京都大学 / 0048 / 新制・論文博士 / 博士(農学) / 乙第12834号 / 論農博第2798号 / 新制||農||1026(附属図書館) / 学位論文||H26||N4857(農学部図書室) / 31372 / (主査)教授 栗原 達夫, 教授 植田 和光, 教授 平竹 潤 / 学位規則第4条第2項該当 / Doctor of Agricultural Science / Kyoto University / DFAM

Matrix metalloproteinases (MMPs) play a significant role in degrading the extra- cellular matrix in cancer development and metastasis. Overexpression of matrix metalloproteinases in tumour tissues relative to normal tissues has been exploited as a target for peptide-based therapeutics, to improve therapeutic index of currently used agents. The stability of MMP-activated prodrugs in normal tissue or organs is a significant challenge for their success in the clinic. In an in vitro study, the stability of twenty six prodrugs was studied in mouse liver, kidney, lung and tumour homogenates using HPLC and LC/MS. Selected agents were studied in vivo. Each prodrug has a characteristic amino acid sequence with dominant FITC N-terminal end cap. All prodrugs were conjugated to a colchicine derivative (ICT 2552) which is a vascular disrupting agent causing tumour vasculature shutdown and consequently, tumour necrosis. ICT 3146, ICT 3019, ICT 3120 and ICT 3115 prodrugs showed significant stability in normal tissues and considerable activation in certain tumour tissues compared to the lead compound ICT 2588. Also, the selectivity of promising prodrugs to the MMP family was confirmed by using leupeptin (serine, cysteine and threonine protease inhibitor), pepstatin A (aspartate protease inhibitor), phosphoramidon (nepralysin inhibitor), ilomastat (metalloproteinase inhibitor) and BML-P115 (matrix metalloproteinase inhibitor). Moreover, members of the MMP family responsible for cleaving the selected prodrugs were identified using recombinant MMP enzymes. Furthermore, a LC/MS-MS method was developed to specifically detect and quantify MMP-16 protein expression in H460 tumour. MMP- 16 was responsible for the cleavage of ICT 3146 and ICT 3115. Therefore, MMP-activated prodrugs could be a useful therapeutic approach to avoid off-site toxicities of currently used anti-tumour agents. / The full text will be available at the end of the extended embargo: 5th March 2027

Introduction Cytochromes P450 (CYPs) are the major family of enzymes responsible for detoxification and metabolism of a wide range of both endogenous and xenobiotics chemicals in living organisms. The use of CYPs to activate prodrugs to cytotoxins selectively in tumours has been explored including AQ4N, Phortress and Aminoflavone. CYP1A1, CYP1B1, CYP2W1, and CYP4F11 have been identified as expressed in tumour tissue and surrounding stroma at high frequency compared to most normal tissues. Aim is to investigate the differential metabolism of novel chloromethylindoline by high frequency expressed CYPs in tumours. This differential may be exploited to elicit a selective chemotherapeutic effect by metabolising inert small molecules to potent cytotoxins within the tumour environment. Materials and Methods Sensitive and specific LC/MS/MS techniques have been developed to investigate the metabolism of chloromethylindolines. Recombinant enzymes and transfected cell lines were used to investigate the metabolic profiles with a focus on production of the cytotoxic derivatives of chloromethylindolines. Results Detailed metabolic studies show that (1-(Chloromethyl)-1,2-dihydropyrrolo [3,2-e]indol-3(6H)-yl)(5-methoxy-1H-indol-2-yl) methanone (ICT2700) and other chloromethylindolines are converted by CYP1A1 mediated hydroxylation at the C-5 position leading to highly potent metabolites. In vitro cytotoxicity studies showed differentials of up to 1000-fold was achieved between CYP1A1 activated compared to the non-metabolised parent molecules. The reactivity of metabolites of ICT2700 was also explored using glutathione as a nucleophile. The metabolites were identified by a combination of LC/MS and LC MS/MS techniques. Investigations using mouse and human liver microsomes show that a large number of metabolites are created though none were shown to be associated with a potential anticancer effect. Studies focused on CYP2W1 show that this isoform metabolised ICT2706 to a cytotoxic species and a pharmacokinetic study showed a good distribution of ICT2706 into mouse tissues including tumour. However metabolism of ICT2726 by CYP2W1 resulted only in a non-toxic metabolite profile and may have potential as a biomarker for functional CYP2W1 in tissues. Preliminary studies show that palmitic acid hydroxylation is a useful marker of functional CYP4F11. Summary and conclusion The in vitro results show that the chloromethylindolines are a novel class of agent with potential as prodrugs that following specific hydroxylation by CYP1A1 and CYP2W1 are converted to ultra-potent cytotoxins. Other metabolites are also evident which are not cytotoxic. Studies in vivo show that selected chloromethylindolines possess a good pharmacokinetic profile and show potential as prodrug anticancer agents that require activation by CYP1A1 or CYP2W1. The methods, results, progress and suggestions for future work are presented in this thesis.

