* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download FORUM SERIES Genetic Toxicity Assessment
Environmental impact of pharmaceuticals and personal care products wikipedia , lookup
Compounding wikipedia , lookup
Drug interaction wikipedia , lookup
Polysubstance dependence wikipedia , lookup
Pharmaceutical marketing wikipedia , lookup
Neuropsychopharmacology wikipedia , lookup
Environmental persistent pharmaceutical pollutant wikipedia , lookup
Toxicodynamics wikipedia , lookup
Neuropharmacology wikipedia , lookup
Prescription costs wikipedia , lookup
Pharmacognosy wikipedia , lookup
Prescription drug prices in the United States wikipedia , lookup
Pharmacokinetics wikipedia , lookup
Pharmaceutical industry wikipedia , lookup
Pharmacogenomics wikipedia , lookup
Theralizumab wikipedia , lookup
TOXICOLOGICAL SCIENCES 96(1), 16–20 (2007) doi:10.1093/toxsci/kfl191 Advance Access publication December 28, 2006 FORUM SERIES Genetic Toxicity Assessment: Employing the Best Science for Human Safety Evaluation Part I: Early Screening for Potential Human Mutagens David Jacobson-Kram*,1 and Joseph F. Contrera† *Office of New Drugs; and †Office of Pharmaceutical Science, Center for Drug Evaluation and Research, U.S. Food and Drug Administration, Silver Spring, Maryland 20993 Received September 18, 2006; accepted November 28, 2006 Results of genetic toxicology tests are used by FDA’s Center for Drug Evaluation and Research as a surrogate for carcinogenicity data during the drug development process. Mammalian in vitro assays have a high frequency of positive results which can impede or derail the drug development process. To reduce the risk of such delays, most pharmaceutical companies conduct early non-GLP (good laboratory practices) studies to eliminate drug candidate with mutagenic or clastogenic activity. Early screens include in silico structure activity assessments and various iterations of the ultimate regulatory mandated GLP studies. Key Words: mutagenesis; clastogenesis; structure activity; bacterial mutation. Phase 1 clinical trails for new drugs are often performed in healthy volunteers. For people involved in these early trials, there is no risk/benefit assessment as would exist for patients given an approved drug or patients in later clinical trials. Risks to healthy subjects must be carefully assessed and must be exceedingly low. With only rare exceptions, carcinogenicity data for a new drug are submitted with the New Drug Application. The consequence of this timing is that hundreds or thousands or individuals will have been exposed to the drug, generally on a repeated basis and at pharmacological doses without understanding potential carcinogenicity. To protect subjects in clinical trials from exposure to potential carcinogens, Food and Drug Administration (FDA) requires sponsors to submit data from genetic toxicology studies prior to beginning ‘‘first in human (FIH)’’ studies. The required tests are delineated in International Committee on Harmonization (ICH) guidelines S2A and S2B which inform on methods for 1 To whom correspondence should be addressed at the Center for Drug Evaluation and Research, 10903 New Hampshire Avenue, Silver Spring, MD 20993. Fax: (301) 796-9856. E-mail: [email protected]. Published by Oxford University Press 2006. conducting the assays and the required battery of tests, respectively (ICH S2A, 1995; ICH S2B, 1997). S2B specifies that prior to FIH studies, a bacterial gene mutation assay and an assay for chromosomal damage in mammalian cells in vitro are required. A test for chromosome damage in rodent bone marrow is required prior to phase 2. In practice, however, most drug sponsors perform all three assays prior to phase 1. A typical battery would consist of an (1) Salmonella mutagenicity test, (2) a chromosomal aberration assay performed either in CHO cells or human peripheral blood lymphocytes or the mouse lymphoma thymidine kinase gene mutation assay, and (3) a bone marrow micronucleus test in mouse or rat bone marrow. One could rightfully argue that this battery will not detect all carcinogens. If fact, approximately two-thirds of drugs which give positive responses in rodent cancer bioassays are negative in genetic toxicology studies (Snyder and Green, 2001). Examination of the list of drugs judged to be nongenotoxic carcinogens shows that most induce their effects via exaggerated pharmacological responses, immune suppression, or hormonal imbalance. Nongenotoxic carcinogens generally require chronic exposure to induce tumors, and cessation of exposure generally is sufficient to prevent tumor development. Because so many phase 1 clinical trials are performed in healthy subjects, a positive response in a genetic toxicology assay