Influence of cognitive impairment and race on plasma p‐Tau₂₁₇ in two diverse cohorts

Abstract

INTRODUCTION

Factors influencing plasma Alzheimer's disease (AD) biomarkers remain incompletely understood. Here we evaluated Fujirebio plasma p‐Tau₂₁₇ in two diverse cohorts among whom 91% underwent cerebrospinal fluid (CSF) analysis.

METHODS

Non‐Hispanic White (NHW, n = 113), Black/African American (B/AA, n = 66), and Chinese American (ChA, n = 38) participants recruited from two universities were included. We examined if plasma p‐Tau₂₁₇ correlated with CSF and clinical factors, differed between racial groups, and associated with novel CSF proteins.

RESULTS

CSF p‐Tau₁₈₁ strongly correlated with CSF p‐Tau₂₁₇ (R ² = 0.912) which moderately correlated with plasma p‐Tau₂₁₇ (R ² = 0.694). Plasma p‐Tau₂₁₇ levels were higher with greater cognitive impairment but lower in B/AA than NHW participants even after adjusting for CSF p‐Tau₁₈₁. This resulted in greater positive predictive value for NHW than B/AA participants, and could be mediated by complement or lysosomal pathways.

DISCUSSION

Severity of cognitive impairment and race both influence plasma p‐Tau₂₁₇ levels beyond race‐associated differences in CSF p‐Tau₁₈₁.

Highlights

• Cognitive impairment associates with plasma p‐Tau₂₁₇ independent of CSF biomarkers.

• Black/African Americans had lower plasma p‐Tau₂₁₇ than non‐Hispanic White Americans.

• CSF p‐Tau₁₈₁ could not explain lower plasma p‐Tau₂₁₇ in Black/African Americans.

• Plasma p‐Tau₂₁₇ difference results in more false positive cases according to race.

• Novel CSF processes were associated with race‐related plasma p‐Tau₂₁₇ difference.

1. BACKGROUND

Objective biomarkers for Alzheimer's disease (AD) can greatly enhance its early detection and treatment. Currently established biomarkers include cerebrospinal fluid (CSF) levels of core AD proteins (Aβ42, Aβ40, t‐Tau, and p‐Tau₁₈₁) and substrate‐specific positron emission tomography (PET) targeting amyloid and tau aggregates. ¹ Blood‐based biomarkers for AD have long been championed as accessible and convenient ² , but their broader adaptation in the past has been limited by suboptimal performance or reproducibility. ³ , ⁴ , ⁵ , ⁶ More recently, mass spectrometry‐based approaches ⁷ or highly sensitivity immunoassays ⁸ have improved the detection of AD‐related protein levels in blood. The latest generation of blood‐based AD biomarkers have been found to strongly associate with post mortem AD neuropathologic changes ⁹ , discriminate between tau PET profiles ⁹ yet better correlate with amyloid PET ¹⁰ , ¹¹ , ¹² , and predict subsequent cognitive decline in the asymptomatic AD stage. ¹³ At the same time, their real‐world applicability has often fallen short of these promising group‐level epidemiological findings, with potential confounds from renal clearance, cardiovascular co‐morbidities ¹⁴ , ¹⁵ , and potential blood–brain barrier (BBB) disruption in the setting of non‐AD neurodegeneration ¹⁶ and progressive dementia. ¹⁷ While plasma p‐Tau₂₁₇ may better predict brain AD neuropathologic changes than plasma p‐Tau₁₈₁, insufficient number of participants having CSF p‐Tau₂₁₇ measures using widely available assays in prior studies makes comparison of two phosphorylated forms of tau difficult to interpret.

We were the first to report lower CSF levels of AD‐related tau biomarkers (t‐Tau, p‐Tau₁₈₁) in older Black/African American (B/AA) than non‐Hispanic White (NHW) adults ¹⁸ , and this has now been replicated in older ¹⁹ , middle‐aged ²⁰ , and younger ²¹ adults. This phenomenon was not observed until a sufficient number of diverse participants were recruited into modern AD biomarker research, and poses a direct challenge to the equitable application of AD fluid biomarkers for older B/AA adults and potentially Hispanic, Asian, and other populations yet to regularly participate in AD biomarker research. This is because a threshold for CSF t‐Tau or p‐Tau₁₈₁ determined in a largely NHW population would under‐estimate the number of B/AA adults with AD neuropathologic changes. The reason for these differences remains unknown, although differences are variably associated with alternate inflammatory pathways following amyloid deposition to social determinants of health. Because p‐Tau₂₁₇ represents another threonine site phosphorylated on tau, we hypothesize that p‐Tau₂₁₇ levels would also differ between B/AA and NHW adults. Based on promising findings related to plasma p‐Tau₂₁₇ across analytical platforms, cohorts, and even racial/ethnic groups ¹¹ , ²² , we evaluated its performance and examined its relationship with established CSF AD biomarkers and demographic/clinical variables in two well‐characterized diverse cohorts.

2. METHODS

2.1. Ethical conduct of study

This study was approved by the Rutgers University and Emory University Institutional Review Boards. Written informed consents were previously obtained from all participants in accordance with the Declaration of Helsinki and the Belmont report for the long‐term storage and future analysis of samples.

