Integration not isolation: arguing the case for quantitative and systems pharmacology in drug discovery and development

Drug Discovery Today  Volume 16, Numbers 23/24  December 2011

REVIEWS

Reviews  GENE TO SCREEN

Integration not isolation: arguing the case for quantitative and systems pharmacology in drug discovery and development Balaji M. Agoram1 and Oleg Demin2 1 2

MedImmune LLC, Granta Park, Cambridge, UK CB21 6GH Institute for Systems Biology SPb, Moscow, Russia

Quantitative and systems pharmacology (QSP) is an emerging modelling technique that combines the flexibility of systems biology and tractability of compartmental pharmacokinetic–pharmacodynamic modelling techniques. Historically, there has been extensive use of QSP within the field of pharmacokinetics to optimise drug biopharmaceutical properties. However, application to target and biomarker selection, and design of preclinical and clinical studies is limited, but growing rapidly. In this article we highlight the impact of QSP within drug discovery and development by citing examples from within the field of pharmacology and we argue for a more systematic integration of QSP within the drug discovery and development paradigm. Introduction Quantitative techniques, such as pharmacokinetic–pharmacodynamic (PKPD) modelling and systems biology are commonly proposed as possible solutions to the problem of reduced productivity within pharmaceutical R&D [1]. However, modelling and simulation techniques, as presently employed during drug discovery and development, have limitations with respect to achieving the overall objective of decreased attrition. Empirical, ‘top-down’ PKPD models, which are the workhorses of quantitative decision making within R&D, are empirically grounded, easy to develop and use, portable, and good at extrapolating within a limited field of vision, across different doses and subpopulations, but are limited in their ability to predict the safety and efficacy profile across different targets and biomarkers. During preclinical and clinical development, these models are routinely used to maximise the information obtained from in vivo experiments, while minimising resource utilisation [2]. Complex, ‘bottom-up’ mechanistic models are long-term approaches to understanding the dynamics of all intermediates of a particular pathophysiology network as a whole, and hence are unable to adequately impact short-term strategy. These models typically attempt to understand a particular disease at a molecular and cellular network level, where multiple pathways intersect. This approach is typically focused on applying a quantitative Corresponding author:. Agoram, B.M. ([email protected]) 1359-6446/06/$ - see front matter ß 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.drudis.2011.10.001

framework to help interpret high throughput -omics data [3]. As such, neither of these approaches is tuned to answer some crucial questions arising in drug discovery and development, including the optimal target within a particular pathway of interest, a suitable biomarker to evaluate modulation of the target, the appropriate animal disease model to use, and the temporal relationship between target modulation, biomarker changes, and changes in clinical outcomes, among others. To bridge this gap, a new class of models, loosely termed quantitative and systems pharmacology (QSP) models, is being developed [1]. These models, of intermediate complexity between systems biology and PKPD models, have the flexibility of the systems biology approach and the tractability of the PKPD approach, and can be developed in a time scale that is compatible with routine decision making within the industrial drug R&D setting. A brief summary of the features of the three different mathematical modelling techniques are shown in Table 1, and an assessment of the stage of R&D where they could have maximal impact is proposed in Fig. 1. The objective of this article is to illustrate the value of QSP within drug R&D through case studies from both PK and PD fields. A summary of the outputs is discussed, followed by an assessment of the current state of this methodology along with emerging developments in this field. Detailed insights into the technical details on the development of systems models are beyond the scope of this article. www.drugdiscoverytoday.com

1031

REVIEWS

Drug Discovery Today  Volume 16, Numbers 23/24  December 2011

TABLE 1

Summary of typical characteristics of systems biology, QSP and PKPD models QSP

PKPD modelling

Typically cellular or molecular level

Conceptually developed at the scale of interest, but is usually at a tissue/organ scale to provide compatibility with PKPD models

Typically developed at the organism scale reflecting typical data collected

Highly granular; typically all network intermediates and their mechanistic linkage are captured (network-level)

Intermediate in granularity; typically considers key network intermediates, might contain mechanistic and empiric relationships between intermediates (pathway level)

Low granularity; typically no further layer of complexity than observation (input/output-type model; target level)

