Mass spectrometric simultaneous quantification of tau species in plasma shows differential associations with amyloid and tau pathologies

1The detection of amyloid-β (Aβ) and tau pathologies by cerebrospinal fluid (CSF) and positron emission tomography (PET) biomarkers has enabled an in vivo biological diagnosis of Alzheimer’s disease (AD). Recently, the availability of ultrasensitive technologies has led to the development of assays capable of measuring the levels of Aβ, tau and neurodegeneration biomarkers in blood. In particular, phosphorylated tau variants (p-tau181 (refs.,), p-tau217 (refs.,) and p-tau231 (ref. )) have shown high diagnostic performance in differentiating AD from other neurodegenerative disorders and have been validated against neuropathology–.

2With the emergence of several immunotherapies that efficiently remove Aβ aggregates from the brain, the need for blood biomarkers to facilitate participant recruitment and monitor disease progression in clinical trials is even more pressing. Although a single biomarker (for example, p-tau231 or p-tau217) may work well to identify AD, a biomarker panel may provide further information on disease stage and treatment effects. Moreover, currently available blood p-tau biomarkers appear to have different associations with AD pathology, but a direct comparison of different p-tau variants is hindered by the large heterogeneity in p-tau immunoassays tested in such studies–. Therefore, a method that systematically measures multiple p-tau and tau species in a single-shot analysis, not dependent on the platform, may give greater insight into stage-specific changes that are critical to monitor pathology changes in drug response and disease progression in clinical management.

3Furthermore, studies have shown that tau does not exist as a full-length protein in CSF and blood but, rather, in the form of fragments that contain a mix of N-terminal and mid-region tau–. Because tau fragments containing specific phosphorylations have different lengths, quatification using immunoassays is limited by the inevitable requirement of the two epitopes being present on the targeted p-tau fragment. This has been shown to influence the quantification of a specific phosphorylation impacting their biological interpretation—for example, in CSF, N-terminal p-tau181 immunoassays report earlier increases in the AD continuum than the mid-region p-tau181 (ref. ). Mass spectrometry (MS) techniques allow simultaneous quantification of several epitopes and enable the detection of a specific p-tau to a broader range of fragments (Extended Data Fig. 1).

4The main aim of the current study was to simultaneously quantify the plasma concentrations of six different phosphorylated tau (p-tau181, p-tau199, p-tau202, p-tau205, p-tau217 and p-tau231) and two non-phosphorylated tau peptides using a targeted MS method to investigate their relationship with AD pathology at different stages of disease development. For this, we assessed each biomarkerʼs performance in identifying Aβ pathology using samples from a memory clinic setting (n = 157). Then, the biomarkers were evaluated in a well-characterized research cohort with amyloid and tau PET (n = 57).

5We first used our method to measure samples from a memory clinic population (cohort 1; Supplementary Table 1). Participants were grouped according to clinical diagnosis and CSF biomarker profile in cognitively unimpaired (CU), mild cognitive impairment (MCI), AD dementia, MCI not due to AD and other dementias (Extended Data Fig. 2). We found that p-tau199, p-tau202 and non-phosphorylated tau (tau195–205 and tau212–221) showed no significant differences between AD clinical stages neither in non-AD individuals. Plasma p-tau181, p-tau205, p-tau217 and p-tau231 had significant increases along the AD continuum, being highest in AD dementia except for p-tau181. All p-tau biomarkers were unaltered in MCI and dementia groups without AD pathology. These results were corroborated in cohort 2 (Supplementary Table 2) when participants were classified by cognitive status and amyloid PET positivity (Fig. 1 and Extended Data Fig. 3). Plasma p-tau205, p-tau217 and p-tau231 presented similar mean fold changes (FCs) from CU Aβ-negative (CU−) to CU Aβ-positive (CU+) individuals (FCp-tau205 = 1.97, FCp-tau217 = 2.05 and FCp-tau231 = 2.05). However, in later stages of the disease, p-tau231 levels were not significantly higher in cognitively impaired Aβ-positive (CI+) compared to CU+ (FCp-tau231 = 1.5), whereas p-tau205 and p-tau217 presented more dynamic differences (FCp-tau205 = 2.26 and FCp-tau217 = 2.2). Plasma p-tau181 concentrations were not significantly higher in CU+ compared to CU− (FCp-tau181 = 1.24) or CI+ to CU+ (FCp-tau181 = 1.14). In all cases, biomarker levels were not significantly different in cognitively impaired Aβ-negative (CI−) or non-AD neurodegenerative diseases (progressive supranuclear palsy (PSP) and frontotemporal dementia (FTD)) compared to CU− (Supplementary Table 3).

