Timing of biomarker changes in sporadic Alzheimer disease in estimated years from symptom onset

Abstract

Objective:

A clock relating amyloid PET to time was used to estimate the timing of biomarker changes in sporadic Alzheimer disease (AD).

Methods:

Research participants were included who underwent CSF collection within two years of amyloid PET. The ages at amyloid onset and AD symptom onset were estimated for each individual. The timing of change for plasma, CSF, imaging, and cognitive measures was calculated by comparing restricted cubic splines of cross-sectional data from the amyloid PET positive and negative groups.

Results:

The amyloid PET positive sub-cohort (n=118) had an average age of 70.4 ± 7.4 years (mean ± standard deviation) and 16% were cognitively impaired. The amyloid PET negative sub-cohort (n=277) included individuals with low levels of amyloid plaque burden at all scans who were cognitively unimpaired at the time of the scans. Biomarker changes were detected 15–19 years before estimated symptom onset for CSF Aβ42/Aβ40, plasma Aβ42/Aβ40, CSF pT217/T217, and amyloid PET; 12–14 years before estimated symptom onset for plasma pT217/T217, CSF neurogranin, CSF SNAP-25, CSF sTREM2, plasma GFAP, and plasma NfL; and 7–9 years before estimated symptom onset for CSF pT205/T205, CSF YKL-40, hippocampal volumes, and cognitive measures.

Interpretation:

The use of an amyloid clock enabled visualization and analysis of biomarker changes as a function of estimated years from symptom onset in sporadic AD. This study demonstrates that estimated years from symptom onset based on an amyloid clock can be used as a continuous staging measure for sporadic AD and aligns with findings in autosomal dominant AD.

Introduction

Alzheimer disease (AD) is the most common cause of dementia and is defined by the presence of amyloid plaques comprised of amyloid-β peptide, neurofibrillary tangles comprised of hyperphosphorylated tau, and neurodegeneration.¹ AD has a prolonged preclinical phase during which AD pathology accumulates over 10–20 years but individuals remain cognitively unimpaired.^1–4 Amyloid positron emission tomography (PET) enables visualization and quantification of amyloid plaque burden in the brain.^{5, 6} Multiple studies have demonstrated that after reaching a threshold level or “tipping point,” amyloid accumulates consistently across individuals, allowing formulation of an “amyloid clock” that relates amyloid plaque burden to time.^{3, 7–10} The age at this tipping point or “amyloid onset” is highly predictive of the age of clinical symptom onset in sporadic AD.^{7, 8}

Concentrations of amyloid-β peptides and tau species in the plasma and cerebrospinal fluid (CSF) reflect amyloid and tau pathology. Plasma and CSF ratios of amyloid-β peptide 42 to 40 (Aβ42/Aβ40) are correlated with amyloid plaque burden.^{11, 12} Plasma and CSF tau phosphorylation occupancies (ratio of the phosphorylated to non-phosphorylated tau concentrations) at threonine 181, 205, 217, and 231 are correlated with various imaging biomarkers of AD pathology.^12–14 Plasma and CSF pT217/T217 have particularly high associations with amyloid PET and early stage tau tangle pathology as measured by tau PET.^{12, 14} In contrast, CSF pT205/T205 is well correlated with tau PET but relatively poorly correlated with amyloid PET^{12, 15}, making it a more specific biomarker of tau pathology.

Plasma and/or CSF concentrations of a variety of AD biomarkers reflect processes such as synaptic dysfunction, neuronal injury, and inflammation. Neurogranin is a post-synaptic calmodulin-binding protein that modulates synaptic plasticity and long-term potentiation.^{16, 17} SNAP-25 is an essential component of the sensitive factor attachment protein receptor (SNARE) complex, which initiates fusion of synaptic vesicles to the pre-synaptic membrane.^{18, 19} VILIP-1 is a neuronal calcium sensor protein.^20–22 Neurofilament light (NfL) is a neuronal cytoskeletal protein.²³ YKL-40, also known as chitinase-3-like protein 1, is a secreted glycoprotein that is mainly expressed in the brain by astrocytes.^{24, 25} Triggering receptor expressed on myeloid cells 2 (TREM2) is a key regulator of microglial activation, survival, and phagocytosis.²⁶ Glial fibrillary acidic protein (GFAP) is a marker of reactive astrogliosis²⁷.

To better understand the timing of biomarker changes in sporadic AD, we used previously described models to estimate the age at amyloid onset and age at symptom onset for each individual⁷. An individual’s age at the time of biomarker measurement minus their estimated age at symptom onset is referred to as their estimated years from symptom onset (EYO), which is a key staging measure used in studies of autosomal dominant AD (ADAD).^{2, 28} In this study, values for plasma, CSF, imaging, and cognitive measures were plotted as a function of EYO on both an individual and cohort level. For the sub-cohort with amyloid plaques by PET (Centiloids>7), the first EYO at which measures were significantly different from a control group without amyloid plaques (Centiloids≤7) was calculated. Overall, this study examines the feasibility of using EYO based on an amyloid clock as a continuous staging measure for studying the pathophysiology of sporadic AD.

