Postmortem associations between Alzheimer’s disease (AD) pathology and plasma pTau217, GFAP, and NfL in AD and AD-related dementias

Abstract

Background and Objectives:

Alzheimer’s disease (AD) and its related disorders (ADRDs) are characterized by a high frequency of co-pathologies. We aimed to determine the specificity of plasma pTau217, GFAP, and NfL for AD neuropathological change (ADNC) in the presence of common ADRD co-pathologies.

Methods:

pTau217, GFAP, and NfL were measured using S-PLEX immunoassays from Meso Scale Discovery in banked plasma samples from two groups of participants in the Massachusetts ADRC Longitudinal Cohort study: 1) participants spanning the cognitive spectrum, who underwent brain autopsy, and blood collection within 6 years prior to death; and 2) participants with normal cognition and no neurological diagnosis during 5 years of follow-up, but no autopsy data [normal controls (NC)]. Cross-sectional associations between biomarker levels and ADNC, primary neuropathological diagnosis (NPDx1), and presence of non-AD co-pathologies were evaluated using linear regression models controlling for age, sex, and time to death.

Results:

187 participants with brain autopsy [NPDx1: AD n=85; other n=102; mean age: 74.3 years, 38.5% female; interval blood collection-death (mean±SD): 2.8±1.6 years] and 67 NC without brain autopsy (mean age: 66.5 years, 71.6% female) were included. pTau217, but not GFAP, levels increased stepwise with increasing Thal phases [β=0.61; 95% CI (0.24–0.97) to β=0.91 (0.55–1.27)] and Braak stages [β=0.59; (0.16–1.01) to β=0.74 (0.33–1.15)]. While 23% of individuals with a non-AD NPDx1 had increased pTau217 levels using a cutoff defined by the contrast between ADNC and NC, the majority (62%) had intermediate/high ADNC co-pathology and the remaining pTau217+ individuals had borderline increased levels. In contrast, 48% of individuals without ADNC had increased GFAP levels. pTau217 and GFAP were not different in the presence or absence of CAA, a-synuclein or TDP-43 proteinopathies, or primary tauopathies. NfL was not specifically associated with ADNC.

Discussion:

Plasma pTau217, but not GFAP or NfL, levels accurately reflect the presence of ADNC in the brain even in individuals with an NPDx1 of a non-AD dementia. Thus, a positive plasma pTau217 test in an individual with a suspected non-AD dementia should not necessarily be considered a misdiagnosis of the presumed non-AD dementia or as a false positive, but rather as evidence of ADNC co-pathology.

Introduction

Central to the current framework for the diagnosis and staging of Alzheimer’s disease (AD) is the definition of AD as a biological process which can be accurately identified with biomarkers of AD’s essential amyloid-β (Aβ) plaques and tau neurofibrillary tangles (NFT), constituting the AD neuropathological changes (ADNC).¹ The development of ultrasensitive assays for blood-based biomarkers (BBM) used for AD, especially phosphorylated tau (pTau), is conceptually changing how AD is diagnosed and monitored not only in research settings, but also in primary and secondary care.^{2, 3} Much of the knowledge around BBMs is, however, based on validation against other surrogate biomarkers (e.g. PET and CSF) for Aβ and tau pathology. While a few studies have investigated plasma AD biomarker levels in relation to autopsy evidence and staging of ADNC in the brain,^4–9 less is known about contributions of various other neuropathologies common in patients with AD such as neuronal α-synuclein and TDP-43 proteinopathies, primary tauopathies, and neurodegeneration associated with cerebrovascular disease. The high prevalence of concurrent neurodegenerative and cerebrovascular diseases in older adults¹⁰ and the difficulty differentiating them using clinical criteria complicate the analysis. Indeed, in the absence of valid biomarkers for these other pathologies, the effects of co-pathologies on AD biomarkers and whether AD biomarkers accurately reflect the presence of ADNC co-pathology in other neurodegenerative and cerebrovascular diseases may only be discerned with brain autopsy.

We used antemortem plasma biomarker measurements from participants with postmortem examination in the Massachusetts Alzheimer’s Disease Research Center Longitudinal Cohort study (MADRC-LC) to further define associations between three highly validated BBMs of neurodegenerative diseases [pTau217, glial fibrillary acidic protein (GFAP), and neurofilament light chain (NfL)] and histopathological findings of Aβ, pTau, TDP-43, α-synuclein, and vascular pathology. pTau217 is a leading candidate among AD BBM with high sensitivity and accuracy for ADNC.^{3, 11} GFAP is an intermediate filament predominantly expressed by CNS astrocytes which is upregulated during reactive astrogliosis.¹² NfL has emerged as an important, albeit non-specific, biomarker for neuro-axonal injury in a range of neurological diseases.¹³ Our aim was to establish how well pTau217, GFAP, and NfL detect ADNC¹⁴ in individuals with other neurodegenerative diseases and the degree to which concurrent neurodegenerative and cerebrovascular diseases affect the levels of these BBMs in the context of ADNC.

