Assessment of Plasma Total Tau Level as a Predictive Biomarker for Dementia and Related Endophenotypes

Key Points

Question  Can plasma total tau be used as a biomarker for dementia and related endophenotypes?

Finding  In this cohort study of samples from 1453 participants in the Framingham Heart Study and 367 individuals in the Memento study, plasma total tau was associated with endophenotypes of dementia and with the risk of incident clinical Alzheimer disease dementia.

Meaning  Measuring plasma total tau may help with risk stratification and subsequent enrollment of high-risk individuals in dementia prevention trials.

Importance  Blood-based biomarkers have the potential to improve the identification of persons with the greatest dementia risk for inclusion in dementia prevention trials through low-cost and minimally invasive screening.

Objective  To investigate the use of plasma total tau as a blood biomarker for dementia and related endophenotypes.

Design, Setting, and Participants  This prospective cohort study used data from the US community-based Framingham Heart Study with replication in the Memento study, a multicenter cohort of persons with mild cognitive impairment or subjective cognitive complaints recruited from memory clinics across France. Total tau levels were measured from stored plasma samples in Framingham Heart Study participants during 2004 to 2011. Dementia follow-up occurred across a median of 6 years (interquartile range, 5-8 years) for persons 65 years and older who were dementia free at baseline. Plasma and/or cerebrospinal fluid samples were obtained from Memento study participants from April 19, 2011, to June 22, 2016. Dementia follow-up took place over a median of 4 years (interquartile range, 3-5 years). Data analysis was performed from January to November 2018.

Exposures  Plasma total tau level measured using single-molecule array technology.

Main Outcomes and Measures  Incidence of dementia of any cause (all dementia) and dementia due to clinical Alzheimer disease (AD dementia).

Results  Among the 1453 participants in the Framingham dementia study sample, the mean (SD) age was 75 (7) years; 792 (54.5%) were female. Among the 367 individuals in the replication cohort, the mean (SD) age was 69 (9) years; 217 (59.1%) were female. Of 134 cases of incident all dementia in the Framingham sample, 105 were AD dementia. After adjustment for age and sex, each SD unit increase in the log of plasma total tau level was associated with a 35% increase in AD dementia risk (hazard ratio [HR], 1.35; 95% CI, 1.10-1.67). The addition of plasma total tau to a model including age and sex improved the stratification of participants for risk of AD dementia (net reclassification improvement, 0.382; 95% CI, 0.030-0.716). Higher plasma total tau level was associated with poorer cognition across 7 cognitive tasks (P < .05) and smaller hippocampi (hippocampal volume: β [SE] = 0.002 [0.001]; P = .003) as well as neurofibrillary tangles (β [SE] = 0.95 [0.45]; P = .04) and microinfarcts (odds ratio, 3.04; 95% CI, 1.26-7.37) at autopsy. In the replication cohort, plasma total tau level weakly correlated with cerebrospinal fluid total tau level (Spearman correlation coefficient, 0.16; P = .07), but plasma total tau was at least as strongly associated with incident AD dementia as cerebrospinal fluid total tau (log plasma total tau: HR, 2.33; 95% CI, 1.00-5.48; log cerebrospinal fluid total tau: HR, 2.14; 95% CI, 1.33-3.44) after adjustment for age and sex.

Conclusions and Relevance  The findings suggest that plasma total tau levels may improve the prediction of future dementia, are associated with dementia endophenotypes, and may be used as a biomarker for risk stratification in dementia prevention trials.

