- Email: [email protected]
In silico screening of drug databases for TSE inhibitors
BioSystems 80 (2005) 117–122
In silico screening of drug databases for TSE inhibitors Stephan Lorenzen ∗ , Mathias Dunkel, Robert Preissner Charite Institute, Institute of Biochemistry, Monbijoustr. 2, 10117 Berlin, Germany Received 21 October 2004; accepted 21 October 2004
Abstract Today, thousands of different chemical compounds are used as drugs for a wealth of different indications. Here, we demonstrate the use of a novel conformational drug database for the search of compounds with a positive influence on Transmissible Spongiform Encephalopathies (TSEs) by using two- and three-dimensional structural similarity to compounds with known effect. Both methods are combined to deduce a list of 16 candidate drugs. The proposal of a small number of putative inhibitors out of about 2300 approved essential drugs allows testing by expensive or time-consuming methods with the advantage that all agents are well known and suitable for use in humans. © 2004 Elsevier Ireland Ltd. All rights reserved. Keywords: Drug database; 2D; 3D; Inhibitor; Prion; Scrapie
1. Introduction Approved drugs have the great advantage that their pharmacological profile, pharmacokinetics, and side effects in humans are well known. Recently, many drugs have been tested successfully against diseases different from their original indication (for reference, see Wermuth, 2004). For example, the antimalaria agent quinacrine has been discussed widely as potential inhibitor of Transmissible Spongiform Encephalopathies (TSE) diseases (see, as an example, Barret et al., 2003) and even been tested with patients by Nakajima et al. (2004). Since the ‘reinvention’ of tradi∗ Corresponding author. Tel.: +49 30 450 528 395; fax: +49 30 450 528 942. E-mail address: [email protected] (S. Lorenzen).
tional drugs also has the advantage of dropping many expensive clinical tests, the detection of secondary effects of therapeutic agents in use is of rising interest. A common task in pharmaceutical research is the search for new lead compounds against diseases that show a greater specificity and/or fewer side effects than already-known agents. A widely used search method is high throughput screening (HTS) of large compound collections. However, since this approach is expensive and time consuming and further on only can be used once a suitable test assay is developed, the in silico design and proposal of new lead structures becomes more important. A central origin of this strategy is the experience that similarities in structures are indicative of similarities in activities of drugs. Thus, a structural search of large compound databases is of great interest.
0303-2647/$ – see front matter © 2004 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.biosystems.2004.10.004
118
S. Lorenzen et al. / BioSystems 80 (2005) 117–122
While a few years ago, chemical databases have only been available to large pharmaceutical companies, the public availability of these tools has risen continuously (Voigt et al., 2001). Today, about two million chemical compounds are available commercially (Bradley, 2002). Here, we show the use of SuperDrug (SD) (Goede et al., 2005; http://bioinf.charite.de/superdrug), a new data base of essential WHO approved drugs (2003), for 2D and 3D search for new lead structures starting from compounds active against prion diseases.
2. Methods Starting with a collection of 17 potent inhibitors experimentally found by Kocisko et al. (2003) using HTS of a collection of 2000 drugs and natural products as lead compounds, we searched for similar drugs in the SuperDrug database. The database consists of about 2300 structures and 111,000 conformers of drugs against all medical indications. As a measure of 2D similarity, we used the Tanimoto coefficient (Delaney, 1996): each compound is characterized by a list of 966 bits representing structural details. The similarity between two substances is calculated as the fraction of identical bits in the structural fingerprints of two chemical compounds. However, this similarity measure is based on the presence and absence of defined chemical groups and residues and can thus, only detect ‘simple’ chemical similarities. A Tanimoto coefficient of 0.85 between two compounds is a more or less reliable indicator of similar activities (Martin et al., 2002). However, compounds with scores in this range are often chemically closely related to each other, so that no new chemical classes can be found. On the other side, a Tanimoto coefficient below 0.70 is a poor indicator. Contrarily, the 3D superposition of structures takes into account the similar distribution of atoms of both structures in space. A strong advantage is the structural similarity of chemically ‘unsimilar’ features, for example, a seven-membered ring can be superimposed with a six-membered ring, whereas the chemical similarity between both structures is low and would thus, not be detected by 2D-descriptor-based searches. Thus, also chemically dissimilar but structurally related drugs could be detected, which often also show activity. 3D
superpositions with a scoring function of s = Pe−rmsd (P: percentage of superposed atoms, rmsd: root mean square deviation) were performed as described by Thimm et al. (2005). In addition to similarities to active agents, similarities to ineffective substances can also be informative. Hits found by the methods described above that have already been tested as inactive or show conspicuous similarity to inactive substances can be omitted from the list, and in addition to criteria for activity, also criteria for inactivity can be derived and used to exclude further compounds. Depending on the application of the drug to be developed, further filtering can be applied, e.g., excluding compounds violating the Lipinski et al. (2001) rule of five or predicted to be non-blood brain barrier permissive. The general approach is shown in Fig. 1. 3. Results Kocisko et al. (2003) experimentally found 17 compounds active against prion propagation in a cell assay.