Materials and Methods

Study population

We measured plasma pTau217, GFAP and NfL using MESO SCALE DISCOVERY^® (MSD, Rockville, MD) S-PLEX immunoassays in a cohort of 1,100 participants in MADRC-LC, a longitudinal observational study of cognitive aging focused on AD and AD-related dementias (ADRDs; eMethods). Annual assessments in the MADRC-LC include a general and neurological exam, a set of forms to document past family and medical history and medications, a semi-structured interview to document cognitive symptoms and score the Clinical Dementia Rating scale (CDR© Dementia Staging Instrument),¹⁵ a battery of neuropsychological tests consisting of the National Alzheimer’s Coordinating Center (NACC) Uniform Data Set (UDS),¹⁶ and blood collection for all consenting participants. Cognitive status and clinical diagnosis are determined at each visit by a consensus team after a detailed examination and review of all available information according to 2011 NIA-AA diagnostic criteria for mild cognitive impairment (MCI)¹⁷ and AD.¹⁸ A subset of participants undergoes imaging and/or CSF biomarker sub-studies in affiliated protocols and all participants are invited to join a brain donation program. A standardized neuropathological assessment is performed on each donated brain by a team of board-certified neuropathologists following current neuropathological criteria for evaluation of neurodegenerative disease including immunohistochemistry for Aβ, tau, α-synuclein, and phosphorylated TDP-43, as well as assessment for vascular disease (eMethods).^{14, 19–21} Whenever two or more comorbid neuropathologies are found, MADRC neuropathologists rank them in order from higher to lower likelihood of contribution to the donor’s cognitive impairment from NPDx1 to NPDx5 based on pathologic severity.

Standard Protocol Approvals, Registrations, and Patient Consents

The study was approved by the Mass General Brigham Institutional Review Board (2006P002104) and all activities were performed according to the ethical standards of the Declaration of Helsinki. All participants or their assigned surrogate decision makers provided written informed consent for MADRC-LC participation and consent for autopsy was confirmed at death.

Sample collection and biomarker analysis

Banked plasma samples from MADRC-LC participants were retrieved from the Harvard Biomarkers Study biobank (samples collected prior to 2019) or the MassGeneral Institute for Neurodegenerative Disease (MIND) biorepository (samples collected in 2019 or later). Blood collection was performed without prior fasting mostly between 10:00 AM and 01:00 PM. Samples were collected in K₂EDTA tubes, centrifuged and frozen within 4 hours of collection, and stored at −80°C until use. pTau217, GFAP, and NfL were measured using the MSD S-PLEX^® pTau217 and S-PLEX Neurology Panel 1 assay kits at an MSD laboratory (eMethods). Three samples were <LLOQ for p Tau217 and one sample was >ULOQ for GFAP, and were assigned the value of LLOQ or ULOQ, respectively. Duplicate CVs were 8.0±11.0% (mean±SD) for pTau217, 3.7±3.9% for GFAP, and 5.3±5.1% for NfL.

Statistical analyses

Statistical analyses were performed using STATA/SE version 18.0 for Mac (StataCorp, College Station, TX). Biomarker concentrations were scaled and log10 transformed to satisfy assumptions of normal distribution for statistical analysis while values in text and graphs are presented without log transformation unless otherwise specified. Outliers exceeding +3SD over the mean were assigned the highest value of the remaining samples (GFAP: n=5; NfL: n=2). All reported p-values were adjusted for multiple hypothesis testing using the Bonferroni or the Benjamini-Hochberg’s method depending on number of comparisons with α<0.05 as significant threshold. Differences between neuropathological diagnostic groups and associations with neuropathological findings were evaluated using linear regression with the biomarker as outcome measure and adjusting for age at blood collection and sex. Time between blood draw and death was included for all analyses including postmortem neuropathological variables. To determine classification utility of the biomarkers, area under the curve (AUC) values were computed using logistic regression models and optimal cutpoints were estimated using the Youden index.

Data availability statement

Anonymized data not published within this article will be made available upon request from any qualified investigator.

Results

Sample characteristics

1,100 MADRC-LC participants had blood banked between 2008–2023 and at least one year of clinical follow-up. Of these, 206 had an autopsy performed between 2009–2023 and a blood sample collected within 6 years of death (mean 2.8±1.6 years). Nineteen participants were excluded due to having a primary neuropathological diagnosis (NPDx1) not consistent with a neurodegenerative or cerebrovascular disorder (n=9), a lack of defined underlying proteinopathy (n=5), or having an NPDx1 not fitting into any of the other diagnostic categories [adult onset leukodystrophy with neuroaxonal spheroids (AOLNAS), multiple sclerosis, normal pressure hydrocephalus, prion disease, thalamic degeneration; n=1 for each condition] resulting in a final cohort of 187 participants (eTable 1). The participants were divided into six main diagnostic categories based on their NPDx1:

1. ADNC (n=85), requiring intermediate or high ADNC burden^{14, 19} or a combination of Braak NFT stage ≥III and a Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) neuritic plaque score ≥2 (denoting moderate or frequent neuritic plaques) in older cases collected prior to the modern classification system where the Thal Aβphase was not available (n=3).

2. TDP-43 proteinopathy (TDP; n=20) including frontotemporal lobar degeneration (FTLD-TDP, n=15) and amyotrophic lateral sclerosis (ALS, n=5) defined by presence of neuronal cytoplasmic TDP-43 inclusions in relevant brain and/or spinal cord regions. ²⁰ 34 donors with TDP-43 proteinopathy that did not meet the threshold for a diagnosis of FTLD-TDP, such as limbic-predominant age-related TDP-43 encephalopathy-neuropathological change (LATE-NC) or Hippocampal Sclerosis with TDP-43 pathology (HSCL), as co-pathology were included under their NPDx1 category.²¹ TDP-43 stainings were not performed in 20 participants.

