Prescription drug Drug Classification No. 629 Harvoni tab. Dosage Forms and strength Each tablet contains Sofosbuvir ……….……………………………………………………………………………………… 400 mg Additives(tar colorant) : Yellow #5 Form A orange, diamond-shaped, film-coated tablet, debossed with "GSI" on one side
Meliordiscovery.com0022-3565/08/3251-1–9$20.00THE JOURNAL OF PHARMACOLOGY AND EXPERIMENTAL THERAPEUTICS Copyright 2008 by The American Society for Pharmacology and Experimental Therapeutics JPET 325:1–9, 2008 Printed in U.S.A. Perspectives in Pharmacology Exploiting Complexity and the Robustness of NetworkArchitecture for Drug Discovery Marc K. Hellerstein Department of Nutritional Sciences, University of California, Berkeley, California; Division of Endocrinology and Metabolism,Department of Medicine, University of California, San Francisco, California; and KineMed, Inc, Emeryville, California Received September 7, 2007; accepted January 16, 2008 ABSTRACT
The issue of complexity stands at the center of contemporary
sembled living systems (in vivo models) be studied and that infor- drug discovery and development. The central problem in drug mative in vivo biomarkers of the activity of biochemical pathways development today is attrition of drug candidates identified by responsible for disease be available. These biomarkers should the modern molecular target-based discovery approach, due to be sensitive, predictive of functional endpoints, and have high two related features of complex metabolic networks: their fun- enough throughput for efficient screening of large numbers of damentally unpredictable response to targeted interventions agents. To the extent that such biomarkers unambiguously reflect and their "robustness" (tendency to maintain stable function in the activity of pathways that mediate disease or therapeutic re- the face of internal or external perturbations). Complexity and sponse (i.e., are "authentic"), their utility will be increased. Exam- adaptations are, therefore, generally seen as obstacles to drug ples are presented of pathway-based screening of approved discovery. Here, the converse proposition is presented—that the drugs for unexpected actions. Results support the principle that complexity and adaptive responses of highly interconnected met- agents that have one action typically have many actions, including abolic networks can be exploited for therapeutic discovery. Un- unanticipated actions, reflecting connectivity relationships of anticipated connectivity relationships may result in "off-target" complex networks. Pathway-based screening in vivo represents changes in metabolic fluxes, leading to unexpected therapeutic an alternative to the high attrition of the molecular target-based actions of agents. Exploiting this approach requires that fully as- discovery paradigm.
Complexity and Modern Target-Based
by Lewis Thomas 30 years ago (Thomas, 1979): "It is essen- tially impossible to just step in and fix a complex social system,like an urban center or a hamster . . if there are things you are Understanding the complex metabolic networks that con- dissatisfied with and anxious to fix, you cannot set about fixing trol function in cells and organisms is widely seen as the next them with much hope of helping. This realization is one of the great challenge in biologic research. In the field of therapeu- sore discouragements of our century . . whatever you propose tics, inability to predict the consequences of manipulating to do based on common sense will almost inevitably make mat- molecular targets in complex living systems is the central ters worse than better . . you cannot meddle within one part of problem presently limiting the development of effective and a complex system from the outside without . . setting off disas- safe drugs (Noble, 2001; Duyk, 2003). The problem of com- trous events that you hadn't counted on in other remote parts.
plexity, therefore, stands at the center of drug discovery and Intervening is a way of causing trouble".
development (DDD) in the postgenomic era.
If Thomas is correct that "meddling" mostly causes trouble The most common perspective on complexity is as an ob- in complex systems, this would indeed be very bad news for stacle to successful application of insights generated from anyone interested in medical interventions (i.e., physicians, reductionist approaches. This perspective was well expressed the pharmaceutical industry, public health workers). Thisreview presents an alternative view of complexity, emphasiz- Article, publication date, and citation information can be found at ing its inherent creative possibilities and uses in drug dis- covery. The role of complexity in the high attrition rate of ABBREVIATIONS: DDD, drug discovery and development; ALS, amyotrophic lateral sclerosis.
drug candidates is discussed first, as this provides the heu- ristic model for understanding how the complexity and ad-aptations of biologic systems can be an ally in drug discovery.
Adaptations (Network Architecture) Can
Create Unpredicted Therapeutic Actions
Attrition and the Unpredictability of
It is clear that the modern DDD system is not working as advertised (Duyk, 2003) (Food and Drug Administration,www.fda.gov/oc/initiatives/criticalpath; Government Account-ability Office, www.gao.gov/new.items/d0749.pdf). Approvals ofnew chemical entities are at historic lows, despite increasingresearch budgets and powerful new tools for identifying thera- peutic targets and for generating compounds with activityagainst these targets.
Attrition is the Achilles' heel of molecular target-based drug discovery (Duyk, 2003) (Food and Drug Administration, www.fda.gov/oc/initiatives/criticalpath). Never have therebeen as many molecular targets or drug candidates as the Complexity Creates Unpredicted Targets
present, but more than 98% of drug candidates fail and ⬎90%of candidates entering phase II fail to gain approval. These failure rates are higher than 20 years ago (Duyk, 2003) (Food Drug Target
and Drug Administration, www.fda.gov/oc/initiatives/criti-calpath; Government Accountability Office, www.gao.gov/new.items/d0749.pdf). Modern drug discovery research isprimarily carried out ex vivo and in intentionally simple systems— e.g., using high-throughput enzyme assays, cul- tured cells for screening, etc. These tools are highly efficientat generating targets and candidates but have low power forpredicting the in vivo efficacy or toxicity of agents and targetsthat are identified. Lacking a reliable understanding of thefull range of activities (both beneficial and toxic) of novel Fig. 1. Fundamental reason for high attrition of drug candidates: outputs
of complex biochemical networks are unpredictable and are the true
compounds in living organisms and having no reliable way to therapeutic targets. This may be exploited for drug discovery, however. A, link molecular actions to physiologic or pathophysiologic pro- adaptations (network architecture) can create unpredicted actions of cesses, it is not possible to predict which interventions will be agents. B, complexity may create unpredicted molecular targets withtherapeutic utility.
effective and safe in humans.