Assumption rich; typically need experimental data to calibrate

Assumption-rich; might need experimental in vitro data to calibrate

Low in assumptions; no further in vitro data required to calibrate

QSP models are widely used to predict PK properties of candidates In spite of its relative novelty in characterising the safety and efficacy profiles of therapeutic candidates, QSP models have historically been successfully applied within the PK field to predict clinical PK. Whole-body physiologically based pharmacokinetic (PBPK) models (Fig. 2), for example, are routinely used during drug discovery and development for several applications [4,5]. During preclinical development, empirical PK parameters obtained in animal species are allometrically scaled to predict human PK. Coupling the predicted human PK with PD thresholds, such as required coverage of target (e.g. 90% receptor occupancy at trough at steady state) provides an idea of the required dosing regimen of the candidate. This approach has several limitations, including the inability to predict tissue level PK, requirement of in vivo data for model calibration, and uncertain translation across different patient subpopulations, such as age, disease state, among others. PBPK models, account for fundamental processes determining drug disposition at the tissue/organ level, including permeability across tissue barriers, organ blood flow, and population variability and disease factors affecting these processes at the tissue-level and therefore provide a more detailed understanding and utilisable prediction of clinical PK [6]. Furthermore, in contrast to macroscopic lumped parameters, such as whole-body volume of distribution, the parameters describing PBPK models can be obtained from in vitro experiments, thus reducing the need [(Figure_1)TD$IG]for in vivo experiments [6,7]. The utility of this systems approach is

Discovery Relative contribution of modelling techniques (a.u)

Reviews  GENE TO SCREEN

Systems biology

Systems biology

Target selection

Preclinical

illustrated by the next generation of these PBPK models. The next generation PBPK models have expanded capability to predict oral PK based on in vitro solubility and permeability inputs by accounting for complex processes of drug dissolution, gut absorption [8] and population variability of predicted PK, by accounting for enzyme expression differences connected to genetic polymorphisms [9]. Similar to small molecules, the use of PBPK models to understand the disposition of biologics has been around for a few decades [10], but recent articles are beginning to reveal applications of this technology within the field of biologics discovery. The disposition mechanisms of many macromolecule biologics are different from that of synthetic molecules. For example, distribution to a tissue is typically confined to extracellular space and lymph flow also has an important role in the transport between blood and tissue interstitium. Mechanisms, such as the neonatal Fc receptor (FcRn) binding protect some biologics from rapid clearance. Because one of the reasons for pursuing some of these biologics is for the PK advantages they provide, it would be desirable to predict their PK before undertaking expensive synthesis steps. For example, Niederalt et al. (Mechanistic analysis of fusion proteins: PBPK applied in an Albuferon case study: http://www.page-meeting.org/?abstract=1866) reported the use of PBPK models to predict the plasma and tissue PK of an albumin-conjugated interferon molecule. They developed a PBPK model incorporating tissue distribution, competitive binding of albumin-conjugated compound to FcRn, and

Early Clinical

QSP

Candidate selection

Late Clinical

PKPD

IND; clinical start

PoC; Ph3 start

Market authorisation Drug Discovery Today

FIGURE 1

Current employment of mathematical modelling techniques during drug discovery and development. Although the use of PKPD modelling in preclinical and clinical development is well recognised, there is a clear need for mechanistic but rapidly developed and deployed models to address some crucial questions before and during preclinical development, such as the selection of animal models, translational biomarkers, and assessment of comparative efficacy/safety. Abbreviations: a.u: arbitrary units; IND: investigational new drug; PoC: proof of concept; Ph3: phase 3. 1032

www.drugdiscoverytoday.com

REVIEWS

[(Figure_2)TD$IG]

Vt

Qt

Lung

Bone Brain GI Liver

Arterial blood

Venous blood

Adipose

Spleen Kidney Heart

PSA t, σt Drug Discovery Today

FIGURE 2

A typical physiologically based PK (PBPK) model for small and large molecules. The broken and unbroken lines signify lymph and blood flows, respectively. Contrary to standard compartmental PK models, which need in vivo data for calibration, input parameters for the PBPK model are biochemical parameters (PSAt or st) which may be predicted from their structure and physiological parameters (Qt and Vt). Abbreviations: GI: gastrointestinal tract; PSAt: permeability-surface area product; st: reflection coefficient; Qt: plasma/lymph flow rate; Vt: tissue volume.

target-mediated internalisation in different tissues to obtain an integrated whole-body PBPK model. The model was found to accurately predict observed cynomolgus monkey and human PK for the test molecule. Such a model can be readily applied to the design of a biologic derivative with optimised PK properties (e.g. the model could be used to evaluate the impact of modified FcRn binding on overall PK and on tissue penetration [11]). Such hypothesis testing is more likely to be successful for biologics than for small molecules because some key determinants of PK, such as tissue penetration and lymphatic redistribution are class and size dependent rather than compound-dependent.