6Plasma p-tau205, p-tau217 and p-tau231 were the biomarkers with higher correlations with Aβ PET global standardized uptake value ratio (SUVR) (p-tau205, r = 0.52, P < 0.001; p-tau217, r = 0.70, P < 0.001; and p-tau231, r = 0.60, P < 0.001), followed by p-tau181 (r = 0.42, P = 0.002) (Supplementary Table 4). When stratifying amyloid PET into quartile (Q) groups, although all three phosphos were statistically similar, the same trend as before was observed with disease stage. Plasma p-tau205, p-tau217 and p-tau231 presented higher levels in Q3 of amyloid PET uptake compared to Q1 (FCp-tau205 = 2.24, FCp-tau217 = 2.1 and FCp-tau231 = 1.90), but p-tau205 and p-tau217 showed higher concentrations in Q4 (FCp-tau205 = 5.43 and FCp-tau217 = 5.36) than p-tau231 (FCp-tau231 = 3.65) (Fig. 2a and Extended Data Fig. 4). Changes in the levels of p-tau species in relation to brain amyloid accumulation also suggested that p-tau181 and p-tau231 increase earlier than p-tau205 and p-tau217, although they all changed close to amyloid load PET uptake abnormality (centiloid < 24) (Fig. 2b). Plasma p-tau217 and p-tau205 showed higher mean FCs with amyloid deposition than p-tau231 and p-tau181. Voxel-based analysis indicated an overall topographic overlap of the brain regions in which plasma biomarkers are associated with [18F]AZD4694 retention, mostly across the posterior cingulate, precuneus, prefrontal and temporal cortices. However, in this dataset, p-tau217 and p-tau205 showed wider voxel associations with larger t values, particularly when compared to p-tau181 (Fig. 2c).

7Correlations of plasma tau peptides with tau PET uptake indicated that p-tau205, p-tau217 and p-tau231 were the site-specific phosphorylations with higher associations (p-tau205, r = 0.49, P < 0.001; p-tau217, r = 0.58, P < 0.001; p-tau231, r = 0.43, P < 0.001) but not significant for p-tau181, p-tau199 and p-tau202 (Supplementary Table 4). Plasma p-tau217 and p-tau231 showed the first significant differences in Braak III–IV (P = 0.047 and P = 0.050, respectively), whereas, for p-tau205, the difference was significant only at Braak V–VI (P < 0.001). Interestingly, p-tau205 and p-tau217 levels were significanly higher in Braak V–VI compared to Braak III–IV (P = 0.012 and P = 0.008, respectively) but not for p-tau231 (P = 0.79) or other p-tau species (Fig. 3a and Extended Data Fig. 6). Visual inspection of locally estimated scatterplot smoothing (LOESS) analyses highlighted plasma p-tau217 as the biomarker with the steepest slope in relation to tau pathology stage and p-tau205 as the latest one to become abnormal, between Braak II and Braak III (Fig. 3b). Voxel-wise analyses showed associations between plasma biomarkers and [18F]MK6240 uptake in the superior, middle and inferior temporal gyri as well as in the entorhinal cortex and hippocampus bilaterally, which were particularly evident for p-tau205 and p-tau217 (Fig. 3c).

8To characterize the contribution of plasma tau species to both amyloid and tau signals, we performed regression models using amyloid PET (A: indexed by global amyloid PET SUVR), tau PET (T: indexed by global tau PET SUVR) or both amyloid and tau (A+T) (Fig. 4). Plasma p-tau205, p-tau217 and p-tau231 were the phosphorylation sites that better associated with the imaging findings. For plasma p-tau231, the lowest Akaike information criterion (AIC) values were observed in the A and A+T models (AICA = 6.94 and AICA+T = 6.66), with similar regression coefficients (R2A = 0.55 and R2A+T = 0.57). This suggests that amyloid PET SUVR (A) is the model that better explains plasma p-tau231 for being the simplest. Plasma p-tau217 was mainly mediated by A+T (R2 = 0.702, AIC = −5). Plasma p-tau205 presented the same AIC for T and A+T (AICT = 34.8 and AICA+T = 34.8), with similar regression values (R2T = 0.457 versus R2A+T = 0.469), indicating that T is the main mediator of p-tau205. Finally, we determined the ability of the tau species to identify amyloid and tau pathologies (Supplementary Table 5). Plasma p-tau217 presented the highest accuracy distinguishing amyloid PET-positive from PET-negative individuals (area under the curve (AUC) = 0.85 (95% confidence interval (CI): 0.74–0.95)), further increased when using the phospho/non-phospho ratio p-tau217/212–221 (AUC = 0.94 (95% CI: 0.87–1)). Plasma p-tau217 and p-tau205 showed similar performances to discriminate Braak I–IV versus Braak V–VI (AUC = 0.81 (95% CI: 0.59–1) and AUC = 0.81 (95% CI: 0.58–1), respectively), improved when using the ratio p-tau217/tau212–221 (AUC = 0.96 (95% CI: 0.63–1), DeLongʼs test P = 0.04) but not significantly superior for the ratio p-tau205/tau195–209 (AUC = 0.84 (95% CI: 0.63–1), DeLongʼs test P = 0.12).