This inability to predict the consequences of modulating Crabtree and Newsholme, 1987) and is not generally predict- molecular targets should hardly have been a surprise. Al- able a priori by analyzing the components in isolation.
though drugs interact with specific physical elements in an The notion that alterations in isolated nodes of complex organism (e.g., an active site on a protein, a ligand-depen- networks have unpredictable consequences is also important dant transcription factor), the actual therapeutic target in in modern genetics. A clear embodiment of this principle is vivo is distal and much more complex. What typically mat- the phenotype of numerous inborn errors of metabolism. In ters for phenotype is the output or flow of molecules through glycogen storage disease-type I (glucose-6-phosphatase defi- a functionally important pathway, which in turn operates ciency), for example, the ultimate phenotype in adults is not within a highly connected and interactive biochemical net- hypoglycemia but includes liver tumors, growth failure, and work (Fig. 1). Thus, functionally important biochemical path- platelet dysfunction (Chou et al., 2002). These would not be ways embedded in larger cellular and organismal networks anticipated a priori. Indeed, if glucose-6-phosphatase had represent the true targets of drugs. In a fibrotic or cirrhotic been discovered by modern "reverse biology", it might have liver, for example, there can be no reduction of fibrosis (ac- been classified as a liver tumor suppressor.
cumulated collagen in the extracellular matrix) without a These principles of metabolic control also have fundamen- reduction in the synthesis rate or an increase in the break- tal implications concerning evolutionary biology and the se- down rate of collagen. The synthesis and breakdown rates of lection of traits. If genetic mutations affect metabolic pheno- collagen in the liver, therefore, represent pathways with type in an unpredictable manner and with the possibility of intrinsic functional significance for fibrosis. In contrast, multiple effects at a distance, it is evident that very different molecular targets, such as transforming growth factor-␤ or phenotypes can emerge from random alterations at critical matrix metalloproteinases, for example, may appear attrac- nodes. The notion that higher level control over network tive in fibrogenic disorders, but their activities have no in- interactions (i.e., control architecture) is the main determi- trinsic functional significance; that is, activity of any isolated nant of function, rather than the genes or proteins in isola- element may or may not influence global output of or flux tion, of course implies that the same biochemical maps can through the pathway. In technical terms, the control have very different macroscopic phenotypes.
strength for any component of a pathway in a complex net- In context of pharmaceutical discovery, it is worth empha- work may range from zero to unity (Kaczer and Burns, 1973; sizing that metabolic fluxes refer to the flow of molecules Exploiting Complexity for Therapeutic Discovery
through endogenous pathways, regardless of the function of will probably induce numerous secondary adaptive changes.
the pathway, and do not include xenobiotic metabolism (i.e., This experimental observation is congruent with general the- drug metabolism). Thus, anabolic pathways (e.g., protein oretical analyses of network behavior. Barabasi (2002) has synthesis, lipogenesis, DNA replication), catabolic pathways emphasized that targeted disruptions against the most con- (e.g., ␤ oxidation of lipids, proteolysis, RNA degradation), as nected nodes in complex adaptive networks can have pro- well as intermediary metabolic pathways (e.g., tricarboxylic found effects. For instance, disabling a central airport such acid cycle, glycolysis, ribonucleotide synthesis, amino acid as Chicago or Dallas can have ripple effects throughout the metabolism), other biosynthetic pathways (modification of entire air traffic within the United States. The key is to lipids, glycosaminoglycan synthesis, etc.), and cellular devel- identify those nodes that induce large effects on flux distri- opment pathways (e.g., proliferation, differentiation and butions as these will be the targets most likely to have not death of epithelial cells or other cell types) are all included.
just one but many therapeutic (and perhaps undesirable)actions. This principle is discussed in greater detail below.
Robustness of Flux Distributions
Implications of Network Rigidity for
A second fundamental issue for therapeutics that arises from biological complexity is the tendency of evolved meta-bolic networks to resist external manipulation. Several lines These insights from metabolic engineering have several of evidence indicate that evolved biological networks actively implications for drug discovery. First, most molecular tar- maintain and defend stable function (flux distributions) in gets, particularly those in signaling or regulatory pathways, the face of internal or external perturbations. Technically, are highly unlikely a priori to have functional consequences this feature is termed robustness (Stephanopoulos and Val- or therapeutic utility (Bailey, 2001; Hellerstein, 2008). This lino, 1991; Fischer and Sauer, 2005) discovery might be quite troublesome for many contemporary The best characterized examples of robustness come from biologists if they were aware of it (which may explain why the field of metabolic engineering (Stephanopoulos et al., this principle is not widely appreciated).
1991; Fischer and Sauer, 2005). Despite the capacity to con- Robustness has more optimistic implications for therapeu- trol the expression of essentially every gene and the level of tics. Gene products that exert an effect on any flux probably every protein in bacterial cells (the simplest of organisms), affect many fluxes (Fischer and Sauer, 2005). This observa- metabolic engineers have learned how difficult it is to direct tion emerges from the same connectivity relationships that cells to efficient production of desired molecules (proteins, prevent most interventions from having any effect. Once any biofuels, etc.).
significant flux rate is altered, a cascade of other adaptive The failure to redirect fluxes by targeted manipulation is flux changes will almost certainly be induced, with unpre- explained by concepts such as "control architecture" (the dictable consequences. This principle has recently been ex- system of feed-back and feed-forward connectivity relation- ploited in the field of metabolic engineering for "strain im- ships that maintain a characteristic pattern of fluxes within provement" (Stephanopoulos et al., 2004). Random genetic a biochemical network) and "network rigidity" (the degree to alterations are introduced, and resulting metabolic fluxes are which characteristic flux distributions at different nodes are measured. Improvements in metabolite production from un- fixed and defended). Metabolic engineers appreciate the im- anticipated molecular targets have resulted, driven by the portance of the internal control systems that connect path- network's internal adaptive programs.
ways and thwart simplistic attempts to redirect flux (Bailey,2001; Stephanopoulos et al., 2004).