The discovery of minimally invasive and cost-effective blood-based biomarkers for dementia has the potential to transform clinical research and practice by permitting widespread low-cost screening, risk stratification, and efficient identification of persons with the greatest dementia risk for inclusion in dementia prevention trials. Single-molecule array technology permits the detection of total tau (t-tau) from plasma with a greater than 1000-fold improvement in sensitivity over conventional enzyme-linked immunosorbent assay (ELISA).^1,2 Plasma t-tau level measured with this technology is greater in persons with Alzheimer disease (AD) dementia or mild cognitive impairment (MCI) compared with control participants.^3-5 However, the usefulness of plasma t-tau as a diagnostic biomarker is limited by the large overlap in values observed between diagnostic groups, which may signal a lack of specificity for AD. The latest National Institute on Aging–Alzheimer Association Research Framework (NIA-AA) describes t-tau as a biomarker of neuronal injury rather than of neurofibrillary tangle burden, noting that an elevated t-tau level reflects neuronal injury from various causes.⁶ Nevertheless, plasma t-tau shows early promise as a predictive biomarker for dementia with 2 longitudinal studies demonstrating that plasma t-tau is associated with endophenotypes of dementia, including cognitive decline.^7,8 However, evidence from representative community-based studies is limited, and studies to date have not been powered to examine the association with risk for dementia. Further studies are required to evaluate plasma t-tau as a predictive biomarker for incident dementia, permitting risk stratification in the setting where blood-based biomarkers would be most useful—in the general community and the age groups of relevance to early dementia prevention trials. Accordingly, we examined the association of plasma t-tau level as a predictive biomarker for dementia of any cause (all dementia) and incident dementia of the Alzheimer type (AD dementia) in the large community-based Framingham Heart Study (FHS), with replication performed in an independent multicenter cohort from France. We also examined endophenotypes associated with dementia, including cognitive function, hippocampal volume, and tau neurofibrillary tangle burden.

The FHS is a community-based, prospective study spanning 3 generations of participants from Massachusetts. It began in 1948 with the recruitment of the original cohort of 5209, which has been reexamined once every 2 years.⁹ The FHS offspring cohort commenced in 1971 with the enrollment of 5124 participants who were offspring of members of the original cohort or spouses of these offspring.¹⁰ The offspring cohort has had 9 quadrennial examinations. In 2002, 4095 grandchildren of the original cohort were enrolled into a third-generation (Gen3) cohort, now studied across 3 examination cycles.¹¹ All participants provided written informed consent. The institutional review board at the Boston University Medical Center approved the study protocols and consent forms.

Blood samples were obtained following an overnight fast at examination cycle 28 for the original cohort (2004-2005), cycle 8 for the offspring cohort (2005-2008), and cycle 2 for the Gen3 cohort (2008-2011) (N = 6471). Samples were immediately centrifuged, aliquoted, and stored at –80°C. Plasma samples were analyzed from February to March 2017 using a Simoa Tau 2.0 Kit and a Simoa HD-1 analyzer (Quanterix). This assay is a single-molecule ELISA (digital ELISA) validated as fit-for-purpose research use only. The limit of detection is 0.019 pg/mL. The assay uses a set of monoclonal antibodies reacting to both normal and phosphorylated tau and can detect all tau isoforms. Samples sent for analysis had never been thawed (see eMethods in the Supplement for full assay methods). The analytical range was between 0.06 and 360 pg/mL. The intra-assay coefficient of variation was 4.1%, and the interassay coefficient of variation was 7.5%. As an additional quality control, we included 292 phantom samples, which were duplicates but were marked with dummy participant identification codes. These further confirmed assay precision except in a subsample (471 of 6417) with suboptimal correspondence between phantom and original samples across 6 consecutive days of running the assays. These samples, all from Gen3 participants, were excluded from further analysis. Comparison of Gen3 participants who were included and excluded based on t-tau assay quality is provided in eTable 1 in the Supplement.

Framingham Heart Study participants are under continual surveillance for incident dementia through routine cognitive screening and comprehensive monitoring.¹² Full methods for dementia surveillance and flagging of suspected cognitive impairment are described in the eMethods in the Supplement. A review committee comprising a neurologist and neuropsychologist adjudicated dementia diagnosis in accordance with the Diagnostic and Statistical Manual of Mental Disorders (Fourth Edition).¹³ A diagnosis of AD dementia was based on the criteria of the National Institute of Neurological and Communicative Disorders and Stroke and the AD and Related Disorders Association for definite, probable, or possible AD.¹⁴