can be a serious liability in the progression of a drug’s clinical development. This often leads to the imposition of a clinical hold for repeated dose clinical studies unless or until the sponsor is able to demonstrate that human exposure does not present a risk to trial participants. Even if the trial moves forward, the positive genotoxicity result will be shown in the informed consent form and ultimately in the drug label. In early 2006, the Center for Drug Evaluation and Research (CDER) published a guidance entitled ‘‘Guidance for Industry and Review Staff- Recommended Approaches to Integration of Genetic Toxicology Study Results’’ (http://www.fda.gov/cder/ EARLY SCREENING FOR POTENTIAL HUMAN MUTAGENS guidance/6848fnl.htm). The guidance discusses various approaches for dealing with a positive study and these include assessment of the weight of evidence, consideration of the mode-of-action, and performance of additional tests to clarify a positive or equivocal result. Clearly, the best situation from a public health perspective and from an efficient drug development model is to move forward with drugs that are clearly negative for potential genetic toxicity. Most pharmaceutical companies have long recognized this and have developed strategies to avoid the liabilities associated with positive genotoxicity responses. 17 and development. QSAR methods are also being used to help qualify and assess the potential risk of contaminants and degradants in drug products at FDA-CDER, the safety of food additives and food contact substances at FDA-Center for Food Safety and Applied Nutrition (Bailey et al., 2005), and environmental hazards (Environmental Protection Agency [EPA], 2006). From a resource, cost, and timeliness perspective, it is not feasible to experimentally test the millions of chemicals in industry combinatorial databases, and QSAR screening can provide a ‘‘virtual’’ alternative to laboratory testing. The ‘‘In Silico First’’ Paradigm for Pharmaceuticals STRUCTURE-ACTIVITY SCREENING Recent advances in the computer technology and chemoinformatics have made available additional resources for safety assessment such as structure similarity searching and improved quantitative structure-activity relationship (QSAR) software models. Essential for effective and predictive QSAR models is a chemically diverse training data set of toxicological information linked to chemical structures. The use of a digital representation of chemical structure (e.g., the MDL molfile or Simplified Molecular Input Line Entry System code) as the primary substance identification in the training database makes possible the identification of clusters of substances having similar chemical structure/substructural features and similar pharmacological or toxicological activities. Toxicological information may be available for compounds that are related to a test compound and may provide some insight on the possible properties of the test compound (Fig. 1). The application of these methods is rapidly expanding and gaining in credibility. In silico QSAR predictive toxicology offers a rapid and cost effective initial screening for potential drug candidate molecules to identify and prioritize compounds for further testing A paradigm shift to utilize in silico testing first to determine a biological testing priority is underway (McGee, 2005). In the pharmaceutical industry, computational toxicology is now being used as a sentinel tool for the early assessment of the toxicological potential of candidate molecules in lead selection and drug discovery (Pearl et al., 2001). For pharmaceuticals, in silico methods are useful decision support tools that compliment traditional animal and laboratory based testing. In the case of pharmaceutical impurities that impart no benefit to the consumer, in silico screening for mutagenicity, carcinogenicity, and developmental toxicity is a useful tool to assess potential toxicological risk, whether further testing is necessary, and the nature of confirmatory laboratory studies that would be needed (Kruhlak et al., 2006). The FDA-CDER, Office of Pharmaceutical Science, Informatics and Computational Safety Analysis Staff (ICSAS) is an applied regulatory research unit within the FDA that has been developing comprehensive electronic relational databases of FDA regulated substances that are linked to chemical structure. (http://www.fda.gov/cder/Offices/OPS_IO/default.htm). To achieve this goal, ICSAS is engaged in Cooperative Research and Development Agreements (CRADAs) with a number of software developers (MultiCASE, Inc., Multicase.com, MDL Information Systems, Inc., MDL.com and Leadscope, Inc., LeadScope.com) to extract and compile toxicology data into an organized, electronically