Historical Context: The History of
Recently, the robustness of flux distributions in the face of single gene deletions has been explicitly quantified. Stableisotope labeling methods were used to establish the effects on It is also worth reflecting on the history of 20th century metabolic fluxes in over 130 bacterial mutants, each lacking drug discovery to provide a context for the notion that com- a specific gene (Fischer and Sauer, 2005). It turned out that plexity can be the ally of therapeutic discovery. The current very few genes, when absent, alter flux distributions through model of DDD is different from the approach used to discover central metabolic pathways. Even without killing a bacterial the most useful drugs of mankind (Le Fanu, 1999). The cell, classic linear metabolic control models would predict unappreciated fact about most of our powerful drugs is the that a reasonably high percentage of gene deletions in central role of empirical observation in their discovery.
metabolic pathways should alter flux through their cognate Antipsychotic drugs, for example, have profoundly altered pathways. Sauer's results point out the redundancy of com- modern society. The phenothiazine antipsychotic drugs were plex networks and the degree to which flux distributions are discovered without any basic understanding of their mecha- defended by cells and organisms. By extension, it is likely nism of action or biological target. A researcher testing chlor- that most gene products in complex mammalian systems also promazine (Thorazine) as an analgesic agent noted that rats do not exert significant control strength over fluxes through appeared to be calmed by the drug and unperturbed by pain- key pathways, even when their activity is reduced substan- ful stimuli. Human volunteers given chlorpromazine exhib- tially (except in trivial cases, such as linear pathways with a ited a "profound quietude". The first patient with psychiatric single entry route).
disease given chlorpromazine had been institutionalized for Interestingly, however, the few gene products that did more than 20 years and was unable to take care of himself.
alter flux distributions altered more than one flux (Fischer Within 4 weeks, he was helping to "plan the institution's and Sauer, 2005). Accordingly, a less recognized consequence Christmas party" (Le Fanu, 1999).
of the rigidity of metabolic networks is that changing one flux Even less rational or targeted is the serendipitous obser- vation of useful actions of drugs after they were tested and systems, one might monitor the molecular fluxes that medi- approved for entirely different purposes. Some of our most ate phenotype, function, and disease (Crabtree and News- widely used drugs were discovered in this matter. Sildenafil holme, 1987; Hellerstein, 2004, 2008; Hellerstein and Mur- citrate (Viagra) was originally tested as an antianginal treat- phy, 2004; Turner and Hellerstein, 2005). Optimally, these ment. The observation of stimulated erectile function was fluxes should represent the critical pathways that drive dis- serendipitous and only detected because it was hard to miss.
ease or that mediate therapeutic response.
No one set out with the goal of overcoming erectile dysfunc- Some examples of outputs of pathways that have intrinsic tion by modifying this pathway. A similar story applies to significance in disease include synthesis and breakdown of minoxidil (Rogaine), the treatment for male-pattern bald- tissue collagen in fibrotic disorders or of axonal myelin in ness. Both of these unexpected activities were discovered multiple sclerosis, mobilization of cholesterol from athero- only because of the physical visibility of the phenotype. Other sclerotic lesions in the vessel wall, proliferation of prostate classic examples of secondary observations that led to impor- epithelial cells in benign prostatic hyperplasia, insulin-me- tant drugs include the development of antimalarials, diuret- diated glucose utilization by peripheral tissues in obesity and ics, and antidiabetic agents from sulfa antibiotics; antide- prediabetes, the proliferation of pancreatic ␤ cells in response pressants from the antituberculosis drug isoniazid; and to insulin resistance, the formation of new brain cells in the many others. These examples all fit the theme of "one action, hippocampus in disorders of cognition, and so on. An agent many actions" for agents in highly connected systems.
that altered the flow through any of these pathways would be The other key point here is the relative lack of importance hard to ignore because such pathways can be said to have of pre-existing mechanistic understanding or therapeutic in- intrinsic functional significance for disease. These metabolic tent in the development of many drugs. Although extremely or cellular pathways provide a conceptual framework for inefficient for testing and identifying candidates, the combi- designing authentic biomarkers of disease activity and ther- nation of serendipity or trial-and-error based on functional apeutic response.
outcome measures was more highly predictive of clinicalsuccess and resulted in much lower attrition rates than the Inventories of Individual Elements of
molecular target-based approach.
Networks: "-omics" Approaches
The classic approach has serious limitations, of course (inefficiency, failure to predict toxicities, low throughput), Considerable effort has been spent in recent years analyz- and the "low-lying fruit" may have been found already, so a ing the transcriptome (i.e., gene expression microarrays), the return of DDD to crude macroscopic phenotypic measures is proteome (global patterns of protein expression), and the not possible or desirable. The key question is whether mod- metabolome (metabolite profiles) (Debouck and Metcalf, ern technologies exist that can combine the capacity of mac- 2000; Berry, 2001; Weston and Hood, 2004). Statistical ap- roscopic phenotypic measures to predict successful clinical proaches have been developed for interpreting gene expres- outcomes with the throughput, breadth, and creativity of sion data, including gene set enrichment analysis, signifi- molecular target-based discovery.