Clinical neuropsychologists and trained research assistants administered validated neuropsychological tests.^15-17 Cognitive outcomes included tests of verbal (episodic) memory, visual memory, verbal reasoning, processing speed, executive function, visuospatial integration, and estimated premorbid intellectual function. Hippocampal volume was calculated from brain magnetic resonance images using a semiautomatic multiatlas segmentation algorithm¹⁸ and expressed as a percentage of intracranial volume. The density of neurofibrillary tangles in the medial temporal lobe, the earliest site of tau accumulation in AD,^19,20 was rated in a semiquantitative fashion using Bielschowsky silver–stained sections. The presence of microinfarcts and senile plaques were also rated by a neuropathologist. The eMethods in the Supplement provide further details of these procedures.

We leveraged FHS data to create 3 analysis samples (Figure 1). First, we examined the associations between plasma t-tau and the risk of incident dementia in 1453 participants from the original and offspring study cohorts aged at least 65 years. Follow-up for dementia was from the baseline examination to the time of incident event up to a maximum of 10 years (median, 6 years [interquartile range, 5-8 years]) through 2016. For persons with no incident events, follow-up was censored at the time of death or the date the participant was last known to be dementia free, also through 2016.

Second, in 3832 persons aged 25 to 98 years from all 3 FHS cohorts, we examined the cross-sectional associations between plasma t-tau and domains of cognitive function, as well as hippocampal volume in 3238 persons with magnetic resonance imaging of the brain.

Third, in a subsample of 42 FHS brain donors, we investigated the association between plasma t-tau and the burden of neurofibrillary tangles, microinfarcts, and amyloid plaques.

Plasma t-tau levels were log transformed to normalize their distributions and then standardized within the sample. Cox proportional hazards regression models were implemented to estimate the associations between plasma t-tau and incident dementia, adjusting for age and sex. Results were expressed as hazard ratios (HRs) and 95% CIs. We confirmed that the assumption of proportionality of hazards was met. Models with age at dementia diagnosis or censoring used as the time scale and adjusting only for sex were found to yield nearly identical results. We assessed the incremental value of adding plasma t-tau level to a model involving age and sex by comparing cumulative hazards in those greater and less than the plasma t-tau median level and by calculating the integrated discrimination improvement and continuous net reclassification improvement (NRI) applicable to survival data, which considers events, nonevents, and participants who were censored.^21,22 The 95% CI was calculated using bootstrap.²³ The continuous NRI shares the same properties as the categorical NRI with the exception that the continuous NRI quantifies upward and downward movement as any change in predicted probabilities.²² The continuous version of the NRI is most appropriate when no established risk categories exist.²² Higher values indicate superior discrimination. Using the event rate observed in the sample, we estimated the sample size needed for both a 5- and 10-year dementia prevention trial to detect a 25% lowering of the dementia event rate with 80% power and with 2-tailed α set at .05. We then examined how the sample size estimates changed when using plasma t-tau level and APOE ε4 presence to risk stratify for trial inclusion.

The associations between plasma t-tau and dementia endophenotypes were examined using linear regression, adjusting for age, sex, time between the blood sample obtainment and outcome assessment, education for the cognitive outcomes, and age squared for hippocampal volumes (association with age was nonlinear).²⁴ For analyses of incident dementia, cognitive function, and hippocampal volume, a second model included additional adjustments for covariates previously associated with these outcomes: systolic blood pressure, use of antihypertensive medication, prevalent diabetes, prevalent cardiovascular disease and stroke, high-density lipoprotein cholesterol level, body mass index, and positivity for an APOE ε4 allele.^25,26

To strengthen the validity of our findings, we replicated the association between plasma t-tau and incident dementia in an independent sample and compared t-tau level concordance between plasma and cerebrospinal fluid (CSF). Our replication sample was the Memento study, a longitudinal, multicenter clinical cohort of 2323 dementia-free outpatients with either subjective cognitive complaints or objective MCI.²⁷ Plasma and CSF samples were obtained from April 19, 2011 to June 22, 2016. A subsample of Memento participants (Memento-CSF subcohort) underwent lumbar puncture. There were 367 participants with both plasma and CSF samples available, 140 of whom had samples of both biofluids obtained on the same day. The Memento cohort used the same Quanterix assay to quantify t-tau in plasma and in CSF and the same diagnostic criteria for all dementia and AD dementia as in FHS, and all dementia cases were reviewed by an independent adjudication committee (eMethods in the Supplement).

