid Accumulators (n = 24) and individuals with ΔPiB SUVR < 0.04 as β-amyloid Non-accumulators (n = 35) (Figure 5D and E). Median ΔPiB SUVR for Accumulators (ΔPiB SUVR = 0.082, IQR = 0.078) was significantly higher compared to Non-accumulators (ΔPiB SUVR = 0.003, IQR = 0.028) (Figure 5F). Therefore, this approach allowed us to identify only Accumulators with clearly increasing PiB retention over time compared to Non-accumulators. Correlation of ΔPiB SUVR with each individual's age at Baseline revealed no strong association over time (Figure S5A, rho = 0.2287, p = 0.0814). As expected, APOE ε4 carriers had a higher median ΔPiB SUVR compared to non-carriers (Figure 5G and S5B). Out of 35 Non-accumulators, 2 subjects were β-amyloid+ at Baseline, whereas out of 24 Accumulators, 14 subjects were β-amyloid+ at Baseline (Figure S5B). Moreover, we observed a significant correlation between high Baseline PiB SUVR and increased ΔPiB SUVR over time (Figure 5H, rho = 0.4275, p = 0.0007).
Alternatively, subjects were categorized according to the change of plasma p-tau181 levels over time (Δp-tau181). Plasma p-tau181 levels were split by median at Δp-tau181 = 0.13pg/mL, to distinguish between p-tau181 Accumulators and Non-accumulators (Figure 5I). Plasma Δp-tau181 was significantly higher in p-tau181 Accumulators and APOE ε4 carriers (Figure 5J and 5K). Higher Baseline p-tau181 levels were not overly associated with increasing Δp-tau181 (Figure 5L, rho = 0.2489, p = 0.0596). In contrast, age at Baseline was significantly correlated with higher plasma Δp-tau181 (Figure S5C, rho = 0.4881, p = 0.0001).
3.7 Longitudinal analysis indicates CD8+ T cell alterations in AD biomarker accumulators
CD3+ CD4− CD8+ T cells were subclustered via FlowSOM (Figure S6A). Without multiple comparison analysis, we increased the number of possible subclusters for the FlowSOM analysis. Those changes led to the identification of 21 CD8+ T cell subpopulations (Figure S6B). All clusters were categorized according to their major T cell subpopulations, as previously described.
In a linear mixed model (LMM), we aimed to explore how AD biomarker response variables such as “Cortical β-amyloid PET” and “Plasma p-tau181 levels” were associated with covariates such as “Relative cell cluster abundance at each time point considered” (= relAbundance), “Group” (Accumulators versus Non-accumulators), and “Age” (Formula S1).60 As a readout, we used an interaction term describing the combined effect of changes in relative abundance of immune cell clusters and assignment to the Accumulator group (= relAbundance:Accumulator). Using this interaction term, we investigated whether the individual AD biomarker increase over time was significantly associated with changes in relative cell cluster abundance of Accumulators, or only with natural aging (Figure 6A).
When focusing on the response variable “Cortical β-amyloid PET”, we found significantly associated interaction terms for CD8+ T cell clusters 16, 17, and 18, all belonging to the subcluster category of TEMRA/Effector cells with CD27– marker expression. The most significant association was observed for CD57+ TEMRA/Effector cells (CD8+ T cell cluster 17, relAbundance:Accumulator p value = 0.01480, Figure 6B-6D). In line with our previous findings, significant associations were also found for naïve CD8+ T cells (CD8+ T cell cluster 1 and 2, Figure 6B-6D).
For the response variable “Plasma p-tau181 levels”, we found significant associations for CD8+ T cell clusters 6, 9, and 10, which all belong to PD-1+, potentially exhausted EM T cells (Figure 6E-6G). These associations were consistent with the results of Cross-sectional I (Figure 2K). Clusters 6, 9, and 10 were further characterized by low CD127 and high KLRG1 expression. CD57 expression was detected for clusters 9 and 10 (Figure S6B).