2.2. Diversity, equity, and inclusion

Two cohorts were included in the study (Table 1). The Rutgers University cohort consists of older Chinese American (ChA), B/AA, and NHW adults recruited through the Center for Healthy Aging Research, Cognitive Neurology and AD Clinic, and the greater New Jersey area between 2021 and 2024 through direct outreach and community partnerships. In particular, B/AA recruitment was facilitated via a community‐based outreach coordinator and consultation with the local National Association for the Advancement of Colored People; ChA recruitment was conducted via culturally and linguistically appropriate materials and presentations by native Mandarin speakers. The previously described Emory University cohort ¹⁸ included older B/AA and NHW adults recruited through the AD Research Center, Cognitive Neurology Clinic, and the greater Atlanta area between 2013 and 2015 through direct outreach, community partnerships, and leadership by non‐NHW researchers. All authors are women, People of Color, or both.

TABLE 1.

Participants included in the current study.

Parameter	Rutgers (n = 98)				Emory (n = 123)
Race	ChA (n = 38)	B/AA (n = 11)	NHW (n = 44)	Other (n = 5)	B/AA (n = 55)	NHW (n = 68)
Women (%)	19 (50%)	10 (91%)	24 (54%)	4 (80%)	29 (53%)	39 (57%)
Age, median (IQR)	67 (64, 73)	70 (64, 74)	68 (62, 73)	72 (62, 82)	68 (64, 73)	69.5 (65, 76)
Education, median (IQR)	18 (16, 18)	16 (14, 18)	16 (13, 18)	17 (10, 18)	16 (14, 18)	16 (14, 18)
Having ≥1 APOE e4 allele	10 (27%)	3 (27%)	17 (39%)	1 (20%)	29 (53%)	35 (51%)
Diagnosis
NC	27	2	13	0	23	28
MCI	9	4	16	2	23	25
AD dementia	2	4	8	3	9	15
Other dementia	0	1	7	0	0	0
CSF analysis a	18 (47%)	11 (100%)	45 (100%)	100%	55 (100%)	68 (100%)
Aβ42, mean (SD)	656 (385)	543 (191)	622 (292)	537 (433)	214 (117)	207 (148)
t‐Tau, mean (SD)	327 (236)	453 (356)	453 (251)	419 (181)	47.7 (31.7)	71.5 (47.8)
p‐Tau181, mean (SD)	50.4 (41.2)	73.4 (62.0)	63.9 (40.1)	66.2 (36.9)	17.8 (9.4)	25.6 (12.6)
Plasma p‐Tau217, mean (SD)	0.16 (0.22)	0.34 (0.46)	0.37 (0.29)	0.27 (0.20)	0.22 (0.28)	0.27 (0.28)

Abbreviations: AD, Alzheimer's disease; APOE, apolipoprotein E; B/AA, Black/African American; ChA, Chinese American; CSF, cerebrospinal fluid; IQR, interquartile range; MCI, mild cognitive impairment; NC, normal cognition; NHW, non‐Hispanic White.

^a CSF analysis performed using Lumipulse for Rutgers participants, and AlzBio3 assays on Luminex for Emory participants. All biomarker concentrations are reported in pg/mL. Other race/ethnicity in Rutgers cohort: Hispanic (n = 2), Asian Indian (n = 1), Japanese (n = 1), and Filipino (n = 1) American participants.

RESEARCH IN CONTEXT

1. Systematic review: We reviewed the literature using traditional sources (e.g., PubMed, Google Scholar; keywords: biomarkers, tau, blood/plasma, Alzheimer's disease [AD], and related terms). Elevated plasma p‐Tau₂₁₇ is strongly associated with cerebral amyloid and tau positron emission tomography (PET) positivity, post mortem Alzheimer's disease (AD), reduced renal clearance, and cardiovascular diseases. Only two studies included sufficient number of Black/African American (B/AA) participants, but neither had sufficient number with cerebrospinal fluid (CSF) p‐Tau₁₈₁ which is known to be lower in B/AA than non‐Hispanic White (NHW) participants.

2. Interpretation: We showed very strong correlation between p‐Tau₁₈₁ and p‐Tau₂₁₇ in the CSF, and each moderately correlated with plasma p‐Tau₂₁₇. In two independent cohorts, we also found plasma p‐Tau₂₁₇ to associate with clinical diagnosis and B/AA race independent of CSF p‐Tau₁₈₁, potentially compounding racial disparities linked to tau biomarkers.

3. Future directions: Further work is needed to equitably apply plasma p‐Tau₂₁₇ and identify factors underlying reported racial disparities.

2.3. Participant characterization

All Rutgers and Emory participants underwent detailed clinical, neuropsychological, and magnetic resonance imaging (MRI) analysis, with Mini‐Mental State Examination (MMSE) scores available only in the Emory cohort. Race/ethnicity was by self‐report. All B/AA and NHW participants underwent cerebrospinal fluid (CSF) and non‐fasting plasma sample collection between 8AM and noon on the same day, with CSF donation an optional component for the ChA participants due to our more recent engagement with this group. Diagnosis of normal cognition (NC), mild cognitive impairment (MCI), AD dementia, or non‐AD dementia was derived through a consensus process. Non‐AD dementia participants included two patients with dementia with Lewy bodies (CSF positive for a‐synuclein RT‐QuIC) and four patients with behavioral variant frontotemporal dementia (bvFTD; two with GRN mutations and two with CSF not consistent with AD). Apolipoprotein E (APOE) genotyping was performed at Center for Applied Genomics (Children's Hospital of Philadelphia) for Rutgers participants and at Emory for Emory participants. 