accessible format and to improve the predictive performance of computational toxicology software to meet the goals of the FDA Critical Path initiative to improve regulatory and drug development tools (U.S. FDA, 2004). MultiCASE MC4PC and MDL-QSAR software modules are being developed and distributed publicly by the CRADA partners. These software platforms are being used by the pharmaceutical industry as an initial screen in the compound selection process in drug development. QSAR modules developed under these CRADAs are now available for many endpoints including mutagenicity, rodent carcinogenicity, reproductive and developmental toxicity. Structural Alerts and In Silico Prediction of Mutagenicity FIG. 1. Identifying a Test Compound and Estimating Potential Toxicity Using Structural Similarity. Structural alerts are defined as molecular functionalities (structural features) that are associated with toxicity, and their 18 JACOBSON-KRAM AND CONTRERA presence in a molecular structure alerts the investigator to the potential hazard of a chemical. Structural alerts have historically been identified and applied by human experts in the field of toxicology. The structural alerts or toxicophores for Salmonella positive mutagenic compounds developed by Ashby and Tennant (1991) can be considered human expert generated structure-activity relationships (SAR). Predictive toxicology software can be classified as either qualitative rule-based SAR or QSAR software. Softwares such as DEREK (Deductive Estimation of Risk from Existing Knowledge) and Oncologic are SAR programs that incorporate rules associating toxicophores with mutagenicity that were developed by panels of human experts. In contrast, QSAR softwares are statistically based programs that produce computer generated equations (models) relating chemoinformatic information such as molecular descriptors, physical chemical attributes and other substructural molecular attributes with toxicity. This approach is multidimensional and considers many more molecular parameters than simple toxicophores in a molecule. Major QSAR softwares currently used by the pharmaceutical and chemical industry include MC4PC (MultiCASE Inc.), MDL-QSAR (MDL Information Systems), and TOPKAT (Accelrys, Inc., Accelrys.com). Predictive Models for Mutagenicity FDA-CDER–developed genetic toxicity models are based upon results of toxicology studies conducted by the National Institute of Environmental Health Sciences National Toxicology Program (NTP) (Gold and Zeiger, 1997), studies in FDA/ CDER archives, and study data collected by the EPA and incorporated into the National Institutes of Health, National Library of Medicine (2005) GENETOX database. A weight-of-evidence scoring system was employed for mutagenicity and other genetic toxicity data where study results were assigned activity scores ranging from a low of 10 units to a high of 80 units. Chemicals with multiple study results for a single test system were assigned a cumulative activity score based on the degree of concordance between those results (Matthews et al., 2006a,b). For example, a test compound that was tested five times in a particular assay, and four of those results were positive but one was negative, would be assigned a relatively high activity score (55 units) reflecting the high degree of confidence in the positive findings and low degree of confidence in the single negative finding. This is consistent with the estimated interlaboratory error rate for reproducibility of Salmonella test data from the NTP that was estimated to be 85% (Piegorsch and Zeiger, 1991). For QSAR modeling purposes, weight-of-evidence scoring allows the computer to more readily distinguish between the varying degrees of confidence in training set data and results in increased overall predictive performance compared to other approaches. In support of the applicability of QSAR models, essentially all of the Salmonella positive mutagenic com- pounds with structurally alerting molecular features identified by Ashby and Tennant (1991) were correctly predicted as highrisk mutagens by MDL-QSAR employing E-state descriptors (Contrera et al., 2005a), and MC4PC using molecular fragment structural features (Matthews et al., 2006b). QSAR offers many of the benefits of conventional human expert structural alert analysis with greatly expanded potential. QSAR can perform highly rapid analyses of large numbers of chemicals and can process multiple toxicological and physical chemical parameters. Validation studies have been performed on both the MC4PC and MDL-QSAR Salmonella mutagenicity modules (Contrera et al., 2005a; Matthews et al., 2006b). The modules have good sensitivity and have been optimized by FDA/CDER to provide high