An analysis of the attrition of drug candidates from four major pharmaceutical companies

Journal name:
Nature Reviews Drug Discovery
Volume:
14,
Pages:
475–486
Year published:
DOI:
doi:10.1038/nrd4609
Published online

Abstract

The pharmaceutical industry remains under huge pressure to address the high attrition rates in drug development. Attempts to reduce the number of efficacy- and safety-related failures by analysing possible links to the physicochemical properties of small-molecule drug candidates have been inconclusive because of the limited size of data sets from individual companies. Here, we describe the compilation and analysis of combined data on the attrition of drug candidates from AstraZeneca, Eli Lilly and Company, GlaxoSmithKline and Pfizer. The analysis reaffirms that control of physicochemical properties during compound optimization is beneficial in identifying compounds of candidate drug quality and indicates for the first time a link between the physicochemical properties of compounds and clinical failure due to safety issues. The results also suggest that further control of physicochemical properties is unlikely to have a significant effect on attrition rates and that additional work is required to address safety-related failures. Further cross-company collaborations will be crucial to future progress in this area.

At a glance

Figures

  1. Composition of the data set by phase and progression status.
    Figure 1: Composition of the data set by phase and progression status.

    The data set is composed of information on 812 oral small-molecule drug candidates from AstraZeneca, Eli Lilly and Company, GlaxoSmithKline and Pfizer. For further details, see the main text and Supplementary information S1 (box).

  2. Reasons for failure.
    Figure 2: Reasons for failure.

    a | Primary cause of failure for terminated compounds. b | Differences in the cause of failure for the first half (2000–2005) and second half (2006–2010) of the decade. c | Differences in the cause of failure in preclinical, Phase I and Phase II development.

  3. Distributions of selected calculated properties for the data set.
    Figure 3: Distributions of selected calculated properties for the data set.

    a | Molecular mass (Da). b | Calculated logP. c | Calculated logD7.4. d | Acid–base properties. For further details, see the main text and Supplementary information S1 (box).

  4. Comparison of selected physicochemical property distributions for the data set compared with compounds in patent applications and launched drugs.
    Figure 4: Comparison of selected physicochemical property distributions for the data set compared with compounds in patent applications and launched drugs.

    a | Molecular mass (Da). b | Calculated logP. c | Calculated logD7.4. d | Aromatic ring count. e | Fraction of sp3 atoms. The drug candidate set studied here was shown to have statistically significantly different mean values from both the patent set and the launched drug set for calculated logP, calculated logD7.4, aromatic ring count and the fraction of sp3 atoms. The mean molecular mass of the drug candidate set was statistically significantly different from the launched drug set but not from the patent data set. For an explanation of the depiction of the plots, see Supplementary information S1 (box).

  5. Distributions of different target classes within the data set.
    Figure 5: Distributions of different target classes within the data set.

    GPCR, G protein-coupled receptor; NHR, nuclear hormone receptor.

  6. Analysis of potential links between physicochemical properties and non-clinical toxicology outcomes.
    Figure 6: Analysis of potential links between physicochemical properties and non-clinical toxicology outcomes.

    a | Comparisons of selected property distributions of the compounds progressing to Phase I (n = 386) versus those failing in preclinical toxicology studies (n = 211). None of the differences in the mean values were statistically significant. b | Predictive power of the Pfizer 3/75 rule for the compounds progressing to Phase I versus those failing in preclinical toxicology studies. c | Ionization classes of the compounds progressing to Phase I versus those failing in preclinical toxicology studies. Probability of observing preclinical toxicology = 0.22 for zwitterions, 0.37 for the combined other classes; χ2 = 5.0; probability χ>2 = 0.026. d | Target class distributions of the compounds progressing to Phase I versus those failing in preclinical toxicology studies. GPCR, G protein-coupled receptor; NHR, nuclear hormone receptor, tPSA, topological polar surface area.