cance analysis of microarray to gene-set analyses, and others(Subramanian et al., 2005; Dinu et al., 2007). These ap-proaches do not, however, provide a fundamental solution to Navigating the Unpredictability and
the problem of complexity and the interconnectedness of Robustness of Complex Networks
metabolic pathways. Gene expression data, proteomics, or To answer this question, it is worth considering the intu- other such inventories of the individual elements of complex itively familiar problem of driving a car through traffic.
networks provide no information about the central control There exists no computer program that can drive through feature of these systems: namely, the connectivity relation- traffic, but every 16-year-old learns to do it. What a teenager ships among the components. It is precisely these feed-back has that a computer does not have is vision—the ability to see and feed-forward interactions that carry the higher level and monitor where they are going. Having vision, it is a organizational rules responsible for defending or altering simple matter to apply feedback control and navigate flux distributions in the network. For this reason, inventories through traffic, stop when a pedestrian is in the cross-walk, of individual elements cannot reveal flux through pathways, and so on. In systems theory, this is termed "observability". A particularly through specific pathways that drive disease key point is that prediction (of the ambulatory behavior and pathogenesis or therapeutic response. The latter can only be location of each pedestrian in the city) is not necessary when learned by measuring actual fluxes that are present in the there is real-time observability. This is the notion that led assembled system. RNA levels, for example, often exhibit Norbert Weiner to coin the term "cybernetics" (from Greek little or no correlation with actual fluxes through pathways kubernetike, "the art of the steersman", or kubernetes, "the [e.g., gene expression data are misleading with regard to pilot of a ship, a helmsman, a guide, a governor") for the lipogenic and adipogenic fluxes in adipose tissue of obese science of automated control systems (Weiner, 1948). The mice (Turner et al., 2007a)].
helmsman is the key to ensuring that a desired destination or These principles of metabolic control suggest that invento- outcome is reached.
ries of network components, whether rational or unbiased, In the context of biologic control and drug development, can not result in definitive functional interpretation. It is what would need to be observed to allow efficient navigation notable, in this context, that the practitioners of microarray toward desired outcomes? The key functional output in bio- analysis and other -omics approaches implicitly recognize logic systems is the flow of molecules through relevant path- that flux through pathways is the bottom line. The authors ways (Kacszer and Burns, 1973; Crabtree and Newsholme, who described gene set enrichment analysis of microarray 1987; Noble, 2001). Thus, to understand and control complex data (Subramanian et al., 2005), for example, state that "an Exploiting Complexity for Therapeutic Discovery
increase of 20% in all genes encoding members of a metabolic Stable Isotope-Mass Spectrometric Methods
pathway may dramatically alter the flux through the path- for Measuring the Dynamics of Critical
way and may be more important than a 20-fold increase in a Pathways in Vivo
single gene", or maybe not. Catalogs of all elements of a From an operational standpoint then, the question can be system can not, in principle, provide definitive functional reduced to the following: can functionally important outputs of complex metabolic networks be characterized in vivo in Accordingly, the most interpretable and, thus, the optimal such a manner that efficient and predictive characterization strategy for characterizing the dynamics of pathways in com- of unexpected drug actions is made feasible? plex networks is to directly measure fluxes through the key Recent developments in stable isotope-mass spectrometric pathways, rather than relying on gene expression profiling or techniques may provide this opportunity. A wide variety of other indirect indices.
intermediary metabolic, biosynthetic, and cellular pathwayflux rates can be measured in vivo by use of new kinetictechniques (some are shown in Table 1). Stable isotope labels Exploiting Complexity for
have no toxicity or risk and can be used safely in humans, so the identical measurements can typically be translated fromanimal models into man. A stable isotope-labeled tracer is Taken together, these considerations have potentially pro- administered, and the pattern or rate of incorporation or found, although largely unexplored, implications for drug loss of label is monitored in molecules of interest by mass discovery. Although expected flux alterations may not occur spectrometry. By applying simple biochemical rules (e.g., in response to a targeted intervention, because of the com- precursor-product relationship, dilution principle, combi- plexity and adaptations of biologic networks, unexpected flux natorial analysis of polymerization biosynthesis) or more changes may result and may have therapeutic utility. This is complex mathematical models, quantitative flux rates of shown schematically in Fig. 1A. Alternatively, interventions molecules can be measured through anabolic, catabolic, at unexpected sites (Fig. 1B) may induce the metabolic out- intermediary metabolic, and cellular developmental (birth, put initially sought, thereby uncovering new drug targets. By differentiation, and death) pathways (Hellerstein, 1995; extension, the more complex and interactive a system is, the Hellerstein and Neese, 1999; Wolfe, 2005; Busch et al., more likely it is that unanticipated therapeutic targets and 2006, 2007).
agents will exist.
Some examples of dynamic pathway fluxes that we have Accordingly, if we had reliable ways of observing the activ- measured using stable isotope-mass spectrometric tech- ities of pathways that are intrinsically important in and niques in vivo include synthesis and turnover rates of pro- predictive of disease progression, these measurements would teins [e.g., liver collagen in fibrogenic states (Gardner et al., allow the potential therapeutic benefit of network connec-tions to be exploited. Having a direct readout of the impact of any intervention or perturbation within the network could Some pathway-based drug targets (diseases within parentheses) make complexity our therapeutic ally.
Insulin-mediated glucose utilization Pancreatic ␤-cell proliferation (diabetes mellitus) What Is Required to Exploit Complexity?
Reverse cholesterol transport (cardiovascular There are several basic requirements for systematic exploi- De novo lipogenesis (hepatic steatosis, tation of complexity for drug discovery. First, the model sys- Mitochondrial proliferation (obesity, tem studied clearly must comprise the fully assembled sys- tem of interest. To exploit connectivity relationships within a Muscle protein synthesis and breakdown (frailty, network, the complete repertoire of connections has to be Axonal microtubule dynamics (ALS, Parkinson's intact. Ex vivo or reductionist approaches can provide little Myelination (multiple sclerosis)Neurogenesis (depression, cognition, brain injury) Second, it is optimal if the outputs measured as markers Microglia proliferation (Alzheimer's and exhibit intrinsic functional significance, as defined above.
