Brain age correlates with plasma NfL in amyloid positive individuals

Saurabh Sihag, Alejandro RibeiroAlzheimer 2024 · DOI

TL;DR

This study uses a machine learning model to predict "brain age" from cortical thickness data and finds that this predicted brain age is positively correlated with plasma neurofilament light (NfL) levels, but only in individuals who are amyloid-positive. This suggests that amyloid status plays a crucial role in how brain aging markers relate to neuronal damage indicators.

What Problem Does This Paper Address?

Understanding the neurobiological underpinnings of aging, particularly accelerated or premature aging, is a critical challenge in neuroscience. "Brain age gap" – the difference between a person's predicted brain age and their chronological age – is a promising metric for assessing this accelerated aging. Simultaneously, plasma neurofilament light (NfL) is emerging as a key biomarker for neuronal damage and white matter integrity loss, often seen in neurodegenerative diseases like Alzheimer's. However, the relationship between these two markers, especially in the context of underlying Alzheimer's pathology (like amyloid accumulation), remains unclear. This paper aims to bridge this gap by investigating how brain age, predicted via machine learning, correlates with plasma NfL levels, considering amyloid status as a potential mediator.

Methods

The researchers employed an "anatomically interpretable ML pipeline" for brain age prediction. Specifically, they used cortical thickness measures from the Desikan-Killiany atlas, derived from MRI scans. The machine learning model used for brain age prediction was pre-trained on a dataset of healthy individuals from the OASIS-3 dataset. This model then predicted brain age for individuals in the ADNI cohort. The "brain age gap" was calculated as the difference between the predicted brain age and the individual's chronological age. The study then assessed the correlation between this brain age gap and plasma NfL levels, stratifying the analysis by amyloid status.

Datasets

  • ADNI (Alzheimer's Disease Neuroimaging Initiative) cohort: This was the primary dataset used for the analysis. It included:
  • 206 healthy controls (HC)
    • 372 individuals with Mild Cognitive Impairment (MCI)
    • 118 individuals with Dementia (AD)
  • OASIS-3 dataset: This dataset was used to train the machine learning model for brain age prediction. It comprised 611 healthy individuals.

Key Insights and Ideas

  • Amyloid status is a critical moderator: The most striking finding is that the correlation between brain age and plasma NfL is highly dependent on amyloid status. This suggests that amyloid pathology might be a necessary condition for plasma NfL to reflect accelerated brain aging.
  • Brain age is elevated in AD: As expected, individuals with dementia (AD) showed a significantly higher predicted brain age compared to healthy controls, reinforcing the validity of the brain age metric in this context.
  • Entorhinal cortex contribution: In healthy, amyloid-positive individuals, the entorhinal region's contribution to the predicted brain age showed a particularly strong positive correlation with plasma NfL. This highlights specific brain regions that might be particularly vulnerable to the combined effects of aging and amyloid pathology.

Key Results

  • Brain age elevation in AD: The AD cohort exhibited significantly higher brain age compared to the HC cohort.
  • No difference in brain age gap by amyloid status: The study found no significant difference in the brain age gap between amyloid-positive and amyloid-negative individuals across all cohorts.
  • Correlation in amyloid-positive HC: For cognitively healthy individuals who were amyloid-positive, there was a significant positive correlation between brain age gap and plasma NfL (ρ=0.509, p-val = 0.002). The entorhinal region's contribution to brain age was strongly correlated (ρ=0.72, p=2.45e-6).
  • No correlation in amyloid-negative HC: In contrast, cognitively healthy individuals who were amyloid-negative showed no significant correlation between brain age gap and plasma NfL.
  • Correlation in amyloid-positive MCI/AD: In the combined MCI and AD cohorts, brain age was correlated with plasma NfL only in amyloid-positive individuals (ρ=0.352, p-val = 0.003).

Limitations and Future Work

  • Cross-sectional data: The study relies on cross-sectional data, which limits the ability to establish causality or track changes over time.
  • Subsets for NfL: Plasma NfL data was available for only a subset of individuals in each cohort, which might affect the generalizability of findings.
  • Future Work: The authors suggest that amyloid status may mediate the association between plasma NfL and brain age. Future research could explore this mediation effect more formally and investigate longitudinal changes to better understand the dynamic interplay between these biomarkers and disease progression. Further investigation into the specific regional contributions to brain age in different pathological contexts is also warranted.

Summary updated

Abstract

Abstract Background Brain age prediction using machine learning (ML) models provides novel opportunities to characterize neurobiology of accelerated aging. Brain age gap, i.e., the gap between the predicted age and the chronological age for an individual, is a meaningful metric to assess accelerated or premature aging. Plasma neurofilament light (NfL) is a well‐recognized biomarker that can reflect the severity of structural atrophy and the loss of white matter integrity in Alzheimer's disease. Here, we leveraged an anatomically interpretable ML pipeline for brain age prediction (see ref. [1]) to investigate the correlations between brain age and plasma neurofilament light chain (NfL) in older adults in the ADNI cohort. Method Brain age was derived from the cortical thickness measures (Desikan‐Killiany atlas, available online as part of the ADNI database) for 206 healthy individuals (HC), 372 individuals with MCI diagnosis, and 118 individuals with dementia diagnosis (AD). Further, contributions from individual brain regions to the predicted brain age were characterized for each individual. The amyloid status was available for all individuals and plasma NfL (within 30 days of the MRI scan acquisition) was available for a subset of individuals (Table 1). The ML model used to derive brain age had been trained on the healthy individuals from the OASIS‐3 dataset. We assessed the associations between brain age and plasma NfL for HC and the combined MCI and AD cohorts. Result Brain age was significantly elevated in AD cohort with respect to HC cohort (Table 1, Figure 1). No significant differences were found between the brain age gap for amyloid positive and amyloid negative individuals (Figure 1). For HC cohort, brain age was positively correlated with plasma NfL in amyloid positive individuals (with brain age contributions from entorhinal region exhibiting the most significant positive correlation (ρ=0.72, p=2.45e‐6) with plasma NfL), but not amyloid negative individuals (Figure 2). In the combined MCI and AD cohort, brain age was correlated with plasma NfL in only amyloid positive individuals. Conclusion Our results suggest that amyloid status may mediate the association between plasma NfL and brain age in older individuals. [1] Sihag et al., Explainable Brain Age Prediction using coVariance Neural Networks, NeurIPS, 2023.
BibTeX
@inproceedings{sihag2024brain,
  author = {Saurabh Sihag and Alejandro Ribeiro},
  title = {{Brain age correlates with plasma NfL in amyloid positive individuals}},
  booktitle = {Alzheimer},
  year = {2024},
  doi = {10.1002/alz.089232},
}
Figure