Synthetic cannabinoids are recreational drugs designed to mimic the effects of Δ9-tetrahydrocannabinol (THC), the main psychoactive component present in cannabis. These drugs exhibit severe toxic effects upon consumption due to their high binding affinity and potency at the cannabinoid receptors (CB1 and CB2). Synthetic cannabinoids have proliferated over the last decade and become a major public health and analytical challenge, critically impacting the clinical and forensic communities. Indazole carboxamide and indole carboxamide class synthetic cannabinoids have been particularly rampant, and are the compound classes most frequently reported to governmental agencies worldwide. However, the metabolic and pharmacological properties of many of these compounds remains unknown. Elucidating these characteristics allows members of the clinical and forensic communities to identify causative agents in patient samples, as well as render conclusions regarding their toxic effects. The aim of this research study was to assess the in vitro Phase I metabolic profile of five synthetic cannabinoids and report the major metabolites identified; compounds evaluated included MDMB-CHNINACA; APP-CHMINACA (PX-3); 5F-APP-PICA (PX-1); 5F-MDMB-PINACA (5F-ADB); and FUB-AMB. These analytes were incubated for 120 minutes with human liver microsomes, followed by analysis of the extracts via ultra high performance liquid chromatography – tandem mass spectrometry (UHPLC-MS/MS). The high-resolution mass spectrometry tool utilized (quadrupole-time of flight mass spectrometry, QTOF) allowed for a thorough characterization of the metabolites, including the assignment of a chemical formula and structure, and accurate mass. The metabolic stability and kinetic profiles of 5F-ADB and FUB-AMB were evaluated by aliquoting the incubation samples at various time points throughout the procedure. It was observed that these compounds were metabolized rapidly, resulting in short half-lives and relatively elevated metabolic clearances. A variety of metabolites were identified for most of the species studied, and this was dependent on the chemical structure of the parent molecule. The major metabolites identified overall for the species were products of amide or ester hydrolysis; hydroxylation (including polyhydroxylation) of the pentyl side chain or cyclohexylmethyl moiety; and oxidative defluorination. It is proposed that these metabolites (especially analyte-specific metabolite) be included in laboratory assay panels to facilitate unequivocal identification of the synthetic cannabinoid agent of interest. For select compounds (5F-ADB and FUB-AMB), authentic forensic human blood samples which screened positive for these analytes were provided by a renowned forensic toxicology laboratory. These samples were tested to verify that the major metabolites identified in the in vitro studies were also present in blood in vivo; the resultant data from the 5F-ADB and FUB-AMB samples showed that the major hydroxylated and hydrolysis metabolite, respectively, were present in greater abundance than the parent molecule, which was most often absent or not present in an appreciable quantity. Additionally, it was observed in the time studies of 5F-ADB and FUB-AMB that the metabolites containing carboxylic acid functional groups were detected in incubation samples longer than the hydroxylated metabolites, potentially indicative of longer detection windows in human samples. These findings have important toxicological implications; many synthetic cannabinoid metabolites, including those identified in this study may have pharmacological activity and contribute to a drug user’s overall impairment profile; identifying them in blood in the absence of parent compound can point to the causative agent. The results demonstrate that it is imperative that synthetic cannabinoid assays screen for known pharmacologically active metabolites; this is particularly important for drugs with short half-lives. The results of this research can be applied to the prediction of metabolic pathways for synthetic cannabinoids as well as non-drug substances with similar structural elements whose metabolic profile has not yet been elucidated, and whose pharmacological activity is currently unknown. Additionally, the results provide reference standard manufacturers and research scientists with further insight into the metabolic products of synthetic cannabinoids and related compounds for the synthesis of materials for the development of laboratory assays. / Chemistry