Parkinson's diseases) The less able a metric is to be dissociated from the disease Amyloid-beta turnover (Alzheimer's disease)Dendrite microtubule synthesis (synaptic itself—i.e., the more authentic it is as a biomarker (Turner plasticity, learning) and Hellerstein, 2005; Hellerstein, 2008)—the better it will Collagen synthesis (fibrosis of liver, lung, kidney, predict successful clinical response and the more value it will Keratinocyte/keratin turnover (psoriasis, atopic have as a screening tool.
Third, it is optimal if relatively high-throughput measure- Joint-space dynamics (osteoarthritis, rheumatoid ments are possible to allow broad screening approaches. Fi- Angiogenesis (tumor growth) nally, the capacity to monitor in real time through noninva- Tumor cell proliferation (prognosis, therapeutic sive sampling and rapid analysis is a plus.
Lymphangiogenesis (metastatic spread) Together, these features would allow true observability for DNA-cytosine methylation (gene silencing) diseases of interest and would simplify systematic attempts Ribonucleotide synthesis (cancer cell to discover unexpected targets and drugs.
2007), neuronal microtubule turnover in neurodegenerative Pathways as Therapeutic Targets in
diseases (Fanara et al., 2007), turnover of skin keratin in Neurobiology: Prediction of Clinical Response
hyperproliferative disorders of the skin (Lindwall et al., It is particularly instructive that several pathway mea- 2006)]; lipid fluxes [e.g., reverse cholesterol transport rates in surements in the field of neurobiology have proven to be atherosclerosis (Turner et al., 2007b), lipogenesis in obesity predictive of clinical response in animal models of disease.
and in response to dietary factors (Hellerstein et al., 1996), Neurobiologic diseases have not traditionally been monitored assembly and secretion of triglycerides by the liver in hyper- through biochemical kinetics but are generally investigated lipidemic states (Vedala et al., 2006)]; intermediary meta- using endpoints such as behavior, electrophysiology, neuro- bolic and carbohydrate-related fluxes [e.g., whole-body glyco- transmitters, or histopathology. We recently measured the lysis rates in insulin resistance (Beysen et al., 2007), hepatic pathway of hippocampal neurogenesis in adult animals gluconeogenesis, and glycogenolysis in diabetes (Chris- (Shankaran et al., 2006). Adult neurogenesis involves the tiansen et al., 2000)]; and the dynamics of cells as reflected in proliferation and maturation of progenitor cells in selected DNA replication and breakdown [e.g., birth and death rates areas of the brain and is essential for learning and formation of tumor cells in chronic lymphocytic leukemia (Messmer et of new memories, as well as being implicated in the thera- al., 2005), lymphocyte kinetics in human immunodeficiency peutic action of antidepressant drugs. By screening a wide virus infection (Hellerstein et al., 2003), and hippocampal variety of compounds in mice, we discovered that several neurogenesis in response to antidepressants (Shankaran et classes of drugs, including statins and antiepileptics, exhib- al., 2006)]. These mass spectrometric measurements are ca- ited previously undescribed stimulatory effects on hippocam- pable of relatively high throughput, which is essential for any pal neurogenesis (Fig. 2). When compounds were further broad screening initiative.
evaluated for functional activity in standard behavioral mod- As a proof-of-concept test of this approach for exploiting els of antidepressant activity or cognition (the forced swim complexity, we recently screened a number of approved and novel object recognition tests, respectively), improve- drugs for their effects on multiple pathways in vivo. The ments were confirmed. Likewise, measurement of microglial drugs tested were selected for pluripotency and included proliferation rates in rat brain in response to lipopolysaccha- statins, salicylates, retinoids, calcium-channel blockers, ride provided a screening approach for identifying agents glitazones, and others. The in vivo pathways studied in- that suppress neuroinflammation (Shankaran et al., 2007).
cluded insulin-mediated glucose utilization, hippocampal Several drugs were identified that had unexpected inhibitory neurogenesis, liver collagen synthesis, antigen-driven lym- actions on this pathway, and an agent identified by this phocyte proliferation, brain microglial proliferation (Table means (the retinoid isotretinoin) was subsequently shown to 1), and others.
delay symptoms in the EAE-MOG (experimental allergic en- It is interesting that we observed a "hit rate" (i.e., fre- cephalitis-myelin oligodendrocyte glycoprotein) mouse model quency of discovering previously unknown therapeutic ac- of multiple sclerosis.
tions) of roughly one for every 10 pathways studied per drug.
More recently, a novel pathway was identified (microtu- That is, if three drugs were tested against 7 to 10 pathways, bule dynamics in neurons) that has potent functional conse- there would be two to three new indications discovered. In quences in neurodegenerative conditions (Fanara et al., view of the fact that the agents that we screened had been 2007). Microtubules are essential as the "conveyer belts" approved for years or even decades, this is a remarkably high used for the transport of molecular cargo along axons. Whenthe dynamics of microtubule assembly/disassembly was mea- discovery rate for new indications. For example, Shankaran sured in neurons, the SODG93A mouse model of amyotrophic et al. (2006) found that a widely used class of cardiovascular lateral sclerosis (ALS) exhibited extraordinary hyperdynam- agents (statins such atorvastatin and simvastatin) has po- icity of microtubules in peripheral nerves and the central tent neurogenic stimulatory actions in the hippocampus (Fig.
