Control and Prediction of the Organic Solid State A Basic Technology project of the Research Councils UK Computational Prediction of Pharmaceutical Crystal Structures a severe test of modelling supramolecular assembly Sarah (Sally) L Price Department of Chemistry, UCL www.cposs.org.uk Contrast Crystals - geological Durable & hard as strong forces between atoms Grown on geological timescales2cm 10 to 1000 years Organic/pharmaceutical and protein crystals Often very difficult (impossible?) to get even the very small crystals needed for solving structures by diffraction Compromise strong covalent bonds & weak intermolecular forces many Protein crystals in 1-2 ml drop Large crystals parabanic acid ~ 3-5mm Paracetamol form II Electron micrograph X-ray diffraction gives the atomic scale model Much greater resolution in organics than proteins No or limited water Protons often located from diffraction data Solving structures from powder samples increasing The Cambridge Crystallographic Database of organic crystal structures has >500,000 entries Most are the crystal structure of the first crystal found that was suitable for using X-ray diffraction, as chemists used to be only interested in the molecular structure Principles of crystal packing small molecules to proteins Close packed solvent may fill small voids dynamic water surrounds most protein molecules Forms hydrogen bonds, p-p stacking, X...X Intermolecular, more diversity Intramolecular, amino acids Conformation vital ~ isolated molecule (Y), some torsions vary dominant issue & major constraint Emphasis on inter - vs intra- molecular forces differ Exercises in prognostication: Crystal structures and protein folding JD Duntiz & HA Scheraga, 2004 PNAS 101, 14309 global optimisation problems to identify the structure(s) of lowest potential (or free) energy Search challenge Accuracy of energy evaluation Kinetic factors ~ preferred pathways to assembly, may be involved Objective blind tests “need to be maintained so they can continue to document progress and monitor excessive claims” Polymorphism - a common phenomenon?? Polymorphism - the ability of a substance to adopt more than one crystal structure since different physical properties, now a major cause for concern when products transform from one polymorph to another. L. Yu et al. •Pigments - change colour. •Chocolate - need polymorph of cocoa butter that melts at 37 oC. • Explosives - change of polymorphic form leads to different detonation properties & industrial accidents. 2000, J. Am. Chem. Soc. 122, 585. Pharmaceuticals must be marketed in one controlled polymorphic form Change of polymorph changes effective dose Want to choose the crystalline form for optimum properties & control production Regulatory requirement for pharmaceuticals that all reasonable experiments are performed in order to identify the maximum number of crystalline forms Difficulty in establishing that all polymorphs are known McCrone (1963) “the number of polymorphs of a material depends on the amount of time and money spent in research on that compound” - Some appear after decades of crystallisation work on compound - Some “disappear” after a more stable polymorph is discovered. Which drugs may have undiscovered polymorphs? 1998 Abbott Laboratories anti-HIV drug Ritonavir produced new polymorph during manufacture after 2 years Problem affected plants in different countries Required reformulation “ Unfortunately, there is nothing we can do today to prevent a hurricane from striking any community or polymorphism from striking any drug” Sun, Abbott Laboratories, press conference. Can we computationally predict whether the drug is in the polymorphism equivalent of Louisiana or Hertfordshire, Herefordshire or Hampshire ? Why calculate crystal energy landscapes? ~ the thermodynamically feasible crystal structures to confirm that most stable polymorph is known to design new molecular materials prior to synthesis to see what structures are plausible undiscovered polymorphs Thermodynamics vs. crystallization conditions (T, P, solvent, supersaturation, impurities, …..) to help solve structures from powder XRD or other experimental evidence as a complement to polymorph screening and “Quality by Design” crystallization processes in pharmaceutical development. 