Plasma phosphorylated tau₂₁₇ strongly associates with memory deficits in the Alzheimer’s disease spectrum

Abstract

Plasma phosphorylated tau (p-tau) biomarkers open unprecedented opportunities for identifying carriers of Alzheimer’s disease pathophysiology in early disease stages using minimally invasive techniques. Plasma p-tau biomarkers are believed to reflect tau phosphorylation and secretion. However, it remains unclear to what extent the magnitude of plasma p-tau abnormalities reflects neuronal network disturbance in the form of cognitive impairment.

To address this question, we included 103 cognitively unimpaired elderly and 40 cognitively impaired, amyloid-β-positive individuals from the TRIAD cohort, in addition to 336 cognitively unimpaired and 216 cognitively impaired, amyloid-β-positive older adults from the BioFINDER-2 cohort. Participants had tau PET scans, amyloid PET scans or amyloid CSF, p-tau₂₁₇, p-tau₁₈₁ and p-tau₂₃₁ blood measures, structural T₁-MRI and cognitive assessments. In this cross-sectional study, we used regression models and correlation analyses to assess the relationship between plasma biomarkers and cognitive scores. Furthermore, we applied receiver operating characteristic curves to assess cognitive impairment across plasma biomarkers. Finally, we categorized participants into amyloid (A), p-tau (T₁) and tau PET (T₂) positive (+) or negative (−) profiles and ran non-parametric comparisons to assess differences across cognitive domains.

We found that plasma p-tau₂₁₇ was more associated with cognitive performance than p-tau₁₈₁ and p-tau₂₃₁ and that this relationship was particularly strong for memory scores (TRIAD: β_p-tau217 = −0.53, β_p-tau181 = −0.35 and β_p-tau231 = −0.24; BioFINDER-2: β_p-tau217 = −0.52, β_p-tau181 = −0.24 and β_p-tau231 = −0.29). Associations in amyloid-β-positive participants resembled these results, but other cognitive scores also showed strong associations in cognitively impaired individuals. Moreover, plasma p-tau₂₁₇ outperformed plasma p-tau₁₈₁ and plasma p-tau₂₃₁ in identifying memory impairment (area under the curve values for TRIAD: p-tau217 = 0.86, p-tau181 = 0.77 and p-tau231 = 0.75; and for BioFINDER-2: p-tau₂₁₇ = 0.86, p-tau₁₈₁ = 0.76 and p-tau₂₃₁ = 0.81) and in identifying executive function impairment only in the BioFINDER-2 cohort (p-tau₂₁₇ = 0.82, p-tau₁₈₁ = 0.76 and p-tau231 = 0.76). Lastly, we showed that subtle memory deficits were present in A+T₁+T₂− participants for plasma p-tau₂₁₇ (P = 0.007) and plasma p-tau₁₈₁ (P = 0.01) in the TRIAD cohort and for all biomarkers across cognitive domains in A+T₁+T₂− and A+T₁+T₂− individuals (P < 0.001 in all) in the BioFINDER-2 cohort. The A+T₁+T₂− individuals showed cognitive deficits in both cohorts (P < 0.001 in all).

Together, our results suggest that plasma p-tau₂₁₇ stands out as a biomarker capable of identifying memory deficits attributable to Alzheimer’s disease and that memory impairment certainly occurs in amyloid-β- and plasma p-tau-positive individuals who have no significant amounts of tau in the neocortex.

Fernández Arias et al. investigate the relationship between plasma biomarkers and cognition in Alzheimer's disease. They show that plasma p-tau₂₁₇ concentrations are more strongly associated with cognitive performance than other plasma p-tau biomarkers, and outperform the other biomarkers in predicting memory impairment.

Introduction

Alzheimer’s disease (AD) is defined by the accumulation of extracellular amyloid-β (Aβ) plaques and intracellular neurofibrillary tangles¹ in the brain. Aβ is understood to initiate a cascade of events that eventually lead to dementia, where tau hyperphosphorylation and aggregation as neurofibrillary tangles in the brain and cognitive decline are present.² Neurofibrillary tangles consist of insoluble fibrillary deposits of hyperphosphorylated tau (p-tau),³ but tau pathology can also be detected early in CSF and plasma by measuring soluble p-tau levels.^4-6

Literature has consistently shown that cognitive deficits attributable to AD are more related to tau than to Aβ pathology.^7-9 Studies where brain tau is measured in vivo have shown that the relationship between tau accumulation and cognitive scores in the AD spectrum varies depending on the domain.^10-15 CSF and plasma phosphorylated tau (p-tau) research has also reported that p-tau biomarker concentrations are differentially related to cognitive scores by domain.¹⁶

Cognitive decline owing to AD occurs most commonly in a sequential manner, with memory problems often identified first.^17-22 Recent research has hinted that AD biomarkers pinpoint the onset of cognitive deficits by domain or modality using PET-based Braak stages,^23,24 where memory appears to be significantly impaired from PET-Braak Stage II onwards, and the rest of the cognitive domains appear to be affected by PET-Braak Stage IV.^25-27

The literature has also indicated the feasibility of tracing the trajectory of cognitive decline through application of the ATN framework established by the National Institutes on Aging and Alzheimer’s Association (NIA-AA).²⁸ The ATN framework refers to the use of amyloid (A), tau (T) and neurodegeneration (N) biomarkers to stage AD, where amyloid positivity is followed by tau positivity, then neurodegeneration. Importantly, recent work has sparked a discussion on the split of the ‘T’ into CSF or plasma p-tau positivity and tau PET positivity, because it is argued that tau PET and biofluid markers might convey different pieces of information.^29,30 In this respect, it has recently been found that, when defining tau positivity with both CSF p-tau and tau-PET, global cognition was significantly hindered at baseline only for the group with all biomarkers positive when compared with controls, although there was a trend towards worse cognitive scores for the tau discordant group.²⁹ These developments have led the Alzheimer’s Association Workgroup to update criteria for the diagnosis and staging of AD.³¹ The new framework recognizes fluid biomarkers that might increase as a reaction to Aβ pathology (T₁) separately from tau PET biomarkers (T₂) and gives them special status as Core 1 biomarkers together with amyloid biomarkers. Importantly, the association between positivity defined by plasma p-tau biomarkers and performance in specific cognitive domains remains to be tested under this novel framework.

Here, we investigated how plasma p-tau₂₃₁, p-tau₁₈₁ and p-tau₂₁₇ relate to cognitive performance, to what extent they can detect cognitive impairment, and whether classifying individuals in AT₁T₂ profiles adds any valuable information to understanding cognitive deficits in a sample of cognitively unimpaired (CU) elderly and cognitively impaired, Aβ-positive individuals.

We hypothesized that plasma p-tau₂₁₇ and plasma p-tau₁₈₁ would display a stronger association with cognitive performance than plasma p-tau₂₃₁, and that plasma p-tau₂₁₇ would show the strongest correlations with memory scores. We also expected plasma p-tau₂₁₇ to outperform the other plasma biomarkers in predicting cognitive impairment, because plasma p-tau₂₁₇ is particularly sensitive to AD pathology.^32-40 We also expected plasma p-tau₁₈₁ to be superior in predicting cognitive impairment compared with plasma p-tau₂₃₁, because literature depicts the former as a marker that might be more associated with cognitive decline.^41,42 Finally, we predicted that participants that are amyloid positive (A+) and plasma p-tau positive (T₁+) but tau PET negative (T₂−) would exhibit only memory deficits, whereas other cognitive domains would also be disrupted in participants who are positive in all biomarkers (A+T₁+T₂+). Our work advances the study of cognition in relationship to the most well-known plasma p-tau biomarkers and sheds light on the usefulness of splitting tau into two categories for the detection of cognitive deficits.