Analyses were performed using SAS software, version 9.4 (SAS Institute). Missing data were excluded from analysis. Results with 2-tailed P < .05 were considered significant.

Cohort characteristics are displayed in Table 1. Descriptive statistics for all other analysis samples are shown in eTables 2 through 5 in the Supplement. During follow-up, we observed 134 of 1453 (9.2%) cases of incident dementia; 105 were due to possible (12), probable (87), or definite (6) AD dementia.

The association between plasma t-tau and age was quadratic across the lifespan but linear in the older sample studied for incident dementia (eFigure 1 and eTable 6 in the Supplement). Higher plasma t-tau levels were associated with known AD dementia risk factors, including female sex, lower educational attainment, and a higher vascular risk factor burden (eTable 7 in the Supplement). Plasma t-tau levels did not differ by APOE ε4 status. Plasma t-tau levels by incident dementia status as well as scatterplots of plasma t-tau against each endophenotype are given in eFigures 2-8 in the Supplement.

After adjustment for age and sex, each SD unit increase in the log of plasma t-tau was associated with a 29% greater risk of incident all-cause dementia (HR, 1.29; 95% CI, 1.07-1.55; P = .007) and a 35% increase in the risk of incident AD dementia (HR, 1.35; 95% CI, 1.10-1.67; P = .004). Results were unchanged after additional adjustment for the presence of an APOE ε4 allele and vascular risk factors (all dementia: HR, 1.31; 95% CI, 1.08-1.57; P = .005; AD dementia: HR, 1.38; 95% CI, 1.12-1.71; P = .003).

Persons with plasma t-tau levels greater than the median had a 62% greater risk of incident all dementia (HR, 1.62; 95% CI, 1.10-2.37) and a 76% greater risk of AD dementia (HR, 1.76; 95% CI, 1.13 to 2.74) after accounting for age and sex (Table 2 and Figure 2). Plasma t-tau level improved risk discrimination for all dementia and AD dementia beyond age and sex. Results were comparable when limiting the outcome to incident probable or definite AD dementia (eTable 8 in the Supplement). Given that the APOE ε4 allele is currently used to power AD dementia prevention trials, we then stratified results by the presence of an APOE ε4 allele. Plasma t-tau level was associated with improved risk discrimination for all dementia and AD dementia in both APOE ε4 carriers and noncarriers. Using plasma t-tau level greater than the median to select participants for inclusion in a 5-year prevention trial reduced the estimated sample size by 38% (from 8756 to 5460) for the outcome of all dementia and by 50% (from 14 618 to 7292) for AD dementia. Selecting participants who were APOE ε4 carriers and had plasma t-tau levels greater than the median reduced the required sample size by 69% (from 8756 to 2712) for all dementia and by 80% (from 14 618 to 2896) for AD dementia (more details are given in eTable 9 in the Supplement).

Higher levels of plasma t-tau were associated with poorer performance across all cognitive outcomes and domains tested (model 1: β ranged from −0.18 to −0.010; model 2: β ranged from −0.17 to −0.009) except the Wide-Range Achievement Test reading subtest (model 1: β [SE], −0.009 [0.010], P = .37; model 2: β [SE], −0.001 [0.010], P = .95), a measure that is commonly used to estimate premorbid function¹⁵ and therefore not expected to be associated with plasma t-tau (Table 3). Higher plasma t-tau levels were also associated with smaller hippocampal volumes (model 1: β [SE], −0.002 [0.001], P = .003; model 2: β [SE], −0.003 [0.001], P = .001).

