CORRESPONDENCE LINK TO ORIGINAL ARTICLE funding often outweighs the efforts required to rigorously challenge the novel findings; Failed trials for central nervous system thus, verification in an independent laboratory is unlikely. disorders do not necessarily invalidate As the costs of clinical studies are so much higher than those of preclinical development, preclinical models and drug targets one might assume that pharmaceutical com- panies would conduct robust replications of key findings. This is indeed the case during Anton Bespalov, Thomas Steckler, Bruce Altevogt, Elena Koustova, Phil Skolnick, lead optimization, candidate selection, test- Daniel Deaver, Mark J. Millan, Jesper F. Bastlund, Dario Doller, Jeffrey Witkin, ing different administration routes and the Paul Moser, Patricio O’Donnell, Ulrich Ebert, Mark A. Geyer, Eric Prinssen, use of primary preclinical disease models. Theresa Ballard and Malcolm Macleod However, this is far less common for stud- ies in the more complex disease models that are used in late stages of development and A recent article identified five key technical modest effects on survival. Numerous other are potentially more relevant for predicting determinants that make substantial contribu- drug candidates have reported efficacy in a clinical efficacy. tions to the outcome of drug R&D projects superoxide dismutase 1 (SOD1) mouse model In general, the rigorousness with which (Lessons learned from the fate of AstraZeneca’s of ALS (one of the common animal models preclinical data is obtained — and the result- drug pipeline: a five-dimensional framework. for this disease), but none of these candidates ing robustness of the data — is quite low; Nat. Rev. Drug Discov. 13, 419–431 (2014))1. produced an efficacy signal in clinical trials. few studies report randomization, blinding, Careful consideration of such determinants The ALS Therapy Development Institute later sample size calculations or attrition. might be particularly valuable in the fields of rigorously retested more than 100 of those neurology and psychiatry, in which successful molecules in the SOD1 model (using adequate Generalizability drug development has declined precipitously statistical power, treatment groups matched Every laboratory has a unique combination over the past decade. This decline has largely for litter and gender, blinding, uniform end of protocols, suppliers of tools and reagents, been fuelled by a high failure rate in the trans- point criteria, tracking of non-ALS deaths source of animals, and animal husbandry lation of preclinical efficacy findings, caused and quantitative analysis of transgene copy characteristics. As drugs are used in highly by multiple factors (see Supplementary infor- number prior to assigning mice to a study), and heterogeneous patient populations, efficacy mation S1 (table)), including limited training they were unable to replicate any of the previ- observed in a single lab is more likely to be and poor protocol design, inadequate animal ously reported preclinical efficacy findings2. successfully translated when similar findings models, insufficiently validated therapeutic In this context, the lack of clinical efficacy is can also be obtained under different condi- targets and problems with data handling and not surprising. tions in other laboratories. There is empirical reporting. Similar problems related to deficiencies evidence to support this assumption: the Here, we focus on three factors that can be in experimental design (such as inadequate broader the range of circumstances and labo- addressed immediately in order to re-evaluate blinding and randomization) had previously ratory environments in which preclinical the therapeutic potential of older drugs and been observed in studies with animal models efficacy can be demonstrated, the higher the targets and to increase the probability of suc- of stroke and multiple sclerosis3. The potential likelihood of detecting efficacy signals in clini- cess for future preclinical-to-clinical translation: impact of such experimental design problems cal studies6. A recent European Union-funded data robustness, data generalizability and target can be assessed retrospectively for drugs that initiative (the Multicentre Preclinical Animal engagement data, a factor that was also high- have already been approved or abandoned, Research Team (MultiPART)) established lighted in the recent article1. We argue that the and steps can be taken to improve the robust- web-based platforms for multicentre animal many failed clinical trials in neuropsychiatry do ness of experiments for drugs in development studies, and the National Institute of Ageing not necessarily invalidate the potential of a drug (see Supplementary information S2 (table)). supports an