Materials and methods

Study samples

We assessed 143 individuals from the Translational Biomarkers of Aging and Dementia (TRIAD) cohort⁴³: 103 CU Aβ-positive (+) or -negative (−) older adults (of whom 23 were Aβ+) and 40 cognitively impaired (CI) Aβ+ individuals. Among the CI Aβ+ participants, 30 had mild cognitive impairment (MCI) and 10 had a diagnosis of AD dementia. All participants had plasma assessments of p-tau₁₈₁, p-tau₂₁₇ and p-tau₂₃₁, in addition to amyloid-PET with ¹⁸F-AZD4694 and tau-PET with ¹⁸F-MK6240. They also had a review of their medical history, a neurological examination by a physician and a neuropsychological examination. An interview was conducted with each participant and their study partner, who was a person with close ties to the participant. CU individuals had no objective cognitive impairment. MCI and AD individuals were diagnosed based on the NIA-AA criteria for MCI owing to AD⁴⁴ and for probable AD dementia,⁴⁵ respectively, by a multidisciplinary team of neurologists, neuropsychologists and nurses. Exclusion criteria included systemic conditions that were not adequately controlled through a stable medication regimen, active substance abuse, recent head trauma, recent major surgery or MRI/PET safety contraindications. Medications that some patients used included those to treat or prevent high blood pressure (e.g. bisoprolol, metroprolol, ramipril, calcium channel blockers such as amlodipine, irbesartan), cardiovascular disease (e.g. atorvastatin, rosuvastatin, Pro-AAS), gastrointestinal problems (e.g. docusate, dexilant, pantoprazole), bladder problems (e.g. myrbetriq), issues with connective tissue (e.g. glucosamine, chondroitin), swelling or pain (e.g. rivasa, celebrex, asaphen), lung disease (e.g. salbutamol), thyroid disease (e.g. synthroid) or depression (e.g. citalopram). None of the drugs was taken by more than eight participants. A large portion of them also took vitamin supplements (e.g. multivitamins, vitamins D or B₁₂, etc.).

We also included 552 participants from the Swedish BioFINDER-2 study (NCT03174938): 336 CU older adults (78 Aβ+) and 216 CI Aβ+ older adults. Among the CI Aβ+ participants, 95 had a diagnosis of MCI and 121 of AD dementia. All participants had plasma assessments of p-tau₁₈₁, p-tau₂₁₇ and p-tau₂₃₁, in addition to a CSF Aβ₄₂/Aβ₄₀ or ¹⁸F-flutemetamol Aβ-PET, and tau-PET with ¹⁸F-RO948. In BioFINDER-2, whenever Aβ-PET was available, Aβ status was defined with PET. Most CU (90.8%) and MCI (81.1%) participants had their Aβ status defined based on Aβ-PET, whereas most AD dementia patients (95%) had their Aβ status defined by CSF Aβ₄₂/Aβ₄₀, because participants at this later stage do not undergo Aβ-PET per study protocol. BioFINDER-2 recruits both patients with cognitive symptoms, referred from primary care to specialized care, and CU volunteers. For AD dementia, participants had to fulfil the diagnostic and statistical manual-5 criteria for dementia attributable to AD (major neurocognitive disorder) and demonstrate biomarker evidence of Aβ-positivity according to NIA-AA guidelines, with exceptions evaluated on a case-by-case basis.⁴⁶ Among individuals referred to the memory clinic who did not meet the diagnostic and statistical manual-5 criteria for dementia, those scoring below −1.5 standard deviations (SD) in any cognitive domain based on age and education-adjusted norms were classified as having MCI. Those who did not meet the criteria for MCI were classified as having subjective cognitive decline and included in the CU group. General exclusion criteria included: (i) significant unstable systemic illness making study participation difficult; (ii) current significant alcohol or substance misuse; and (iii) refusal of lumbar puncture, MRI or PET. Information on medication was not available for the BioFINDER-2 cohort.

TRIAD was approved by the Montreal Neurological Institute (MNI) PET working committee and the Douglas Mental Health University Institute Research Ethics Board (IUSMD16-61, IUSMD16-60). BioFINDER-2 was approved by the Ethical Review Board in Lund, Sweden, which is part of the Swedish Ethical Review Authority (#2016-1053). All patients gave their written informed consent to participate in the study, and all participants were volunteers.

Neuropsychological testing

Neuropsychological evaluation encompassed memory, executive, language and visuospatial cognitive domains. In TRIAD, memory was assessed using the delayed recall Logical Memory (LM) subtest of the Wechsler Adult Intelligence Scale 4th edition (WAIS-IV),⁴⁷ the Rey Auditory Verbal Learning Test (RAVLT)⁴⁸ delayed recall and the Free and Cued Selective Reminding Test delayed recall (FCSRT).⁴⁹ Executive function (EF) was assessed using Trail-Making Test–B time (TMT-B),⁵⁰ Digit Span’s total score from the WAIS-IV and inhibition scores from the Delis-Kaplan Executive Function System (D-KEFS).⁵¹ Language was assessed with the D-KEFS category fluency and the Boston Naming Test (BNT).⁵² Visuospatial function was assessed with the Digit-Symbol Substitution Test (DSST)⁵³ and Birmingham Object Recognition Battery (BORB)⁵⁴ orientation tests. In BioFINDER-2, tests evaluating the same cognitive domains were included. For memory, we used delayed and immediate recall scores from the Alzheimer’s Disease Assessment Scale (ADAS).⁵⁵ Executive function was assessed with the TMT-A and TMT-B tests.⁵⁰ Language was evaluated with the Animal Fluency test and the BNT.⁵² Visuospatial function was assessed using the Symbol Digit Modality Test (SDMT)⁵⁶ and the cube test and incomplete letter components from the Visual Object and Space Perception (VOSP) battery.⁵⁷

Raw test scores were z-transformed using mean and SD values from larger samples of CU older adults from each cohort that had available data on specific cognitive tests, APOE ɛ4 genotyping, years of education and any of the plasma p-tau biomarkers included in the study. In total, 147, 140, 143 and 142 individuals were available to calculate memory, EF, language and visuospatial scores, respectively, in the TRIAD cohort. For the BioFINDER-2 cohort, 347, 345, 342 and 341 individuals were available to calculate memory, EF, language and visuospatial scores, respectively. The z-scores were averaged across all tests within each cognitive domain, resulting in a composite score for each domain. Cognitive impairment was defined as 1.5 SD below the average for each domain, which indicates clinical impairment when cognitively impaired individuals are matched to cognitively unimpaired elderly in age and years of education.^44,58,59

Plasma and CSF biomarkers

Blood samples were collected following previously described protocols.⁶⁰ For both cohorts, APOE ɛ4 genotyping was performed. Plasma p-tau₁₈₁ and plasma p-tau₂₃₁ were measured in the Clinical Neurochemistry Laboratory, University of Gothenburg, by scientists blinded to participant clinical and biomarker information. Both plasma biomarkers were assessed using an in-house Single Molecular Array (Simoa) method (Simoa HD-X instruments, Quanterix), as described previously.^60,61 In TRIAD, plasma p-tau₂₁₇ concentrations were measured using a Simoa assay developed by Janssen,^62,63 by scientists blinded to clinical and biomarker data. In BioFINDER-2, plasma p-tau₂₁₇ was measured with a Mesoscale Discovery (MSD) immunoassay developed by Lilly Research Laboratories and quantified at the Memory Clinic Research Unit (Lund University, Sweden).³⁹

Abnormality for plasma biomarkers was predefined in accordance with appropriate use recommendations.⁶⁴ In TRIAD, a threshold of 15.085 pg/ml was used for plasma p-tau₁₈₁, a threshold of 17.652 pg/ml was used for plasma p-tau₂₃₁ and a threshold of 0.083 pg/ml was determined for the Janssen plasma p-tau₂₁₇ assay, as detailed elsewhere.^30,65 To define p-tau positivity in BioFINDER-2, we determined that those presenting with a z-scored biomarker value (normalized to a CU Aβ− population) >1.5 were positive.⁶⁶ The threshold for p-tau₂₁₇ positivity was 0.27 pg/ml, and it was 12.0 pg/ml for p-tau₁₈₁ and 8.12 pg/ml for p-tau₂₃₁.

