Multiplex Mass Spectrometry Analysis of Amyloid Proteins in Human Plasma for Alzheimer’s Disease Diagnosis

Abstract

Direct analysis of amyloid proteins in human plasma will promote rapid screening of brain amyloidosis-the earliest pathological signature of Alzheimer’s disease. We developed a microflow liquid chromatography-targeted mass spectrometry assay for quantitation of 4 intact β-amyloid proteins starting from 1 milliliter of human plasma samples. This method showed 90% accuracy for predicting brain amyloid using plasma Aβ42/Aβ40 values from 36 cognitively normal individuals in a prospective clinical study (raw data deposited in MassIVE, Dataset ID MSV000087451). Our method may contribute to early diagnosis of Alzheimer’s disease.

Graphical Abstract

Introduction

Alzheimer’s disease (AD) is a devastating neurodegenerative disease that affects more than 30 million people (including 5 million Americans) worldwide, and this number is expected to reach more than 120 million by the year 2050¹. Although AD is typically diagnosed postmortem², detection of the key constituents of AD pathology, the β-amyloid (Aβ) plaque and pathological forms of tau are measurable with PET scans of the brain^2–5, or by immunoassays of Aβ1-42 and Aβ1-40, as well as Tau/pTau (ELISA or xMAP) in CSF^6–9. On the other hand, minimally invasive (e.g., blood-based), reproducible, and cost-effective assays for specific biomarkers will facilitate early diagnosis and accelerate therapeutic development for AD. Major progress has been made in recent years in detecting AD biomarkers in human plasma samples, including Aβ42/Aβ40^{10, 11}, and tau phosphorylated at Thr181 and Thr217^12–14. Immunoprecipitation (IP)-mass spectrometry (MS)-based assays are the state-of-art method for plasma Aβ42/Aβ40 analysis^{10, 11}. Compared to the traditional immunoassays, MS-based assays can achieve high specificity as well as high precision using stable isotope-labeled internal standards. However, the bottom-up nanoflow LC-MS method requires enzymatic digestion of IP products for quantifying the amyloid proteins at the peptide level¹¹, which increases labor and reduces accuracy and reproducibility¹⁵.

Intact protein including top-down LC-MS analysis has progressed rapidly due to the recent advances in LC separation, high resolution mass spectrometry, and data analysis^16–18. In addition, there is a renewed interest in microflow rather than highflow or nanoflow LC-MS for clinical analysis due to its sensitivity, robustness, and throughput¹⁹. Here we developed a microflow LC-targeted MS1 assay for quantification of 4 intact β-amyloid proteins from human plasma IP samples. Our work is built on the earlier works by Nakamura et. al., and Schindler et. al^{10, 11}. The main thrust of this work is to demonstrate a novel LC-targeted MS1 assay for AD diagnosis. Compared to the published nanoflow LC-MS method¹¹, our method is equally accurate and sensitive, with the additional benefits of higher throughput, significantly reduced labor, and potentially more robustness from a simplified workflow (without digestion), as demonstrated in a prospective clinical study. Our new workflow may promote large-scale screening of the older population for early diagnosis of AD.

Methods

Participants

Thirty-six cognitively normal older participants were recruited from the Berkeley Aging Cohort Study (BACS). All individuals underwent blood draws immediately prior to amyloid and tau PET imaging at Lawrence Berkeley National Laboratory. Participants also underwent clinical assessments that included detailed cognitive testing; the Mini-Mental State Examination (MMSE, table 1) was a measure of overall cognitive ability.

Table 1:

Participant Characteristics

	Participants (N=36)
Age (years)	78 (6)
Education (years)	17 (2)
Sex (F/M)	19/17
MMSE	29 (1.6)
APOE4 carriers	5*
PiB DVR	1.142 (0.215)
PiB Centiloid	21 (31)
PIB positive/negative	16+/20−

All values are mean (SD)

MMSE: Mini-Mental State Examination

^*3 missing

Plasma collection and processing

At the same session of brain imaging, blood was drawn into one 10 ml tube precoated with K₂EDTA (BD cat # 366643). The samples were kept at room temperature during transportation and centrifuged to separate plasma from blood cells within two hours of phlebotomy. The plasma was then aliquoted into 0.5 ml each in a low protein-binding tube (Eppendorf cat # 022431081) and stored at −80°C. For method development, 50 ml pooled human plasma from Innovative Research (cat # IPLAWBK2E50ML) was aliquoted into 1.0 ml each in a low protein-binding tube and stored at −80°C.