3. Neuronal α-synuclein disease (SYN; n=29) in the form of Lewy body disease (LBD) defined by presence of neuronal α-synuclein Lewy bodies and neurites in relevant brain regions. Clinically, these encompassed dementia with Lewy bodies and/or Parkinson’s disease with and without dementia. Their Braak Lewy body stages ranged between 3–6 (mean 4.7±1.1).

4. Primary tauopathies (TAU; n=37) such as progressive supranuclear palsy (PSP), Pick’s disease, and corticobasal degeneration (CBD), defined by their hallmark pTau aggregates in neurons, astrocytes, and oligodendrocytes.

5. Cerebrovascular disease (CVD; n=13).

6. Cerebral amyloid angiopathy (CAA; n=3).

In addition, 67 MADRC-LC study participants who had a clinical consensus diagnosis of “normal” with stable, unimpaired cognition (global CDR=0 at all visits) and no neurological diagnosis for at least 5 years of follow up, but who did not undergo brain autopsy, were included as normal controls (NC).

Demographic information, cognitive characteristics, and frequency of various co-pathologies for all participants are presented in Table 1. Initial analysis showed that plasma levels of pTau217, GFAP, and NfL increased with age and that GFAP levels were higher in males than females (eTable 2). Levels of all three biomarkers were negatively associated with cognitive status as defined by combined MoCA/MMSE scores²² at the time of the blood collection (i.e., higher levels with lower scores). Biomarker levels were not associated with disease duration as determined by clinician’s assessment of age of symptom onset or with time between blood draw and death (eTable 2).

Table 1.

Demographic, clinical, and histopathological information at the time of blood collection

Primary Neuropathological Diagnosis category	ADNC	CAA	TDP	SYN	TAU	CVD	NC
n	85	3	20	29	37	13	67
Sex, female	34 (40%)	2 (67%)	8 (40%)	8 (28%)	15 (41%)	5 (39%)	48 (72%)
Race, White	84 (99%)	3 (100%)	20 (100%)	28 (97%)	36 (97%)	12 (92%)	63 (94%)
Ethnicity, non-Hispanic	0	0	0	1 (3%)	0	1 (8%)	0
APOEe4 (≥1 allele)	56 (66%)	0 (0%)	6 (30%)	9 (31%)	6 (16%)	3 (23%)	19 (28%)
Age at death (years)	82.7 (57–98)	84.8 (66–98)	71.8 (51–82)	78.7 (61–97)	71.7 (32–92)	88.2 (83–96)	N/A
Time blood collection to death (years)	3.1 (0.0–6.0)	1.8 (0.8–3.0)	2.6 (0.2–5.7)	2.9 (0.6–6.0)	2.1 (0.4–6.0)	3.6 (1.2–4.9)	N/A
Cognitive status (NCI/MCI/Dem)*	3/18/64	0/1/2	0/3/17	3/12/14	2/11/24	1/9/3	67/0/0
Global CDR	1 (0–3)	1 (0.5–1)	2 (0.5–3)	0.5 (0–3)	1 (0–3)	0.5 (0–1)	0 (0–0)
MoCA/MMSE**	13 (0–29)	25 (21–29)	12 (0–22)	21 (3–29)	21 (0–29)	23 (14–29)	28 (22–30)
Number (frequency) of participants with evidence of AD and ADRD pathologies at autopsy
ADNC	85 (100%)	1 (33%)	2 (10%)	11 (38%)	1 (2.7%)	7 (54%)	n.d.
CAA	39 (46%)	3 (100%)	3 (15%)	8 (28%)	3 (8.1%)	3 (23%)	n.d.
TDP	20 (24%)	1 (33%)	20 (100%)	6 (21%)	6 (16%)	2 (15%)	n.d.
SYN	13 (15%)	0	0	29 (100%)	1 (2.7%)	1 (7.7%)	n.d.
TAU	1 (1.2%)	0	2 (10%)	3 (10%)	37 (100%)	0	n.d.
CVD	74 (87%)	3 (100%)	17 (85%)	27 (93%)	27 (73%)	13 (100%)	n.d.

Data presented as median (range) or n (%). ADNC=Alzheimer’s disease neuropathological changes; CAA=Cerebral amyloid angiopathy; TDP=TAR DNA-binding protein 43 kDa proteinopathy; SYN=Neuronal α-synuclein disease; TAU=Primary tauopathy; CVD=Cerebrovascular disease; NC=normal controls; CDR=Clinical Dementia Rating; MoCA=Montreal Cognitive Assessment; MMSE=Mini Mental State Examination; n.d.=not determined.

^*Cognitive status determined by global CDR (gCDR): No cognitive impairment (NCI): gCDR=0; Mild cognitive impairment (MCI): gCDR=0.5; Dementia (Dem): gCDR≥1.