Fig. 1. Search algorithm for new inhibitors. Starting with known lead compounds, a data base is searched to create a pool of putative drugs. These compounds are compared to known inhibitors and noninhibitors, and drugs with similarities to inactive structures are removed from the list of proposed inhibitors. Combining structural features of ineffective substances with property filtering rules allows the exclusion of further candidates. Drugs surpassing this sieve are proposed as new inhibitors.
S. Lorenzen et al. / BioSystems 80 (2005) 117–122
119
Fig. 2. 2D-search results from Kocisko’s drugs and structural similarities to query compounds (Tanimoto coefficient). The search was performed in SuperDrug (SD) and MicroSource Discovery (MS). Engrayed: ineffective in Kocisko’s assay. Bold, IC50 of ≤10 M in Kocisko’s assay; bold green, IC50 of ≤1 M in Kocisko’s assay and bold red, proposed by us as new agents.
120
S. Lorenzen et al. / BioSystems 80 (2005) 117–122
Of these, 6 are natural products and 11 are drugs of diverse groups besides such already mentioned antimalarial drugs as quinacrine, the authors found phenothiazines which were already discussed as antiprion agents by Korth et al. (2001), and also a steroid (budenoside), lovastatin and the antihistaminic astemizole. Remarkably, quinacrine shows a three-ring system similar to that in phenothiazines, and parts of this system are also conserved in the antimalarial compound astemizole. Thus, 6 of the 11 substances found in this study (quinacrine, four phenothiazins and astemizole) fall into one group. Since the natural compounds identified by Kocisko are large compounds with molecular weights beyond the dimensions of applicable drugs, we did not consider the six natural products in our approach. The drugs identified by Kocisko were used for 2D searches in the SD data base and a new data base containing the 2000 substances screened by Kocisko (MicroSource, MS). For each query substance, we considered the 20 best hits in both data bases (Fig. 2). Interestingly, hits with high scores within the MicroSource (MS) data base often represent substances that are active in Kocisko’s assay: For example, a search with amodiaquine identifies quinacrine, chloroquine, and hydroxychloroquine as best results, which also have been identified experimentally by Kocisko, and prochlorperazine as query leads to other phenothiazines identified by Kocisko. Six of the 11 drugs belonging to the group described above (amodiaquine, prochlorperazine, quinacrine, thioridazine, thiothixene, and triflu-
operazine) are found by searches with other substances belonging to the group. Within the search results of these groups, the density of active substances is quite high. Contrarily, a search with the steroid budesonide led to several other steroids, most of which were tested as ineffective (data not shown). Here, we lack further information on which criteria determine the activity of steroids. As indicated, the 2D search leads mostly to compounds that are similar to the query substances. On the other hand, substances of other chemical classes can be found by the 3D search. As an example, Fig. 3a shows a superposition between prochlorperazine (Fig. 3b) and the antidepressant opipramol (Fig. 3c). The sixmembered ring in the original structure is changed to a seven-membered ring in opipramol, the electron density of the sulphur atom is replaced by a double bond. Similar to the original structure, also the new component contains a piperazine ring. Since the chemical similarity between both structures is low, this hit would not have been detected by pure 2D comparison. Also the ring systems of clopenthixol, methantheline, and rupatadin can be found by 3D superposition with phenothiazins in a similar way. Search results found by 3D screening were inspected visually and added to the list of hits if they appeared promising. The next step is a comparison between the hits and compounds proven to be inactive. As indicated in Fig. 2, a considerable number of hits were themselves not effective in Kocisko’s assay (engreyed). By excluding substances with obvious similarities to ineffective
Fig. 3. (a) The three-dimensional superposition between prochlorperazine (ATC N05AB04, thick sticks) and opipramol (ATC N06AA05, thin sticks) shows striking similarities although (b) a six-membered ring in prochlorperazine is changed to (c) a seven-membered ring in opipramol.
