Lumipulse-Measured Cerebrospinal Fluid Biomarkers for the Early Detection of Alzheimer Disease

Abstract

Background and Objectives

CSF biomarkers of Aβ42 and phosphorylated tau (p-tau181) are used clinically for the detection of Alzheimer disease (AD) pathology during life. CSF biomarker validation studies have largely used clinical diagnoses and/or amyloid PET imaging as the reference standard. The few existing CSF-to-autopsy studies have been restricted to late-stage AD. This CSF-to-autopsy study investigated associations between CSF biomarkers of AD and AD neuropathologic changes among brain donors who had normal cognition at the time of lumbar puncture (LP).

Methods

This was a retrospective study of brain donors from the National Alzheimer's Coordinating Center who had normal cognition at the time of LP and who had measurements of CSF Aβ42 and p-tau181 performed with Lumipulse assays. All brain donors were from Washington University Knight ADRC. Staging of AD neuropathologic change (ADNC) was made based on National Institute on Aging–Alzheimer's Association criteria. For this study, participants were divided into 2 categories: “AD−” (no AD/low ADNC) and “AD+” (intermediate/high ADNC). Accuracy of each biomarker for discriminating AD status was evaluated using area under the curve (AUC) statistics generated using predicted probabilities from binary logistic regressions that controlled for age, sex, APOE ε4, and interval between LP and death.

Results

The average age at LP was 79.3 years (SD = 5.6), and the average age at death was 87.1 years (SD = 6.5). Of the 49 brain donors, 24 (49%) were male and 47 (95.9%) were White. 20 (40.8%) had autopsy-confirmed AD. The average interval from LP until death was 7.76 years (SD = 4.31). CSF p-tau181/Aβ42 was the optimal predictor of AD, having excellent discrimination accuracy (AUC = 0.97, 95% CI 0.94–1.00, p = 0.003). CSF p-tau181 alone had the second-best discrimination accuracy (AUC = 0.92, 95% CI 0.84–1.00, p = 0.001), followed by CSF Aβ42 alone (AUC = 0.92, 95% CI 0.85–1.00, p = 0.007), while CSF t-tau had the numerically lowest discrimination accuracy (AUC = 0.87, 95% CI 0.76–0.97, p = 0.005). Effects remained after controlling for prevalent comorbid neuropathologies. CSF p-tau181/Aβ42 was strongly associated with CERAD ratings of neuritic amyloid plaque scores and Braak staging of NFTs.

Discussion

This study supports Lumipulse-measured CSF Aβ42 and p-tau181 and, particularly, the ratio of p-tau181 to Aβ42, for the early detection of AD pathophysiologic processes.

Classification of Evidence

This study provides Class II evidence that Lumipulse measures of p-tau181/Aβ42 in the CSF accurately discriminated cognitively normal participants with and without Alzheimer disease neuropathologic change.

Introduction

CSF p-tau181_, Aβ42, and total tau (t-tau) biomarkers are used clinically for the detection of Alzheimer disease (AD) pathology during life.^1,2 The CSF biomarker profile of AD includes decreased Aβ42, increased levels of p-tau181, and elevated p-tau181/Aβ42 ratio.³ CSF biomarkers of Aβ42 and p-tau181 have been extensively studied against clinical and other biomarker end points (i.e., PET imaging and magnetic resonance imaging).^3-7 They accurately differentiate participants with and without cognitive impairment, predict clinical progression (i.e., cognitive decline and diagnostic conversion), and correspond to cross-sectional and longitudinal changes of other gold-standard biomarkers of amyloidosis, p-tau, and neurodegeneration.^8-10 The elevated p-tau181/Aβ42 ratio has been shown to be the best predictor of AD.^11,12

Clinical-pathologic correlation studies are necessary for biomarker development and validation. There are few CSF-to-autopsy studies in AD. Many of the existing CSF-to-autopsy studies were performed between 2003 and 2009 using population-based and biobank samples and were restricted to CSF sampling among patients with advanced-stage disease.^13-16 Discrepancies in those studies exist. For example, some studies found an association between CSF Aβ and tau and AD neuropathologic changes (ADNCs), such as neuritic plaques and neurofibrillary tangles.^13,14,16 Yet, other studies found no association of CSF Aβ42 and tau with neuritic plaques and neurofibrillary tangles.¹⁵

More recent CSF-to-autopsy investigations have been conducted. A recently published study investigated the association between CSF biomarkers of AD and ADNC in 114 brain donors from 2 autopsy cohorts in Antwerp and Amsterdam. This study examined CSF biomarkers (measured using ELISAs) from participants in mid- to late-stage AD disease course and found no association between CSF Aβ42 and autopsy-confirmed ADNC; by contrast, CSF p-tau181 and t-tau were indeed associated with AD.¹⁷ Associations were moderated by APOE ε4 status and interval between LP and death. Another study¹⁸ investigated the association between antemortem CSF biomarkers (Aβ40, Aβ42, t-tau, p-tau181, p-tau/Aβ42, Aβ42/Aβ40) using Elecsys in vitro diagnostic immunoassays and ADNC in 101 patients (the mean interval between LP and death was 2.9 years). All CSF biomarkers were associated with the various neuropathologic measures of AD with p-tau/Aβ42 and Aβ42/Aβ40 having optimal diagnostic performance. FTLD subtypes had unique CSF biomarker profiles. Notably, the sample investigated by this group comprised participants with cognitive impairment (across the continuum) at the time of the LP.