2.4. CSF and plasma collection and processing

CSF was collected into 15 mL polypropylene tubes via 24‐gauge atraumatic needle and syringe aspiration between 8AM and noon. Emory CSF samples were immediately inverted several times, aliquoted (500 µL), and frozen at −80°C. Rutgers CSF samples were centrifuged 600 × g at room temperature for cellular studies ²³ , and supernatants were then inverted, aliquoted (500 µL), and frozen at −80°C until analysis. For each participant, 20 mL of non‐fasting whole blood was drawn into K₂‐EDTA tubes and immediately centrifuged at 2400 rpm for 15 min at 4°C to derive plasma which was aliquoted (500 µL) and frozen at −80°C until analysis.

2.5. CSF and plasma biomarker analysis

CSF levels of Aβ42, t‐Tau, and p‐Tau₁₈₁ were previously measured in Emory participants using the AlzBio3 assay (Fujirebio Diagnostics Inc, Malvern, PA) on the bead‐based Luminex platform (Austin, TX), with a threshold of t‐Tau/Aβ42 ≥ 0.39 for AD neuropathologic changes. ²⁴ Additional biomarkers previously measured in this cohort included CSF levels of Aβ40 (INNOTEST, Fujirebio Diagnostics, Malvern, PA), neurofilament light chain (NfL; NF‐light, Uman Diagnostics, Umeå, Sweden) ¹⁸ , neurogranin (Ng), complement‐related proteins (C3, C3b, C1q, and C4), and 15 inflammatory proteins including soluble trigger receptor expressed on myeloid cells 2 (sTREM2) and soluble tumor necrosis factor receptors 1 and 2 (sTNFR1, sTNFR2; MilliporeSigma, Burlington, MA) ²⁰ .

CSF levels of Aβ40, Aβ42, t‐Tau, and p‐Tau₁₈₁ were measured in Rutgers participants using the automated Lumipulse G1200 platform (Fujirebio Diagnostics Inc, Malvern, PA) ²⁵ with a threshold of t‐Tau/Aβ42 ≥0.58 for AD neuropathologic changes converted from AlzBio3‐based threshold. For Lumipulse, our laboratory achieves median intermediate precision of 4.2% for Aβ40, 6.4% for Aβ42, 6.2% for t‐Tau, and 4.5% for p‐Tau₁₈₁ across calibrator and cartridge lots. In a subset of this cohort (n = 44), CSF levels of NfL, Ng, sTNFR1, sTNFR2, complement‐related proteins, and other inflammatory proteins were measured using aptamer‐based assays (SomaLogic, Boulder, CO) as part of a larger multi‐center biomarker study (1084 analytes in 350 participants; Hu, unpublished) ²³ .

All plasma p‐Tau₂₁₇ levels were measured using Lumipulse G1200 (Fujirebio Diagnostics Inc, Malvern, PA) following manufacturer's protocol. Plasma samples were thawed on ice, vortexed for 10‐15 seconds, and then centrifuged at 2000 × g for 5 min at room temperature. For CSF p‐Tau₂₁₇, a preliminary experiment using five CSF samples and various dilutions (1:10, 1:20, 1:50, and 1:100) determined the optimal dilution of 1:50 for using the plasma‐based assay. For participants in the CSF‐plasma p‐Tau₂₁₇ correlation, CSF samples were thawed, diluted in distilled water before loading onto Lumipulse for analysis.

To assess the impact of BBB permeability on the relationship between CSF and plasma p‐Tau₂₁₇, CSF and plasma albumin were measured in 36 Rutgers participants using a bromocresol green‐based BCG albumin assay kit (Cat# MAK‐124, Millipore Sigma, Burlington, MA, USA) following the manufacturer's protocol (absorbance at 620 nm). The Albumin Quotient (Q‐Alb) was calculated as the ratio of CSF albumin (mg/dL) to plasma albumin (g/dL).

2.6. Intermediate precision analysis for plasma p‐Tau₂₁₇

To determine intermediate precision for plasma p‐Tau₂₁₇ between adjacent aliquots, we identified 13 Rutgers participants with six frozen aliquots of plasma representing a range of CSF p‐Tau₁₈₁ concentrations. Plasma p‐Tau₂₁₇ levels for these samples were measured in the morning and in the afternoon on three separate days by two experienced Lumipulse operators.

2.7. Statistical analysis

All statistical analysis was performed in IBM SPSS 29.0 (Aramonk, NY). To determine intermediate precision within the same day, coefficients of variation (CV, in %) were calculated for each day's AM and PM run of 13 samples across 3 days, and the average concentrations and pair‐wise %CV values were plotted to generate the intra‐day precision profile. To determine intermediate precision across days, average concentrations and overall %CV were calculated separately for AM and PM runs.

For relationships between plasma p‐Tau₂₁₇ and CSF p‐Tau measured at different threonine epitope (181 vs. 217) and by different assay platforms (Lumipulse vs. Luminex), we compared the correlation between plasma p‐Tau₂₁₇ and (1) CSF p‐Tau₂₁₇ measured by Lumipulse in 72 Rutgers participants (same analyte, same platform); (2) CSF p‐Tau₁₈₁ in 79 Rutgers participants measured by Lumipulse (different analyte, same platform); and (3) CSF p‐Tau₁₈₁ in 123 Emory participants measured by Luminex (different analyte, different platform).