^**Combined MoCA and MMSE.²²

Biomarker associations with ADNC

Using a linear regression model adjusting for age, sex, and time between blood draw and death, we found that plasma pTau217 levels were positively associated with Aβ deposits (Thal phase 3–5 vs 0–2; p<0.0001; Table 2), neuritic plaques [CERAD neuritic plaque (NP) score ≥1 (sparse, moderate or frequent) vs 0 (none); p<0.0001], and NFTs (Braak stage III-VI vs 0-II; p<0.0001). pTau217 levels increased stepwise with increasing Thal, CERAD NP, and Braak NFT scores (Fig 1; eTable 3), showing a correlation between higher pTau217 levels and higher ADNC burden. Remarkably, the associations between pTau217 levels and ADNC burden remained when limiting the analysis to participants with no cognitive impairment or MCI (global CDR≤0.5; n=63; Table 2) and pTau217 could differentiate between Braak stages III-IV vs I-II in this group (p<0.005), demonstrating the sensitivity of pTau217 to detect early ADNC. Moreover, pTau217 levels were independently associated with both Thal Aβ phase and Braak NFT stage as both associations remained statistically significant when both variables were added in the same regression model, although effect sizes were smaller (Table 2; eTable 3).

Table 2.

Associations between plasma biomarkers and ADNC.

Biomarker	Group	Neuropathological feature	Covariate	β (95% CI)	p
pTau217	All	Thal (3–5 vs 0–2)	--	1.56 (1.27–1.85)	<0.0001
		Thal (3–5 vs 0–2)	Braak	0.67 (0.27–1.06)	<0.002
		CERAD (≥1 vs 0)	--	1.73 (1.48–1.97)	<0.0001
		Braak (III–VI vs 0–II)	--	1.74 (1.50–1.98)	<0.0001
		Braak (III–VI vs 0–II)	Thal	1.23 (0.85–1.60)	<0.0001
	NCI/MCI	Thal (3–5 vs 0–2)	--	1.29 (0.77–1.82)	<0.005
		CERAD (≥1 vs 0)	--	1.47 (1.05–1.90)	<0.005
		Braak (III–VI vs 0–II)	--	1.75 (1.32–2.19)	<0.005
		Braak (III–IV vs I–II)	--	1.44 (0.90–1.99)	<0.005
GFAP	All	Thal (3–5 vs 0–2)	--	0.93 (0.61–1.25)	<0.0001
		Thal (3–5 vs 0–2)	Braak	0.31 (−0.17–0.79)	n.s.
		CERAD (≥1 vs 0)	--	1.03 (0.75–1.32)	<0.0001
		Braak (III–VI vs 0–II)	--	1.03 (0.74–1.33)	<0.0001
		Braak (III–VI vs 0–II)	Thal	0.80 (0.35–1.26)	<0.002
	NCI/MCI	Thal (3–5 vs 0–2)	--	0.77 (0.28–1.25)	<0.01
		CERAD (≥1 vs 0)	--	0.88 (0.46–1.30)	<0.005
		Braak (III–VI vs 0–II)	--	0.72 (0.22–1.21)	<0.01
		Braak (III–IV vs I–II)	--	0.55 (−0.08–1.19)	n.s.
NfL	All	Thal (3–5 vs 0–2)	--	−0.09 (−0.39–0.22)	n.s.
		Thal (3–5 vs 0–2)	Braak	−0.06 (−0.54–0.43)	n.s.
		CERAD (≥1 vs 0)	--	−0.05 (−0.35–0.25)	n.s.
		Braak (III–VI vs 0–II)	--	−0.10 (−0.40–0.20)	n.s.
		Braak (III–VI vs 0–II)	Thal	−0.02 (−0.48–0.45)	n.s.
	NCI/MCI	Thal (3–5 vs 0–2)	--	0.71 (0.19–1.23)	<0.02
		CERAD (≥1 vs 0)	--	0.71 (0.23–1.19)	<0.01
		Braak (III–VI vs 0–II)	--	0.67 (0.13–1.21)	<0.05
		Braak (III–IV vs I–II)	--	0.85 (0.14–1.56)	<0.05

Linear regression adjusting for the specified covariate, age, sex, and time between blood draw and death. Biomarker levels were scaled and log-transformed prior to analysis. NCI = No cognitive impairment (Global CDR = 0). MCI = Mild cognitive impairment (Global CDR = 0.5). Braak NFT stages: 0 = none, I-II = entorhinal, III-IV = limbic, and V-VI = neocortical. CERAD (Consortium to Establish a Registry for Alzheimer’s Disease) neuritic plaque score: 0 = none, 1 = sparse, 2 = moderate, and 3 = frequent. Thal phases of Aβ deposition: 0 = none, 1 = neocortex; 2 = hippocampus, 3 = striatum, 4 = brainstem, and 5 = cerebellum.

Figure 1. Plasma biomarker levels in relation to amyloid-β and tau pathology as measured by Thal Aβ phases, CERAD neuritic plaque scores, and Braak NFT stages.

Biomarker levels were scaled and log-transformed prior to analysis. Groups were compared stepwise using linear regression adjusting for age, sex, and time between blood draw and death, and displayed p-values were FDR corrected for multiple comparisons. *p<0.05; **p<0.01; ***p<0.005; ****p<0.001.