There is one study that examined the predictive potential of core AD CSF biomarkers in participants with normal cognition at the time of LP.¹⁹ That study sampled 720 community-dwelling participants and quantified AD biomarkers by CSF analysis and/or PET imaging. Among 57 participants who died and donated their brains for neuropathologic examination, 20 were defined as biomarker positive or negative based on the CSF p-tau181/Aβ42 ratio as determined by the Elecsys immunoassay (positivity was defined by amyloid PET for the remaining participants). The CSF p-tau181/Aβ42 ratio had high sensitivity and specificity for predicting ADNC (reported as either “intermediate likelihood” or “high likelihood” by NIA-Reagan criteria or intermediate/high ADNC by National Institute on Aging–Alzheimer's Association [NIA-AA] criteria). Univariate models showed strong concordance between the individual biomarkers of Aβ42, p-tau181, and t-tau and ADNC.¹⁹

The goal of this study was to better understand the use of antemortem CSF biomarkers for the early detection of ADNC. We examined the association between antemortem CSF biomarkers of AD and postmortem ADNC using the National Alzheimer's Coordinating Center (NACC) data sets from participants who had normal cognition (NC) at the time of CSF collection. CSF AD biomarkers were quantified by the Lumipulse assay as opposed to Elecsys (as performed in the above-described study by Long et al.) to allow for a comparison across 2 of the major assay platforms used for the prediction of underlying AD. In doing so, this study is among the first to compare CSF AD biomarkers by Lumipulse against AD neuropathology. We tested the ability of antemortem Lumipulse measures of CSF Aβ42, p-tau181, and t-tau to accurately differentiate brain donors with and without intermediate/high ADNC. According to NIA-AA criteria, intermediate and high ADNCs (AD+) are considered sufficient to account for observed cognitive impairment.²⁰ We also tested the concordance of the CSF AD biomarkers with neuropathologic ratings of neuritic amyloid plaque densities (CERAD score) and Braak staging of neurofibrillary tangles. We hypothesized that elevated p-tau181, decreased Aβ42, and an increased p-tau181/Aβ42 ratio in the CSF would accurately discriminate AD+ from AD− brain donors.

Primary Research Question

To determine whether Lumipulse measures of p-tau181, Aβ42, and p-tau181/Aβ42 in the CSF can accurately discriminate cognitively normal participants with and without Alzheimer disease neuropathologic change.

Methods

Participants

The sample included 49 brain donors from the National Alzheimer's Coordinating Center Uniform Data Set (NACC-UDS) and Neuropathology Data Set (NACC-NDS) who had normal cognition at the time of lumbar puncture (LP). Baseline UDS visits were made between 2005 and 2015, and LPs were performed between 1999 and 2015. The most recent LP and corresponding UDS visit were used. The NACC was created by the National Institute on Aging (NIA) in 1999 to establish a publicly available database to facilitate the study of AD and AD-related dementias. There are currently approximately 37 Alzheimer's Disease Research Centers (ADRCs) across the United States that conduct standardized assessments approximately annually. Since 2005, each ADRC has been contributing standardized data collected to the central repository NACC to form the UDS. CSF biomarker data obtained from LP were added to the NACC database in 2015. Some of the NACC-UDS participants agreed to brain donation and postmortem neuropathologic evaluations. The data obtained from these postmortem evaluations make up the NACC-NDS. The participant data from the NACC-UDS and NACC-NDS queried are provided in Tables 1 and 2.

Table 1.

Demographics and Clinical Features of the Sample by Alzheimer Disease Status

	AD+ (n = 20)	AD− (n = 29)	p Value
Age at LP, mean (SD), range	79.1 (5.3) 67–90	79.4 (5.9) 69–91	0.83
Age at death, mean (SD), range	88.1 (5.4) 76–95	86.3 (7.1) 70–100	0.33
Interval between LP and death, mean (SD) years, range	9.0 (4.1) 1–16	6.9 (4.3) 1–17	0.09
Sex, n % male	10 (50.0%)	14 (48.3%)	0.91
Race, n % White	19 (95.0)	28 (96.6)	0.79
Years of education, mean (SD)	15.9 (4.1)	16.0 (2.5)	0.89
APOE ε4 carrier status, n (%)	11 (55.0)	3 (10.3)	<0.01
Global CDR at the last visit, n (%)			<0.01
0	5 (25)	20 (69.0)
0.5	2 (10.0)	5 (17.2)
1.0	5 (25)	2 (6.9)
2.0	8 (40.0)	2 (6.9)
Lumipulse CSF p-tau181/Aβ42 ratio, mean (SD), range	0.19 (0.09) 0.06–0.36	0.05 (0.03) 0.02–0.14	<0.001
Lumipulse CSF Aβ42, mean (SD), range	413.05 (97.63) 220.00–575.00	898.48 (423.11) 224.00–2100.00	<0.01
Lumipulse CSF p-tau181, mean (SD), range	77.03 (40.49) 20.20–169.60	37.63 (23.27) 16.10–135.30	0.0005
Lumipulse CSF t-tau, mean (SD), range	495.70 (250.67) 139.00–1,071.00	307.31 (201.05) 102.00–954.00	0.0083

Abbreviations: AD = Alzheimer disease; APOE = apolipoprotein E.

Independent-samples t tests and χ² analysis compared differences between the AD status groups. The APOE ε4 variable included carrier vs noncarrier. AD+ was defined as intermediate or high NIA-AA Alzheimer disease neuropathologic change (ADNC). Interval between CDR at the last visit and death was 1.76 (SD = 1.65) years, on average (range: 0–8 y).

Table 2.