To determine factors associated with plasma p‐Tau₂₁₇, continuous variables were examined if normally distributed (age, CSF, Aβ42) or after transformation if not normally distributed (log₁₀‐transformation for t‐Tau, p‐Tau₁₈₁, p‐Tau₂₁₇; square root transformation for MMSE in the Emory cohort). For linear regression models of transformed CSF or plasma p‐Tau₂₁₇, categorical variables (gender, race/ethnicity, having ≥1 APOE ε4 allele, diagnosis, chronic renal failure [CRF], congestive heart failure [CHF]) were entered as fixed factors, and continuous variables (age, CSF Aβ42, p‐Tau₁₈₁) were entered as co‐variates, with stepwise removal of variables. Variance inflation factors (VIFs) were also determined in models for p‐Tau₂₁₇ to account for collinearity between clinical diagnosis and biomarkers in the Rutgers cohort; between clinical diagnosis, MMSE, and biomarkers in the Emory cohort. MMSE was introduced in the Emory cohort to determine if a screening test – without clinicians and additional testing – could account for the effect of diagnosis. Diagnostic performance of plasma p‐Tau₂₁₇ was evaluated through receiver operating characteristics (ROC) curve analysis in the Rutgers sample first, with abnormal CSF t‐Tau/Aβ42 or Aβ42/Aβ40 as the outcome. A threshold optimizing sensitivity and specificity was chosen via Youden's index, and the same numerical threshold was validated separately in NHW and B/AA Emory participants.

To explore additional CSF factors associated with plasma p‐Tau₂₁₇, levels of 1084 analytes in the larger Rutgers biomarker study were first examined for normality and log₁₀‐transformed if not normally distributed. Raw or log₁₀‐transformed values were then analyzed according to principal component analysis (PCA; correlation matrix, Varimax rotation) which derived 102 principal components (PCs). Because of the smaller sample size (n = 44), adjustment for multiple comparison was made by identifying only PCs whose correlations with plasma p‐Tau₂₁₇ had smaller p‐value than CSF p‐Tau₂₁₇.

3. RESULTS

3.1. Plasma p‐Tau₂₁₇ has high intermediate precision within and across days

Plasma p‐Tau₂₁₇ levels were measured in 13 samples in two separate runs within the same day across 3 days using adjacent aliquots (78 measurements total). For assays within the same day, precision profile showed maximum CV < 10% at plasma p‐Tau₂₁₇ > 0.04 pg/mL, and mean CV < 5% at plasma p‐Tau₂₁₇ > 0.24 pg/mL (Figure 1A). For assays done 24 h apart over 3 non‐consecutive days, precision profile showed maximum CV < 10% at p‐Tau₂₁₇ > 0.45 pg/mL, with upper limit of the 95% confidence interval of the mean CV curve falling below 8% at the same threshold (Figure 1B).

FIGURE 1.

Performance and correlates of the automated plasma p‐Tau₂₁₇ assay. Intermediate precision for the assay was determined over six runs (two runs per day for 3 non‐consecutive days) for adjacent aliquots from 13 participants. Intra‐day (A) and inter‐day (B) coefficients of variation (%) are shown in precision profiles. The relationship between CSF p‐Tau₂₁₇ and p‐Tau₁₈₁ – both measured on the automated assays – was marginally better represented by a sigmoidal than a linear function (C), while plasma p‐Tau₂₁₇ levels better correlated with automated CSF p‐Tau assay results than CSF p‐Tau₁₈₁ measured on the Luminex platform (D). In panel D, each data point represents an individual having plasma p‐Tau₂₁₇ measured by Lumipulse, with Rutgers participants having CSF p‐Tau₂₁₇ and/or CSF p‐Tau₁₈₁ measured by Lumipulse, and Emory participants having CSF p‐Tau₁₈₁ measured by Luminex. CSF, cerebrospinal fluid.

3.2. Relationship between CSF p‐Tau₁₈₁, CSF p‐Tau₂₁₇, and plasma p‐Tau₂₁₇

Analysis of CSF p‐Tau₁₈₁ and p‐Tau₂₁₇ in 72 Rutgers participants (same biofluid, different antibodies; representing a range of CSF p‐Tau₁₈₁ concentrations) showed a very strong linear correlation (adj R ² = 0.888, p = 5 × 10⁻²⁹), although a sigmoidal curve was a slightly better fit (adj R ² = 0.912, Figure 1C). Plasma p‐Tau₂₁₇ was not correlated with a measure of BBB permeability (Q‐Alb, p = 0.426). Multivariate analysis showed that controlling for CSF Aβ42 levels modestly improved the overall model (adj R ² = 0.944 with R² change of 0.056, p = 4 × 10⁻¹⁰; VIF < 1.05 for each). Plasma p‐Tau₂₁₇ moderately correlated with CSF p‐Tau₂₁₇ (different biofluids, same antibodies; Figure 1D, adj R ² = 0.690, p = 2.5 × 10⁻¹⁶) and p‐Tau₁₈₁ (different biofluids and antibodies; adj R ² = 0.568, p = 1.5 × 10⁻¹⁵) measured by Lumipulse, suggesting additional factors to influence plasma p‐Tau₂₁₇ levels beyond CSF p‐Tau levels when measured on the same platform. By comparison, plasma p‐Tau₂₁₇ only weakly correlated with CSF p‐Tau₁₈₁ measured by Luminex in the Emory cohort (different biofluids, antibodies, and assay platforms; Figure 1D, adj R ² = 0.255, p = 1.6 × 10⁻⁹).