QSP can be used in all stages of drug R&D to understand safety/efficacy profile In this section we will review, through examples, how QSP can be applied to: (i) identify and evaluate targets, compounds, and biomarkers; (ii) understand preclinical models of disease, and; (iii) understand the mechanism of action of a compound.

QSP can be used in target and compound identification, and biomarker characterisation Target identification New targets are generally discovered by mining genomic information. Typically, constraint-based modelling, is used to reconstruct stoichiometry of metabolic pathways from genomics information using flux analysis [12,13]. However, models of that type do not take into account mechanistic aspects of intracellular metabolic pathways, which are enriched with various inhibitory and activatory mechanisms and strongly nonlinear reaction processes. Thus, they cannot derive dynamical changes in metabolite concentrations from known mechanistic features of the disease and therapy,

such as the drug action on target, pathway regulation and cell dynamics. So, kinetic modelling techniques, such as QSP can be used to describe dynamical properties of the disease progression and therapy applied. The article by Noble et al. [14] demonstrates this approach. Noble et al. focuses on the mathematical (kinetic) modelling of four-enzyme section of the shikimate pathway (Aro B, D, E and K) of Streptococcus pneumoniae. It is an essential pathway in plants, fungi, and microorganisms that is absent in mammals. The enzymes therefore provide attractive targets for new antimicrobial compounds [15]. The developed and validated kinetic model has been used to determine the contribution of each enzyme to the final product formation rate, to profile intermediate concentrations, and predict responses to inhibition effects of possible antipneumonia compounds. Using the model, conditions most appropriate for high-throughput screening were identified.

Compound identification Once a particular target is identified, it is necessary to identify a candidate with appropriate PKPD properties. The use of QSP at this stage is illustrated by the following examples. The high affinity and consequent low doses, and receptor binding nature of biologics results in complex in vitro affinity versus in vivo potency relationships. For example, darbepoetin alfa is less potent at the erythropoietin receptor than erythropoietin alfa, but yet is more potent in vivo in increasing haemoglobin than erythropoietin alfa [16]. A similar, complex relationship has been seen for granulocyte colony stimulating factors (GCSF) filgrastim and peg-filgrastim. The complex affinity–potency profiles for these classes of proteins were explained on the basis of the interdependent relationship between PK, receptor binding and internalisation by Sarkar et al. [17]. These proteins bind to their cognate receptors and the resulting complex is internalised. The internalised complex is then either recycled back to the surface or endosomally cleared [18]. Higher affinity to the receptor could result in higher receptor occupancy but less recycling to the surface and hence could result in lower overall potency. Similarly, a larger protein with lower receptor affinity could be protected from non-receptor mediated clearance through the kidney and serum proteases, and hence have higher serum persistence and hence higher in vivo potency. This work illustrated that creation of systems-level models can be invaluable in identifying the optimal PK and PD (e.g. receptor affinity) for such complex relationships.

Biomarker discovery and evaluation Statistical modelling is usually applied to discover possible biomarkers on the basis of microarray datasets generated through high-throughput techniques [19]. However, further information on the spatial and temporal dynamics of the biomarker and its mechanistic linkage to the disease are required before the biomarker can be thoroughly understood and used in diagnosis, monitoring progression, or reversal of a particular disease. An example of this situation is the use of vascular endothelial growth factor (VEGF) as a prognostic and diagnostic marker of cancer. The VEGF family holds enormous potential as biomarkers of angiogenesis and tumour growth. An important challenge in the effort to evaluate VEGF-A as a soluble diagnostic and prognostic biomarker of a variety of cancers is a lack of quantitative understanding of the temporal and spatial heterogeneities in its distribution. As one of the steps to address this problem, a simple www.drugdiscoverytoday.com

