Emerging cerebrospinal fluid biomarkers in autosomal dominant Alzheimer disease

Abstract

INTRODUCTION:

Four less well-studied but promising “emerging” cerebrospinal fluid (CSF) biomarkers are elevated in late onset Alzheimer disease (AD): neurogranin (Ng), synaptosomal-associated protein-25 (SNAP-25), visinin-like protein 1 (VILIP-1) and chitinase-3-like protein 1 (YKL-40).

METHODS:

CSF Ng, SNAP-25, VILIP-1 and YKL-40 were measured in families carrying autosomal dominant AD (ADAD) mutations.

RESULTS:

The four emerging CSF biomarkers were significantly elevated in the mutation carriers (n=235) versus non-carriers (n=145). CSF SNAP-25, VILIP-1 and YKL-40 were altered very early in the AD time course, approximately 15-19 years before estimated symptom onset. All CSF biomarkers predicted important AD-related outcomes including performance on a cognitive composite, brain amyloid burden as measured by amyloid positron emission tomography and the estimated years from symptom onset.

DISCUSSION:

Early abnormalities in CSF tTau, pTau, SNAP-25, VILIP-1 and YKL-40 suggest that synaptic damage, neuronal injury and neuroinflammation begin shortly after the commencement of brain amyloid accumulation.

1. Background

Alzheimer disease (AD) is the most common cause of dementia in older adults [1] and is characterized neuropathologically by extracellular amyloid plaques comprised primarily of amyloid β-peptide (Aβ), intraneuronal neurofibrillary tangles comprised primarily of the microtubule-associated protein tau, neuronal dysfunction and eventual cell death. Neuropathological and biomarker findings demonstrate that the pathological changes of AD begin more than 10-20 years before the onset of dementia symptoms in the asymptomatic, preclinical phase of the disease [2-5]. Cerebrospinal fluid (CSF) biomarkers of AD are used to diagnose individuals with AD brain pathology in clinical and research settings and are altered both in preclinical and symptomatic AD [6-8]. The best-established CSF biomarkers for AD are the concentrations of the 42 amino acid form of the amyloid-β peptide (Aβ42), total tau (tTau), and phosphorylated tau181 (pTau) [6].

Less well-studied but promising “emerging” CSF biomarkers of other disease processes have been described that are altered in individuals with symptomatic AD compared to controls, including neurogranin (Ng) [9-16], synaptosomal-associated protein-25 (SNAP-25) [17], visinin-like protein 1 (VILIP-1) [18-20] and chitinase-3-like protein 1, also known as YKL-40 [9, 21-23]. Neurogranin is a post-synaptic calmodulin-binding protein that is enriched in dendritic spines and modulates synaptic plasticity and long-term potentiation [24-26]. 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 [27]. Elevated CSF levels of Ng and SNAP-25 are thought to reflect synaptic damage. VILIP-1 is a neuronal calcium sensor protein, and elevated levels may be a marker of neuronal injury [28]. YKL-40 is a secreted glycoprotein that, in the brain, is mainly expressed by astrocytes. It is elevated in a variety of neurological conditions including stroke and multiple sclerosis and is thought to be a general marker of neuroinflammation [29].

In this study, we evaluated these four emerging CSF biomarkers, as well as CSF Aβ42, Aβ42/Aβ40, tTau and pTau, in families with fully-penetrant autosomal dominant Alzheimer Disease (ADAD) mutations. Study of biomarkers in ADAD families is uniquely informative because the age of symptom onset for mutation carriers is predicted by the age at symptom onset in affected family members and the specific ADAD mutation [30, 31]. The estimated years from expected symptom onset (EYO) is calculated as the age of an individual minus the individual’s expected age at symptom onset. Studying biomarkers as a function of EYO allows investigators to determine how biomarkers change over the entire AD time course, including the long preclinical phase [4, 30, 31]. We first examined when the emerging CSF biomarkers became altered in the AD time course. Next, we evaluated how the established and emerging CSF biomarkers, individually and in combination, predicted performance on a cognitive composite and brain amyloid burden as measured by amyloid positron emission tomography (PET). Finally, we assessed whether CSF biomarkers predicted EYO and, therefore, could be useful in staging AD.

2. Methods

2.1. Participants

Individuals with a first-degree family member with a known mutation in Presenilin 1 (PSEN1), Presenilin 2 (PSEN2), or Amyloid Precursor Protein (APP) were enrolled in the Dominantly Inherited Alzheimer Network (DIAN) observational study at one of fourteen sites in the USA, UK, Germany, and Australia. Participants with genetic, clinical, CSF biomarker and neuroimaging data that passed quality control from the 12th semiannual data freeze (last data from June 30, 2017) were included in the analyses. Eighteen individuals carrying the APP Dutch and Flemish mutations were excluded because they manifest a different clinical syndrome (severe cerebral amyloid angiopathy). The institutional review board at Washington University (St. Louis, MO, USA) provided human studies approval. Participants or their caregivers provided written informed consent in accordance with the Declaration of Helsinki and their local institutional review board.

