Physical and Molecular Properties of Agrochemicals: an Analysis of Screen Inputs, Hits, Leads, and Products
Total Page:16
File Type:pdf, Size:1020Kb
CROP PROTECTION RESEARCH 731 CHIMIA 2003, 57, No.11 Chimia 57 (2003) 731–734 © Schweizerische Chemische Gesellschaft ISSN 0009–4293 Physical and Molecular Properties of Agrochemicals: An Analysis of Screen Inputs, Hits, Leads, and Products Eric D. Clarke* and John S. Delaney Abstract: This work provides a comprehensive overview of agrochemical properties in terms of the way they change during progression from screen hit to product and in terms of their limits as expressed in modern commercial products. Most herbicides and fungicides readily meet the Lipinski ‘rule of five’ criteria for drug- like compounds with many meeting the more constrained limits reported for pharmaceutical leads. Keywords: Agrochemicals · Bioavailability · Design guidelines · Property profiles · ‘Rule of five’ 1. Introduction tion coefficient, 5 for the number of hydro- ferentiate agrochemicals by type and stage, gen bond donors and 10 for the number of but that differences in property profiles are Identifying the particular balance of intrin- hydrogen bond acceptors. Subsequent pub- subtle rather than dramatic [8]. This work sic potency and bioavailability required by lications, notably from AstraZeneca and has now been further developed and re- a given chemical class to express a desired GlaxoSmithKline, indicated that pharma- viewed to give a comprehensive account of biological effect is a key part of the chem- ceutical leads are less structurally complex the physical and molecular properties of istry design and optimization process. than marketed drugs and their property lim- agrochemicals. Bioavailability in itself provides the chal- its are more constrained [2]. lenge of understanding the balance of mo- Within the agrochemical industry Briggs bility and stability related properties re- was quick to follow up the Lipinski ‘rule of 2. Properties quired to allow expression of activity in five’with ‘ground rules of three’outlined in lead generation screens and on progression a 1997 talk [3]; and in a recent poster Briggs This section gives brief details of the from screen hit to lead series to product. In and co-workers set limits to the physical molecular and physical properties used in practice, chemists, organic, physical and properties of fungicides [4]. The most cited this work. The structures of the fungicide computational, seek to define and exploit work on agrochemicals relating to the Lip- azoxystrobin [9], the herbicide mesotrione these balances in terms of physical and mo- inski ‘rule of five’comes from Tice who fo- [10] and the insecticide thiamethoxam [11] lecular properties. In 1997 Lipinski and co- cused on insecticides and foliar applied her- are given as examples of modern agro- workers at Pfizer published what is widely bicides [5]. Our own initial contributions to chemicals (Fig. 1). regarded the key paper defining physico- understanding property profiles of agro- chemical and structural properties profiles chemicals, based on an analysis of com- 2.1. Partition Coefficients for optimal oral availability of drugs [1]. pounds and physical properties taken from The octanol/water partition coefficients Their work unlocked results and analysis the Pesticide Manual [6] have been pre- (log P oct) used are estimated values, des- from other pharmaceutical companies, sented in talks and posters [7]. In many re- ignated ELOGP, defined as the mean value which further explored the concepts of spects these reported property profiles for from three distinct prediction methods. drug-likeness and lead-likeness in terms of drugs and agrochemicals can be considered These methods are CLOGP based on struc- physical and molecular properties [2]. The broadly similar, the most marked difference tural fragments, AlogP based on atom con- Lipinski ‘rule of five’ placed upper limits being that agrochemicals have a lower tributions and ACD/LogP based on a com- for four properties; namely 500 for molec- number of hydrogen bond donors. In 2002 bination of atom and fragment contribu- ular weight, log P 5 for octanol/water parti- we significantly expanded the scope of tions [12]. By definition, logP oct values studies on agrochemicals to cover all types relate to the un-ionized form of acids and *Correspondence: Mr. E.D. Clarke (insecticides, fungicides and herbicides), bases. Syngenta all stages of the progression stream (input, Jealott’s Hill International Research Centre hit, lead, product) and a diverse range of 2.2. Delta log P Bracknell Berkshire physical and molecular properties. Prelimi- Delta log P is defined as the difference RG42 6EY nary reports of this work concluded that pa- between octanol/water and alkane/water United Kingdom rameters relating to molecular size, atom partition coefficients (∆log P) [13]. In prac- Tel.: + 44 1344 414067 Fax: + 44 1344 455629 types, hydrogen bonding, ionization state, tice this