1033

Reviews  GENE TO SCREEN

Drug Discovery Today  Volume 16, Numbers 23/24  December 2011

REVIEWS

Reviews  GENE TO SCREEN

equilibrium model of the VEGF-A dynamics has been developed by Latham et al. [20]. This model does not take into account detailed mechanistic relationships between VEGF-A and intra- and extracellular processes involved in cancer progression, but uses a simple mass balance equation describing the whole-body mass conservation of VEGF-A to link tumour mass and serum VEGF-A levels. Despite the approximations inherent in the equation, the model predicts dependence of serum VEGF-A concentration on tumour mass, and thus provides a starting point towards the evaluation of serum VEGF-A as a biomarker of cancer diagnosis and treatment. Similarly, Budu-Grajdeanu et al. [21] have reported a simple QSP model of lupus nephritis accounting for immune cell dynamics, pro- and anti-inflammatory mediators, and global tissue damage/ dysfunction. They calibrated the model with biomarkers, urinary monocyte chemotactic protein-1 (uMCP-1) and urinary protein:creatinine ratio (u-P:C) data from patients. The authors indicated that the calibrated model could be used to better understand the disease mechanisms specific to individual patient flares and thus individualise lupus dosing regimen for each subject, something that individual biomarkers of lupus nephritis have not been shown to be capable of.

Understanding preclinical disease models One of the common challenges of preclinical development is the selection of an animal model of disease that suitably mimics the human disease condition to test drug candidates. The following example illustrates the use of QSP models to understand an animal model of rheumatoid arthritis (RA) – a complex multifactorial disease. Low-dose corticosteroid therapy has been recently implicated as a possible way of treating RA [22]. To identify optimal dosages and regimes of the exogenously administered corticosteroids, interregulation processes between cytokines and endogenous and exogenous corticosteroids must be taken into account. To address the problems, a disease progression model of collagen-induced RA in rats has been developed by Earp et al. [23,24]. The model integrates intracellular gene expression regulation, cell dynamics, extracellular biomarker performance and clinically measured endpoints at whole organism level. Details of mutual regulations of IL-1b, IL-6 and TNF-a expression with corticosterone and glucocorticoid receptor have been taken into account. Then, levels of extracellular concentrations of IL-1b, IL-6 and TNF-a have been taken into account as key factors contributing to regulation of cell dynamics of osteoblasts/osteoclasts and equations to describe such endpoints as paw edema and bone mineral density in cancellous and cortical fraction of bone in femur and lumbar regions have been derived. Owing to the model complexity authors employed a step-by-step verification/parameter identification strategy. This strategy of parameter identification resulted from model ability to integrate various levels of experimental (in vitro, in vivo and ex vivo) and clinical information complexity is specific characteristic of hierarchical QSP models. The model of RA in rats, then, has been applied to demonstrate relationships between dynamics of biomarkers (IL-1b, IL-6 and TNF-a), measured endpoints (paw edema and bone mineral density) and disease progression. In particular, it has been found that IL-1b and TNF-a most significantly induce paw edema but IL-6 exerted most influence on bone mineral density. Upon verification 1034

www.drugdiscoverytoday.com

Drug Discovery Today  Volume 16, Numbers 23/24  December 2011

the model has been applied to describe dexamethasone effects in collagen-induced RA in rats [23]. The model predicts that IL-6 and IL-1b are most sensitive to inhibition of dexamethasone and lower doses of corticosteroids might be sufficient to suppress the cytokines most relevant to bone erosion. This is a good example of how QSP can be used to generate an integrated understanding of preclinical models of disease progression and how the insights gained might be used to design clinical treatment regimens.

