Gifnic Logo Forestry Logo
The Food and Environment Research Agency
| Home | Background | Search | Administration |
Executive Summary
The UK Forest Resource
Secondary Metabolites From Trees
Non-Timber Markets For Trees
Extraction Technologies For Tree Metabolites
Adding Value To Tree Metabolites
Further Research
Modelling Tools

Modelling Tools
The Search For Novel Activities For Tree Metabolites Using QSAR Predictive Models
Complementary to the traditional approaches of identifying new uses for natural products (e.g. on-line searches of phytochemical databases) Fera is developing computational models (Quantitative Structure-Activity Relationship models; QSAR) that can predict the activity of a compound based solely on its structure (Price & Watkins 2003). QSAR techniques are routinely used in the discovery of new compounds with desired activities, where the molecules may be pharmaceuticals, protein polymers, catalysts, pesticides etc.
QSAR models have been credited in the discovery of the drug norfloxacin, the herbicides metamitron and bromobutide and the fungicide myclobutinail, which are marketed around the world (Boyd 1990). Fera has also used this technique to identify and predict the activity of tree metabolites as bird repellents that are now the subject of commercial evaluation (Watkins et al. 1999). Conventional strategies for the discovery and screening of natural products are often costly, laborious and technically exacting. QSAR and other computer-aided molecular design approaches adopted by Fera promises to increase the speed and efficiency of the discovery process, drastically reducing developmental costs.
QSAR is proving useful in the natural products area, because plants, despite, their diversity of form, produce many compounds that are structurally similar (e.g. alkaloids). We know what they look like structurally, and if we can measure their activity against a particular biological process, we could predict related molecules with enhanced activity. For example, there is a series of natural compounds related to cinnamic acid, which occurs in fruit trees and deters bullfinches from eating the fruit buds (Figure 1).
Figure 1. Computer-generated simulation of cinnamic acid.
Cinnamic acids are the active principles underlying the resistance of 'Doyenné du Comice' pear trees to feeding damage by bullfinches
Through modelling the effects of 20 or so of these chemicals, we were able to show that cinnamamide was the most effective bird repellent and was also a very good slug repellent (Watkins 1996). Cinnamamide also occurs naturally in bracken and related plants (Figure 2).
Figure 2. Computer-generated simulation of cinnamamide, the most effective repellent in the cinnamic acid family of molecules
Our aim for this project was to describe potential activities for metabolites identified in the UK's commercial tree species. However, due to resource limitations, this element of the project has been reserved for a selected set of tree metabolites and activities: the antimicrobial and insecticidal activities of monoterpenes from Pinus sylvestris.
Fera has developed QSAR models that describe the activities of molecules for a range of applications, In this example (Figure 3) we have developed a QSAR model using published and in-house generated data on a series of 34 monoterpene compounds with a wide range of insecticidal activities (Figure 4). This model provides a robust quantitative prediction of the insecticidal activity of monoterpenes.
Figure 3. Scatter-plot of observed vs. predicted activity of the QSAR model for insecticidal monoterpenes.
A 3 descriptor model was generated using multiple regression analysis techniques and then refined using a simple 2 layer neural network with the same descriptors. A genetic function algorithm was used to optimize the properties of the net, and the same 27 monoterpenes (shown in blue) were used to train the net. As before the remaining 6 (shown in pink) were used as a test set to test the robustness of the net.
Figure 4. Computer-generated simulation of monoterpenes used to construct the insecticide model
The least active compound borneol The most active compound thymol
Using the model we can not only predict the activity of our original 'test set' of 6 herb-derived monoterpenes, but also the activity of monoterpenes from our tree species. For example, Pinene (Figure 5), a monoterpene present in Pinus sylvestris, but not used to construct the mathematical model above, is predicted by the model to have an LD50 against houseflies of 190 mg/kg. This is a weak insecticidal activity and therefore Pinene would not be the best candidate for development as a pesticide.
Figure 4. Pinene found in Pinus sylvestris, (predicted as weak insecticide)
We have also constructed models for the prediction of antibiotic potency of monoterpenes against Escherichia coli (e.g. gastroenteritis), Staphylococcus aureus (e.g. Staphylococcal food poisoning) and Candida albicans (e.g. candidiasis or "thrush"). The potency of three monoterpenes derived from Pinus sylvestris, L-Carvone, β-Myrcene and Citral (Figure 5) is described in Table 1, and shows that L-Carvone from P. sylvestris is predicted to be one of the most potent monoterpenes in killing E. coli. It should be noted that less potent antimicrobial monoterpenes such as linalool are already being extracted from herbs for incorporation into plastic food wraps (Biever 2003), the natural origins of these chemicals being an attractive selling point. However, natural does not mean safe: QSAR models could assist the development of these product applications by predicting those metabolites that are effective and safe (Price & Watkins 2003).
Figure 5. Computer-generated simulations of three monoterpenes form Pinus sylvestris
Table 1. MIC (Minimum Inhibitory Concentrations in parts per million) predicted for three Pinus sylvestris derived monoterpenes for the three potential pathogens, Escherichia coli, Staphylococcus aureus and Candida albicans.
MIC E.coli MIC S.aureus MIC C.albicans
L-Carvone 1000 6000 3000
β-Myrcene 10000 11000 3000
Citral 4000 5000 1000
The most effective herb-derived monoterpenes used in constructing the model were for Perillaldehyde (E. coli, MIC 1000); Citronellol (S. aureus MIC 5000) and Citronellol (C. albicans, MIC 1000).
This technique can be used with any biological or chemical activity so long as a dataset of activity is available or can be generated. Expansion of this approach, with the compiling of a database of computer-generated structures for those metabolites presented in this report, would aid the screening and reduce the development costs for natural tree products. This would provide the first ever structural and property database with which to identify and quantify the efficacy of tree metabolites for use in novel applications.
|   Home   |   Search   |   Top
Reference List
Biever,C. (2003) Herb extracts wrap up lethal food bugs. New Scientist 178, No. 2399, pp. 26
Boyd,D.B. (1990) Successes of computer-assisted molecular design. Reviews in computational chemistry (eds Lipkowitz,K.B. & Boyd,D.B.), pp. 355-371. VCH Publishers Inc., New York
Price, N. R. and Watkins, R. W. (2003) Quantitative structure activity relationships (QSAR) in predicting the environmental safety of pesticides. Pesticide Outlook . In Press
Watkins,R.W. (1996) Efficacy of cinnamamide as a repellent for vertebrate and invertebrate pests. Pesticide Outlook 7, pp. 21-24
Watkins,R.W., Lumely, J. A., Gill, E., Bishop, J., Langton, S. D., MacNicoll, A., Price, N. R., & Drew, M. G. B. (1999) Quantitative structure-activity relationships (QSAR) of cinnamic acid bird repellents. Journal of Chemical Ecology 25, pp. 2825-2845