  7. Analysis of potential links between physicochemical properties and non-clinical safety failures.
    Figure 7: Analysis of potential links between physicochemical properties and non-clinical safety failures.

    a | Comparisons of calculated logP and calculated logD7.4 distributions of the compounds progressing to Phase II (n = 159) versus those failing for clinical safety in Phase I (n = 45). The set of compounds failing for clinical safety were shown to have statistically significantly different mean calculated logP values from those progressing. The difference in the calculated logD7.4 values was not statistically significantly different between the two sets. b | Ionization classes of the compounds progressing to Phase II versus those failing for clinical safety in Phase I. Probability of observing clinical safety failure in Phase I = 0.11 for zwitterions, 0.23 for the combined other classes; χ2 = 1.9, probability >χ2 = 0.17. c | Target class distributions of the compounds progressing to Phase II versus those failing to show clinical safety in Phase I. Probability of observing clinical safety failure in Phase I = 0.10 for ion channels, 0.23 for the combined other classes; χ2 (ion channel versus others in class) = 2.2, probability > χ2 = 0.14. GPCR, G protein-coupled receptor; NHR, nuclear hormone receptor.

  8. Analysis of potential links between physicochemical properties and failures due to poor pharmacokinetics.
    Figure 8: Analysis of potential links between physicochemical properties and failures due to poor pharmacokinetics.

    a | Comparisons of selected calculated property distributions of the compounds progressing to Phase II (n = 159) versus those failing owing to poor pharmacokinetics (PK) in Phase I (n = 28). None of the differences in the mean values were statistically significant. b | Ionization classes of the compounds progressing to Phase II versus those failing owing to poor PK in Phase I. Probability of observing PK failure in Phase I = 0.29 for zwitterions, 0.13 for the combined other classes; χ2 = 3.7, probability >χ2 = 0.054. c | Target class distributions of the compounds progressing to Phase II versus those failing for PK in Phase I. Probability of observing PK failure in Phase I = 0.23 for G protein-coupled receptors (GPCRs), 0.0 for nuclear hormone receptors (NHRs), χ2 (GPCRs versus other classes) = 5.7, probability >χ2 = 0.017; χ2 (NHRs versus other classes) = 5.1, probability > χ2 = 0.024. tPSA, topological polar surface area.