specificity (> 75%) and low false positive rates (< 15%), Salmonella mutagenicity, and other endpoints. The utility and accuracy of a model are dependent on the domain of applicability or ‘‘coverage’’ of a particular model that is a function of the molecular characteristics of the test molecule relative to the molecules in the training data set. If a test molecule is not well represented in the training data molecular library, the test molecule will be outside of the domain of the model and will have a poor statistical probability of an accurate prediction. Various metrics have been developed to define the position of a test chemical relative to a model’s chemical space (Tropsha et al., 2003). Another QSAR limitation is related to the molecular attributes of a test molecule. Molecules such as inorganic chemicals, large organic molecules (> 1200 Da) or polymers, fibers, salts, organometallic chemicals, gases, and complex mixtures of chemicals are currently unsuitable for QSAR analysis. These are not limiting factors for human expert based SAR methods such as DEREK. FDA-CDER Pilot New Molecular Entity Screening Program A large-scale external validation study is underway in which all pharmaceuticals submitted to CDER in 2005 including new molecular entities will undergo QSAR predictive toxicology screening using CDER-developed MC4PC and MDL-QSAR models. The predictive performance of the QSAR models will be evaluated to determine their utility as additional decision support tools for reviewers on toxicological endpoints such as carcinogenicity, genetic toxicity, and reproductive and developmental toxicity. SCREENING IN BACTERIAL MUTATION ASSAYS Most pharmaceutical companies screen drug candidates early in development for potential mutagenicity in a bacterial reverse mutation assay. Positive results in such an assay are generally sufficient to forgo further development. Often, pharmaceutical companies will have back-up molecules in case the lead structure is found to have toxic potential. These early tests are non Good Laboratory Practices (GLP) and EARLY SCREENING FOR POTENTIAL HUMAN MUTAGENS require only small amounts of test material, often on the order of 20–150 mg. Different test strategies have evolved. For example, many companies will screen using two bacterial tester strains, most often TA98 and TA100. The assay is performed in the presence and in the absence of rat liver S9 up to 5 mg of test material per plate, often using single plates. Generally, these two strains are sufficient to detect most Salmonella mutagenicity positive chemicals. Other approaches include use of all tester strains, with and without S9 but with a lower top dose, for example 500 lg per plate. This strategy conserves test material but risks observing a positive response at higher doses in the GLP study. Some companies use all tester strains with doses up to 5 mg per plate but only in the presence of S9 since compounds uniquely positive in the absence of S9 are rare (E. Zeiger, unpublished observations). Use of the ‘‘spot test’’ is a qualitative approach to screening for bacterial mutation. The solubulized test material is applied in a small well in the center of a plate containing selective agar and the tester strain. The material diffuses out into the agar and forms a concentration gradient. If the chemical induces reverse mutations, a ring of revertants will form around the well. In this way, only two plates (with and without S9) are required for each tester strain. Yet another approach involves the use of the Ames II assay. This variation involves the use of a liquid microtiter system and uses the traditional strain TA98 to detect frame shift mutations and ‘‘TAMix’’ (Fluckiger-Isler et al., 2004). The latter is a mixture of six Salmonella strains used to detect base pair substitutions; these strains are patented and must be licensed for use. While the strategies between companies can vary, all these approaches have been found to be highly predictive of the ultimate ICH/GLP compliant bacterial mutation assay. This is not terribly surprising since the screening tests generally utilize the same tester stains as the regulatory-required test. A confounding variable during this early phase of screening is that the candidate drug substance can contain significant levels of impurities. A ‘‘false positive’’ could result from a mutagenic impurity resulting in the discontinuation of development of a potentially valuable drug. SCREENING FOR CLASTOGENIC ACTIVITY Positive results for bacterial mutation and in vivo chromosome damage assays are relatively rare in Investigational New Drug submissions to CDER. Salmonella mutagenicity– positive compounds are effectively weeded out using the approaches described