Clinical, biomarker and imaging visits were performed every three years for asymptomatic individuals younger than three years from their familial age of dementia onset, and every year for individuals who either were three or fewer years from their familial age of dementia onset or had dementia symptoms.

2.2. Procedures

Clinical dementia status was assessed using the Clinical Dementia Rating (CDR) scale, which also yields the Clinical Dementia Rating-Sum of Boxes (CDR-SB) [32]. Clinicians assigning the CDR were blinded to the mutation status of participants. The Mini-Mental State Examination (MMSE) [33] was completed as a part of the clinical assessment. The DIAN cognitive composite consists of the delayed recall score from the DIAN Word List Test, the Logical Memory delayed recall score from the Wechsler Memory Scale-Revised (WMS-R), the Digit Symbol Coding test total score from the Wechsler Adult Intelligence Scale-Revised, and the MMSE total score [34]. Scores from each test were converted to z-scores using the baseline mean and standard deviation of non-carriers and averaged to form a composite z-score.

The presence or absence of an ADAD mutation was determined using PCR-based amplification of the appropriate exon followed by Sanger sequencing [4]. The participant's estimated years from expected symptom onset (EYO) was defined as the participant’s age minus their estimated age at symptom onset [4]. The estimated age at symptom onset was based on an individual’s actual age of symptom onset if they were a symptomatic mutation carrier (MC), as well as the age at symptom onset for family members. For the asymptomatic MC and the non-carriers (NC), the estimated age at expected symptom onset was calculated based on the participant's age relative to either the family mutation-specific expected age at dementia onset [30] or parental age at first progressive cognitive decline if the expected age at symptom onset for the mutation was unknown.

CSF was collected in accordance with the Alzheimer’s Disease Neuroimaging Initiative (ADNI) protocols. Briefly, ~15 mL CSF was collected in polypropylene tubes at 8:00 am following overnight fasting and was immediately frozen on dry ice. Frozen samples were shipped to the DIAN Biomarker Core, thawed, aliquoted (0.5 ml), flash-frozen and stored at − 84°C until analysis.

2.3. CSF biomarker assays

The assays used for Ng, SNAP-25, and VILIP-1 have been described in detail elsewhere [35]. Briefly, these analytes were measured with microparticle-based immunoassays using Single Molecule Counting [SMC]^™ technology originally developed for the Erenna® System by Singulex® and now part of EMD Millipore (Burlington, MA), and employed antibodies developed in the laboratory of Dr. Jack Ladenson at Washington University in St. Louis. YKL-40 was measured with a plate-based enzyme-linked immunoassay (MicroVue ELISA; Quidel, San Diego, CA) [36]. Within-person longitudinal samples were run on the same assay plate.

Aβ42, Aβ40, tTau, and pTau were measured with the corresponding Elecsys immunoassays on the automated Roche cobas e 601 analyzer in the laboratory of Leslie Shaw in the ADNI Biomarker Core at the University of Pennsylvania as previously described [37, 38]. The Elecsys immunoassays are electrochemiluminescence immunoassays employing a quantitative sandwich principle. A single lot of immunoassay reagents for each analyte was used to measure all samples.

2.4. Brain imaging

Structural brain MRI was performed according to the ADNI protocol [39] and is described in detail elsewhere [40]. T1-weighted images were acquired on a 3 Tesla scanner. Volumetric segmentation and cortical surface reconstruction were performed using FreeSurfer 5.3 and subcortical volumes were corrected for intracranial volume using a regression approach. Cortical thickness and volume measures were averaged across hemispheres.

Amyloid PET was performed with ¹¹C-Pittsburgh Compound B (¹¹C-PiB) as previously described [40]. Data was converted to regional standardized uptake value ratios (SUVRs) relative to the cerebellar grey matter and partial volume corrected using a regional spread function technique.

2.5. Statistical analysis

The mean differences between the MC and NC were calculated at each EYO point and standardized [4]. Results were inverted (multiplied by negative 1) for measures where lower levels represent abnormality (Aβ42/Aβ40, Aβ42, MMSE, cognitive composite, hippocampal volume, precuneus thickness and entorhinal cortical thickness) so that the degree of abnormality could be compared more directly. Correlations between the baseline values of CSF biomarkers were estimated using bivariate linear mixed effects (LME) models treating families as random effects and adjusting for covariates including baseline age, sex, years of education and APOE ε4 status [41].