The mean (SD) age at death was 82 (9) years (21 of 42 [50.0%] male), and 11 had confirmed AD. Each SD unit increase in the log of plasma t-tau was associated with a higher burden of neurofibrillary tangles in the medial temporal lobe (β [SE], 0.95 [0.45]; P = .04) and a higher burden of microinfarcts (odds ratio [OR], 3.04; 95% CI, 1.26-7.37), but not cortical neuritic (OR, 1.42; 95% CI, 0.59-3.39) or diffuse (OR, 1.20; 95% CI, 0.50-2.88) plaque burden (P > .05).

Over a median follow-up of 3.97 years (interquartile range, 3.02-4.66 years) in the Memento-CSF subcohort (mean [SD] age, 69 [9] years; 217 of 367 [59.1%] women), there were 76 cases of incident all dementia (55 were probable AD). Of these persons, all but 4 had MCI at baseline. In the largest sample with plasma t-tau and after adjustments for age and sex, each SD unit increase in the log of plasma t-tau levels was associated with a nonsignificant 14% greater risk of incident all-cause dementia (HR, 1.14; 95% CI, 0.86-1.51; P = .35) and a significant 54% increase in the risk of incident AD dementia (HR, 1.54; 95% CI, 1.04-2.28; P = .03).

The Spearman correlation coefficient between the log of plasma and CSF t-tau was 0.16 (P = .07) in 140 participants with plasma and CSF samples obtained on the same day. In this sample and when including both plasma and CSF t-tau in a model involving age and sex, each SD unit increase in the log of plasma t-tau was associated with an HR of 2.33 (95% CI, 1.00-5.48) for incident AD dementia, whereas each SD unit increase in the log of CSF t-tau was associated with an HR of 2.14 (95% CI, 1.33-3.44).

In our large population-based study, plasma t-tau level was associated with the risk of incident AD dementia, a finding that we replicated in an independent cohort. Although plasma t-tau did not correlate significantly with corresponding values in CSF, plasma t-tau was at least as strong a predictive biomarker for incident AD dementia as CSF t-tau in the Memento CSF subcohort. Higher plasma t-tau level was also associated with improved dementia risk stratification and with key endophenotypes of dementia, including poorer cognitive performance and smaller hippocampal volumes. Although plasma t-tau levels did not parallel those in CSF, plasma t-tau levels improved risk stratification for AD dementia.

Two other studies^7,8 using the same plasma t-tau assay recently demonstrated that higher plasma t-tau was associated with cognitive decline, hippocampal atrophy, and decreased cortical glucose metabolism across short follow-up periods. We extend earlier findings to suggest that plasma t-tau level is associated with the risk of incident dementia and improved dementia risk stratification over age and sex. In our study, plasma t-tau was associated with endophenotypes of dementia in a larger sample and at younger ages than previously described.

Tau is primarily expressed in central nervous system neurons,²⁸ with plasma levels thought to reflect neuronal damage and subsequent drainage of tau from the brain parenchyma to the CSF and blood.^8,29 Thus, elevated t-tau level in plasma may reflect neuronal damage from numerous sources. The newly revised NIA-AA Research Framework describes CSF t-tau as a nonspecific marker of neuronal injury.⁶ Similarly, higher plasma t-tau levels have been shown to correlate with lower gray matter densities across multiple brain regions in the Alzheimer Disease Neuroimaging Initiative cohort, suggesting a lack of specificity.³⁰ Moreover, plasma t-tau was weakly correlated with CSF t-tau in our study. This finding is consistent with previous studies showing that the associations of plasma t-tau with CSF t-tau have been weak or nonexistent.^8,31 In our study, plasma t-tau was associated with medial temporal lobe neurofibrillary tangle burden and the presence of microinfarcts at autopsy. Moreover, greater plasma tau level was associated with poorer performance across multiple cognitive domains and not only those typically impaired in AD. These results support the suggestion that plasma t-tau may lack specificity for neurofibrillary tangle burden.

Our findings suggest that plasma biomarkers do not need to parallel their CSF equivalent to be useful. In our study, plasma t-tau was associated with a greater risk of incident AD dementia and with improved dementia risk stratification. Despite a weak correlation between plasma and CSF t-tau, plasma t-tau was at least as strongly associated with the development of incident AD dementia. Although CSF t-tau had greater precision, this may be offset by the invasiveness of CSF sampling. Thus, regardless of its source, our results suggest that plasma t-tau is a useful biomarker for assessing the risk of AD dementia.