above. In addition, both the bacterial mutation assay and the in vivo chromosome damage assays have relatively high specificities and low sensitivities (low frequencies false positives) so that positive results are less common (Kirkland et al., 2005). The in vitro mammalian cell assays, chromosomal aberration and mouse lymphoma, have high sensitivities and low specificities so that positive results 19 are much more common. Nevertheless, even though many of these results may be ‘‘false positives’’ in that they are later found not to be carcinogenic or perhaps carcinogenic for a different reason, a positive result in an in vitro mammalian cell assay can delay or possibly derail a drug development program. Non-GLP screening for potential clastogenic activity is also commonly done by many pharmaceutical companies, especially large pharma. Performing an abbreviated, non-GLP in vitro metaphase assay is an option but requires significant amounts of test material (100–500 mg), is relatively expensive ($4000–$5000), and is time (3–4 weeks) and labor intensive. More commonly, drug developers rely on in vitro assessment of micronucleus induction using CHO cells as the target. In this assay, cells are grown and treated in slide chambers and fixed in situ. Treatment in the absence of S9 is generally for 24 h to mimic the protracted exposure in the GLP chromosomal aberration assay and 4 h in the presence of S9. Cells are also treated with cytochalasin B to generate binucleated cells so that only those cells that divided during the treatment are scored for the presence of micronuclei. Unless limited by solubility or toxicity, the high dose in such an assay is set at 5 mg/ml. While still not a ‘‘high-throughput’’ assay, the in vitro micronucleus assay can generally be performed with 50–100 mg of test material. Since the endpoint in this assay is different from that evaluated in the GLP in vitro metaphase analysis, the correlation of the two assays has been investigated (Miller et al., 1997). The percentage concordance is approximately 79%. One reason for discordance is that the micronucleus assay detects aneugenic materials while metaphase assays generally do not. HIGH-THROUGHPUT SCREENS Advances in combinatorial chemistry have made possible the generation of thousands of unique chemical structures in very short time frames. While the assays described above are useful, they cannot be considered high throughput. More facile assays have been developed in an attempt to fill this gap. For example, Billinton et al. (1998) developed a yeast assay in which expression of a green fluorescent protein is linked to the upregulation of the DNA damage responsive RAD54 gene. Subsequent validation studies have assessed the utility of this assay (Van Gompel et al., 2005). Recently, a variation of this assay using a human cell line, ‘‘GreenScreen HC GADD45aGFP genotoxicity assay’’ has been published (Hastwell et al., 2006). This test uses the human lymphoblastoid cell line TK6 in which the promotor for a DNA damage–inducible gene, GADD45a, is linked a green fluorescent protein gene. The use of the human cell lines permits testing both in the presence and absence of metabolic activation. In a validation study of 75 well-characterized genotoxic and nongenotoxic chemicals, the assay appeared to have good sensitivity and specificity. 20 JACOBSON-KRAM AND CONTRERA OMIC STUDIES ICH S2B. (1997). Genotoxicity: A standard battery for genotoxicity testing of pharmaceuticals. Available at: www.ich.org. Accessed January 1, 2007. Intuitively, one might expect that changes in gene expression following exposure to a mutagenic agent would be a more sensitive endpoint than the quantification of a specific locus mutation. In fact, the Health and Environmental Sciences Institute organized a collaborative study in which expression arrays were examined after exposure to 12 direct acting mutagens (Pennie et al., 2004). Contrary to expectations, mutations at the thymidine kinase locus and levels of DNA adducts were much more sensitive endpoints than changes in gene expression. Indeed, significant changes in expression were only seen after plating efficiency (a measure of cytotoxicity) declined to 1.3% of control. While these types of studies may help shed light on mechanisms of mutagenesis, they currently do not appear to be of value in drug screening. Kirkland, D., Aardema, M., Henderson, L., and Muller, L. (2005). Evaluation of the ability of a battery of three in vitro genotoxicity tests to discriminate rodent carcinogens and non-carcinogens I. Sensitivity, specificity and relative predictivity. Mutat. Res. 584, 1–256. REFERENCES Ashby, J., and Tennant, R. W., (1991). Definitive relationships among chemical structure, carcinogenicity and mutagenicity for 301 chemicals tested by the U.S. NTP. Mutat. Res. 257, 229–306. Bailey, A. B., Chanderbhan, R., Collazo-Braier, N., Cheeseman, M. A., and Twaroski, M. L. (2005). The use of structure-activity relationship analysis in the food contact notification program. Regul. Toxicol. Pharmacol. 