To determine the EYO at which the MC first became significantly different from the NC in the baseline level and rate of intra-individual change for each measure, linear or linear spline mixed effects models with one knot (depending on which model better fit the data) were used to evaluate the intra-individual rate of change [42]. The mutation group (MC or NC), baseline EYO, time, and all possible two-way or three-way interactions among them were included as fixed effects. Sex, years of education, and APOE ε4 status were considered as covariates, but only those effects that were significant were retained in the models. Random slopes and intercepts for each individual with an unstructured covariance matrix were included to account for the within-subject correlation due to repeated measures, and random intercepts were used to account for correlation within the familial cluster. The adjusted difference in the mean level at baseline and difference in the rate of change between the MC and NC was tested using the approximate t-test derived from the models to determine the first EYO point where the difference became significant.

To evaluate the ability of baseline CSF biomarkers individually or in combination to predict outcomes, LME models with random intercepts for familial cluster were utilized. For each outcome measure, a base model was built using significant predictors among potential covariates: baseline age, sex, years of education and APOE ε4 status. Baseline CSF biomarkers were then added to the model as predictors one at a time, then with Aβ42, then with Aβ42 and pTau. The significance of the coefficients for the CSF biomarker(s) were used to determine their prediction performance, in addition to the predictors in the base models and the retained CSF biomarkers.

To prevent inadvertent disclosure of mutation status to research participants, plots do not include the EYO scale, points from EYO outliers are omitted, and points after three data collections are not shown. However, all data was included in the analyses. All analyses were conducted using SAS 9.4 (SAS Institute Inc., Cary, NC). A p-value <0.05 was considered statistically significant.

3. Results

3.1. Baseline characteristics of study participants

The baseline characteristics of the study participants are detailed in Table 1. Notably, the 235 mutation carriers (MC) included individuals expected to have widely varying levels of AD brain pathology, while the 145 non-carriers (NC) consisted mainly of cognitively normal individuals expected to have no or very little AD brain pathology due to their relatively young age. At their baseline CSF collection, the MC had a CSF profile consistent with AD brain pathology, with lower Aβ42 and Aβ42/Aβ40 and higher tTau and pTau compared to the NC (p<0.0001). The emerging CSF biomarkers (Ng, SNAP-25, VILIP-1 and YKL-40) were all higher in the MC versus the NC (p<0.0001). Most brain imaging, clinical, and cognitive measures performed within one year of the baseline CSF collection also demonstrated significant abnormalities in the MC compared to NC.

Table 1. Characteristics of study participants at baseline.

Continuous measures are presented as the mean ± standard deviation. For each type of measure, the number of individuals with at least one measurement is provided. The significance of the difference between the MC and NC was calculated by linear mixed effects models with random intercepts for familial clusters.

Characteristic	All (n=380)	Mutation carrier (MC) (n=235)	Mutation Non- carrier (NC) (n=145)	p-value
Estimated years from symptom onset	−8.2 ± 11.6	−7.6 ± 10.8	−9.2 ± 12.7	N.S.
Participants with longitudinal CSF, n (%)	160 (42%)	105 (44%)	55 (38%)	N.S.
Interval between longitudinal LPs (years)	2.9 ± 1.4	2.8 ± 1.4	3.20 ± 1.5	N.S.
Age (years)	38.6 ± 11.1	38.4 ± 10.4	38.8 ± 12.1	N.S.
Sex, n (% female)	216 (57%)	127 (54%)	89 (61%)	N.S.
Education (years)	14.6 ± 2.9	14.3 ± 3.1	14.9 ± 2.5	N.S.
Presence of APOE ε4 allele, n (%)	121 (32%)	72 (31%)	49 (34%)	N.S.
Established CSF biomarkers (n = 380)
Aβ42 (pg/ml)	1123 ± 605	961 ± 623	1385 ± 471	<0.0001
Aβ42/Aβ40	0.07 ± 0.03	0.06 ± 0.03	0.09 ± 0.01	<0.0001
tTau (pg/ml)	243.6 ± 141.9	287.9 ± 159.1	171.6 ± 58.5	<0.0001
pTau (pg/ml)	24.5 ± 19.8	30.8 ± 22.7	14.2 ± 5.3	<0.0001
Emerging CSF biomarkers (n = 358)
Ng (pg/ml)	2008 ± 1096	2269 ± 1189	1572 ± 741	<0.0001
SNAP-25 (pg/ml)	4.3 ± 1.8	4.6 ± 1.9	3.7 ± 1.3	<0.0001
VILIP-1 (pg/ml)	158.2 ± 71.5	173.4 ± 77.9	132.9 ± 50.2	<0.0001
YKL-40 (ng/ml)	159.3 ± 83.4	173.1 ± 88.7	136.4 ± 68.2	<0.0001
Amyloid PET (n = 319)
Amyloid PET SUVR	1.67 ± 1.00	2.06 ± 1.12	1.06 ± 0.17	<0.0001
Structural brain measures (n = 356)
Hippocampal volume (mm3)	8492 ± 1054	8310 ± 1201	8779 ± 676	<0.0001
Entorhinal cortical thickness (mm)	3.44 ± 0.34	3.43 ± 0.35	3.46 ± 0.32	N.S.
Precuneus thickness (mm)	2.31 ± 0.2	2.27 ± 0.23	2.38 ± 0.14	<0.0001
Clinical measures (n = 380)
CDR>0, n (%)	89 (23%)	83 (35%)	6 (4%)	<0.0001
CDR-SB	0.86 ± 2.33	1.37 ± 2.84	0.04 ± 0.24	<0.0001
Cognitive measures (n = 380)
DIAN cognitive composite	−0.37 ± 0.92	−0.60 ± 1.03	−0.02 ± 0.59	<0.0001
MMSE	27.6 ± 4.4	26.7 ± 5.3	29.1 ± 1.2	<0.0001