Understanding the mechanism of action of compounds In some cases, observed data can exhibit complex, non-intuitive behaviour. Although empirical PKPD models can be developed to help describe these observations, their predictive ability beyond the range of observations used to develop the models is uncertain. QSP models might be required to help integrate empirical data with a quantitative description of the underlying mechanism, which can help interpret observations and be used to extrapolate across different doses with relatively high confidence than a PKPD model. An example of this instance is the clinical data on 5lipoxygenase inhibition with zileuton in asthma [25]. Overproduction of leukotrienes (LT), which are products of arachidonic acid metabolism by the enzyme 5-lipoxygenase (5LO), is a major cause of bronchoconstriction and inflammation in asthma. Zileuton (Zyflo1) is a marketed, orally administered redox inhibitor of 5LO that causes bronchodilation in patients with asthma, as measured by forced expiratory volume in 1 s (FEV1). Literature data indicate similar FEV1 response after doses of 400 and 600 mg zileuton for 1 week, but increased sustained FEV1 response after 8 days at the 600 mg dose, but not the 400 mg dose. The difference in response could not be explained by changes in PK. Following the strategy described in previous section of the article, a systems model was developed integrating all known in vitro, in vivo and clinical data on the relevant components of 5LOmediated inflammatory pathophysiology and possible regulatory mechanisms involved in the response at the intracellular, cellular and organism levels (Can systems modeling approach be used to understand complex PK-PD relationships? A case study of 5-lipoxygenase inhibition by zileuton: http://www.page-meeting.org/ ?abstract=1746). This mathematical model contained the following components: (i) cell dynamics model of eosinphil (EO) maturation, migration, activation and death (Fig. 3); (ii) detailed biochemical model of 5-LO operation; (iii) semi-mechanistic model of LT biosynthesis in leukocytes; (iv) biophysical model of bronchoconstriction; and (v) PK model of zileuton and its inhibition of the intracellular 5LO pathway. All model parameters were estimated on the basis of available literature data. Using this model multiple hypotheses were generated to explain the observed delayed dose-response to zileuton administration in subjects with asthma. The model suggested that acute bronchodilation after zileuton administration was due to direct inhibition of LT synthesis. Doses of 400 and 600 mg maximally achieved this inhibition hence no dose–response is observed. In the asthmatic state high levels of activated inflammatory cells in the lung are driven by two positive feedback mechanisms through LT activation of EO and IL-5 induced cellular proliferation and activation. Sustained high levels of inhibition of LT synthesis (85%) are required to interrupt these positive feedback

Drug Discovery Today  Volume 16, Numbers 23/24  December 2011

REVIEWS

[(Figure_3)TD$IG]

Blood

IL5

Airways

IL5

IL5

LT

LT EOLT

EOLT EO

EO

EOact

Hn

EO

Reviews  GENE TO SCREEN

Bone marrow

EOact

Hn Drug Discovery Today

FIGURE 3

The mechanistic model of 5-lipoxygenase pathway, illustrating eosinophil cell generation, maturation, and trafficking through pro-inflammatory cytokines in bone marrow, blood, and airways. Abbreviations: EO: eosinophils; LT: leukotrienes; IL5: interleukin-5; Hn: histamine. Green and red arrows signify synthesis and degradation, respectively.

mechanisms, reduce the number of resident inflammatory cells and stimulate bronchodilation. This occurs only at the 600 mg dose and not the 400 mg dose. The delay in the observation of dose–response is characteristic of EO cell lifespan in the airways. A minimal, QSP modelling approach was successful in providing plausible explanation for complex PKPD observations. Similarly, a systems model of calcium homoeostasis was recently reported to help understand the mechanisms involved in maintaining calcium levels in plasma and thus maintaining bone mineral density [26]. Simulations with the model provided plausible explanations for some complex clinical observations including the effects of parathyroid hormone and RANK ligand inhibition on calcium levels and bone resorption.

Concluding remarks The above examples illustrate the use of the QSP approach, which combines the flexibility and tractability of systems biology and PKPD approaches, respectively, across various stages of drug discovery and development. Many of the questions in the examples listed above cannot be answered using an empiric, ‘top-down’, PKPD modelling approach, where data are required to estimate model parameters and furthermore, extrapolation of results to new scenarios not represented by the calibration data set is saddled with unknown uncertainty. A ‘bottom-up’ systems biology modelling approach is likely to take a substantial period of time to set up, validate, and implement and therefore, incompatible with the timescale of typical R&D decision making. The strength of the ‘centre-out’ QSP approach lies in the fact that, the model is only as complex as the question to answer; therefore, such models can be either expanded in scope into more detailed ones or simplified to fit empirical data with few assumptions [27]. The above examples also highlight the fact that pharmaceutical R&D has, in the past, successfully employed QSP to reduce candidate attrition owing to undesirable PK properties and select candidates with optimal PKPD properties. The next generation of QSP compounds in the PK field are attempting to address more complex issues, such as lung tissue retention for inhaled lung-targeted agents (A physiologically-based mathematical