Immunoprecipitation of amyloid proteins from plasma

Endogenous amyloid proteins (Aβ38, APP669-711, Aβ40 and Aβ42) were simultaneously immunoprecipitated from a 1 ml of plasma at 4 °C using a monoclonal antibody (6E10 from Biolegend, cat # 803003) conjugated to the M-270 Epoxy Dynabeads (Thermo Fisher Scientific, cat # 14302D). The antibody-bead conjugation was carried out according to the manufacturer protocol at a ratio of 10 μg antibody per mg beads. A volume of 125 μl 5x master mix containing 2.5x phosphate-buffered saline (PBS), 1% n-nonyl-β-D-thiomaltoside (NTM), 1% n-dodecyl-β-D-maltoside (DDM) and 5x complete protease inhibitor cocktail from Roche was added into each 500 μl plasma sample. Two 0.5 ml plasma samples from each individual participant were thawed at 4 °C and combined. The total 1 ml of plasma was spiked with 5 μl solution containing 10 pg/μl ¹³C Phe & Ile Aβ38 (AnaSpec, cat # AS-65220), 40 pg/μl ¹⁵N APP669-711, 100 pg/μl ¹⁵N Aβ40, and 10 pg/μl ¹⁵N Aβ42 (rpeptide, cat # AP-1101-1, A-1101-1 and A-1102-1 respectively), with 500 ng/μl BSA as matrix in 10% acetonitrile (ACN), 5% methanol, and 0.2% formic acid (FA). The plasma amyloid proteins and internal standard were immunoprecipitated by incubating 0.5 mg antibody-conjugated beads with the plasma overnight (~16 hrs) at 4 °C. The beads after IP were transferred to a new low-protein binding tube, washed a total of six times with 5 min each on a rotator, including twice with 1xPBS, three times with 50 mM (NH₄)₂CO₃, and one time with 10 mM (NH₄)₂CO₃. Proteins bound to the beads were then eluted with 60 μl 70% ACN elution buffer containing 5 mM HCl and 100 ng/μl BSA. The elution step was done on a rotator for 30 min. Finally, eluted proteins were transferred to a new low protein-binding tube, dried under vacuum, and stored at – 80 °C before LC-MS analysis.

LC-MS analysis of plasma IP samples

Dried amyloid proteins were reconstituted with 24 μl of 10% ACN, 5% methanol and 0.2% FA. After the undissolved pellet was spun down, amyloid proteins in the supernatant were transferred to a sample vial. Typically, three repeated injections were analyzed for each IP sample. For each technical replicate, a 3 μl aliquot was subjected to liquid chromatography mass spectrometry (LC-MS) analysis on a Thermo Orbitrap Q-Exactive Plus Mass Spectrometer interfaced with a Thermo Dionex 3000 HPLC system. Samples were loaded via a 20 μl sample loop onto a Waters 150 μm × 100 mm nanoEase M/Z Protein BEH C4 column using 95% buffer A (3% ACN and 0.2%FA) and 5% buffer B (97% ACN and 0.2% FA) for 5 min. The column was placed in a column heater set at 50 °C with a constant 1.5 μl/min flow rate. Proteins were then resolved using a steep 1.9-minute gradient from 5% B to 20% B, followed by a 11-minute linear gradient from 20% buffer B to 32% buffer B, then followed by a 2-min gradient to 50% B. The column was then washed with 95% B for 3 min, followed by re-equilibration with 5% B for 5 min.

Proteins eluted from the column were ionized using a Thermo Nanospray Flex ion source with a stainless-steel emitter. A 275 °C capillary temperature and 60% of S-lens RF level were used. For targeted MS1 analysis, the m/z of the most abundant charge state of amyloid proteins (5+ for both light and heavy labeled proteins) were selected at the expected retention time with a 3-Thomson (Th) isolation window (0.4 Th offset) and automatic gain control (AGC) values of 1e5. Maximum fill time for each protein was 400 ms, and Orbitrap resolution was 70,000. Table S1 lists the details about the MS parameters for each protein.

