Dynamic modeling of the hypoxia induced HIF-1 response: a quantitative data-driven approach

Hypoxia is a reduction in the normal level of tissue oxygen tension and is a common feature inmany diseases including cancer [1]. Hypoxia triggers several oxygen-sensing pathways thatalter cell metabolism and allow the tumour to adapt to the hypoxic micro-environment. Hypoxiccells display a high resistance against conventional therapies, like radiotherapy andchemotherapy. Therefore, these hypoxia induced pathways have become of increasing interestfor additional tumour therapies. The best studied hypoxia-response is driven by a transcriptionalprogram initiated by (regulated) stabilization of hypoxia inducible factor 1 (HIF-1). The HIF-1response is complex and depends highly on both the severity and duration of hypoxia.Moreover, the response shows cell-type specific characteristics. Understanding the dynamicbehaviour of the HIF-1 response will facilitate future progress in targeting tumour-hypoxia.However, understanding a complex bio-regulatory network is often difficult and thereforecomputational models are becoming of increasing interest.Past years, several computational models of the HIF-1 pathway have been presented inliterature. To our knowledge, Kohn et al. [2] were the first to publish a computational model.Next, Yu et al. [3] simplified their model. They both used a qualitative approach to model theHIF-1 response and successfully described a ‘switch-like’ oxygen-response of HIF-1 activity.However, experimental data of Jiang et al. [4] showed an ‘exponential-like’ oxygen-response ofHIF-1 activity. The aim of this study is to develop a computational model using a quantitativeapproach which is able to reproduce current experimental data.A minimal model of the HIF-1 response has been developed, containing eight molecular speciesand nine unknown model parameters. The parameters have been estimated by minimizing thedifference between data of Jewell at al. [5] and the model output using a least squares criterion.Furthermore, the identifiability of the network components has been studied extensively bymeans of Fisher Information Matrix (FIM) analysis and a Monte Carlo approach. A LocalParametric Sensitivity Analysis (LPSA) and Multi Parametric Sensitivity Analysis (MPSA) havebeen carried out to assess model robustness and revealed key-sensitive model components.This has contributed to the optimal design of new experiments to improve model reliability. Tovalidate the model, we demonstrate that the model is capable of reproducing two other datasetsof Jewell et al. [5] and Jiang et al. [4]. This has lead to new insight in which parts of thepathway are specifically oxygen-sensitive and indicates that the complex formation of HIF-1áand PHD is the oxygen-sensitive step rather than the actual hydroxylation of HIF-1á. Thishypothesis is open for experimental validation.