Interventions Testing Program target or an animal model. Rather, these failures A hallmark of the scientific method is that seeks to validate the efficacy of treatments indicate a need for improved experimental the replication of findings both within and for ageing across several test sites with ade- designs and a robust translational strategy to between laboratories. However, such replica- quately powered, rigorous experiments using better inform compound and dose selection tion is limited by cost, human resources, time genetically heterogeneous mice of both sexes. for clinical trials. We conclude that many of the and bioethical considerations. Additionally, Generalizability of preclinical data is not drugs and targets in neuropsychiatry that have “there is an almost irresistible pressure to stop only an issue concerned with laboratory been discarded because of negative clinical trial when the result is about what one expects it conditions and animal strains, age and sex. outcomes may deserve re-evaluation using con- to be,” according to Terry Quinn4. However, For example, there is a remarkable paucity temporary knowledge, methodology and tools. the more novel the findings of an experiment of studies employing chronic or subchronic appear, the less likely they are to be true5, drug administration. Given that even a second Robustness especially in the context of poorly designed dose of most drugs can alter the biological The problem of robustness in preclinical and underpowered studies. This problem milieu (for example, tolerance, sensitization data is best illustrated by an example from reflects the pressure to publish novel findings or receptor regulation), chronic dosing stud- research in amyotrophic lateral sclerosis in high-impact journals before being scooped ies are important for generating the best (ALS), a severe progressive neurodegenera- by a rival laboratory or funding runs out. The predictions of effects in patients. It is largely tive disease. There is currently one approved conventional value attached to such publi- unknown whether a compound’s efficacy has medication for ALS, riluzole, which has only cations for career advancement and future been confirmed in properly designed studies NATURE REVIEWS | DRUG DISCOVERY www.nature.com/nrd ©2016 Mac millan Publishers Li mited. All ri ghts reserved. CORRESPONDENCE with chronic administration in animals before 20 7 initiation of a clinical trial . This information 18 Without biomarker is particularly important for those indications With biomarker 16 for which preclinical models do not require repeated administration to detect drug effi- 14 cacy, while the treatment duration in clinical 12 trials can range from weeks to several months, 10 depending on the indication. 8 Target engagement Number of projects 6 In the context of hypothesis-driven drug 4 research, any observation of clinical efficacy is serendipitous if the molecule does not engage 2 its biological targets at the dose tested. Various 0 σ modelling tools can be used to assess target 1 3 4 H 3 NK 3 6 2A 1 M1 α7 PDE10 DA D DA D DA D 5-HT 5-HT GABA mGlu GlyT Other engagement, and direct target occupancy NMDAR Target assays including positron emission tomog- | raphy (PET) are increasingly available. The Figure 1 Analysis of the use of biomarkers in the development of novel treatments for schizophrenia. The Thomson Reuters Cortellis database (searchedNature on January Reviews 21, | 2014)Drug Discoverywas used aim is to demonstrate that, at relevant doses, to identify drug development projects in schizophrenia between 1994 and 2014. We could not find the drug is present in the same compartment evidence of biomarker-driven (for example, using positron emission tomography (PET), MRI or elec- as the target and in appropriate free concen- troencephalography (EEG)) dose selection for 80% of 72 novel drugs that were evaluated in Phase II trations to bind to it. PET studies demonstrate clinical proof-of-concept studies in this time period. 5-HT2A, 5-hydroxytryptamine receptor 2A; receptor occupancy but not targeted down- 5-HT6, 5-hydroxytryptamine receptor 6; α7, α7 nicotinic receptor; σ, σ receptor; DA D1, dopamine stream effects; such approaches may not be receptor 1; GABA, γ-aminobutyric acid; GlyT1, sodium- and chloride-dependent glycine trans- appropriate for novel drugs working through porter 1; H3, histamine receptor 3; M1, muscarinic acetylcholine receptor 1; mGlu, metabotropic allosteric or non-competitive molecular glutamate receptor; NK3, neurokinin receptor 3; NMDAR, N-methyl-d‑aspartate receptor; PDE10, mechanisms, and for some targets PET tracers phosphodiesterase 10. are not yet available. Scientists at Pfizer considered evidence of exposure at the site of action, target binding Path forward