Plasma GFAP levels were similarly associated with both Aβ and tau pathology (p<0.0001 for all measures; Table 2), but while there was a marginal increase in GFAP levels with increasing CERAD NP scores, levels did not increase further with more advanced Thal Aβ phases or Braak NFT stages (Fig 1; eTable 3). GFAP levels were also associated with Thal Aβ, CERAD NP, and Braak NFT stages in participants with a global CDR≤0.5 but, unlike pTau217, could not differentiate between Braak stages III-IV vs I-II in this group (Table 2). In addition, unlike pTau217, GFAP was primarily associated with Braak NFT stage since there was no independent association between GFAP and Thal Aβ phases remaining when adjusting the regression model for Braak NFT stages (Table 2).

Lastly, NfL levels were not associated with Thal Aβ, CERAD NP, or Braak NFT scores in the full cohort (Table 2; Fig 1), but associations were observed when limiting the analysis to the participants with a global CDR≤0.5 (Table 2).

Of note, these results did not change when performing a sensitivity analysis including disease duration and cognitive status (MoCA/MMSE) in the regression models (eTable 4).

Biomarker levels in relation to NPDx1

Plasma levels of pTau217 and GFAP were increased in participants with an NPDx1 of ADNC (i.e., intermediate or high ADNC burden) compared to all other neuropathological categories (excluding CAA due to the small sample size; Fig 2A, C). Compared to NC, pTau217 levels were also higher in individuals with an NPDx1 of SYN (p<0.001), CVD (p<0.001), or TAU (p<0.05), but the differences between the groups were small. Using a pTau217 threshold to discriminate between ADNC NPDx1 and NC of 8.3 pg/mL (AUC 0.97, sensitivity 96%, specificity 100%), 23% of the individuals in the non-AD NPDx1 categories had borderline or moderately increased pTau217 levels.

Figure 2. Plasma biomarker levels in relation to neuropathological category in the presence vs absence of Alzheimer’s disease neuropathological change (ADNC) co-pathology.

Left column (A, C, E): Biomarker levels in relation to the primary neuropathological diagnosis (NPDx1) regardless of the presence or absence of additional co-pathology. Right column (B, D, F): Biomarker levels in participants with a single NPDx1 and no co-pathologies. Participants with multiple co-pathologies are shown separately as AD+ (ADNC plus one or more co-pathologies) or Mult (multiple non-AD co-pathologies). CVD was not considered a co-pathology in this analysis due to its high prevalence across categories and since no increase in pTau217 levels was observed in individuals with CVD compared to controls. Biomarker levels are presented in pg/mL for pTau217 and GFAP but were scaled and log-transformed prior to analysis. Groups were compared using linear regression adjusting for age and sex, and displayed p-values were FDR corrected for multiple comparisons. ADNC was defined as high or intermediate ADNC burden. AD=Alzheimer’s disease; CAA=Cerebral amyloid angiopathy; TDP=TAR DNA-binding protein 43 kDa proteinopathy; SYN=Neuronal α-synuclein disease; TAU=Primary tauopathy; CVD=Cerebrovascular disease; NC=Normal controls. *p<0.05; **p<0.01; ***p<0.005; ****p<0.001. Color lines represent median (yellow), percentile 75 (green) and percentile 25 (red). Dotted line represents optimal cutpoints between ADNC and NC as defined by the Youden index.

GFAP levels were increased in all neuropathological categories compared to NC (TDP: p<0.002; all other groups: p<0.001) and, considering a GFAP threshold to discriminate between ADNC NPDx1 and NC of 82.4 pg/mL (AUC 0.88, sensitivity 93%, specificity 82%), 57% of the individuals with a non-AD NPDx1 had GFAP levels exceeding this threshold.

NfL levels displayed a different pattern than the other two biomarkers. While NfL levels were increased in ADNC NPDX1 compared to NC, the highest NfL levels were observed in individuals with an NPDx1 of TDP followed by individuals with an NPDx1 of TAU, CVD, ADNC, and SYN (Fig 2E). Thus, NfL levels were increased in all neuropathological categories compared to NC (p<0.001 for all comparisons), and the majority of all individuals (78%) with a non-AD NPDx1 had NfL levels exceeding the NfL threshold best discriminating between ADNC NPDx1 and NC (206.8 pg/mL; AUC 0.87, sensitivity 81%, specificity 93%).

Effects of ADNC co-pathology on biomarker levels

The majority of the participants (n=170; 91%) had evidence of more than one disease pathology in the postmortem brain examination, with CVD being the most common co-pathology (present in 85% of all participants) followed by CAA (30%), ADNC (22%), TDP (21%), SYN (10%), and TAU (4%). 86 participants (46%) exhibited three or more co-pathologies (Fig 3). To determine if ADNC co-pathology contributed to the increased pTau217 and GFAP levels observed in cases with other non-AD NPDx1 categories, we analyzed biomarker levels across NPDx1s including only individuals with a single neurodegenerative disease pathology. As the majority of participants (85%) had evidence of vascular pathology on postmortem examination, it was not feasible to exclude these individuals. A sensitivity analysis did, however, demonstrate that there were no differences in pTau217 levels between individuals with CVD in the absence of ADNC (n=66) and NC [n=67; pTau217: β=0.11 (−0.08–0.32), p=0.2], suggesting that the inclusion of vascular pathology would not affect pTau217 results further. Removing individuals with ADNC co-pathology from the non-AD NPDx1 categories showed that increased levels of pTau217 in these individuals were mostly driven by ADNC co-pathology as only 8.8% of individuals without ADNC co-pathology had increased pTau217 levels, most of them only marginally above the threshold (Fig 2B). Furthermore, there were no differences in pTau217 levels between NC and the non-AD NPDx1 categories except for borderline increased levels in individuals with an NPDx1 of SYN [β=−0.49 (−0.88 to −0.10), p<0.05] or TAU [β=−0.35 (−0.62 to −0.09), <0.05]. Similarly, when stratifying individuals based on whether they had ADNC regardless of the NPDx1, pTau217 levels were higher in the presence of ADNC in all non-AD categories except the TAU category, in which there was only one participant (Fig 4A).