Mass spectrometry data analysis

Mass spectrometry data were analyzed using the Skyline software package. The integrated peak areas of isotopic ions [M+1] to [M+5] (for Aβ38, APP669-711 and Aβ40) or isotopic ions [M+2] to [M+4] (for Aβ42) with the 5+ charge state were summed. To get the Aβ42/Aβ40 ratio, endogenous (¹⁴N) Aβ42 and Aβ40 concentrations from each sample were quantified by normalization to the corresponding internal ¹⁵N labeled proteins. The final Aβ42/Aβ40 ratio was obtained by dividing the calculated Aβ42 concentration by the calculated Aβ40 concentration. The CVs for the Aβ42/Aβ40 ratio from the same individual were mostly <5%.

All mass spectrometry and quality control analyses were performed prior to sample unblinding. For batch normalization: Due to the column lifetime, all 36 individual samples and QC samples (standard Aβ proteins and plasma IP samples for QC calibration, see below) were run on two C4 columns. Samples run on the same column were considered the same batch. Two plasma IP samples, one with a relatively low Aβ42/Aβ40 ratio (low QC calibrator), another with a relatively high Aβ42/Aβ40 ratio (high QC calibrator), were run on both columns for inter-batch normalization. The Aβ42/Aβ40 ratios for the low and high QC calibrators from the first batch were 0.1221 and 0.1731, respectively, and the ratios for the low and high QC calibrators from the second batch were 0.116 and 0.153, respectively. A linear regression equation for the QC calibrators was determined as the following:

$$\text{Normalized}\text{value}\text{for}\text{the}\text{second}\text{batch}=1.38*\text{raw}\text{value}\text{for}\text{the}\text{second}\text{batch}-\mathrm{0.038.}$$

After normalization with the linear regression equation, the Aβ42/Aβ 40 ratios for both low and high QC calibrators, respectively, were the same for the two batches. Subsequently, all Aβ42/Aβ40 ratios for individual samples in the second batch were normalized using the same linear regression equation.

Amyloid PET imaging

Amyloid PET was used as the reference standard for brain amyloid pathology. Participants were scanned with ¹¹C Pittsburgh-B (PiB) according to a well-established method³. Briefly, PET studies were performed on a Siemens Biograph PET/CT TruePoint 6 scanner (Siemens Medical Systems, Erlangen, Germany). Ninety minutes of PIB-PET data were acquired, quality checked, motion corrected, and processed with a combination of the SPM8 software package, and in-house code using a graphical approach to calculate distribution volume ratios (DVR) with Logan Plots. PiB PET DVR values were used to determine amyloid positivity with a cutoff value of > 1.065 DVR²⁰. PET DVR values were also converted to Centiloid values (cutoff > 10).

Data analysis and statistics

Data analyses were performed in a double-blind and independent manner. The plasma-Aβ measurements were performed at Newomics Inc. without clinical information. The scientist performing MS analyses was completely blind to any PET imaging data. All of the PET imaging data were analyzed at UC Berkeley without any information on plasma-Aβ measurements.

Data were analyzed using Excel (Fig. 1, 2, 3d and S1) and Origin (Fig. 3a–c) software for statistical analysis and graph plotting. Student t tests were performed for continuous variables and a p value < 0.05 was considered statistically significant between two groups. Receiver operating characteristic analysis was performed to evaluate the ability of plasma Aβ42/Aβ40 to diagnose amyloid PET status. The cutoff value was determined by maximal Youden Index (Sensitivity + Specificity −1). Spearman correlations were used to evaluate the relationship between amyloid PET Centiloid and plasma Aβ42/Aβ40 values.

Fig. 1. Microflow LC-targeted MS1 analysis of ¹³C and ¹⁵N Aβ proteins.

a, Comparison of the analytical performance of targeted and full scan MS1 analyses. Solid and dotted peaks indicate the Extracted Ion Chromatograms of ¹⁵N Aβ42 using targeted and full scan MS1 methods, respectively. 0.94 pg ¹⁵N Aβ42 and 750 ng BSA matrix were loaded on column. b, Extracted Ion Chromatograms of ¹³C and ¹⁵N Aβ proteins using targeted MS1 analysis. m/z indicates the monoisotopic mass to charge ratio of the most abundant charge state (5+) of the Aβ proteins analyzed. 7.5 pg ¹³C Aβ38, 60 pg ¹⁵N APP669-711, 150 pg ¹⁵N Aβ40, 15 pg ¹⁵N Aβ42, and 750 ng BSA matrix were loaded on column. c-f, Limit of quantification analysis of Aβ proteins. The mean signal intensities from four replicates and the corresponding amount of Aβ proteins were transformed into log2 scale. Error bars represent the stand errors of means. The linearities were evaluated with the coefficient of determination (R²). We obtained a limit of quantification of 0.06 pg for Aβ38 (c), 0.94 pg for APP669-711 (d), 0.59 pg for Aβ40 (e), and 0.23 pg for Aβ42 (f). For Aβ42 in the subfigure f, the data from the two lowest concentrations were not plotted because they were out of the linear range for the titration curve.