Neuropathologic Features of AD+ and AD− Brain Donors

Pathology, n (%)	AD+ N = 20	AD– N = 29	p Value
Atherosclerosis	10 (50.0)	11 (37.9)	0.59
Arteriolosclerosis	12 (60.0)	10 (34.5)	0.14
Microinfarcts	6 (30.0)	5 (17.2)	0.48
Infarcts/lacunes	8 (40.0)	5 (17.2)	0.15
Cerebral amyloid angiopathy	9 (45.2)	2 (6.9)	<0.01
Lewy body disease	6 (30.0)	3 (10.3)	0.10
LATE-NC	7 (35.0)	6 (20.7)	0.30
Hippocampal sclerosis	1 (5.0)	3 (10.3)	0.60
Non-ADNC tauopathies	10 (50.0)	19 (65.5)	0.30
ARTAG	9 (45.0)	15 (51.7)
Argyrophilic grain disease (AGD)	2 (10.0)	9 (31.0)
PART, definite	0 (0)	6 (20.7)
Globular glial tauopathy (GGT)	0 (0)	1 (3.4)
Corticobasal degeneration	0 (0)	1 (3.4)
Thal phase for amyloid plaques
Phase 0	0 (0)	6 (20.7)	<0.001
Phase 1	0 (0)	10 (34.5)
Phase 2	0 (0)	3 (10.3)
Phase 3	2 (10.0)	5 (17.2)
Phase 4	4 (20.0)	3 (10.3)
Phase 5	14 (70.0)	2 (6.9)
Braak score
0	0 (0)	0 (0)	<0.001
I/II	0 (0)	24 (82.8)
III/IV	4 (25.0)	5 (17.2)
V/VI	16 (75.0)	0 (0)
CERAD score
0	1 (5.0)	27 (93.1)	<0.001
1	5 (25.0)	2 (6.9)
2	3 (15.0)	0 (0)
3	11 (55.0)	0 (0)

The chi-square analysis or Fisher exact test compared differences between the AD status groups. Brain donors were classified as having atherosclerosis, arteriolosclerosis, or cerebral amyloid angiopathy if they had moderate or severe ratings of pathology.

Sample Size Derivation

At the time of this study, there were 1,063 participants who had normal cognition at the time of LP and available CSF biomarker data. Of these 1,063, 85 had available neuropathologic data (based on NPFORMVER) and then 50 of the 85 had CSF biomarker data analyzed by Lumipulse. 1 additional participant was excluded because Braak staging was uncertain and not assigned because of the presence of other competing tauopathies. Taken together, the current sample included those who had available CSF Aβ42, t-tau, and p-tau181 biomarker data analyzed by Lumipulse, had normal cognition at the time of the LP, and had available postmortem neuropathology data. Because we restricted the sample to assays performed by Lumipulse, all cases included in the study were from the Washington University Knight ADRC. 34 of the 49 participants in this study were also included in the Washington University Knight ADRC study by Long et al. that investigated the Elecsys assay.¹⁹

CSF Biomarkers

This study examined AD CSF biomarkers Aβ42, p-tau181, and t-tau.²¹ Not all ADRCs conduct LP for CSF biomarker analyses, and only 6 of 30 sites were represented in this study. At this time, CSF biomarker assay methods are not standardized across ADRCs, and ADRCs voluntarily contribute CSF biomarker data to NACC. In this sample, we only included participants whose CSF biomarker data were analyzed using Lumipulse (n = 49). 30 were excluded who had CSF biomarker measurements performed using other assay methods (ELISA n = 9, Luminex n = 21) because of the known significant differences in absolute concentrations across methods.²² There was also insufficient sample size to evaluate the other assay types separately. All Lumipulse-based samples were collected at 1 site, the Knight ADRC. Samples were collected and processed as previously described.⁷ Participants underwent LP after overnight fasting, and samples were collected the following morning at approximately 8 am. Samples were collected in a 50-mL polypropylene tube using an atraumatic Sprotte 22-gauge spinal needle. CSF was immediately placed on ice and centrifuged at low speed within 2 hours of collection. CSF was transferred to another 50-mL tube and aliquoted at 500 uL into polypropylene tubes and stored at −80°C.²³ For biomarker analysis, samples were brought to room temperature, vortexed, and transferred to polystyrene cuvettes. Concentrations of each biomarker were measured by chemiluminescent enzyme immunoassay using a fully automated platform (LUMIPULSE G1200, Fujirebio, Malvern, PA). A single lot of each reagent was used for all samples. Quality control analyses included confirmation that concentrations fell within 20% of the manufacturer-reported mean value for each analyte and assessment of intraday and interday variability by running 3 pooled CSF controls in duplicate in each batch of samples. Average intraday and interday coefficients of variation were less than 5% across all analytes.