References

  1. Hay, M. et al. Clinical development success rates for investigational drugs. Nat. Biotechol. 32, 4051 (2014).
  2. Bunnage, M. Getting pharmaceutical R&D back on target. Nat. Chem. Biol. 7, 335339 (2011).
  3. Kola, I. & Landis, J. Can the pharmaceutical industry reduce attrition rates? Nat. Rev. Drug Discov. 3, 711716 (2004).
  4. Lipinski, C. A. Drug-like properties and the causes of poor solubility and poor permeability. Adv. Drug Delivery Rev. 23, 325 (1997).
  5. Leeson, P. D. & Springthorpe, B. The influence of drug-like concepts on decision-making in medicinal chemistry. Nat. Rev. Drug Discov. 6, 881890 (2007).
    This paper shows, perhaps for the first time, a link between physicochemical properties, notably lipophilicity and molecular mass, and in vitro promiscuity.
  6. Gleeson, M. P. Generation of a set of simple, interpretable ADMET rules of thumb. J. Med. Chem. 51, 817834 (2008).
  7. Hughes, J. D. et al. Physicochemical drug properties associated with in vivo toxicological outcomes. Bioorg. Med. Chem. Lett. 18, 48724875 (2008).
    From the observation of a link between logP and PSA and toxicology in preclinical in vivo toxicology studies, this study led to what is now known as the 3/75 rule.
  8. Peters, J.-U., Schnider, P., Mattei, P. & Kansy, M. Pharmacological promiscuity: dependence on compound properties and target specificity in a set of recent Roche compounds. ChemMedChem 4, 680686 (2009).
  9. Gleeson, M. P., Hersey, A., Montanari, D. & Overington, J. Probing the links between in vitro potency, ADMET and physicochemical parameters. Nat. Rev. Drug Discov. 10, 197208 (2011).
  10. Luker, T. et al. Strategies to improve the in vivo toxicology outcomes for basic candidate drug molecules. Bioorg. Med. Chem. Lett. 21, 56735679 (2011).
  11. Kenny, P. W. & Montanari, C. A. Inflation of correlation in the pursuit of drug-likeness. J. Comput. Aided Mol. Des. 27, 113 (2013).
  12. Van Der Graaf, P. H. et al. Can the flow of medicines be improved? Fundamental pharmacokinetic and pharmacological principles toward improving Phase II survival. Drug Discov. Today 17, 419424 (2012).
    Based on an analysis of development failures, this study suggests a set of criteria that a compound should meet to successfully progress through clinical development.
  13. Cook, D. et al. Lessons learned from the fate of AstraZeneca's drug pipeline: a five-dimensional framework. Nat. Rev. Drug Discov. 13, 419431 (2014).
  14. Watson, D. E. et al. Relating molecular properties and in vitro assay results to in vivo drug disposition and toxicity outcomes. J. Med. Chem. 55, 64556466 (2012).
  15. Wager, T. T. et al. Improving the odds of success in drug discovery: choosing the best compounds for in vivo toxicology studies. J. Med. Chem. 56, 97719779 (2013).
  16. Muthas, D., Boyer, S. & Hasselgren, C. A critical assessment of modeling safety-related drug attrition. Med. Chem. Commun. 4, 10581065 (2013).
    This analysis of the application of physicochemical descriptors to modelling safety attrition suggests that many approaches, including some of those cited here, may not be generally applicable to other data sets.
  17. Leeson, P. D. & St-Gallay, S. A. The influence of the 'organizational factor' on compound quality in drug discovery. Nat. Rev. Drug Discov. 10, 749765 (2011).
  18. Lovering, F., Bikker, J. & Humblet, C. Escape from Flatland: increasing saturation as an approach to improving clinical success. J. Med. Chem. 52, 67526756 (2009).
    This paper suggests that molecules containing a greater degree of saturation, assessed using the fraction of sp3 atoms, may be more likely to progress through various clinical stages.
  19. Morphy, R. The influence of target family and functional activity on the physicochemical properties of pre-clinical compounds. J. Med. Chem. 49, 29692978 (2006).
  20. Vieth, M. & Sutherland, J. J. Dependence of molecular properties on proteomic family for marketed oral drugs. J. Med. Chem. 49, 34513453 (2006).
  21. Tetko, I. et al. Accurate in silico logP predictions: one can't embrace the unembraceable. QSAR Comb. Sci. 28, 845849 (2009).
  22. Waring, M. J. Lipophilicity in drug discovery. Expert Opin. Drug Discov. 5, 235248 (2010).
    This review analyses the link between lipophilicity and a range of preclinical ADME and toxicology parameters, deriving an optimal range for logD and logP values that are expected to increase the chance of identifying compounds of development candidate quality.
  23. Bowes, J. et al. Reducing safety-related drug attrition: the use of in vitro pharmacological profiling. Nat. Rev. Drug Discov. 11, 909922 (2012).
  24. Waring, M. J. Defining optimum lipophilicity and molecular weight ranges for drug candidates — molecular weight dependent lower logD limits based on permeability. Bioorg. Med. Chem. Lett. 19, 28442851 (2009).
  25. Waring, M. J. & Johnstone, C. A quantitative assessment of hERG liability as a function of lipophilicity. Bioorg. Med. Chem. Lett. 17, 17591764 (2007).
  26. Tomizawa, K., Sugano, K., Yamada & H. Horii, I. Physicochemical and cell-based approach for early screening of phospholipidosis-inducing potential. J. Toxicol. Sci. 31, 315324 (2006).