In BioFINDER-2, CSF Aβ₄₂/Aβ₄₀ was measured using the fully automated Roche Elecsys NeuroTool Kit, and abnormal CSF status was defined based on previously derived cut-offs determined using Gaussian mixture modelling, with a threshold of ≥0.08 for Aβ positivity.

PET scans

In TRIAD, ¹⁸F-AZD4694 PET and ¹⁸F-MK6240 PET scans were conducted with a brain-dedicated Siemens High Resolution Research Tomograph (HRRT). ¹⁸F-AZD4694 PET images were acquired 40–70 min after bolus injection and reconstructed with the ordered-subset expectation maximization algorithm (OSEM) on a four-dimensional volume with three frames (3 × 600 s), as previously described.^{67 18}F-MK6240 PET images were acquired at 90–110 min after bolus radiotracer injection and reconstructed with the OSEM algorithm on a four-dimensional volume with four frames (4 × 300 s). A 6-min transmission scan with a rotating ¹³⁷Cs point source was performed after each PET acquisition for attenuation correction. Corrections to PET images were applied for decay, motion, dead time, random and scattered coincidences. T₁-weighted MRIs were acquired at the MNI on a 3 T Siemens Magnetom using a standard head coil. Non-uniformity and field-distortion corrections were conducted, and images were processed using an in-house pipeline. PET images were automatically registered to the T₁-weighted image space, and the T₁-weighted images were linearly and non-linearly registered to the Alzheimer’s Disease Neuroimaging Initiative (ADNI) reference space. Fluid attenuated inversion recovery (FLAIR) sequences were used to obtain Fazekas scores. ¹⁸F-MK6240 images were meninges stripped in native space before they were transformed and blurred to minimize interference of meningeal spillover, as previously described.²⁴ The whole cerebellum grey matter was used as the reference region to calculate [18F]AZD4694 standardized uptake value ratio (SUVR) maps, and the inferior cerebellar grey was used as the reference region to generate [18F]MK6240 SUVR maps. Spatial smoothing allowed the PET images to achieve an 8 mm full-width at half-maximum resolution.

Aβ-PET SUVR was estimated for each participant by averaging the SUVR from the precuneus, prefrontal, orbitofrontal, parietal, temporal and cingulate cortices included in the regions of interest (ROIs). Aβ positivity was defined as a global ¹⁸F-AZD4694 SUVR of >1.55.⁶⁷ Temporal meta-ROI SUVR (tau PET) was calculated by averaging ¹⁸F-MK6240 SUVR values from entorhinal, parahippocampal, amygdala, fusiform, inferior and middle temporal cortices; and positivity was defined when SUVR > 1.24, as previously reported.⁶⁸

In BioFINDER-2, Aβ-PET was quantified using ¹⁸F-flutemetamol on a digital GE Discovery MI scanner in BioFINDER-2. Scans were acquired 90–110 min after the injection of ∼185 MBq of ¹⁸F-flutemetamol. The SUVR was obtained by normalizing the neocortical composite values to the whole cerebellum as a reference region. FreeSurfer (v.5.3) parcellation of the T₁-weighted MRI scan was used to transform the PET data to the native T₁-space of participants in order to obtain mean regional SUVR values in predefined neocortical ROIs, including prefrontal, lateral temporal, parietal, anterior cingulate and posterior cingulate/precuneus.⁶⁹ Aβ-PET data were binarized into normal and abnormal using cut-offs derived from Gaussian mixture modelling, with a threshold of ≥1.033.⁷⁰ Tau PET was measured using ¹⁸F-RO948 in BioFINDER-2 70–90 min following injection (list-mode acquisition) using the GE Discovery MI digital scanner, with SUVR images obtained using the inferior cerebellar cortex as the reference region. A high-resolution T₁-weighted MRI was acquired (3 T MAGNETOM Prisma; Siemens Healthineers) for PET image coregistration and template normalization. Tau-PET positivity was determined by a previously validated cut-off of 1.36 based on a temporal meta-ROI, computed as a weighted average of entorhinal, amygdala, parahippocampal, fusiform and inferior and middle temporal ROIs.^71-73

A/T₁/T₂ classification

We categorized participants into different biomarker profiles based on whether they were Aβ (A), p-tau (T₁) or tau PET (T₂) positive (+) or negative (−), inspired by the revised criteria for diagnosis and staging of AD by the Alzheimer’s Association Workgroup.³¹ The procedure was done for each plasma biomarker. This taxonomy system rendered the following groups for each biomarker: (i) A−T₁−T₂− (controls); (ii) A+T₁−T₂− (tau negative); (iii) A+T₁+T₂− (tau discordant); (iv) A+T₁+T₂+ (tau positive); and (v) A−T₁+ (amyloid negative). In both cohorts there were a few participants (n < 10) who were A+T₁−T₂+. We decided to include them in the A+T₁+T₂+ group to avoid bias in our analyses, because previous research has indicated that their cognitive scores might be worse than those of A+T₁+T₂− individuals.⁷⁴ Supplementary Figs 1–6 summarize the distribution of AT₁T₂ profiles in each cohort according to diagnostic groups.

Statistical analyses

Statistical analyses were performed in R v.4.3.0. All analyses were performed in the whole sample (i.e. both CU and CI Aβ+) unless otherwise specified. Dunn’s tests were conducted to assess differences for continuous variables across diagnosis, where results were significant if P < 0.05. The χ² test or Fisher’s exact test was used for categorical variables. Benjamini–Hochberg correction for multiple comparisons⁷⁵ was applied to both Dunn’s and χ² tests or Fisher’s exact tests.

We computed partial correlations (Spearman’s ρ) using the ppcor package of R⁷⁶ to assess the independent association between biomarkers and cognitive scores. Partial correlation analyses were also performed for Aβ+, CI Aβ+ and CU subsamples. Moreover, we conducted regression models to assess associations further in the whole sample (simple models). P-tau biomarkers or tau PET were entered as predictors, with cognitive scores as the outcome variable. We corrected for age, years of education, sex, deep white-matter hyperintensity Fazekas^77,78 scores and APOE ɛ4 status. Additionally, we ran models including all plasma biomarkers and models including both plasma biomarkers and tau PET (combined models). Variance inflation factor and tolerance scores were obtained to assess the presence of collinearity in regression models. We adopted two strategies to examine the strength of the relationship between plasma biomarkers and cognitive scores. First, we used 83.4% confidence intervals (CIs) of standardized β coefficients from the simple models to assess whether differences were significant.^79,80 Second, we used Dominance Analysis^81,82 to determine the relative importance of the contribution of plasma biomarkers to the variance explained in all possible subset models given all the predictors and covariates that we included. A predictor would be considered to dominate another completely when its additional contribution to each of the models to be compared is greater than that of the other predictor. A second level of dominance can be established by averaging the contribution of predictors within each model size. When the average of one predictor is higher than that of other predictor, conditional dominance is established. Yet, a third level of dominance (general dominance) can be ascertained by averaging the contribution of variables across models of all possible sizes. The additional contribution of a predictor to a certain subset model in ordinary least squares regression is defined as the change in R² when adding the predictor to the model.