2010 5th CCDC Blind Test results – can we predict a crystal structure? O 2N O O - + N NO 2 N COOH O O S CH3 + COO- N Cl Cl - 2/13* OH N OH SO2 HOOC NH OH Polymorphs 3 and 4 2/10* x/y H2O S O O + 2/11 1/13 CH3 N H A salt N Cl 2/15 N CH3 0 or 2 excl H* /10 x = # correct within 3 submitted y = # groups submitting *own success Main issue is accuracy of calculating relative energies of different crystal structures Successful approaches to calculating lattice energy (biological force-fields rarely adequate) Plane wave density functional theory (i.e. crystal Y) supplemented by empirically damped -C6/R6 dispersion Elatt=Eelectronic+Edisp Model for intermolecular forces with electrostatic model derived from isolated molecule Y DEintra from Y Elatt = Uinter+ DEintra Non-spherical atoms developed by fitting to Use theory of intermolecular crystal structures forces, moving toward nonNeumann, M. A.; Perrin, M. A. J.Phys.Chem.B 2005, 109, 15531 empirical models Success in 4th for C6Br2ClFH2 with no experimental input Misquitta AJ, Welch GWA, Stone AJ, Price SL 2008.Chem Phys Lett 456 Rarely only one feasible crystal structure H3C Requires a uniquely favourable close packing defining all 3 dimensions O H N N N O H O 2N O CH3 Example with energy gap of ~12 kJ mol-1 Unique close packed plane Unique stacking from electrostatics MU Schmidt 1999 Erice Pigment Yellow 74 O CH3 More typical from 2007 blind test Cl Br Br H H F Landscapes will show the expected hydrogen bond motifs defining ribbons/layers BUT different •packings of ribbons •stackings of layers More predicted structures than known polymorphs Relative energies sensitive to method Basic method for crystal energy landscapes ~ thermodynamically feasible crystal structures Use quantum mechanics to predict molecular structure and represent the charge distribution within the molecule (repeat with multiple conformers for flexible molecules, using intramolecular energy penalty DEintra) Use search method to generate plausible crystal structures (~3000 MOLPAK or ~105 CrystalPredictor for each rigid conformation, or >106 for flexible CrystalPredictor) for Z’=1,... Use advanced models of the intermolecular forces (distributed multipoles to represent lone pair & p electron density) to minimize the intermolecular lattice energy Uinter of each crystal structure. Refine conformation within crystal to minimize Elatt= Uinter + DEintra > Basic Crystal (Lattice) Energy Landscape Estimate lattice modes, elastic tensor & harmonic free energies for rigid molecules and confidence in relative stabilities. Calculate other properties: PXRD, morphologies Karamertzanis PG, Kazantsev AV, Issa N, Welch GWA, Adjiman CS, Pantelides CC, Price SL 2009. J Chem Theory Comput 5, 1432 Price SL, Leslie M, Welch GWA, Habgood M, Price LS, Karamertzanis PG, Day GM 2010. Phys Chem Chem Phys 12:8478-8490. Why do we overpredict polymorphism ? 1 Neglect of thermal motion Cyclopentane C5H10 MD 30K ~ form III MD 160K ~ form I Plastic phases Free energy landscape for benzene has ~ a minimum for each known form in a metadynamics study Both have many lattice energy minima, and ~ only the observed structures when thermal motion modelled. Solid state transitions unusually facile for these hydrocarbons Torrisi A, Leech CK, Shankland K, David WIF, Ibberson RM, Benet-Buchholz J, Boese R, Leslie M, Catlow CRA, Raiteri, P. et al. Angew.Chem.,Int.Ed. 2005, 44, 3769 Price SL 2008. J Phys Chem B 112:3746 Contrast solid state of 5-fluorouracil, with no polymorphic transitions -94 Form I Z’=4 Lattice Energy / kJ mol-1 -96 C2/c P-1 -98 -100 -102 75% of these structures are free energy minima at 310 K P2/c P21 P21/c P212121 