In summary, the longitudinal cohort analysis also confirmed adaptive immune cell signatures that are associated with the accumulation of AD biomarkers such as brain β-amyloid deposition and increase in plasma p-tau181 levels. Importantly, these immune changes were observed during an early, preclinical stage of AD, when cognitive impairment is usually not yet present. The majority of study participants remained cognitively healthy, with the exception of two individuals who converted to MCI.
4 DISCUSSION
This work demonstrates that pathological alterations in AD biomarkers such as brain β-amyloid deposition revealed by amyloid PET imaging and increases in plasma p-tau181 levels during early stages of AD pathology are associated with systemic changes within the T cell compartments of the adaptive immune system. The main strengths of our study are (I) the combined association of peripheral T cell alterations with AD biomarkers and cognitive status, (II) the consideration of both cross-sectional differences and longitudinal accumulation of cerebral β-amyloid and plasma p-tau181, and (III) the detailed analysis of latent viral background infections.
In cross-sectional analysis setups, T cells showed comparable trends in cognitively healthy β-amyloid+ individuals. We observed reduced relative abundances of naïve T cells and increased relative abundances of differentiated effector subtypes such as highly reactive CD8+ CD57+ TEMRA cells. These observations suggest an adaptive immune activation event with a shift from non-activated, naïve T cells to cells that have undergone initial antigenic stimulation and are in a stage of replicative senescence or present as reactivatable memory cells. The general consensus of many previous studies has been that functional non-naïve T cell subpopulations such as memory T cell phenotypes,61, 62, 26 activated CD8+ HLA-DR+ T cells,63 or CD8+ TEMRA cells27 are increased in the blood of AD patients. However, previous studies mostly lack a detailed analysis of β-amyloid status and/or plasma AD biomarkers. Our results add to the existing body of evidence the important information that adaptive immune changes are in fact associated with cerebral β-amyloid status and its change over time, rather than to clinical diagnosis alone. For instance, the increased number of CD8+ TEMRA cells in AD has previously been interpreted as specific to the cognitive states of MCI and AD.27 However, we show that elevated numbers of CD8+ TEMRA cells can be observed as early as in the stage of preclinical AD (HCS β-amyloid+), in which cognitively healthy subjects already present with substantial cerebral β-amyloid depositions and altered plasma AD biomarker levels. Therefore, it is essential to analyze cohorts across the entire AD continuum. As for other neurodegenerative diseases, specific T cell responses have also been associated with amyotrophic lateral sclerosis (ALS)64 and early, preclinical stages of Parkinson's disease.65 Importantly, our findings on CD8+ TEMRA/Effector cells can be reproduced in a setup, in which diagnostic groups are assigned based on cognitive state and levels of plasma AD biomarker p-tau181. Abnormalities in p-tau181 levels are specific for AD-related neuropathological change and not affected by potential co-morbidities such as Lewy body disease (LBD) or limbic age-related TDP-43 encephalopathy.66 It should also be noted that previous studies that found only minimal67 or no changes68 in peripheral immune cell populations in β-amyloid+ subjects used smaller cohort sizes and conventional flow cytometry with antibody panels unable to delineate implicated cell subtypes in detail.
Our cross-sectional results suggest that cerebral β-amyloid accumulation is associated with specific CD8+ T cell alterations. However, such changes could also be caused by latent reactivated viral infections, as is suspected in AD.58 Correlative studies have shown increased DNA content from Herpesviridae such as HSV-1 in post-mortem brains of AD patients, specifically within β-amyloid plaques.69-71 Reactivation or recent infections with other Herpesviridae such as HSV-2, CMV, and EBV have also been associated with AD development to a lesser extent.72-74 A different hypothesis postulates that Aβ serves as an antimicrobial peptide and aggregates as part of an innate immune response in order to provide an antiviral, protective function.75 Here, we found that preclinical β-amyloidosis-related alterations of CD8+ T cells are independent of latent, reactivated or acute Herpesviridae background infections. Importantly, in contrast to studies mentioned before, we did not measure brain or cerebrospinal fluid (CSF) virus load but serum antibody titers, since we were interested in the peripheral effect of anti-viral immune reactions.