Area under the receiver operating characteristic (ROC) curve (AUC) values were calculated for all plasma p-tau biomarkers for detection of cognitive impairment. We tested for differences in AUCs using DeLong’s method in the pROC package of R.⁸³ Benjamini–Hochberg correction for multiple comparisons was applied. A cut-off using Youden’s method⁸⁴ was calculated, which rendered optimal sensitivity, specificity and accuracy values to determine impairment in each cognitive domain for each biomarker. CIs for sensitivity, specificity and accuracy were calculated, performing bootstrap with 10 000 replicates. AUC analyses were also performed in Aβ+ and CI Aβ+ subgroups.

We conducted non-parametric comparisons with Dunn’s post hoc tests across A/T₁/T₂ groups, where results were significant if P < 0.05.

Results

Demographic, clinical and biomarker data are summarized in Table 1. All variables in regression models had a variance inflation factor <5 and a tolerance >0.2, which are not deemed to indicate problematic levels of multicollinearity.⁸⁵ Bonferroni outlier tests discarded the presence of outliers. Raw cognitive scores per diagnostic group are available in Supplementary Tables 1 and 2.

Table 1.

Demographics, biomarker and cognitive data

Parameter	TRIAD			BIOFINDER-2
	CU	MCI	AD	CU	MCI	Dementia
Number (%)	103 (72.0%)	30 (21%)	10 (7%)	336	95	121
Sex (female; %)	63.1%	60.0%	40.0%	53.6%	49.5%	51.2%
Age, mean (SD), years	71.5 (5.5)	71.7 (5.0)	68.4 (8.3)	66.2 (11.4)a,b	72.3 (8.46)a	74.1 (6.71)b
Education, mean (SD), years	15.2 (3.6)	15.4 (3.6)	14.3 (4.7)	12.5 (3.42)	13.2 (4.87)	12.1 (4.09)
APOE ɛ4 status	25.2%b	46.7%	70%b	46.1%a,b	77.9%a	71.9%b
18F-AZD4694 SUVRc, mean (SD)	1.46 (0.35)a,b	2.33 (0.46)a	2.35 (0.37)b	NA	NA	NA
18F-Flutemetamol SUVRc, mean (SD)	NA	NA	NA	1.01 (0.206)a,b	1.46 (0.300)a	1.72 (0.229)b
Missing	–	–	–	31 (9.1%)	18 (18.8%)	115 (95.0%)
CSF positivityd, n (%)	NA	NA	NA	88 (25.9%)a,b	96 (100%)a	121 (100%)b
18F-MK6240 temporal metaROI SUVRe, mean (SD)	0.86 (0.09)b,c	1.32 (0.6)a	2.1 (0.94)a	NA	NA	NA
18F-RO948 temporal metaROI SUVRe, mean (SD)	NA	NA	NA	1.16 (0.14)a,b	1.47 (0.43)a,f	1.87 (0.68)b,f
Plasma p-tau217g, mean (SD)	0.06 (0.05)a,b	0.14 (0.06)a	0.23 (0.13)b	0.19 (0.14)a,b	0.43 (0.29)a,f	0.61 (0.33)b,f
Plasma p-tau181, mean (SD)	10.93 (7.48)a,b	16.48 (6.59)a	28.15 (16.17)b	6.67 (4.34)a,b	9.97 (6.38)a,f	12.5 (6.47) b,f
Plasma p-tau231, mean (SD)	14.88 (9.15)a,b	19.97 (7.79)a	25.41 (12.1)b	6.24 (3.91)a,b	10.7 (5.62)a	10.7 (4.02)b
DWMH Fazekas (grade): 0/1/2/3	59/37/6/1a,b	5/20/4/1a	1/8/1/0b	43/199/65/17 (12)a,b	5/49/30/7 (4)a	3/60/40/11 (7)b
Memory composite, mean (SD)	0.0 (0.7)a,b	−1.8 (1.3)a	−3.5 (0.7)b	0 (0.7)a,b	−1.9 (1.1)a,f	−3.0 (1.0)b,f
Executive function composite, mean (SD)	−0.1 (0.8)a,b	−0.7 (0.9)a	−1.6 (2.1)b	0 (0.9)a,b	−2.1 (2.1)a,f	−4.6 (3.5)b,f
Language composite, mean (SD)	0.0 (0.8)a,b	−0.7 (1.5)a,f	−1.7 (1.4)b,f	0 (0.8)a,b	−1.2 (1.1)a,f	−2.1 (1.6)b,f
Visuospatial composite, mean (SD)	0.0 (0.7)a,b	−0.8 (0.8)a	−1.7 (2.0)b	0 (0.8)a,b	−1.1 (1.4)a,f	−2.2 (1.9)b,f
MMSE (SD)	29.1 (1.1)a,b	27.9 (1.8)a	24.2 (4.3)b	28.9 (1.3)a,b	26.7 (2.0)a,f	21.6 (4.1)b,f
CDRh (0/0.5/1/2)	101/2/0/0a,b	5/25/0/0a,f	0/6/4/0b,f	NA	NA	NA

Dunn’s tests were conducted to assess differences between groups, with Benjamini–Hochberg correction for multiple comparisons, except for sex, APOE ɛ4 status and CDR sum of boxes, where contingency χ² tests and Fisher’s exact tests were conducted. AD = Alzheimer’s disease; CDR = Clinical Dementia Rating; CU = cognitively unimpaired; DWMH = deep white matter hyperintensities; MCI = mild cognitive impairment; MMSE = Mini-Mental State Examination; NA = not available; SD = standard deviation; SUVR = standardized uptake value ratio.

^aSignificant differences between CU and MCI groups.

^bSignificant differences between CU and AD or dementia groups.

^cAmyloid PET scans differed by cohort.

^dAmyloid positivity was defined by CSF in the BIOFINDER-2 cohort.

^eTau PET tracers differed by cohort.

^fSignificant differences between MCI and AD or dementia groups.

^gPlasma p-tau₂₁₇ is Janssen’s for TRIAD, and Alzpath for BIOFINDER-2.

^hCDR scores were not available in the BioFINDER-2 cohort.

Associations between plasma biomarkers and cognitive scores

We tested the hypothesis that plasma biomarkers would relate differentially to cognitive performance by domain. In both cohorts, partial correlation analyses confirmed that the strength of associations varied by cognitive domain and that the relationship between cognitive scores and plasma p-tau₂₁₇ shed the highest values among plasma biomarkers from memory to EF, to language and to visuospatial ability (Fig. 1 and Supplementary Figs 7–9). Regression analyses showed a similar pattern (Table 2). For brevity, and given that the highest coefficients were observed for memory scores, we focus on this cognitive domain in the present section. The 83.4% CIs for β coefficients showed that the independent relationship between plasma p-tau₂₁₇ and memory was significantly stronger than the independent relationship between plasma p-tau₁₈₁ and memory in BioFINDER-2 (but not in TRIAD), but it was stronger than that between p-tau₂₃₁ and memory in both cohorts. In both cohorts, the independent associations between p-tau₁₈₁ or p-tau₂₃₁ and memory were not significantly different from each other. Dominance analyses revealed complete dominance of tau PET over all other variables for the association with memory scores in both cohorts (Table 3 for complete dominance). Plasma p-tau₂₁₇ displayed complete dominance over the other plasma biomarkers and all covariates except for years of education in TRIAD. In BioFINDER-2, plasma p-tau₂₁₇ achieved complete dominance only over p-tau₁₈₁, p-tau₂₃₁ and APOE. Conditional and general dominance results are shown in Supplementary Tables 3 and 4 and Supplementary Fig. 10.