nervous system. This hyperdynamicity was present most 2); that a ligand for the retinoid receptor (isotretinoin, Accu- strikingly in the microtubule subpopulations that are typi- tane) has inhibitory effects on microglial proliferation and cally the most stable; it was present before symptoms of neuroinflammation (Shankaran et al., 2007); that a cardio- disease; it was associated with abnormalities of cargo trans- vascular drug alters lymphocyte kinetics and has immuno- port along axons; and it worsened as the disease progressed suppressive actions; and so on. All of these actions were (Fanara et al., 2007). Moreover, treatment with agents that previously unknown and could be claimed to be new and reduced microtubule turnover rates improved transport of cargo, prevented the death of spinal cord neurons, delayed The extremely high frequency of unanticipated actions of the onset of neurologic signs of disease, and prolonged life approved drugs that we have observed by screening against remarkably in these mice (⬃30% extension of life span, which complex pathways supports the model, derived from the prin- is greater than previous reports for any agent). Most impor- ciple of robustness (see above) that agents that do anything tantly, the degree of normalization of this pathway by drugs are likely to do many things (Stephanopoulos et al., 2004; accurately predicted clinical signs of disease and survival.
Fischer and Sauer, 2005). A recent embodiment of this ap- A point about translating results from animal models to hu- proach is the work of Chong et al. (2006), who looked for mans is worth noting. It is true that one can never be sure antimalarial activity by repurposing approved drugs. These whether the control systems discovered in an animal model will investigators found antimalarial actions of several approved apply to humans. A key feature of the stable isotope measure- drugs not previously known or expected to alter host—para- ments described here, however, is that most of the techniques site interactions, congruent with the notion of one action, for measuring flux rates through pathways in vivo in animal many actions.
models can be used in identical fashion in human subjects. This Exploiting Complexity for Therapeutic Discovery
Fraction New Cells (
Fraction New Cells(
Fig. 2. A, screening approved drugs
Data: Mean ± SEM * p < 0.05 vs Vehicle. Fluoxetine and Imipramine were used as positive controls for stimulation of hippocampal neuro-genesis (Shankaran et al., 2006). B, dose-response of simvastatin on hip-pocampal neurogenesis (Shankaranet al., 2006).
Fraction New Cells (
Dose of Simvastatin (mg/kg/day, p.o.)
Dose of Simvastatin (mg/kg/day, p.o.)
Data: Mean +/- SEM * p < 0.05 vs Vehicle translatability of methods results in the capacity to test imme- Challenges and Obstacles for
diately, and with modest expense, time, and number of subjects, whether data from preclinical models apply in man.
To move pathway-based approaches into general use for indications discovery, several challenges must be faced. The Implications for Indications Discovery
most obvious is to identify which pathways should be mea- These results in living animals provide experimental sup- sured as biomarkers for DDD. Unlike molecular targets, port for the theoretical prediction based on fundamental which in a general sense exist as unambiguous physical units features of metabolic networks that systematic attempts to that do not require editorial decisions, key pathways exist in identify new therapeutic actions of agents ("indications dis- the eye of the beholder. There exist hundreds of pathways covery") are likely to be rewarding if tools for measuring the that might be targeted for therapeutic modulation. Criteria effects of agents on pathway fluxes are available and capable of intrinsic functional significance, authenticity in a disease of relatively high throughput in living organisms.
process, measurability in vivo, capacity for translation into humans, and medical significance of the disease(s) influence molecules through selected pathways may have intrinsic the attractiveness of different pathways for DDD.
functional significance, biomarkers of metabolic pathway Perhaps the most difficult challenge in practice will be to fluxes are probably more predictive of clinical response than prove that information from pathway measurements predicts static metrics, such as the expression level or activity of functional outcomes. In preclinical models, this is a relatively proteins or genes in isolation. If this general strategy can be straightforward process. Agents that are discovered to mod- efficiently reduced to practice, it represents an alternative to ulate a pathway flux in an animal model can be tested rap- the high attrition, high cost, and long time-lines of the con- idly against standard animal models of the intended disease.
temporary molecular target-based drug discovery approach.
Some examples of this were mentioned above [e.g., normal-izing the hyperdynamicity of microtubules in neurons im- proved transport of molecular cargo along axons in vivo and Bailey JE (2001) Reflections on the scope and the future of metabolic engineering and its connections to functional genomics and drug discovery. Metab Eng 3:111–
markedly delayed symptoms and death in the mouse model of ALS (Fanara et al., 2007)]. Other examples have been Barabasi A-L (2002) Linked: The New Science of Networks, Perseus Books, Cam- demonstrated [e.g., infusion of reconstituted HDLs increases Berry S (2001) Drug discovery in the wake of genomics. Trends Biotechnol 19:239 –
cholesterol efflux rate from tissues in vivo in rodents (Turner et al., 2007b) and appears to prevent atherosclerosis in ani- Beysen C, Murphy EJ, McLaughlin T, Riiff T, Lamendola C, Turner HC, Awada M, Turner SM, Reaven G, and Hellerstein MK (2007) Whole-body glycolysis measured mal models as well as people (Nissen et al., 2003)]. Demon- by the deuterated-glucose disposal test correlates highly with insulin resistance in strating the predictive strength of a pathway for human vivo. Diabetes Care 30:1143–1149.
Busch R, Kim YK, Neese RA, Schade-Serin V, Collins M, Awada M, Gardner JL, disease outcomes is more difficult. The prognostic value of Beysen C, Marino ME, Misell LM, et al. (2006) Measurement of protein turnover tumor cell proliferation rate in chronic lymphocytic leukemia rates by heavy water labeling of nonessential amino acids. Biochim Biophys Acta.
patients is currently being explored in a multicenter prospec- Busch R, Neese R, Voogt J, Awada M, Hayes G, Hellerstein M (2007) Measurement tive clinical trial based on a strong cross-sectional correlation of cell proliferation by heavy water labeling. Nat Protoc 2:3045–3057.