Methods

Participants

The study analyzed data from research participants enrolled in studies of memory and aging at the Charles F. and Joanne Knight Alzheimer Disease Research Center (ADRC) affiliated with Washington University. The cohort consists of community-dwelling older adults, including participants with and without cognitive impairment, who were recruited primarily from the greater St. Louis, Missouri metropolitan region. Participants who underwent an amyloid PET scan with the Pittsburgh compound B (PiB) radiotracer within two years of CSF collection were included. All procedures were approved by the Washington University Human Research Protection Office. Written informed consent in accordance with the Declaration of Helsinki was obtained from each participant or their legally authorized representative when appropriate.

Clinical and cognitive assessments

Individuals underwent a comprehensive clinical assessment at enrollment and then every two to three years for participants younger than 65 years old and yearly for participants age 65 years and older. The clinical assessment included a detailed interview of a collateral source, a neurological examination of the participant, the Clinical Dementia Rating^® (CDR^®)²⁹, Clinical Dementia Rating Sum of Boxes (CDR-SB), and the Mini-Mental State Examination (MMSE).³⁰ Individuals with a CDR of 0 were categorized as “cognitively unimpaired.” Individuals with a CDR of 0.5 or greater were considered to have a dementia syndrome and the probable etiology of the dementia syndrome was formulated by clinicians based on clinical features in accordance with standard criteria and methods.³¹ A global cognitive composite was formed as previously described.³²

Plasma and CSF collection

Plasma and CSF were collected at approximately 8 am after overnight fasting. Blood was drawn into two 10 mL syringes pre-coated with 0.5 M EDTA, then transferred to two 15 mL polypropylene tubes containing 120 μl 0.5 M EDTA. The samples were kept on wet ice (<2 hours) until centrifugation for 2,000 × g × 15 minutes to separate plasma from blood cells. The plasma was then transferred to a single 50 mL polypropylene tube, gently mixed, aliquoted into polypropylene tubes and stored at −80°C. CSF was collected via lumbar puncture in a 50 mL polypropylene tube by gravity drip using an atraumatic Sprotte 22 gauge spinal needle. The entire sample was gently inverted to disrupt potential gradient effects and centrifuged at low speed to pellet any cellular debris. CSF was then aliquoted into polypropylene tubes and stored at −80°C.

Plasma and CSF assays

Plasma Aβ42/Aβ40, pT217/T217, and pT181/T181 were measured by C2N Diagnostics in their clinical laboratory.³³ Plasma NfL and GFAP were measured via Quanterix Neuro 2-Plex B Advantage assay kits on a SIMOA HD-X analyzer at Washington University according to manufacturer’s specifications.³⁴ CSF Aβ42, Aβ40, p-tau181, and total tau (t-tau) concentrations were measured with an automated immunoassay platform (LUMIPULSE G1200, Fujirebio) according to manufacturer’s specifications.¹² CSF pT181/T181, pT205/T205, pT217/T217, and pT231/231 were measured via an immunoprecipitation-mass spectrometry assay.¹² CSF neurogranin, SNAP-25, and VILIP-1 were measured with microparticle-based immunoassays using Single Molecule Counting [SMC]^™ technology.^{35, 36} CSF soluble triggering receptor expressed on myeloid cells 2 (sTREM2) was measured by an in-house plate-based ELISA.²⁶ CSF NfL and YKL-40 were measured with commercial Enzyme-Linked Immunosorbent Assays (ELISAs), manufactured by Uman Diagnostics and Quidel, respectively.^{34, 35}

Amyloid PET and structural brain MRI

Participants underwent amyloid PET using ¹¹C-Pittsburgh Compound B (PiB) in coordination with a 3 Tesla structural MRI scan. T₁-weighted MRIs were processed using Freesurfer 5.3 to generate regions of interest used for the processing of PET data. Estimates of regional volumes were adjusted for intracranial volume using a regression approach. Data from the 30–60 minute post-injection window for PiB were converted to standardized uptake value ratios (SUVRs) using the cerebellar grey as a reference and partial volume corrected using a geometric transfer matrix.³⁷ Values from the following regions were averaged together to represent mean cortical SUVR for PiB: bilateral orbitofrontal, medial orbitofrontal, rostral middle frontal, superior frontal, superior temporal, middle temporal, and precuneus. Mean cortical SUVRs were converted to Centiloid units to facilitate comparisons with other datasets.^{38, 39} Hippocampal volume was calculated as the sum of both the right and left hippocampus and was normalized to the mean intracranial volume in the cohort.

Models for amyloid time and estimated years from symptom onset

Because amyloid accumulates consistently across individuals after reaching a threshold level or tipping point, longitudinal amyloid PET data can be used to create an amyloid clock relating amyloid burden to time in years. For this study, a previously described amyloid clock created in an overlapping cohort was used that demonstrated a Pearson correlation of 0.83 between estimated time intervals and actual time intervals⁷. Amyloid time is a non-linear transformation of mean cortical SUVR and is defined as the estimated time in years from a “tipping point” that was identified at SUVR 1.20 (Centiloids=7), after which amyloid burden increased in nearly all individuals. Notably, this tipping point, which was obtained with the highly sensitive PIB tracer in a single large cohort, was lower than in some other studies that used less sensitive tracers or multiple cohorts⁴⁰. Further, this tipping point was derived by anchoring rates of change to the estimated amyloid burden halfway through the follow-up period rather than at baseline. The time between levels of amyloid burden was calculated by integrating the inverse of the modeled rate of change between an SUVR of interest and SUVR 1.20. Amyloid time, which has also been called “amyloid duration” or “amyloid chronicity”⁴¹, can be conceptualized as the time of detectable amyloid accumulation.

