Plasma biomarkers for Alzheimer's Disease in relation to neuropathology and cognitive change

Abstract

Plasma biomarkers related to amyloid, tau, and neurodegeneration (ATN) show great promise for identifying these pathological features of Alzheimer's Disease (AD) as shown by recent clinical studies and selected autopsy studies. We have evaluated ATN plasma biomarkers in a series of 312 well-characterized longitudinally followed research subjects with plasma available within 5 years or less before autopsy and examined these biomarkers in relation to a spectrum of AD and related pathologies. Plasma Aβ42, Aβ40, total Tau, P-tau181, P-tau231 and neurofilament light (NfL) were measured using Single molecule array (Simoa) assays. Neuropathological findings were assessed using standard research protocols. Comparing plasma biomarkers with pathology diagnoses and ratings, we found that P-tau181 (AUC = 0.856) and P-tau231 (AUC = 0.773) showed the strongest overall sensitivity and specificity for AD neuropathological change (ADNC). Plasma P-tau231 showed increases at earlier ADNC stages than other biomarkers. Plasma Aβ42/40 was decreased in relation to amyloid and AD pathology, with modest diagnostic accuracy (AUC = 0.601). NfL was increased in non-AD cases and in a subset of those with ADNC. Plasma biomarkers did not show changes in Lewy body disease (LBD), hippocampal sclerosis of aging (HS) or limbic-predominant age-related TDP-43 encephalopathy (LATE) unless ADNC was present. Higher levels of P-tau181, 231 and NfL predicted faster cognitive decline, as early as 10 years prior to autopsy, even among people with normal cognition or mild cognitive impairment. These results support plasma P-tau181 and 231 as diagnostic biomarkers related to ADNC that also can help to predict future cognitive decline, even in predementia stages. Although NfL was not consistently increased in plasma in AD and shows increases in several neurological disorders, it had utility to predict cognitive decline. Plasma Aβ42/40 as measured in this study was a relatively weak predictor of amyloid pathology, and different assay methods may be needed to improve on this. Additional plasma biomarkers are needed to detect the presence and impact of LBD and LATE pathology.

Keywords: Alzheimer’s Disease; Biomarker; Dementia; Neuropathology; Plasma.

Fig. 1

Plasma biomarkers at last blood draw by pathologic groups. Boxplots of the distributions of the plasma biomarkers from the blood draw closest to death by pathologic group. One NfL value of 1154 in FTLD patient was removed from plots for visualization but retained in statistical analyses. Effect sizes and both raw and multiple-comparisons adjusted p values are available in Supplementary Table 2, On-line Resource. Statistics for pairwise comparisons are corrected for multiple comparisons using Tukey’s method to maintain a family error rate of 0.05, and are graphically summarized as follows: *p < 0.05, **p < 0.01, ***p < 0.001

Fig. 2

Plasma biomarkers in groups defined by different staging of AD Neuropathology. Boxplots of the distributions of the plasma biomarkers from the blood draw closest to death divided by a CERAD neuritic plaque density score, b Braak neurofibrillary tangle stage, and c NIA-Reagan Institute criteria stage of ADNC. One NfL value of 1154 in FTLD patient was removed from plots for visualization, but retained in statistical analyses. Effect sizes and both raw and multiple-comparisons adjusted p values are available in Supplementary Table 3, On-line Resource. Statistics for pairwise comparisons are corrected for multiple comparisons using Tukey’s method to maintain a family error rate of 0.05, and are graphically summarized as follows: *p < 0.05, **p < 0.01, ***p < 0.001

Fig. 3

Plasma biomarkers in relation to Hippocampal sclerosis (HS), Limbic Age-related TDP-43 Encephalopathy (LATE), Lewy Body Disease (LBD) and ADNC. Boxplots of the distributions of the plasma biomarkers from the blood draw closest to death in individuals with High ADNC and/or other non-AD pathologies: a hippocampal sclerosis of aging defined as neuronal loss in the CA1 and subiculum out of proportion with the degree of AD pathology, b hippocampal staining positive for TDP-43 proteinopathy representing LATE neuropathologic changes (LATE-NC), and c Lewy body disease of the limbic (transitional) or neocortical (diffuse) type. These plots and analyses exclude participants who did not have either High ADNC or the non-AD pathology being assessed. TDP-43 immunostaining was available in a select subset of cases, with their demographic data available in Supplementary Table 4, On-line Resource. One NfL value of 1154 in a FTLD patient was removed from plots for visualization, but retained in statistical analyses. Effect sizes and both raw and multiple-comparisons adjusted p values are available in Supplementary Table 5, On-line Resource. Statistics for pairwise comparisons are corrected for multiple comparisons using Tukey’s method to maintain a family error rate of 0.05, and are graphically summarized as follows: *p < 0.05, **p < 0.01, ***p < 0.001