Chong CR, Chen X, Shi L, Liu JO, and Sullivan DJ Jr (2006) A clinical drug library between chronic lymphocytic leukemia tumor cell prolifera- screen identifies astemizole as an antimalarial agent. Nat Chem Biol 8:415– 416.
tion rates and aggressiveness of disease (Messmer et al., Chou JY, Matern D, Mansfield BC, and Chen YT (2002) Type I glycogen storage diseases: disorders of the glucose-6-phosphatase complex. Curr Mol Med 2:121–
2005). Lower turnover rate of T-cells in human immunodefi- ciency virus-1-infected humans with "virologic failure" while Christiansen MP, Linfoot PA, Neese RA, and Hellerstein MK (2000) Effect of dietary energy restriction on glucose production and substrate utilization in type 2 diabe- on antiretroviral treatment correlates with attenuated viral tes. Diabetes 49:1691–1699.
pathogenicity and was better associated with clinical pro- Crabtree B and Newsholme EA (1987) A systematic approach to describing and analyzing metabolic control systems. Trends Biomed Sci 12:4 –12.
gression than plasma viral load (Deeks et al., 2002). We have Debouck C and Metcalf B (2000) The impact of genomics on drug discovery. Annu Rev also reported that whole-body glycolytic utilization of an oral Deeks SG, Hoh R, Grant RM, Wrin T, Barbour JD, Narvaez A, Cesar D, Abe K, deuterated-glucose load correlates very closely with the re- Hanley MB, Hellmann NS, et al. (2002) CD4⫹ T cell kinetics and activation in sults of hyperinsulinemic-euglycemic glucose clamps (Beysen human immunodeficiency virus-infected patients who remain viremic despite
long-term treatment with protease inhibitor-based therapy. J Infect Dis 185:315–
et al., 2007), which in turn predict the efficacy of insulin- sensitizing agents.
Dinu I, Potter JD, Mueller T, Liu Q, Adewale AJ, Jhangri GS, Einecke G, Famulski KS, Halloran P, and Yasui Y (2007) Improving gene set analysis of microarray data
by SAM-GS. BMC Bioinformatics 8:242–254.
Duyk G (2003) Attrition and translation. Science 302:603– 605.
Fanara P, Banerjee J, Hueck RV, Harper MR, Awada M, Turner H, Husted KH,
Summary and Conclusions
Brandt R, and Hellerstein MK (2007) Stabilization of hyperdynamic microtubules
is neuroprotective in ALS. J Biol Chem 282:23465–23472.
Contemporary DDD is based on a molecular target-based Fischer E and Sauer U (2005) Large-scale in vivo flux analysis shows rigidity and paradigm, which suffers from some fundamental limitations.
suboptimal performance of Bacillus subtilis metabolism. Nat Genet 37:636 – 640.
Gardner JL, Turner SM, Bautista A, Lindwall G, Awada M, and Hellerstein MK These are predictable and understandable from principles of (2007) Measurement of liver collagen synthesis by heavy water labeling: effects of metabolic control theory. In a general sense, the disappoint- profibrotic toxicants and antifibrotic interventions. Am J Physiol Gastrointest
Liver Physiol 292:G1695–G1705.
ing output of new drugs in recent years reflects the failure of Hellerstein MK (1995) Methods for measurement of fatty acid and cholesterol me- molecular target-based DDD to account for the dynamic sys- tabolism. Curr Opin Lipidol 6:172–181.
Hellerstein MK (2004) New stable isotope-mass spectrometric techniques for mea- tems aspects of drug actions in complex biological networks.
suring fluxes through intact metabolic pathways in mammalian systems: intro- This review has focused on a related consequence of bio- duction of moving pictures into functional genomics and biochemical phenotyping.
Metab Eng 6:85–100.
logic unpredictability and robustness that has potential Hellerstein MK (2008) A critique of the molecular target-based drug discovery value for therapeutic discovery: namely, the occurrence of paradigm based on principles of metabolic control: advantages of pathway-based
discovery. Metab Eng 10:1–9.
unexpected, off-target therapeutic activities of drugs. Unex- Hellerstein MK, Hoh RA, Hanley MB, Cesar D, Lee D, Neese RA, and McCune JM pected off-target effects arise from the same network of con- (2003) Subpopulations of long-lived and short-lived T cells in advanced HIV-1
infection. J Clin Invest 112:956 –966.
nectivity relationships in biochemical systems that thwart Hellerstein MK and Murphy E (2004) Stable isotope-mass spectrometric measure- intended actions of candidates from being achieved. To the ments of molecular fluxes in vivo: emerging applications in drug development.
Curr Opin Mol Ther 6:249 –264.
extent that these unanticipated actions are beneficial in dis- Hellerstein MK and Neese RA. (1999) Mass isotopomer distribution analysis at eight ease states and can be systematically identified, this conse- years: theoretical, analytic and experimental considerations. Am J Physiol Endo-
crinol Metab 276:E1146 –E1162.
quence of unpredictability may be exploited for discovery of Hellerstein MK, Schwarz J-M, and Neese RA (1996) Regulation of hepatic de novo new therapies.
lipogenesis in humans. Ann Rev Nutr 16:523–557.
The most direct strategy for systematic discovery of actions Kaczer H and Burns JA (1973) The control of flux. Symp Soc Exp Biol 27:65–104.