Figure 1.

The association of plasma p-tau₂₁₇ with memory is stronger than that of other p-tau epitopes and memory and it is closer to the association with tau in the brain. The plots show the relationship between residuals after regressing age, sex, APOE ɛ4, years of education and Fazekas scores on memory scores and residuals after regressing age, sex, APOE ɛ4, years of education and Fazekas scores on plasma p-tau. (A–D) TRIAD cohort. (E–H) BioFINDER-2 cohort. AD = Alzheimer's disease; CU = cognitively unimpaired; MCI = mild cognitive impairment; p-tau = phosphorylated tau; TRIAD = Translational Biomarkers of Aging and Dementia cohort.

Table 2.

Standardized β coefficients for plasma p-tau biomarkers with cognitive scores as outcome variable

Parameter	TRIAD			BIOFINDER-2
	β (83.4% CI)	t-value	P-value	β (83.4% CI)	t-value	P-value
Composite memory scores
Plasma p-tau217	−0.53 (−0.63 to −0.43)	−7.14	<0.001	−0.52 (−0.57 to −0.47)	−15.0	<0.001
Plasma p-tau181	−0.35 (−0.46 to −0.24)	−4.46	<0.001	−0.24 (−0.30 to −0.19)	−5.8	<0.001
Plasma p-tau231	−0.24 (−0.36 to −0.13)	−2.97	0.004	−0.29 (−0.35 to −0.24)	−7.1	<0.001
Composite executive function scores
Plasma p-tau217	−0.48 (−0.65 to −0.32)	−5.97	<0.001	−0.38 (−0.44 to −0.33)	−9.8	<0.001
Plasma p-tau181	−0.29 (−0.45 to −0.12)	−3.37	0.001	−0.23 (−0.29 to −0.17)	−5.6	<0.001
Plasma p-tau231	−0.21 (−0.33 to −0.09)	−2.43	0.02	−0.20 (−0.26 to −0.14)	−4.5	<0.001
Composite language scores
Plasma p-tau217	−0.35 (−0.46 to −0.23)	−4.19	<0.001	−0.35 (−0.40 to −0.29)	−8.9	<0.001
Plasma p-tau181	−0.15 (−0.27− 0.03)	−1.8	0.07	−0.17 (−0.23 to −0.12)	−4.2	<0.001
Plasma p-tau231	−0.14 (−0.25 to −0.02)	−1.61	0.11	−0.18 (−0.24 to −0.12)	−4.6	<0.001
Composite visuospatial scores
Plasma p-tau217	−0.44 (−0.55 to −0.32)	−5.36	<0.001	−0.35 (−0.41 to −0.30)	−8.7	<0.001
Plasma p-tau181	−0.24 (−0.36 to −0.12)	−2.85	0.005	−0.24 (−0.30 to −0.18)	−5.7	<0.001
Plasma p-tau231	−0.14 (−0.26 to −0.02)	−1.57	0.12	−0.19 (−0.25 to −0.13)	−4.2	<0.001

Standardized β coefficients from 12 separate models with years of education, sex, age, APOE ɛ4 status and Fazekas scores as covariates. CI = confidence interval; p-tau = phosphorylated tau; TRIAD = Translational Biomarkers of Aging and Dementia cohort.

Table 3.

Complete dominance for composite memory scores

Parameter	Tau PET	P-tau217	P-tau181	P-tau231	ApoE	Fazekas	Education	Sex	Age
TRIAD
Tau PET	0.5	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0
P-tau217	0.0	0.5	1.0	1.0	1.0	1.0	0.5	1.0	0.5
P-tau181	0.0	0.0	0.5	0.5	0.5	0.5	0.5	0.5	0.5
P-tau231	0.0	0.0	0.5	0.5	0.5	0.5	0.5	0.5	0.5
APOE ɛ4	0.0	0.0	0.5	0.5	0.5	0.5	0.5	0.5	0.5
Fazekas	0.0	0.0	0.5	0.5	0.5	0.5	0.5	0.5	0.5
Education	0.0	0.5	0.5	0.5	0.5	0.5	0.5	1.0	1.0
Sex	0.0	0.0	0.5	0.5	0.5	0.5	0.0	0.5	0.5
Age	0.0	0.5	0.5	0.5	0.5	0.5	0.0	0.5	0.5
BIOFINDER-2
Tau PET	0.5	1.0	1.0	1.0	1.0	1.0	1.0	1.0	1.0
P -tau217	0.0	0.5	1.0	1.0	1.0	0.5	0.5	0.5	0.5
P-tau181	0.0	0.0	0.5	0.5	0.5	0.5	0.5	0.5	0.5
P-tau231	0.0	0.0	0.5	0.5	0.5	0.5	0.5	0.5	0.5
APOE ɛ4	0.0	0.0	0.5	0.5	0.5	0.5	0.5	0.5	0.5
Fazekas	0.0	0.5	0.5	0.5	0.5	0.5	0.5	0.5	0.5
Education	0.0	0.5	0.5	0.5	0.5	0.5	0.5	0.5	0.5
Sex	0.0	0.0	0.0	0.5	0.5	0.5	0.5	0.5	0.5
Age	0.0	0.5	0.5	0.5	0.5	0.5	0.5	0.5	0.5

1.0 = dominance; 0.5 = absence of dominance; 0 = the dominance of the variable to which the reference is compared. P-tau = phosphorylated tau; TRIAD = Translational Biomarkers of Aging and Dementia cohort.

In subsequent analyses, we fitted models with tau PET and covariates alone, and with tau PET, plasma biomarkers and covariates for comparative reference. These analyses showed that, in both cohorts, p-tau₂₁₇ is the only plasma biomarker that remained significantly or marginally associated with memory, EF and visuospatial scores when including all tau biomarkers in the model (e.g. for memory, TRIAD: standardized β = −0.22, P = 0.09; BioFINDER-2: standardized β = −0.29, P < 0.001), suggesting that it adds value even when tau PET is accounted for. Full statistics of the model including all biomarkers are displayed in Supplementary Tables 5–8.

In line with the analyses that were performed on the main sample, partial correlation analyses in Aβ+ individuals revealed that the strongest associations between plasma biomarkers and cognitive scores were observed for plasma p-tau₂₁₇ and memory (Fig. 2 for memory and Supplementary Figs 11–13 for EF, language and visuospatial ability). In CI Aβ+ participants, partial correlations were still among the highest for the association between p-tau₂₁₇ and memory (TRIAD: ρ = −0.45, P = 0.004; BioFINDER-2: ρ = −0.42, P < 0.001), but associations between plasma p-tau biomarkers and cognitive domains beyond p-tau₂₁₇ and memory were sometimes as high in the TRIAD cohort (e.g. for EF: ρ = −0.45, P = 0.003). Results for CI Aβ+ individuals are displayed by Fig. 3 and Supplementary Figs 14–16. Often, in both the previous subsamples, associations were not significant for plasma p-tau₁₈₁ and p-tau₂₃₁. Analyses in CU individuals revealed inconsistencies across cohorts. Although the TRIAD cohort rendered only one significant result, most results were significant for BioFINDER-2 (Supplementary Figs 17–20). In particular, significant negative correlations were found for plasma p-tau₂₁₇ and plasma p-tau₂₃₁ biomarkers on each occasion.

Figure 2.

The association of plasma p-tau₂₁₇ with memory remains stronger than that of other p-tau epitopes and memory in amyloid-positive individuals. The plots show the relationship between residuals after regressing age, sex, APOE ɛ4, years of education and Fazekas scores on memory scores and residuals after regressing age, sex, APOE ɛ4, years of education and Fazekas scores on plasma p-tau. (A–D) TRIAD cohort. (E–H) BioFINDER-2 cohort. AD = Alzheimer's disease; CU = cognitively unimpaired; MCI = mild cognitive impairment; p-tau = phosphorylated tau; TRIAD = Translational Biomarkers of Aging and Dementia cohort.