Neuropathology

The neuropathologic data included in this study were generated by the Knight ADRC Neuropathology Core, following an established protocol. For each case, the left hemibrain was sliced after formalin fixation: the supratentorial portion, in the coronal plane; the cerebellum, parasagittally; the brainstem, axially. 16 brain areas were sampled and processed into paraffin for histology: the middle frontal gyrus; the anterior cingulate gyrus (at the level of the genu of the corpus callosum); the precentral gyrus; the superior and middle temporal gyri; the inferior parietal cortex including the angular gyrus; the posterior cingulate gyrus and precuneus (at the level of the splenium); the calcarine sulcus and parastriate cortex; the amygdala with entorhinal cortex and fusiform gyrus; the hippocampus with parahippocampal and fusiform gyri (at the level of the lateral geniculate nucleus); the caudate, putamen, nucleus accumbens, and olfactory cortex; the anterior pallidum and basal forebrain with nucleus basalis (including the anterior commissure); the thalamus (including subthalamic nucleus); the superior cerebellar cortex with deep white matter and the dentate nucleus; the midbrain; the pons; and the medulla. When available, the rostral cervical spinal cord was also sampled and processed. Slide-mounted 6-micron-thick sections of the resulting formalin-fixed, paraffin-embedded tissue blocks were stained with hematoxylin and eosin (H&E) and by immunohistochemistry using antibodies for amyloid-β (10D5, Eli Lilly, Indianapolis, IN), phosphorylated tau (PHF-1, a gift from Dr. Peter Davies), phosphorylated alpha-synuclein (Cell Applications, San Diego, CA) or nonphosphorylated alpha-synuclein (LB509, MilliporeSigma Burlington, MA), and phosphorylated TAR DNA-binding protein of 43 kDa (pTDP-43, Cosmo Bio USA, Carlsbad, CA). Slides were evaluated systematically to generate semiquantitative scores for observed neurodegenerative features including those directly relevant to staging ADNC (Aβ plaques, neuritic plaques, neurofibrillary tangles), neuronal loss, gliosis, lesions that characterize other common and uncommon neurodegenerative proteinopathies (e.g., other tau-immunoreactive neurons, neuronal inclusions, glial inclusions, and cell processes; alpha-synuclein immunoreactive Lewy bodies and glial cytoplasmic inclusions; and pTDP-43-immunoreactive inclusions and cellular processes), cerebral vasculopathies (e.g., arteriolosclerosis and cerebral amyloid angiopathy), and other relevant findings (e.g., white matter pallor, microinfarctions, and microhemorrhages). These regional semiquantitative scores were then evaluated according to established historical and contemporary neuropathologic criteria to render all appropriate neuropathologic diagnoses and stages for each case.

Neuropathology data obtained from the NACC database for this study were derived from neuropathologic examinations of participants who died between 2006 and 2020; these data had originally been collected and provided to NACC using standardized NACC Neuropathology Forms and Coding Guidebooks²⁴ versions 9 (n = 15), 10 (n = 29), and 11 (n = 5). However, the 2012 NIA-AA criteria for ADNC were not included until version 10. Therefore, to harmonize ADNC scoring across the cohort, older cases reported using version 9 were re-evaluated for this study by the Knight ADRC NPC according to NIA-AA criteria. Furthermore, RJP re-reviewed the data for non-ADNC tauopathies from all 49 cases to facilitate accuracy and harmonization, given the limitations of the NACC neuropathology data collection instrument.

For the analyses in this study, we computed a binary ADNC variable to indicate positive (AD+) or negative (AD−) AD status. AD+ was defined by ADNC ≥2 (i.e., intermediate or high ADNC) while AD− included those with no or low ADNC. All other neuropathologic diagnoses of neurodegenerative diseases and cerebrovascular disease reported in this study (Table 2) were made following NACC guidelines and established criteria.

Statistical Analyses

Demographic, clinical, and neuropathologic characteristics were compared between the AD+ and AD− groups using the independent-samples t test or χ². There were 2 primary binary outcome variables: AD+ vs AD−. The primary independent variables included were CSF Aβ42, p-tau181, t-tau, and p-tau181/Aβ42 ratio. For each CSF analyte, a binary logistic regression was performed to test their association with AD+/AD− status, controlling for age at the time of LP (age at death–interval from LP until death), sex, interval between LP and death (NACCYOD [year of death]-CSFLPYR [year of LP]), and APOE ε4 carrier status (absent/present). Using the predicted probabilities from the binary logistic regression, the area under the receiver operating curve (AUC) was calculated to determine the accuracy of each analyte to discriminate brain donors considered AD+ from those considered AD−. The AUC, 95% CI, and associated p values are the primary statistics reported. Guidelines from Hosmer and Lemeshow were used for interpretation of AUC discrimination classification accuracy.²⁵ Models were repeated with prevalent vascular neuropathologic comorbidities included in the model (i.e., arteriolosclerosis and atherosclerosis), as well as after excluding participants with other tauopathies. For each CSF analyte, the binary logistic regressions were also repeated excluding covariates from the model. Using the unadjusted predicted probabilities from the binary logistic regression, the unadjusted area under the receiver operating curve (AUC) was calculated.

Ordinal logistic regression tested the association between p-tau181/Aβ42 and rating scales of AD neuropathology, including the Braak stage for neurofibrillary tangle pathology (0 = none, I-II = mild, III-IV = moderate, V-VI = severe) and CERAD neuritic amyloid plaque score (0 = no neuritic plaques, 1 = sparse neuritic plaques, 2 = moderate neuritic plaques, 3 = frequent neuritic plaques). We focused on the p-tau181/Aβ42 ratio to reduce the number of analyses and because it has been shown to be the best biomarker of AD status. For the Braak stage and CERAD score, analysis of covariance (ANCOVA) with Tukey post hoc test was also performed to compare p-tau181/Aβ42 between each stage. The aforementioned covariates were included in these ordinal logistic regressions and ANCOVA models. For the ANCOVA models, estimated marginal mean differences were reported. A p value less than or equal to 0.05 was considered statistically significant for all analyses.

Standard Protocol Approvals, Registrations, and Patient Consents

The NACC database was approved by the University of Washington Institutional Review Board. Written informed consent of each participant was obtained at the individual ADRC at which the data were collected. A formal data request (request #10014) was submitted to NACC for this study and was approved on June 7, 2022.

Data Availability

All clinical and neuropathologic deidentified data accessed for this study are publicly available through the NACC data set. Requests by qualified investigators can be made at naccdata.org. Deidentified data can also be made available to qualified investigators on request to the Washington University Knight ADRC.