Download references

Author information

Affiliations

  1. AstraZeneca, Alderley Park, Cheshire SK10 4TG, UK.

    • Michael J. Waring &
    • Garry Pairaudeau
  2. Thomson Reuters, 77 Hatton Garden, London EC1N 8JS, UK.

    • John Arrowsmith,
    • Sam Mandrell &
    • Alex Weir
  3. GlaxoSmithKline, Stevenage, Hertfordshire SG1 2NY, UK.

    • Andrew R. Leach,
    • Paul D. Leeson &
    • Stephen D. Pickett
  4. Present address: Paul Leeson Consulting, The Malt House, Main Street, Congerstone, Nuneaton, Warwickshire CV13 6LZ, UK.

    • Paul D. Leeson
  5. Pfizer, Cambridge, Cambridgeshire CB21 6GS, UK.

    • Robert M. Owen
  6. Pfizer, Groton, Connecticut 06340, USA.

    • William D. Pennie
  7. Present address: Takeda Pharmaceuticals, Cambridge, Massachusetts 02139, USA.

    • William D. Pennie
  8. Eli Lilly, Indianapolis, Indiana 46285, USA.

    • Jibo Wang &
    • Owen Wallace
  9. Present address: Novartis Institutes for Biomedical Research, Cambridge, Massachusetts 02139, USA.

    • Owen Wallace

Competing interests statement

The authors are minor stock holders and employees of their respective companies. The research was funded jointly by the four pharmaceutical companies AstraZeneca, Eli Lilly and Company, GlaxoSmithKline and Pfizer. The authors from Thomson Reuters received payment from these four companies to compensate them for their part in this work.

Corresponding author

Correspondence to:

Author details

  • Michael J. Waring

    Michael J. Waring is a Principal Scientist in the Oncology Medicinal Chemistry Group at AstraZeneca. He earned his Ph.D. from the University of Manchester, UK, under the supervision of Professor Timothy Donohoe, followed by postdoctoral research with Professor Philip Magnus at the University of Texas at Austin, USA. Waring joined AstraZeneca in 2001 and has worked on projects at all stages of drug discovery across both oncology and cardiovascular disease areas, culminating in the discovery of 14 development candidates.

  • John Arrowsmith

    John Arrowsmith is a medicinal chemist by training, with a Ph.D. in synthetic organic chemistry. He has more than 30 years of biopharmaceutical research and development experience as an executive director with Pfizer in the United Kingdom, United States and Japan. His expertise includes drug discovery and development, intellectual property, licensing, competitive intelligence and portfolio prioritization. He spent 3 years with Thomson Reuters in London, UK, as a pharmaceutical adviser. He has authored and contributed to numerous scientific papers, books and patents, and is the co-inventor of Tikosyn (dofetilide; an anti-arrhythmic drug).

  • Andrew R. Leach

    Andrew R. Leach leads a global group that applies structural biology, biophysics and biological mass spectrometry platforms across multiple therapeutic areas. He joined GlaxoSmithKline in 1994 following a Ph.D. from the University of Oxford, UK, postdoctoral research at the University of California, San Francisco, USA, and a fellowship at the University of Southampton, UK. His research interests have included the development and application of computational chemistry and informatics methods to drug discovery, and the development of new hit and lead discovery technologies, including fragment-based drug discovery and approaches to attrition reduction.

  • Paul D. Leeson

    Paul D. Leeson is an independent medicinal chemistry consultant. His industrial career began at Smith Kline and French, and has taken him to Merck Sharp and Dohme, Wyeth, AstraZeneca (where he was Head of Medicinal Chemistry at the Charnwood, UK, site and leader of AstraZeneca's Global Chemistry Forum) and GlaxoSmithKline. Leeson's drug discovery contributions have been in the cardiovascular, neuroscience, respiratory and inflammation therapy areas, and he has a special interest in compound quality.