The prevention of dementia has become a major health priority, with clinical trials seeking to modify the early stages of AD.^32,33 The feasibility of AD dementia prevention trials depends on the ability to select individuals at high risk of developing dementia.³⁴ Whereas we do not expect plasma t-tau cutoffs to enhance diagnostic certainty for any single patient, our results suggest that plasma t-tau could be associated with improved risk stratification at a population level, targeting persons for inclusion in prevention trials, thus improving the power and precision of clinical trials and potentially accelerating therapeutic pipelines and drug discovery. Use of plasma t-tau in this manner could be likened to the measurement of the APOE ε4 allele, which is not a biomarker of AD pathology providing diagnostic certainty for AD dementia but is still routinely used to power clinical trials by selecting at-risk individuals.^33,34

Limitations include our predominantly white sample, which limits generalizability to other racial/ethnic groups. Subtyping for AD dementia was mostly based on clinical diagnosis without neuropathological confirmation. Further replication of our results in cohorts with differing population characteristics is needed to ascertain the plasma t-tau thresholds most associated with future dementia.

Whereas plasma t-tau may lack diagnostic specificity for AD, we show that plasma t-tau may be a useful albeit nonspecific predictive biomarker that improves risk stratification for dementia. Our findings were derived from a community setting, where plasma t-tau testing may help with risk stratification and the enrollment of high-risk individuals into dementia prevention trials.

Accepted for Publication: December 5, 2018.

Corresponding Author: Matthew P. Pase, PhD, Melbourne Dementia Research Centre, The Florey Institute for Neuroscience and Mental Health. 30 Royal Parade, Parkville, Victoria 3052, Australia (matthewpase@gmail.com).

Published Online: March 4, 2019. doi:10.1001/jamaneurol.2018.4666

Author Contributions: Drs Pase and Beiser are co–first authors. Drs Dufouil and Seshadri are co–senior authors. Dr Seshadri had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: Pase, Beiser, Satizabal, Dufouil, Seshadri.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Pase, Seshadri.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Beiser, Himali, Dufouil.

Obtained funding: Pase, DeCarli, Dufouil, Seshadri.

Administrative, technical, or material support: Satizabal, DeCarli, Dufouil.

Supervision: Beiser, Seshadri.

Conflict of Interest Disclosures: Dr Decarli reported being a consultant to Novartis on a clinical trial of LCZ696 for heart failure. No other disclosures were reported.

Funding/Support: Dr Pase is funded by a National Heart Foundation of Australia Future Leader Fellowship (ID 102052) as well as funding from the National Health and Medical Research Council APP1158384. The Framingham Heart Study is supported by contracts N01-HC-25195 and HHSN268201500001I from the National Heart, Lung and Blood Institute; grants AG054076, AG033193, AG033040, AG049505, AG049607, AG052409, and AG059421 from the National Institute on Aging; and grants NS017950 and UH2 NS100605 from the National Institute of Neurological Disorders and Stroke. Dr DeCarli directs the UC Davis Alzheimer’s Disease Center with funding from grant P30 AG010182 from the National Institutes of Health. The MEMENTO cohort was sponsored by the Fondation Plan Alzheimer 2008-2012 and the Plan Maladies NeuroDégénératives 2014-2019. This work was also supported by Bordeaux University Hospital, Inserm, and the University of Bordeaux.

Role of the Funder/Sponsor: The funders had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Additional Contributions: Patrice Sutherland, BS, Framingham Heart Study, and the Framingham Heart Study laboratory staff assisted in organizing and overseeing the analysis of plasma total tau. Isabelle Pellegrin, MD, PhD, Bordeaux Hospital, and the Bordeaux hospital laboratory staff helped organize and oversee the analysis of t-tau in plasma and cerebrospinal fluid samples. All were employees of the institutions stated and received no further compensation for their work on this article.