The longitudinal cohort analysis suggests again that subtypes of naïve adaptive immune cell subpopulations and CD8+ TEMRA/Effector cells are associated with cerebral β-amyloid accumulation in cognitively healthy subjects. As expected, we found CD27– CD45RA+ TEMRA/Effector cells showing CD57 expression across the full spectrum, including terminally differentiated CD57+, as in the cross-sectional analysis, and higher proliferative CD57– subtypes.57 In line with cross-sectional results, we also found an association of PD-1+ EM cells with increasing plasma p-tau181 levels. Those cells are potentially exhausted due to persistent antigenic stimulation and upregulate inhibitory receptors including Programmed Death-1 (PD-1 or CD279).76 In Cross-sectional I, we observed a reduction of these PD-1+ EM in HCS β-amyloid+ compared to HCS β-amyloid–, probably indicating a shift of abundances from exhausted cell types with impaired effector functions to highly antigen-specific TEMRA/Effector cells. Another possibility would be that PD-1+ EM represent a putative precursor for TEMRA/Effector cells, since their interleukin-7 (IL-7) receptor (CD127) expression is already reduced, while some subtypes upregulate KLRG1 and CD57 expression.
Since major immune changes were observed in a preclinical stage of AD, the adaptive immune system might respond directly or indirectly to an initial change in AD biomarkers. Possible epitope targets of the identified T cells are still subject to speculation at the moment and can only be revealed by targeted epitope mapping for clonally expanded cells. As the immune system is exposed to autologous Aβ peptide long before its deposition in β-amyloid plaques, it is very likely that Aβ-specific T cells are negatively selected in the thymus during early development, or are subject to peripheral tolerance. Therefore, the native Aβ1-42 peptide itself is unlikely to elicit such an (auto-)immune response, since several tolerance barriers would have to be overcome. More recently, Chen and colleagues reported the involvement of activated CD8+ effector T cells in tauopathy mouse models, which might drive brain atrophy in response to tau, suggesting that T cells may recognize tau itself.77 It is also important to mention that CD8+ cytotoxic T cells are not the subtype that is expected to respond to mainly extracellular threats such as Β-amyloid depositions. However, there are ways how such an adaptive immune response could be initiated. For example, expanded T cell clones discovered in some neurodegenerative/neurological disorders showed specificity against viral antigens.78, 79 These T cell responses could trigger bystander activations of innate antigen-presenting cells and adaptive immune responses on the one hand. On the other hand, infectious diseases might also be the trigger for the development of an autoreactive immune response through molecular mimicry between peptides derived from infectious agents and self-peptides. Such antigen cross-recognition does not require perfect homology of epitope sequences, but may occur when peptides share common amino acid motifs. Also, the concept of T cell neo-epitopes, against which there is tolerance in only a few cases, has to be further explored as potential source of TEMRA cell activation.12
5 CONCLUSION
We conclude that changes within the peripheral blood T cell compartments precede the clinical diagnosis of AD, are associated with brain β-amyloidosis and/or plasma AD biomarkers, and are independent of latent viral infections. Future research will be directed toward confirmatory longitudinal studies with larger cohort size and longer follow-up. This information could be used to further investigate whether immune cell changes might even precede established AD plasma biomarkers, as they can respond to low peptide antigen concentrations, whereas certain peptide levels above the detection limit are required for successful plasma analysis. Early AD-associated immune cell changes might provide guidance and prompt more expensive downstream analysis techniques if a population at risk is found. It has to be investigated whether the observed immune cell changes consistently disappear with progressing AD pathology as observed for the MCI β-amyloid+ subjects of Cross-sectional I who had the highest cerebral β-amyloid load. It might be that due to constant exposure to an antigenic stimulus, which drains from the preclinical AD brain into the periphery via cerebral lymphatics, an initial immune response is abrogated by mounting tolerance mechanisms or immunological exhaustion. In addition, it remains to be elucidated whether these initial immune responses are indeed protective against toxic protein aggregates in the brain or other antigenic entities related to AD, or whether they even exacerbate the pathology. In fact, in aging studies, cognitively healthy, potentially resilient centenarians seemed to have higher numbers of effector T cells in peripheral blood compared to younger control subjects.80, 81 In contrast, many studies have shown a lack of proper immune responses in mouse models of β-amyloidosis and AD.82-84 Thus, it is perhaps desirable to have less exhausted immune cells such as PD-1+ EM cells and more antigen-specific effector T cells. To further determine the importance of CD8+ CD57+ TEMRA/Effector cells in AD pathology, the presence of this cell type in AD brain tissue and long-term studies would need to be confirmed.