The age at which an individual reached SUVR 1.20 (“amyloid onset”) was estimated by subtracting the amyloid time value corresponding to a SUVR from the age at the scan (e.g., if an individual is 70 years old and has an SUVR of 2.20, which corresponds to an amyloid time of 9.2 years, the individual was approximately 60.8 years old at amyloid onset). The age at symptom onset was previously found to be highly correlated with the estimated age at amyloid onset and was calculated as 0.70 × (estimated age at SUVR 1.20) + 34.7 years.⁷ For example, an individual with amyloid onset at 60.8 years has an estimated age at symptom onset of 77.3 years ([0.70 × 60.8 years] + 34.7 years). EYO was calculated as the individual’s age at the time of a cognitive assessment or biomarker measurement minus their estimated age at symptom onset. For example, a 70 year old individual with an estimated age at symptom onset of 77.3 years would have an EYO of −7.3 years (70 years minus 77.3 years).

Visualization of measures over time

The amyloid PET positive sub-cohort included individuals with at least one amyloid PET scan in the range (Centiloids 7 to 88) that enabled estimation of their age at the tipping point (SUVR 1.20, Centiloids=7). The amyloid PET negative sub-cohort included individuals with low levels of amyloid plaque burden at all scans (Centiloids ≤7) who were cognitively unimpaired at the time of the scans. For each measure, the baseline values for the amyloid positive group were standardized by the mean of the measure in the amyloid negative group and the standard deviation of the measure in the amyloid positive group (standardized value=[individual value-mean of amyloid negative group]/standard deviation of amyloid positive group). To visualize measures over time, standardized differences were plotted as a function of EYO and amyloid time.

Timing of change

Analyses of the timing of biomarker were performed using an approach similar to that used for studies of ADAD¹³. For the amyloid positive group, restricted cubic splines were fit to the baseline values of each measure as a function of EYO. Based on the restricted cubic splines, the mean and standard error of the biomarker at each integer EYO were estimated for the amyloid positive group. The differences between the mean of the amyloid positive group at each integer EYO and the overall mean of the amyloid negative group were then compared using two sample t-tests. The standard error for the mean differences were estimated using Satterthwaite approximation based on the standard error from both the amyloid positive and negative groups. The first integer EYO when a measure was significantly different between the amyloid positive and negative groups was determined.

Participants in the amyloid positive group with longitudinal data were used for analyses of timing of change based on longitudinal data. For each measure, the annual rate of change was estimated using a linear regression for each participant and the EYO halfway through the follow-up period was calculated. Restricted cubic splines were then fit to the estimated annual rate of change by the EYO halfway through the follow-up period. Based on the restricted cubic splines, the mean and standard error of the annual rate of change at each integer EYO for the amyloid positive group were estimated. The first integer EYO when the rate of change in a measure was significantly different between the amyloid positive and negative groups was determined using the same approach as for the cross-sectional analysis. The integer EYO at which the maximal rate of change occurred was determined by estimating the rate of change at each integer EYO based on the cubic spline model.

As some measures may become abnormal with older age regardless of AD pathology, we also calculated the first integer EYO at which a measure became significantly different from the amyloid negative group after adjustment for age. For each measure, a linear regression with age as the independent variable was fitted using the data from the amyloid negative group only and the age adjusted residuals were estimated from the regression. The same regression from the amyloid negative group was then applied to the data in the amyloid positive group to determine the age adjusted residuals for amyloid positive group. The first integer EYO when a measure was different from the amyloid negative group was determined based on the age adjusted residuals using the same method described for the unadjusted values.

Similar cross-sectional and longitudinal analyses for each biomarker were also performed as a function of amyloid time. Biomarkers were transformed with the natural logarithm prior to standardization as appropriate. All p values were based on 2-sided tests and values <0.05 were considered significant. All analyses were conducted using SAS 9.4 (SAS Institute Inc., Cary, NC).

Results

Participants

All individuals included in the study cohort (n=395) had available data on CSF biomarkers and had undergone at least one amyloid PET scan with PiB. Participant characteristics at the time of the baseline CSF samples are shown in Table 1. The amyloid PET positive sub-cohort (n=118) included individuals with at least one amyloid PET scan in the range (SUVR 1.20 to 3.00, Centiloids 7 to 88) that enabled estimation of their age at the amyloid PET tipping point (SUVR=1.20, Centiloids=7). The average age was 70.4 ± 7.4 years (mean ± standard deviation), 55% were female, 56% were APOE ε4 carriers, and 16% were cognitively impaired (CDR>0). The number of amyloid PET positive individuals with longitudinal data for plasma and CSF analytes ranged from 40–60 individuals and the average length of follow-up ranged from 4.9 to 6.9 years (Supplemental Table 1). The amyloid PET negative sub-cohort (n=277) included individuals with low levels of amyloid PET signal at all scans (SUVR≤1.20, Centiloids≤7) who were cognitively unimpaired (CDR=0) at the time of the scan. The average age was 63.5 ± 9.1 years, 59% were female, and 23% were APOE ε4 carriers.