Results

Sample Characteristics

Table 1 summarizes participant and sample characteristics. 47 participants identified as White (95.9%), and 24 identified as male (49%). The mean years of education was 15.9 (SD = 3.2). The average age at LP was 79.3 (SD = 5.6) years, and the average age at death was 87.1 (SD = 6.5) years. The mean interval between LP and death was 7.76 years (SD = 4.31, 1.00–17.00 years). Of the 49 brain donors who had normal cognition (at the time of LP), 20 were AD+. Compared with the AD− group, the AD+ group was approximately 2 years older on average at death and had a longer interval between LP and death (9.0 vs 6.9 years). In addition, the AD+ group was more likely to be a carrier of the APOE ε4 allele and have a higher CDR score at the visit closest to death (mean, SD years between the last visit and death = 1.76, 1.65) (p < 0.01). There were no statistically significant differences between the AD+ and AD− groups regarding age, sex, race, or years of education.

Independent-samples t tests or bivariate correlations showed no statistically significant associations between the CSF biomarkers and demographic variables (i.e., age at LP visit, age at death, sex, race, and years of education) or with the interval between LP and death. CSF Aβ42 levels were associated with APOE ε4, such that ε4 carriers had lower concentrations compared with noncarriers (mean difference = 384.49, p = 0.002). There were no ε4 effects for t-tau or p-tau181.

Neuropathology Characteristics

Table 2 lists the neuropathology characteristics of the cohort. Rates of cerebral amyloid angiopathy were higher among the AD+ group compared with the AD− group (p < 0.01).

CSF Biomarkers: Association With ADNC Status

A baseline model that consisted of demographics (sex and age at LP), years between LP and death, and APOE ε4 status had an AUC of 0.72 (95% CI 0.57–0.87, p = 0.02) for discriminating AD+ and AD− brain donors. Based on binary logistic regressions controlling for sex, age at LP, years between LP and death, and APOE ε4 status, a higher CSF p-tau181/Aβ42 ratio strongly and accurately differentiated AD+ vs AD− status at autopsy (AUC = 0.97). Antemortem CSF p-tau181 levels performed second-best (AUC = 0.92), followed by CSF Aβ42 (AUC = 0.92). CSF t-tau (AUC = 0.87) had the weakest but still excellent discrimination accuracy for AD status. Figure 1 shows ROC curves, and Table 3 provides a summary of the logistic regression models. Sensitivity models showed that the AUC statistics remained similar when atherosclerosis and arteriolosclerosis were added to the models (CSF p-tau181/Aβ42 AUC = 0.97, CSF p-tau181 = 0.92, CSF Aβ42 AUC = 0.96, and CSF t-tau AUC = 0.89) and when participants with non-ADNC tauopathies were excluded (n = 29 excluded) (CSF p-tau181/Aβ42 AUC = 0.95, CSF p-tau181 = 0.93, CSF Aβ42 AUC = 0.94, and CSF t-tau AUC = 0.90).

Figure 1. ROC Curve of Lumipulse-Measured CSF Biomarkers Discriminating AD+ and AD− Brain Donors.

ROC curve analysis of AD CSF biomarkers by AD status. ROC curves and AUC statistics generated using predicted probabilities from binary logistic regressions that controlled for age at LP, sex, years between LP and death, and APOE ε4 carrier status. CSF p-tau181/Aβ42 had the strongest discrimination accuracy of AD status, followed by CSF p-tau181, then CSF Aβ42, and CSF t-tau. AUC = area under the curve.

Table 3.

ROC Curve Analysis of CSF Biomarkers and Alzheimer Disease

	AD+ vs AD− Adjusted Models: Age at Visit, Sex, Years Between LP and Death, and APOE ε4 Carrier Status
Biomarker	AUC	95% CI	p Value
Lumipulse CSF p-tau181/Aβ42	0.97	0.94–1.00	0.003
Lumipulse CSF p-tau181	0.92	0.84–1.00	0.001
Lumipulse CSF Aβ42	0.92	0.85–1.00	0.007
Lumipulse CSF t-tau	0.87	0.76–0.97	0.005

	AD+ vs AD− Unadjusted Models
Biomarker	AUC	95% CI	p Value
Lumipulse CSF p-tau181/Aβ42	0.97	0.92–1.00	<0.001
Lumipulse CSF p-tau181	0.82	0.69–0.94	0.002
Lumipulse CSF Aβ42	0.91	0.83–0.99	0.002
Lumipulse CSF t-tau	0.73	0.58–0.88	0.01

Abbreviation: AUC = area under the curve.

Binary logistic regression was used to investigate the association between the core AD CSF biomarkers and AD status, as classified by the binary AD status variable used throughout the study. As shown on top, logistic regression and ROC curve analysis were controlled for age at visit, sex, years between LP and death, and APOE ε4 carrier status. Unadjusted models are shown on bottom.

To facilitate clinical interpretation, binary logistic regressions were repeated removing all covariates (sex, age at LP, years between LP and death, and APOE ε4 status). Again, a higher CSF p-tau181/Aβ42 ratio strongly and accurately differentiated AD+ vs AD− status at autopsy (AUC = 0.97). CSF Aβ42 levels performed second-best (AUC = 0.91), followed by CSF p-tau181 (AUC = 0.82). CSF t-tau (AUC = 0.73) had the weakest discrimination accuracy for AD status. Figure 2 shows the unadjusted ROC curves, and Table 3 provides a summary of the unadjusted logistic regression models.

Figure 2. Unadjusted ROC Curve of Lumipulse-Measured CSF Biomarkers Discriminating AD+ and AD− Brain Donors.

ROC curve analysis of AD CSF biomarkers by AD status. ROC curves and AUC statistics generated using predicted probabilities from binary logistic regressions removing all covariates. CSF p-tau181/Aβ42 had the strongest discrimination accuracy of AD status, followed by CSF p-tau181, then CSF Aβ42, and CSF t-tau. AUC = area under the curve.