Introduction: Cytochromes P450 (CYPs) are the major family of enzymes responsible for detoxification and metabolism of a wide range of both endogenous and xenobiotics chemicals in living organisms. The use of CYPs to activate prodrugs to cytotoxins selectively in tumours has been explored including AQ4N, Phortress and Aminoflavone. CYP1A1, CYP1B1, CYP2W1, and CYP4F11 have been identified as expressed in tumour tissue and surrounding stroma at high frequency compared to most normal tissues. Aim is to investigate the differential metabolism of novel chloromethylindoline by high frequency expressed CYPs in tumours. This differential may be exploited to elicit a selective chemotherapeutic effect by metabolising inert small molecules to potent cytotoxins within the tumour environment. Materials and Methods: Sensitive and specific LC/MS/MS techniques have been developed to investigate the metabolism of chloromethylindolines. Recombinant enzymes and transfected cell lines were used to investigate the metabolic profiles with a focus on production of the cytotoxic derivatives of chloromethylindolines. Results: Detailed metabolic studies show that (1-(Chloromethyl)-1,2-dihydropyrrolo [3,2-e]indol-3(6H)-yl)(5-methoxy-1H-indol-2-yl) methanone (ICT2700) and other chloromethylindolines are converted by CYP1A1 mediated hydroxylation at the C-5 position leading to highly potent metabolites. In vitro cytotoxicity studies showed differentials of up to 1000-fold was achieved between CYP1A1 activated compared to the non-metabolised parent molecules. The reactivity of metabolites of ICT2700 was also explored using glutathione as a nucleophile. The metabolites were identified by a combination of LC/MS and LC MS/MS techniques. Investigations using mouse and human liver microsomes show that a large number of metabolites are created though none were shown to be associated with a potential anticancer effect. Studies focused on CYP2W1 show that this isoform metabolised ICT2706 to a cytotoxic species and a pharmacokinetic study showed a good distribution of ICT2706 into mouse tissues including tumour. However metabolism of ICT2726 by CYP2W1 resulted only in a non-toxic metabolite profile and may have potential as a biomarker for functional CYP2W1 in tissues. Preliminary studies show that palmitic acid hydroxylation is a useful marker of functional CYP4F11. Summary and conclusion: The in vitro results show that the chloromethylindolines are a novel class of agent with potential as prodrugs that following specific hydroxylation by CYP1A1 and CYP2W1 are converted to ultra-potent cytotoxins. Other metabolites are also evident which are not cytotoxic. Studies in vivo show that selected chloromethylindolines possess a good pharmacokinetic profile and show potential as prodrug anticancer agents that require activation by CYP1A1 or CYP2W1. The methods, results, progress and suggestions for future work are presented in this thesis.

Em face das recentes exigências das agências regulatórias quanto à aprovação de novos fármacos, os estudos de biotransformação têm-se tornado uma etapa indispensável para a identificação e otimização de compostos bioativos. O objetivo desses estudos é identificar, já nas fases iniciais da descoberta de fármacos, candidatos que apresentam propriedades indesejáveis como a (i) presença de metabólitos ativos ou tóxicos; (ii) inibição de enzimas metabolizadoras; (iii) depuração metabólica inadequada, entre outras. Neste estudo, foi realizada a caracterização metabólica e a identificação de possíveis inibidores das enzimas do citocromo P450 de oito promissores candidatos a fármacos, identificados através de ensaios virtuais como inibidores da TcGAPDH, Cruzaina e TcDHODH, todas do Trypanosoma cruzi, agente causador da doença de Chagas. Esses compostos foram testados contra as três principais isoformas do citrocromo P450: CYP 3A4, CYP 2D6 e CYP2C9. Os valores de IC50 de 1,4 µM e 1,3 µM contra a CYP2C9 foram encontrados para os compostos Nequimed53 e Nequimed125, enquanto o Nequimed42 inibiu a CYP 3A4 com um valor de IC50 de 7,12 µM. Posteriormente foi conduzida a caracterização metabólica dos compostos Nequimed53 e 125 com foco na identificação dos principais metabólitos, sítios de metabolismo e vias de biotransformação através da técnica de LC-ESI-QqTOF-MS. Para ambos os compostos, a biotransformação por microssomas extraídos de fígado de ratos deu-se preferencialmente por uma única via dependente de NADPH. No caso do Nequimed54, o metabólito formado apresentou uma variação da razão m/z de +16, indicando a ocorrência da hidroxilação do composto parental, enquanto que para o composto Nequimed125, o metabólito formado apresentou uma variação da razão m/z de -28, condizente com a perda de um fragmento etila do composto parental. / In the light of recent demands from regulatory agencies for the acceptance of new drugs, the biotransformation studies have become an essential step for the identification and optimization of bioactive compounds. The objective of these studies is to identify compounds that have undesirable properties such as (i) the presence of toxic or active metabolites, (ii) inhibition of metabolizing enzymes, (iii) excessive metabolic clearance, inter alia. In this study we characterized the metabolism and cytochrome P450 inhibition of eight compounds identified by virtual screening as inhibitors of TcGAPDH, Cruzain and TcDHODH which are of interest as targets for intervention in treatment of Chagas Disease. These compounds were tested against cytochrome P450 isoforms 3A4, 2D6 and 2C9. IC50 values of 1.4 µM and 1.3 µM against CYP 2C9 were observed for Nequimed53 and Nequimed125.while Nequimed42 inhibited CYP 3A4 with an IC50 of 7.1 µM. Subsequently, we characterized the in vitro metabolism of Nequimed53 and 125 with a focus on metabolite identification and biotransformation pathways using the LC-ESI-MS-QqTOF technique. For each, the biotransformation by rat liver microsomes occurred by a single NADPH-dependent pathway. For Nequimed54, the observed metabolite [M+16]+ indicated hydroxylation of parent compound. The metabolite [M-28]+ observed for Nequimed125 indicated desethylation of the parent compound.

Submitted by Luciana Ferreira (lucgeral@gmail.com) on 2016-02-23T13:23:39Z No. of bitstreams: 2 Dissertação - Flávia Cristina da Silva - 2015.pdf: 1680001 bytes, checksum: 3d5116e971ca03d4464ee6b64ab4a913 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Approved for entry into archive by Luciana Ferreira (lucgeral@gmail.com) on 2016-02-23T13:26:06Z (GMT) No. of bitstreams: 2 Dissertação - Flávia Cristina da Silva - 2015.pdf: 1680001 bytes, checksum: 3d5116e971ca03d4464ee6b64ab4a913 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) / Made available in DSpace on 2016-02-23T13:26:06Z (GMT). No. of bitstreams: 2 Dissertação - Flávia Cristina da Silva - 2015.pdf: 1680001 bytes, checksum: 3d5116e971ca03d4464ee6b64ab4a913 (MD5) license_rdf: 23148 bytes, checksum: 9da0b6dfac957114c6a7714714b86306 (MD5) Previous issue date: 2015-02-26 / The discovery and development of drugs consist of a complex process, requiring the integration of various strategic areas such as knowledge, innovation, technology, management and high investments in Research, Development and Innovation (RD&I). No drug can be approved for use in humans without first go through extensive studies aimed at ensuring its effectiveness and safety. On the other hand, a drug that inhibits the activity of a metabolic enzyme cytochrome P450 family (CYP450) can affect the pharmacokinetics of other drugs, resulting in drug-drug interactions (DDIs), which potentially lead to side effects and toxic effects. The main oxidative enzymes responsible for drug metabolism have as main representatives CYP450 superfamily, wherein the CYP3A4 isoform is the most important because it is responsible for metabolizing approximately 50% of the drugs on the market. Several computational methods have been developed as a strategy to predict human metabolism in the early stages of research and development of drugs. In silico models of metabolism have advantages such as faster, lower cost and ease of operation when compared to traditional models in vitro and in vivo. The work aimed mainly at the development of Quantitative Relations between models chemical structure and activity / property (QSAR / QSPR) robust and predictive, to identify CYP3A4 substrates and inhibitors. To this were collected, integrated and prepared larger data sets available in the literature substrates and inhibitors of CYP3A4. Several QSAR models were generated and validated for both properties using a workflow that contemplated carefully the recommendations of the Organization for Economic Co-operation Development (OECD). The combination of different descriptors and machine learning methods have led to obtain robust and predictive QSAR models, with correct classification rate (CCR) ranging from 0.65 to 0.83 and 0.69 to 0.89 of coverage, showing a statistically significant values for classification of compounds with high accuracy whether or not substrates of CYP3A4 substrates. The binary Morgan RFgenerated model to classify compounds inhibitors and non-inhibitors also proved highly robust and predictive with sensitivity values of 0.77 and accuracy of 0.76, and the Morgan-RF model multiclass obtained