Fig. 4

ROC analyses, comparing plasma biomarkers in the Low Pathology group vs the last blood draw in the High ADNC group. ROC curves and associated thresholds, specificities, sensitivities, and areas under the curve (AUCs) for the use of each plasma biomarker to distinguish patients who were classified as Low Pathology at autopsy from those who were classified as High ADNC. Because of the older age and stability of plasma biomarkers from baseline to last blood draw in the Low Pathology group, baseline plasma biomarkers in this group were compared to the final blood draw High ADNC group to achieve closer age matching

Fig. 5

Longitudinal changes of plasma biomarkers in relation to ADNC. Longitudinal progression in biomarkers in the 10 years prior to death in all study participants were divided by their degree of ADNC. Horizontal dashed lines represent the thresholds derived from ROC analyses presented in Fig. 4. Thick lines represent predictions of the trajectories of the biomarkers for a demographically average participant, derived from mixed effects models with covariates added for age, sex, interval from last visit to death, as well as each variable’s interaction with time. All models included random intercepts and slopes by participant. A version of this figure and analysis excluding individuals with concomitant non-AD pathologies is available in Supplementary Fig. 2, On-line Resource

Fig. 6

Plasma P-tau and NfL biomarkers and longitudinal cognitive change in relation to AD pathology (excluding FTLD, HS, LBD, and Other pathologies). Longitudinal progression on the Dementia Rating Scale (DRS) in the 5-year interval from baseline in study participants divided by their a degree of ADNC, or b–d baseline plasma biomarker levels, after excluding FTLD, HS, LBD, and other significant pathologies. Cutoffs for each biomarker are those derived from ROC analyses presented in Fig. 4. Thick lines represent predictions of the trajectories of the DRS for a demographically average participant, derived from mixed effects models with covariates added for age, sex, interval from last visit to death, education as well as each variable’s interaction with time. To account for different starting levels of impairment the baseline DRS score was included as an interaction with time. All models included random intercepts and slopes by participant. Statistics for Exponential time term by biomarker interaction: NIA-Reagan Low vs Int, p = 0.24, Low vs High p = 9.4 × 10^–12, pTau181 p = 6.6 × 10^–7, pTau231 p = 0.0022, NfL p = 0.021

Fig. 7

ADNC and baseline plasma P-tau and NfL biomarkers and longitudinal cognitive change in subjects with normal cognition or MCI at baseline. Longitudinal progression on the Dementia Rating Scale (DRS) in the 5-year interval from baseline in participants with normal cognition of mild cognitive impairment (MCI) divided by their a degree of ADNC, or b–d baseline plasma biomarker levels, after excluding FTLD, HS, LBD, and other significant pathologies. Cutoffs for each biomarker are those derived from ROC analyses presented in Fig. 4. Thick lines represent predictions of the trajectories of the DRS for a demographically average participant, derived from mixed effects models with covariates added for age, sex, interval from last visit to death, education as well as each variable’s interaction with time. To account for different starting levels of impairment the baseline DRS score was included as an interaction with time. All models included random intercepts and slopes by participant. Statistics for exponential time term by biomarker interaction: NIA-Reagan Low vs Int, p = 0.66, Low vs High p = 0.0074, pTau181 p = 0.0019, pTau231 p = 0.044, NfL p = 0.00524

Fig. 8

Plasma biomarker Z-scores and Braak stage. Local regression curves derived by locally estimated scatterplot smoothing of the Z-score of each plasma biomarker from the blood draw closest to death across Braak stage. The individual biomarker Z-scores were derived by setting the mean of the distribution to 0 and its standard deviation to 1