model to predict lung retention and inhaled pharmacokinetics of therapeutic candidates: http://www.page-meeting.org/ ?abstract51452), prediction of penetration across the blood–brain barrier (Towards quantitative prediction of in vivo brain penetration using a physiology based CNS disposition model: http:// www.page-meeting.org/?abstract51691), and between-subject variability in metabolism [9]. In spite of their potential to transform drug R&D, systems modelling approaches face significant challenges before they become de rigueur in R&D. Foremost among them is the availability of reliable knowledge and experimental data on the system of interest. For many systems, sufficient quantitative mechanistic understanding might not be available to develop a QSP model to adequately describe complex behaviour of the system. In such cases, further in vitro and in vivo experiments might need to be conducted to generate data to develop and calibrate the QSP model so that it can answer the question of interest within tolerable limits of uncertainty. Operational hurdles also exist, such as lack of robust computational tools, lack of a common modelling and simulation language, lack of easy-to-use computational software, limited portability of models, and personnel trained in mathematical and life sciences who can also effectively translate the models into strategy and communicate to project teams [28]. Cultural hurdles also remain within pharmaceutical industry, where uncertain experimental data are almost always preferred to uncertain simulated data. To overcome these issues and to ensure that the QSP approach becomes an integral part of all aspects of drug discovery and development, it is necessary that PKPD modellers, who are embedded within project teams, recognise the value of this approach and ensure that data and knowledge generation occur simultaneously within project teams through the use of these models. A collaborative approach with disease biology experts in developing and deploying these models is crucial to exert influence on decision making within project teams. In spite of these hurdles, application of systems models in general, and QSP in particular, is poised to grow within pharmaceutical R&D. With their ability to provide quantitative insights www.drugdiscoverytoday.com

1035

REVIEWS

into disease pathways, describe behaviour of potential biomarkers, and mechanism of action of compounds, QSP modelling can form an integral part of translational research, especially for complex multifactorial diseases, such as diabetes [29,30] and

Drug Discovery Today  Volume 16, Numbers 23/24  December 2011

other auto-immune diseases. A consortium approach to the development and deployment of these models is taking hold and is likely to overcome some of the technical difficulties mentioned above (Pharma 2020: Virtual R&D Which path will you take?).

References Reviews  GENE TO SCREEN

1 Allerheiligen, S.R. (2010) Next-generation model-based drug discovery and development: quantitative and systems pharmacology. Clin. Pharmacol. Ther. 88, 135–137 2 Lalonde, R.L. et al. (2007) Model-based drug development. Clin. Pharmacol. Ther. 82, 21–32 3 Kreeger, P.K. and Lauffenburger, D.A. (2010) Cancer systems biology: a network modeling perspective. Carcinogenesis 31, 2–8 4 Nestorov, I. (2003) Whole body pharmacokinetic models. Clin. Pharmacokinet. 42, 883–908 5 Rowland, M. (1986) Physiologic pharmacokinetic models and interanimal species scaling. In Pharmacokinetics: Theory and Methodology. International Encyclopedia of Pharmacology and Therapeutics, Section 122 (Tucker, G., ed.), pp. 69–88, Pergamon Pr 6 Jones, H.M. et al. (2009) Modelling and PBPK simulation in drug discovery. AAPS J. 11, 155–166 7 Poulin, P. and Theil, F.P. (2000) A priori prediction of tissue: plasma partition coefficients of drugs to facilitate the use of physiologically-based pharmacokinetic models in drug discovery. J. Pharm. Sci. 89, 16–35 8 Agoram, B. et al. (2001) Predicting the impact of physiological and biochemical processes on oral drug bioavailability. Adv. Drug Deliv. Rev. 50 (Suppl. 1), S41–S67 9 Jamei, M. et al. (2009) A framework for assessing inter-individual variability in pharmacokinetics using virtual human populations and integrating general knowledge of physical chemistry, biology, anatomy, physiology and genetics: a tale of ‘bottom-up’ vs ‘top-down’ recognition of covariates. Drug Metab. Pharmacokinet. 24, 53–75 10 Baxter, L.T. et al. (1995) Biodistribution of monoclonal antibodies: scale-up from mouse to human using a physiologically based pharmacokinetic model. Cancer Res. 55, 4611–4622 11 Ghetie, V. et al. (1997) Increasing the serum persistence of an IgG fragment by random mutagenesis. Nat. Biotechnol. 15, 637–640 12 Jerby, L. et al. (2010) Computational reconstruction of tissue-specific metabolic models: application to human liver metabolism. Mol. Syst. Biol. 6, 401 13 Raghunathan, A. et al. (2010) Systems approach to investigating host-pathogen interactions in infections with the biothreat agent Francisella. Constraints-based model of Francisella tularensis. BMC Syst. Biol. 4, 118 14 Noble, M. et al. (2006) The kinetic model of the shikimate pathway as a tool to optimize enzyme assays for high-throughput screening. Biotechnol. Bioeng. 95, 560–573 15 Davies, G.M. et al. (1994) (6S)-6-fluoroshikimic acid, an antibacterial agent acting on the aromatic biosynthetic pathway. Antimicrob. Agents Chemother. 38, 403–406