CSF Biomarkers: Braak and CERAD Associations

Figures 3 and 4 present violin plots showing the distribution of p-tau181/Aβ42 values by CERAD scores and Braak NFT staging. Using multivariable ordinal logistic regression, greater p-tau181/Aβ42 values were associated with increased odds of having a higher Braak NFT stage (OR = 2.32, 95% CI 1.36, 3.29, p < 0.01). Additional ANCOVAs compared CSF p-tau181/Aβ42 across each Braak stage. CSF p-tau181/Aβ42 was higher in Braak V-VI vs Braak I-II (mean difference = 0.16, p < 0.01) and vs Braak III-IV (mean difference = 0.10, p < 0.01).

Figure 3. Distribution of Lumipulse-Measured CSF p-tau181/Aβ42 by CERAD Ratings.

Violin plots showing the distribution of the CSF p-tau181/Aβ42 ratio by CERAD scores for neuritic amyloid plaques.

Figure 4. Distribution of Lumipulse-Measured CSF p-tau181/Aβ42 by Braak Ratings.

Violin plots showing the distribution of the CSF p-tau181/Aβ42 ratio by Braak staging for neurofibrillary degeneration.

Using multivariable ordinal logistic regression, greater p-tau181/Aβ42 values were associated with increased odds of having a higher CERAD staging score (OR = 2.94, 95% CI 1.74, 4.15, p < 0.01). ANCOVAs showed significant differences in CSF p-tau181/Aβ42 between those with no neuritic plaques and those with moderate neuritic plaques (mean difference = −0.14, p < 0.01) and frequent neuritic plaques (mean difference = −0.18, p < 0.01). There was also a significant difference in CSF p-tau181/Aβ42 between participants with sparse neuritic plaques and those with moderate neuritic plaques (mean difference = −0.12, p = 0.012) and also between those with sparse and those with frequent neuritic plaques (mean difference = −0.14, p < 0.01).

Discussion

This study examined the association between antemortem CSF biomarkers of AD (using Lumipulse) and postmortem AD neuropathology among 49 brain donors from the NACC data set who were cognitively normal at the time of LP. The interval between LP until death was approximately 8 years, on average. Our analyses showed the strongest associations between CSF p-tau181/Aβ42 and AD+ (defined as intermediate or high ADNC, according to NIA-AA criteria) (AUC = 0.97, p = 0.003). High p-tau181, low Aβ42, and high t-tau also accurately discriminated AD+ brain donors from AD− donors, but with slightly lower AUCs. Finally, greater values of CSF p-tau181/Aβ42 were associated with increased odds of having both higher Braak and CERAD staging scores in participants.

The goal of this study was to validate antemortem Lumipulse-measured CSF biomarkers for the early detection of AD. Previous studies have primarily investigated AD CSF biomarker validity in patients with mid-stage to late-stage disease and have relied on clinical end points.²⁶ However, the validation of biomarkers for AD neuropathology relies on comparison with findings at autopsy. We are aware of only one other study that validated the leading diagnostic CSF AD biomarkers against neuropathology in participants with normal cognition.¹⁹ That study used the Elecsys assay, whereas our study used Lumipulse. In this sample of participants with normal cognition, Lumipulse CSF p-tau181/Aβ42 emerged as the most sensitive predictor of AD status including early AD, followed by CSF p-tau181, then Aβ42, and t-tau. Each of these CSF biomarkers had significant and strong discrimination accuracy independent of demographics alone. In comparison with the study by Long et al. (which used Elecsys), both studies show high sensitivity and specificity for CSF p-tau181/Aβ42 and support the comparability of Lumipulse and Elecsys for the prediction of AD. This aligns with the research that has conducted head-to-head comparison of these assays using amyloid PET as the reference and found them to be strongly comparable.²⁷ Overall, our results support the utility of core AD CSF biomarkers measured by Lumipulse, particularly the CSF p-tau181/Aβ42 ratio, to predict early AD neuropathologic changes.

The current state of AD research is focused on the early detection of AD because this is an opportune time for intervention and prevention or slowing of cognitive decline. Recent FDA approvals of disease-modifying drugs for AD, that is, lecanemab (Leqembi, Eisai. Nutley, NJ) and aducanumab (Aduhelm, Biogen. Cambridge, MA), are most appropriate for people who are in the early disease stages.^28-30 Biomarkers that can accurately detect AD pathology as early as possible are thus needed to ensure that appropriate use recommendations of these drugs are being followed. Our results support the usefulness of AD CSF biomarkers to serve this role. We acknowledge that LP is often viewed as invasive and, therefore, blood-based biomarkers are hopefully a promising alternative in the future.

Our study has limitations. First, this study has a small sample size, mostly because the data set was restricted to participants who had CSF analyzed by Lumipulse. The sample size is in the context of requiring both CSF and neuropathology data, and only a few studies have reported on these jointly because of the challenges associated with LP and brain donation. Even in this small sample, significant effects were detected and concern for lack of statistical power is attenuated, given the absence of null findings. However, we acknowledge that the small sample size can also lead to unstable estimates and larger CSF-to-autopsy studies are needed for validation and cutoff derivation and to improve understanding on the specificity of the CSF biomarkers to AD. Second, the sample is demographically homogeneous and predominantly White, limiting the internal and external validity of the findings. The requirements of an LP procedure and consent for and successful coordination of brain donation for inclusion in this study likely contributed to the lack of racial diversity because previous research has shown these procedures to be barriers to research participation and/or to have racial disparities.^31,32 Blood draw for plasma-based biomarkers is an appealing alternative to LP. Plasma biomarkers are unlikely to replace gold-standard biomarkers that provide a direct window into the CNS and are needed for making highly accurate and confident diagnoses. Third, participants in ADRCs may preferentially include those with concerns for their cognitive function and/or their risk of AD/ADRD. Therefore, these findings may best generalize more closely to a clinic-based setting, than to randomly selected members of the community. In addition, our study was limited to the use of CSF Aβ42. Future studies should include the Aβ42/40 ratio and the ratio approved by the FDA for detection of amyloid. Finally, CSF collection and curation, CSF biomarker measurements, and neuropathologic evaluations for this study were all performed at one center; although all these procedures and techniques involve robust protocols, it remains to be demonstrated whether identical results might be obtained with similar cohorts in validation studies across centers.