Abbreviations: Aβ40, amyloid-β 40; Aβ42, amyloid-β 42; CDR, Clinical Dementia Rating; CDR-SB, Clinical Dementia Rating, Sum of Boxes; CSF, cerebrospinal fluid; DIAN, Dominantly Inherited Alzheimer Network; LP, lumbar puncture; MMSE, Mini-Mental State Examination; Ng, neurogranin; N.S., not significant; PET, positron emission tomography; pTau, phosphorylated tau181; SNAP-25, synaptosomal-associated protein-25; SUVR, standardized uptake value ratio; tTau, total tau; VILIP-1, visinin-like protein 1; YKL-40, chitinase-3-like protein 1.

3.2. CSF, imaging, clinical and cognitive measures by EYO

The unadjusted (not adjusted for covariates) levels of all CSF, imaging, clinical and cognitive measures as a function of EYO in the MC and NC were examined (Fig. 1). To visualize the order of biomarker changes, the difference between the MC and NC for the baseline unadjusted standardized levels of the measures was calculated and plotted as a function of EYO, where the y-axis indicates the degree of abnormality (Fig. 2). These plots suggest that amyloid PET as well as the established and emerging CSF biomarkers all change many years prior to the estimated age at expected symptom onset. Because ADAD mutations cause the over-production of Aβ42, CSF Aβ42 levels were higher in the MC versus the NC prior to significant amyloid deposition in the brain, but as brain amyloid burden increased, Aβ42 levels in the MC became equivalent and then lower in the MC versus the NC. The structural brain, clinical, and cognitive measures appear to change shortly before the onset of symptoms (Fig. 2, bottom panel).

Fig. 1. CSF, imaging, clinical and cognitive measures as a function of EYO in the ADAD mutation carriers and non-carriers.

The unadjusted levels of each measure are shown as a function of EYO. Values for the mutation carriers are shown in red and values for the noncarriers are shown in blue. Each point represents one measurement. Thin lines between points connect measurements within the same individual. Thick lines are the LOESS curves for all values (red for the mutation carriers and blue for the non-carriers) and are depicted to aid with visualization of trends in the data. For Aβ42, tTau, pTau, Ng, SNAP-25, and VILIP-1, units are in pg/ml. For YKL-40, units are in ng/ml. For amyloid PET, units are in mean cortical SUVR. For hippocampus, units are in mm³. For entorhinal cortex and precuneus, units are in mm. The cognitive composite is a Z-score, with a mean of 0 and units in standard deviations from the mean. CDR, CDR-SB and MMSE are the raw test score. To prevent inadvertent disclosure of mutation status to research participants, the plot does not include the EYO scale, points from EYO outliers were omitted, and points after >3 data collections are not shown.

Fig. 2. Comparison of baseline CSF, imaging, clinical and cognitive measures as a function of EYO.

The difference between values in the ADAD mutation carriers and non-carriers for each unadjusted standardized measure at baseline is shown as a function of EYO. The y-axis represents the degree of abnormality consistent with AD pathology so that trajectories are more comparable. Lines are LOESS curves for the values. To prevent inadvertent disclosure of mutation status to research participants, plots do not include the EYO scale.

3.3. Divergence in baseline levels and rates of change for CSF, imaging, clinical and cognitive measures

Linear mixed effects models were used to determine the EYO when CSF, imaging, clinical and cognitive measures first became significantly more abnormal in the MC versus the NC (Table 2). Baseline brain amyloid burden by amyloid PET became significantly higher in the MC versus the NC at EYO −22. Baseline CSF Aβ42/Aβ40 and Aβ42 became significantly lower in the MC versus the NC at EYO −17 and −14, respectively. Baseline levels of CSF tTau, pTau, SNAP-25, VILIP-1 and YKL-40 became significantly higher in the MC versus the NC at EYO −19 to −15. In contrast, baseline structural brain measures did not become significantly smaller and baseline clinical and cognitive measures did not become significantly worse in the MC versus the NC until just prior to symptom onset, at EYO −3 to 0. There was no EYO at which baseline CSF Ng was significantly higher or entorhinal cortical thickness was significantly lower in the MC versus the NC because of high variance in values.

Table 2. Divergence in measures between the ADAD mutation carriers and non-carriers.