Pbcn In two solvates Pc Pca21 Pna21 Form I Form II -104 1.55 1.6 1.65 1.7 -3 Density / g cm 1.75 1.8 Form II found experimental search from dry nitromethane Form II & solvate Hulme AT, Tocher DA, SLP, 2005 J. Am Chem Soc, 127, 1116 Karamertzanis PG, Raiteri P, Parrinello M, Leslie M, SLP 2007 J Phys Chem B 112:4298. Do we need to do Molecular Dynamics to model thermal motion? Only if expect facile phase transitions. Dynamics of nucleation & growth will determine which structures are observed hydration of uracil in water gives close F···F of form I Hamad, S, Moon, C, Catlow, CRA, Hulme, AT, SLP, 2006 J. Phys. Chem. B, 110 3323 in nitromethane get R22 (8) of form II Solid-State Forms of b-Resorcylic Acid: How Exhaustive Should a Polymorph Screen Be? Braun DE, Karamertzanis PG, Arlin J-B, Florence AJ, Kahlenberg V, Tocher DA, Griesser UJ, Price SL 2011 Cryst Growth Des 11: 210-220. New polymorph I predicted, Added confidence to PXRD solution and evidence for proton disorder Similar structures, unlikely to be distinguishable polymorphs How? Relative stability? Catemer polymorph? Why do we overpredict polymorphism ? 2 The right crystallization experiment has yet to be performed Huge range of crystallization methods which have generated new polymorphs – deliberate to failed cocrystallization Experimental conditions vary kinetics of nucleation & growth Can we use crystal energy landscapes to find the right crystallization conditions? The right crystallization experiment has not yet been performed on carbamazepine ? Early predictions of a chain structure Better methods, modelling flexibility, induction etc – chains still competitive, also for related molecules Florence AJ, Johnston A, Price SL, Nowell H, Kennedy AR, Shankland N 2006. J Pharm Sci 95:1918-1930. Exptal searches are productive N O dimers chains isostructural relationships N NH2 O NH2 O NH2 O NH2 DHC CYH CYT CBZ form I form I form I form I 1992 2007 2008 2003 form II form II form II form II 2006 2008 2008 1987 form III form III 2007 1981 form IV form IV 2010 2002 1:1 CBZ:DHC solid solution form V 2006 In prepn Success of catemeric CBZ V required seeded sublimation CBZ form V CBZ form V DHC form II (seed) CBZ form V Arlin J-B, Price LS, Price SL, Florence AJ, in prepn Pbca a/Å b/Å c/Å Expt 9.1245(5) 10.4518(5) 24.8224(11) Rigid prediction 9.3124 10.5979 24.8819 Flex prediction 9.4816 10.3426 24.7227 Finding the right crystallization conditions may be even harder Racemic crystal could not be formed without racemization - or obliging synthetic chemist Challenge: what about cases where barrier is high but no so high? e.g. Changing to an unfavourable conformation – c.f. ritonavir Lancaster, RW; Karamertzanis, PG; Hulme, AT; Tocher, DA; Covey, DF; Price, SL, Chem.Commun., 2006, 47, 4921 (Dis)Appearing polymorphs Only need to nucleate more stable form once to get seeds Develop other routes to most stable form May lead to loss of control of crystallisation of metastable form Other forms of seeding/templating May need impurities to producing a polymorph 1 mol% ethamindosulphathiazole stabilizes form I sulphathiazole Attempts to reproduce form 2 progesterone failed – could only get moderately unstable samples when crystallised in presence of pregnenolone Lancaster RW, Karamertzanis PG, Hulme AT, Tocher DA, Lewis TC, Price SL 2007. J Pharm Sci 96:3419-3431. N. Blagden, R. J. Davey, R. Rowe and R. Roberts, Int. J. Pharm., 1998, 172, 169-177. 50 year old samples from Innsbruck Liquid Chromatography-Mass Spec Form 2 11 impurities total 4.85% Form 1 3 impurities total ~1.5%, Aldrich 1.3% different impurities irreproducible cocktail of impurities needed for long-lived form 2? Lancaster RW, Harris LD, Pearson D CrystEngComm, ASAP Why do we overpredict polymorphism ? 3 The right crystallization experiment cannot be performed Crystal may be unstable relative to other products, inherent in possible range of crystallization experiments Solvates may form Proton transfer – salt or cocrystal if 0< DpKa <3 Cocrystal may be less stable than components Why do we overpredict polymorphism ? 