9Plasma p-tau biomarkers have shown high value in detecting AD in patients with cognitive complaints and predicting the future development of AD. Thus, because of their additional value of being less invasive and cost-effective, they are now being used as recruitment and treatment monitoring tools in clinical trials for anti-Aβ therapies and will soon be implemented for diagnostic purposes in clinical practice. In this study, by using an MS method to simultaneously quantify six different phosphorylated and two non-phosphorylated tau peptides in plasma, we show that not all tau phosphorylations detected in blood reflect the same brain pathological changes and evolve in the same direction with proxies of disease progression. Our results support a previous study in autosomal dominant AD using an MS method, which found that hyperphosphorylation at specific sites of the tau protein is a dynamic process with predictable progression in the phosphorylation pattern when measured in CSF. Here we demonstrate that this dynamic process can also be detected in plasma across the full clinical spectrum of sporadic AD.

10Our results corroborate previous observations that p-tau231 presents an early change in preclinical AD, emerging before amyloid PET abnormality,. Furthermore, we observed that p-tau231 levels are not further increased in individuals with more advanced symptomology, which verifies immunoassay findings. This could be explained by the strong association of p-tau231 with amyloid pathology, increasing with the early accumulation of Aβ but plateauing in later stages when amyloid deposition stabilizes. Plasma p-tau181 showed similar trends and emergence as p-tau231, but the mean FCs of this biomarker were more subtle, showing weaker and non-significant associations with amyloid and tau pathologies, respectively. Several studies comparing different plasma p-tau assays have reported p-tau181 having lower accuracies in detecting AD compared to p-tau217 and p-tau231 (refs.,,,). Another possible explanation might be a limitation of the MS technique. In particular, the phosphorylation at position Thr181 prevents digestion by trypsin between residues Lys180 and Thr181, which allows the detection of the phosphorylated peptide 175–190 but not the non-phosphorylated version of the exact same peptide. This was also observed previously by us and others in CSF,. It is possible that trypsin cleaves a small percentage of the phosphorylated peptide, which would be undetected and lead to an underestimation of this phosphorylation. In addition, we previously showed that an endogenous tau peptide (tau 175–190) is present in CSF, in both phosphorylated and non-phosphorylated forms, but does not change in AD. It is likely that this endogenous peptide is also secreted to plasma and, because it is the same as the one generated by trypsin, is, thus, possible to affect results.

11In contrast, p-tau217 has been shown as the p-tau that exhibits larger FCs in symptomatic AD phases in CSF and plasma,,. We corroborate that plasma p-tau217 presents continuous and higher increases along the AD continuum and better associates with amyloid and tau pathologies determined by PET,,. By simultaneously measuring different p-tau species, we observed that p-tau217 performs superiorly to other p-tau species identifying both pathologies. Furthermore, our results support previous findings suggesting that plasma p-tau217 is a dynamic biomarker associated with amyloid pathology in early stages and with tau burden later, specifically reflecting the progression of ADʼs main biological features.