Fig. 2. Microflow LC-targeted MS1 analysis of Aβ proteins in plasma.

1 ml pooled plasma were spiked with different amounts of heavy (isotopically labeled) Aβ proteins to test the linearity of the IP-targeted MS1 assay. a, Extracted Ion Chromatograms of Aβ proteins from plasma IP. 10x light: 10 times of the signals from endogenous proteins from the pooled plasma sample. Heavy: signals from the isotopically labeled proteins that were spiked into the plasma samples before IP. b and c, linear responses of the intensity ratio of heavy/light plot with the amount of heavy protein spiked into the pooled plasma for Aβ40 and Aβ42, respectively. Error bars represent stand errors of means from three repeated injections.

Fig. 3. Correlation between plasma Aβ42/Aβ40 values and brain amyloid PET status.

a, Aβ42/Aβ40 decreased in amyloid PET positive individuals. Error bars represent 95% confidence intervals for the mean Aβ42/Aβ40. There were 36 individuals participating in the prospective clinical study, including 20 PET− and 16 PET+ ones. b, Receiver operating characteristic analysis demonstrate that plasma Aβ42/Aβ40 was predictive of amyloid PET status. The area under the curve (AUC) is noted with 95% confidence intervals. For the cutoff listed, 88% sensitivity and 90% specificity were achieved. c, Aβ42/Aβ40 inversely correlated with amyloid PET status as measured on the Centiloid scale (L shape). The spearman ρ value is noted. The horizontal lines in a and c indicate the cutoff value of 0.1485 for Aβ42/Aβ40 based on the maximum Youden Index. The vertical line in c indicates a cutoff of 10 for amyloid PET positivity on the Centiloid scale. d, Relationship between the plasma Aβ42/Aβ40 value and MMSE rating. Individual amyloid PET status was specified. High percentage of individuals with low plasma Aβ42/Aβ40 and positive PET was identified among participants with a normal MMSE value.

Results and Discussions

We first developed a microflow LC- MS method to rapidly separate amyloid proteins and determine the relative abundance of their different charge states. The most abundant charge state (5+) of the amyloid proteins was then selected for targeted high resolution MS1 analysis with a predefined m/z range of 3 Thomson (Th) and retention time window. The representative full scan MS1 spectrum (m/z range from 350 to 1,500) and targeted MS1 spectrum of ¹⁵N Aβ42 are shown in Fig. S1a and Fig. S1b, respectively. Targeted MS1 is more sensitive and exhibits a better linearity for amyloid proteins at low concentrations, compared to full scan MS1 (Fig. 1a). It is worth noting that the standard amyloid proteins were diluted in a relative high concentration of BSA matrix to reduce the loss of the amyloid proteins during the sample prep and LC separation. Although BSA was well separated from Aβ proteins, relatively high chemical background mostly from the small m/z range was detected during the LC gradient for Aβ proteins in the full scan MS1 mode (Fig. S1a), which could saturate the C-trap of the Q-Exactive Plus and reduce the dynamic range for amyloid protein detection and quantification. Similar effects have been observed for MS analysis of tryptic peptides in the presence of proteome digests background²¹.

We then developed the multiplex targeted MS1 assay for the four standard ¹³C or ¹⁵N labeled amyloid proteins for AD diagnosis: Aβ38, APP669-711, Aβ40, and Aβ42. We did a series of dilutions (4-fold) of the mixture to determine their limit of quantification. The four amyloid proteins were well separated from each other using our micro-flow LC method (Fig. 1b). This method showed good linear correlations between the amounts of amyloid proteins loaded and the peak area intensity from targeted MS1, with the coefficient of determination (R²) above 0.995 for all four proteins (Fig. 1c–f). The limit of quantification was determined at a minimal signal to noise ratio (S/N) of 10, with a Coefficient of Variation (CV) of less than 20% within the linear range. We obtained a limit of quantification of 0.06 pg for Aβ38, 0.94 pg for APP669-711, 0.59 pg for Aβ40 and 0.23 pg for Aβ42 for our assay (Fig. 1).