3.3. Effects of cognitive impairment and race/ethnicity on plasma p‐Tau₂₁₇

Because prior studies have suggested p‐Tau₂₁₇ to associate with presence of AD pathologic changes but also clinical dementia, we next examined if clinical diagnosis (NC, MCI, and dementia) was associated with plasma p‐Tau₂₁₇ levels. This showed greater plasma p‐Tau₂₁₇ levels in participants with AD dementia than those with NC, even after adjusting for CSF p‐Tau₁₈₁ levels (Figure 2A). A regression model incorporating demographic (age, sex, and race/ethnicity), clinical (diagnosis, APOE ε4 carrier status, CRF, and CHF), and biomarker (CSF p‐Tau₁₈₁ and Aβ42, both measured by Lumipulse) variables showed greater plasma p‐Tau₂₁₇ levels to associate with greater CSF p‐Tau₁₈₁ levels and lower CSF Aβ42 levels (each linked to AD neuropathologic changes) as well as clinical AD dementia diagnosis (p = 0.012, Table 2). While neither CRF (n = 6, p = 0.422) nor CHF (n = 1, p = 0.857) influenced plasma p‐Tau₂₁₇ levels (p = 0.422), B/AA race (n = 11; observed power of 0.874) – but not ChA ethnicity (n = 18) – was associated with lower plasma p‐Tau₂₁₇ levels (p = 0.013). This difference in B/AA participants was surprising as we expected the race‐associated difference to be already accounted by CSF p‐Tau₁₈₁ levels in the model. Because of the small number of B/AA participants in the Rutgers cohort, we sought to replicate these findings in a cohort with similar numbers of B/AA and NHW participants.

FIGURE 2.

Effect of clinical diagnosis (NC, MCI, AD dementia, non‐AD dementia) on plasma p‐Tau_217. Differences according to diagnosis persisted after controlling for CSF p‐Tau₁₈₁ levels in the Rutgers (A) and Emory (B) cohorts. AD, Alzheimer's disease; CSF, cerebrospinal fluid; NC, normal cognition; MCI, mild cognitive impairment.

TABLE 2.

Factors influencing plasma p‐Tau₂₁₇ levels in the Rutgers cohort (n = 73 with complete data; CSF Aβ42 Z‐transformed according to NC participants’ levels).

Parameter	B (95% confidence interval)	p‐value
Constant	−2.592 (−3.019, −2.165)	<0.001
log10(CSF p‐Tau181, Lumipulse)	0.966 (0.731, 1.201)	<0.001
Z CSF Aβ42	−0.167 (−0.236, −0.099)	<0.001
Race/Ethnicity
NHW	Reference
ChA	0.018 (−0.119, 0.154)	0.798
Other Asian American	−0.095 (−0.366, 0.176)	0.486
Hispanic	−0.581 (−1.071, −0.091)	0.021
B/AA	−0.216 (−0.385, −0.047)	0.013
Diagnosis
NC	Reference
MCI	0.077 (−0.062, 0.215)	0.272
AD dementia	0.247 (0.056, 0.437)	0.012
Non‐AD dementia	0.020 (−0.188, 0.228)	0.848
Having ≥1 APOE ε4 allele	−0.119 (−0.238, 0)	0.050

Note: Coefficients and confidence intervals are bolded if p < 0.05.

Abbreviations: AD, Alzheimer's disease; B/AA, Black/African American; ChA, Chinese American; CSF, cerebrospinal fluid; MCI, mild cognitive impairment; NC, normal cognition; NHW, Non‐Hispanic White.

3.4. Replicating effects of cognitive impairment and race on plasma p‐Tau₂₁₇

In the Emory cohort, we first validated that a clinical diagnosis of AD dementia (p < 0.001) but also now a diagnosis of MCI (p = 0.002) to associate with greater plasma p‐Tau₂₁₇ levels after controlling for CSF p‐Tau₁₈₁ levels (Figure 2B). When clinical diagnosis, history of CRF, history of CHF, MMSE, APOE genotype, race, and CSF AD biomarkers were entered into a regression model, B/AA participants continued to show lower plasma p‐Tau₂₁₇ levels than NHW participants (observed power of 0.509) beyond the race‐associated difference in CSF p‐Tau₁₈₁ levels. The inclusion of MMSE diminished the impact of clinical diagnosis, suggesting that incorporating a simple screening test may diminish the role of formal diagnosis in interpreting plasma p‐Tau₂₁₇.

Since CSF Aβ42 will not be available in most instances when plasma p‐Tau₂₁₇ is used to predict CSF p‐Tau, our models suggest plasma p‐Tau₂₁₇ to better predict the p‐Tau₁₈₁/Aβ42 ratio than p‐Tau₁₈₁ levels alone. Substituting the ratio's individual components in the model by the ratio did not diminish the impact of race (p = 0.020 and p = 0.010 for B/AA race in the two cohorts, data not shown).