Estimated differences between the ADAD mutation carriers (MC) and non-carriers (NC) in the baseline level and intra-individual rate of change for each measure were generated from linear or linear spline mixed effects models. The fixed effects included in the models were baseline EYO (which was highly correlated with baseline age), mutation status (MC or NC), time, and all possible two-way or three-way interactions among the fixed effects. Sex, years of education, and APOE ε4 status were considered as covariates, but only those covariates that were significant were retained in the final models. Random effects included random intercepts for familial clusters as well as individual random intercept and slope for repeated measures. Baseline measures were differences in the mean level ± standard error. Longitudinal measures were differences in the annualized rate of change ± standard error. Only significant differences (p<0.05) are shown below. If the baseline level or intra-individual rate of change did not vary significantly between the MC and NC, the time of divergence is reported as N.A. (not applicable) for that measure.

	Baseline level		Intra-individual annual rate of change
Characteristic	Time of divergence (EYO)	Difference in mean level (MC-NC)	Time of divergence (EYO)	Difference in rate of change (MC-NC)
Established CSF biomarkers
tTau (pg/ml)	−19	35.7 ± 15.3	N.A.	N.A.
pTau (pg/ml)	−18	4.91 ± 2.03	N.A.	N.A.
Aβ42/Aβ40	−17	−0.006 ± 0.003	−25	−0.005 ± 0.0008
Aβ42 (pg/ml)	−14	−210 ± 94	−19	−39.3 ± 19.2
Emerging CSF biomarkers
VILIP-1 (pg/ml)	−18	16.9 ± 8.3	−25	7.47 ± 2.69
SNAP-25 (pg/ml)	−15	0.43 ± 0.18	N.A.	N.A.
YKL-40 (ng/ml)	−15	16.1 ± 7.7	N.A.	N.A.
Ng (pg/ml)	N.A.	N.A.	N.A.	N.A.
Amyloid PET
Amyloid PET SUVR	−22	0.19 ± 0.09	−23	0.04 ± 0.02
Structural brain measures
Precuneus thickness (mm)	−3	−0.049 ± 0.022	−4	−0.021 ± 0.007
Hippocampal volume (mm3)	−1	−449 ± 140	−2	−126 ± 44
Entorhinal cortical thickness (mm)	N.A.	N.A.	−1	−0.057 ± 0.028
Clinical measures
CDR-SB	0	0.70 ± 0.26	−11	0.215 ± 0.100
CDR	−2	0.13 ± 0.05	−2	0.060 ± 0.025
Cognitive measures
DIAN cognitive composite	−3	−0.33 ± 0.10	−3	−0.102 ± 0.033
MMSE	−2	−1.77 ± 0.60	−12	−0.34 ± 0.16

Abbreviations: Aβ40, amyloid-β 40; Aβ42, amyloid-β 42; ADAD, autosomal dominant Alzheimer disease; CDR, Clinical Dementia Rating; CDR-SB, Clinical Dementia Rating, Sum of Boxes; CSF, cerebrospinal fluid; DIAN, Dominantly Inherited Alzheimer Network; EYO, estimated years from expected symptom onset; LP, lumbar puncture; MC, mutation carriers; MMSE, Mini-Mental State Examination; N.A., not applicable; NC, mutation non-carriers; Ng, neurogranin; PET, positron emission tomography; pTau, phosphorylated tau181; SNAP-25, synaptosomal-associated protein-25; SUVR, standardized uptake value ratio; tTau, total tau; VILIP-1, visinin-like protein 1; YKL-40, chitinase-3-like protein 1.

The intra-individual rates of change for the CSF, imaging, clinical and cognitive measures as a function of EYO became significantly different in the MC versus the NC at some point for most measures (Table 2, see Supplementary Table 1 for detailed analyses on rates of change as a function of EYO). CSF Aβ42/Aβ40 and Aβ42 began declining, and VILIP-1 levels began increasing more rapidly, in the MC versus the NC very early in the AD time course, at EYO −25, −19, and −25, respectively. Given the complexity of Aβ42 levels in the MC at early EYOs due to mutation related over-production, the rate of change of Aβ42 or Aβ42/Aβ40, rather than the absolute level, may be a more reliable indicator of when Aβ42 levels start to change in the MC. Around the same time that the rates of change of CSF biomarkers became different in the two groups, the MC began accumulating brain amyloid more quickly than the NC as measured by amyloid PET (EYO −23). The first changes in clinical and cognitive measures were observed around EYO −12 or −11, but this may be an artifact of the ceiling effect and floor effect of MMSE and CDR-SB, respectively, since at this EYO nearly all individuals scored at or near the maximum score for MMSE and minimum score for CDR-SB. The MC did not develop significantly faster rates of brain atrophy or more rapid decline on the DIAN cognitive composite or global CDR versus the NC until shortly before symptom onset (EYO −4 to −1).

