Tatiana Karelina (1), Oleg Demin (1), Sridhar Duvvuri (2), Tim Nicholas (2)
Institute for System Biology, Moscow, Russia, (2) Pfizer Worldwide Research and Development, Groton, CT, USA
Objectives: Amyloid (Ab) is one of the main targets in Alzheimer’s disease (AD) therapy. The goal of this work was to model quantitatively longitudinal biomarker dynamics and to check whether different trial scenarios can help to distinguish between Ab toxicity hypotheses.
Methods: The proposed translational model of Ab pathology describes Ab production, clearance and distribution in brain, CSF, plasma and other tissues, as well as aggregation in brain. Multiple mouse and human data were used for model calibration, including on longitudinal data for healthy and AD subjects (brain, plasma and CSF Ab concentrations). It was externally verified using PET data for both populations and allows for translation between mouse, human and monkey. Model uncertainty was analyzed using hessian. Three simple assumptions of relationship between Ab and Adas-cog score have been formulated: (i) proportional or (ii) threshold Ab influence on Adas-cog score and (iii) Ab functional obligatoriness for neurons [1]. Their combinations gave 5 hypotheses which were implemented as explicit algebraic functions, describing the disease progression corresponding to the observed longitudinal sporadic AD data. Model was developed and analyzed using DBSolve Optimum software and R.
Results: Different therapeutic regimens were simulated: inhibition of Ab production by 20 and 50 % and activation of insoluble Ab destruction to 150% for two ages of therapy start: 70 (early) and 75 (late) years. Model predictions for different hypotheses revealed that for early therapy start no significant difference with placebo group would be seen for at least two years for all hypotheses. For later therapy start hypotheses of Ab obligatoriness predict cognitive decline at the beginning (for production inhibition) which is compensated only after two years by depletion of toxic forms. Destruction activation is less dangerous in this respect, but it requires not less than a year to observe at least 4 points of Adas-cog score diff vs placebo. Under the hypothesis of proportional toxicity and Ab obligatoriness model predictions approximately correspond to the published results of avagacestat [2] and bapineusumab [3] trials.
Conclusions: Mechanistic model allows framework for analysis of clinical trials, hypotheses formulation and optimal therapeutic design.
References:
[1] J Neurosci. 2008; 28(53):14537.
[2] Arch Neurol. 2012; 69(11): 1430.
[3] Lancet Neurol. 2010; 9(4):363.
Reference: PAGE 25 (2016) Abstr 5891 [www.page-meeting.org/?abstract=5891]
Poster: Drug/Disease modeling - CNS