3.5. Race‐related impact of using plasma p‐Tau₂₁₇ to screen for abnormal CSF AD biomarkers

We additionally examined if a plasma p‐Tau₂₁₇ threshold derived in the Rutgers cohort was applicable to the Emory cohort. ROC curve analysis showed plasma p‐Tau₂₁₇ ≥ 0.150 (Youden's index of 0.848) in the Rutgers cohort to identify elevated CSF t‐Tau/Aβ42 with sensitivity of 90.5% and specificity of 94.3% (AUC = 0.940, Figure 3A). Applying this threshold in the Emory cohort resulted in sensitivity of 90.3% and specificity of 81.1% for NHW participants, but only sensitivity of 73.7% and specificity of 72.5% for B/AA participants (Figure 3A). Performance of plasma p‐Tau₂₁₇ to identify decreased CSF Aβ42/Aβ40 was also better in NHW than B/AA participants (Figure 3B). When the two cohorts were analyzed together, plasma p‐Tau₂₁₇ of 0.150 showed slightly better overall accuracy in NHW than B/AA participants (87% vs. 77%, p = 0.145), but a much higher positive predictive value (PPV) in NHW than in B/AA (87% vs. 58%, p = 0.003). Using p‐Tau₂₁₇ to identity decreased Aβ42/Aβ40 gave a similar outcome (85% vs. 65%, p = 0.04, Figure 3B).

FIGURE 3.

Impact of a plasma p‐Tau₂₁₇ threshold developed from a diverse cohort. ROC curve analysis of plasma p‐Tau₂₁₇ for elevated CSF t‐Tau/Aβ42 (A) or reduced Aβ42/Aβ40 (B) in the mixed Rutgers cohort (red line), B/AA Emory participants (dark green line), and NHW Emory participants (blue line; AUC). Filled arrows indicate the optimal p‐Tau₂₁₇ threshold derived in the Rutgers and applied to the NHW Emory participants, and empty arrow indicates the same threshold applied to the B/AA Emory participants. When examined according to diagnosis and core CSF AD biomarker status (t‐Tau/Aβ42 in (C), the same threshold had greater positive predictive value in NHW than B/AA participants. AD, Alzheimer's disease; AUC, area under the curve; B/AA, Black/African American; CSF, cerebrospinal fluid; NHW, non‐Hispanic White; ROC, receiver operating characteristics.

3.6. Other CSF factors influencing plasma p‐Tau₂₁₇ levels

Finally, we explored if other CSF factors influenced plasma p‐Tau₂₁₇ levels beyond core AD biomarkers. Common markers of neurodegeneration (NfL, Ng) and neuroinflammation (sTREM2, sTNFR1, and sTNFR2) did not improve the association model between plasma p‐Tau₂₁₇ and CSF core AD biomarkers in either cohort (Table S1). Among 102 PCs derived from PCA of 1,086 CSF SomaLogic analytes, 5 showed nominal correlation with plasma p‐Tau₂₁₇ at p < 0.038 (correlation between plasma and CSF p‐Tau₂₁₇; see the Methods section). Regression analysis showed plasma p‐Tau₂₁₇ to inversely associate with PC37 and PC7 scores independent of diagnosis, CSF p‐Tau₁₈₁, and CSF Aβ42 (Table S2).

An examination of the two PCs showed synaptotagmin 2, ATP synthase subunit b (ATP5F1B), and ADP ribosylation factor like GTPase 2 binding protein (ARL2BP/BART) as top analytes loading positively onto PC37; and multiple lysosomal proteins (cathepsin A, F, V, D; prolylcarboxypeptidase, alpha galactosidase) as top analytes loading positively onto PC7 (Table S3). Two complement proteins (C4a, C3b) also negatively – albeit modestly – loaded onto PC7. Among these proteins, only C3b levels were available in the Emory cohort. Expanding the regression analysis from Table 3 to include log₁₀(C3b) showed at least some of the additional effect from race to be mediated by introducing a protein from PC7 (p = 0.038, Table S4).

TABLE 3.

Factors asscoiated with plasma p‐Tau₂₁₇ levels in the Emory cohort of older B/AA and NHW participants, with coefficient, 95% confidence interval, and p‐value shown for each term (CSF Aβ42 values were Z‐transformed according to NC participants’ levels).

Parameter	B (95% confidence interval)	p‐value
Constant	1.730 (0.731, 2.729)	<0.001
Diagnosis
NC	Reference
MCI	0.088 (−0.008, 0.183)	0.071
AD dementia	−0.035 (−0.215, 0.145)	0.701
Chronic renal insufficiency	0.682 (0.356, 1.008)	<0.001
Congestive heart failure	0.335 (0.067, 0.603)	0.015
B/AA race	−0.091 (−0.181, −0.001)	0.048
Having ≥ 1 APOE ε4 allele	0.007 (−0.086, 0.100)	0.883
MMSE (square root transformed)	−0.468 (−0.625, −0.310)	<0.001
log10(CSF p‐Tau181, Luminex)	0.640 (0.430, 0.851)	<0.001
Z CSF Aβ42	−0.128 (−0.170, −0.086)	<0.001

Note: Coefficients and confidence intervals are bolded if p < 0.05.