In summary, we propose that cellular immune patterns in the periphery could be used as a proxy for early AD biomarker changes. A better understanding of these immune patterns in combination with detailed epitope specificity analysis might pave the way for the identification of new composite biomarker candidates and complement the diagnostic toolbox for the detection of AD, in particular preclinical AD. Moreover, understanding the pathophysiological significance of these processes may give rise to new therapeutic strategies.
AUTHOR CONTRIBUTIONS
Conceptualization, C.G., T.K., L.K., V. Treyer, M.T.F., and A.G.; Methodology, C.G., T.K., and V. Tosevski; Software, R.F., C.R., and V. Tosevski; Formal Analysis, C.G., T.K., and R.F.; Investigation, C.G., T.K., A.M., and C.R.; Resources, L.K., V. Treyer, and A.G.; Data Curation, R.F.; Writing—Original Draft, C.G. and T.K.; Writing—Review & Editing, C.G., T.K., R.F., L.K., C.H., V. Treyer, M.T.F., and A.G.; Visualization, A.M.; Supervision and Project Administration, C.H. and R.M.N.; Funding Acquisition, C.G., L.K., C.H., R.M.N., M.T.F. and A.G.
ACKNOWLEDGMENTS
The authors thank all the volunteers who kindly participated in the studies. The authors thank all study physicians and neuropsychologists for their contribution to the assessments and conduct of the study. Samples from participants were collected at the Center for Prevention and Dementia Therapy (Institute for Regenerative Medicine, University of Zurich) by study nurses led by Esmeralda Gruber. The authors would like to thank the Cytometry Facility (University of Zurich) for technical assistance, and the Institute of Medical Virology (University of Zurich) for viral serology analysis.
This work was supported by institutional funding of the University of Zurich, grants from the Synapsis Foundation – Dementia Research Switzerland (No. 2015-CDA01 to M.T.F. and No. 2019-PI06 to R.M.N., C.G., and A.G.), the Swiss National Science Foundation (SNF 33CM30-124111, SNF 320030-125387/1 to C.H.), the Mäxi Foundation (to C.H.) and the Velux Foundation (Grant Project 993 to L.K.).
Open access funding provided by Universitat Zurich.
CONFLICT OF INTEREST STATEMENT
T.K. is currently an employee of Charles River Associates, Switzerland; L.K. is an employee of Roche, Switzerland; C.H. and R.M.N. are employees and shareholders of Neurimmune AG, Switzerland; M.T.F. is co-founder and CSO of the Women's Brain Project; M.T.F. has received in the past 2 years personal fees from Roche, Lundbeck, GW, for activities unrelated to this paper. Author disclosures are available in the supporting information.
CONSENT STATEMENT
All study participants gave written informed consent. All studies and further use for the current analyses were approved by the local ethics committee (Kantonale Ethikkommission Zürich) and conducted in accordance with their guidelines and the Declaration of Helsinki.