S. Lorenzen et al. / BioSystems 80 (2005) 117–122
121
Fig. 4. Proposed new inhibitors. Sixteen putative drugs against TSE with structural formulas and therapeutic indications.
drugs, the hit list was narrowed further. To obtain a manageable size of the list, in cases of highly 2Dsimilar structures only one representative compound was included in the list. Our final list of substances that we regard as potential prion inhibitors contains 16 substances of different structural families not tested by Kocisko (Fig. 4). The variety of the substances ranges from drugs that are highly similar to known inhibitors, e.g., perazine, which thus has a high probability to show an effect, up to completely new lead compounds such as opipramol or oxomemazine. Remarkably, the therapeutic indication of the proposed compounds often differs from the
indication of the lead, indicating the potential of this approach to propose substances of different structural classes with new properties (‘scaffold hoppers’).
References Barret, A., Tagliavini, F., Forloni, G., Bate, C., Salmona, M., Colombo, L., De Luigi, A., Limido, L., Suardi, S., Rossi, G., Auvre, F., Adjou, K.T., Sales, N., Williams, A., Lasmezas, C., Deslys, J.P., 2003. Evaluation of quinacrine treatment for prion diseases. J. Virol. 77, 8462–8469. Bradley, M.P., 2002. An overview of the diversity represented in commercially available databases. Mol. Divers. 5, 175–183.
122
S. Lorenzen et al. / BioSystems 80 (2005) 117–122
Delaney, J.S., 1996. Assessing the ability of chemical similarity measures to discriminate between active and inactive compounds. Mol. Divers. 1, 217–222. Goede, A., Dunkel, M., Mester, N., Frommel, C., Preissner, R., 2005. SuperDrug: a conformational drug database. Bioinformatics. PMID: 15691861, in press. Kocisko, D.A., Baron, G.S., Rubenstein, R., Chen, J., Kuizon, S., Caughey, B., 2003. New inhibitors of scrapie-associated prion protein formation in a library of 2000 drugs and natural products. J. Virol. 77, 10288–10294. Korth, C., May, B.C., Cohen, F.E., Prusiner, S.B., 2001. Acridine and phenothiazine derivatives as pharmacotherapeutics for prion disease. Proc. Natl. Acad. Sci. U.S.A. 98, 9836–9841. Lipinski, C.A., Lombardo, F., Dominy, B.W., Feeney, P.J., 2001. Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv. Drug Deliv. Rev. 46, 3–26. Martin, Y.C., Kofron, J.L., Traphagen, L.M., 2002. Do structurally similar molecules have similar biological activity. J. Med. Chem. 45, 4350–4358.
Nakajima, M., Yamada, T., Kusuhara, T., Furukawa, H., Takahashi, M., Yamauchi, A., Kataoka, Y., 2004. Results of quinacrine administration to patients with Creutzfeldt-Jakob disease. Dement Geriatr. Cogn. Disord. 17, 158–163. Thimm, M., Goede, A., Hougardy, S., Preißner, R., 2005. Comparison of 2D similarity and 3D superposition. Application to searching a conformational drug database. J. Chem. Inf. Comput. Sci. 44, 1816–1822. Voigt, J.H., Bienfait, B., Wang, S., Nicklaus, M.C., 2001. Comparison of the NCI open database with seven large chemical structural databases. J. Chem. Inf. Comput. Sci. 41, 702– 712. Wermuth, C.G., 2004. Selective optimization of side activities: another way for drug discovery. J. Med. Chem. 47, 1303– 1314. WHO, 2003. The Selection and Use of Essential Medicines. Report of the WHO Expert Committee, 2002 (including the 12th Model list of essential medicines). World Health Organ Tech. Rep. Ser. 914 (i–vi), 1–126.