This study validates Lumipulse-measured CSF biomarkers of p-tau181/Aβ42 for the detection of preclinical ADNC, with subsequent neuropathologic confirmation. While CSF biomarker analysis has long been considered the gold standard for the antemortem detection of ADNC, comparison against neuropathology is necessary for biomarker validation. Our results encourage the use of LP and CSF biomarker analysis to assist with diagnosis and the identification of optimal candidates for AD-related therapeutics and drug trials.

Glossary

AD

Alzheimer disease

ADNC

AD neuropathologic change

ADRC

Alzheimer's Disease Research Center

ANCOVA

analysis of covariance

AUC

area under the curve

LP

lumbar puncture

NACC

National Alzheimer's Coordinating Center

NACC-NDS

National Alzheimer's Coordinating Center Neuropathology Data Set

NACC-UDS

National Alzheimer's Coordinating Center Uniform Data Set

NIA

National Institute on Aging

NIA-AA

National Institute on Aging–Alzheimer's Association

Appendix. Authors

Name	Location	Contribution
Michelle Safransky, MS	Boston University Alzheimer’s Disease Research Center, Boston University CTE Center, Department of Neurology, Boston University Chobanian & Avedisian School of Medicine, MA	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data
Jenna R. Groh, BA	Boston University Alzheimer’s Disease Research Center, Boston University CTE Center, Department of Neurology, Boston University Chobanian & Avedisian School of Medicine, MA	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data
Kaj Blennow, MD, PhD	Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital; Department of Psychiatry and Neurochemistry, Institute of Neuroscience and Physiology, The Sahlgrenska Academy at the University of Gothenburg, Mölndal, Sweden; Institut du Cerveau et de la Moelle épinière (ICM), Pitié-Salpêtrière Hospital, Sorbonne Université, Paris, France; University of Science and Technology of China and First Affiliated Hospital of USTC, Hefei, Anhui, P.R. China	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Henrik Zetterberg, MD, PhD	Clinical Neurochemistry Laboratory, Sahlgrenska University Hospital; Department of Psychiatry and Neurochemistry, Institute of Neuroscience and Physiology, The Sahlgrenska Academy at the University of Gothenburg, Mölndal, Sweden; Department of Neurodegenerative Disease, UCL Institute of Neurology; UK Dementia Research Institute at UCL, UCL Institute of Neurology, University College London, United Kingdom	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Yorghos Tripodis, PhD	Department of Biostatistics, Boston University School of Public Health, MA	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Brett Martin, MS	Biostatistics and Epidemiology Data Analytics Center (BEDAC), Boston University School of Public Health, MA	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data
Jason Weller, MD	Boston University Alzheimer’s Disease Research Center, Boston University CTE Center, Department of Neurology, Boston University Chobanian & Avedisian School of Medicine, MA	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Breton M. Asken, PhD, ATC	University of Florida, Gainesville, FL	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Gil Dan Rabinovici, MD	Memory & Aging Center, Department of Neurology, Weill Institute for Neurosciences; Department of Radiology and Biomedical Imaging, University of California, San Francisco	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Wendy Wei Qiao Qiu, MD, PhD	Boston University Alzheimer’s Disease Research Center, Boston University CTE Center, Department of Neurology, Boston University Chobanian & Avedisian School of Medicine; Department of Psychiatry; Department of Pharmacology and Experimental Therapeutics, Boston University Chobanian & Avedisian School of Medicine, MA	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Ann C. McKee, MD	Boston University Alzheimer’s Disease Research Center, Boston University CTE Center, Department of Neurology, Boston University Chobanian & Avedisian School of Medicine, Boston; VA Boston Healthcare System, US Department of Veteran Affairs, Jamaica Plain; Department of Pathology and Laboratory Medicine, Boston University Chobanian & Avedisian School of Medicine; VA Bedford Healthcare System, US Department of Veteran Affairs, MA	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data
Thor D. Stein, MD, PhD	Boston University Alzheimer’s Disease Research Center, Boston University CTE Center, Department of Neurology, Boston University Chobanian & Avedisian School of Medicine, Boston; VA Boston Healthcare System, US Department of Veteran Affairs, Jamaica Plain; Department of Pathology and Laboratory Medicine, Boston University Chobanian & Avedisian School of Medicine; VA Bedford Healthcare System, US Department of Veteran Affairs, MA	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data
Jesse Mez, MD, MS	Boston University Alzheimer’s Disease Research Center, Boston University CTE Center, Department of Neurology, Boston University Chobanian & Avedisian School of Medicine; Framingham Heart Study, MA	Drafting/revision of the manuscript for content, including medical writing for content; analysis or interpretation of data
Rachel L. Henson, MS	Department of Neurology, Knight Alzheimer’s Disease Research Center, Washington University School of Medicine	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data
Justin Long, MD, PhD	Department of Neurology, Knight Alzheimer’s Disease Research Center, Washington University School of Medicine	Drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data
John C. Morris, MD	Department of Neurology, Knight Alzheimer’s Disease Research Center, Washington University School of Medicine	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; analysis or interpretation of data
Richard J. Perrin, MD, PhD	Department of Neurology, Knight Alzheimer’s Disease Research Center, Washington University School of Medicine	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data
Suzanne E. Schindler	Department of Neurology, Knight Alzheimer’s Disease Research Center, Washington University School of Medicine	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data
Michael L. Alosco, PhD	Boston University Alzheimer’s Disease Research Center, Boston University CTE Center, Department of Neurology, Boston University Chobanian & Avedisian School of Medicine; Department of Neurology, Boston Medical Center; Department of Anatomy & Neurobiology, Boston University Chobanian & Avedisian School of Medicine, MA	Drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data

