Optimal anti-amyloid-beta therapy for Alzheimer's disease via a personalized mathematical model

Wenrui Hao, Suzanne Lenhart, Jeffrey R. PetrellaPLOS Computational Biology 2022 · DOI

TL;DR

This paper introduces a novel computational framework that uses personalized mathematical models to optimize anti-amyloid-beta therapy for Alzheimer's disease (AD). By integrating individual patient biomarker data, the model simulates treatment outcomes and suggests personalized regimens to maximize cognitive benefits while minimizing side effects, offering a new paradigm for AD treatment planning.

What Problem Does This Paper Address?

Alzheimer's disease (AD) is a devastating neurodegenerative disorder affecting millions, and the recent FDA approval of aducanumab, the first disease-modifying therapy, highlights the growing importance of personalized medicine. However, significant challenges remain: the complex and variable nature of AD progression, the uncertainty surrounding the efficacy of amyloid-beta plaque removal for cognitive decline, and the need for tailored treatment strategies. Current approaches often rely on population-level data, which may not translate optimally to individual patients. This paper aims to bridge this gap by developing a personalized mathematical model that can predict disease trajectories and optimize treatment plans for individual AD patients.

Methods

The core of this research lies in a data-driven, personalized mathematical modeling approach. The authors developed a "sparse empirical cascade model" that describes AD progression through four key clinical biomarkers: amyloid-beta (Aβ) and tau pathology, neuronal loss, and cognitive impairment. This model is built upon a system of ordinary differential equations (ODEs).

Key methodological contributions include:

  • Personalized Model Calibration: The model is parameterized using longitudinal biomarker data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) cohort. This allows for the creation of a unique model for each individual patient, capturing their specific disease trajectory.
  • "Equation-by-Equation" Parameter Estimation: To address the complexity of optimizing parameters in a high-dimensional space, the authors employ an "equation-by-equation" approach. This method estimates parameters sequentially, following the cascade of AD pathology, which provides a good initial guess for the optimization algorithm and improves convergence.
  • Optimal Control Theory for Treatment Optimization: The framework integrates optimal control theory to simulate and optimize anti-amyloid-beta therapies. This allows for the incorporation of time-varying treatment controls (drug dosage) and the modeling of side effects. The objective is to minimize cognitive impairment and amyloid levels while simultaneously minimizing treatment-related side effects.
  • In-Silico Clinical Trials: The developed model is used to conduct simulated clinical trials for aducanumab and donanemab, evaluating different dosages and treatment durations (78 weeks and 10 years).

Datasets

The primary dataset used in this study is the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset. Specifically, they utilized data from ADNI-1, which includes longitudinal biomarker data (CSF, MRI, cognitive assessments) from subjects with mild cognitive impairment (MCI), early AD, and cognitively normal elderly controls. This dataset is crucial for parameterizing the personalized models for individual patients.

Key Insights and Ideas

  • Personalization is Key: The study strongly emphasizes that a one-size-fits-all approach to AD treatment is insufficient. Individualized models, informed by a patient's unique biomarker profile, are essential for optimizing therapeutic outcomes.
  • Modeling Side Effects Matters: The authors explicitly incorporate side effects of anti-amyloid-beta therapies (like ARIA) into their optimization objective. This is a critical step towards developing realistic and safe treatment plans, as side effects can significantly impact treatment adherence and efficacy.
  • Donanemab Shows Promise: The in-silico simulations suggest that donanemab may have a more significant effect on slowing cognitive decline compared to aducanumab, even though both drugs show comparable amyloid-beta clearance. This finding is particularly relevant given the ongoing development of various anti-amyloid therapies.
  • Long-Term Treatment is Crucial: The simulations extend to 10-year treatment periods, highlighting the potential need for sustained therapy to achieve meaningful long-term benefits in managing AD.

Key Results

  • Model Validation: The parameter estimation process yielded average relative errors of approximately 10% for the AD group and 15% for the CN and LMCI groups, indicating a good fit of the model to the ADNI data.
  • Treatment Efficacy: In silico simulations showed that both aducanumab and donanemab effectively cleared amyloid beta. Donanemab demonstrated a more pronounced effect on slowing cognitive decline compared to aducanumab.
  • Cognitive Decline Reduction: For an AD patient, long-term treatment (10 years) starting at age 60 showed a reduction in cognitive decline of 8% for aducanumab and 20% for donanemab. Median cognitive percentage changes for the AD group with long-term treatment were 5.2% for aducanumab and 13.1% for donanemab. Similar trends were observed for the LMCI group.
  • Dose-Dependent Effects: The study simulated low and high doses for aducanumab, showing that higher doses generally led to greater amyloid plaque reduction, aligning with clinical trial observations.

Limitations and Future Work

The authors acknowledge several limitations and suggest future directions:

  • Model Simplification: The current model is a simplification of the complex AD pathophysiology. While it captures key biomarkers, it doesn't encompass all molecular and cellular mechanisms.
  • Data Availability: The accuracy of personalized models is dependent on the availability and quality of longitudinal biomarker data.
  • Future Directions:
    • Extension to Other Therapies: The framework can be extended to include other AD treatment classes, such as anti-tau therapies or combination therapies.
  • Spatiotemporal Modeling: Incorporating spatial information from imaging biomarkers could provide a more comprehensive understanding of treatment effects on whole-brain neuropathology.
  • Further Validation: The authors plan to further validate and refine the model using data from ongoing clinical trials, such as the Aducanumab phase 3 studies.
  • Real-World Data Integration: Testing the model with real-world clinical data beyond ADNI would be a valuable next step.

Summary updated

BibTeX
@article{hao2022optimal,
  author = {Wenrui Hao and Suzanne Lenhart and Jeffrey R. Petrella},
  title = {{Optimal anti-amyloid-beta therapy for Alzheimer's disease via a personalized mathematical model}},
  journal = {PLOS Computational Biology},
  year = {2022},
  doi = {10.1371/journal.pcbi.1010481},
}
Figure