3.4. Baseline CSF biomarkers as predictors of imaging, clinical and cognitive measures

The ability of baseline levels of CSF biomarkers to predict baseline levels and intra-individual rates of change of structural brain, clinical and cognitive measures was evaluated. In the NC, baseline CSF biomarker levels did not significantly predict either the baseline levels or rates of change of any of the imaging, clinical or cognitive measures (data not shown). However, in the MC, most CSF biomarkers were highly predictive of baseline values for most measures, including the DIAN cognitive composite, amyloid PET SUVR, and hippocampal volume (Table 3, see Supplementary Table 2 for other imaging, clinical and cognitive measures). Additionally, most CSF biomarkers were highly predictive of the rate of change of structural brain, clinical, and cognitive measures; the rate of change of amyloid PET was an exception and was significantly predicted by only tTau, pTau and Ng (Table 3).

Table 3. Baseline CSF biomarkers as predictors of the difference between the ADAD mutation carriers and non-carriers in the baseline level and intra-individual rate of change of imaging, clinical and cognitive measures.

Linear mixed effects models were used to estimate the amount of change in imaging, clinical and cognitive measures per unit change in CSF biomarkers. Fixed effects included in the models were biomarker baseline level, mutation status (MC or NC), time, and all possible two-way or three way-interaction among the fixed effects. Baseline age, sex, years of education, APOE ε4 status and their interaction with mutation status were considered as covariates, but only those covariates that were significant were retained in the models. Random effects included random intercepts for familial clusters, as well as individual random intercept and slope for repeated measures. The results for entorhinal cortical thickness, precuneus thickness, CDR, CDR-SB and MMSE are shown in Supplementary Table 2.

Baseline level		Aβ42 (pg/ml)	Aβ42/Aβ40	tTau (pg/ml)	pTau (pg/ml)	Ng (pg/ml)	SNAP-25 (pg/ml)	VILIP-1 (pg/ml)	YKL-40 (ng/ml)
DIAN cognitive composite	Estimate	0.0003	6.9	−0.002	−0.01	−0.0001	−0.1	−0.002	−0.003
	S.E.	0.00008	1.6	0.0003	0.002	0.00004	0.03	0.0006	0.0007
	p-value	0.0002	<0.0001	<0.0001	<0.0001	0.003	<0.0001	0.0009	0.0004
Amyloid PET SUVR	Estimate	−0.0007	−11.9	0.001	0.01	0.0001	0.05	0.002	0.003
	S.E.	0.00009	1.7	0.0004	0.003	0.00005	0.03	0.0008	0.0008
	p-value	<0.0001	<0.0001	0.0007	0.0001	0.04	N.S.	0.004	0.0003
Hippocampal volume (mm3)	Estimate	0.2	4097.2	−2.2	−16.4	−0.2	−97.1	−2.3	−4.6
	S.E.	0.1	2235.5	0.4	3.1	0.06	39.2	0.9	0.9
	p-value	0.08	0.07	<0.0001	<0.0001	0.008	0.02	0.01	<0.0001
Rate of change		Aβ42 (pg/ml)	Aβ42/Aβ40	tTau (pg/ml)	pTau (pg/ml)	Ng (pg/ml)	SNAP-25 (pg/ml)	VILIP-1 (pg/ml)	YKL-40 (ng/ml)
DIAN cognitive composite	Estimate	0.00009	1.7	−0.0007	−0.005	−0.00007	−0.04	−0.0009	−0.0009
	S.E.	0.00003	0.6	0.0001	0.0009	0.00002	0.01	0.0003	0.0002
	p-value	0.008	0.004	<0.0001	<0.0001	0.007	0.0003	0.007	0.0003
Amyloid PET SUVR	Estimate	−0.00002	−0.2	0.0003	0.002	0.00004	−0.005	0.0003	0.00003
	S.E.	0.00002	0.4	0.0001	0.0006	0.00001	0.009	0.0002	0.0002
	p-value	N.S.	N.S.	0.002	0.009	0.002	N.S.	N.S.	N.S.
Hippocampal volume (mm3)	Estimate	0.2	2662.4	−0.9	−6.1	−0.09	−66.1	−1.3	−1.6
	S.E.	0.03	588.5	0.1	0.9	0.02	13.8	0.3	0.2
	p-value	<0.0001	<0.0001	<0.0001	<0.0001	<0.0001	<0.0001	<0.0001	<0.0001

Abbreviations: Aβ40, amyloid-β 40; Aβ42, amyloid-β 42; ADAD, autosomal dominant Alzheimer disease; CSF, cerebrospinal fluid; CDR, Clinical Dementia Rating; CDR-SB, Clinical Dementia Rating, Sum of Boxes; DIAN, Dominantly Inherited Alzheimer Network; MMSE, Mini-Mental State Examination; Ng, neurogranin; N.S., not significant; PET, positron emission tomography; pTau, phosphorylated tau181; S.E., standard error; SNAP-25, synaptosomal-associated protein-25; SUVR, standardized uptake value ratio; tTau, total tau; VILIP-1, visinin-like protein 1; YKL-40, chitinase-3-like protein 1.