Table 1. Participant characteristics.

Individuals in the amyloid PET negative group had an amyloid PET SUVR ≤1.20 (Centiloids ≤7) at all scans. Individuals in the amyloid PET positive group had an amyloid PET SUVR between 1.20 and 3.00 (Centiloids 7 to 88) at one or more scans. Values at the baseline amyloid PET scan are shown. Continuous measures are presented as mean ± SD.

Characteristic	Entire cohort (n=395)			Amyloid PET negative (n=277)		Amyloid PET positive (n=118)
Demographics
Age at CSF collection (years)	395	65.6 ± 9.2	277		63.5 ± 9.1	118	70.4 ± 7.4
Gender (n, % female)	395	229, 58%	277		164, 59%	118	65, 55%
APOE ε4 carrier status (n, % ε4 carrier)	388	128, 33%	272		63, 23%	116	65, 56%
Years of education	395	16.1 ± 2.5	277		16.2 ± 2.4	118	15.7 ± 2.8
Race (Black/White/Other)	395	44/349/2	277		36/239/2	118	8/110/0
Clinical/cognitive measures
CDR (0/0.5/1+, % >0)	394	375/17/2, 5%		277	277/0/0, 0%	117	98/17/2, 16%
CDR-SB	394	0.12 ± 0.60		277	0.01 ± 0.07	117	0.40 ± 1.05
MMSE	394	29.1 ± 1.3		277	29.3 ± 1.0	117	28.7 ± 1.7
Global cognitive composite	388	−0.10 ± 0.73		276	0 ± 0.70	112	−0.36 ± 0.74
Imaging measures
Amyloid PET SUVR	395	1.3 ± 0.5		277	1.0 ± 0.1	118	1.9 ± 0.7
Amyloid PET Centiloid	395	9.5 ± 24.2		277	−2.2 ± 3.0	118	37.0 ± 29.3
Interval between amyloid PET and CSF collection	395	0.38 ± 0.48		277	0.38 ± 0.47	118	0.39 ± 0.52
Hippocampal volume	394	7825 ± 987		277	7972 ± 933	117	7478 ± 1028
Interval between MRI and CSF collection	394	0.25 ± 0.35		277	0.25 ± 0.35	117	0.24 ± 0.33
CSF biomarkers by Lumipulse automated assay
Aβ42 (pg/ml)	395	840 ± 349		277	939 ± 336	118	608 ± 260
Aβ40 (pg/ml)	395	10661 ± 3232		277	10444 ± 3289	118	11173 ± 3049
Aβ42/Aβ40	395	0.0792 ± 0.0203		277	0.0896 ± 0.0102	118	0.0547 ± 0.0169
Total tau (pg/ml)	395	303 ± 186		277	251 ± 126	118	427 ± 240
p-tau181 (pg/ml)	395	39.7 ± 25.6		277	31.4 ± 12.1	118	59.0 ± 36.3
CSF biomarkers by immunoassay
Neurogranin (pg/mL)	351	1945 ± 1021		244	1767 ± 908	107	2353 ± 1145
VILIP-1 (pg/ml)	371	198.6 ± 76.9		259	186.3 ± 70.3	112	227.0 ± 84.1
SNAP-25 (pg/ml)	358	5.00 ± 1.93		252	4.62 ± 1.58	106	5.90 ± 2.35
YKL-40 (ng/ml)	342	210.1 ± 83.2		247	199.8 ± 84.0	95	236.9 ± 75.2
NfL (pg/ml)1	383	6.49 ± 0.50		266	6.40 ±0.50	117	6.72 ± 0.42
sTREM2 (pg/ml)	332	1742 ± 573		224	1655 ± 517	108	1922 ± 641
CSF biomarkers by IPMS
pT217/T217 (%)	395	4.32 ± 2.79		277	3.11 ± 0.56	118	7.17 ± 3.71
pT181/T181 (%)	395	29.48 ± 4.40		277	27.96 ± 2.87	118	33.03 ± 5.26
pT231/T231 (%)	325	9.14 ± 9.55		218	4.96 ± 4.66	107	17.66 ± 11.20
pT205/T205 (%)	391	0.97 ± 0.31		274	0.92 ± 0.26	117	1.11 ± 0.36
Plasma biomarkers
Aβ42/Aβ40	395	0.10 ± 0.01		277	0.11 ± 0.01	118	0.10 ± 0.01
pT217/T217 (%)	297	0.87 ± 1.21		211	0.57 ± 1.04	86	1.62 ± 1.28
pT181/T181 (%)	298	15.59 ± 4.59		212	14.92 ± 4.40	86	17.25 ± 4.67
NfL (pg/ml)1	227	2.48 ± 0.53		164	2.38 ± 0.52	63	2.74 ± 0.47
GFAP (pg/ml)1	224	4.77 ± 0.52		161	4.62 ± 0.43	63	5.16 ± 0.53

¹CSF NfL, plasma NfL and GFAP are log transformed.