In silico screening of drug databases for TSE inhibitors Stephan Lorenzen ∗ , Mathias Dunkel, Robert Preissner Charite Institute, Institute of Biochemistry, Monbijoustr. 2, 10117 Berlin, Germany Received 21 October 2004; accepted 21 October 2004
Abstract Today, thousands of different chemical compounds are used as drugs for a wealth of different indications. Here, we demonstrate the use of a novel conformational drug database for the search of compounds with a positive influence on Transmissible Spongiform Encephalopathies (TSEs) by using two- and three-dimensional structural similarity to compounds with known effect. Both methods are combined to deduce a list of 16 candidate drugs. The proposal of a small number of putative inhibitors out of about 2300 approved essential drugs allows testing by expensive or time-consuming methods with the advantage that all agents are well known and suitable for use in humans. © 2004 Elsevier Ireland Ltd. All rights reserved. Keywords: Drug database; 2D; 3D; Inhibitor; Prion; Scrapie
1. Introduction Approved drugs have the great advantage that their pharmacological profile, pharmacokinetics, and side effects in humans are well known. Recently, many drugs have been tested successfully against diseases different from their original indication (for reference, see Wermuth, 2004). For example, the antimalaria agent quinacrine has been discussed widely as potential inhibitor of Transmissible Spongiform Encephalopathies (TSE) diseases (see, as an example, Barret et al., 2003) and even been tested with patients by Nakajima et al. (2004). Since the ‘reinvention’ of tradi∗ Corresponding author. Tel.: +49 30 450 528 395; fax: +49 30 450 528 942. E-mail address: [email protected] (S. Lorenzen).
tional drugs also has the advantage of dropping many expensive clinical tests, the detection of secondary effects of therapeutic agents in use is of rising interest. A common task in pharmaceutical research is the search for new lead compounds against diseases that show a greater specificity and/or fewer side effects than already-known agents. A widely used search method is high throughput screening (HTS) of large compound collections. However, since this approach is expensive and time consuming and further on only can be used once a suitable test assay is developed, the in silico design and proposal of new lead structures becomes more important. A central origin of this strategy is the experience that similarities in structures are indicative of similarities in activities of drugs. Thus, a structural search of large compound databases is of great interest.
0303-2647/$ – see front matter © 2004 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.biosystems.2004.10.004
118
S. Lorenzen et al. / BioSystems 80 (2005) 117–122
While a few years ago, chemical databases have only been available to large pharmaceutical companies, the public availability of these tools has risen continuously (Voigt et al., 2001). Today, about two million chemical compounds are available commercially (Bradley, 2002). Here, we show the use of SuperDrug (SD) (Goede et al., 2005; http://bioinf.charite.de/superdrug), a new data base of essential WHO approved drugs (2003), for 2D and 3D search for new lead structures starting from compounds active against prion diseases.
2. Methods Starting with a collection of 17 potent inhibitors experimentally found by Kocisko et al. (2003) using HTS of a collection of 2000 drugs and natural products as lead compounds, we searched for similar drugs in the SuperDrug database. The database consists of about 2300 structures and 111,000 conformers of drugs against all medical indications. As a measure of 2D similarity, we used the Tanimoto coefficient (Delaney, 1996): each compound is characterized by a list of 966 bits representing structural details. The similarity between two substances is calculated as the fraction of identical bits in the structural fingerprints of two chemical compounds. However, this similarity measure is based on the presence and absence of defined chemical groups and residues and can thus, only detect ‘simple’ chemical similarities. A Tanimoto coefficient of 0.85 between two compounds is a more or less reliable indicator of similar activities (Martin et al., 2002). However, compounds with scores in this range are often chemically closely related to each other, so that no new chemical classes can be found. On the other side, a Tanimoto coefficient below 0.70 is a poor indicator. Contrarily, the 3D superposition of structures takes into account the similar distribution of atoms of both structures in space. A strong advantage is the structural similarity of chemically ‘unsimilar’ features, for example, a seven-membered ring can be superimposed with a six-membered ring, whereas the chemical similarity between both structures is low and would thus, not be detected by 2D-descriptor-based searches. Thus, also chemically dissimilar but structurally related drugs could be detected, which often also show activity. 3D
superpositions with a scoring function of s = Pe−rmsd (P: percentage of superposed atoms, rmsd: root mean square deviation) were performed as described by Thimm et al. (2005). In addition to similarities to active agents, similarities to ineffective substances can also be informative. Hits found by the methods described above that have already been tested as inactive or show conspicuous similarity to inactive substances can be omitted from the list, and in addition to criteria for activity, also criteria for inactivity can be derived and used to exclude further compounds. Depending on the application of the drug to be developed, further filtering can be applied, e.g., excluding compounds violating the Lipinski et al. (2001) rule of five or predicted to be non-blood brain barrier permissive. The general approach is shown in Fig. 1. 3. Results Kocisko et al. (2003) experimentally found 17 compounds active against prion propagation in a cell assay.