Study Funding

This research was supported by a Medical Student Summer Research Program at Boston University. This work was supported by grants from the National Institutes of Health (P30AG072978, P30AG066444, P01AG003991, P01AG026276, U19AG032438, U19AG032438-09S1, R01AG068319, P01AG003991, P30AG066444, R01AG070883, R01NS075321, R01NS097799, U19AG069701, U19NS110456, R01AG058676, R01AG074909, U19AG024904, U19AG07879). HZ is a Wallenberg Scholar and a Distinguished Professor at the Swedish Research Council supported by grants from the Swedish Research Council (#2023-00356; #2022-01018 and #2019–02397), the European Union's Horizon Europe research and innovation programme under grant agreement No. 101053962, Swedish State Support for Clinical Research (#ALFGBG-71320), the Alzheimer Drug Discovery Foundation (ADDF), USA (#201809–2016862), the AD Strategic Fund and the Alzheimer's Association (#ADSF-21-831376-C, #ADSF-21-831381-C, #ADSF-21-831377-C, and #ADSF-24-1284328-C), the Bluefield Project, Cure Alzheimer's Fund, the Olav Thon Foundation, the Erling-Persson Family Foundation, Stiftelsen för Gamla Tjänarinnor, Hjärnfonden, Sweden (#FO2022-0270), the European Union's Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No. 860197 (MIRIADE), the European Union Joint Programme - Neurodegenerative Disease Research (JPND2021-00694), the National Institute for Health and Care Research University College London Hospitals Biomedical Research Centre, and the UK Dementia Research Institute at UCL (UKDRI-1003). KB is supported by the Swedish Research Council (#2017-00915 and #2022-00732), the Swedish Alzheimer Foundation (#AF-930351, #AF-939721 and #AF-968270), Hjärnfonden, Sweden (#FO2017-0243 and #ALZ2022-0006), the Swedish state under the agreement between the Swedish government and the County Councils, the ALF agreement (#ALFGBG-715986 and #ALFGBG-965240), the European Union Joint Program for Neurodegenerative Disorders (JPND2019-466-236), the Alzheimer's Association 2021 Zenith Award (ZEN-21-848495), and the Alzheimer's Association 2022–2025 Grant (SG-23-1038904 QC).

Disclosure

M.J. Safransky and J.R. Groh report no disclosures relevant to this manuscript. K. Blennow reports serving as a consultant and at advisory boards for Acumen, ALZPath, BioArctic, Biogen, Eisai, Lilly, Moleac Pte. Ltd, Novartis, Ono Pharma, Prothena, Roche Diagnostics, and Siemens Healthineers; he has served at data monitoring committees for Julius Clinical and Novartis; he has given lectures, produced educational materials, and participated in educational programs for AC Immune, Biogen, Celdara Medical, Eisai and Roche Diagnostics; and is a cofounder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is a part of the GU Ventures Incubator Program, outside the work presented in this paper. H. Zetterberg reports serving on scientific advisory boards and/or as a consultant for Abbvie, Acumen, Alector, Alzinova, ALZPath, Amylyx, Annexon, Apellis, Artery Therapeutics, AZTherapies, Cognito Therapeutics, CogRx, Denali, Eisai, Merry Life, Nervgen, Novo Nordisk, Optoceutics, Passage Bio, Pinteon Therapeutics, Prothena, Red Abbey Labs, reMYND, Roche, Samumed, Siemens Healthineers, Triplet Therapeutics, and Wave; has given lectures in symposia sponsored by Alzecure, Biogen, Cellectricon, Fujirebio, Lilly, Novo Nordisk, and Roche; and is a cofounder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is a part of the GU Ventures Incubator Program (outside submitted work). Y. Tripodis, B. Martin, J. Weller, and B. Asked report no disclosures relevant to this manuscript. G.D. Rabinovici reports that he receives research support from NIA, NINDS, Alzheimer's Association, American College of Radiology, Rainwater Charitable Foundation, Alliance for Therapeutics in Neurodegeneration, Avid, GE Healthcare, LMI (all for New IDEAS); Genentech; is a paid consultant (SAB, last 36 months) for Alector, Eli Lilly, Merck; is a paid DSMB member for Johnson & Johnson; and he is an associate editor of JAMA Neurology. W.Q. Qiu, A.C. McKee, T.D. Stein, J. Mez, R.L. Henson, and J. Long report no disclosures relevant to this manuscript. J.C. Morris reports that he consults for Barcelonaβeta Brain Research Foundation Scientific Advisory Board and Diverse VCID Observational Study Monitoring Board; is on the advisory board for Cure Alzheimer's Fund Research Strategy Council and LEADS Advisory Board, University of Indiana. Neither J.C. Morris nor his family owns stock or has equity interest (outside of mutual funds or other externally directed accounts) in any pharmaceutical or biotechnology company. R.J. Perrin reports that his laboratory receives cost recovery funding from Biogen for tissue procurement and processing services related to ALS clinical trials. Neither R.J. Perrin nor his family owns stock or has equity interest (outside of mutual funds or other externally directed accounts) in any pharmaceutical or biotechnology company. S.E. Schindler reports that she has served on scientific advisory boards and consulted for Eisai. M.L. Alosco reports that he receives royalties from Oxford University Press Inc., and research support from Life Molecular Imaging Inc and Rainwater Charitable Foundation Inc. Go to Neurology.org/N for full disclosures.