12To our knowledge, this is the first study reporting the quantification of p-tau205 in blood. Although its accuracy detecting amyloid pathology was inferior to p-tau217 and p-tau231, this epitope (1) performed similarly to p-tau217 in identifying advanced tau pathology, (2) was the only p-tau biomarker that was better explained by the tau PET signal and (3) presented the latest changes to abnormality in relation to Braak staging. Postmortem staging of neurofibrillary tangles (NFTs) has traditionally been performed by immunostaining using the AT8 antibody, which reacts against hyperphosphorylated tau at positions 202/205 (ref. ), making the in vivo quantification of this biomarker an important tool for examining neuropathological associations. In addition, findings in autosomal dominant AD suggest that CSF p-tau205 levels increase with the beginning of neuronal dysfunction, years later than the rise in p-tau217 (ref. ). Altogether, these results point to p-tau205 as a late biomarker possibly linked to tau burden. Considering the recent success of the TRAILBLAZER-2 donanemab trial based on recruiting participants with intermediate tau PET burden, we postulate that the window between p-tau217 and p-tau205 could be used to better identify this group.

13Finally, plasma p-tau199, p-tau202 and the non-phosphorylated tau species tau195–209 and tau212–221 did not change significantly along the AD continuum. No previous studies have assessed the levels of p-tau199 in blood, and plasma p-tau202 concentrations have shown no correlations with CSF measures. Assays targeting non-phosphorylated versions of tau—or total-tau (t-tau)—although very successful in CSF, have shown limited accuracy in blood. This is probably due to the contribution of peripheral tau to the quantification masking the measurement of brain-specific tau, which has been estimated to account for approximately 20% of the total tau protein detected in blood. Despite this, the ratio phosphorylated/non-phosphorylated peptides for p-tau217, as shown here and previously,, performs better identifying AD pathology than the peptide alone. However, in this work, the ratio was not significantly better for p-tau205. Future studies should address the biological significance of these ratios considering the fragmentation pattern of tau in blood and the contribution of central nervous and peripheral nervous system tau to these quantifications, as well as to which degree the benefit of normalization by using such ratios is due to minimizing methodological variability.

14A few methodological factors should be considered on the interpretations of our results. Although this study introduces a method for the simultaneous quantification of different plasma tau species, the number of samples with neuroimaging data available was limited. Furthermore, assessment of tau pathology in the brain using PET quantifies aggregated tau in NFTs, whereas the tau pool detected in plasma is composed of soluble forms of the protein, which could be a limitation when establishing correlations. Differences in the information provided by fluid biomarkers and imaging tracers could have also affected the relationship of the plasma tau species with amyloid pathology. Amyloid PET uptake becomes abnormal later than fluid Aβ,, and close to fluid p-tau,, explaining why plasma tau biomarkers showed similar trends in early amyloid PET uptake. In addition, the analysis of the accuracy of the ratio phosphorylated/non-phosphorylated peptides was limited to p-tau217 and p-tau205, because our panel lacked the non-phosphorylated versions of p-tau181 and p-tau231 due to technical caveats discussed before. Finally, larger studies are needed to confirm the differential associations of these biomarkers with amyloid and tau pathologies, and longitudinal studies should address their prognostic value over clinical progression as well as their evolution during the disease course.

15In conclusion, we have developed a MS method to simultaneously quantify six different phosphorylated (p-tau181, p-tau199, p-tau202, p-tau205, p-tau217 and p-tau231) and two non-phosphorylated (tau195–209 and tau212–221) tau peptides in plasma. Our results indicate that p-tau217, p-tau231 and p-tau205 are the plasma tau forms that best reflect AD-related brain pathological changes although with different emergence along the AD continuum and associations with amyloid and tau pathologies. Plasma p-tau231 was found to be the earliest p-tau biomarker; p-tau217 showed the highest FCs and diagnostic performance; and p-tau205 was observed to become abnormal in later stages of the disease. A comprehensive understanding of the pathological information that each blood p-tau reflects is paramount to decide which biomarker to use in each stage of the disease and to guarantee a correct read-out in clinical trials for anti-Aβ and anti-tau therapies.

16We evaluated the performance of the plasma tau MS method by analyzing plasma from 24 participants in the Gothenburg H70 Birth Cohort Studies (12 AD and 12 controls). Participants were selected and grouped according to CSF biomarker values (Aβ42/40 ratio < 0.62 and ptau > 60 pg ml−1).

17A total of 157 patients from the Cognitive Neurology Center, Lariboisière Fernand Widal Hospital, Université Paris Cité, was included in the study. Patients went through a comprehensive neurological examination, neuropsychological assessment and CSF and plasma biomarker analysis. Cognitive follow-up data were acquired for an average time of 6 months. Samples from patients were grouped according to the clinical diagnosis at the memory clinic and biological CSF marker profile: controls (n = 23), AD–MCI (n = 24), AD (n = 27), non-AD MCI (n = 50) and other dementias (n = 33).