Abbreviations: Aβ42, amyloid-β 42; Aβ40, amyloid-β 40; APOE ε4, apolipoprotein E epsilon 4 allele; CDR, Clinical Dementia Rating where CDR, 0 indicates cognitive normality, CDR, 0.5 indicates very mild dementia, and CDR, 1 indicates mild dementia; CDR-SB, CDR Sum of Boxes; CSF, cerebrospinal fluid; MMSE, Mini-Mental State Examination (score of 30 is best and a score of 0 is worst); n, number of individuals; NfL, neurofilament light chain; PET, positron emission tomography; p-tau181, tau phosphorylated at site 181; pT181/T181, percentage of tau phosphorylated at threonine 181; pT205/T205, percentage of tau phosphorylated at threonine 205; pT217/T217, percentage of tau phosphorylated at threonine 217; pT231/T231, percentage of tau phosphorylated at threonine 231; SNAP-25, synaptosomal-associated protein-25; SUVR, standardized uptake value ratio; sTREM2, soluble triggering receptor expressed on myeloid cells 2; t-tau, total tau; VILIP-1, visinin-like protein 1; YKL-40, chitinase-3-like protein 1, GFAP, glial fibrillary acidic protein.

Biomarker values as a function of different time measures

Amyloid PET Centiloid values over the course of AD were visualized on an individual level with spaghetti plots. For all individuals in the cohort, amyloid PET Centiloid values as a function of age are shown in Figure 1A. While this plot depicts the actual changes in amyloid PET Centiloids as a function of time, individuals start to accumulate amyloid at different ages, and therefore the trajectories across the cohort are not aligned and appear chaotic. For individuals in the amyloid PET positive sub-cohort, amyloid PET Centiloid values as a function of amyloid time (the age at the scan minus the estimated age at amyloid onset) are shown in Figure 1B. The same data are shown as a function of the estimated years from symptom onset (EYO, the age at the scan minus the estimated age at symptom onset) in Figure 1C. Note that the amyloid PET Centiloid values shown in Figure 1A–C for the amyloid PET positive sub-cohort are the same but are aligned on the x-axis by different time measures.

Figure 1. Amyloid PET values by age, amyloid time, and estimated years from symptom onset.

Amyloid PET Centiloids as a function of age (A); amyloid time (age at scan minus estimated age at SUVR 1.20) (B); estimated years from symptom onset (age at scan minus estimated age at symptom onset) (C). Red points/lines represent APOE ε4 carriers and blue points/lines represent APOE ε4 non-carriers. Circles represent cognitively unimpaired individuals (CDR, 0 at the time of biomarker assessment) and triangles represent cognitively impaired individuals (CDR>0 at the time of biomarker assessment).

The same approach was used to visualize hippocampal volume (Figure 2A–D), CSF Aβ42/Aβ40 (Figure 2E–H), and CSF pT217/T217 (Figure 2I–L) as a function of different time measures. First, biomarker values as a function of age are shown for the entire cohort (Figure 2A, E, I). Second, biomarker values within two years of an amyloid PET scan as a function of amyloid PET Centiloids are shown for the entire cohort (Figure 2B, F, J). This approach uses amyloid PET Centiloids as a proxy for time, but it is not informative about the values on a time scale (i.e., years), and data obtained longer than two years from an amyloid PET scan are not represented. Third, for individuals in the amyloid PET positive sub-cohort, values are shown as a function of amyloid time (Figure 2C, G, K) and EYO (Figure 2D, H, L). Notably, alignment of data by an age (estimated age at amyloid onset or estimated age at symptom onset) enables evaluation of trajectories in units of actual time (years) and examination of information that is obtained distant from an amyloid PET scan (e.g., >2 years before or after the scan), increasing the amount of longitudinal data represented. Similar sequences of plots are shown for additional CSF and plasma biomarkers: CSF Aβ42, Aβ40, p-tau181, total tau (Supplemental Figure 1); CSF pT205/T205, pT231/T231, pT181/T181 (Supplemental Figure 2); CSF neurogranin, SNAP-25, VILIP-1 (Supplemental Figure 3); CSF YKL-40, sTREM2, and NfL (Supplemental Figure 4); plasma Aβ42/Aβ40, pT217/T217, NfL and GFAP (Figure 3) and plasma pT181/T181 (Supplemental Figure 5).

Figure 2. Biomarkers by age, amyloid PET Centiloids, amyloid time, and estimated years from symptom onset.

Hippocampal volume (A-D), CSF Aβ42/Aβ40 (E-H), and CSF pT217/T217 (I-L) are plotted as a function of age (A, E, I), amyloid PET Centiloids (B, F, J); amyloid time (age at scan/collection minus estimated age at SUVR 1.20) (C, G, K); or estimated years from symptom onset (age at scan/collection minus estimated age at symptom onset) (D, H, L). Red points/lines represent APOE ε4 carriers and blue points/lines represent APOE ε4 non-carriers. Circles represent cognitively unimpaired individuals (CDR, 0 at the time of biomarker assessment) and triangles represent cognitively impaired individuals (CDR>0 at the time of biomarker assessment).