Figure 3.

The association of plasma p-tau₂₁₇ with memory remains stronger than that of other p-tau epitopes and memory in cognitively impaired individuals. The plots show the relationship between residuals after regressing age, sex, APOE ɛ4, years of education and Fazekas scores on memory scores and residuals after regressing age, sex, APOE ɛ4, years of education and Fazekas scores on plasma p-tau. (A–D) TRIAD cohort. (E–H) BioFINDER-2 cohort. AD = Alzheimer's disease; MCI = mild cognitive impairment; p-tau = phosphorylated tau; TRIAD = Translational Biomarkers of Aging and Dementia cohort.

ROC analyses for the detection of cognitive impairment

Our results indicated that plasma p-tau₂₁₇ was superior to p-tau₁₈₁ (TRIAD: AUC_p-tau217 = 0.86 versus AUC_p-tau181 = 0.77, P = 0.007; BioFINDER-2: AUC_p-tau217 = 0.86 versus AUC_p-tau181 = 0.76, P < 0.001) and p-tau₂₃₁ (TRIAD: versus AUC_p-tau231 = 0.75, P = 0.012; BioFINDER-2: versus AUC_p-tau231 = 0.81, P < 0.001) at predicting memory impairment, and as effective as tau PET (TRIAD: versus AUC_tauPET = 0.91, P = 0.11; BioFINDER-2: versus AUC_tauPET = 0.87, P = 0.86) (Fig. 4A and E). However, in TRIAD, the predictive power across plasma biomarkers and tau PET did not differ significantly for EF (Fig. 4B), language (Fig. 4C) and visuospatial (Fig. 4D) impairment. In BioFINDER-2, plasma p-tau₂₁₇ was superior to both p-tau₁₈₁ and p-tau₂₃₁ in predicting executive impairment (P < 0.05 for both; Fig. 4F), but was not significantly superior to them at predicting language and visuospatial impairment (P > 0.05 for all; Fig. 4G and H, respectively). In TRIAD, it was determined that a cut-off of 0.115 [95% CI (0.075, 0.145)] pg/ml for p-tau₂₁₇ would identify memory impairment with 0.9 specificity, 0.72 sensitivity and 0.87 accuracy. For p-tau₁₈₁, a cut-off value of 17.34 [95% CI (9.38, 20.38)] pg/ml indicated memory impairment with 0.91 specificity, 0.59 sensitivity and 0.85 accuracy. Furthermore, we revealed that a cut-off value of 16.71 [95% CI (16.06, 23.28)] pg/ml indicated memory impairment for p-tau₂₃₁ with 0.64 specificity, 0.79 sensitivity and 0.67 accuracy. We estimated that a ¹⁸F-MK6240 SUVR value of 1.06 [95% CI (0.93, 1.2)] identified memory impairment with 0.96 specificity, 0.76 sensitivity and 0.92 accuracy. In BioFINDER-2, for memory impairment, the p-tau₂₁₇ threshold was 0.2891 pg/ml [95% CI (0.256, 0.3568) pg/ml], with specificity at 0.82, sensitivity at 0.79 and accuracy at 0.81. The p-tau₁₈₁ threshold was 7.478 pg/ml [95% CI (6.531, 8.2) pg/ml], with 0.68 specificity, 0.77 sensitivity and 0.70 accuracy. For p-tau₂₃₁, the threshold was 7.145 pg/ml [95% CI (6.562, 8.203) pg/ml], with 0.74 specificity, 0.82 sensitivity and 0.77 accuracy. The ¹⁸F-RO948 SUVR cut-off value was 1.284 [95% CI (1.242, 1.304)], with 0.90 specificity, 0.77 sensitivity and 0.85 accuracy. CIs for specificity, sensitivity and accuracy are available in Supplementary Table 9. Cut-off values for the rest of the cognitive domains can also be found in the Supplementary material, ‘Results’ section 1–2. Youden J values can be found in Supplementary Table 10.

Figure 4.

Receiver operating characteristic curves for cognitive impairment across biomarkers. Plasma p-tau₂₁₇ identifies memory impairment with an accuracy that is significantly higher than that of other plasma biomarkers and not significantly different from that of tau PET in both cohorts. (A–D) TRIAD cohort. (E–H) BioFINDER-2 cohort. AUC = area under receiver operating characteristic curve; p-tau = phosphorylated tau; TRIAD = Translational Biomarkers of Aging and Dementia cohort.

Plasma p-tau₂₁₇ AUC values for detecting memory impairment were significantly higher than those of plasma p-tau₂₃₁, except for CI Aβ+ individuals in BioFINDER-2 (Supplementary Figs 21 and 22 for Aβ+ and CI Aβ+, respectively). Results for the comparison between plasma p-tau₂₁₇ and plasma p-tau₁₈₁ in the domain of memory were less consistent. Differences were significant for Aβ+ individuals in BioFINDER-2, and in CI Aβ+ individuals for TRIAD. In BioFINDER-2, plasma p-tau₂₁₇ was superior to plasma p-tau₂₃₁ across all other domains in Aβ+ individuals and in CI Aβ+ individuals only in the visuospatial domain. Tau PET was superior to all biomarkers in detecting memory impairment in both groups, except plasma p-tau₂₃₁ in CI Aβ+ participants in TRIAD. No further significant differences were found.

Cognitive deficits across biomarker profiles

Our analyses revealed that, in both cohorts, A+T₁+T₂+ subjects presented cognitive scores substantially lower than most other AT₁T₂ groups across all cognitive domains irrespective of whether we used plasma p-tau₂₁₇, p-tau₁₈₁ or p-tau₂₃₁ to define plasma biomarker (‘T₁’) positivity (Fig. 5 and Supplementary Figs 23–29). In contrast to the results for TRIAD, amyloid-positive (A+T₁−T₂−) participants from the BioFINDER-2 cohort had significantly lower scores for all cognitive domains when compared with controls (A−T₁−T₂−). When compared with the tau-discordant group (A+T₁+T₂−), no differences were found across cognitive domains or plasma biomarkers in both cohorts. Tau-discordant individuals showed significantly lower memory scores compared with controls for p-tau₂₁₇ and p-tau₁₈₁ in both cohorts (TRIAD: P_p-tau217 = 0.0073 and P_p-tau181 = 0.01; BioFINDER-2: P_p-tau217 < 0.001 and P_p-tau181 < 0.001). Significant memory impairment was not observed for these individuals when T₁ = p-tau₂₃₁ in TRIAD but was observed in BioFINDER-2 (Supplementary Fig. 23). Additionally, cognitive deficits in any of the other domains (executive function, language and visuospatial) were not significant when comparing tau-discordant subjects with controls in TRIAD (Supplementary Figs 24, 26 and 28), but were significant in BioFINDER-2 (Supplementary Figs 25, 27 and 29). Amyloid-negative individuals with positive p-tau (A−T₁+) did not present lower cognitive scores in any cognitive domains in either of the cohorts.

Figure 5.

Memory impairment is likely to be detected in participants with no significant amount of tau in the brain, and is outstanding in tau-positive participants. (A and B) TRIAD cohort. (C and D) BioFINDER-2 cohort. P-values reflect the result of post hoc analyses using Dunn’s test when comparing different groups with the group of individuals that are negative in all biomarkers (control group). Significance is defined by P < 0.05; ns = not significant. A−T₁−T₂− = controls; A+T₁−T₂− = tau negative; A+T₁+T₂− = tau discordant; A+T₁+T₂+ = tau positive; A−T₁+ = amyloid negative; TRIAD = Translational Biomarkers of Aging and Dementia cohort.