18Participants from the Translational Biomarkers of Aging and Dementia (TRIAD) cohort who had amyloid and tau PET imaging as well as plasma volume available at the time of the MS experiments were included in the study (n = 57). The TRIAD cohort is well characterized in terms of biomarker and clinical data and contains participants ranging from CU young (<30 years of age) and older adult (>50 years of age) individuals to patients with MCI and AD dementia. CU participants had a Mini-Mental State Examination (MMSE) score >24 and a Clinical Dementia Rating (CDR) score of 0. MCI participants had a CDR score of 0.5—subjective and objective impairments in cognition but preserved activities of daily living. Patients with AD dementia had a CDR score ≥0.5 and met the National Institute on Aging and Alzheimerʼs Association criteria for AD determined by diagnostic,. For the purposes of this study, CU, MCI and AD dementia participants were grouped as CU (young and older adults) and CI (MCI and dementia) as well as according to their Aβ status (based on Aβ PET visual rating) as Aβ-positive (+) or Aβ-negative (−). The resulting clinical and biomarker-defined groups were: 18 CU−, 8 CU+, 18 CI+ and 7 CI−.

19All participants provided written informed consent, and all three studies were approved by their regional ethics committee. REB approval for TRIAD IUSMD-16-60 2021 (Centre intégré universitaire de santé et de services sociaux (CIUSSS) de l'Ouest-de-l'Île-de-Montréal–Mental Health and Neuroscience). The Paris Lariboisière cohort was approved by the ethic committee of Bichat University, Paris, France (CEERB GHU Nord n°10-037).

20MS detection of phosphorylated and non-phosphorylated tau peptides was performed by adapting a previous method for CSF to blood. In this case, EDTA plasma samples (1 ml) were thawed, vortexed for 30 s at 2,000 r.p.m. and spun down for 10 min at 4,000g. Tau protein was extracted by IP using beads (Dynabeads M-280 sheep anti-mouse IgG, Thermo Fisher Scientific, 11202D) cross-linked with a combination of antibodies targeting non-phosphorylated tau: Tau12 (BioLegend, 806501), HT7 (Thermo Fisher Scientific, MN1000) and BT2 (Thermo Fisher Scientific, MN1010). Antibodies were conjugated to the beads at a concentration of 4 µg antibody/50 µl beads. Automated IP was performed using the KingFisher Flex System (Thermo Fisher Scientific). Samples were incubated with the beads for 2 h at room temperature, followed by washes with PBS, PBS 0.05% Triton X-100, PBS and 50 mM ammonium bicarbonate (AMBIC) and elution with 0.5% formic acid. Quality control samples (which were a pool of several plasma samples) and recombinant tau (0.001 µg per sample) were included in each plate as IP control and to monitor intensity signal for normalization purposes. Further tau enrichment was performed by adding perchloric acid (15 µl, 60% v/v) to the samples, which induces precipitation of the vast majority of proteins but not tau. After centrifugation at 3,000g for 30 min at 4 °C, supernatants were transferred to a 96-well SPE plate (Oasis PRiME HLB 96-well µElution plate, 3 mg of sorbent per well; Waters) and desalted. The SPE plate was washed with 2 × 200 µl of 5% methanol (v/v) and eluted into a microtiter plate with 200 µl of 50% acetonitrile and 0.1% trifluoroacetic acid. Then, samples were speed-vac-dried. Tryptic digestion was performed by resuspending the samples with trypsin solution (sequencing grade, Promega) (0.1 µg per sample at a concentration 2.5 µg/ml−1 in 50 mM AMBIC) and incubating overnight at 37 °C. After 18 h, proteolysis was quenched with trifluoroacetic acid (TFA) final concentration 0.1%, and samples were lyophilized and stored at −20 °C.

21For liquid chromatography–mass spectrometry (LC–MS) analysis, samples were resuspended in 50 µl of 0.01% TFA and run in singlicates. LC settings were the same as previously described. MS analysis was performed on a hybrid Orbitrap mass spectrometer (Fusion Tribrid, Thermo Fisher Scientific, for the discovery and memory clinic cohorts and Lumos for TRIAD), fitted with an EasySpray nano-ESI ion source. The mass spectrometer was operated in the positive ion mode, with the following settings for the parallel reaction monitoring (PRM) scan: Activation Type: HCD; Detector Type: Orbitrap; Orbitrap Resolution: 60,000; Scan Range: 250–1,200; RF Lens: 30%; Easy-IC: On; Isolation Type: Quadrupole; and Isolation Window: 0.7 m/z. Maximum Injection Time, Normalized AGC Target, Optimal Collision Energy and FAIMS Voltage were determined experimentally for each peptide. The endogenous tryptic peptides targeted in this study are shown in Supplementary Table 6. Heavy-labeled AQUA peptide standards were prepared in a mix with adjusted concentrations for each peptide and spiked in during the sample preparation (Supplementary Table 7). LC–MS data were acquired using Xcalibur 4.5 and Tune 3.5 software (Thermo Fisher Scientific) and analyzed with Skyline 22.2 software (McCoss Laboratory, University of Washington). The analysis of the plasma samples was performed blinded to any participant information.