Figure 3. Histopathological evidence of AD and ADRD pathology (horizontal) within the different primary neuropathological categories (vertical).

AD=Alzheimer’s disease; CAA=Cerebral amyloid angiopathy; TDP-43=TAR DNA-binding protein 43 kDa proteinopathy; SYN=Neuronal ɑ-synuclein disease; TAU=Primary tauopathy; CVD=Cerebrovascular disease.

Figure 4. Effects of Alzheimer’s disease neuropathological change (ADNC) on plasma biomarker levels in participants with evidence of isolated (CAA, TDP, SYN, TAU, and CVD) or multiple (Multiple) non-AD pathologies irrespectively of primary neuropathological diagnosis or clinical syndromes.

Individuals with CVD were not removed from the other histopathological groups in this analysis due to the high prevalence of CVD pathology across categories and since no increase in pTau217 levels was observed in individuals with CVD compared to controls. pTau217 and GFAP levels are presented in pg/mL but were scaled and log-transformed prior to analysis. Groups were compared using linear regression adjusting for age and sex, and displayed p-values were FDR corrected for multiple comparisons. ADNC was defined as high or intermediate ADNC burden. AD=Alzheimer’s disease; CAA=Cerebral amyloid angiopathy; TDP=TAR DNA-binding protein 43 kDa proteinopathy; SYN=Neuronal α-synuclein disease; TAU=Primary tauopathy; CVD=Cerebrovascular disease. *p<0.05; **p<0.01; ***p<0.005; ****p<0.001. Color lines represent median (yellow), percentile 75 (green) and percentile 25 (red). Dotted line represents optimal cutpoints between ADNC and NC as defined by the Youden index.

By contrast, GFAP was less specific for ADNC since 48% of individuals with a non-AD NPDx1 and no ADNC co-pathology had GFAP levels exceeding the upper threshold for NC, although levels were only significantly increased in individuals with an NPDx1 of TAU [β=−0.83 (−1.20 to −0.47), p<0.005] or TDP [β=−0.59 (−1.06 to −0.12), p<0.05] (Fig 2D). Consequently, the AUC to differentiate between the presence or absence of ADNC co-pathology for GFAP was only 0.75 (78% sensitivity, 73% specificity), compared to an AUC for pTau217 of 0.91 (89% sensitivity, 93% specificity). GFAP levels were, however, still numerically higher in the presence vs. absence of ADNC co-pathology in all non-AD NPDx1 categories. This increase was statistically significant for the groups with CAA, SYN, TAU, and multiple non-AD pathologies (Fig 4B).

NfL levels were increased in individuals with an NPDx1 of CVD compared to NC [β=−1.11 (−1.83 to −0.39); p<0.005] but did not differ in individuals with an NPDx1 of CVD in the presence or absence of ADNC co-pathology [β=0.03 (−0.76 to 0.81); p=0.9], making it difficult to assess the individual contributions of the different co-pathologies on NfL levels due to the high frequency of CVD in this sample (Fig 2F, Fig 4C).

Effects of CAA, TDP, SYN, and CVD co-pathology on biomarker levels in individuals with and without ADNC

Next, we addressed whether additional co-pathologies increased biomarker levels in individuals with an NPDx1 of ADNC. As the number of participants with isolated co-pathologies was small, we elected not to exclude participants with multiple co-pathologies from this analysis. Using this approach, the presence of CAA, TDP, SYN, and CVD co-pathology in individuals with an NPDx1 of ADNC did not result in increased pTau217, GFAP, or NfL levels, whereas there was a small decrease in pTau217 levels in the presence of TDP co-pathology [β=−0.42 (−0.71 to −0.13), p<0.05; Fig 5]. All but two participants with TDP and ADNC had LATE+/−HSCL (eFig 1). pTau217 levels were slightly lower in ADNC in the presence vs absence of LATE+/−HSCL [β=−0.35 (−0.64 to −0.07), p<0.05; eFig 2], while the presence of LATE+/−HSCL did not affect levels of GFAP and NfL. GFAP and NfL levels were high in two participants with ADNC and concurrent non-AD TAU, but the sample size was too small for a meaningful analysis. Similarly, there were no differences in pTau217 or GFAP levels in individuals with and without CAA, TDP, SYN, TAU, or CVD when analyzing all participants without ADNC (eFig 3). TDP was associated with higher NfL levels [β=0.80 (0.33 to 1.26), p<0.02], which predominantly was observed in participants with ALS-TDP and FTLD-TDP (eFig 1). SYN was, in contrast, associated with lower NfL levels [β=−0.90 (−1.43 to −0.36), p<0.05] in the participants without ADNC, but the compounding effects of the different co-pathologies make the analysis of NfL levels difficult.

