AI in Alzheimer’s drug development: Where do disease-modifying interventions go from here?

Based on the success of amyloid antibodies, many therapeutic modalities for tau and inflammation pathways use similar biologics, often combined with a brain shuttle to enhance brain penetration. Because of the complex blood-brain-barrier system (BBB), concentrations reach between 0.1 and 0.3% of plasma levels. The addition of a shuttle increases this by about 4-to 8-fold. Physiology-Based Pharmacokinetic Modeling (PBPK) can help identify the best combination of half-life and shuttle selection for optimal brain penetration ^​1​.

This works quite well for amyloid antibodies that can bind to extracellular amyloid and amplify their response by engaging with microglia-dependent clearance pathways of extracellular amyloid. In contrast, for tau antibodies, the challenge is to reach sufficient levels inside the extremely small (30 nm) and crowded synaptic cleft to capture the seed-competent tau molecule that is released from the presynaptic nerve ending. A seed-competent tau molecule is a specific, misfolded form of the tau protein, which can induce other normal tau proteins to misfold and aggregate, propagating the spread of neurodegenerative diseases like Alzheimer’s. Spatial Monte-Carlo simulation ​^2,3​ based on a realistic geometric and biologically relevant model of the synaptic cleft, can help identify the minimal concentration needed to achieve a sufficient level of interaction between the tau antibody and its substrate.

As the disease progresses with increasing demyelination taking place, seed-competent tau might also enter along the axonal projections as well. A spatially resolved mechanistic QSP model is available that includes various processes such as uptake, lysosomal breakthrough, aggregation with monomeric tau into soluble tau oligomers and neurofibrillary tangles, degradation via the autosome-lysosome and the proteasome pathway, transport along microtubule and secretion at the next presynaptic compartment. This model is calibrated using image-based Tau SUVR PET (Standardized Uptake Value Ratio (SUVR) used in Positron Emission Tomography) imaging readout.

A better solution would be to tackle tau and neuroinflammation targets with small molecules that can achieve adequate brain target engagement. Indeed, tau spends almost all its time inside neuronal cells, and blocking aggregation or enhancing degradation can achieve better outcomes over the long term. Similarly, a number of small molecules have been approved for peripheral inflammation that could serve as a template for addressing microglia- or astrocyte-relevant pathways.

As anti-amyloid therapy becomes part of the standard-of-care, the question arises about possible synergy with tau or neuroinflammation therapeutic modalities. Both preclinical and clinical data strongly suggest that soluble amyloid oligomers increase neuronal firing and, therefore, secretion of misfolded tau protein. To address this question, a computational neuroscience and biophysically realistic QSP model was developed that links changes in amyloid oligomers to neuronal firing and integrates with the mechanistic tau propagation model described above ^​4​. The model can be used to identify unexpected non-linear interactions and optimize timing, duration, and patient selection of amyloid-tau combination therapies.

In a certain sense, combination therapy is already well established, as many Alzheimer’s patients take CNS active medications, such as standard-of-care acetylcholinesterase inhibitors, antidepressants, benzodiazepines, antipsychotics, and anti-epileptics. As an example, serotonin dynamics can affect cognitive readout even with the same amount of amyloid reduction, possibly explaining the difference between aducanumab’s ENGAGE and EMERGE trial^​5​. Using the properties of neuronal firing, this platform allows simulation of the effect of these symptomatic medications on functional cognitive readouts, such as CDR-SOB or ADAS-Cog, at the individual patient level ^​6​. CDR-SOB is the Clinical Dementia Rating – Sum of Boxes measurement, a clinical endpoint to measure cognitive changes in the early stages of Alzheimer’s disease. ADAS-Cog, or the Alzheimer’s Disease Assessment Scale-Cognitive Subscale, is a neuropsychological test considered a “gold standard” in Alzheimer’s clinical trials.

The pharmacodynamic interactions between comedications and disease-modifying interventions described above can lead to a large variability in functional readout, especially in small Phase 2 studies with a limited number of patients ^​7​. These studies often show a beneficial effect on biomarkers, but the clinical signal on cognitive readout is blurred by the different effects of comedications. Using the concept of virtual twins by investigating the pharmacodynamic effects at the individual patient level allows for a better estimation of the functional effect of a novel therapeutic modality, possibly rescuing promising therapeutic approaches.

The increased use of Non-Animal Methods (NAM), such as human Induced Pluripotent Stem Cells, brain organoids, or brain-on-a-chip, allows for focusing on human disease-specific pathways, circumventing the limited translatability of preclinical animal models. While this can certainly identify more relevant human targets, these models do not address the complex anatomical connectivity in the human brain that ultimately drives behavior and functional outcome. A possible solution consists of implementing findings from these models into complex computational neuroscience circuits based on human brain connectomics. As an example, using multi-electrode array readouts of electrophysiological activity in human neurons with the Fronto-Temporal Dementia V337M Tau mutations, a model was developed to predict the impact of tau pathology on a clinically relevant functional scale^​8​.

There is an increasing consensus that, besides amyloid and tau pathology, aberrant neuroinflammation plays a major role in the progression of Alzheimer’s disease. While we have a better understanding of how to address peripheral inflammation and immune disorders, the role of microglia and astrocytes in driving aberrant neuroinflammation in the human Alzheimer’s brain is incomplete, not in the least because of the lack of good biomarkers and the dynamic nature of the cellular phenotypes. One way to integrate the increasing body of knowledge is to develop mathematical models of microglia and astrocyte pathways of amyloid and tau clearance, release of anti- and proinflammatory cytokines, proliferation, and survival. We have developed a detailed QSP model of the TREM2, TLR4 and TNFa-R pathway (RIPK1) in microglia based on preclinical data but constrained with available clinical data with the objective of developing actionable predictions on the impact of specific therapeutic interventions, either as a stand-alone or in combination with other interventions.

Continue to read how to build a comprehensive strategy for Alzheimer’s drug development by reading this complementary post on building a robust clinical pharmacology and model-informed drug development strategy where we highlight how QSP modeling can address key translational and target engagement challenges from discovery through early clinical research.

To learn more about Certara’s Alzherimer’s Disease QSP model, contact us below.