3.5. Combinations of CSF biomarkers as predictors of cognitive performance, brain amyloid burden and EYO

Pair-wise correlations between the baseline CSF biomarkers in the MC were determined (Table 4). The correlation between CSF Aβ42 and the other CSF biomarkers was the lowest, suggesting that Aβ42 may be the least redundant with the other biomarkers. The ability of combinations of baseline CSF biomarkers to predict the baseline DIAN cognitive composite and amyloid PET SUVR was evaluated in the MC. These two measures were chosen for further analyses because they were strongly predicted by CSF biomarkers and were well modeled by linear relationships. Baseline levels of all CSF biomarkers individually predicted the baseline DIAN cognitive composite in the MC (Supplemental Table 3). In combination with Aβ42, all other CSF biomarkers (tTau, pTau, Ng, SNAP-25, VILIP-1 or YKL-40) improved prediction of the baseline DIAN cognitive composite. When combined with both Aβ42 and pTau, Ng or SNAP-25 further improved prediction significantly, although it was a very small improvement (adjusted R² of 0.566 and 0.572, respectively, versus 0.555).

Table 4. Correlations between baseline CSF biomarkers in ADAD mutation carriers.

The upper number in the cell is the correlation coefficient (ρ) estimated using a bivariate linear mixed model that accounts for familial clusters. The lower number in the cell is the significance of the correlation. Calculations of significance were adjusted for baseline age, years of education, sex and APOE ε4 status.

	tTau	pTau	Ng	VILIP-1	SNAP-25	YKL-40	Aβ42/Aβ40	Aβ42
tTau	1	0.955	0.868	0.832	0.731	0.616	−0.690	−0.426
		<0.0001	<0.0001	<0.0001	<0.0001	<0.0001	<0.0001	<0.0001
pTau		1	0.774	0.744	0.681	0.587	−0.698	−0.498
			<0.0001	<0.0001	<0.0001	<0.0001	<0.0001	<0.0001
Ng			1	0.844	0.655	0.568	−0.605	−0.222
				<0.0001	<0.0001	<0.0001	<0.0001	0.009
VILIP-1				1	0.744	0.567	−0.515	−0.157
					<0.0001	<0.0001	<0.0001	0.08
SNAP-25					1	0.479	−0.579	−0.355
						<0.0001	<0.0001	<0.0001
YKL-40						1	−0.686	−0.516
							<0.0001	<0.0001
Aβ42/Aβ40							1	0.817
								<0.0001
Aβ42								1

Abbreviations: Aβ40, amyloid-β 40; Aβ42, amyloid-β 42; ADAD, autosomal dominant Alzheimer disease; CSF, cerebrospinal fluid; EYO, estimated years from expected symptom onset; Ng, neurogranin; pTau, phosphorylated tau181; SNAP-25, synaptosomal-associated protein-25; tTau, total tau; VILIP-1, visinin-like protein 1; YKL-40, chitinase-3-like protein 1.

In the MC, the baseline amyloid PET SUVR was predicted by all baseline CSF biomarkers except Ng or SNAP-25 (Supplementary Table 4). Prediction of baseline amyloid PET SUVR was improved by the combination of CSF Aβ42 with either tTau, pTau, VILIP-1 or YKL-40 (adjusted R² of 0.314, 0.316, 0.332, and 0.328, respectively, versus 0.296). The combination of CSF Aβ42, pTau and a third CSF biomarker was no better at predicting amyloid PET SUVR than the combination of Aβ42 and pTau.

The ability of baseline CSF biomarkers to predict EYO was also examined (Supplementary Table 5). EYO was significantly predicted by Aβ42, Aβ42/Aβ40, tTau, pTau, Ng, SNAP-25, or VILIP-1, but not YKL-40. In combination with Aβ42, addition of tTau, Ng, SNAP-25 or VILIP-1 significantly improved prediction of EYO. When combined with Aβ42 and pTau, no additional biomarker improved prediction of EYO.

4. Discussion

Despite some data suggesting that amyloid PET positivity lags behind abnormalities in CSF biomarkers [43], we found that amyloid PET SUVR changes a small but measurable amount very early in the AD time course (EYO −22), shortly before absolute levels of the established CSF biomarkers (Aβ42, tTau and pTau) show significant alterations. In this study, the EYO at which the established CSF biomarkers became significantly more abnormal in the MC versus the NC (EYO −19 to −14) was approximately four years earlier than in previous reports [4, 42, 44-46], likely because of increased precision with the automated Elecsys immunoassays and increased statistical power due to the larger number of samples analyzed. Interestingly, CSF SNAP-25, VILIP-1 and YKL-40 also changed very early in the AD time course (EYO −18 to −15). These findings suggest that the processes reflected by these emerging CSF biomarkers, as well as by CSF tTau and pTau, and which include synaptic damage, neuronal injury and neuroinflammation, begin at or shortly after the commencement of amyloid accumulation as defined by amyloid PET. In contrast to hypothetical models that propose a significant lag period between amyloid accumulation and neuronal injury [47], amyloid deposition appears to be associated with other pathological changes almost immediately.