Figure 3. Plasma biomarkers by age, amyloid PET Centiloids, amyloid time, and estimated years from symptom onset.

Plasma Aβ42/Aβ40 (A-D), pT217/T217 (E-H), NfL (I-L) and GFAP (M-P) are plotted as a function of age (A, E, I, M), amyloid PET Centiloids (B, F, J, N); amyloid time (age at scan/collection minus estimated age at SUVR 1.20) (C, G, K, O); or estimated years from symptom onset (age at scan/collection minus estimated age at symptom onset) (D, H, L, P). Red points/lines represent APOE ε4 carriers and blue points/lines represent APOE ε4 non-carriers. Circles represent cognitively unimpaired individuals (CDR, 0 at the time of biomarker assessment) and triangles represent cognitively impaired individuals (CDR>0 at the time of biomarker assessment).

Differences in biomarker values over the course of AD

Plots of biomarker values over the course of AD on a cohort level were generated by modeling the standardized difference in a biomarker measure between amyloid PET positive and negative groups as a function of EYO. Importantly, the standardized difference reflects both the magnitude of difference between amyloid PET positive and negative groups as well as the variability within groups. Plots are shown for imaging and cognitive measures (Figure 4A), CSF tau phosphorylation (Figure 4B), various CSF biomarkers of AD (Figure 4C) and plasma biomarkers (Figure 4D). Relatively small standardized differences were observed for some of the CSF biomarkers, including neurogranin, SNAP-25, VILIP-1, and sTREM2, which likely reflects high inter-individual variability in these biomarkers. Differences in key measures as a function of EYO are shown in Figure 5: CSF Aβ42/Aβ40, pT217/T217, pT205/T205; amyloid PET, hippocampal volume, and CDR-SB. Supplemental Figures 6–7 demonstrate differences as a function of amyloid time.

Figure 4. Differences in imaging, cognitive, CSF and plasma biomarkers as a function of estimated years from symptom onset.

LOESS curves were used to model the standardized differences between the amyloid positive group and the amyloid negative group as a function of estimated years from expected symptom onset, with higher values corresponding to greater abnormality. Curves are shown for imaging and cognitive measures (A), CSF tau phosphorylation (B), a variety of CSF biomarkers (C) and plasma biomarkers (D).

Figure 5. Differences in key biomarkers as a function of estimated years from symptom onset.

LOESS curves were used to model the standardized differences between the amyloid positive group and the amyloid negative group as a function of estimated years from expected symptom onset, with higher values corresponding to greater abnormality.

The earliest EYO or amyloid time value when biomarker measures varied significantly between amyloid PET positive and negative groups was identified (Table 2). Some individuals had data collected prior to reaching the amyloid PET tipping point, resulting in some estimates for the time of biomarker change occurring at amyloid time<0. The biomarkers demonstrating very early changes included CSF Aβ42/Aβ40 (EYO −19 years), plasma Aβ42/Aβ40 (EYO −18 years), CSF pT217/T217 (EYO −18 years) and CSF pT181/T181 (EYO −17), as well as amyloid PET, CSF Aβ42 and CSF NfL (EYO −15 years). CSF pT231/T231 and CSF t-tau also had early detectable changes (EYO −14 years). CSF biomarkers of neuronal injury (VILIP-1), synaptic dysfunction (neurogranin and SNAP-25), microglial activation (sTREM2), as well as plasma pT217/T217, plasma NfL and plasma GFAP, had detectable changes at EYO −14 to −12 years. A change in CSF pT205/T205 was detected at EYO −9 years, later than for other CSF tau species, but at approximately the same time when differences in hippocampal volume and cognitive measures (CDR-SB, global cognitive composite, MMSE) became significant (EYO −8 to −7 years). Additional models adjusted for age resulted in similar major findings, except for CSF NfL, CSF YKL-40, and CSF sTREM2; adjusting for the association between age and these biomarkers reduced differences between amyloid positive and negative groups such that the timing of difference could not be detected (Supplemental Table 2).

Table 2. Timing of biomarker differences.

The time from symptom onset at which biomarker values, or the rate of change of biomarkers, in the amyloid PET positive group were significantly different from those in the amyloid PET negative group was estimated. For values listed as not available (NA), the biomarker values never became significantly different between the amyloid PET positive and negative groups.

Measure	Cross-sectional		Rate of change
	EYO (years)	Amyloid time (years)	EYO (years)	Amyloid time (years)
CSF Aβ42/Aβ40	−19	−4	−17	−3
Plasma Aβ42/Aβ40	−18	−3	NA	NA
CSF pT217/T217 (%)	−18	1	−16	−2
CSF pT181/T181 (%)	−17	−3	−13	2
Amyloid PET	−15	−2	−14	0
CSF Aβ42 (pg/ml)	−15	−1	−17	−3
CSF NfL (pg/ml)	−15	−2	NA	NA
CSF pT231/T231 (%)	−14	1	−13	2
CSF t-tau (pg/ml)	−14	2	−16	8
CSF VILIP-1 (pg/ml)	−14	3	−12	6
CSF p-tau181 (pg/ml)	−13	3	−17	−4
Plasma pT217/T217 (%)	−13	2	−15	−2
Plasma GFAP (pg/ml)	−13	2	NA	NA
CSF Neurogranin (pg/ml)	−13	3	NA	NA
CSF sTREM2 (pg/ml)	−12	2	NA	NA
Plasma NfL (pg/ml)	−12	2	NA	NA
CSF SNAP-25 (pg/ml)	−12	3	−15	2
CSF pT205/T205 (%)	−9	7	−4	7
Plasma pT181/T181 (%)	−8	7	−11	3
Global cognitive composite	−8	7	−1	NA
Hippocampal volume	−8	8	−6	11
CDR-SB	−8	10	NA	NA
CSF YKL-40 (ng/ml)	−8	12	NA	NA
MMSE	−7	10	NA	NA