Le Fanu J (1999) The Rise and Fall of Modern Medicine, Little Brown, London, UK.
of agents is to measure the flow of molecules through critical Lindwall G, Hsieh E, Misell L, Chai C, Turner S, and Hellerstein MK (2006) Heavy pathways, as metabolic engineers have been forced to do for water labeling of keratin as a non-invasive biomarker of skin turnover in vivo in
rodents and humans. J Invest Dermatol 126:841– 848.
altering outputs of bacterial systems (Stephanopoulos and Messmer BT, Messmer D, Allen SL, Kolitz JE, Kudalkar P, Cesar D, Murphy EJ, Vallino, 1991; Bailey, 2001; Stephanopoulos et al., 2004; Koduru P, Ferrarini M, Zupo S, et al. (2005) In vivo measurements document thedynamic cellular kinetics of chronic lymphocytic leukemia B cells. Clin Invest Fischer and Sauer, 2005). Moreover, because the flow of Exploiting Complexity for Therapeutic Discovery
Nissen SE, Tsunoda T, Tuzcu EM, Schoenhagen P, Cooper CJ, Yasin M, Eaton GM, in preclinical and clinical drug development. Curr Opin Drug Discov Dev 8:115–
Lauer MA, Sheldon WS, Grines CL, et al. (2003) Effect of recombinant ApoA-I Milano on coronary atherosclerosis in patients with acute coronary syndromes: a Turner SM, Roy S, Sul HS, Neese RA, Murphy EJ, Samandi W, Roohk DJ, and randomized controlled trial. J Am Med Assoc 290:2292–2300.
Hellerstein MK (2007a) Dissociation between adipose tissue fluxes and lipogenic Noble D (2001) From genes to whole organs: connecting biochemistry to physiology.
gene expression in ob/ob mice. Am J Physiol Endocrinol Metab 292:E1101–E1109.
Novartis Found Symp 239:111–123.
Turner S, Voogt J, and Hellerstein MK (2007b) Effect of reconstituted HDL on Shankaran M, King C, Lee J, Busch R, Wolff M, and Hellerstein MK (2006) Discov- cholesterol efflux from tissues in vivo, in Atherosclerosis, Thrombosis, and Vascu- ery of novel hippocampal neurogenic agents by using an in vivo stable isotope lar Biology 8th Annual Meeting; 2007 Apr 19 –21; Chicago, IL.
labeling technique. J Pharmacol Exp Ther 319:1172–1181.
Vedala A, Wang W, Neese R, Christiansen M, and Hellerstein M (2006) Delayed Shankaran M, Marino ME, Busch R, Keim C, King C, Lee J, Killion S, Awada M, and secretory pathway contributions to VLDL-triglycerides from plasma NEFA, diet, Hellerstein MK (2007) Measurement of brain microglial proliferation rates in vivo and de novo lipogenesis in humans. J Lipid Res 47:2562–2574.
in response to neuroinflammatory stimuli: application to drug discovery. J Neu- Weiner N (1948) Cybernetics or Control and Communication in the Animal and the rosci Res 85:2374 –2384.
Machine, The MIT Press, Cambridge, MA.
Stephanopoulos G, Alper H, and Moxley J (2004) Exploiting biological complexity for Weston AD and Hood L (2004) Systems biology, proteomics, and the future of health strain improvement through systems biology. Nat Biotechnol 10:1261–1267.
care: toward predictive, preventative, and personalized medicine. J Proteome Res Stephanopoulos G and Vallino JJ (1991) Network rigidity and metabolic engineering in metabolite overproduction. Science 252:1675–1681.
Wolfe RR and Chinkes DL (2005) Isotope tracers in metabolic research, in Principles Subramanian A, Tamayo P, Mootha VK, Mukherjee S, Ebert BL, Gillette MA, and Practice of Linetic Analysis, 2nd ed, Wiley-Liss, Hoboken, NJ.
Paulovich A, Pomeroy SL, Golub TR, Lander ES, et al. (2005) Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression Address correspondence to: Dr. Marc K. Hellerstein, Professor (DH Callo-
profiles. Proc Natl Acad Sci U S A 102:15545–15550.
way Chair), Department of Nutritional Sciences and Toxicology, 309 Morgan Thomas L (1979) The Medusa and the Snail: More Notes of a Biology Watcher. Viking Hall, University of California, Berkeley, CA 94720. E-mail: march@nature.
Press, New York.
Turner SM and Hellerstein MK (2005) Emerging applications of kinetic biomarkers
Froum.qxd 1/23/06 10:57 AM Page 71 A Retrospective Study of 1,925 Consecutively Placed Immediate Implants From 1988 to 2004 Barry Wagenberg, DMD1/Stuart J. Froum, DDS2 Purpose: The purpose of the present study was to evaluate implant survival rates with immediateimplant placement (IIP) into fresh extraction sockets and to determine risk factors for implant failure.Materials and Methods: A retrospective chart review was conducted of all patients in whom IIP wasperformed between January 1988 and December 31, 2004. Treatment required atraumatic toothextraction, IIP, and mineralized freeze-dried bone allograft with an absorbable barrier to cover exposedimplant threads. Implant failure was documented along with time of failure, age, gender, medical his-tory, medications taken, postsurgical antibiotic usage, site of implant placement, and reason forimplant failure. Statistical analysis was performed using chi-square and logistic regression analysismethods. Results: A total of 1,925 IIPs (1,398 machined-surface and 527 rough-surface implants)occurred in 891 patients. Seventy-one implants failed to achieve integration; a total of 77 implantswere lost in 68 patients. The overall implant survival rate was 96.0% with a failure rate of 3.7% pre-restoration and 0.3% postrestoration. Machined-surface implants were twice as likely to fail as rough-surface implants (4.6% versus 2.3%). Men were 1.65 times more likely to experience implant failure.Implants placed in sites where teeth were removed for periodontal reasons were 2.3 times more likelyto fail than implants placed in other sites. Patients unable to utilize postsurgical amoxicillin were 3.34times as likely to experience implant failure as patients who received amoxicillin. Conclusions: With a1- to 16-year survival rate of 96%, IIP following tooth extraction may be considered to be a predictableprocedure. Factors such as the ability to use postsurgical amoxicillin and reason for tooth extractionshould be considered when treatment planning for IIP. INT J ORAL MAXILLOFAC IMPLANTS 2006;21:71–80