Fig. 1. Search algorithm for new inhibitors. Starting with known lead compounds, a data base is searched to create a pool of putative drugs. These compounds are compared to known inhibitors and noninhibitors, and drugs with similarities to inactive structures are removed from the list of proposed inhibitors. Combining structural features of ineffective substances with property filtering rules allows the exclusion of further candidates. Drugs surpassing this sieve are proposed as new inhibitors.
S. Lorenzen et al. / BioSystems 80 (2005) 117–122
119
Fig. 2. 2D-search results from Kocisko’s drugs and structural similarities to query compounds (Tanimoto coefficient). The search was performed in SuperDrug (SD) and MicroSource Discovery (MS). Engrayed: ineffective in Kocisko’s assay. Bold, IC50 of ≤10 M in Kocisko’s assay; bold green, IC50 of ≤1 M in Kocisko’s assay and bold red, proposed by us as new agents.
120
S. Lorenzen et al. / BioSystems 80 (2005) 117–122
Of these, 6 are natural products and 11 are drugs of diverse groups besides such already mentioned antimalarial drugs as quinacrine, the authors found phenothiazines which were already discussed as antiprion agents by Korth et al. (2001), and also a steroid (budenoside), lovastatin and the antihistaminic astemizole. Remarkably, quinacrine shows a three-ring system similar to that in phenothiazines, and parts of this system are also conserved in the antimalarial compound astemizole. Thus, 6 of the 11 substances found in this study (quinacrine, four phenothiazins and astemizole) fall into one group. Since the natural compounds identified by Kocisko are large compounds with molecular weights beyond the dimensions of applicable drugs, we did not consider the six natural products in our approach. The drugs identified by Kocisko were used for 2D searches in the SD data base and a new data base containing the 2000 substances screened by Kocisko (MicroSource, MS). For each query substance, we considered the 20 best hits in both data bases (Fig. 2). Interestingly, hits with high scores within the MicroSource (MS) data base often represent substances that are active in Kocisko’s assay: For example, a search with amodiaquine identifies quinacrine, chloroquine, and hydroxychloroquine as best results, which also have been identified experimentally by Kocisko, and prochlorperazine as query leads to other phenothiazines identified by Kocisko. Six of the 11 drugs belonging to the group described above (amodiaquine, prochlorperazine, quinacrine, thioridazine, thiothixene, and triflu-
operazine) are found by searches with other substances belonging to the group. Within the search results of these groups, the density of active substances is quite high. Contrarily, a search with the steroid budesonide led to several other steroids, most of which were tested as ineffective (data not shown). Here, we lack further information on which criteria determine the activity of steroids. As indicated, the 2D search leads mostly to compounds that are similar to the query substances. On the other hand, substances of other chemical classes can be found by the 3D search. As an example, Fig. 3a shows a superposition between prochlorperazine (Fig. 3b) and the antidepressant opipramol (Fig. 3c). The sixmembered ring in the original structure is changed to a seven-membered ring in opipramol, the electron density of the sulphur atom is replaced by a double bond. Similar to the original structure, also the new component contains a piperazine ring. Since the chemical similarity between both structures is low, this hit would not have been detected by pure 2D comparison. Also the ring systems of clopenthixol, methantheline, and rupatadin can be found by 3D superposition with phenothiazins in a similar way. Search results found by 3D screening were inspected visually and added to the list of hits if they appeared promising. The next step is a comparison between the hits and compounds proven to be inactive. As indicated in Fig. 2, a considerable number of hits were themselves not effective in Kocisko’s assay (engreyed). By excluding substances with obvious similarities to ineffective
Fig. 3. (a) The three-dimensional superposition between prochlorperazine (ATC N05AB04, thick sticks) and opipramol (ATC N06AA05, thin sticks) shows striking similarities although (b) a six-membered ring in prochlorperazine is changed to (c) a seven-membered ring in opipramol.