Figure 5. Effects of co-pathologies on biomarker levels in participants with a primary neuropathological diagnosis (NPDx1) of Alzheimer’s disease neuropathological change (ADNC).

Biomarkers were compared in participants with an NPDx1 of intermediate or high ADNC burden in the presence or absence of co-pathologies on neuropathological examination. Individuals with multiple co-pathologies were not excluded from the analysis. Groups were compared using linear regression adjusting for age and sex. CAA=Cerebral amyloid angiopathy; TDP=TAR DNA-binding protein 43 kDa proteinopathy; SYN=Neuronal α-synuclein disease; TAU=Primary tauopathy; CVD=Cerebrovascular disease. *p<0.05. Color lines represent median (yellow), percentile 75 (green) and percentile 25 (red).

Discussion

In this study, we used a well-characterized longitudinal cohort with autopsy neuropathological data to define associations of BBM with ADNC, non-AD-pathologies, and mixed pathologies. Our findings confirmed that pTau217 is highly specific for ADNC even in non-AD dementias because, while almost 25% of individuals with a nonAD NPDx1 had increased levels of pTau217, the vast majority of these individuals concurrently had intermediate or high ADNC burden at autopsy. This observation has important implications for the interpretation of plasma pTau217 levels in the clinical setting since a “positive” plasma pTau217 test should not necessarily exclude an alternative ADRD diagnosis and/or be considered a false positive if an ADRD diagnosis is highly suspected. Rather, it would support the consideration of ADNC as a factor in or contributor to the patient’s cognitive impairments given its strong correlation with intermediate and high ADNC burden even in participants with a non-AD NPDx1 shown here. Moreover, our findings call for caution about relying heavily on plasma pTau217 levels to guide the prescription of the new FDA-approved anti-Aβ monoclonal antibodies²³ and highlight both the importance of a careful and thorough patient’s examination and the unmet need for equally definitive BBM of α-synuclein, TDP-43, and non-AD tau pathologies.

We observed a marginal increase in pTau217 concentrations at the group level in individuals with a primary tauopathy or α-synuclein proteinopathy as their NPDx1, but levels were distinctly lower than in ADNC and none of the individuals had clearly AD-range levels. The causes for this modest increase in pTau217 levels in individuals with a non-AD NPDx1 are unclear. While it is difficult to exclude that the underlying non-AD pathology contributes to the increased pTau217 levels in these individuals, their levels were close to the threshold and may reflect an overlap in pTau217 levels between ADNC positive and negative individuals in this range. A two-cutoff approach, which classifies some results as borderline, has been suggested to improve the diagnostic accuracy of pTau217.^{1, 3, 24} We used categorical Thal Aβ phases, CERAD NP scores, and Braak NFT stages to assess ADNC, which may be less sensitive for early or atypical distributions of pathology compared to more quantitative, regionally resolved analyses and it is possible that the increased soluble pTau217 levels in these individuals reflect very early ADNC changes, not yet reflecting abundant NP and paired helical filament (PHF)-tau NFTs on neuropathological examination. Together, these data are consistent with those from a few partially overlapping autopsy studies failing to show increased pTau217 levels in small numbers of individuals without ADNC.^{4–6, 9} A caveat is that our cohort did not contain any individuals with rapidly progressing neurodegenerative dementias (RPD) such as prion disease, which recently have been shown to be associated with increased plasma pTau217 levels.²⁵ The increased pTau217 levels in RPDs prompt caution that, despite the rapidly growing literature associating increased pTau217 levels with ADNC, pTau217 levels can likely also increase secondary to other proteinopathies with more extensive tissue damage.

Our data confirm previous neuropathology studies showing that pTau217 is independently associated with both Aβ plaques and tau NFTs.^{4, 6, 9} pTau217 levels are increased early in AD, at the time when amyloid PET becomes positive but before tau PET turns positive,^26–28 at least with current radioligands. Together this would suggest that pTau217 is a marker for Aβ plaques. However, the discrepancy in the timing of plasma pTau217 and tau PET is proposed to be caused by the tau PET tracer binding to insoluble PHF-tau protein aggregates²⁹ while plasma and CSF pTau217 detect soluble pTau released from neurons. pTau217 has thus been suggested to reflect an early physiological tau response to Aβ plaque development.^{6, 30–32} Recently, plasma levels of MTBR-tau243, binding to an endogenously cleaved, microtubule-binding region of tau, have been shown to correlate better than plasma pTau217 with tau PET and show promise as a BBM for insoluble tau aggregate pathology in AD.³³