We next developed an IP-targeted MS1 method for amyloid proteins using commercial pooled plasma. We spiked the four isotopically labeled amyloid proteins mentioned above into 1 ml plasma samples before IP for precise quantification of endogenous proteins. For MS analysis, we targeted the endogenous and labeled amyloid proteins sequentially in a single LC run (Fig. 2a). To determine the linearity of our plasma Aβ IP-targeted MS1 analysis, we spiked various concentrations of isotopically labeled amyloid proteins covering the physical concentration ranges into aliquots of the pooled plasma (containing the same fixed amounts of endogenous light amyloid proteins) and quantified the MS intensity ratio of heavy/light for each sample. We focused on the quantitation of Aβ40 and Aβ42 in the subsequent studies because they provided better accuracy in predicting PET results (see below). This method showed a good linear response between the amounts of labeled amyloid proteins spiked in and the ratio of heavy/light peak area intensity from targeted MS1 analysis, with the coefficient of determination (R²) above 0.999 for both Aβ40 and Aβ42 (Fig. 2b–2c). The CV of the intensity ratios from the lowest to the highest spiked concentrations for Aβ40 was 4.3%, 1.1%, 1.2%, 2.5% and 1.6%, respectively; and the corresponding CV for Aβ42 was 14.0%, 3.7%, 1.6%, 1.8% and 2.7%, respectively. We also plotted the data using a log2 scale, and the slopes of the linear regression lines for both Aβ40 and Aβ42 were very close to 1 (Fig. S2), indicating that the MS intensity ratio of heavy/light for both Aβ40 and Aβ42 accurately reflects the actual concentration ratio in the samples. The data reproducibility across multiple days was confirmed by additional experiments (Fig. S3).

To assess the reproducibility of our IP-targeted MS1 assay, we spiked the same amount of aliquoted isotopically-labeled amyloid proteins into three individual tubes (1 ml each) of pooled human plasma samples, and performed IP-MS in parallel. We quantified the endogenous amyloid proteins in each tube by normalizing them to the corresponding spiked-in labeled amyloid protein. The average concentrations of plasma Aβ were 32.2 pg/ml for Aβ40 and 9.3 pg/ml for Aβ42, respectively, and Aβ42/Aβ40 was 0.288, for the pooled commercial plasma. The CV for Aβ40, Aβ42, and Aβ42/Aβ40 was 1.4%, 3.1%, and 2.4%, respectively (Table S2). The average concentrations of plasma Aβ proteins from the commercial pooled plasma were significantly lower than that from our clinical study and other reports (see below). This could presumably be due to the younger age of the donors for the commercial pooled plasma, as well as the different protocols for blood collection and sample processing. High resolution (to resolve isotopic peaks) and mass accuracy (typically less than 5 ppm) are necessary for accurate quantification at MS1. We detected interference peaks that show similar m/z to some isotopic ions of Aβ42 with 5+ charge state, so we only quantified the isotopic ions from [M+2] to [M+4] for Aβ42.

