Comparing positron emission tomography imaging and cerebrospinal fluid measurements of β-amyloid

Abstract

Objective: We examined agreement and disagreement between 2 biomarkers of β-amyloid (Aβ) deposition (amyloid positron emission tomography [PET] and cerebrospinal fluid [CSF] Aβ1-42 ) in normal aging and dementia in a large multicenter study.

Methods: Concurrently acquired florbetapir PET and CSF Aβ were measured in cognitively normal, mild cognitive impairment (MCI), and Alzheimer's disease participants (n = 374) from the Alzheimer's Disease Neuroimaging Initiative. We also compared Aβ measurements in a separate group with serial CSF measurements over 3.1 ± 0.8 years that preceded a single florbetapir session. Additional biomarker and cognitive data allowed us to further examine profiles of discordant cases.

Results: Florbetapir and CSF Aβ were inversely correlated across all diagnostic groups, and dichotomous measurements were in agreement in 86% of subjects. Among subjects showing the most disagreement, the 2 discordant groups had different profiles: the florbetapir(+) /CSF Aβ(-) group was larger (n = 13) and was made up of only normal and early MCI subjects, whereas the florbetapir(-) /CSF Aβ(+) group was smaller (n = 7) and had poorer cognitive function and higher CSF tau, but no ApoE4 carriers. In the longitudinal sample, we observed both stable longitudinal CSF Aβ trajectories and those actively transitioning from normal to abnormal, but the final CSF Aβ measurements were in good agreement with florbetapir cortical retention.

Interpretation: CSF and amyloid PET measurements of Aβ were consistent in the majority of subjects in the cross-sectional and longitudinal populations. Based on our analysis of discordant subjects, the available evidence did not show that CSF Aβ regularly becomes abnormal prior to fibrillar Aβ accumulation early in the course of disease.

Figure 1

The inverse association between florbetapir cortical retention ratios and CSF Aβ_1-42 is shown for normal, EMCI, LMCI, and AD individuals (a). Predefined cutoffs are shown each marker (CSF Aβ_1-42 =192 pg/mL; florbetapir = 1.11) that were derived from independent samples (see Methods). Subjects with concordant florbetapir and CSF Aβ_1-42 are in the upper left (florbetapir−/CSF Aβ−) and bottom right (florbetapir+/CSF Aβ+) quadrants while subjects with discordant florbetapir and CSF Aβ_1-42 are in the upper right (florbetapir+/CSF Aβ−) and bottom left (florbetapir−/CSF Aβ+) quadrants. A florbetapir scan for an example discordant LMCI subject (florbetapir−/CSF Aβ+) is shown (b; see asterisk on scatterplot in A), indicating that the visual read is consistent with the qualitative florbetapir measurement (florbetapir cortical retention ratio = 0.98) despite abnormal CSF status. The percent of individuals from each diagnostic group in each of the 4 scatterplot quadrants is shown in the bar graph (c). The proportion of subjects who are abnormal on both markers (black bars) increases as diagnostic severity increases, but the proportions of discordant subjects (grey and striped bars) is similar across diagnostic groups and between the two types of discordance. The proportion of ApoE4 allele carriers who are concordant on both markers increases with diagnostic severity (d; black bars), the proportion of ApoE4 carriers is moderate for the two discordant groups (grey and striped bars) in the Normal and EMCI subjects.

Figure 2

Subjects who remain discordant after applying a +/− 5% confidence interval to each cutoff.

Figure 3

Longitudinal CSF Aβ_1-42 data is plotted against time for each subject in the longitudinal sample (a), with florbetapir+ individuals in the top row and florbetapir− subjects in the bottom row. Subjects are plotted separately by diagnosis at the time of florbetapir (left column, 27 Normal; middle column, 17 MCI; right column, 16 AD). Time of zero corresponds to the florbetapir scan, each colored line corresponds to an individual subject, and each point on the line corresponds to a CSF Aβ_1-42 value from a single LP. Dotted lines in each panel represent the CSF Aβ_1-42 cutoff value (192 pg/mL) that divides abnormal values (below line) from normal values (above line). In the top panel, CSF Aβ values that are concordant with florbetapir appear below the dotted line (both abnormal), while in the bottom panel CSF Aβ values that are concordant with florbetapir appear above the dotted line (both normal). Representative florbetapir scans are shown for three discordant subjects: (b) a CSF Aβ+/florbetapir− normal 80 yo male (florbetapir = 1.06, labeled “1” on plot); (c) a CSF Aβ−/florbetapir+ 84 yo MCI male (florbetapir cortical retention ratio = 1.12, labeled “2” on plot); and (d) a CSF Aβ+/florbetapir− 81 yo AD male (florbetapir = 0.99, labeled “3” on plot). All longitudinal CSF samples for an individual subject were included in the same immunoassay analytical run to minimize variance due to run to run and reagent lot to lot variabilities.