values of 0.68 sensitivity and 0.69 for accuracy. The map of predicted probability proved useful as it could encode major structural fragments to classify compounds inhibitors or not CYP3A4 inhibitors. In conclusion, have been developed and validated many QSAR to predict the interaction with the CYP450 enzyme that may be useful in the early stages of the development of new drugs. The next step is the online availability of the models obtained in LabMol server (http://labmol.farmacia.ufg.br). / A descoberta e o desenvolvimento de fármacos consistem um processo complexo, sendo necessária a integração de várias áreas estratégicas como conhecimento, inovação, tecnologia, gerenciamento e altos investimentos em Pesquisa, Desenvolvimento e Inovação (PD&I). Nenhum fármaco pode ser aprovado para uso em humanos sem que antes passe por extensivos estudos que visem garantir sua eficácia e segurança. Um fármaco que inibe a atividade metabólica de uma enzima da família citocromo P450 (CYP450), pode afetar a farmacocinética de outros fármacos, resultando em interações fármaco-fármaco (DDIs), que podem conduzir potencialmente a efeitos colaterais e tóxicos. As principais enzimas oxidativas responsáveis pelo metabolismo de fármacos possuem como principais representantes a superfamília CYP450, em que a isoforma CYP3A4 é a mais importante, pois é responsável por metabolizar aproximadamente 50 % dos fármacos disponíveis no mercado. Diversos métodos computacionais têm sido desenvolvidos como estratégia para predizer o metabolismo humano nos primeiros estágios de pesquisa e desenvolvimento de fármacos. Modelos in silico do metabolismo apresentam vantagens como maior rapidez, menor custo e maior facilidade de operação, quando comparados aos modelos tradicionais in vitro e in vivo. O trabalho teve como objetivo central o desenvolvimento de modelos de Relações Quantitativas entre estrutura química e atividade/propriedade (QSAR/QSPR) robustos e preditivos, visando identificar substratos e inibidores de CYP3A4. Para isso, foram compilados, integrados e preparados os maiores conjuntos de dados disponíveis na literatura de substratos e inibidores de CYP3A4. Vários modelos de QSAR foram gerados e validados para ambas as propriedades usando um fluxo de trabalho que contemplou criteriosamente as recomendações da Organization for Economic Co-operation Development (OECD). A combinação de diferentes descritores e métodos de aprendizado de máquina levaram a obtenção de modelos QSAR robustos e consistentes, com taxa de classificação correta (CCR) que variam entre 0,65-0,83 e cobertura de 0,69-0,89,demonstrando valores estatisticamente significativos para classificação com alta precisão de compostos em substratos ou não substratos de CYP3A4. O modelo Morgan-RF binário gerado para classificar compostos em inibidores e não inibidores se mostraram também altamente robusto e preditivo com valores de sensibilidade de 0,77 e acurácia de 0,76, e o modelo Morgan-RF multiclasse obteve valores de 0,68 para sensibilidade e 0,69 para acurácia. O mapa de probabilidade predita se mostrou útil, pois conseguiu codificar fragmentos estruturais importantes para classificar compostos em inibidores ou não inibidores de CYP3A4. Como conclusões foram desenvolvidos e validados diversos modelos de QSAR para prever a interação com a enzima CYP450 que podem ser úteis nos estágios iniciais do desenvolvimento de novos fármacos. O próximo passo será a disponibilização online dos modelos obtidos no servidor do LabMol (http://labmol.farmacia.ufg.br).

CYP2D6 metabolizes codeine to morphine, the active analgesic metabolite. Variation in brain CYP2D6 activity may affect brain morphine levels after codeine administration and thereby influence analgesia. We investigate the effect of inhibiting brain CYP2D on codeine-induced analgesia. METHODS: Rats received intracerebroventricular (i.c.v.) injections of CYP2D inhibitors or vehicle controls. Rats were then given subcutaneous codeine injections and analgesia was measured with the tail-flick test. Morphine and codeine concentrations in brain and plasma were measured. CYP2D activity in brain and liver were assessed in vitro. RESULTS: Compared to vehicle treatment, i.c.v. inhibitor treatments resulted in lower codeine-induced analgesia, lower morphine levels in brain but not in plasma after codeine injections, and lower CYP2D activity in brain membranes but not in liver microsomes. CONCLUSIONS: Inhibiting brain CYP2D reduces codeine’s metabolism to morphine, resulting in less analgesia. Variation in brain CYP2D6 activity may influence response to codeine and other CYP2D6 substrates.

