Comparative neurofilament light chain trajectories in CSF and plasma in autosomal dominant Alzheimer's disease

Abstract

Disease-modifying therapies for Alzheimer's disease (AD) are likely to be most beneficial when initiated in the presymptomatic phase. To track the benefit of such interventions, fluid biomarkers are of great importance, with neurofilament light chain protein (NfL) showing promise for monitoring neurodegeneration and predicting cognitive outcomes. Here, we update and complement previous findings from the Dominantly Inherited Alzheimer Network Observational Study by using matched cross-sectional and longitudinal cerebrospinal fluid (CSF) and plasma samples from 567 individuals, allowing timely comparative analyses of CSF and blood trajectories across the entire disease spectrum. CSF and plasma trajectories were similar at presymptomatic stages, discriminating mutation carriers from non-carrier controls 10-20 years before the estimated onset of clinical symptoms, depending on the statistical model used. However, after symptom onset the rate of change in CSF NfL continued to increase steadily, whereas the rate of change in plasma NfL leveled off. Both plasma and CSF NfL changes were associated with grey-matter atrophy, but not with Aβ-PET changes, supporting a temporal decoupling of Aβ deposition and neurodegeneration. These observations support NfL in both CSF and blood as an early marker of neurodegeneration but suggest that NfL measured in the CSF may be better suited for monitoring clinical trial outcomes in symptomatic AD patients.

Fig. 1. Association between CSF and plasma NfL.

Plasma log₁₀(NfL) levels were associated with CSF log₁₀(NfL) at baseline (a) and longitudinally (b) in mutation carriers (MC; red) but less so in the non-carriers (NC; grey). There were 274 MC (149 Females; 125 Males) and 162 NC (90 Females; 72 Males) with concurrent baseline CSF and plasma NfL measurements. Within-person annualised rate of change in NfL was extracted from linear mixed effects models (see “Methods” section) for 146 MC and 88 NC with concurrent longitudinal CSF and plasma NfL available. Cross-sectional and longitudinal associations are presented with scatteplot showing unadjusted linear relationship (red and grey bolded lines) between plasma and CSF log₁₀(NfL). The shaded area around each unadjusted linear fit line represents the 95% confidence interval. See Supplementary Table 3 for associations between CSF and plasma log10(NfL) levels after adjusting for baseline age, sex, and baseline BMI.

Fig. 2. CSF and plasma NfL trajectories across the disease course.

Baseline plasma (a) and CSF (b) log₁₀(NfL) concentrations in mutation carriers (MC; red) begin to increase, relative to non-carriers (NC; grey), around 10-15 years prior to estimated symptom onset. Similarly, within-person rate of change in plasma (c) and CSF (d) log₁₀(NfL) levels in MC begin to increase, relative to NC, around 15-20 years prior to estimated symptom onset. The curves and credible intervals are drawn from the actual distributions of model fits derived by the Hamiltonian Markov chain Monte Carlo analyses (see “Methods” section). The shaded areas represent the 95% credible intervals around the model estimates. The first point in the disease course (using estimated years to symptom onset) where NC and MC differed was determined to be the first point where the 95% credible intervals around the difference distribution between NC and MC did not overlap (see Supplementary Figs. 1 and 2 for corresponding longitudinal spaghetti plots and difference distribution plots).

Fig. 3. Association between rate of change in NfL and AD clinical groupings.

Rate of change per year in plasma and CSF log₁₀(NfL) across non-carriers (NC; grey, plasma n = 88; CSF n = 89); Presymptomatic (Presym) mutation carriers (MC; yellow, plasma n = 79; CSF n = 82; individuals with CDR = 0 across all visits); Converters (orange, plasma n = 13; CSF n = 13; MC with CDR = 0 at baseline and CDR > 0 at all subsequent visits); Symptomatic (Sym) MC (red, plasma n = 48; CSF n = 50; MC with CDR > 0 across all visits). (a) Annualised rate of change of plasma log₁₀(NfL). Presym MC had a significantly higher rate of change compared to NC. Converters had significantly higher rate of change compared to both NC and presym MC. Sym MC had significantly higher rates of change compared to NC, presym MC and converters. (b) Annualised rate of change of CSF log₁₀(NfL). Presym MC had a significantly higher rate of change compared to NC. Converters had significantly higher rate of change compared to both NC and presym MC. Sym MC had significantly higher rates of change compared to NC, presym MC, and converters. (c) Ratio of absolute plasma to CSF NfL levels. Sym MC had a significantly lower ratio compared to NC and presym MC. NC, Presym MC, and Converters had similar plasma/CSF ratios. The boxes map to the median, 25th and 75th quintiles, and the whiskers extend to 1.5 × interquartile range (IQR). The violin plots illustrate kernel probability density (i.e. the width of the shaded area represents the proportion of the data located there). Comparisons were done with linear mixed effects models adjusting for baseline age, baseline BMI, and sex. Corresponding unstandardized beta estimates, standard errors, and multiple comparison corrected exact p-values are reported in Supplementary Table 4. n.s. > 0.05; *p < 0.05, **p < 0.01, ***p < 0.001.

Fig. 4. Longitudinal association between NfL, brain atrophy, and amyloid deposition.

a, b Relationship between within-person rate of change in log₁₀(NfL) and precuneus grey matter volume for plasma (a) and CSF (b). Linear mixed effects models were adjusted for baseline age*time, baseline BMI*time, and sex*time. Results revealed a significant association between rate of change in precuneus grey matter and rate of change in plasma and CSF log₁₀(NfL) in the Sym MC group (plasma: n = 58 and CSF: n = 60), but not in NC (plasma: n = 81 and CSF: n = 82) or Presym MC (plasma: n = 76 and CSF: n = 78). c, d Relationship between within-person rate of change in log₁₀(NfL) and rate of change in precuneus PiB-PET. There were no significant associations between longitudinal PiB-PET and log₁₀(NfL) within NC (plasma: n = 89 and CSF: n = 65), Presym MC (plasma: n  =  81 and CSF: n = 60), and Sym MC (plasma: n = 64 and CSF: n = 47). Shown are non-carrier (NC; grey square), presymptomatic (Presym) mutation carriers (MC; yellow circle), and symptomatic (Sym) MC (including converters to the symptomatic phase, red triangle). The shaded area around each unadjusted linear fit line represents the 95% confidence interval. Solid linear fit line represents a significant association (p < 0.05); dashed linear fit line represents non-significant association (p > 0.05). Corresponding unstandardized beta estimates, standard errors, and exact p-values are reported in Supplementary Table 5. Note that not all participants with longitudinal NfL measurements had imaging parameters available.