S. Lorenzen et al. / BioSystems 80 (2005) 117–122
121
Fig. 4. Proposed new inhibitors. Sixteen putative drugs against TSE with structural formulas and therapeutic indications.
drugs, the hit list was narrowed further. To obtain a manageable size of the list, in cases of highly 2Dsimilar structures only one representative compound was included in the list. Our final list of substances that we regard as potential prion inhibitors contains 16 substances of different structural families not tested by Kocisko (Fig. 4). The variety of the substances ranges from drugs that are highly similar to known inhibitors, e.g., perazine, which thus has a high probability to show an effect, up to completely new lead compounds such as opipramol or oxomemazine. Remarkably, the therapeutic indication of the proposed compounds often differs from the
indication of the lead, indicating the potential of this approach to propose substances of different structural classes with new properties (‘scaffold hoppers’).
References Barret, A., Tagliavini, F., Forloni, G., Bate, C., Salmona, M., Colombo, L., De Luigi, A., Limido, L., Suardi, S., Rossi, G., Auvre, F., Adjou, K.T., Sales, N., Williams, A., Lasmezas, C., Deslys, J.P., 2003. Evaluation of quinacrine treatment for prion diseases. J. Virol. 77, 8462–8469. Bradley, M.P., 2002. An overview of the diversity represented in commercially available databases. Mol. Divers. 5, 175–183.
122
S. Lorenzen et al. / BioSystems 80 (2005) 117–122
Delaney, J.S., 1996. Assessing the ability of chemical similarity measures to discriminate between active and inactive compounds. Mol. Divers. 1, 217–222. Goede, A., Dunkel, M., Mester, N., Frommel, C., Preissner, R., 2005. SuperDrug: a conformational drug database. Bioinformatics. PMID: 15691861, in press. Kocisko, D.A., Baron, G.S., Rubenstein, R., Chen, J., Kuizon, S., Caughey, B., 2003. New inhibitors of scrapie-associated prion protein formation in a library of 2000 drugs and natural products. J. Virol. 77, 10288–10294. Korth, C., May, B.C., Cohen, F.E., Prusiner, S.B., 2001. Acridine and phenothiazine derivatives as pharmacotherapeutics for prion disease. Proc. Natl. Acad. Sci. U.S.A. 98, 9836–9841. Lipinski, C.A., Lombardo, F., Dominy, B.W., Feeney, P.J., 2001. Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings. Adv. Drug Deliv. Rev. 46, 3–26. Martin, Y.C., Kofron, J.L., Traphagen, L.M., 2002. Do structurally similar molecules have similar biological activity. J. Med. Chem. 45, 4350–4358.
Nakajima, M., Yamada, T., Kusuhara, T., Furukawa, H., Takahashi, M., Yamauchi, A., Kataoka, Y., 2004. Results of quinacrine administration to patients with Creutzfeldt-Jakob disease. Dement Geriatr. Cogn. Disord. 17, 158–163. Thimm, M., Goede, A., Hougardy, S., Preißner, R., 2005. Comparison of 2D similarity and 3D superposition. Application to searching a conformational drug database. J. Chem. Inf. Comput. Sci. 44, 1816–1822. Voigt, J.H., Bienfait, B., Wang, S., Nicklaus, M.C., 2001. Comparison of the NCI open database with seven large chemical structural databases. J. Chem. Inf. Comput. Sci. 41, 702– 712. Wermuth, C.G., 2004. Selective optimization of side activities: another way for drug discovery. J. Med. Chem. 47, 1303– 1314. WHO, 2003. The Selection and Use of Essential Medicines. Report of the WHO Expert Committee, 2002 (including the 12th Model list of essential medicines). World Health Organ Tech. Rep. Ser. 914 (i–vi), 1–126.