Discussion

Our study aimed to explore how plasma p-tau species associate with cognitive performance across the AD spectrum. We found that plasma p-tau₂₁₇ concentrations were more closely associated with cognitive performance in comparison to plasma p-tau₁₈₁ or plasma p-tau₂₃₁ concentrations. Our analyses revealed that this relationship was particularly strong for memory. Moreover, plasma p-tau₂₁₇ outperformed plasma p-tau₁₈₁ and plasma p-tau₂₃₁ in predicting memory impairment in both cohorts and EF impairment in the BioFINDER-2 cohort but found no differences with respect to other cognitive domains in either cohort. Importantly, we showed that different biomarker profiles were associated with distinct cognitive profiles, but it was unclear whether cognitive profiles for amyloid-positive and tau-discordant individuals differed. Overall, our results support plasma p-tau₂₁₇ as a particularly sensitive biomarker associated with cognitive impairment attributable to AD, especially in the domain of memory.

Our findings pointed to plasma p-tau₂₁₇ as a biomarker that is more closely associated with cognition than plasma p-tau₁₈₁ and plasma p-tau₂₃₁ across cohorts and samples. Plasma p-tau₂₁₇ and memory deficits showed the highest associations in the whole sample and in the AD spectrum, whereas the relationship between p-tau₂₁₇ and other cognitive scores was equally strong in individuals with clinical impairment attributable to AD. The latter finding supports the idea that cognitive deficits are widespread in clinical AD. Plasma p-tau₁₈₁ and p-tau₂₃₁ do not seem as sensitive to cognitive deficits as p-tau₂₁₇. In CU individuals, plasma p-tau₂₁₇ and p-tau₂₃₁ were consistently correlated with cognitive scores in BioFINDER-2 (Supplementary Figs 17–20), but not in TRIAD. Previous research revealed that plasma p-tau₁₈₁ and p-tau₂₃₁ were correlated with cognition in cognitively unimpaired individuals.^41,86 The unique relationship between plasma p-tau₂₁₇ and cognitive deficits in AD has only recently been highlighted by several studies.^32-40 For example, p-tau₂₁₇ concentrations at baseline predicted cognitive decline in a sample of autosomal-dominant AD patients³⁴ and in a sample of cognitively unimpaired Aβ+ individuals.³³ A recent study specifically hinted at a strong relationship between p-tau₂₁₇ and memory in cognitively unimpaired individuals.³² Here, we showed that plasma p-tau₂₁₇ is the biomarker most sensitive to cognitive deficits across groups, and that memory deficits display the strongest associations with plasma biomarkers in the spectrum. The association between p-tau₂₁₇ concentrations and memory performance was generally the strongest.

We also showed that p-tau₂₁₇ outperformed p-tau₁₈₁ and plasma p-tau₂₃₁ in detecting memory impairment in both cohorts and EF impairment in BioFINDER-2 but found no differences regarding language and visuospatial cognitive domains. Furthermore, the performance of p-tau₂₁₇ fell within a range to be considered excellent, according to Mandrekar et al.,⁸⁷ in detecting memory impairment (TRIAD: AUC = 0.86; BioFINDER-2: 0.86), and was not significantly different from tau PET (TRIAD: AUC = 0.91; BioFINDER-2: 0.87). This was also true for the performance of p-tau₂₃₁ to detect memory impairment and the performance of p-tau₂₁₇ to detect EF impairment in the BioFINDER-2 cohort. To the same standards,⁸⁷ the remaining AUC values across biomarkers and cognitive domains were acceptable at best (range AUC = 0.653–0.79). As expected, biomarkers did not perform so well in subgroups, probably because the subsamples were more homogeneous (i.e. most were already impaired). Thus, the significance of differences was less meaningful overall. Previous studies have sought to differentiate participants in terms of amyloid positivity,^38,88-93 cognitive status/diagnosis^90,92,94,95 or tau positivity,^38,92 but differentiation based on cognitive impairment by domain has not been investigated, to the best of our knowledge. Our results further emphasize the role of plasma p-tau₂₁₇ in uncovering memory deficits in AD.

Lastly, we unveiled cognitive deficits across domains when amyloid positivity was combined with tau PET positivity. Our results were inconsistent for amyloid-positive and tau-discordant groups. In BioFINDER-2, we found that cognition in both amyloid-positive and tau-discordant groups was worse than in controls across all domains and plasma biomarkers; whereas in TRIAD, only memory for the tau-discordant group was impaired when T₁ = p-tau₂₁₇ or T₁ = p-tau₁₈₁. Few studies have looked at cognitive deficits from a perspective that classified participants based on tau discordance.^29,74,96-98 Rather than investigating static cognition, these studies focused on the rate of cognitive decline based on the ATN profile. All studies exposed that the only group whose cognition declined significantly was the tau-concordant group. Tau-discordant (plasma-positive or CSF-positive and PET-negative) groups did not manifest significant longitudinal cognitive decline in these studies. A recent study has shown that the ability of plasma p-tau₂₁₇ to detect cognitive decline is inferior to that of tau PET.⁹⁹ Our results confirmed that the tau-positive group is the most cognitively strained and additionally revealed that subtle cognitive deficits, memory more prominently, are perceptible in patients who do not have a widely deleterious load of tangles in the brain.

Our latter findings have a few connotations. First, they suggest that there seems to be a gradient of cognitive impairment across the AD spectrum that goes along with the occurrence of neuropathology. By examining several cognitive domains, we were able to observe how different cognitive deficits were distinctly associated with biomarker profiles. Cognitive impairment might already be present in participants with AD pathology who have no significant amount of tau in the brain, irrespective of whether this pathology is defined by amyloid positivity or both plasma p-tau and amyloid positivity. Thus, our study supports previous claims suggesting that p-tau biomarkers and tau PET biomarkers are not interchangeable^29,100,101 and, in turn, the new criteria for the diagnosis and staging of AD.³¹ Mattsson-Carlgren et al.,¹⁰¹ for instance, claim that CSF p-tau₁₈₁ and p-tau₂₁₇ positivity precede tau PET positivity by several years, and that they mediate the relationship between amyloid PET and tau PET.¹⁰⁰ Our study is less clear about the differentiation between plasma p-tau positivity and amyloid PET positivity with regard to cognition, because differences between amyloid-positive and tau-discordant groups were not significant. Therefore, our findings support the split of ‘T’ into T₁ and T₂ but are inconclusive regarding the contrast between amyloid positivity and plasma p-tau positivity.

Second, memory deficits arise as the only cognitive impairment in A+T₁+T₂− individuals in both cohorts, which is unsurprising, because memory decline has classically been pinpointed as one of the earliest cognitive symptoms of AD.^18,102-104 Importantly, given that the A−T₁+ group did not display significant cognitive impairment when compared with the control group, one could argue that it is the combination of amyloid and tau that drives cognitive deficits in AD, which is in line with well-established research.¹⁰⁵ Finally, it is noteworthy to emphasize that statistically significant memory deficits were not apparent for the A+T₁+T₂− group when determining p-tau positivity with plasma p-tau₂₃₁ in the TRIAD cohort. However, the larger sample size of BioFINDER-2 seemed enough to pick up a significant effect that was missing in TRIAD. Scientific work over the past few years has consistently suggested that plasma p-tau₂₃₁ is an earlier biomarker that reflects amyloid PET.^{32,88,89,106,107} If that is the case, it would be reasonable that cognition was not noticeably affected in people who are amyloid and p-tau₂₃₁ positive but not tau PET positive, because research has reliably shown that cognitive decline relates to tau rather than to amyloid.¹⁶ In summary, our results align, in part, with recent studies on staging of plasma p-tau biomarkers and support the idea that cognitive changes are likely to co-occur with anomalies in plasma p-tau biomarkers.