property has been directly predict- E-Mail: [email protected] lipophilicity and aqueous solubility can dif- ed using the LFER equation for ∆log P CROP PROTECTION RESEARCH 732 CHIMIA 2003, 57, No.11 molecular properties were determined for N N O O NO CH NO 2 3 2 compounds in each set [8]. NN O O N 3.2. Data Format and Visualization CN O O O N Cl H C CH SO CH S All data analysis and visualization work 3 3 O 2 3 O has been carried out using Microsoft Excel Azoxystrobin Mesotrione Thiamethoxam 2000. For each dataset as a whole and sub- divided into activity areas, mean, standard Fig. 1. Examples of modern agrochemical products deviation (SD), 90th and 10th percentile val- ues were determined for the nine properties. hexadecane (∆log P = log P octanol – log P using the Daylight programming toolkit Property variations for the different sets of hexadecane) via PC based Absolv software [16] ran a series of SMARTS sub-structur- compounds have also been displayed as [14]. al searches (patterns) against each molecule ‘radar’ plots [8]. To enable this the mean and reported the number of times each sub- values for a given dataset were normalized 2.3. Aqueous Solubility structure occurred in the molecule. In each against the mean and standard deviation of The aqueous solubility of compounds case the number of hits against the the screen input set to give a scaled value was calculated using an in-house program SMARTS pattern (“a” defining an atom in equal to (Dataset Mean – Input Mean)/(In- called ESOL. This method was developed an aromatic system, “[!#6]” defining a non- put SD). Property means identical to the to produce solubility predictions to an ac- carbon atom) was divided by the number of screen input mean are plotted on the 0 ring curacy comparable with Yalkowsky’s gen- heavy (non-hydrogen) atoms and the result of the radar plot. Higher/lower values are eral solubility equation but without the multiplied by 100 to give a percentage. shown as displacements away from/toward need for measured melting points [15]. The the center of the plot. The units of displace- linear model uses four parameters (log P 2.6. Formal Charge at pH 7 ment are the number of input standard de- oct, molecular weight, number of rotatable Acount of formally charged centers was viations (SD) from the input mean. bonds, proportion of heavy atoms defined made using a program kindly provided by as ‘aromatic’by Daylight SMARTS [16]) to Peter Kenny (AstraZeneca, Alderley Park, describe solubility in water (log Sw in ppm). UK) [18]. The program recognizes func- 4. Results The performance of ESOL (R2 = 0.51) was tionality predominantly charged at pH 7 comparable to the GSE (R2 = 0.48) when (e.g. oxyanions, sulphonyls, aliphatic Figs 2 to 5 show the four radar plots judged against measured aqueous solubili- amines, guanidines) regardless of their reg- which give the scaled values obtained for ty values taken from the Pesticide Manual istered protonation state using SMARTS the nine representative parameters for all [6] for 469 agrochemical products; full de- patterns and the Daylight programming compounds in the hit, lead and product sets tails of the ESOL method will be described toolkit [16]. and on division into insecticide, fungicide elsewhere. and herbicide types. These plots highlight the shifts in property profile that can occur 2.4. Hydrogen Bonding 3. Analysis as compounds progress from screen hit, Hydrogen bonding capacity was as- through lead series to agrochemical prod- sessed using two methods. Values obtain This section gives detail of the datasets uct. In Table 1 we give the screen inputs from Absolv [14] for the Abraham descrip- used, data formats and data visualization. values used to normalize the plots and the tors A (H-bond acidity) and B (H-bond ba- means and standard deviations for hits, lead sicity) gave a nuanced description of a mol- 3.1. Datasets series, products and post-1967 products. ecule’s overall hydrogen bonding capacity Four sets of compounds were used in The property ranges for post-1967 agro- while counts of donors (any NH or OH this work, representing the progression chemical products delimited by 10th & 90th group) and acceptors (any double bonded HTS input → HTS hit → lead series → percentiles shown in Table 2 serve to define oxygen or aromatic nitrogen except NH – agrochemical product. These were a 9900 preferred property profiles. All references strong acceptors only) gave a simpler view random subset of the Syngenta company to products for the rest of this paper will re- of the same. It was found that H-bond donor compound collection (HTS inputs), a 6500 fer to post-1967 products unless otherwise count correlated quite well with Abraham’s set of confirmed high throughput in vivo stated. A(R2 = 0.78) while H-bond acceptor count screen actives (HTS hits), 660 compounds showed little correlation with Abraham’s B which justified synthetic work (lead series), (R2 = 0.20). The poor correlation between and the 1380 single organic products listed 5.