While pTau217 is associated with ADNC, our data confirm that GFAP is not. GFAP levels were increased in ADNC but also increased in over 50% of the individuals with a non-AD NPDx1, even in those without substantial ADNC burden at autopsy. This is consistent with GFAP being upregulated not only in reactive astrocytes surrounding Aβ plaques and tau NFTs in the AD brain,^{34, 35} but also in response to many other challenges to the local microenvironment.¹² Increased plasma levels of GFAP are observed in a range of other conditions including FTLD, traumatic brain injury, vascular dementia, and multiple sclerosis.³⁶ We observed that although plasma GFAP is associated with both Aβ plaques and tau NFTs, the association with Aβ is attenuated when adjusting for tau.^{7, 8} Reactive astrogliosis caused by Aβ may trigger pathological tau phosphorylation resulting in AD progression and providing a link between Aβ and tau,³⁷ although the mechanisms by which this might occur are unknown. Brain pTau levels are well known to correlate with cognitive decline³⁸ and, while not specific for AD, plasma GFAP levels have previously been shown to discriminate between symptomatic and pre-symptomatic AD, predict disease progression, and correlate with cortical thinning and cognitive impairment in AD.^{7, 39–41} GFAP was furthermore associated with all-cause dementia, including AD, in a large-scale proteomic profiling of over 50,000 adults without dementia at baseline and was shown to predict AD more than 10 years prior to diagnosis.⁴² This suggests that increased plasma GFAP levels can be used as an early non-specific signal for incipient neurodegeneration, which should trigger additional testing in positive individuals.

Plasma levels of NfL demonstrated a different expression pattern compared to pTau217 and GFAP, consistent with NfL being a non-specific marker for neuro-axonal damage.¹³ While plasma NfL was not correlated with ADNC in the whole cohort, such association was observed in the subgroup with NCI/MCI. An uncoupling between NfL and ADNC was demonstrated in autosomal dominant AD (ADAD) where the rate of NfL change during longitudinal follow-up was associated with gray-matter atrophy but not with Aβ accumulation as measured by PET imaging.⁴³ While NfL measurements can be used to measure disease “activity”, provide prognostic information, and follow treatment efficacy in several neurological conditions,¹³ the ubiquitous increase in NfL in most neurodegenerative diseases commonly co-occurring with AD complicates the interpretation of change in NfL in AD.⁴⁴

FTLD is a heterogeneous group of neurodegenerative pathologies largely characterized by misfolded TDP-43 (FTLD-TDP) or tau (FTLD-Tau) proteins, manifesting themselves as frontotemporal dementia (FTD), an umbrella term of often overlapping clinical syndromes.⁴⁵ We elected to group participants with the different FTD syndromes together under their respective proteinopathies for our main analysis due to the small sample sizes even though their genetic, biological, and regional heterogeneity may mask associations of interest and influence biomarker interpretation. GFAP and NfL are increased in autopsy-confirmed FTLD-TDP and FTLD-Tau^{45, 46}, consistent with our findings of elevated levels of GFAP and NfL in individuals with TDP and TAU pathology. It has been proposed that NfL and the GFAP/NfL ratio can be used to discriminate between FTLD-TDP and FTLD-Tau, as NfL is relatively higher in FTLD-TDP while GFAP displays poor discriminatory ability between the two groups.^{45, 46} Compared to other neurodegenerative diseases, plasma levels of NfL are often highest in ALS, a TDP-43 proteinopathy affecting large myelinated motor neurons, where both the aggressive neurodegeneration and the anatomical distribution are thought to contribute to the release of large amounts of NfL to CSF and plasma.⁴⁷ pTau181 levels have previously been shown to be increased in ALS,⁴⁸ but we did not observe a corresponding increase in pTau217 levels in our cohort although numbers were small (ALS-TDP: n=5). By contrast, 3 of these 5 individuals had increased levels of pTau181 measured in parallel (6.3–9.9 pg/mL; NC mean±SD: 2.3±1.5 pg/ml; eFig 4).

Our study has multiple strengths. We studied a well characterized longitudinal cohort with standardized clinical assessments and pre-analytical procedures. We used highly sensitive biomarker assays with high precision across the study. Our sample size (n=187) is larger than many of the existing BBM-neuropathology correlation studies, and we assessed multiple non-AD co-pathologies in addition to ADNC. The time between plasma collection and death (mean 2.8 years) was relatively short. Limitations of the study are that most participants died with advanced disease, making it difficult to generalize findings to earlier stages of the disease. While the MADRC-LC cohort recently has become more diverse, this autopsy cohort is predominantly consisting of non-Hispanic White individuals, limiting the generalizability to a more diverse population. It is furthermore not a population-based cohort and may have different proportions of co-pathologies compared to the general population. Our NC did not have postmortem examination performed, and it is possible that some of them would have evidence of pre-symptomatic neurodegenerative disease on neuropathological examination. We attempted to minimize this likelihood by using blood samples from individuals without cognitive symptoms (i.e., global CDR=0) for at least 5 years after the blood collection. Finally, while the high frequency of co-morbidities (with 91% of all participants having more than one disease pathology) may provide insights into the complicated interplay between different pathological processes, it precludes the possibility of studying single pathologies. We elected not to exclude individuals with vascular pathology from the analysis of single pathologies due to its high prevalence and the absence of differences in pTau217 levels between CVD and NC. The underlying vascular pathologies may, however, be heterogeneous in mechanism, severity, and timing, and it is possible that interactions between Aβ and vascular pathology may result in enhanced biomarker expression.⁴⁹

In summary, utilizing a well characterized neuropathology cohort, we show that plasma pTau217 levels, but not GFAP or NfL levels, accurately reflect the presence of ADNC in the brain even in individuals with a primary neuropathological diagnosis of an ADRD. Thus, a positive plasma pTau217 test in an individual with a suspected non-AD dementia should not be considered a false positive, but rather as evidence of ADNC co-pathology.