CYP2D6 metabolizes codeine to morphine, the active analgesic metabolite. Variation in brain CYP2D6 activity may affect brain morphine levels after codeine administration and thereby influence analgesia. We investigate the effect of inhibiting brain CYP2D on codeine-induced analgesia. METHODS: Rats received intracerebroventricular (i.c.v.) injections of CYP2D inhibitors or vehicle controls. Rats were then given subcutaneous codeine injections and analgesia was measured with the tail-flick test. Morphine and codeine concentrations in brain and plasma were measured. CYP2D activity in brain and liver were assessed in vitro. RESULTS: Compared to vehicle treatment, i.c.v. inhibitor treatments resulted in lower codeine-induced analgesia, lower morphine levels in brain but not in plasma after codeine injections, and lower CYP2D activity in brain membranes but not in liver microsomes. CONCLUSIONS: Inhibiting brain CYP2D reduces codeine’s metabolism to morphine, resulting in less analgesia. Variation in brain CYP2D6 activity may influence response to codeine and other CYP2D6 substrates.

Pagrindinis šio darbo tikslas buvo kiekybinio struktūros ir aktyvumo ryšio modelių, prognozuojančių su vaistų metabolizmu susijusias savybes, kūrimas. Modeliai, prognozuojantys CYP3A4 slopinimą ir žmogaus kepenų mikrosomų katalizuojamo metabolizmo regioselektyvumą, buvo sukurti naudojant GALAS (angl. Global, Adjusted Locally According to Similarity; Globalus, lokaliai pakoreguotas pagal panašumą) modeliavimo metodą, kuris geba įvertinti prognozės patikimumą, taip apibrėždamas modelio pritaikymo sritį. Sukurtų modelių prognozės buvo tikrinamos naudojant eksperimentinius naujų cheminių junginių duomenis. Visų globalių modelių prognozės gerėjo po korekcijų pagal panašumą, o neteisingų spėjimų skaičius buvo ženkliai mažesnis tarp aukšto patikimumo prognozių. Visgi daugiau nei pusė išorinių duomenų nepatenka į šių modelių pritaikymo sritį. GALAS modeliai gali būti gana paprastai apmokomi, pridedant naujus duomenis į lokalią modelio dalį ir apskaičiuojant reikiamą korekciją. Po tokios apmokymo procedūros CYP3A4 slopinimo modelis prisitaikė prie PubChem duomenų bazės cheminių junginių ir taip pat prie vaistų, turinčių naują cheminį karkasą. Pridėjus naujų junginių ir apmokius regioselektyvumo modelį, jis pradėjo prognozuoti naujas metabolizmo vietas. Pastarasis modelis taip pat buvo pritaikytas atskirų fermentų katalizuojamo metabolizmo prognozavimui. / The main objective of this study was the development of QSAR models for drug metabolism-related properties. Novel GALAS (Global, Adjusted Locally According to Similarity) modeling method was used, which is a combination of baseline global QSAR model and local similarity based corrections. GALAS modeling method allows forecasting the reliability of prediction thus defining the model applicability domain. Models predicting CYP3A4 inhibition and regioselectivity of metabolism in human liver microsomes were developed and validated using external test sets. In all cases the baseline models already showed acceptable results, and the overall accuracy of predictions increased after the similarity based corrections. Moreover, the numbers of mispredictions reduced significantly when only results of higher reliability were taken into account. However, the original models are applicable only for less than a half of external datasets. Since the similarity correction procedure of GALAS modeling method allows simple model training, the possibility to expand the applicability domain has been tested. The CYP3A4 inhibition model was successfully adapted to PubChem data and compounds with a novel chemical scaffold. After training the regioselectivity model new metabolism sites could be identified in compounds of new chemical class. Moreover, this model was adapted for human cytochrome P450 isoform profiling.