42, 225–235. Billinton, N., Barker, M. G., Michel, C. E., Knight, A. W., Heyer, W.-D., Goddard, N. J., Fielden, P. R., and Walmsley, R. M. (1998). Development of a green fluorescent protein reporter for a yeast genotoxicity biosensor. Biosens. Bioelectron. 13, 831–838. Contrera, J. F., Matthews, E. J., Kruhlak, N. L., and Benz, R. D. (2005). In silico screening of chemicals for bacterial mutagenicity using electrotopological E-state indices and MDL QSAR software. Regul. Toxicol. Pharmacol. 43, 313–323. Environmental Protection Agency (EPA). (2006). New Chemicals Program ecological structure activity relationships. Available at: http://www.epa.gov/ oppt/newchems/tools/21ecosar.htm. Accessed January 1, 2007. Fluckiger-Isler, S., Baumeister, M., Braun, K., Gervais, V., Hasler-Nguyen, N., Reimann, R., Van Gompel, J., Wunderlich, H. G., and Engelhardt, G. (2004). Assessment of the performance of the Ames II assay: a collaborative study with 19 coded compounds. Mutat. Res. 558, 181–197. Gold, L. S., and Zeiger, E. (1997). Handbook of Carcinogenic Potency and Genotoxicity Databases. CRC Press, Boca Raton, FL. Hastwell, P. W., Chai, L. L., Roberts, K. J., Webster, T. W., Harvey, J. S., Rees, R. W., and Walmsley, R. M. (2006). High-specificity and high-sensitivity genotoxicity assessment in a human cell line: validation of the GreenScreen HC GADD45a-GFP genotoxicity assay. Mutat. Res. 607, 160–175. ICH S2A. (1995). Genetic toxicity studies: Guidance on specific aspects of regulatory tests for pharmaceuticals. Available at: www.ich.org. Accessed January 1, 2007. Kruhlak, N. L., Contrera, J. F., Benz, R. D., and Matthews, E. J. (2007). Progress in QSAR toxicity screening of pharmaceutical impurities and other FDA regulated products. Adv. Drug Deliv. Rev. (in press). Matthews, E. J. Kruhlak, N. L., Cimino, M. C., Benz, R. D., and Contrera, J. F. (2006a). An analysis of genetic toxicity, reproductive and developmental toxicity, and carcinogenicity data: I. Identification of carcinogens using surrogate endpoints. Regul. Toxicol. Pharmacol. 44, 83–96. Matthews, E. J., Kruhlak, N. L., Cimino, M. C., Benz, R. D., and Contrera, J. F. (2006b). An analysis of genetic toxicity, reproductive and developmental toxicity, and carcinogenicity data: II. Identification of genotoxicants and reprotoxicants using in silico methods. Regul. Toxicol. Pharmacol. 44, 97–110. Miller, B., Albertini, S., Locher, F., Thybaud, V., and Lorge, E. (1997). Comparative evaluation of the in vitro micronucleus test and the in vitro chromosome aberration test: industrial experience. Mutat. Res. 392, 45–59. National Institutes of Health, National Library of Medicine. (2005). U.S. Environmental Protection Agency, Office of Prevention, Pesticides and Toxic Substances—GENE-TOX Agent Registry. Online computer database— TOXNET file. U.S. Department of Health and Human Services. Bethesda, MD 20894. McGee, P. (2005). Modeling success with in silico tools. Drug Discov. Dev. 8, 24–28. Pearl, G. M., Livingston-Carr, S., and Durham, S. K. (2001). Integration of computational analysis as a sentinel tool in toxicological assessments. Curr. Top. Med. Chem. 1, 247–255. Pennie, W., Pettit, S. D., and Lord, P. G. (2004). Toxicogenomics in risk assessment: An overview of an HESI collaborative research program. Environ. Hlth. Perspec. 112, 439–448. Piegorsch, W. W., and Zeiger, E. (1991). Measuring intra-assay agreement for the Ames Salmonella assay. In L. Hotorn (ed.) Statistical Methods in Toxicology, Lecture Notes in Medical Informatics, pp. 35–41. vol. 43. Springer-Verlag, Heidelberg, Germany. Tropsha, A., Gramatica, P., and Gombar, V. K. (2003). The importance of being earnest: Validation is the absolute essential for successful application and interpretation of QSPR models. Quant. Struct. Act. Relat. 22, 1–9. U.S. Food and Drug Administration (FDA). (2004). Challenge and opportunity on the critical path to new medical products, pp. 1–21. Available at: http:// www.fda.gov/oc/initiatives/criticalpath/whitepaper.html. Accessed January 1, 2007. Snyder, R. D. and Green J. W. (2001). A review of the genotoxicity of marketed pharmaceuticals. Mutat. Res. 488, 151–169. Van Gompel, J., Woestenborghs, F., Beerens, D., Mackie, C., Cahill, P. A., Knight, A. W., Billinton, N., Tweats, D. J., and Walmsley, R. M. (2005). An assessment of the utility of the yeast GreeenScreen assay in pharmaceutical screening. Mutagenesis 20, 449–454.