Abbreviations: Aβ42, amyloid-β 42; CDR-SB, CDR Sum of Boxes; CSF, cerebrospinal fluid; MMSE, Mini-Mental State Examination (score of 30 is best and a score of 0 is worst); NfL, neurofilament light chain; PET, positron emission tomography; p-tau181, tau phosphorylated at site 181; pT181/T181, percentage of tau phosphorylated at threonine 181; pT205/T205, percentage of tau phosphorylated at threonine 205; pT217/T217, percentage of tau phosphorylated at threonine 217; pT231/T231, percentage of tau phosphorylated at threonine 231; SNAP-25, synaptosomal-associated protein-25; sTREM2, soluble triggering receptor expressed on myeloid cells 2; t-tau, total tau; VILIP-1, visinin-like protein 1; YKL-40, chitinase-3-like protein 1, GFAP, glial fibrillary acidic protein.

Rate of change in biomarker values over the course of AD

For individuals in the amyloid PET cohort with longitudinal data (see Supplementary Table 1), the rate of change in key biomarkers was plotted as a function of EYO halfway through the follow-up period (Figure 6) and the timing of significant change was examined (Table 2). Amyloid as measured by PET had a significant rate of increase at −14 years and a maximum rate of change at EYO −5 years (Figure 6A). CSF Aβ42/Aβ40 had a significant rate of decline at EYO −17 years and a maximum rate of change at EYO −11 years (Figure 6B). CSF pT217/T217 had a significant rate of increase at EYO −16 years and a maximum rate of change at EYO −9 years (Figure 6C). CSF pT205/T205 and hippocampal volume demonstrated a different pattern of change, with a later increase in the rate of change (EYO −4 and −6 years, respectively, Figure 6D–E). Because this limited cohort included few individuals with cognitive impairment, the rate of change in CDR-SB could not be rigorously evaluated (Figure 6F). The rate of change in biomarkers as a function of EYO halfway through the follow-up period is also shown for CSF Aβ42, Aβ40, p-tau181, total tau, pT231/T231, pT181/T181 (Supplemental Figure 8); CSF neurogranin, SNAP-25, VILIP-1, NfL, YKL-40, sTREM2 (Supplemental Figure 9); plasma Aβ42/Aβ40, pT217/T217, pT181/T181, NfL and GFAP (Supplemental Figure 10). Similar plots are shown for the various measures as a function of amyloid time (Supplemental Figures 11–14).

Figure 6. Rate of change of biomarkers as a function of estimated years from symptom onset.

For each amyloid positive participant with longitudinal data, the rate of change in selected biomarkers was calculated and plotted as a function of the estimated years from symptom onset halfway through the biomarker follow-up period for amyloid PET Centiloids (A), CSF Aβ42/Aβ40 (B), CSF pT217/T217 (C), CSF pT205/T205 (D), hippocampal volume (E), and CDR-SB (F). Red points represent APOE ε4 carriers and blue points represent APOE ε4 non-carriers. Open points represent cognitively unimpaired individuals (CDR, 0 at the last assessment) and filled points represent cognitively impaired individuals (CDR>0 at the last assessment).

Discussion

We used previously described models based on amyloid PET to visualize and analyze plasma, CSF, imaging, and cognitive measures as a function of estimated years from AD symptom onset on both an individual and cohort level. Although other studies have evaluated the timing of biomarker changes in sporadic AD using amyloid PET as a proxy for time^42–44, or using other types of longitudinal analyses⁴⁵, these studies did not transform the amyloid PET measure into a time scale or evaluate biomarkers as a function of estimated years from symptom onset. The approach demonstrated herein provides an intuitive context for understanding biomarker changes over the time course of AD. Further, it allows for examination of complex biomarker changes including non-linear and non-monotonic trends. Applying a continuous staging measure to sporadic AD, rather than a categorical stage⁴⁶, may facilitate our understanding of AD pathophysiology and improve identification of individuals at high risk for progression to dementia.

In this study of sporadic AD, we found that changes in plasma and CSF biomarkers that reflect amyloidosis (plasma and CSF Aβ42/Aβ40 and CSF pT217/T217), as well as amyloid PET, were detected 15 to 19 years before estimated symptom onset. Changes in CSF biomarkers of neuronal injury (VILIP-1), synaptic dysfunction (neurogranin and SNAP-25), microglial activation (sTREM2), and a plasma biomarker of reactive astrogliosis (GFAP)—were detected between 12–14 years before estimated symptom onset. Changes in plasma and CSF biomarkers of neuroaxonal injury (NfL) were detected 12 and 15 years before estimated symptom onset correspondingly, but after controlling for age, there was no longer significant differences between amyloid positive and negative groups for CSF NfL. Changes in CSF pT205/T205, CSF YKL-40, hippocampal volumes, and cognitive measures were detected 8–9 years before estimated symptom onset.