22The reproducibility of the method was tested by measuring the same sample repeatedly and at spaced times (Supplementary Table 8). Linearity was assessed by preparing plasma samples of different volumes (250 µl, 500 µl, 750 µl and 1,000 µl) and analyzed with the method (Extended Data Fig. 8). A recovery test was performed to guarantee that beads were not saturated during the IP and there was no tau protein remaining in the supernatant after the sample preparation. The total tau concentration was measured in plasma samples using Quanterix total tau SIMOA Kit (101552) before and after IP (Extended Data Fig. 8). Results from MS showed high significant correlations with SIMOA measurements of p-tau181, p-tau217 and p-tau231 (only available epitopes) in TRIAD samples (Extended Data Fig. 9). The method was validated with a discovery cohort that included plasma from 24 participants (12 AD and 12 controls) in the H70 clinical studies (Institute of Neuroscience and Physiology, University of Gothenburg) (Extended Data Fig. 10). Supplementary Table 9 shows the differences in FCs when the phosphorylated peptides were normalized only with their respective standards or with the heavy standards plus the non-phosphorylated peptides.

23Amyloid and tau imaging were obtained using [18F]AZD4694 (40–70 min after injection) and [18F]MK6240 (90–110 min after injection) PET, respectively,. PET scans were acquired with a Siemens High Resolution Research Tomograph (Siemens Medical Solutions), and imaging data were processed, in conjunction with each individualʼs magnetic resonance imaging, as previously described–, using the cerebellar gray matter and the inferior cerebellar gray matter as reference regions for Aβ and tau PET SUVR calculation, respectively. Amyloid status was based on visual reading of [18F]AZD4694 PET by two neurologists blinded to clinical diagnosis, and the global amyloid SUVR refers to the average SUVR of the precuneus, cingulate, inferior parietal, medial prefrontal, lateral temporal and orbitofrontal cortices. For tau PET, a global index of tau pathology was given by the average SUVR in the meta-ROI region. For the in vivo classification of Braak stages, average tau PET SUVR was estimated for the transentorhinal (Braak I–II), limbic (Braak III–IV) and isocortical (Braak V–VI) cortices, allowing for the determination of a regional positivity status, which was then applied in an ordinal logistic regression to determine one’s Braak stage classification, as previously described.

24The R statistical software package (4.0.0) was used to perform non-imaging statistical analyses. Chi-square or Fisher tests were conducted for categorical variables, and t-tests or analysis of variance (ANOVA) were conducted for continuous variables when appropriate. Correlations between biomarkers were assessed with Spearman rank tests. As mentioned above, participants were grouped according to cognitive and amyloid statuses. In addition, amyloid PET global SUVRs were split into quartiles, which was also used to segregate participants into groups (Q1 = (−Inf, 1.29); Q2 = (1.29, 1.7); Q3 = (1.7, 2.45); Q4 = (2.45, Inf)). Biomarker z-scores and mean FCs were calculated using average values of the CU− as reference. Linear models had the plasma biomarkers as dependent variables and tested the effect of imaging biomarkers, as independent factors, both in isolation and in an additive manner, and had age and sex as covariates. Adjusted R2 values and adjusted AIC were used to compare the fit of these regression models. Biomarker values that were z-scored were plotted against amyloid PET (centiloids) and Braak stage classification, and curves were smoothed using the LOESS method. Finally, voxel-wise regression analyses were conducted on Rminc to evaluate the associations between the different plasma markers and in vivo Aβ PET and tau PET, adjusted for age and sex. Random field theory was applied on the t-parametric maps to correct for multiple comparisons. No a priori sample size calculation was performed. In addition, not all biomarkers were normally distributed, but, although no data transformation was applied, normality of the ANOVA/linear model residuals was visually inspected using quantile plots.

25Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.