1036

www.drugdiscoverytoday.com

16 Egrie, J.C. et al. (2003) Darbepoetin alfa has a longer circulating half-life and greater in vivo potency than recombinant human erythropoietin. Exp. Hematol. 31, 290–299 17 Sarkar, C.A. and Lauffenburger, D.A. (2003) Cell-level pharmacokinetic model of granulocyte colony-stimulating factor: implications for ligand lifetime and potency in vivo. Mol. Pharmacol. 63, 147–158 18 Gross, A.W. and Lodish, H.F. (2006) Cellular trafficking and degradation of erythropoietin and novel erythropoiesis stimulating protein (NESP). J. Biol. Chem. 281, 2024–2032 19 Qiu, P. et al. (2007) Dependence network modeling for biomarker identification. Bioinformatics 23, 198–206 20 Latham, A.M. et al. (2010) An integrative model for vascular endothelial growth factor A as a tumour biomarker. Integr. Biol. (Camb) 2, 397–407 21 Budu-Grajdeanu, P. et al. (2010) Mathematical framework for human SLE Nephritis: disease dynamics and urine biomarkers. Theor. Biol. Med. Model 7, 14 22 Svensson, B. et al. (2005) Low-dose prednisolone in addition to the initial diseasemodifying antirheumatic drug in patients with early active rheumatoid arthritis reduces joint destruction and increases the remission rate: a two-year randomized trial. Arthritis Rheum. 52, 3360–3370 23 Earp, J.C. et al. (2008) Modeling corticosteroid effects in a rat model of rheumatoid arthritis II: mechanistic pharmacodynamic model for dexamethasone effects in Lewis rats with collagen-induced arthritis. J. Pharmacol. Exp. Ther. 326, 546–554 24 Earp, J.C. et al. (2008) Modeling corticosteroid effects in a rat model of rheumatoid arthritis I: mechanistic disease progression model for the time course of collageninduced arthritis in Lewis rats. J. Pharmacol. Exp. Ther. 326, 532–545 25 Liu, M.C. et al. (1996) Acute and chronic effects of a 5-lipoxygenase inhibitor in asthma: a 6-month randomized multicenter trial. Zileuton Study Group. J. Allergy Clin. Immunol. 98 (5 Pt 1), 859–871 26 Peterson, M.C. and Riggs, M.M. (2010) A physiologically based mathematical model of integrated calcium homeostasis and bone remodeling. Bone 46, 49–63 27 Marathe, A. et al. (2008) Integrated cellular bone homeostasis model for denosumab pharmacodynamics in multiple myeloma patients. J. Pharmacol. Exp. Ther. 326, 555–562 28 Vicini, P. (2010) Multiscale modeling in drug discovery and development: future opportunities and present challenges. Clin. Pharmacol. Ther. 88, 126–129 29 Cedersund, G. and Stralfors, P. (2009) Putting the pieces together in diabetes research: towards a hierarchical model of whole-body glucose homeostasis. Eur. J. Pharm. Sci. 36, 91–104 30 Dalla Man, C. et al. (2007) Meal simulation model of the glucose–insulin system. IEEE Trans. Biomed. Eng. 54, 1740–1749