One major limitation of our study is that findings in ADAD may not be generalizable to typical sporadic late onset AD (LOAD). This is particularly true for findings involving CSF Aβ42 because ADAD mutations cause over-production of Aβ42 [48]. In a LOAD cohort the differences between CSF biomarkers in individuals with and without AD brain pathology may be different because of the increased prevalence of other neurological diseases or conditions in older individuals that may alter CSF biomarker levels. Another limitation is that we did not evaluate all promising CSF biomarkers such as neurofilament light chain (NfL) or soluble triggering receptor expressed on myeloid cells 2 (sTREM2) [49, 50], which have been measured [49] or are in the process of being measured in baseline samples from a smaller cohort of families with ADAD. Finally, some of the results, especially the EYO at which measures became significantly different in the MC versus the NC, are sensitive to a variety of factors including measurement precision, power (number of measurements), and statistical modeling approach, and therefore different studies may reach different conclusions about the exact timing of biomarker changes.

Despite these limitations, our work has important implications for the field. We have found that CSF Ng, SNAP-25, VILIP-1 and YKL-40 are all significantly altered in ADAD mutation carriers and predict important AD-related outcomes measures. Relative to a single emerging CSF biomarker, the combination of one emerging CSF biomarker with CSF Aβ42 often improved prediction of performance on a cognitive composite or brain amyloid burden as defined by amyloid PET. In this side-by-side comparison, one emerging biomarker was not consistently superior in the prediction of a variety of outcome measures. Instead, all four of the emerging CSF biomarkers could potentially substitute for tTau and pTau as markers of AD brain pathology in studies where tau-related biomarkers may be directly affected (e.g. drug trials of anti-tau antibodies where CSF biomarkers are outcome measures). Finally, a unique strength of studying biomarkers in ADAD is that it provides reliable estimates of where individuals are located in the AD time course (estimated years from symptom onset, EYO). Importantly, we found that CSF biomarkers predict EYO. Clinical trials for AD, particularly prevention trials, may benefit from including AD stage as estimated by CSF or imaging biomarkers as a factor in their analyses because different drugs may be efficacious at different stages of the disease.

Highlights.

• Ng, SNAP-25, VILIP-1 and YKL-40 are emerging CSF biomarkers of Alzheimer disease

• The emerging CSF biomarkers are elevated in autosomal dominant Alzheimer disease

• CSF SNAP-25, VILIP-1 and YKL-40 increase 15-19 years before estimated symptom onset

• The emerging CSF biomarkers predict Alzheimer disease-related outcomes

• CSF biomarkers predict estimated years from symptom onset in ADAD mutation carriers

Research in Context.

Systematic review:

Literature was reviewed on the association of late onset Alzheimer disease (AD) and CSF levels of neurogranin (Ng), synaptosomal-associated protein-25 (SNAP-25), visinin-like protein 1 (VILIP-1) and chitinase-3-like protein 1 (YKL-40).

Interpretation:

CSF SNAP-25, VILIP-1 and YKL-40 were significantly elevated in autosomal dominant AD mutation carriers approximately 15-19 years before estimated symptom onset, suggesting that the pathological processes reflected by these biomarkers begin very early in the AD time course. Additionally, all four CSF biomarkers predicted important AD-related outcomes including performance on a cognitive composite, brain amyloid burden as measured by amyloid positron emission tomography and the estimated years from symptom onset.

Future directions:

CSF Ng, SNAP-25, VILIP-1 and YKL-40 may be useful biomarkers of AD brain pathology in drug trials. CSF biomarkers may help stage preclinical AD, which could enable investigators to examine whether different therapies are more efficacious at different points in the AD time course.

Abbreviations

Aβ40

amyloid-β 40

Aβ42

amyloid-β 42

AD

Alzheimer disease

ADAD

autosomal dominant Alzheimer disease

ADNI

Alzheimer’s Disease Neuroimaging Initiative

CDR

Clinical Dementia Rating

CDR-SB

Clinical Dementia Rating, Sum of Boxes

CSF

cerebrospinal fluid

DIAN

Dominantly Inherited Alzheimer Network

EYO

estimated years from expected symptom onset

LOAD

late onset Alzheimer disease

LP

lumbar puncture

MC

mutation carrier

MMSE

Mini-Mental State Examination

NC

mutation non-carrier

Ng

neurogranin

N.A.

not applicable

N.S.

not significant

PET

positron emission tomography

pTau

phosphorylated tau181

SNAP-25

synaptosomal-associated protein-25

SUVR

standardized uptake value ratio

tTau

total tau

VILIP-l

visinin-like protein 1

YKL-40

chitinase-3-like protein 1