Abbreviations: AD, Alzheimer's disease; APOE, apolipoprotein E; B/AA, Black/African American; CSF, cerebrospinal fluid; MCI, mild cognitive impairment; MMSE, Mini‐Mental State Examination; NC, normal cognition; NHW, non‐Hispanic White.

4. DISCUSSION

Convenient and effective plasma AD biomarkers have the potential of greatly advancing ready access to early and accurate diagnosis, but nuances can be overlooked in large studies with mixed cohort compositions, enrollment sources, and dementia severity. We showed in one diverse cohort reflecting one regional Northeastern US population a stronger correlation between two related biomarkers (p‐Tau₂₁₇ and p‐Tau₁₈₁) measured in the same CSF compartment than the same biomarker (p‐Tau₂₁₇) measured in two related biofluids (CSF and plasma), which could not be explained by intermediate precision or dynamic range of the assays alone. We further linked greater cognitive impairment to higher but B/AA race to lower plasma p‐Tau₂₁₇, and replicated these as well as one novel finding related to CSF non‐amyloid/non‐tau biomarkers in an independent cohort of B/AA and NHW participants. Importantly, a plasma p‐Tau₂₁₇ threshold optimized in a cohort with limited number of B/AA participants (11 out of 98) showed 50% higher PPV in NHW than in B/AA participants. We discuss these findings below.

The origins of plasma AD biomarkers remain unclear, with potential input from the brain via glymphatic clearance, extracellular vesicles, receptor‐mediated transport across intact BBB, gross BBB disruption, or a combination of these processes. While a clear understanding of the mechanism giving rise to biomarker changes is not necessary for its validation or clinical application if said biomarker's levels are not readily influenced by common clinical variables, the low signal‐to‐noise ratio of plasma AD biomarkers (including p‐Tau₂₁₇, p‐Tau₁₈₁, and Aβ42/Aβ40) demands greater attention to demographic and clinical factors which predispose to over‐ or under‐detection. Because we were able to confirm the excellent correlation between CSF p‐Tau₂₁₇ and CSF p‐Tau₁₈₁ using automated assays (one of the first reports to do so), previous reports of better predictive value of p‐Tau₂₁₇ over p‐Tau₁₈₁ ²⁶ , ²⁷ likely resulted from greater CV using non‐automated assays especially given their low concentrations. This excellent correlation also challenges proposed biological differences for AD‐related phosphorylation at threonine 181 and 217 derived from assay‐based observations.

This is not the first report to identify common factors influencing measured p‐Tau levels and thus performance as a screening tool for AD pathology or AD risks ¹¹ , ¹⁴ , ¹⁵ , ²⁷ , but significant influences from severity of cognitive impairment on plasma p‐Tau₂₁₇ in two separate cohorts suggests comparisons between the usual participants with and without dementia may over‐estimate the value of plasma p‐Tau₂₁₇ as a fluid biomarker. The conflation of severity markers (often non‐specific across neurodegenerative disease types) with etiologic biomarkers is not new in AD, but is increasingly overlooked in cohort studies. Whereas neurodegeneration is often cited post hoc to account for this performance inflation (i.e., more AD neuropathology leads to more neurodegeneration), we did not find their typical markers (NfL, Ng) to influence plasma p‐Tau₂₁₇ levels. Because we cannot consistently account for the accuracy of clinical cognitive assessment in diverse populations, appropriate determination of novel biomarkers’ performance against core AD biomarkers should be ideally carried out in people having similar degrees of cognitive impairment (e.g., MCI) from different etiologies (AD vs. suspected non‐amyloid pathology) to eliminate the confounds from diagnostic mislabels.

The impact of B/AA race on plasma p‐Tau₂₁₇ again emphasizes the importance of enrolling diverse participants in biomarker studies. One previous study found plasma p‐Tau₂₁₇ to potentially have greater race‐associated differences than plasma p‐Tau₁₈₁, ²⁷ but did not formally assess its predictive accuracy due to the small number of B/AA adults with autopsy‐ or amyloid PET‐confirmed AD (n = 9). The more recent Bio‐Hermes study including 103 older B/AA adults (27 amyloid PET positive) found them to have significantly lower plasma p‐Tau₁₈₁ and p‐Tau₂₁₇ levels than older NHW adults. ¹¹ However, only 11 out of 1001 (1%) Bio‐Hermes participants had paired CSF p‐Tau measurements. Because of the consistently reported difference in CSF p‐Tau₁₈₁ levels between B/AA and NHW older adults ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²⁸ , omission of CSF p‐Tau₁₈₁ leaves open the possibility that lower plasma p‐Tau₂₁₇ merely corresponds to the lower CSF p‐Tau levels. Across the two cohorts presented here, we analyzed plasma p‐Tau₂₁₇ in 69 older B/AA participants (40 or 58% with low CSF Aβ42, 100% with paired CSF p‐Tau₁₈₁), confirming plasma p‐Tau₂₁₇ as a fluid marker with racial disparities beyond a carry‐over from race‐associated CSF tau‐related biomarker disparities ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²⁸ , as inclusion of CSF p‐Tau₁₈₁ was insufficient to eliminate the effect of race on plasma p‐Tau₂₁₇. This is regardless of whether CSF p‐Tau₁₈₁ was measured by bead‐based assays in a semi‐automated (Luminex, Emory cohort) or fully‐automated (Lumipulse, Rutgers cohort) format, although p‐Tau₁₈₁ levels measured by the two platforms have a linear relationship. ²⁹ In contrast, we ²⁵ and others ²⁹ have reported a non‐linear relationship in CSF Aβ42 levels measured by the two platforms at high concentrations.