We validated our microflow LC-targeted MS1 assay for amyloid proteins using plasma samples from 36 cognitively normal individuals in a prospective clinical study (Table 1). We spiked the same amount of aliquoted isotopically labeled amyloid proteins into 1 ml of each plasma sample before IP, and 12.5% of the IP products was loaded for each LC-MS replicate run. We quantified the endogenous amyloid proteins by normalizing them to the corresponding labeled amyloid proteins spiked in. Average concentrations of plasma Aβ was 329.5 pg/ml for Aβ40 and 48.0 pg/ml for Aβ42 for these 36 individuals, with a standard deviation of 89.4 and 12.8 pg/ml, respectively. The average concentrations of plasma Aβ from this study were similar to what has been reported previously²². Furthermore, all individual concentrations were within the linear ranges for our developed assay (Fig 2b–c). We calculated the ratio of plasma Aβ42/Aβ40 for each individual, and compared the result to the data obtain from the brain amyloid PET scan. The CVs of Aβ42/Aβ40 for instrument variation from majority of the IP samples are within 5%. As expected, individuals with a positive amyloid PET had a significantly lower plasma Aβ42/Aβ40 compared to individuals with a negative amyloid PET (mean = 0.1359 vs. 0.1770, p < 0.001) (Fig 3a). The average Aβ42/Aβ40 ratio from this study is slightly higher than what has been reported¹¹. Receiver Operating Characteristic analysis demonstrated that plasma Aβ42/Aβ40 was a good predictor of brain amyloid PET status, with an Area Under Curve (AUC) of 0.89 (95% Confidence Interval 0.77–1.00) (Fig. 3b). A plasma Aβ42/Aβ40 cutoff of < 0.1485 was considered positive and had the maximum Youden Index with an accuracy close to 90% (Fig. 3b). Strikingly, plasma Aβ42/Aβ40 was inversely corelated with amyloid PET on the Centiloid scale (Fig. 3c), with a Spearman ρ = −0.63. There were 2 individuals with positive plasma Aβ42/Aβ40 and negative PET scans, and 2 individuals with negative plasma Aβ42/Aβ40 and positive PET scans; 1 of these individuals was within 5% of the threshold values. Plasma Aβ42/Aβ40 value was lower with older age and in APOE ɛ4 carriers (Fig. S4), similar to what has been reported previously¹¹. In addition, a high percentage of these cognitively normal individuals has both a low plasma Aβ42/Aβ40 value and corresponding positive PET status, even though majority of them have a normal Mini-Mental State Examination (MMSE) value (Fig. 3d). Our data support the notion of affordable rapid screening of the older population, particularly those at risk of AD dementia, by measuring their plasma Aβ42/Aβ40 values.

We also analyzed the data on APP669-711 and calculated the ratio of APP669-711/Aβ42 for ROC analysis. The AUC from APP669-711/Aβ42 was only 0.57, and that from the composite of APP669-711/Aβ42 and Aβ42/Aβ40 was 0.792, both were dramatically less than that from Aβ42/Aβ40 alone. Through extensive literature search, we have found another recent publication using plasma APP669-711 for AD diagnosis by immunoassay²³. The authors obtained an AUC of 0.790 and 0.854 for Aβ42/APP669-711 and Aβ42/Aβ40, respectively, whereas the AUC for both markers were close to 0.9 in Ref. 10. The discrepancy in using plasma APP669-711 as a reference marker might be due to the different sample preparation procedures and assay platforms (LC-MS1 assay in this study, MALDI-MS in Ref. 10, and the Immunoassay in Ref. 23) used in these studies. Nevertheless, the performance and value of using plasma APP669-711 as an additional AD biomarker warrant further investigations.

In conclusion, we have developed a microflow LC-targeted MS1 assay for multiplex analysis of intact amyloid proteins from human plasma IP samples. We have demonstrated high diagnostic accuracy for brain amyloid pathology using our assay in a prospective clinical study. Compared to the bottom-up nanoflow LC-MS method^{11, 22}, this method has the advantage of significantly reduced labor cost for sample preparation, easy adaption for further automation, and higher robustness and throughput. We were able to reliably quantify the Aβ42/Aβ40 biomarker from all plasma samples analyzed so far, whereas 3% of the samples were dropped from the study in Ref.11 because of low data quality. Our current microflow LC-MS method is more than 6 min shorter than the nanoflow LC-MS method, and it can be further shortened by simply increasing the flow rate of the sample loading and column washing, similar to what have been adopted in the nanoflow LC-MS method^{11, 22}. We have observed ion suppression for Aβ42 from the co-eluting protein(s) in our plasma IP samples when we used a much shorter LC gradient. Further improvement in the throughput of our current LC-MS method will likely require optimization of plasma sample preparation and LC gradient to reduce ion interference. Future studies will multiplex amyloid proteins with other proteoforms such as pTau¹², and incorporate our microflow LC-nanospray MS platform¹⁶, to achieve even higher accuracy, throughput, and robustness of LC-MS assays for blood-based screening and early diagnosis of AD.

Supplementary Material

Supplementary Material

Mass spectra of full scan MS1 and targeted MS1 corresponding to Figure 1a (Figure S1)

Log2 plotted data of targeted MS1 analysis of Aβ proteins in plasma corresponding to Figure 2b and 2c (Figure S2)

Plasma Aβ proteins IP-MS assay reproducibility across multiple days (Figure S3)

Relationship between Aβ42/Aβ40 values and age, with different APOE ɛ4 status (Figure S4)

MS parameters for targeted MS1 analysis of Aβ proteins (Table S1)

Reproducibility analysis of IP-MS assay of plasma Aβ proteins (Table S2)