Alzheimer's disease profiled by fluid and imaging markers: tau PET best predicts cognitive decline

Abstract

For early detection of Alzheimer's disease, it is important to find biomarkers with predictive value for disease progression and clinical manifestations, such as cognitive decline. Individuals can now be profiled based on their biomarker status for Aβ42 (A) or tau (T) deposition and neurodegeneration (N). The aim of this study was to compare the cerebrospinal fluid (CSF) and imaging (PET/MR) biomarkers in each ATN category and to assess their ability to predict longitudinal cognitive decline. A subset of 282 patients, who had had at the same time PET investigations with amyloid-β and tau tracers, CSF sampling, and structural MRI (18% within 13 months), was selected from the ADNI dataset. The participants were grouped by clinical diagnosis at that time: cognitively normal, subjective memory concern, early or late mild cognitive impairment, or AD. Agreement between CSF (amyloid-β-1-42(A), phosphorylated-Tau181(T), total-Tau(N)), and imaging (amyloid-β PET (florbetaben and florbetapir)(A), tau PET (flortaucipir)(T), hippocampal volume (MRI)(N)) positivity in ATN was assessed with Cohen's Kappa. Linear mixed-effects models were used to predict decline in the episodic memory. There was moderate agreement between PET and CSF for A biomarkers (Kappa = 0.39-0.71), while only fair agreement for T biomarkers (Kappa ≤ 0.40, except AD) and discordance for N biomarkers across all groups (Kappa ≤ 0.14) was found. Baseline PET tau predicted longitudinal decline in episodic memory irrespective of CSF p-Tau181 positivity (p ≤ 0.02). Baseline PET tau and amyloid-β predicted decline in episodic memory (p ≤ 0.0001), but isolated PET amyloid-β did not. Isolated PET Tau positivity was only observed in 2 participants (0.71% of the sample). While results for amyloid-β were similar using CSF or imaging, CSF and imaging results for tau and neurodegeneration were not interchangeable. PET tau positivity was superior to CSF p-Tau181 and PET amyloid-β in predicting cognitive decline in the AD continuum within 3 years of follow-up.

Fig. 1. Amyloid-β (A)/tau (T) profiles measured using CSF and imaging biomarkers across diagnostic groups.

A CSF biomarkers. B PET biomarkers. With PET imaging, A+T+ profiles were less prevalent in the CN and SMC groups, and A-T+ profiles were less prevalent in the CN, SMC, EMCI, and LMCI groups, than with CSF. Sunburst graphs with biomarkers visualized in hierarchical fashion. CSF biomarkers, A (Amyloid-β 42) first level (C), T (pTau181) first level (D). PET biomarkers, A (Amyloid PET) first level (E), T (Tau PET) first level (F). AD = Alzheimer’s disease; CN = cognitively normal; EMCI = early MCI; LMCI = late MCI; MCI = mild cognitive impairment; SMC = subjective memory concern.

Fig. 2. Correlations between CSF and imaging (PET and MRI) biomarkers across diagnostic groups.

A = PET amyloid-β ¹⁸F-florbetapir PET and ¹⁸F-florbetaben levels; AD = Alzheimer’s disease; CN = cognitively normal; EMCI = early MCI; LMCI = late MCI; MCI = mild cognitive impairment; ROI = region of interest; SMC = subjective memory concern; T = PET tau ¹⁸F-flortaucipir level; T1 = T measured in the temporal meta-ROI; T2 = T measured in the inferior temporal cortex ROI; T3 = T measured in the entorhinal cortex ROI; N1 = neurodegeneration measured as hippocampal volume adjusted for intracranial volume.

Fig. 3. Concordance and discordance between amyloid-β, tau, and neurodegeneration (ATN) biomarkers.

Amyloid-β (panel A), tau (panel B), and neurodegeneration (panel C) concordance/discordance profiles between pairs of biomarkers (CSF vs imaging). Cohen’s Kappa statistics allowed numerical comparisons between pairs of profiles obtained using different assessment methods to obtain a coefficient that measured the degree of concordance between the methods. Agreement was defined as coefficient values >0.4 (fair agreement) ranging up to 1 (perfect agreement). Aβ amyloid-β, AD Alzheimer’s disease, CN cognitively normal, EMCI early MCI, HVa hippocampal volume adjusted for intracranial volume, LMCI late MCI, MCI mild cognitive impairment, pTau tau phosphorylated at Thr181, SMC subjective memory concern, tTau total.

Fig. 4. ATN (amyloid-β, tau and neurodegeneration) profiles composed with the six biomarkers, by diagnosis group and cognition.

The ATN profiles from the CSF biomarker results (A) were different from those from the imaging biomarker results (B); Sankey diagrams that represent the correspondence between the CSF and imaging panels in the same subject show disagreement of the ATN classification between the two panels for the CU (C) and CI (D) groups. AD = Alzheimer’s disease; CN = cognitively normal; EMCI = early MCI; LMCI = late MCI; MCI = mild cognitive impairment; SMC = subjective memory concern.

Fig. 5. Amyloid-β and tau profiles (CSF and PET together) and CSF and PET profiles (amyloid-β and tau together) tested as predictors for cognitive decline.

A Among the amyloid-β profiles, only the concordant positive profile (CSF+/PET+) showed a significant negative interaction effect on longitudinal episodic memory decline; B Among the tau profiles, only the profiles positive for imaging biomarkers, irrespective of CSF status, (CSF+/PET+, CSF-/PET+) showed a significant negative interaction effect on longitudinal episodic memory decline. C CSF profiles: there was a significant interaction between Aβ42 and time (stable or cognitive decline) irrespective of whether pTau was positive or negative; D PET profiles: ^The PET profile A-T+ was excluded from the models for this analysis since only one subject had valid data for longitudinal analysis (one more had only baseline measurements). There was a significant interaction between PET tau and time (cognitive decline) only when PET Aβ was positive (A+T+), meaning that PET amyloid-β is predictive only if PET tau is positive and, given that there were two cases (one cross-sectional and one longitudinal case) with PET A-T+ , PET tau is a preferable predictor. *p ≦ 0.05, **p ≦ 0.01. The ADNI_MEM baseline, as expected, had a substantial impact on the model estimations, and its coefficient was consequently about tenfold higher than that of the second highest factor; hence the range selected to display the factor coefficient has been tailored to show differences among the smaller factors. The full representation of the coefficient for each model tested is presented in the SI. AD = Alzheimer’s disease; CN = cognitively normal; EMCI = early MCI; LMCI = late MCI; MCI = mild cognitive impairment; SMC = subjective memory concern.