Although this cohort included limited longitudinal data, the rates of change demonstrated clear differences between some biomarkers, such as an earlier maximal rate of change for CSF Aβ42/Aβ40 as compared to amyloid PET, and an increasing rate of change in pT205/T205 later in AD. These different patterns in the rates of change suggest that these biomarkers are reflecting different processes in AD. For example, CSF Aβ42/Aβ40 likely represents the sequestration of Aβ42 in amyloid plaques while amyloid PET reflects total amyloid plaque burden.¹² CSF pT205/T205 may reflect processes related to accumulation of neurofibrillary tangle pathology and/or neurodegeneration, as its pattern of change is different from CSF Aβ42/Aβ40 and pT217/T217, and is instead more similar to hippocampal volume.¹²

A major limitation of this approach is that estimates for years from symptom onset could only be derived for individuals with at least one amyloid PET scan within a range of values over which amyloid burden changes consistently across individuals: SUVR 1.20 to 3.00 (Centiloids 7 to 88).⁷ Not including individuals with high levels of amyloid burden (SUVR>3.00, Centiloids>88) excluded most individuals with symptomatic AD, reducing the ability to evaluate biomarker changes that occur at or following symptom onset. The use of biomarker clocks that change consistently over a longer time window, both very early in the disease as well as after symptom onset, would increase the data available for analyses and extend the window for evaluating biomarker changes. An additional limitation was that the model for estimated age at symptom onset, as derived in a prior publication, was based on a relatively small number of individuals (n=19) who progressed from cognitively unimpaired to cognitively impaired and includes only estimated age at amyloid onset (SUVR 1.20) and age; sex and APOE genotype were not significant predictors in this model and thus were not included. While this model was created with a small number of informative individuals, this prior publication also showed that the model for estimated age at symptom onset was consistent with a logistic regression model predicting symptoms in a larger cohort of n=180 individuals. Another study using different cohorts and methods found that APOE genotype and sex were significant predictors in some models relating the estimated age at amyloid onset to symptom onset.⁸ Therefore, it is possible that even more accurate estimates could be derived for age at symptom onset.

Because the EYO when measures become significantly different is influenced by factors such as the variance of biomarker assays, the exact timing of change may not be consistent across studies, but the relative ordering of biomarker changes would be expected to be consistent. Some measures, especially plasma pT217/T217, pT181/T181, NfL, and GFAP, had more missing data and therefore estimates may have higher inaccuracy. Tau PET was not represented, given a large amount of missing data; future analyses are needed to evaluate the timing of changes in tau PET. Additionally, the modeling approach assumes that all individuals progress through the disease in a similar fashion, but because AD is heterogeneous and may be influenced by other factors and pathologies, the relative ordering of biomarker changes may not be the same for all individuals. However, the similarities in the timing and ordering of biomarker changes in this study of sporadic AD and studies of ADAD support a similar pathophysiology and time course for most individuals with sporadic and autosomal dominant forms of AD.^{2, 13, 35, 47, 48} Studies in ADAD that directly compare the estimated age at symptom onset based on either an amyloid clock or family history would be informative. Furthermore, studies in more representative populations, including more diverse groups and older individuals, are needed.

Overall, this study demonstrates the feasibility of using estimated years from symptom onset as a continuous staging measure for studying the pathophysiology of sporadic AD. Estimated years from symptom onset based on family history has served as an extremely valuable tool for studying biomarker changes in ADAD and has enabled more efficient clinical trials.^{28, 49, 50} Applying this approach to additional studies of sporadic AD may improve understanding of the timing and magnitude of biomarker changes and could potentially improve identification of cognitively unimpaired individuals at high risk of progression to AD dementia.

Glossary

Aβ

amyloid-β

Aβ42/Aβ40

ratio of amyloid-β peptide 42 to 40

AD

Alzheimer disease

APOE

apolipoprotein E

CDR

Clinical Dementia Rating

CDR-SB

CDR Sum of Boxes

CI

confidence interval

CSF

cerebrospinal fluid

CV

coefficient of variation

EYO

estimated years from symptom onset

GFAP

glial fibrillary acidic protein

IPMS

immunoprecipitation mass spectrometry

MMSE

mini mental state exam

NfL

neurofilament light chain

PET

positron emission tomography

PiB

Pittsburgh compound B

p-tau

phosphorylated tau

SNAP-25

synaptosomal-associated protein-25

SUVR

standardized uptake value ratio

sTREM2

soluble triggering receptor expressed on myeloid cells 2

t-tau

total tau

VILIP-1

visinin-like protein 1

YKL-40

chitinase-3-like protein 1

Data availability

Data are available to qualified investigators upon request to the Knight ADRC (knightadrc.wustl.edu/Research/ResourceRequest.htm).