Notwithstanding the mounting evidence, researchers have so far not identified a solution to address CSF tau‐related racial disparities 7 years after our initial report. ¹⁸ Here we found that differences in some CSF proteins (e.g, cathepsins, C3b) may begin to elucidate mechanisms by which tau‐related protein levels differ according to race. From the perspective of hypothesis generation, one putative mechanism has to do with lysosomes’ role in tau processing. Because tau peptides undergo C‐terminal truncation in the lysosome before secretion into the CSF and interstitial fluid (ISF) ³⁰ , ³¹ , ³² , ³³ , the inverse relationship between secreted lysosomal cathepsin proteins (PC7) and p‐Tau₂₁₇ could translate into greater lysosomal dysfunction in B/AA participants. This can result from social determinants of health such as environmental toxic exposure (including mercury ³⁴ and air pollution ³⁵ ), or an upstream mechanism hypothesized to underlie other B/AA health disparities such as APOL1‐associated kidney disease (associated with leaky lysosomes). ³⁶ , ³⁷ , ³⁸ , ³⁹ For the latter hypothesis, complications of endothelial disease and lower tau secretion would each correlate with lysosomal dysfunction without diminishing the role of AD neuropathologic changes in B/AA brain health. ⁴⁰ , ⁴¹ Similarly, synaptotagmin 2 (top loading analyte for PC37) negatively regulates lysosomal exocytosis ⁴² , ⁴³ , ⁴⁴ beyond its better‐known function in synaptic vesicle release. These hypotheses should be prospectively tested in cohorts with sufficient B/AA participants taking into account APOL1 variants, plasma lysosomal proteins, and plasma p‐Tau₂₁₇. Plasma lysosomal protein levels – which already predict renal functions without the need for race‐based adjustments ⁴⁵ – should also be prospectively tested to adjust for race‐based differences in plasma p‐Tau₂₁₇ levels.

While we present a study using automated assays with high intermediate precision in two diverse cohorts, this study has several limitations. Not all ChA participants had CSF for paired analysis, and we did not have a sufficient number of older Hispanic or non‐Chinese Asian participants to generalize our findings to these groups. None of the participants had amyloid or tau PET, even though core CSF AD biomarkers have consistently shown high correspondence with amyloid PET. We also could not assess the longitudinal implication for plasma p‐Tau₂₁₇ differences in B/AA participants due to the cross‐sectional nature of these studies, although future longitudinal studies might be challenging with the increasing use of anti‐amyloid therapies. The two cohorts shared many clinical and biomarker assessments, but the more extensive aptamer‐based proteomic analysis was only available in a subset of Rutgers participants and MMSE was only available in Emory participants. Finally, we did not have sufficient matching information to adequately account for the influence on plasma p‐Tau₂₁₇ from social determinants of health, but these findings – if further replicated – can be tested in subsequently recruited cohorts having more detailed social characterization.

In conclusion, plasma p‐Tau₂₁₇ is a potentially powerful clinico‐biological marker which can greatly accelerate the usual AD diagnostic process. At the same time, measuring p‐Tau species in blood cannot circumvent the disparities long known to exist in CSF t‐Tau and p‐Tau₁₈₁ measurements, and the race‐associated effect on p‐Tau levels may be further compounded in plasma. Extreme caution thus should be exercised when applying plasma p‐Tau₂₁₇ to B/AA adults for the purpose of clinical or trial screening, and an informed consent process to minimize mental and financial distress may be necessary to mitigate such biomarker‐related disparities.

CONFLICT OF INTEREST STATEMENT

W.T.H. has received research support from NIH, Robert Wood Johnson Foundation, Fujirebio Diagnostics Inc, Atlanta Family Foundation, and TMCity Foundation; has consulted for Apellis Pharmaceuticals, Beckman Coulter Diagnostics, Biogen, Fujirebio Diagnostics Inc; has patents on CSF‐based diagnosis of FTLD‐TDP, prognosis of MCI due to AD, and prognosis of spinal muscular atrophy on gene therapy; and has copyright on Mandarin‐based cognitive assessments (licensed to Linus Health). H.G.J. has received research support from Glenn Foundation for Medical Research and American Federation for Aging Research. A.N. has nothing to disclose. M.L. has nothing to disclose. V.S. has nothing to disclose. I.H. has nothing to disclose. M.K. has nothing to disclose. M.J.M. has nothing to disclose. M.M. has nothing to disclose. K.S. has nothing to disclose. K.J.K. has nothing to disclose. Author disclosures are available in the Supporting Information.

CONSENT STATEMENT

Written informed consents were previously obtained from all participants in accordance with the Declaration of Helsinki and the Belmont report for the long‐term storage and future analysis of samples.

Supporting information

Supporting Information

Supporting Information

Nayyar A, Li M, Sotelo V, et al. Influence of cognitive impairment and race on plasma p‐Tau₂₁₇ in two diverse cohorts. Alzheimer's Dement. 2025;21:e14585. 10.1002/alz.14585