4.Plurality of possible structures is hindering crystallization Crystallization is difficult “Commonly found that when good quality large crystals of a substance cannot be grown, the small crystals are poor in quality with substantial mosaic spread” Harding, M. M. J. Synchrotron Radiat. 1996, 3, 250 i.e. structures solved from very small crystals (synchroton) are more likely to be disordered Can crystal energy landscape can warn of possibilities of disorder = combinations of low energy structures? From Eniluracil Crystal Energy Landscape H H H N Non-polar ribbons Polar ribbons O N H -114.9 kJ mol-1 -115.5 kJ mol-1 Parallel ribbons PXRD_1 -116.1 kJ mol-1 PXRD_2? Anti-parallel ribbons -116.4 kJ mol-1 Also little energy discrimination for the stacking variations for C4O C6H interchange Stacking & interdigitation errors hard to avoid & barrier to correction O Experimental: variable disorder in single XRD on 4 crystals Single crystal analysis could be interpreted as polymorphism. Powder patterns are very similar Variable disorder challenging for devising robust production process P21/n disordered anti-parallel non-polar 0.742(3) 0.705(3) 0.738(3) 0.841(3) Simulated PXRD Crystal 4 better R1 P21 Z’=2 minor polar Copley RCB, Barnett SA, Karamertzanis PG, Harris KDM, Kariuki BM, Xu MC, Nickels EA, Lancaster RW, Price SL 2008. Cryst Growth Des 8:3474 Where are we now? Crystal energy landscapes complement experiment, providing the alternatives likely motifs in solid forms range of possible target structures possible types of disorder Can be calculated with “good enough” accuracy for increasing range of molecules & multicomponent systems from aspirin / paracetamol to modern pharmaceuticals H2N O N O Database of computed crystal structures O H O O >150 molecules H3C N N N H3C N H S H H + N H - O O H What are the challenges? Improving accuracy of relative energies periodic electronic structure DFT+D non-empirical anisotropic atom-atom potentials free energies Understanding limitations of thermodynamic predictions ~ kinetic factors that lead to polymorphism Move to modern pharmaceuticals Computational efficiency Grateful Thanks to Matthew Habgood, Doris Braun, Nizar Issa, Gareth Welch, Sharmarke Mohamed Derek Tocher, Louise Price, M Leslie (ex-CCLRC) Bob Lancaster (ex-GSK) (UCL) Andrei Kazantsev, Panos Karamertzanis, Costas Pantelides, Claire Adijman (IC) Alastair Florence, Andrea Johnston, Jean-Baptiste Arlin, Phillipe Fernandes (SU) Other coworkers in CPOSS and many collaborators Other Programs AJ Stone (Cambridge), H Ammon (Maryland), CCDC Computing infrastructure: National Grid Service (database), HPC(x), UCL CCDC & CSP community for blind tests Funding EPSRC (including E-Science) Basic Technology Program of RC UK for funding Control and Prediction of the Organic Solid State www.cposs.org.uk, including “Translation” funding for Knowledge Transfer in CPOSS Industrial Alliance from April 2008. CPOSS Open Day, UCL Wednesday 30 March 2011 10.00, Coffee and registration in the South Cloisters Introduction and welcome, Progress in the fifth International Test of Crystal Structure Prediction, Industrial problems from polymorphism and how we might avoid them, Dr Colin Groom, CCDC Mapping Crystallization Processes Using In-Situ SSNMR, Dr Colan Hughes, Cardiff University First and Second Order Transitions: A Re-appraisal, Dr Terry Threlfall, University of Southampton 12.30, Lunch and poster session 2.00, The role of transformations in pharmaceutical crystallization, Prof. Kieran Hodnett, University of Limerick GIPAW: a "Bragg's Law" for solid state NMR, Prof. Chris Pickard, UCL Experimental screening and characterization of solid forms, Prof. Alastair Florence, University of Strathclyde 3.45, Coffee and poster session cont. 5.00, Removal of posters Sponsored by CPOSS Industrial Alliance – visit www.cposs.org.uk to register