Another interesting observation regarding ROC curves is that the threshold that indicates memory impairment for p-tau₂₁₇ and p-tau₁₈₁ was higher than the threshold for positivity of these biomarkers, whereas the threshold for memory impairment for tau-PET and p-tau₂₃₁ was lower than the one to reach positivity in the TRIAD cohort. In BioFINDER-2, a lower threshold for positivity when compared with the threshold that signals memory impairment was found only for plasma p-tau₂₁₇, and the cut-off for positivity in all other biomarkers was higher than the cut-off indicating memory impairment. In other words, people with memory impairment were likely to be p-tau₂₁₇ positive (or p-tau₁₈₁ positive in TRIAD), whereas memory impairment was compatible with being tau PET negative. Results for p-tau₁₈₁ (in BioFINDER-2) and p-tau₂₃₁ thresholds might be more intricate, but it might be explained by the poor specificity of these biomarkers to detect memory impairment [TRIAD: p-tau₂₃₁ specificity = 0.64, 95% CI (0.58, 0.93); BioFINDER-2: p-tau₁₈₁ specificity = 0.68, 95% CI (0.54, 0.75); p-tau₂₃₁ specificity = 0.74, 95% CI (0.68, 0.81)].

Our study has several limitations. First, linear regression models do not contain global amyloid as a covariate. When global amyloid PET was included (analysis not included in this manuscript), the significance of p-tau biomarkers disappeared. However, when tau PET was excluded from a model with amyloid PET (analysis not included either), p-tau₂₁₇ reappeared as a significant player. This highlights the strong associations between tau phosphorylation with amyloid and tau brain aggregation. The goal of the analyses in our study was to present a proof-of-concept portrait. In the real world, many variables might potentially be interacting with each other. In this case, we are hinting at a mediation on the relationship between cognitive scores and p-tau biomarkers. Moreover, a linear relationship, which is what some of our analyses illustrated, does not imply causality, and it does not rule out the possibility that this relationship might not be linear. It is also important to note that the discrepancies between cohorts might not be attributable only to the larger sample size of the BioFINDER-2 cohort, but also to the use of different biomarkers to assess plasma p-tau, tau PET and amyloid positivity, the use of different cognitive tests and, importantly, diverging strategies both to recruit individuals and to classify them into different diagnostic categories. Regarding the latter, the proportion of cognitively impaired individuals in the A+T₁−T₂− group is much higher in the BioFINDER-2 cohort than in the TRIAD cohort. The most conspicuous differences in terms of results concern AT₁T₂ profiles, where another potential source of discordance might stem from discrepancies in the definition of amyloid positivity. Numerous studies have shown that amyloid CSF undergoes changes prior to amyloid PET,¹⁰⁸ and this may explain, in part, why A+T₁−T₂− individuals from the BioFINDER-2 cohort are cognitively impaired across domains. An earlier threshold for amyloid positivity might include individuals who are less prone to show cognitive impairment, but it might also enlarge the sample size for the amyloid-positive group, potentially making the control group more cognitively homogeneous, while decreasing the chances that amyloid-positive individuals are also p-tau positive. Other studies hint at different neuropathological pathways based on discordant amyloid CSF and PET profiles.¹⁰⁹ Lastly, caution is advised concerning the diagnostic value of plasma p-tau forms owing to different issues, including a need for further research on plasma p-tau biomarkers.^64,110,111 Nevertheless, plasma p-tau₂₁₇ has been demonstrated to perform at the level of well-established CSF p-tau biomarkers,⁶⁵ and the recently reviewed criteria by the Alzheimer’s Association Workgroup support the use of sensitive plasma p-tau biomarkers for diagnostic and staging purposes.³¹ Our manuscript did not examine diagnostic potential, but supports the revised framework by showing that different cognitive profiles emerge in relationship to different biomarker profiles.

Conclusion

In conclusion, our findings indicate that plasma p-tau₂₁₇ stands out as a biomarker in close association with memory impairment attributable to AD. Incorporating plasma p-tau₂₁₇ and p-tau₁₈₁ into the Alzheimer’s Association Workgroup framework as Core 1 biomarkers might have multiple benefits, including a more precise categorization of patients for planning clinical trials, in addition to enriching clinical assessments.

Data availability

The data presented in this study are available from the corresponding author upon reasonable request, and such arrangements are subject to standard data-sharing agreements.

Funding

The project that gave rise to these results received the support of a fellowship from ‘la Caixa’ Foundation (ID 100010434). The fellowship code is LCF/BQ/EU21/11890154. The Translational Biomarkers in Aging and Dementia (TRIAD) is supported by the Weston Brain Institute, Canadian Institutes of Health Research (grants MOP-11-51-31, RFN 152985, 159815 and 162303); Canadian Consortium on Neurodegeneration in Aging (grant MOP-11-51-31-team 1); Brain Canada Foundation (Canadian Foundation for Innovation Project grants 34874 and 33397) and the Fonds de Recherche du Quebec Sante (grant 2020-VICO-279314 TRIAD/BIOVIE Cohort). Pedro Rosa-Neto is supported by the Weston Brain Institute, the Fonds de Recherche du Quebec Sante (grant Chercheur Boursier), the Canadian Institutes of Health Research (CIHR) and the Canadian Consortium on Neurodegeneration in Aging (CCNA).

Competing interests

G.S. has received speaker fees from Springer and Adium. G.T.-B. receives salary and stock from Johnson & Johnson. He also holds patents for Janssen Simoa assays CSF p217+Tau, Plasma p217+Tau, and plasma CD tTau assays, respectively. S.P. has acquired research support (for the institution) from Avid and ki elements through ADDF. In the past 2 years, he has received consultancy/speaker fees from BioArtic, Biogen, Eisai, Eli Lilly, Novo Nordisk, and Roche. H.Z. has served at scientific advisory boards and/or as a consultant for AbbVie, Acumen, Alector, Alzinova, ALZpath, Amylyx, Annexon, Apellis, Artery Therapeutics, AZTherapies, Cognito Therapeutics, CogRx, Denali, Eisai, Merry Life, NervGen, Novo Nordisk, OptoCeutics, Passage Bio, Pinteon Therapeutics, Prothena, Red Abbey Labs, reMYND, Roche, Samumed, Siemens Healthineers, Triplet Therapeutics, and Wave; has given lectures in symposia sponsored by AlzeCure, Biogen, Cellectricon, Fujirebio, Lilly, Novo Nordisk, and Roche; and is a co-founder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is a part of the GU Ventures Incubator Program (outside submitted work). K.B. has served as a consultant and at advisory boards for AbbVie, AC Immune, ALZpath, AriBio, Beckman Coulter, BioArctic, Biogen, Eisai, Lilly, Moleac Pte Ltd, Neurimmune, Novartis, Ono Pharma, Prothena, Quanterix, Roche Diagnostics, Sanofi and Siemens Healthineers; has served at data monitoring committees for Julius Clinical and Novartis; has given lectures, produced educational materials, and participated in educational programmes for AC Immune, Biogen, Celdara Medical, Eisai, and Roche Diagnostics; and is a co-founder of Brain Biomarker Solutions in Gothenburg AB (BBS), which is a part of the GU Ventures Incubator Program, outside the work presented in this paper. O.H. is an employee of Eli Lilly and Lund University. The other authors have no competing interests to disclose.

Supplementary material

Supplementary material is available at Brain online.