  • Sam Mandrell

    Sam Mandrell is a manager within the clinical practice at Thomson Reuters and oversees the CMR Pharmaceutical Performance Benchmarking Programmes. His role focuses on the collection, curation and analysis of pharmaceutical industry data to provide performance benchmarks to assist research and development, and clinical development. He holds a degree in molecular and cellular biology, and before joining Thomson Reuters he worked in a number of research and development roles for biotechnology and contract manufacturing organizations, developing downstream purification processes for monoclonal antibodies and therapeutic proteins.

  • Robert M. Owen

    Robert M. Owen is a member of Worldwide Medicinal Chemistry in the Neuroscience and Pain Research Unit at Pfizer. He earned his Ph.D. from the University of Wisconsin–Madison, USA, under the supervision of Professor Laura Kiessling, followed by postdoctoral research with Professor William Roush. Owen joined Pfizer in 2006 and has worked on a broad range of target classes, including G protein-coupled receptors, kinases, enzymes and ion channels across a range of therapeutic areas. He is currently a medicinal chemistry project lead focusing on the identification of novel pain therapeutics.

  • Garry Pairaudeau

    Garry Pairaudeau is Head of External Sciences at AstraZeneca, with responsibility for open innovation and external collaborations in the chemistry and hit identification areas. He obtained his Ph.D. in chemistry from the University of Southampton, UK, in 1991, followed by postdoctoral work at the University of California, Irvine, USA. He joined AstraZeneca in 1994 as a medicinal chemist and has worked on all phases of drug discovery, with a focus on respiratory and cardiovascular diseases. He was Director of Chemistry for the cardiovascular group before taking up his current position in 2012.

  • William D. Pennie

    William D. Pennie was Head of Pfizer's Compound Safety Prediction Department before joining Takeda Pharmaceuticals as Global Head of Drug Safety in 2014. He has a longstanding research interest in multifactorial models of toxicity and was a co-recipient of the Doerenkamp–Zbinden Honour Award for Alternatives in Biomedicine and the Johns Hopkins CAAT (Center for Alternatives to Animal Testing) Recognition Award for advancing predictive toxicology.

  • Stephen D. Pickett

    Stephen D. Pickett works in the Chemical Sciences Department at GlaxoSmithKline. His current research interests include computational chemistry and chemoinformatics approaches for fragment-based drug discovery, high-throughput screening, screening collection design, quantitative structure–activity relationship modelling and virtual screening. He has authored more than 40 peer-reviewed scientific articles and six patent applications.

  • Jibo Wang

    Jibo Wang is a Research Adviser within Lilly Research Laboratories. He earned his M.S. in computer science and a Ph.D. in biochemistry from the University of Delaware, USA. His current research includes developing methodologies and tools in chemoinformatics, predictive modelling and computational chemistry. He is also interested in developing software for drug discovery scientists.

  • Owen Wallace

    Owen Wallace is Head of Global Discovery Chemistry at Novartis. He received his Ph.D. at Yale University, USA, before joining Bristol–Myers Squibb, working as a medicinal chemist on projects directed towards neuroscience and virology indications. In 2000, he joined Eli Lilly and Company, where his responsibilities included programmes in the endocrine and neuroscience therapeutic areas. In 2010, he moved to the Lilly Research Centre in the UK as the Site Scientific Leader. He returned to the United States in 2013, joining the Novartis Institutes of BioMedical Research.

  • Alex Weir

    Alex Weir is an analytics specialist at Thomson Reuters IP and Science. His work focuses on the development of data integration and analytic software solutions. Before his current role he acted as Programme Manager for the CMR Pharmaceutical R&D Performance Benchmarking Programme. His expertise includes data modelling, data visualization and prescriptive analytics.

Supplementary information

PDF files

  1. Supplementary information S1 (box) (1.5 MB)

    Dataset compilation and statistical analysis

Additional data